Approaches To Discourse Particles

APPROACHES TO DISCOURSE PARTICLES
STUDIES IN PRAGMATICS
General Editor: Bruce Fraser
Associate Editors: Kerstin Fischer, Maj-Britt Mosegaard Hansen
The Studies in Pragmatics series is dedicated to publishing

innovative, authoritative monographs and edited collections from
all micro-, macro- and metapragmatic linguistic perspectives.
Rooted in the interdisciplinary spirit of the Journal of Pragmatics, it
welcomes not only book proposals from linguistics proper but also
pragmatically-oriented proposals from neighboring disciplines such
as interactional sociology, language philosophy, communication sci-
ence, social psychology, cognitive science, and information science.
The goal of the series is to provide a widely read and respected
international forum for high quality theoretical, analytical, and
applied pragmatic studies of all types. By publishing leading edge
work on natural language practice, it seeks to extend our growing
knowledge of the forms, functions, and foundations of human
interaction.
Other titles in this series:
AIJMER & SIMON- Pragmatic Markers in Contrast

VANDENBERGEN
Forthcoming:
HABERLAND, CAFFI, Future Prospects for Pragmatics

HIRAGA & JANNEY
FETZER & FISCHER Lexical Markers of Common Grounds
Proposals for the series are welcome, please contact the General Editor,
Bruce Fraser: bfraser@bu.edu
APPROACHES TO DISCOURSE PARTICLES
EDITED BY
KERSTIN FISCHER
University of Bremen, Germany
Amsterdam • Boston • Heidelberg • London • New York • Oxford

Paris • San Diego • San Francisco • Singapore • Sydney • Tokyo
Elsevier
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, UK
Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands
First edition 2006
Copyright © 2006 Elsevier Ltd. All rights reserved
No part of this publication may be reproduced, stored in a retrieval system

or transmitted in any form or by any means electronic, mechanical, photocopying,
recording or otherwise without the prior written permission of the publisher
Permissions may be sought directly from Elsevier’s Science & Technology Rights
Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333;
email: permissions@elsevier.com. Alternatively you can submit your request online by
visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting
Obtaining permission to use Elsevier material
Notice
No responsibility is assumed by the publisher for any injury and/or damage to persons
or property as a matter of products liability, negligence or otherwise, or from any use
or operation of any methods, products, instructions or ideas contained in the material
herein. Because of rapid advances in the medical sciences, in particular, independent
verification of diagnoses and drug dosages should be made
British Library Cataloguing in Publication Data

A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress
ISBN-13: 978-0-08-044737-7
ISBN-10: 0-08-044737-6
ISSN: 1750-368X
For information on all Elsevier publications

visit our website at books.elsevier.com
Printed and bound in The Netherlands
06 07 08 09 10 10 9 8 7 6 5 4 3 2 1
Studies in Pragmatics (SiP)
General Editor
Bruce Fraser
Boston University, USA
Associate Editors
Kerstin Fischer
University of Bremen, Germany
Maj-Britt Mosegaard Hansen

University of Copenhagen, Denmark
Consulting Editor
Jacob L. Mey
University of Southern Denmark, Denmark
Editorial Board
Kent Bach, San Francisco State University, USA

Diane Blakemore, University of Salford, UK
Shoshana Blum-Kulka, Hebrew University, Israel
Claudia Caffi, University of Genoa, Italy
Alessandro Duranti, UCLA, USA
Marjorie Goodwin, UCLA, USA
Hartmut Haberland, University of Roskilde, Denmark
William F. Hanks, University of California, USA
Sachiko Ide, Tokyo Women’s University, Japan
Mikhail Ilyin, Moscow State Institute of International Relations of the
Foreign Affairs Ministry of the Russian Federation (MGIMO), Russia
Richard W. Janney, University of Munich, Germany
Kasia Jaszczolt, University of Cambridge, UK
Elizabeth Keating, University of Texas, USA
Sotaro Kita, University of Bristol, UK
Ron Kuzar, University of Haifa, Israel
Alec McHoul, Murdoch University, Australia
Brigitte Nerlich, Nottingham University, UK
Etsuko Oishi, Fuji Women’s University, Japan
Srikant Sarangi, Cardiff University, UK
Marina Sbisà, University of Trieste, Italy
Maxim Stamenov, Bulgarian Academy of Sciences, Bulgaria
Deborah Tannen, Georgetown University, USA
This Page is Intentionally Left Blank
Contents
1 Towards an understanding of the spectrum of approaches to discourse

particles: introduction to the volume 1
Kerstin Fischer
Part I: Polysemy-based approaches
2 A dynamic polysemy approach to the lexical semantics of discourse markers

(with an exemplary analysis of French toujours) 21
3 Discourse markers in English: a discourse-pragmatic view 43

Diana M. Lewis
4 The rise of discourse markers in Italian: a specific type of language change 61

Richard Waltereit
5 A functional approach to the study of discourse markers 77

Salvador Pons Bordería
6 Pragmatic markers in translation: a methodological proposal 101

Karin Aijmer, Ad Foolen, and Anne-Marie Simon-Vandenbergen
7 The description of text relation markers in the Geneva model of discourse

organization 115
Eddy Roulet
8 A dynamic approach to discourse particles 133

Henk Zeevat
Part II: Monosemy-based approaches
9 A relevance-theoretic approach to discourse particles in Singapore English 149

Ler Soon Lay Vivien
10 From procedural meaning to processing requirement 167

Thanh Nyan
11 Towards a theory of discourse markers 189

Bruce Fraser
12 What are particles good for? 205

Harald Weydt
viii Contents
13 The Natural Semantic Metalanguage approach to discourse markers 219

Catherine E. Travis
14 Epistemic modalities and the discourse particles of Singapore 243

Anthea Fraser Gupta
15 Integrating prosodic and contextual cues in the interpretation of discourse

markers 265
Li-chiung Yang
16 Formal properties of a subset of discourse markers: connectives 299

Corinne Rossari
17 Discourse marker research and theory: revisiting and 315

Deborah Schiffrin
18 Discourse markers as attentional cues at discourse transitions 339

Gisela Redeker
19 A dynamic-interactional approach to discourse markers 359

Barbara Frank-Job
20 Discourse particles as morphemes and as constructions 375

François Nemo
21 Discourse particles and modal particles as grammatical elements 403

Gabriele Diewald
22 Frames, constructions, and invariant meanings: the functional polysemy of

discourse particles 427
Kerstin Fischer
23 Discourse markers in Italian: towards a “compositional” meaning 449

Carla Bazzanella
References 465
Approaches to Discourse Particles
Kerstin Fischer (Editor)
© 2006 Elsevier Ltd. All rights reserved.
Chapter 1
Towards an understanding of the spectrum of

approaches to discourse particles: introduction
to the volume1
Kerstin Fischer
1. Aims
There are very many studies of discourse particles on the market, and by now it is almost
impossible to find one’s way through this jungle of publications.2 For a newcomer to the
field, it is furthermore often very difficult to find the bits and pieces that constitute an
original model of the meanings and functions of discourse particles. Moreover, the studies
available so far are hardly comparable; the approaches vary with respect to very many
different aspects: the language(s) under consideration, the items taken into account, the
terminology used, the functions considered, the problems focussed on, and the
methodologies employed. Some kind of overview is needed that allows us to sort out the
different research directions, methods, and perspectives. This collection constitutes an
initial attempt to present such a path through the jungle of different approaches.
Any attempt to provide an overview of the state of the art is, however, in danger of
denying the breadth and heterogeneity of the research field and the complexity of the
problems involved. Aiming at the impossible, the major challenge is thus how to present
the spectrum of approaches to discourse particles in a single volume. The procedure taken
here is to combine as many original perspectives on discourse particles as possible in
order to get an idea of what is actually there and then to see what we can learn from each
different perspective. Thus, the volume comprises synchronic and diachronic, formal and
informal approaches; approaches building on text-linguistic models; models of general
cognitive processing or interactively relevant domains of discourse; as well as approaches
concentrating on syntactic, semantic, pragmatic, or prosodic aspects. The result is a very
heterogeneous collection, however, no more (rather, less) heterogeneous than the research
field in general, since the collection presented here can self-evidently only provide a
glimpse into the field with its many valuable findings and perspectives; the field is in fact
even richer.
Reflecting the state of the art in discourse particle research requires presenting the well-
known scholars whose perspectives may already be under discussion, yet whose work has
2 K. Fischer
not yet been displayed in a comparable form, as well as less well-received approaches that
provide a new, interesting look at the issues under consideration. In order to reflect the
current breadth of the discussion, it is essential that no selection according to particular
methodological orientations, perspectives, or data considered is taken. Thus, after a
number of well-known authors had been encouraged to contribute to the volume, the
initial selection was supplemented by a public call for papers.
The second challenge concerns how the spectrum of approaches can be presented in its
richness and variability such that the differences and similarities between the approaches
become apparent and comparable. The current volume attempts to address this problem by
asking the researchers from their various different linguistic backgrounds to describe their
particular ways of addressing some of the most important problem areas that are involved
in the description of discourse particles. Thus, in order to make the different approaches
comparable, all of the authors were asked to compose their articles according to a given
schema. In this way, the strengths and weaknesses, coverage and limitations of each
approach become visible and comparable. Since we are all finite, it can be expected that
our models can only provide some very restricted answers as well. However, an overview
of the available bits and pieces may support us in defining where we currently are and
where we may want to go next. The collection should therefore provide us with a very
first step towards identifying the achievements, as well as the problems, of the field.
The schema corresponds to some (in the editor’s view) central problem areas for which an
account of the meanings and functions of discourse particles has to provide solutions. The
idea is thus that in spite of all the variation, the different theoretical perspectives on
discourse particles may be made visible and comparable. The four central questions
addressed, which are taken to be essential for any comprehensive approach to discourse
particles, are outlined in the following.
The first challenge consists in providing a definition of discourse particles by describing

the characteristics of the items under consideration and by developing criteria for deciding
for every given particle instance whether it is a discourse particle. Additionally, the
relationship between discourse particles and other word classes, such as modal and focus
particles, conjunctions, adverbs, and the like, needs to be explained. The problem areas
the authors of this volume address in the first section of their papers are thus, for instance:
• the distinction between discourse particles and other particles,

• the question whether discourse particles constitute a semantic, syntactic, or functional
class (or no class at all),
• the status of the definition of discourse particles; for instance, whether it states
necessary and sufficient conditions or whether it describes the prototype only,
• which terminology they prefer to use.
Terminological problems are very common in scientific research. In the case of discourse
particles, asking the authors for their definitions may reveal initial assumptions about the
field itself. If the field was homogeneous, much redundancy could be expected in the
definition sections. However, besides some general linguistic properties that are
mentioned by several authors, there is surprisingly little overlap in the different
definitions. In case of redundancy, I would kindly ask the reader to excuse this minor
Introduction 3
drawback in favour of the advantage of being able to see the diversity and richness of the
whole field displayed in this first section.
The second problem area consists in accounting for the different interpretations of the
items of the class, that is, for the functional spectrum of discourse particles. A model of
the semantics/pragmatics of discourse particles has to describe precisely and exhaustively
the different readings a particle may support. As for all lexical items, principles of
learnability, interpretability, and plausibility demand that the readings of discourse
particles be discrete, and that the number of possible interpretations be finite, that is, there
should be a “plausible” number of well-defined, identifiable readings. The criterion of
identifiability includes that the conditions for a given particle to receive a particular
interpretation should be clear. Thus, the different readings of discourse particles should be
described in connection with the different structural contexts in which they occur. To
summarize, the problems the authors address in the second section include the following:
• the many different functional interpretations discourse particles may support,

• the different communicative purposes they may serve,
• the context sensitivity of their meanings,
• their “homonyms” in other word classes,
• the relationship between functional interpretation and syntactic position.
Another major challenge consists in accounting for the relationship between the different
readings and the relationship between the readings and the particle lexeme. Just listing the
different interpretations treats the items under consideration as homonymous; such an
approach does not account for our intuition of the relatedness of these meanings, and it
leaves unexplained how the interpretations observable are learnable and how contextual
occurrences are interpretable. Moreover, the relationship between each particle lexeme
and its interpretations has to be described in order to explain why just this particle can get
just these particular interpretations and not others. Consequently, in their third sections the
authors were asked to present their solutions to questions such as:
• how to account for the relationship between the different readings,

• how to account for the relationship between the different readings and the context,
• how to account for the relationship between the readings and the particle morpheme,
• how to account for the relationship between the different readings in the different word
classes,
• how to account for the relationship between functional interpretation and syntactic
position.
The fourth problem area involved in the description of discourse particles consists in
relating the study of discourse particles to other questions of general linguistic interest,
such as the semantics/pragmatics interface, the nature and levels of discourse, or
communicative functions. General linguistic distinctions should be mirrored in the
architecture of the model, so that particles may finally lose their exceptional status (see
Wilkins, 1992). In addition, the model should be built on those categories that can be
shown to be general categories of the communicative situation to which speakers attend,
and the properties of discourse particles, word class definitions, etc., should follow as
natural results of the interaction of the components of the model. Furthermore, the model
4 K. Fischer
developed should allow the comparison of the individual aspects of the model between
languages.
These four problem areas constitute the schema on the basis of which the articles in this
volume were organised. This procedure is intended to enhance the comparability of the
different approaches for the prospective readers and to support them in identifying the
characteristics and individual and original solutions of each approach. The authors were
furthermore asked to describe briefly their approaches, what their methods and data
consist of, and what they consider to be the most pressing problems, at the beginning of
their chapters, so that each approach can be located within the spectrum of current
approaches to discourse particles.
This presentation of approaches constitutes a first step towards the understanding of the
variability of the spectrum of possible perspectives on the analysis of discourse particles.
Presenting some of the approaches to discourse particles on the linguistic market in this
comparable way may help us understand the heterogeneity of the field and to identify
those parameters in which the various approaches differ, as well as the common
assumptions. Further research will have to show in how far the volume can contribute to
structuring the field of discourse particle research in general. An initial attempt to
systematise the spectrum of approaches on the basis of the results of this volume is
presented in sections 3 and 4.
2. Results from the discussion
Besides presenting their approaches to discourse particles, the authors of this volume have
been engaged in a lively discussion via e-mail with basically two objectives. The first
question was whether we can all agree on some terminology. As Bruce Fraser pointed out,
“we can’t even talk to one another without a clarifying statement”. Thus, there was an
acute problem for the authors to find a common denominator for discussion, and most of
the authors write in their problem statements that they consider the terminological/
definitional issues to be problematic. As a second objective, the problem of the units to
which discourse particles refer, which they mark or bracket, was raised.
The discussion on terminology focused on the two labels discourse particle versus
discourse marker. However, since terminological issues always also mirror conceptual
distinctions, a wide range of issues entered the discussion. The dissent about the
appropriate term to use could of course not be resolved, but the discussion has contributed
to some clarification. Because both terms, discourse particle and discourse marker, are so
controversial, the term discourse word (from the common French label mots de discours)
was also discussed, but it was generally agreed that it was too vague to be useful.
The term discourse particle suggests a focus on small, uninflected words that are only
loosely integrated into the sentence structure, if at all. The term particle is used in contrast
to clitics, full words, and bound morphemes. Using the term discourse particle
furthermore distinguishes discourse particles/markers from larger entities, such as phrasal
idioms, that fulfil similar functions.
Distinguishing discourse particles from speech routines, pauses, adverbs, etc., which are
functionally very similar, however, may lead us to too a narrow picture. It involves
Introduction 5
concentrating on a formally distinct class of items that may rather belong to a much larger
functional continuum. Furthermore, there seem to be typological differences: what is
expressed by particles in one language may be expressed by very heavy speech formulae
in another. While we have to see that traditional Chinese grammar recognises discourse
particles as a word class, in French we have to deal with the rather heavy size and low
flexibility of multiword units. Similarly, what some languages have modal particles for is
expressed by adverbs in another. Many authors therefore argued that the class should be
described entirely from a functional point of view.
The term discourse marker was regarded to be a purely functional term. The term was
furthermore suggested to be the most wide-spread and considered to be the most
inclusive. An advantage is certainly that a functional characterisation of the class may
avoid unnecessary formal limitations. After all, as argued above, many items commonly
discussed as discourse particles are actually inflected and are not small words but
formulas consisting of more than one word or having evolved out of complex units (like
English I know or I mean or Italian guarda). However, there are several open questions
with respect to purely functional definitions.
First of all, many functions that discourse particles/markers can fulfil are also fulfilled by
other items. For instance, conversational management functions are also fulfilled by
speech formulas and nonlexicalized metalinguistic devices, such as au risque de me
répéter. Stance can be expressed by, among others, modal verbs, adverbs, parenthetic
clauses, or tag questions. And linking functions can also be fulfilled by conjunctions and
speech formulas. We therefore have to ask whether a particular functional range can be
found that can be used to identify a discourse particle/marker. It seems that the functions
the items under consideration are suggested to fulfil are extremely diverse and that the
only uniting property is the fact that they can all be fulfilled by such items. The
classification is thus in danger of being circular: we call discourse markers those items
that fulfils discourse marking functions, and we call discourse marking functions those
that are fulfilled by discourse markers.
Secondly, a purely functional classification would not allow a distinction according to the
size of the unit under consideration and the degree of idiomatization or the loss of
semantic meaning of the respective items. At the same time, research on discourse
particles/markers usually concentrates on items that are prototypically particles,
connectives, or fixed phrases like you know and I mean. That is, research may actually be
restricted to the investigation of lexicalized items with non-truthfunctional (bleached),
idiomatic meanings, the prototype being particles, and may not be open to functionally
similar freely constructed phrases or sentences. Other practices, like reformulations,
hesitations, false starts, and pauses, for instance, are, in spite of their functional similarity
with the items considered, usually not taken into account. Thus, indeed particular formal
decisions are usually taken, and the classification therefore does not occur on an entirely
functional basis. At least lexicalization and idiomatization of the respective items are
usually taken for granted. The choice of the term discourse particle versus discourse
marker thus reflects in how far the decisions regarding the formal aspects of the items
under discussion are made explicit.
Another problem connected to the term discourse marker is its unclear semiotic status.
Basically, three different positions can be distinguished. On the one hand, it was argued
6 K. Fischer
that discourse particles/markers do not mark anything, but that they, like every other
linguistic sign as well, create meanings and are thus not substantially different from other
lexical items. Most of the authors of the volume indeed hold that the items under
consideration have an encoded meaning. The question arising then is what the nature of
the meaning involved could be if they are only markers.
On the other hand, it was proposed that marking and creating meanings are not opposed
to each other but that the items under discussion may actually do both: mark particular
meanings, in the same way as sociolinguistic variables reflect particular situations of
formality, power, solidarity and the like, and create those meanings in being used
strategically to construct such a situation. The question then is whether and how those two
processes can be distinguished and whether the term marker is appropriate since it focuses
only on one of two different interpretative mechanisms.
The third position was that discourse particles/markers have a procedural or instructional
meaning and that they therefore are linguistic signs different from other lexical items. In
this view, the term discourse marker is thus fully justified, because they indeed do no
more than guide the interpretation of the “real” signs.
Figure 1 visualises some of the results of the terminological discussion. The items under
consideration can be defined on both the formal and the functional side. On the functional
side, they are taken to fulfil discourse functions, which are understood to be a subset of
pragmatic functions in general. On the formal side, we can distinguish between non-
lexicalized and lexicalized items, the latter of which can be further distinguished into
particles and others. Discourse particles in this typology are items that are both formally
and functionally defined, while discourse markers may be both lexicalized, including
particles, and nonlexicalized items that fulfil discourse functions.
To sum up the terminology discussion, it seems that the issue is not only related to the
decision of taking a semasiological or onomasiological starting point, or to taking into
account formal or only functional criteria. It also seems to involve the consideration of
what pragmatic meanings and the interpretative mechanisms involved are like and thus
anchors the study of discourse particles directly in the general discussion of the field.
The discussion proceeded then to the question which units discourse particles may mark
or which units they connect. One problem addressed was whether these units are segments
of discourse. For instance, in the sentence “Peter came but too late”, what would the
segments be? To hold that discourse particles/markers mark or create relationships
between discourse segments, such as utterances, would thus exclude very many of their
uses. The proposal instead was to say that discourse particles/markers connect discourse
contents rather than segments, including contents not explicitly mentioned. That is, they
could create or mark relationships between actual, virtual (attributed), or presupposed
utterances, as well as aspects of discourse memory.
For some authors, this definition was still too narrow. In their accounts, discourse units
include speech acts, turns at talk, or the participation structure. The “units” involved in
their accounts can be paraphrased as “aspects/planes/domains of discourse”.
Introduction 7
Figure 1. Discourse particles versus discourse markers
lexicalized items
items fulfilling discourse functions
particles
discourse
particles
discourse
markers
The problem with this approach concerns the question whether the term unit of discourse
would be appropriate here at all. On the one hand it can be asked whether a discourse
plane constitutes a unit of discourse rather than, say, a unit of the discourse situation. On
the other hand, it is questionable whether aspects like the participation structure constitute
units of discourse rather than their conditions or circumstances.
Finally, the question was raised whether discourse particles/markers should really be
defined by the property to create or mark relationships between two units at all. Stance
marking, for instance, does not necessarily involve relationships between units. Similarly,
the concept of creating or marking a relationship would also be pushed very far in the case
of interjections, hesitation markers, or feedback signals, which some authors would want
to include in their investigations. In contrast, other authors are content with the fact that
the definition excludes such items. Thus, there is no common view on the role of the unit
relating function for the definition of the category nor on the items that may belong to it.
The results from the discussion reflect the enormous breadth of possible perspectives on
discourse particles/markers and the diversity of views regarding which items should be
considered, how they should be labelled, which functions they fulfil, and which units they
act upon. However, the hope is that the discussion and the spectrum of different
approaches presented by the authors in this volume will allow us in the long run to
understand the field in its breadth a little better. I will turn to the issues last discussed
now: the question of the types of items to be taken into account, the role of the
relationship marking/creating function, and the types of units indexed. I would like to
propose an initial systematisation on the basis of the feature of integratedness: the degree
to which a discourse particle is part of a host unit. In section 4, another systematisation
will be suggested on the basis of the different mechanisms proposed to account for the
polyfunctionality of discourse particles.
8 K. Fischer
3. Towards an understanding of the spectrum of approaches
The proposal made here is that the dimension of integratedness constitutes a useful
criterion to account for some of the variability of the spectrum of approaches to discourse
particles. Two poles on opposite ends of this dimension can be identified: on the one
hand, there are those items that constitute parts of utterances, such as connectives; on the
other, there are completely unintegrated items that may constitute independent utterances,
such as feedback signals or interjections.
Figure 2. Dimension of integratedness
items integrated into host items constituting

utterances independent utterances
An example for the first kind of items from the Verbmobil appointment scheduling
dialogues would be:3
(1) mdmr_3_06: yes, I’m free two to five on Wednesday. so how ’bout meeting three
to five?
Focus on such items can be illustrated with a number of citations from definitions of
discourse particles in this volume; for instance, Fraser speaks about “discourse segments
that host them”, that is, he assumes that discourse particles form parts of sentences. Lewis
even speaks of “syntactic hosts”, Ler addresses “utterances in which they occur”, and
Hansen describes discourse particles as “instructions to the hearer on how to integrate
host utterances into a developing model of the discourse”. Thus, there are a number of
researchers who take discourse particles (or discourse markers) to occur integratedly into
some host utterances regarding which they do particular linguistic work.
However, there are also a number of approaches that define discourse particles by means
of the property of unintegratedness, that is, as items that constitute utterances themselves.
Examples would be:
(2) flmb_6_05: twenty ninth I think we can agree is horrible for both of us,
 and, oh, let’s see, on the thirtieth, the thirtieth’s pretty
horrible too.
(3) mbjr_1_12: alright, so, it’s, sixteenth, one to three, that’s confirmed? thanks,
nice doing business with you,
mder_1_13: yeah,
Proponents of this view of discourse particles are, for instance, Diewald, who speaks of
grammatical unintegratedness; Schiffrin, who defines them as “syntactically detachable”;
Travis, who outlines in detail how she understands discourse particles to be prosodically,
syntactically, and semantically independent; Yang, to whom discourse particles are
Introduction 9
“syntactically independent”; and myself: they are “syntactically, semantically and often
prosodically unintegrated”.
We can conclude that there are approaches to discourse particles that focus on items that
are on the opposite ends of a dimension of integratedness. These different foci have a
number of consequences. First of all, as it already emerged from the discussion reported
on in section 2, very different items may be considered as discourse particles, a
consequence being the heterogeneity of the class. Furthermore, focus on a particular status
of integratedness influences the functional spectrum observable. Thus, approaches that
focus on integrated items usually focus on the connecting function of discourse
particles/markers. For instance, Pons Bordería and Rossari focus on connectives from the
outset, and Fraser, Hansen, Redeker, and Roulet also concentrate on the connecting,
coherence-establishing function.
In contrast, approaches that focus on unintegrated items mainly address the roles
discourse particles may play in the management of conversation. The functions they may
fulfil concern domains such as the sequential structure of the dialogue, the turn-taking
system, speech management, interpersonal management, the topic structure, and
participation frameworks (see, for instance, Frank-Job, Travis, Schiffrin, and Yang).
Figure 3. Dimensions of integratedness and function
items integrated into host items constituting independent

utterances utterances
connecting functions regarding

function conversation management
The dimension of integratedness thus corresponds in part to a focus on particular

functional aspects. A further dimension that correlates with both the integratedness and
the functional spectrum considered concerns the data taken into consideration. While
approaches that focus on integrated items often also account for written text, approaches
that focus on unintegrated items focus on conversation. The reason is that the more
integrated an item is in its surrounding cotext, the more reference elements are retrievable
from the cotext, the more easily it can occur in (de-contextualised) written discourse.
Thus, approaches focusing on the connecting function of discourse particles/markers (in
this volume, for instance, Fraser, Lewis, Hansen, Nemo, Rossari, and Roulet) also
account for items occurring in written discourse. Conversely, the more the interpretation
of items is dependent on aspects of the communicative situation, like speech management,
the participation framework, or the turn-taking system, the less relevant they are for
written communication. The approaches concentrating on these kinds of items usually
restrict themselves to spoken, conversational interaction (proponents of this approach in
this volume are, for instance, Frank-Job, Schiffrin, Travis,Yang, and myself).
10 K. Fischer
Figure 4. Dimensions of integratedness, function, and data
items integrated into host items constituting independent

utterances utterances

spoken and conversation

written text
Thus, structuring the spectrum of approaches to discourse particles according to the

dimension of integratedness provides a useful basis for systematising both the different
foci on the functional spectra considered in the different analyses as well as the
differences in the types of data considered in the different studies. However, the
systematisation proposed so far has presented only the two opposing poles of the
dimension. Many approaches in this volume do not focus on one of these poles (e.g.,
Bazzanella, Diewald, Frank-Job, Schiffrin, Yang) or acknowledge the whole spectrum of
discourse particle uses in other ways. This is in accordance with the empirical fact that
often the same phonological/orthographic form can occur in positions that vary regarding
the integratedness of the item. For instance, the example below shows that so, a good
candidate for connecting uses, as shown in example (1) above, can constitute a completely
independent unit, with no host utterance identifiable:
(4) mggd_1_13: you know, I don’t even feel like thinking any more, /begin laughter/
’cause I did all that at work, /end laughter/. so um
anyway, you can find me at, um at work.
Similarly, in the following examples, so and um occur in very similar contexts, the latter
being a prime candidate for items fulfilling conversation management functions:
(5) mdmr_1_03: I have a meeting from two to four. so how ’bout Tuesday from three
to four?
(6) mkps_6_06: I’ll be busy the entire first week of October. um how about the
Friday? the first?
Thus, while there are these opposing poles identifiable between unintegrated items, such
as so in example (4), which serve conversation management purposes, and connecting
uses, such as so in (1), it may be the same items that occur in both contexts and with both
functions. Correspondingly, the authors in this volume, even if they focus on only one
particular type of usage, regard those uses as a “subclass of discourse markers” (Pons
Bordería, Redeker, Rossari), acknowledging their other functions as well. For instance,
Introduction 11
Fraser gives a list of classes of discourse particles, among them connective particles, on
which he concentrates later on. Consequently, although the authors in this volume may
focus on particular uses of discourse particles, the breadth of the whole functional
spectrum is generally acknowledged and can be taken to be common ground.
A fourth dimension can be correlated with our previous systematisation along the
dimension of integratedness, namely, the different proposals for what may constitute the
host units, what exactly the discourse particle occurrences are integrated in (see Schiffrin,
this volume). The host units specify the reference elements of the respective discourse
particle, that is, that aspect of the cotext or (local or larger, sequential or situational)
context regarding which the respective discourse particle is doing its linguistic work.
Here, along with our previous distinctions, we can see that those approaches that focus on
integrated uses with connecting functions regard different aspects of utterances as the host
units. They distinguish, for instance, like Nyan, “informational or argumentational
contents, the act of utterance, speech acts, the fact that a speech act occurred, S’s attitude/
commitment to a view” as “aspects of host utterances” in relation to which discourse
particles may fulfil their pragmatic functions. In contrast, approaches that focus on
syntactically, semantically and prosodically unintegrated items may also consider host
units in the sense of domains or planes of reference. The units considered here may be
constituted by, for instance, the topic structure, extralinguistic activities, or participation
frameworks (e.g., Bazzanella, Fischer, Frank-Job, Schiffrin).
Figure 5. Dimensions of integratedness, function, data, and host units
items integrated into host items constituting

utterances independent utterances

spoken and conversation

written text
aspects of topics, activities,

utterances sequential structures
While the perspective discussed here so far outlines a model of the spectrum of
approaches that is oriented at two opposing poles, it also hints at general mechanisms that
may constitute a joint basis for all of the approaches to discourse particles. Thus, my
suggestion is that the distinctions we have worked out so far can be used to find a unified
view of the functional spectrum of discourse particles. In particular, irrespective of
whether an item may occur integratedly in a host utterance or unintegratedly, it will point
to something outside itself, either to an aspect of the host utterance, or to a contextual,
12 K. Fischer
situational factor (domains, discourse planes, etc.). Accordingly, it can be found in many
papers that deixis is taken to be a key feature of the pragmatic functioning of discourse
particles (e.g., Aijmer et al., Diewald, Fischer, Frank-Job, Schiffrin), and deixis seems to
be a central mechanism involved in accounts of the polyfunctionality of discourse
particles (see section 4 below). To contain deictic elements is thus one important property
of discourse particles that is recognised in one way or another in most of the approaches
in this volume. So diverse the approaches to discourse particles presented in this volume
thus may be, some of the variability may be accounted for by means of a model that takes
into consideration the different host units associated with the different structural positions
in which the items may occur.
To sum up, the systematisation according to the dimension of integratedness helps us to

understand a number of dimensions of variability between the different approaches. In
particular, using the dimension of integratedness we can take the first steps in a
systematisation of the different approaches regarding differences with respect to:
• the items considered (e.g., predominantly connectives vs. predominantly interjections,

feedback signals, hesitation and segmentation markers, etc.)
• the functions determined (e.g., connecting vs. conversation management related
functions)
• the types of data considered (i.e., written text vs. conversation)
• the types of host units recognised (i.e., aspects of host utterances vs. larger host units
such as topics, activities, participation frameworks).
The dimensions proposed also provide a unified look at a possibly general mechanism
that may account for the whole functional spectrum of discourse particles and that could
constitute the basis for future, comprehensive models, namely, the concept of reference to
different host units that accounts for the different behaviour of discourse particles with
respect to the four dimensions discussed. Yet, other dimensions of structuring the
spectrum of approaches in this volume are also possible (see also Nemo, this volume).
The presentation of the different perspectives on discourse particles presented in this
volume will provide the opportunity to shed increasingly more light on the variability of
different perspectives for future research. As another example for a possible dimension of
variability of the spectrum of approaches, the different proposals for how to account for
the polyfunctionality of discourse particles will be discussed in the next section. That is,
one of the questions addressed by the authors in this volume is the problem of how a
single phonological/orthographic form can have so many different possible readings. On
the basis of the question of how it addresses the problem of polyfunctionality, each
approach will be briefly introduced and a systematisation of the possible approaches to
this problem will be developed.
4. Models of polyfunctionality
Discourse particles/markers are acknowledgedly polyfunctional, and this property has

been the focus of many studies. In this volume, we find a considerable spectrum of
possible ways of dealing with the problem of bridging the gap between the single
phonological/orthographic form and the many different possible interpretations associated
with this form.
Introduction 13
4.1. Approaches to polyfunctionality
Approaches to the polyfunctionality of lexical items, such as discourse particles, can be

ordered taking as a basis the question of which meaning aspects are taken up in the lexical
representation. On the one side of this dimension, there are the so-called monosemy
approaches, on the other the so-called homonymy approaches.
• Monosemy: Each phonological/orthographic form is associated with a single invariant

meaning. This invariant meaning may describe the common core of the occurrences of
the item under consideration, its prototype, or an instruction. Individual interpretations
arise from general pragmatic processes and are not attributed to the item itself.
• Homonymy: There are a number of readings that are identifiable as distinct. No
relationship between the different readings is assumed, and the different senses are
described in numbered or unnumbered lists, sometimes associated with their
conditions of usage, such as, for instance, the structural contexts in which they occur.
Thus, while monosemy approaches assume a single meaning that may be instantiated in
context, homonymy approaches spell out the different interpretations, yet assume them
not to be related. These two opposed ways of dealing with the problem are widely
acknowledged in the literature on lexical semantics (see Lyons, 1977; Cruse, 2000: 109,
114). In between these two oppositely oriented poles of the spectrum, there are numerous
different approaches to the fact that one orthographic/phonological form can have
different interpretations. In particular, it is assumed that there are different distinct
readings and that these different senses are related.
• Polysemy: A single phonological/orthographic form may be used with a number of

different, recognisable interpretations that are assumed to be related.
Sometimes, the term polysemy is only associated with one particular approach to the
relationship between the different readings (e.g., Hansen, this volume), in particular, an
approach that does not assume a single invariant meaning component but general
relationships between distinct readings. The relationships postulated usually apply to
many different independent domains (like metaphorical or metonymic relations) or are the
lexicalized results of other general mechanisms, such as implicature (e.g., Waltereit, this
volume).
• Polysemy in the narrow sense: A single phonological/orthographic form is associated

with a number of distinct readings that are related by a set of general relationships.
These readings do not necessarily share common meaning aspects.
Besides this way of accounting for the relationships between the different senses, there is
a broad spectrum of models that take the monosemy approach as a starting point but
which furthermore attempt to account for the different senses observable by providing
models of mechanisms that relate the invariant meaning to the distinct but motivated
readings. The various mechanisms include at least the following:
• Describing the invariant meaning and providing a general mechanism that allows the
retrieval of the individual interpretations in context.
14 K. Fischer
• Describing the invariant meaning and listing the individual readings. In this approach,
the different senses of the item in use are taken to be richer and more specific in their
semantic content than the core meaning component. The individual readings all
contain the core components plus further specifications. The individual senses are
taken to be lexicalized.
• Describing the invariant meaning and proposing additional mechanisms that contribute
further, more specific, meaning components. The additional mechanisms proposed,
such as syntactic/semantic constructions or prosody, are not specific to single items
but are proposed to be more general.
• Describing the invariant meaning and assuming a system of pragmatic parameters that
select the respective context-dependent meaning.
• Describing the invariant meaning and accounting for the observable senses by means
of mapping to different domains that are part of a general model of discourse. In these
models, the domains referred to provide the specifications of the invariant meaning
component. Since the domains provide accounts of the observable senses, much
attention is devoted to their description in these models.
• Describing the invariant meaning, mapping it onto different domains, and constraining
the possible meanings and their combinations furthermore by providing general
structural mechanisms.
To sum up, there is a broad spectrum of possible ways of accounting for the fact that a
single phonological/orthographic form can have different interpretations. Models vary
particularly regarding two aspects. On the one hand, they differ regarding whether the
different interpretations are assumed to be related and what the relations between the
different interpretations are attributed to (e.g., common meaning components, general
conceptual or rhetorical relations, domains of reference, etc.). On the other, approaches
vary regarding the inclusion of contextual factors, such as syntactic-semantic
constructions or prosody. The uniqueness of each approach results from their original
combinations of solutions to these problems.
4.2. The approaches to the polyfunctionality of discourse particles in this volume
The approaches to the polyfunctionality of discourse particles combined in this volume

reflect the complex picture of possible ways of dealing with the problem in general.4
Some models have not been developed particularly for the description of discourse
particles, and others have been used for the description of the polysemy of other linguistic
material as well.
None of the authors in this volume defends a homonymy approach. That is, there seems to
be general agreement on the relatedness of the different interpretations of a single
phonological/orthographic form, often even across word class boundaries (e.g., Diewald,
Fischer, Nemo, Weydt). The approaches in this volume can be distinguished into those
that assume a number of distinct meanings that are related by general relationships
(polysemy in the narrow sense) and those that assume a common core component that
accounts for the relatedness of the individual interpretations (monosemy). This latter
category can then be distinguished further.
Introduction 15
First of all, representatives of the polysemy approach in the narrow sense are Hansen,
Lewis, and Waltereit. Interestingly, all three defendants of the polysemy approach
combine a diachronic view with the synchronic perspective:
• Maj-Britt Mosegaard Hansen proposes an approach to discourse markers, illustrated

with the French discourse marker toujours, in which no core semantic meaning is
assumed to connect the different readings of an item. Instead, the different senses are
related by metaphoric or metonymic extension. Defining discourse marker meanings
in networks such as radial categories or family resemblances has in her view the
advantage of accounting for both the relatedness of the meanings and for the
possibility of semantic change. Her model is thus designed to allow a particularly
dynamic perspective on discourse marker meanings.
• Diana Lewis defines discourse markers as form-function mappings, combining the

semantics of discourse-relational predications, including speaker-attitudinal, relational
meaning, with certain types of linguistic realisations that are syntactically dependent
on their host (thus excluding interjections). The functional spectrum of discourse
particles in her view comprises rhetorical management, information structuring, and
thematic organisation. She also defends a polysemy approach to discourse particles (in
the narrow sense) which she supports with diachronic analyses, thus presenting the
synchronic state of polysemy as a result of diachronic developments.
• Richard Waltereit takes a diachronic perspective to explain the functional variability of

discourse markers as well. By asking what turns a linguistic item into a discourse
marker, he isolates rhetorical strategies by means of which speakers manipulate the
structure of the discourse or the interaction. In this way, he explains the development
of both new distributional contexts and new functions. The functional spectrum
observable is in his account a result of the coexistence of uses developed in different
diachronic stages in order to fulfil particular communicative needs.
Similarly, Pons Bordería holds that a polysemy approach in which the individual
meanings are related by means of family resemblances would account best for his
functional approach. However, in contrast to the other authors, he argues that it is just far
too early to propose models of polyfunctionality:
• Salvador Pons Bordería defines discourse markedness as a hyperonym of three

different functions: interactive, modal, and connective. He illustrates these functions,
drawing on a wide range of descriptive methods and approaches, with the discourse
markers of colloquial Spanish. He argues that a model whose only purpose is to
explain what discourse particles/markers are is of not much use. Instead, what is
needed is a model of discourse that provides a place to locate these markers within.
With the example of the definition of the units of discourse, he illustrates how an
investigation of discourse markers can be used to develop aspects of a theory of
discourse.
Then there are some approaches that assume a relationship between the different readings
of an item but which allow this relationship to be either specified by common meaning
components or by conceptual or rhetorical links. This is particulary argued for by Aijmer,
Foolen, and Simon-Vandenbergen, and it is implicit in Roulet’s and Zeevat’s papers:
16 K. Fischer
• Karin Aijmer, Ad Foolen, and Anne-Marie Simon-Vandenbergen analyse mainly

English discourse particles, but they also look at the translation of the English
discourse particles into other languages, such as Swedish. Their model of the
polyfunctionality of discourse particles allows both a core meaning representation of
the item under consideration as well as the description of several related senses. The
core functions of discourse particles they hold to be epistemic or affective. The
pragmatic interpretations of a discourse particle in their account is related to the
indexicality of discourse particles and the heteroglossia of dialogues.
• Eddy Roulet outlines not only a model whose purpose is to explain what text relation
markers (TRMs) are but a model of discourse which provides a place to locate these
markers within (cf. Pons Bordería, this volume). In his view, accounting for the
different functions TRMs can fulfil means providing a model of text units, text
relations, and the role of TRMs therein. The TRMs themselves are considered to
indicate illocutionary or interactive relations between a text constitutent and
information stored in discourse memory and to provide instructions for the reader/
hearer to facilitate access to the relevant information. The different readings of each
TRM are conceptualized as distinct but related.
• Henk Zeevat defines discourse particles as context markers, relating the current
sentence to aspects of common ground. He develops a formal semantic and pragmatic
account for a number of English, Dutch, and German particles that have previously
been discussed in connection with presupposition. In Zeevat’s model, polysemy arises
if a discourse particle marks two different relations to the context. The broader
framework for his model could be a theory of speech act marking.
The other approaches in this volume generally assume a single meaning for each
phonological/orthographic form, that is, they take a monosemy approach. First of all, we
find approaches in which general mechanisms account for the polyfunctionality of
monosemous items. Ler employs the relevance theoretic framework, whereas Nyan uses
the argumentation theory approach:
• Ler Soon Lay Vivien takes a relevance theoretic approach to discourse particles,
investigating clause-final discourse particles in Singapore Colloquial English. The
polyfunctionality of discourse particles she addresses by postulating a single
instruction that is applied to different contexts. The inherent meaning of a discourse
particle therefore accounts for the relatedness of the different readings, while the
polyfunctionality results from the fact that each discourse particle use has to be
interpreted against the background of a new context.
• Thanh Nyan’s analysis of French discourse markers is based on Anscombre and

Ducrot’s Argumentation Theory (AT), augmented by neurolinguistic considerations of
pragmatic processing. With AT, she holds discourse markers to have an instructional
core meaning, and she is mainly concerned with their argumentative and cohesive
functions. Taking a monosemy approach, AT possesses a conceptual apparatus
susceptible of providing a general account of how the core meaning descriptions create
different interpretations in context.
Introduction 17
Fraser and Weydt then do not specify any mechanism, but they describe those factors that
determine the interpretation and that may cause pragmatic effects, such as politeness:
• Bruce Fraser outlines a typology of pragmatic markers of which discourse markers

constitute one subclass. Concentrating on those items that fulfil connecting functions,
he attempts a systematisation of their formal and semantic properties. His model of the
polyfunctionality of discourse markers comprises the core meaning description of the
prototype, the interpretation and inferences of the utterances connected by the
discourse marker, and linguistic and extralinguistic context factors.
• Harald Weydt approaches the nature of the meanings and functions of particles by
taking the question as to what particles are good for as a starting point. First of all, he
reports on a study (Weydt et al., 1983) that shows that particles contribute to the
impression of naturalness and friendliness of a dialogue. He then develops his model
of how particles come to fulfil this kind of function step by step. Defending a
monosemic approach to particles, he draws a clear distinction between the encoded
semantic meaning of particles, holding this to be the same across word classes, and the
individual uses, which may be listed in a dictionary. The pragmatic function to create a
harmonious, friendly atmosphere is an aspect of their usage: they show that the
speaker “is aware of what B, the other, thinks and believes.”
Travis goes a step further by spelling out a set of specific readings that include the core
meaning component related to the lexeme:
• Catherine Travis argues that a monosemantic view does not exclude a polysemy (in
the narrow sense) approach to discourse particles. By using Wierzbicka’s Natural
Semantic Metalanguage (Wierzbicka, 1996), she shows that a core meaning
description can indicate how the different readings of a discourse particle are related,
but at the same time different senses can be defined as extensions of the core
meanings. She exemplifies her model on the Spanish discourse particle bueno,
showing furthermore how structural position and intonation contour contribute to the
disambiguation of the different senses.
Similar to Travis’ disambiguation of discourse particle interpretations by means of

contextual factors, there are a number of approaches that take structural information to
function as parameters for the selection of the right interpretation, only one of which is the
invariant meaning component. Whereas Gupta concentrates on the contribution of
sentence types, Yang and Rossari analyse the contribution of very complex contexts:
• Anthea Fraser Gupta proposes a definition of discourse particles as a very particular

word class of Singlish, Singapore Colloquial English. This word class is directly
associated with the function of signaling epistemic modality, which explains the
syntactically and semantically peripheral status of discourse particles. The concrete
functions of discourse particle uses arise in interaction with the sentence types in
which they occur. These sentence types have to be considered as functional, as much
as syntactic, categories.
• Li-Chiung Yang attempts to integrate prosody in a model of the polyfunctionality of

Mandarin Chinese discourse markers. In her model, the four functions of discourse
18 K. Fischer
markers she assumes (signaling phrase relationship, interactive relationship, as well as

cognitive and emotional relationship) are signaled by discourse context, phrase
position, lexical meaning, and prosody (including the prosodic context) to differing
degrees. Prosody, particularly duration and intonation contour, is taken to play an
important part in the disambiguation of the functions of discourse markers. The author
shows how different functions of particular discourse markers are correlated with
particular contours and durational features across speakers and in spite of the lexical
tone assigned to each marker, and most importantly, also across different markers.
• Corinne Rossari approaches the problem of the polyfunctionality of a subset of

discourse markers: she takes connectives to be such items that propose restrictions for
the right as well as the left context. The functional spectrum of (monosemous)
connectives arises in her model due to different types of discourse configurations
because of which the same operation of a particular connective gives rise to different
semantic values. Thus, different discourse configurations are responsible for the
polyfunctionality of the class. Discussing the French examples alors and après tout,
she illustrates how connectives fulfil their main functions regarding the processing of
information states.
Another set of approaches attributes the polyfunctionality of (monosemous) discourse

particles to their reference to particular discourse domains. Among these are Schiffrin,
Redeker, and Frank-Job:
• Deborah Schiffrin’s model of the polyfunctionality of discourse markers includes two

scenarios: on the one hand, there may be polyfunctionality on the lexical level, on the
other, discourse markers in toto perform multiple functions. Individual discourse
markers, such as and, for which she provides an exemplary analysis, may select from a
range of possible meanings depending on the domain that serves as a point of
reference. While and in her view has only a single meaning and a single basic
function, how to continue a cumulative set, what constitutes this set, and what provides
the textual anchor for and may vary greatly. In her model, discourse markers are thus
characterized as indexicals referring to different domains of discourse.
• Gisela Redeker develops a model of discourse coherence with three components:

ideational, rhetorical, and sequential structures, regarding which discourse markers
simultaneously mark semantic, rhetorical, and sequential relations respectively. The
polyfunctionality of discourse markers results from their functioning in the three
domains. Furthermore, Redeker demonstrates in the analysis of transcripts of Dutch
and English discourse markers, as well as in a psycholinguistic experiment, that the
markers of transitions in discourse function as attentional cues for the listener/reader.
• Barbara Frank-Job argues that for an account of the polyfunctionality of (Italian)

discourse markers a combination of a synchronic and a diachronic perspective is
necessary. In a pragmaticalization process, originally deictic elements, indicating
persons, times, or locations, may come to be used to fulfil metalinguistic functions.
The synchronic polyfunctionality of these items results from their reference to three
different levels of conversational structure: the turn-taking system, macrostructure, and
superstructure. Regarding these three levels discourse markers fulfil their functions.
Introduction 19
Nemo takes a construction-based perspective. That is, a complex set of general form-
meaning pairs, constructions that directly interact with the meaning of the respective
particle morpheme, explain how the individual interpretations are created:
• François Nemo suggests an approach to the polyfunctionality (and polycategoriality)

of discourse particles, which he exemplifies on English but. His model rests on
morphemic meanings that interact with constructions, general form-meaning pairs,
advocating a strict distinction between the morpheme’s encoded meanings and the
connective, functional, or categorial interpretations it may receive when it is inserted
in specific (connective or nonconnective) constructions. He also argues for a
distinction between an utterance-type level and a contributional level in order to
account for two subclasses of discourse particles, a large and very diverse class of
utterances modifiers (illocutionary adverbs, evidentials, illocutionary particles, etc.) on
the one hand and a class of contribution modifiers (discourse connectives, etc.) on the
other.
Diewald and Fischer then combine invariant meaning components, constructional, i.e.,
grammatical, information with reference to particular domains of discourse:
• Gabriele Diewald argues that the pragmatic functions of particles are really genuine
grammatical functions, indispensable for the organisation and structuring of spoken
discourse. She holds that particles can be identified by their grammatical function,
which is basically indexical. The apparent homonymy between the different word
classes she explains by different elements indexed by the particle. At the same time,
she argues for a monosemantic view of discourse particles. The abstract semantic
content of a particle accounts for the relatedness of the different senses of a particle
morpheme, while the reference to different domains, the application of the same
semantic template to different reference elements in the sense of Sweetser (1990),
explains the differences between the readings. She supports her model with current
theories on the diachronic development of particles.
• Kerstin Fischer proposes a model of the polyfunctionality of discourse particles that

rests upon the interaction of three components: the invariant meaning of the respective
particle morpheme, constructions modelling the general structural contexts in which
discourse particles may occur, and communicative background frames, models of the
communicative situation at hand. She develops her model by discussing the functions
of the English discourse particle okay in human-to-human and in human-computer
dialogues. The comparison of the two different corpora allows her to identify the
background frames as speaker models of the respective communicative situations.
Finally, Bazzanella combines reference to discourse planes with parameterization:
• Carla Bazzanella distinguishes three macrofunctions for discourse markers, which she
exemplifies mainly on spoken Italian: interactional, metatextual, and cognitive
functions. The choice of the intended reading that speaker and interlocutor/reader
make in a given text is, according to her model, activated on the basis of the co-
occurrence of a number of contextually and cotextually relevant variables. These
variables provide a parameterization of the meaning retrieved. Semantic
20 K. Fischer
correspondence provides thus only one clue to the interpretation of a discourse marker
in use.
To sum up, with the exception of the homonymy approach, which is not followed by any
of the authors, the volume instantiates all of the approaches to the polyfunctionality of
discourse particles outlined in section 4.1. The collection of approaches combined in this
volume thus presents an exciting overview of a broad and varied spectrum of possible
approaches to the phenomenon of the polyfunctionality of discourse particles, and it
presents first steps into furthering our understanding of the spectrum of approaches to
discourse particles in general.
Notes
1
I would like to thank the authors of this volume for their enthusiasm, for their
willingness to follow the structure I have proposed and to share their thoughts in the
discussion, and for the many different ways in which they have turned the editing of
this volume into an inspiring and exciting project.
2
Throughout the volume, both terms, discourse particle and discourse marker, as well as
some other terms, are used. In this introduction, I am mainly using the term discourse
particle for reasons outlined in my own chapter in this volume.
3
Transcription conventions: = pause; = breathing; ? = rising intonation; , = level
intonation; . = terminal intonation. For details on the corpus, see my paper in this
volume.
4
The papers will be presented in the order outlined here, thus reflecting the different
positions taken regarding the polyfunctionality of discourse particles/markers.
Chapter 2
A dynamic polysemy approach to the lexical

semantics of discourse markers (with an
exemplary analysis of French toujours)
1. Introduction
1.1. Approach
The present approach to discourse markers is concerned principally with the lexical
semantics—or coded meaning—of these items and, secondarily, with how such abstract
coded meaning may interact with concrete discourse contexts to produce situated
interpretations of utterances. In other words, the most fundamental guiding hypothesis of
the approach is that any item capable of functioning as a discourse marker will be
endowed with inherent, specifiable meaning, which restricts the possible interpretations of
utterances in which that item appears.1
As a means of getting at this coded meaning and putting it into relief, a number of
problem areas may have to be dealt with, and the approach is thus an interdisciplinary
one, combining insights from a number of linguistic subdisciplines. It is, however, firmly
situated within a broad cognitive-functional framework.
1.2. Methodology
On the most basic level, my chosen methodology can be described as semasiological, i.e.,
as taking its point of departure in specific linguistic forms and investigating the range of
functions these forms may fulfil. An onomasiological approach, in contrast, would start
from a predefined set of discourse functions and attempt to determine how these functions
might be expressed linguistically.
Moreover, the analyses carried out are primarily qualitative, not quantitative. As one of
the principal aims of the approach is to provide descriptions of the coded content of
individual markers which should ideally account for all their various contextual uses, the
22 M.-B. Mosegaard Hansen
frequency of one specific use as opposed to another, or the distribution of the various uses
across speaker categories—while of course in no way irrelevant—are of lesser
importance. In principle, any attested use of a marker is of equal semantic interest,
whether it accounts for 90% of the available data or occurs only once or twice in a vast
corpus. This is not, of course, to deny that distributional frequencies, as well as data of a
situational and/or sociolinguistic nature, may provide valuable clues to the appropriate
description of the meanings of specific markers (see for instance Fischer, 2000a: ch. 3;
this volume), particularly if such descriptions are assumed to be endowed with an internal
structure, i.e., if markers are assumed to be polysemous in the sense to be outlined below.
Finally, the methodology is essentially inductive and interpretive, i.e., hermeneutic.

Among other things, this means that theory and description are developed in tandem, with
a constant interplay between the two levels.2 The approach is therefore continuously
evolving in the light of new data.
Heuristically, my analyses of the meaning of different markers rely to a large degree on

recurrent patterns at various levels:
• At the most global level, the nature of the speech event (where the term “speech event”
is intended to include written discourse), including its goals and external
circumstances, will often support certain interpretive hypotheses over others.
• At a more local level, the sequential environment in which an utterance hosting a
discourse marker occurs is considered to be of the utmost importance. Importantly, this
sequential environment will frequently comprise more than just the immediately
adjacent utterances. At this local level, metadiscursive (elements of) utterances may
also provide strong clues to the meaning and function of a given marker.
• At the microlevel, finally, linguistic and paralinguistic clues internal to the host
utterance, such as syntactic structure and information structure, co-occurrence of more
than one marker, prosody, etc., will suggest appropriate interpretations of the markers
under analysis.
The use of actual corpus data may be more or less essential depending on the type of
marker under investigation (see below). However, even if mainly intuitive data are used as
evidence, the three levels mentioned will, in my view, remain relevant in as much as there
is probably no such thing as a completely neutral context; i.e., in order to determine the
acceptability of the occurrence of a given particle or make a choice between different
possible interpretations, one will in a great many cases have to specify at least certain
aspects of the hypothetical global and local co- and context to which the host utterance is
assumed to contribute.
1.3. Data
Although the model should in principle be applicable to other languages as well, my

object language is primarily modern Standard French, as spoken and written in France by
members of what one might call “mainstream” culture.3 While dialectal and sociolectal
data may certainly provide insights and supporting arguments for a particular analysis,
they cannot be considered decisive. Attested examples are therefore acceptable only if
produced by native speakers of the above-mentioned variety, and it goes without saying
A dynamic polysemy approach to DMs 23
that examples constructed by nonnative speakers such as myself must be checked against
the intuitions of native speakers.
In earlier work (e.g., Hansen, 1998a), I have worked almost exclusively with
spoken—principally interactional—data, but more recently, written and constructed data
have been included. The inclusion of intuitive data is due mainly to a change in the nature
of the items that I am interested in, and I would maintain that some markers, in particular
semantically nontransparent ones like ben (see Hansen, 1998a: ch. 10) or quoi (Beeching,
2002: ch. 8), which are found mainly in informal, impromptu speech, are generally the
object of only very weak intuitions on the part of native speakers, and typically, it is
exceedingly difficult to come up with contexts in which these items would be clearly
unacceptable. Hence, a corpus-based analysis may be indispensible in a number of cases.
However, when dealing with markers that are more semantically “tangible”, constructed
examples (including apparently unacceptable ones) are highly useful in focusing attention
on specific aspects of meaning which may all too easily be overlooked or attributed to
other elements, when richer, authentic contexts are considered.
Although it has frequently been claimed that discourse markers are especially
characteristic of informal spoken language, items that qualify as markers on my definition
(see section 2 below) are in fact found both in written texts and in more formal spoken
discourse.4 Consequently, the analyst should ideally be able to account for particle
semantics independently of the medium or context of realization. I take it that language
users are not operating with separate grammars and lexica for spoken and written
language but that there is a continuum between what has been called the “closeness” and
the “distance” mode of language use (see Koch and Oesterreicher, 1990: ch. 2; Hansen,
1998a: ch. 5) and that speakers/writers have at their disposal a range of linguistic
strategies, some of which may be preferentially employed towards one or the other pole of
this continuum, but none of which are by definition exclusive to a particular mode.
Linguistic descriptions which as a matter of principle are only applicable to one particular
type of language use are therefore unlikely to reflect the actual competence of the
language users.
Hence, if–for practical reasons–only corpus data of a specific type are used in a particular
analysis, this should be made clear and the scope of the conclusions be restricted
accordingly.
1.4. Problem Statement
1.4.1. State of the Art

Given the notorious multifunctionality of discourse markers, a central issue for those
interested in the semantic description of these items has always been, and continues to be,
the question of how to account for this variety of functions, the traditional choice being
between homonymy and monosemy, to which the notion of polysemy has more recently
been added. The picture is further complicated by the fact that these terms are not used in
exactly the same way by different researchers, and I will therefore start by defining what I
take them to mean.
First, on what I call the “homonymy view”, it is assumed that the nuances of meaning
attributable to the presence of a particular linguistic item in a given context are in
principle a matter of the semantics of that item. Hence, if a given form has a number of
seemingly different uses, then these various uses are taken to represent separate lexical
items, any connexions between them being assumed to be essentially arbitrary. I find such
an approach unsatisfying. For one thing, it seems particularly prone to conflate the coded
meaning of a given marker with the situated interpretations of the utterances in which that
marker appears. Secondly, it is inherently unable to explain the frequently quite robust
intuition (often supported by diachronic data) that the so-called homonyms are
nevertheless somehow semantically related.
What I call the “monosemy approach” aims, on the contrary, to simplify semantic
descriptions as much as possible, leaving the burden of interpretation to pragmatics. In
practice, this means that the descriptive goal is to circumscribe an invariable “core
meaning” compatible with all the possible contextual uses of a given item. Theoretically,
this approach is in many ways more appealing than the notion of homonymy.
Descriptively, however, it is not without problems. Firstly, because the descriptions
offered may, depending on the multiplicity of concrete uses of the marker in question, end
up being so abstract and general that they neither exclude nonexistent uses nor distinguish
adequately between different markers. Secondly, postulating monosemy leaves the
researcher at a loss to explain how the range of uses of a given item can vary
systematically, both diachronically and in language acquisition.
My preferred strategy is therefore a polysemy approach. The guiding assumption here is

that items which in at least some contexts fulfil a discourse marking function can have
more than one meaning on the semantic level but that these meanings may be related in a
motivated—if not necessarily fully predictable—way, such that we may describe as many
as possible of the functionally distinct examples of a given homophone/homograph as
instantiations of a single, polysemous lexical item.
1.4.2. Problems
An important problem for any approach to the meaning and functions of discourse
markers is, of course, how to constrain the range of possible distinct functions so that it
does not get out of hand; given that no two concrete contexts of use are entirely identical,
it would in principle be possible to claim that any use of a given item was functionally
distinct from any other use. Presumably, most people would shy away from such a claim,
but where then to draw the line?
I do not believe it possible to draw that line in a totally objective way. Instead, I advocate
observing a principle also adhered to by several other scholars in the field and which
Foolen (1993: 64) has called “methodological minimalism”. Very simply, this principle
enjoins the semanticist not to “multiply meanings beyond necessity”, as it were.5 In
practice, this means that if—faced with a given use of a given marker—one has the choice
between adding a meaning node to an already existing network or explaining the use in
question as a systematic “side-effect” of the occurrence of another meaning in a specific
type of context, one should probably choose the latter option.6
Another problem, which is of interest in itself but also more specifically to the polysemy
approach, is that of diachrony. Succinctly put, we need more in-depth studies of the
diachronic evolution of discourse markers, both on how they evolve syntactically and
semantically from other function classes and on how meaning extensions take place
within the domain of discourse marking.7
We also, in my view, need to emphasize what you might call “diachronic responsibility”
in synchronic description. This does not mean that one must necessarily carry out an in-
depth diachronic study in order to propose a synchronic description of a given marker, but
it does mean that synchronic descriptions, at least where polysemy networks are
postulated, should aim for compatibility with what is already known about the diachronic
evolution of the particles under study. Thus, for instance, we should as a rule avoid
postulating meaning extensions whose directionality is the opposite of what is attested in
the available historical sources, and diachronic evidence can therefore potentially falsify
synchronic descriptions. This having been said, we do of course need to keep in mind that
certain markers are primarily found in the spoken language and that some of their uses
may therefore have existed in that mode long before the time they were first attested in
writing.
2. Definition
Like many others working in this area, I define discourse markers in primarily functional-
pragmatic, rather than formal-syntactic, terms. According to my definition, the role played
by linguistic items functioning as discourse markers is nonpropositional and meta-
discursive, and their functional scope is in general quite variable. The role of markers is,
in my view, to provide instructions to the hearer on how to integrate their host utterance
into a developing mental model of the discourse in such a way as to make that utterance
appear optimally coherent. This means that markers have connectivity (in a wide sense) as
at least a part of their meaning.
Importantly, however, connectivity is not limited to relations between neighboring

utterances or utterance parts, and the notion of a “developing mental model of the
discourse” used in the above definition is meant to reflect that. It must be kept in mind
that discourse is not constituted by language only—the context (situational and cognitive)
is an essential part of it, and the connective role of discourse markers may therefore
pertain to relations between the host utterance and its context in this wider, nonlinguistic
sense.
It is for this reason that the very first utterance produced in a given situation may in fact
be introduced by a marker; thus, for instance, Blakemore’s (1987: 106) example, where so
indicates that the contents of the host utterance should be understood as cohering with an
element of the developing mental model derived from a salient aspect of the nonlinguistic
context:
(1) So, you’ve spent all your money! [As said to a person who has just entered he
room loaded with parcels]
Even when they are not discourse initial, some markers may in fact signal to the hearer
that their host utterance should precisely not be connected to the preceding cotext, but that
its relevance is rather to some aspect of the larger situational context:
(2) [Two linguists sitting on a park bench discussing Peircean semiotics]

A. A mon avis, on peut concevoir le “ground” d’un signe linguistique en tant que
tel comme constitué par le système linguistique dans lequel le signe en question
s’insère. —Tiens, il pleut !
A. In my opinion, you can conceive of the “ground” of a linguistic sign as such as
constituted by the linguistic system of which that sign forms a part. —Hey, it’s
raining!8
Thus, like Roulet (this volume) and Pons Bordería (this volume), I follow Berrendonner
(1983) in maintaining that discourse markers actually never mark a direct connection
between their host utterance and the linguistic cotext,9 but always a connection between
the utterance and the mental discourse model under construction, where the latter will of
course contain information gleaned from, among other things, previous utterances, but
also (as stated above) information from the nonlinguistic context, as well as contextually
relevant encyclopedic knowledge.
It is, further, important to note that markers do not, on my view, merely guide
interpretation with respect to an already given context—indeed, as probably first noted by
O. Ducrot and collaborators a couple of decades ago, they may actively help to construct
that context (Ducrot et al., 1980; see also Nyan, this volume). Thus, the speaker of (3)
may well be understood as (conversationally) implicating that Elizabeth might not remain
submissive, even if the (conventionally) implied conflict between wifely submission and
extensive book reading had never before occurred to the hearer:
(3) Elizabeth has always been a very submissive wife, but she reads a lot of books
It is often said that a defining property of discourse markers is their optionality, i.e., it
should be possible to remove a marker without fundamentally changing the meaning of its
host utterance. In other words, markers are conceived of as fundamentally redundant, as
sign-posts to (virtual) meanings which could equally well be derived from other aspects of
the co- or context. In many cases, this does appear to be correct. But as pointed out by
Rossari (2000: 32), it is not invariably the case: some markers can never be deleted
without radically altering the range of possible interpretations of the discourse:
(4) Max a oublié de se rendre à la réunion. De toute façon, le comité a décidé

d’ajourner cette réunion.
Max forgot to go to the meeting. In any case, the committee decided to adjourn
the meeting.
In (4), the marker de toute façon indicates that an otherwise possible causal relationship
between the two propositions should explicitly not be inferred. As Rossari (2000) points
out, such a reading does not appear possible if the marker is removed.
2.1. A note on terminology
In Hansen (1998a), I used the terms discourse particle and discourse marker
interchangeably. Currently, however, I think the latter is preferable, at least if the term
particle is taken at face value.
The reason is that not all items which are capable of assuming a discourse-marking
function actually fit the traditional description of particles as monomorphemic, non-
inflectable items, and the label discourse particle is therefore misleading because of its
formal component. The term discourse marker, on the other hand, primarily denotes a
function and is therefore unproblematic.
Moreover, I do not conceive of discourse markers as constituting a part of speech, for it

seems that very few linguistic items are exclusively devoted to this function. Rather, a
great many, often formally quite different, linguistic items may have one or more
discourse-marking uses alongside one or more non-discourse-marking uses. In other
words, an item like déjà is formally an adverb in both (5) and (6), but it functions as an
aspectual adverbial in the former and as a discourse marker in the latter. Similarly, dites is
formally a verb in both (7) and (8),10 but only in the latter does it function non-
propositionally as a discourse marker:
(5) La réunion ne commencera que dans une heure, mais Benjamin est déjà arrivé.
The meeting won’t start until an hour from now, but Benjamin has already
arrived.
(6) Je préfère ce restaurant à celui où on était l’autre jour : déjà, la cuisine chinoise
me plaît mieux que la cuisine maghrébine, et puis, l’atmosphère est plus relax ici.
I prefer this restaurant to the one we went to the other day: for one thing, I like
Chinese cooking better than North African, and also, the atmosphere is more
relaxed here.
(7) Si vous désirez autre chose, dites-le-nous !

If you want anything else, let us know!
(8) Dites donc, on est pressé! [As said to someone to who has just jumped ahead of
you in a line]
Why, we are in a hurry, aren’t we!
The advantages of defining discourse marking as simply a functional potential of formally

disparate items are both synchronic and diachronic. On the strictly synchronic level, it
allows us to explain those cases where it is not quite obvious whether a given item does
indeed have discourse marking as (part of) its function in a given context or whether it
instead fulfils some other, more “traditional” function. Thus, in (9), it is not clear whether
alors functions on the propositional level as a temporal anaphor, or rather non-
propositionally marks the second sentence as a conclusion. What is more, it is not even
clear that hearers necessarily have to make a choice between the two (mutually
compatible) interpretations in order to gain a satisfactory understanding of the utterance:
(9) Jean a tiré. Alors, Pierre s’est écroulé.

Jean fired. Then/So, Pierre fell down.
Similarly, in (10), for instance, we cannot tell from the exchange itself whether B’s
utterance is meant to remind A of something he already knows or rather intended to
impart entirely new information. In the latter case, tu sais clearly has a purely discourse
marking function, inviting the hearer to infer the implicit connection between the two
propositions expressed in the exchange (see Dostie and de Sève, 1999). In the former
case, this connective function can plausibly be described as secondary, i.e., as simply a
relevance implicature (Grice, 1975) of the truth-conditional meaning of the utterance:
(10) A. Tu as l’air en forme.

B. Je fais beaucoup de sport, tu sais.
A. You seem to be in good shape.
B. I exercise a lot, you know.
Whatever the analysis of (10), it seems to me that the intuitively most plausible way to
explain the fact that tu sais (and its equivalents in a number of other languages) can fulfil
the particular discourse marking function described is to assume that this use originates
precisely in an implicature of the truth-conditional use of the expression. The same holds,
of course, for (9) above, where the nonpropositional interpretation of alors will, in many
cases, be a natural implicature of the propositional one. Seeing discourse marking as a
functional potential of a wide variety of linguistic items rather than as the defining
characteristic of a specific category of items is thus compatible with the frequently
gradual nature of diachronic evolution.
This means, of course, that the term discourse marker is not a cohyponym of, for instance,
interjection, conjunction, modal particle, focus particle, or sentence adverbial. I consider
these latter terms to be names for specifiable syntactic categories which may or may not
exist in a given language,11,12 whereas discourse marker names a function which may be
fulfilled by items from several of these categories.
Rather, discourse marker should be considered a hyponym of pragmatic marker, the latter
being a cover term for all those nonpropositional functions which linguistic items may
fulfil in discourse (Brinton, 1996: ch. 2; Fraser, 1996; Foolen, 2001: 350). Alongside
discourse markers, whose main purpose is the maintenance of what I have called
“transactional coherence” (Hansen, 1998a: 180ff.), this overarching category of functions
would include various forms of interactional markers, such as markers of politeness, turn-
taking, etc. whose aim is the maintenance of interactional coherence (Hansen,
1998a:180ff.); performance markers, such as hesitation markers; and possibly others.
3. Functional Spectrum
Following the stance outlined above, I take it, firstly, that most discourse markers
instantiate particular non-truth-conditional senses of polysemous lexical items and that the
senses compatible with the discourse-marking function are typically derived from other,
diachronically prior and typically truth-conditional, meanings. Secondly, I take it that
polysemous items in general may be internally structured in more than one way. Thus, we
could, depending on the specific item under investigation, be dealing with either a
meaning chain (Heine et al., 1991: 228-229), a radial category (Lakoff, 1987: 65), or a
network of variously interconnected nodes.
The exact mechanism of extension from one meaning to another may be metaphorical, but
in the case of discourse markers, metonymy (in the sense of the conventionalization of
highly frequent implicatures) seems likely to play at least as important a role. More
specifically, the existence of a number of seemingly unidirectional tendencies of semantic
change identified by Traugott and Dasher (2002: 281) appears to be solidly supported by
empirical evidence and to constitute a plausible foundation for what that Traugott (1989:
31) has called “internal semantic reconstruction”. The tendencies that are most relevant
for present purposes are (i) the tendency for meanings to become increasingly subjective;
(ii) the tendencies for conceptual or truth-conditional meanings to become, respectively,
increasingly procedural or non-truth-conditional; (iii) the tendency for meanings with
subpropositional scope to progressively enlarge their scope, possibly even to the discourse
level; and (iv) the tendency for meanings that originally make reference to the described
event to come to make reference to the speech event itself.
While the different nodes representing the distinct meanings need not exhibit the exact
same syntactic properties, polysemy in the strict sense ends when the item changes its
basic part-of-speech affiliation, in which case we are faced, rather, with a case of
“heterosemy” (Lichtenberk, 1991). Nevertheless, parts of speech are probably not
Aristotelian categories but may shade into one another (Hopper and Thompson, 1984), so
the exact moment at which polysemy turns into heterosemy may be difficult to determine
in concrete cases.
The polysemy approach accounts for the intuition that the different functions of a given
item are semantically related, while allowing for the fact that new nodes may be created,
while others may disappear.13 At the same time, it allows us to explain how new uses of
an item may gradually emerge, first as pragmatic implicatures or “side-effects” of existing
meanings and only later as fully conventionalized, distinct semantic meanings.
Importantly, if a given linguistic item is capable of functioning as a discourse marker in

some of its uses but does not do so in others, we may, in my view, still speak of polysemy
as long as the item continues to belong to the same part of speech in all the uses
described. Thus, I will argue in section 4 below that French toujours is polysemous (as
opposed to either homonymous or heterosemous) because no matter what its specific
function in a given utterance is, it is always basically adverbial in nature, and its different
meanings can be related in a motivated way.
An crucial question is, of course, how hearers go about deciding on a particular

interpretation of a polysemous (or heterosemous) item. It seems probable that they do so
by integrating information from several levels of discourse, using much the same type of
heuristics, of both a bottom-up and a top-down nature, as was outlined in section 1.2
above.
For instance, microlevel syntactic and prosodic information may be used in deciding that
bon is an adjective in (11) but an adverbial functioning as a discourse marker in (12),
since in the former it appears in a syntactically and—in the spoken language—
prosodically integrated premodifying position of an NP, whereas in the latter, it is
syntactically peripheral to, and prosodically detached from, the host clause:
(11) C’était un très bon film, ça

That was a very nice movie.
(12) A. . . . si par exemple ta mère m’avait au bout du fil

B. oui
A. bon, j’appelle, c’est elle qui décroche . . . (CT: 7)
A. . . . if for instance your mother had me at the end of the line
B. yes
A. well, I’m calling, and she answers . . .
Local-level semantic and pragmatic information may privilege the interpretation of tiens
as a main verb in (13), but of the same item as a discourse marker in (14), given that one
can physically hand an article, but not an e-mail message appearing on a computer screen,
to one’s interlocutor:
(13) Ah, voilà l’article que tu m’avais demandé. Tiens !

Ah, here’s the article that you asked me for. Here you are! (lit.: hold!)
(14) A. Qu’est-ce que tu fais ?

B. Je regarde mon courriel. Tiens, un message de Ségolène !
A. What are you doing?
B. I’m checking my e-mail. Hey, a message from Ségolène!
Finally, global-level discourse information about the speech situation, including among
other things, the discourse roles and the social relationship between speaker and hearer,
may ultimately be responsible for the interpretation of c’est-à-dire in (15) as primarily
indicating either a clarification of A’s own previous utterance or a hedged correction of
B’s assumptions (Beeching, 2002: ch. 5):
(15) A. Il ne faut pas oublier que l’ESB est une maladie bovine.
B. Alors, quelles mesures dois-je prendre pour protéger mon perroquet ?
A. C’est-à-dire que ça touche essentiellement les vaches.
A. We mustn’t forget that BSE is a bovine disease.
B. So, what steps should I take to protect my parrot?
A. That is, it’s mainly cows that suffer from it.
4. The approach exemplified
In this section, I will show how the approach outlined above works in practice by way of a
specific example, namely the different uses of the French adverb toujours. For reasons of
space, the analysis presented here will be sketchy. For further details, I refer the reader to
Hansen (to appear).
Syntactically, toujours is adverbial in all its uses, but among these we find both
propositional and nonpropositional ones. Only a subset of the latter qualify as discourse
marking uses according to the present approach. At the propositional end of the spectrum,
we have three different truth-conditional uses of this adverb, a temporal one in which
toujours functions essentially as a (quasi-)universal quantifier over instances and where it
might be described as “globally affirmative”, an iterative one, and a phasal or continuative
one, illustrated in (16) through (18):
(16) Ghislaine a un tempérament de chien, mais Philippe est toujours content.

Ghislaine has a lousy temper, but Philippe is always happy.
(17) Philippe préfère les BD, mais Ghislaine lit toujours les nécrologies en premier.
Philippe prefers the comics, but Ghislaine always reads the obituaries first.
(18) Luc est toujours en train de préparer sa thèse, alors que Corinne a terminé la
sienne il y a longtemps.
Luc is still working on his thesis, whereas Corinne finished hers a long time ago.
These three uses are not always clearly distinct if the host clause is considered in
isolation. In a number of cases—including the ones above—whether one or the other
sense is intended can only be determined by examining the co- and context.
Fairly closely related to these is the nonpropositional use illustrated in (19), where
toujours is still fully integrated into the syntax of its host clause:
(19) Un pingouin, c’est toujours un oiseau.

Whatever else, a penguin is a bird.
This use of the adverb is, however, not completely identical to the truth-conditional uses
from the syntactic point of view: whereas truth-conditional toujours can be both negated
and used in isolation (Cadiot et al., 1985a), this is not the case with the use exemplified in
(19). So, although (21) and (23) are grammatical (although pragmatically odd in most
contexts), toujours in these examples can only be understood as propositional:14
(20) Philippe n’est pas toujours content.

Philippe is not always happy.
(21) #Un pingouin n’est pas toujours un oiseau

A penguin is not always a bird.
(22) A. Est-ce que Ghislaine lit les nécrologies d’abord? —B. Toujours!
A. Does Ghislaine read the obituaries first? —B. Always!
(23) A. Est-ce qu’un pingouin est un oiseau? —B. #Toujours!

A. Is a penguin a bird? —B. Always!
In this particular nonpropositional use, I would prefer to classify toujours as modal,

despite the fact that French has no commonly recognized category of modal particles,
such as exist in the continental Germanic languages. Toujours, however, is not the only
French adverb with functions which can usefully be described as akin to those of modal
particles (see Hansen, 1998b, 2002b; Waltereit, 2001; also Weydt, 1969). In other words,
following the terminological proposal of section 2.1 above, toujours, in examples such as
(19), functions as a type of pragmatic marker, but not as a discourse marker.
Finally, we have two discourse marking uses proper, in which toujours appears with a
connective function. In the first of these, toujours is part of the frozen collocation toujours
est-il que + clause.15 While toujours est-il is formally a matrix clause taking que + clause
as its complement, its invariability together with the fact that the following clause always
constitutes the truth-conditional core of the utterance, suggests that toujours est-il que
should rather be considered an unanalyzable idiomatic connective.16 Hence, toujours in
this use is syntactically peripheral to its host clause:
(24) A. Cet appartement est petit et il est cher. Je n’ai pas envie de le prendre.
B. Comme tu veux. Toujours est-il qu’il est super-bien situé.
A. This apartment is small and it’s expensive. I don’t want it.
B. Whatever you say. Still, the location is great.
In its second discourse marking use, toujours is prosodically or graphically detached at

the end of its host clause:
(25) A. Cet appartement est petit et il est cher. Je n’ai pas envie de le prendre.
B. Il est super-bien situé, toujours.
A. This apartment is small and it’s expensive. I don’t want it.
B. The location is great, though.
4.1. Propositional uses of toujours
Etymologically, toujours is entirely transparent: it is a coalesced form of the universally

quantified NP tous jours (‘all days’). We may therefore assume that the “globally
affirmative” use is the diachronically prior one, from which the others have evolved. Such
a diachronically prior sense is not necessarily also the synchronically most salient one. In
this particular case, however, it is likely to be not only highly salient, but also central in
the sense of being the conceptual origin of several of the other possible senses.
In this use, toujours signals that the state of affairs (SoA) denoted by its host clause
extends indefinitely over time. The iterative use of toujours can be straightforwardly
related to the globally affirmative use: here, the SoA is not assumed to hold at any
possible time t, but the speaker is claiming that in a specific type of context (in the case of
(17) above, that of Ghislaine’s reading the newspaper), a repetition of the relevant SoA
invariably occurs. In other words, the (quasi-)universal quantification over instances is
restricted to particular frame of reference. Given that it seems necessary that this
restriction be specified by either co- or context, there is probably no need to see the
iterative reading as separate from the globally affirmative one, and we might simply want
to pose it as a more specific, pragmatically determined reading of the latter (see also
Muller, 1999).
In the phasal use exemplified in (18), which appears to be a relatively recent development
(Trésor de la langue française 16: 381), toujours asserts the truth of the SoA p at the
moment of reference, and it weakly presupposes the truth of p during some period
preceding, and continuing up to, the reference time. We may assume that this phasal
meaning of toujours originated as a conversational implicature of the temporal meaning,
given that situations compatible with the latter meaning will normally also be compatible
with the former, but not vice versa.
Due (I would claim) to the persistence (Hopper, 1991: 22) of the element of global
affirmation which characterizes the temporal use of the adverb, phasal toujours allows for
the indefinite extension of the current SoA p. Hence, it contains no notion of dynamism,
but is essentially a marker of stasis.
4.2. Nonpropositional uses of toujours
4.2.1. “Modal” use

This lack of dynamism, and consequent lack of direction of evolution, explains certain
properties of the nonpropositional uses of toujours, which seem to be first attested around
the same time as the phasal use. Thus, the “modal” use in (19) above evokes the idea of
the speaker performing an undirected mental scan of indefinite duration of the category of
birds and concluding that, whichever way one looks at penguins, the fact that they belong
to this category remains.17
Thus, the modal meaning of the adverb appears to instantiate Traugott and Dasher’s
(2002) tendency towards increasing subjectification of meanings (see section 3 above).
That is, whereas the propositional sense of the item concerns the “external described
situation”, the modal sense clearly concerns the “internal (evaluative/perceptual/cognitive)
situation” (Traugott, 1989: 34-35).
When used in argumentational contexts, modal toujours marks utterances that are seen as
rhetorically quite weak, in as much as toujours-marked utterances like that in (26) point to
no particular conclusion, positive or negative, beyond themselves:
(26) A. Tu sais, les 10.000 euros que ma tante m’avait légués? Eh bien, il s’avère que
ce sont des francs—belges!
B. Ah zut ! Enfin, c’est toujours de l’argent . . .
A. You know, the 10,000 euros my aunt left me? Well, it turns out they’re francs
—Belgian francs.
B. Aw, man! Anyway, it’s money, I guess . . .
For this reason, toujours is odd in combination with strongly value-laden expressions,
such as énorme in (27):
(27) Solange n’aura peut-être pas le prix Goncourt, mais elle a quand même publié un
roman chez Gallimard. ?C’est toujours énorme!
Solange may not get the Goncourt Prize, but she did get a novel published by
Gallimard. If nothing else, that’s a huge thing!
This essential neutrality explains the otherwise puzzling fact that utterances containing
“modal” toujours can be made to work argumentatively in either direction depending on
the context; witness the following examples (Franckel, 1989: 303):
(28) Tu peux toujours lui téléphoner. Cela ne fera pas de mal.

You can always give him a call. That won’t hurt.
(29) Tu peux toujours courir. Cela ne donnera rien.

You can make as much effort as you like. It won’t make any difference.
In (28), the hearer is (weakly) encouraged to make the phone call, whereas in (29), the
speaker is attempting to dissuade him from wasting his efforts. Because of its static
nature, the SoA marked by toujours can at best act as a weak argument for a positive
conclusion: something which is always and invariably the case can normally make only
little (if any) difference to the outcome of a given situation.
4.2.2. Discourse marking uses

As for the two discourse-marking uses, they likewise exemplify Traugott and Dasher’s
(2002) tendencies of semantic change: like the modal meaning, they are basically
subjective, and they concern the speech event itself, as opposed to the described event
(recall section 3 above).
The locution toujours est-il que always explicitly connects its host utterance to
information gleaned from prior discourse, whether that discourse is monologic or
dialogic. The meaning of this expression can be straightforwardly derived from the
“globally affirmative” truth-conditional use described in sect. 4.1 above, for by using this
connective, the speaker appears to be indicating that she explicitly refrains from taking a
stand on the prior discourse and instead chooses to point out what she knows to be true at
any time, irrespective of what has been said before. This is supported by the fact that the
utterance preceding the one introduced by toujours est-il que very frequently contains
hedges such as je ne sais pas, peut-être, etc., or is interrogative or hypothetical in form:
(30) Nul ne savait d’où il arrivait, ni par quel hasard il s’arrêta en pays toumat.
Toujours est-il que notre roi subit son influence et décréta un jour qu’il était la
réincarnation de David, roi des Hébreux. (J. Lanzmann, La horde d’or, p. 370,
1994; from Corpus Frantext)
Nobody knew where he came from, nor what made him stop in the land of the
Toumat. In any case, our king fell under his influence and one day decreed that he
was the reincarnation of David, King of the Hebrews.
(31) Pourquoi, dans leurs longs hivers, avaient-ils choisi l’étude du français? à cause
de sa clarté, de sa transparence, bon remède à leurs nuits perpétuelles? Toujours
est-il que ces Scandinaves nous ouvraient le chemin. (E. Orsenna, Le grand
amour, p. 21, 1993; from Corpus Frantext)
Why did they choose to study the French language during their long winters?
Because of its clarity, its transparency, which made a good remedy against their
perpetual nights? Whatever the case may be, these Scandinavians paved the way
for us.
The above description of toujours est-il’s semantic content provides for two different
contextual interpretations of this marker, namely a weakly concessive interpretation, as in
(32), and discourse-structuring interpretation, as in (33), where the expression marks the
return to a prior topic following a digression:
(32) Il est possible que Jean réussira brillamment à l’examen. Toujours est-il que son
prof ne l’aime guère.
It is possible that Jean will pass his exam with flying colors. Still, his teacher does
not like him.
(33) . . . le quartier était plutôt discrédité sur le marché des locations. Période de crise
. . . on se demande d’ailleurs quelle période n'est pas de crise? Toujours est-il
qu’aller percher dans le XIIIe ça vous classait chez les loquedus. (Alphonse
Boudard, Mourir d'enfance, 1995; from Corpus Frantext)
. . . the neighborhood was more or less in disrepute on the rental market. A period
of crisis . . . incidentally, one wonders if there was ever a time when there wasn’t
a crisis. Whatever the case may be, to go live in the 13th arrondissement singled
one out as destitute.
The fact that the contents of the host utterance thus makes no difference to the status of
the previous discourse, and vice versa (Nemo, 2000), is compatible with Nguyen’s (1986:
192) observation that in dialogal contexts where the preceding discourse represents a view
endorsed by the hearer, no reaction is expected from the latter following a concession
marked by toujours est-il.
Now, in fact, the only difference between the weakly concessive interpretation of toujours
est-il and the discourse structuring interpretation appears to lie in the nature of what may
be inferred from the preceding context: if the latter seems to evoke certain expectations,
which are subsequently contradicted by the utterance hosting the marker (as in (32),
where one might have expected Jean’s teacher to at least not dislike him, given that
teachers more often than not appreciate students who may be capable of brilliant results),
then the indication that the contents of the host clause will remain in force no matter what
the status of the previous discourse will result in a concessive interpretation. If, on the
other hand, the host clause does not appear to contradict any contextual expectations, then
we get the digression-closing interpretation seen in (33).
Rather than classifying the two interpretations of toujours est-il as representing two
separate coded meanings, I would therefore prefer to regard them as systematic “side-
effects” of the contexts in which they appear. (The more so as the border between them
frequently seems to be somewhat fuzzy.)
Finally, the semantics of right-detached toujours is largely similar to that of toujours est-
il: like the latter, right-detached toujours marks utterances whose content can, according
to the speaker, be asserted independently of how one feels about the previous discourse,
and we may therefore assume that its meaning derives from that of the globally
affirmative temporal adverbial.
The difference between the two discourse marking uses of the item lies in the fact that the
host utterance of a right-detached toujours has a somewhat different argumentative weight
from what we find when toujours est-il is used. That is, right-detached toujours appears to
mark its host utterance as at least a potential counterargument to a (possibly implicit)
conclusion conveyed by the prior discourse: in the speaker’s opinion, the hearer ought at
least to give the contents of the host utterance some serious thought, and a reaction from
the hearer is therefore appropriate (Nguyen, 1988: 42).
This difference might well be attributable to the different syntactic positions of the adverb
in the two constructions. Thus, we might hypothesize that toujours est-il functions at the
level of the “tropic” (see Hare, 1971), i.e., it comments on the truth-value ascription,
whereas right-detached toujours functions at the level of the “neustic” (Hare, 1971), i.e., it
comments on the speech act itself. Moreover, right-detached toujours, due to its position,
could plausibly be attributed the status of an afterthought. In this manner, the host
utterance will appear at first as an unmodified assertion, to which the speaker then adds
the comment that its contents are in principle assertable at any time. It could be that by
focusing thus on the speech act itself, the marker underscores the current discourse
relevance of the utterance, and thereby makes it harder for the hearer to ignore.
4.3. Summary
To sum up the preceding analysis, the uses of toujours in modern French seem to form a
radial category with the temporal “globally affirmative” use at its center, from which all
the other uses can be derived.
In one case, the iterative use, we seem to be dealing with simple contextual modulation
(Cruse, 1986: 52); hence, this use does not in my view constitute a separate node of
coded meaning. With respect to the remaining uses of the adverb, I would, however,
postulate various types of actual meaning extensions: conventionalization of what is
basically a logical implication of the globally affirmative meaning in the case of the
phasal sense and various forms of subjectification in the case of the three non-
propositional senses.
5. Broader perspective
The approach to discourse markers which I have sketched in the preceding pages has the
advantage of being dynamic on several levels. Firstly, the notion of polysemy coupled
with “methodological minimalism” is synchronically dynamic in that—similarly to
monosemy—it allows for contextual modulation of the items whose meaning is analyzed.
This is an important advantage of polysemy over the homonymy approach, given that, as
has been repeatedly pointed out by researchers in conversational analysis, actual
interpretation is necessarily situated and that no two actual situational contexts are exactly
identical.
Secondly, polysemy is dynamic in that it allows for the conventionalization of new senses
of morphemes and constructions, based on frequently occurring contextual modulations of
situated occurrences. These new senses are themselves subject to contextual modulations
and subsequent conventionalization of the latter, such that the most recently created sense
of a given item may in principle be quite far removed from the meaning of its ultimate
diachronic origin.
In this, polysemy stands in opposition to monosemy, which although it allows for

contextual modulation, is nevertheless an essentially static way of viewing meaning, for
two related reasons: (i) The notion of a core meaning which is held constant between
contexts entails that all the possible contextual interpretations of the linguistic item in
question ought to be simultaneously available; (ii) consequently, should certain uses of

the item in question at some stage of either phylogeny or ontogeny give rise to one or
more previously unavailable interpretations, it must be assumed that its core meaning has
undergone a qualitative change. This means that descriptions of semantic change can only
compare successive, but essentially independent and static, synchronic stages, whereas the
dynamic diachronic process as such can have no theoretical status.
Thirdly, the idea that discourse markers are a strictly functional category, which is
orthogonal to parts-of-speech classifications, coupled with the possibility of not just
polysemy but heterosemy, means that what appears to be materially the same linguistic
item may in its various contextual uses seem to shift back and forth between both
functional categories and word classes; yet its different senses may not only strike us as
clearly related, but indeed, such synchronic variation may frequently reflect diachronic
changes.
In a broader perspective, given its essentially dynamic nature, the present approach points
to a conception of the linguistic sign different from the binary, structuralist one which is
commonly accepted. Instead, it seems to call for a triadic conception of the sign, such as is
found in the works of the American philosopher C. S. Peirce, and which crucially defines
signs as vehicles of actual communication, thus incorporating the notion of situated
interpretation into the sign function itself.18
Contrary to structuralist semiology, which postulates a sign function consisting of two

solidary parts, a signifier and a signified (Saussure, 1972/1916: 99), Peircean semiotics
operates with a pragmatic and dialogal sign relation holding between three entities, a
representamen (i.e., an expression or vehicle), an object (i.e., the thing represented), and
an interpretant (i.e., a further, equivalent sign, evoked in the mind of the comprehender
by the original sign (Peirce, 1932: 2.228)). Moreover, the sign does not represent its
object in all its aspects, but only with respect to a “ground”, i.e., a particular frame of
reference. This is illustrated in Fig. 1.
Figure 1. The Peircean sign

Representamen
Ground
Object Interpretant
Lastly, for Peirce, the sign constitutes an action prescription (Peirce, 1932: 2.330). Within
the framework of an instructional semantics (Hansen, 1998a: ch. 4), the representamen
may thus be seen as conveying the semantic instructions which the hearer must carry out
in order to grasp the meaning of the sign, while the interpretant can be understood as the
result of the hearer’s having carried out these instructions, i.e., as a mental representation
in the form of a new and more developed sign, which itself has the status of an action
prescription.
This has two consequences: (i) interpretation (or “semeiosis”) does not necessarily stop
when the first interpretant has been produced—it can in principle continue indefinitely;
(ii) given that an instruction can in principle be carried out in a multitude of ways, the
representamen does not determine a unique interpretant which is valid for all contexts—
rather, it should be seen as offering a more or less restricted range of possible
interpretations. Importantly, the correctness of the different possible interpretations can be
evaluated intersubjectively, in as much as the interpretant, being itself a sign, can in turn
be subjected to further interpretation.
Consequently, Peirce operates with three types of interpretants:
• An immediate interpretant, constituted by the range of potential interpretations of the

sign as such.
• A dynamic interpretant, which is the effect actually produced by the sign on the
recipient in a given context. That is, the dynamic interpretant represents what is
actually “decoded” by the comprehender. According to Andersen (1984: 38), this
decoded content is the result of an abductive process, and thus has the status of a
hypothesis. This means that the dynamic interpretant may be modified or even rejected
in the light of subsequent information. This brings us to the third type of interpretant,
namely,
• The final interpretant, which is the effect which would be produced by the sign in
question on any recipient whose circumstances were such that he was able to grasp the
full meaning of the sign. This final interpretant may only be reached through a process
of intersubjective negotiation.
It seems relevant to postulate the existence of different types of “grounds” corresponding

to the first two types of interpretants: at the level of the immediate interpretant, the
“ground” can thus be understood as the linguistic code (or system) as such, although it
should be noted that this code contains a pragmatic dimension, namely, those non-truth-
conditional, but nevertheless conventional, interpretive frames evoked by a great many
signs “as such”. More specifically, I am thinking of, for instance, the adversative element
inherent in a connective such as but, or the fact that a noun such as bachelor evokes a
sociocultural context which will normally exclude its felicitous use with respect to a 17-
year-old (Fillmore, 1982).
At the level of the dynamic interpretant, the “ground” represents the concrete context in
which the sign is “actualized” in the dialog between linguistic code and situated use. This
interpretation of the role of the “ground” and its relation to the semiotic triad is illustrated
in Fig. 2. What Fig. 2 shows is that the (initial) dynamic interpretant is arrived at through
a dialogic interplay between the sign and its context of appearance: on the one hand, the
sign “as such” will convey a certain image of the context, by way of the conventional
interpretive frames contained in what I call “ground 1”; and on the other hand, the
manner in which speaker and hearer conceive of the specific context in which the sign
appears, and which forms the content of “ground 2”, will influence the way in which the
sign is understood.
Figure 2. Roles and relations of ground

Representamen 1
= the sign “as such”
Ground 1
= language system
Object 1 Interpretant 1
= immediate interpretant
Representamen 2
Ground 2
= specific context
Object 2 Interpretant 2
= dynamic interpretant
This understanding of the semiotic process has implications not only for the synchronic
interpretation of utterances, but for diachronic change as well: for should a sufficiently
large number of comprehenders produce more or less similar chains of inference when
interpreting the situated uses of a given linguistic sign in a sufficiently large number of
contexts, the speaking community may well end up by establishing a new interpretive
habit which will henceforth form a part of the meaning of the sign “as such”. In other
words, thanks to the frequency of a particular kind of dynamic interpretants, level 1 of the
sign in question may be abductively modified, resulting in either polysemy or actual
semantic shift.
Clearly, the above does not represent a fully fledged theory of meaning, but it provides, in
my view, a potentially fruitful standpoint from which to approach the study of how
linguistic meaning invariably interacts with contextual factors. Discourse markers being a
prime example of this, our still small, but rapidly growing field could become a spearhead
discipline in the search for a novel conception of what language is.
Notes
1
This includes not only item with an obvious connective function of a quasi-logical
nature such as puisque (causal ‘since’), but also markers like French ben or hein,
which have been variously classified as “segmentation signals” (Gülich, 1970),
“punctors” (Vincent, 1993), and “markers of conversational structure” (Auchlin,
1981), and which to many researchers have appeared to be largely devoid of semantic
content.
2
Clearly, the hermeneutic method involves a certain inescapable amount of subjectivity.
To some extent, this can—as suggested for instance by Fischer (2000a)—be overcome
by the use of tests of various kinds, but within the field of discourse marker semantics,
such a procedure is complicated by the fact that, to my knowledge at least, no set of
tests is universally accepted as both convincing and relevant to all types of markers.
Furthermore, the degree of usefulness of tests as an objective measure strikes me as
depending on the type of markers studied: in the case of a number of items (e.g., ben,
hein, etc.), solid judgments of (in)acceptability will be quite difficult to come by, and
judgments concerning contrasts between equally acceptable markers will necessarily
rely on ultimately subjective descriptions of the meanings of the items in question.
However, this situation is one which the researcher shares with the language users
whose competence and performance are the objects of analysis. Indeed, among the
tenets of ethnomethodological Conversation Analysis—a method of analysis which I
have used extensively in past work—is the notion that analysts should seek to describe
such systematic properties of social action (including language use) as are real to
members themselves. Hence, the aim of the analyst is not to deny his or her own social
competence in making sense of activities but rather to employ it and seek to explicate
it (Turner, 1974: 214).
3
I fully realize that the precise definition and circumscription of this variety is no mean
feat. In the following, I will, however, take the not inconsiderable liberty of largely
ignoring this particular problem.
4
Certain markers, such as en guise de conclusion (‘by way of concluding’), may as a
matter of fact be more frequent in the latter modes.
5
As such, it is but a semanticist’s version of Occam’s Razor, and this very useful maxim
has never to my knowledge been fully operationalized either.
6
Whilst not forgetting that frequently occurring “side-effects” may, of course, over time
become conventionalized as additional nodes of meaning. Probably, the line between a
mere pragmatic “side-effect” and a new coded meaning originating in such a “side-
effect” should be drawn in terms of whether or not the nuance of meaning in question
can occur independently of whichever prior meaning was basic to those uses that
originally gave rise to the side-effect (Traugott and Dasher, 2002: 35)
7
In fact, studies of this kind have begun to appear in recent years (e.g., Onodera, 1995;
Brinton, 1996; Traugott, 1999; Traugott and Dasher, 2002; Waltereit, 2002; Visconti,
2003; Hansen, 2005; Hansen and Rossari, 2005; to mention just a few). I am merely
urging that this trend continue.
8
It almost goes without saying, but English translations of French discourse markers
exemplified in this paper are meant only as approximate functional equivalents in the
contexts indicated. Hence, they cannot necessarily be generalized to other contexts.
9
Strictly speaking, Berrendonner (1983) speaks only of “pragmatic connectives”, so I am
extending his claim to comprise the somewhat larger category of DMs.
10
That this is the case even with the use in (8) is shown by the fact that it possesses two
forms, the plural seen in (8), and a singular, dis, which vary according to the number
of individuals addressed and according to the social relationship between speaker and
hearer, such that the singular signals an informal relationship with a single addressee,
while the plural signals a plurality of addressees and/or a formal relationship. This
having been said, in contemporary spoken French, one may not infrequently observe a
generalization of the singular forms to all contexts, which suggests that this item may
be in the process of changing its part-of-speech affiliation.
11
It is, however, highly likely that the categories in question are specificable only in
terms of prototypes and not in terms of necessary and sufficient conditions for
membership (Hansen, 1998a, ch. 3; also Pons Bordería, 1998, ch. III).
12
Thus, modal particles, for instance, appear to be specific to the continental Germanic
languages, although certain items in other languages may have meanings and functions
which approximate those identified for the Germanic modal particles.
13
An interesting, if unanswered, question in that connection is how to give a
synchronically valid account of meanings which have evolved out of earlier meanings
which have later fallen into disuse, leaving in effect a conceptual gap. Clearly, if we
still want to postulate the existence a single lexical item, the synchronic network must
have a different structure from that of the actual diachronic evolution of the item, but
what then about the notion of “diachronic responsibility” (see section 1.4.2 above)?
14
Strictly speaking, a nonpropositional reading of (21) is possible if the negation is
understood as metalinguistic, but I am ignoring that possibility here.
15
The fact that this is a fixed expression obviously raises the question of whether it ought
not to be treated as a separate lexical item, which would consequently not need to be
accounted for in a semantic description of toujours. However, in as much as one
intuitively tends to relate the meaning of toujours est-il to that of toujours and given
that the expression used to be compositional (see note 16 below), I have chosen to
include it as a node (albeit a separate one) in the semantic network of toujours.
16
This was not always the case: in older texts, a predicative adjective could be inserted
between toujours est-il and que, and the form of the verb être was variable. In other
words, the locution can be seen to have undergone a process of grammaticalization
over the past 200 to 300 years.
17
In this, as in several other uses, toujours may profitably be compared and contrasted
with the adverbs déjà (‘already’) and encore (‘still, yet’), as is done in Hansen (2002a,
to appear).
18
Space does not permit me to expose Peirce’s quite complex semiotic theory in all its
details, but see Hansen (2002b) for further information, references, and a sketch of
how the theory may be applied to the analysis of verbal interaction.
Chapter 3
Discourse markers in English: a discourse-

pragmatic view
Diana M. Lewis
1. Introduction
1.1. Approach
A glance at a sample of English spontaneous conversation is likely to find it peppered

with expressions such as well, I mean, so, in fact, though, of course, anyway, actually, on
the other hand, commonly described as discourse particles or discourse markers.1
Although they have attracted particular attention from linguists working on the spoken
language, these and similar expressions permeate written language too.
This chapter takes the view that the discoursal use of the expressions mentioned above is
part of the wider phenomenon of speakers’ attitudes towards the ideas they express. The
study of discourse markers is therefore a part of the study of modal and metatextual
comment and is best approached under the rubric of discourse structure. Our discussion of
the meaning of discourse markers will defend a panchronic view of sense spectra; that is,
the view that the synchronic senses of a polysemous lexeme map earlier and ongoing
functional splits.
The English data are drawn mostly from synchronic and diachronic corpora, with a few
constructed examples.
1.2. Problems
The many recent studies of individual discourse-marking expressions, often based on

corpus data, have provided valuable insights into the phenomenon. Nonetheless, our
understanding of discourse markers is still sketchy. First, the category itself is poorly
defined: a plethora of category labels refers to overlapping groups of expressions. Do the
sorts of expression that have been dubbed discourse markers form a natural class or are
they, rather, a collection of misfits? This problem of definition is addressed in section 2.
Second, there has been uncertainty over how to characterize discourse marker meanings,
44 D. M. Lewis
and this has sometimes led to claims that these are purely pragmatic. What, if any, are the
semantic values of discourse markers? Why are the forms used for discourse marking so
typically polyfunctional? The semantic field of discourse marking and the striking
polyfunctionality of many relevant expressions are discussed in section 3. Section 4
argues that discourse markers must be understood in the light of their historical
development. There follows an overview of the discourse-pragmatic approach and of the
implications of discourse marker studies for the semantics-pragmatics interface.
2. Natural class or misfits?
Definitions of discourse markers have often been couched in negative terms: markers are
said to be nonpropositional, to contribute nothing to truth-conditional meaning, to have
little or no semantic value, to be outside the syntax of the sentence, to be optional
elements, etc. Such negative characterizations risk creating a ragbag class of leftovers.
Yet there is no reason to suppose that the expressions that typically function as markers
are so exceptional.
First, distributional analysis and substitution tests of particular discourse marking

expressions clearly reveal that they have conventional meanings that are part of our
knowledge of our language. Discourse markers are not devoid of semantic content, if by
that we understand conventional or coded meaning.
Second, there is no reason why discourse markers should be exempt from syntactic
analysis, as is sometimes suggested. If there appears to be no place for discourse markers
in certain syntactic models, this does not mean that they are “outside” syntactic structure;
rather, it means either that they are not a syntactic category or that our syntactic models
are inadequate. English expressions commonly categorized as markers can usually be
described as subtypes of sentence adverbials, parentheticals, conjunctions, or transparent
predicates, all of which must surely be accounted for by any adequate syntactic theory.2
The motivation for bringing syntactically diverse expressions together under the discourse
marker label is nevertheless the observation of form-function regularity: similarity of
discourse function and similarity of structure. Relevant forms may be used for discourse
marking to varying degrees. In English, discourse marker refers to a range of form-
function mappings, rather than to a closed set of forms. English discourse marker in the
approach described here is a label for an expression that combines the semantics of
discourse-relational predications with syntactic dependency on a clausal host and low
informational salience. Discourse markers are defined by these discourse-semantic,
syntactic, and information-structural parameters.
3. The semantics of discourse marking
3.1. Rhetorical management
Examples (1) to (3) illustrate the discourse-marking expressions in fact, after all, well, and
anyway.3 Example (1) shows a claim that something was successful, followed by a
measure of its success.
Discourse markers in English 45
(1) The JREI4 .. has proved to be an outstanding success. | In fact, the JREI has been
so successful that [...] it is to be an annual event (ELABORATION)
(Speech by John Battle, British Minister for Science, Energy and Industry, 17
March 1998)
The discourse marker in fact indicates that what follows is an elaboration of the previous
idea.
In (2), Moby’s limited experience of the outside world is presented as justification for
believing that his behaviour is only a phase. In other words, the second argument is
presented as justifying the belief expressed in the first segment.5
(2) We’re sure Moby’s behaviour is simply a phase. | After all, he’s only been
experiencing the outside world since his vaccination course was completed a few
weeks ago (JUSTIFICATION)
(Dogs Today, August 1991, BNC-A17)6
The discourse marker after all provides this link from the second segment to the status of
the first segment, which in this case is expressed overtly by we’re sure. In the
JUSTIFICATION relation, a belief or a claim is justified by citing an idea that is both
strongly compatible with and more certain than that belief. The host of after all must
therefore be an assertion,7 while the related segment is acknowledged by the speaker to be
questionable: it usually either contains a modal qualification or is evaluative. The
speaker’s strategy is to bolster the hearer’s acceptance of the first idea (or of the right of
the speaker to say it) by citing the second: the strong compatibility of the two ideas
suggests that if the second is true, the first is probably true too. The nature of the relation
thus accounts for constraints on the types of segments that can be related by a particular
marker; in this case, a relatively uncertain idea must be followed by a more certain one.
In (3), a similar speaker strategy is at work, but aimed this time at reducing the hearer’s
belief in the first segment.
(3) yeah .. we allow dogs in here ... | well you’ve managed to get one in anyway
(RETREAT)
(Dogs Today, August 1991, BNC-A17)
Well, like after all, introduces a compatible idea that is presented as undisputed. But it is
either a narrower claim than the first, or a tangential claim. Anyway also emphasizes both
the validity of the second idea and its independence from the first claim. The two
arguments of the relation are thus a claim followed by a narrower claim in the same field.
The relation is described above as a RETREAT. It might also be labelled a REFORMULATION.
Without the markers, the second segment, you’ve managed to get one in, might be
interpreted as EXEMPLIFICATION of or EVIDENCE for the allowing of dogs. It does provide
evidence, but the discourse markers indicate that the evidence is not conclusive.8 Other
expressions that can signal the RETREAT relation are actually, as in (4), and at least, as in
(5):
46 D. M. Lewis
(4) You may never have heard of the “postmodernist” challenge to history; [...] | but
you will surely delight in this exhibition of a superb professional historian seeing
it off. | Actually, it is slightly unfair to say that Professor Evans “sees off”
postmodernism, ..
(Electronic Telegraph, 27 September 1997)
(5) Many years later they become lovers | —at least, it is dimly possible to construe
the text in that way.
(The Sunday Times, 19 October 1997)
The identification of a discourse marker category stems from the intuition that discourse
relations such as those above have something in common and that the relational meanings
make up a coherent semantic space.
For a relation to work, there must be some common ground or congruence between two
ideas, and in the case of rhetorical relations, this level is the status—the validity, accuracy
or strength—of the related arguments. Rhetorical relations are essentially persuasive and
include sequences such as the above CLAIM + ELABORATION, CLAIM + JUSTIFICATION, and
CLAIM + RETREAT.
Although any list of relations—and many have been proposed—will necessarily be

somewhat arbitrary, extensive text analysis can provide a working model of the
conceptual space of discourse relations. The approach taken here for the description of
relational meanings is based on rhetorical structure theory (RST) (Mann and Thompson,
1987). The relevant advantage of RST is that it identifies both signalled and nonsignalled
relations. It builds its picture of relations not from the semantics of connectives and
discourse markers but from interpretations of whole texts.9 If a text is coherent, its
segments will all be related. Moreover, RST is open-ended rather than taxonomic: it
allows the relational space to be described in a more finely grained or more coarsely
grained way. And it allows for embedded relations.
As noted above, the types of arguments that can be related by a particular marker are
constrained by the relation associated with that marker, i.e., the marker’s semantics. On
the other hand, it is the types of arguments that a given expression typically links that
allow us to identify the meaning of the marker in the first place. To mitigate this
circularity and to appreciate the role of discourse marking in discourse construction and
interpretation in general, we need a wider view of the field of discourse relations. A
bottom-up, or semasiological, approach, based on analysis of individual expressions and
texts, suggests a range of discourse-relational meanings. An initial description of this
range then enables us to take a top-down, or onomasiological, approach, based on
identifying how relations are expressed. Alternating these approaches and working across
different languages should enable us to refine our model of the conceptual space of
discourse relations. We can then better appreciate differences and similarities among
markers and draw comparisons with other means of expressing discourse relations. For
instance, it was seen above that in English the expressions at least, anyway, actually, and
well have in common that they can express an epistemic RETREAT. Yet these expressions
overlap only partially—they are far from interchangeable in other contexts.
To test the intuition that inter-ideational relations constitute a coherent area of conceptual
space, we need to identify the parameters along which relations vary and to describe the
space they occupy.10 Relations seem to describe either a similarity or a dissimilarity
between the arguments, i.e., to be either consonant or dissonant. Consonant relations, such
as ELABORATION , EVIDENCE , JUSTIFICATION, reinforce the status of the related segment
based on the presupposition of consonance, or close compatibility of ideas. Dissonant
relations, such as CONTRAST, RETREAT, CONCESSION, etc., point to some incompatibility
between ideas.
Another parameter may be degree of subjectivity (and intersubjectivity). The opposition

set up in example (6) between cut and dried .. sorted out, .. and flexible.. spontaneous .. is
somewhat subjective.
(6) now the erm judging people want everything to be . . . well they prefer to have
everything cut and dried .. sorted out .. closed off .. decided ... they don’t like
ambiguity or loose ends ... at all ... the perceiving people on the other hand .. want
to be flexible .. spontaneous .. and responsive
(Careers guidance seminar, recorded March 1993, BNC-G3Y)
Example (7) involves a much more objective contrast, that between dated and up to date
.. state of the art.
(7) so it wouldn’t have struck anybody in nineteen thirteen as in any way dated ... on
the contrary .. it would’ve seemed a very very up to date .. state of the art .. kind
of book
(Lecture, London School of Economics, recorded December 1993, BNC-HUH)
These are typical contexts for the markers on the other hand and on the contrary,
respectively. Speakers can exploit the fact that on the other hand encodes a subjective
contrast, while on the contrary encodes an objective one.
Degree of speaker commitment is a possible further dimension. It was seen in example

(2), for instance, that the JUSTIFICATION relation normally involves two claims of differing
strengths and that after all signals strong speaker commitment. Other relations involve
different configurations of speaker commitment. The views of discourse-relational space
that can be built up in this way allow for crosslinguistic comparisons.
3.2. Information structuring
Most discourse relations involve an asymmetry between the related ideas: one is presented
as more salient, more foregrounded, than the other. Discourse markers also therefore often
assume an information structuring role. In fact, indicating the information structure is a
main function of many markers. The role of discourse markers in foregrounding or
backgrounding their host ideas can be seen in the above examples (1) through (3),
repeated in Fig. 1 below. In each case the discourse marker introduces the less salient
idea. (An arrowhead points to a “nuclear” idea, the tail of the arrow a “satellite” idea
related to it.)
48 D. M. Lewis
Figure 1. Rhetorical structure and information structure of examples 1-3
Elaboration
The JREI has proved In fact, the JREI has

to be an outstanding success. been so successful
that [...] it is to be an
annual event.
Justify
We're sure Moby's After all, he's only

behaviour is simply been experiencing
a phase. the outside world
since his
vaccination course
was completed a
few weeks ago.
Retreat
yeah ... we allow well you've managed

dogs in here ... to get one in anyway
3.3. The multi-functionality of discourse markers
Example (8) shows just three of the common uses of English anyway.
(8) a. And I was wrong. | Adams’s coming to the States did crack the ice floe, |
[...], and Protestants and Catholics are talking. | Would they have talked
anyway?
(The Irish Times, 7 February 1998)
b. There is an apocryphal saying by an actor. | (I think it’s apocryphal; | I’ve

never met him, anyway) (RETREAT)
(The Irish Times, 9 May 1998)
c. Thoreau’s lonely hut was actually in the Emersons’ wood-lot. | Anyway, there
they all were, | these anti-slavery, pro-simplicity, serious New Englanders
(TOPIC RESUMPTION)
(The Independent, 24 November 1993)
In (8a), anyway is not speaker oriented but external. Its interpretation is “if Adams had not
come to the States”. It is a VP adverb in focus. In (8b), anyway is speaker oriented and
relational. It has its own tone group and is sentence-adverbial. It signals an epistemic
RETREAT , in that the anyway segment weakens the status of the previous segments. A
claim (“the saying is apocryphal”) is followed by a weakened claim (“I think”) and then
by a narrower claim presented as definitely true (“I’ve never met the person concerned”).
The claim I’ve never met him is only incomplete evidence for the saying being apocryphal
(cf. example (3)), therefore the speaker cannot maintain the first claim. In written
language, the anyway host often appears in brackets, as here, emphasizing its relatively
backgrounded informational status (at least). Anyway in (8c) is also speaker oriented and
relational. But this time it is discourse-organizational and indicates the resumption of the
main topic line after a digression.
Many discourse-marking expressions are multifunctional in the same way as anyway.

What is the relation between these various meanings? How might the expressions and
their semantic values be represented in the mental lexicon? Three main approaches have
been suggested: (i) the homonymy approach—there are two or more quite separate senses,
(ii) the pragmatic (or monosemy) approach—the form has a single core semantics and the
different interpretations reflect pragmatic ambiguity that is resolved by the context, (iii)
the polysemy approach—the form has two or more related meanings. I shall argue that
this third view is better motivated than the first or second views. However, insights are to
be gained from considering all three approaches.
A radical homonymy analysis looks implausible. The types of ambiguity that can arise
between, for example, the different uses of so or anyway do not seem comparable with
classic homonym ambiguities such as “She brought me a box” (plant vs. container: the
common origin of the two senses in Lat. buxus has long since been obscured). The
polyfunctionality of discourse-marking expressions is far from random, as shown by the
regularities observable in their development by subjectification from lexemes of certain
kinds. The sense distinctions of so, anyway, and so on very often have corresponding
intonational and/or structural distinctions. They are intuitively more akin to derivational
drifts between pairs such as awful-awfully. Yet we cannot rule out homonomy solely on
the grounds of semantic overlap and identity of form; from this linguistic evidence we
cannot infer a single representation in the mental lexicon. As described below, at some
point (for individual speakers), diverging senses can lose their apparent relatedness and
become homonyms. But before that point is reached, uses of a form may still be perceived
by speakers as related without being necessarily predictable from one another.
The pragmatic (or monosemy) view holds that “a single semantics is pragmatically
applied in different ways according to pragmatic context” (Sweetser, 1990: 76). It has
long been noted that ideas may be related either in the external sociophysical world or by
the speaker.11 This is the distinction made in section 2, where discourse markers were
defined as speaker oriented, expressing subjective views on relationships between ideas.
A relational expression cannot be used to express simultaneously both external and
speaker-oriented relations. The differences are illustrated in (9) (constructed examples):
(9) a. It’s not green, but red.

[external relation based on real-world incompatibility: red implies not green]
50 D. M. Lewis
b. I like red, | but my sister likes blue.

[speaker-defined relation based on presupposed incongruity between two
ideas]
c. J’s going away next month, | but you probably knew that.
[speaker-defined relation based on presupposed incongruity between
informing hearer of X and hearer knowing X already]
Whereas in (9a) the object cannot be both red and green, there is no inherent
incompatibility in (9b) or (9c). The “single sense, pragmatic ambiguity” view suggests
that but retains the same core sense of contrast or adversativity in all three instances and is
correctly interpreted in each case by pragmatic inference from the context.
While this analysis appears to work well for these but examples, problems arise for some
expressions. One problem is the way meanings are realized across the domains. Although
it implies a pragmatic paradigm, the pragmatic view does not explain the gaps in the
paradigm. Some expressions can be used only to signal a speaker-oriented relation. For
instance, after all can only introduce a reason for a speaker’s stance and cannot indicate
an external causal link (10a), while because/’cos can do either (10a, b). Where there may
be ambiguity, both can occur (10c).
(10) a. then when they got older .. me dad sort of took our John on .. because our
John were more mechanical minded .. and our Colin got pushed out a bit ..
‘cos/*after all our Colin weren’t interested in cars
(Conversation, recorded February 1992, BNC-KB1)
b. I think the Queen’s done an excellent job ... | ’cos/after all she was put in that
job when she was only a young girl (JUSTIFICATION) (’cos in original)
(Television discussion: “The Royals”, BNC-FLE)
c. I gave mum thirty-five pound because after all you know .. I think she needs
it
(Conversation, recorded April 1992, BNC-KDN)
It is not clear why there might be domain-independent senses that were blocked for use in
one domain or another. In other words, the “single core sense” model does not explain
why there is no apparent synchronic productivity. It is not rare, crosslinguistically, for the
same type of relation in two different domains to be expressed by two different lexemes.
All these observations suggest a semantic rather than a pragmatic difference.
Another problem is how to define the core sense of an expression. The pragmatic
approach implies that a knowledge of the core sense plus an interpretation of the host
discourse segment must be adequate to interpret a discourse marker token. It is relatively
easy to posit some core sense for most lexemes used for discourse marking. Yet the
salience of that sense, and the degree to which it would need to be enriched by inferencing
in order to interpret particular instances, vary greatly across lexemes. This variation can
be illustrated by anyway, at least, and in fact.
In the case of anyway (example (8) above), the sense of “whatever the case/
independently/whether or not’ is clearly common across the three main uses. The
examples can be paraphrased as follows: whether or not Adams had come (8a), whether
or not “it’s apocryphal” is true (8b), whether or not I said the foregoing (8c). To arrive at
the appropriate interpretation of anyway, the reader need only identify whether the related
argument is the external situation described, the epistemic status of a proposition
expressed, or a chunk of discourse.
But some forms have far less transparently related uses. An example is at least (11).
(11) a. it goes on until at least nine thirty or ten [external]

(Conversation, recorded April 1992, BNC-KCH)
b. Charles is a stick in the mud | and the other one’s the other way .. seems to
be the other way inclined | .. at least she won’t .. won’t be er short of a
bob or two will she? [speaker-oriented, evaluative]
(Conversation, recorded March 1992, BNC-KCL)
c. but things seem to be moving a little bit .. | at least we’re told they are
(RETREAT) [speaker-oriented, rhetorical]
(Business meeting, recorded January 1994, BNC-JA6)
The examples in (11) may all be perceived as scalar, as “this much and perhaps more”.
But it is hard to describe a “core” semantic value for at least that would be rich enough to
allow the hearer to compute the relevant interpretations for (11b) and (11c) from
contextual clues.12 It seems more plausible that the positive evaluation notion of (11b) and
the RETREAT notion of (11c) are semanticized and nondefeasible in present-day English.
In other words, at least is polysemous, with three related but distinct conventional senses
(also differing in intonation and structure).
Some expressions recruited for discourse marking appear to have split to the point of
having almost opposite meanings. An example is in fact, which can be used either to
introduce a reinforcement of an idea (12a), or to introduce a refutation of an idea (12b).
(12) a. | he’s not (...) nice looking | in fact he’s (...) nothing .. you know .. nice
looking at all | but he’s a nice bloke (ELABORATION)
(Conversation, recorded January 1992, BNC-KCA)
b. The river just to the east of Tarsus is marked as the Goksu River; in fact, it
is the Seyhan River. (ANTITHESIS)
(Times Higher Education Supplement, 1 May 1998)
While both uses share the notion of epistemic certainty, inherited from the PP in fact used
as a VP adverbial, this notion is not always sufficient to compute adequate interpretations
in context. The split between the two uses is reflected in their different prosodic contours
and in their different information structural properties: in fact[elaborative] backgrounds its
host, while in fact[contrastive] foregrounds its. Contrastive in fact always implies an
erroneous claim in the related segment (in this example, “is [erroneously] marked as”). By
contrast, there are few constraints on the use of Elaborative in fact: it is compatible with a
wide range of contexts, and in present-day British English it is extending beyond
elaboration and evolving into an additive marker with even fewer contextual constraints.
Anyway, at least, and in fact exhibit different degrees of divergence among their various
uses. Some polyfunctional discourse-marking expressions, then, have two or more clearly
related and mutually predictable meanings; others have clearly related but nonpredictable
ones (i.e., they have different conventionalized senses which must be learnt). As Croft
52 D. M. Lewis
points out, “there is no a priori reason to think that speakers always recognize identity of
form in linguistic units and then construct a semantic relationship between uses” (1998:
157). A panchronic polysemy analysis accounts for transparent relatedness of meanings
while allowing for lack of predictability. It thus caters for degrees of relatedness and for
the observation from corpus studies that tokens cluster into subsenses, some of which
seem closer and more transparently related than others. Evidence from psycholinguistic
experiment into the processing of polysemous adjectives, nouns, and verbs suggests that
mental representations may include both schematic, semantically-underspecified entries in
the mental lexicon and more fully specified subsenses (Brisand et al., 2001). This type of
representation is plausible for discourse marking expressions too. The polysemy analysis
reflects, in the synchronic subsenses, the diachronic “layering” that has given rise to
them.13 This brings us to the question of the diachronic development of discourse
markers.
4. The development of discourse markers
In English, as in many other languages, discourse markers develop largely through

internal lexical semantic change, and this is the source of the multifunctionality of the
lexemes (Traugott, 1995a, 2003a; Traugott and Dasher, 2002). Discourse-marking senses
tend to arise from repeated usage of a lexeme in particular context types, leading
eventually to functional split. While the split is little advanced, the incipient new sense is
transparently related to the old one and the new interpretation is predictable. When the
split is more advanced, there may be no perceived relation between the senses (again, with
variation across speakers).
Several studies have examined the historical background to some common present-day
English discourse markers and have traced the functional splits that took place one or
more centuries ago (e.g. Jucker, 1997; Lewis, 2002; Powell, 1992; Traugott, 1997). The
histories of some Old English and Middle English markers are traced in Brinton (1996).
Here we shall look briefly at an expression of Modern English—of course—which
acquired a relational sense and then lost it, and which in present-day English displays
further signs of splitting (Lewis, 2003).
How did of course acquire its current identity as a marker of speaker commitment?14 The
usage of of course seems to have shifted from mainly PP to mainly lexicalized adverbial
around the late seventeenth and early eighteenth centuries. Of course of the period
displays two main, closely related subsenses: ‘naturally’, i.e., due to the way the natural
world works (13), and ‘normally’, i.e., due to established human conventions and norms
(14).
(13) it was impossible, that the Dirt, wherewith I was so freely and bountifully
bespatter’d, should stick long upon me, that a little Time would of course dry it
off
(1692, Letter from Mr Humphry Hody to a friend, Lampeter Corpus)
(14) ... the Articles to be Engrost & they will be sent up to the Lds on satureday next
When the Lds have Recd & Read the same then of Course they will give the Lds
Impeached A Certaine day to put In their Answer & then things will goe Currantly
on In order to A speedy triall
(Newsletter, 3 April 1679, Newdigate Corpus)
In the early eighteenth century, of course is found in two common context types.15 The
first is causal contexts (15) :
(15) they learn Love-Songs, [...] This of course makes them wanton, and so they think
of Husbands, before they are capable to choose for themselves
(1729, Sermon, Lampeter Corpus)
The implicatures picked up by of course in such contexts may account for the emergence,
probably in the early- to mid-eighteenth century, of the usage of of course in a relational
sense akin to thereby or therefore (16). For the fourth edition of his dictionary (1773),
Johnson amends his entry for course and enters the expression of course with the sense
‘by consequence’. This relational of course expresses a nonvolitional result which may be
an objective consequence, as in (16a), or a more subjective one, as in (16b).
(16) a. My malt ... does not shrink so much when it comes to be laid on the kiln; of
course it measures to more advantage
(1765, Museum Rusticum et commerciale: or Select papers on agriculture, III.
222 , OED)
b. Surely of all human characters a fanatic philosopher is the most incongruous,
and of course the most truly ludicrous
(1788, Horace Walpole, Walpoliana, Fr. Philos. 50, OED)
The second common early eighteenth-century contexts of of course are epistemic: the
speaker deduces some conclusion (17):
(17) As Homer is the Author nearest to those, his Style must of course bear a greater
Resemblance to the sacred Books than that of any other Writer
(Alexander Pope’s preface to his translation of Homer’s Iliad, 1715).
This type of context is likely to have given rise to the epistemic usage of of course, which
is attested at least from the turn of the nineteenth century (18).
(18) An aristocracy, of course, is naturally the first form of government.

(1792, Mary Wollstonecraft, A Vindication of the Rights of Women)
For both these developments, a discoursal motive is likely. If it is reported that one event
occurs and then that a second event occurs “naturally, as a matter of course”, it will easily
be inferred that the second event follows logically from the occurrence of the first. This
can be exploited by speakers. Depending on the degree to which the second situation is
known or said to be the case, the pragmatic force of the sequence will be either that the
second event was a result of the first or that the occurrence of the second event can be
logically deduced from the report of the first plus knowledge of the world. In either case,
the speaker’s warrant for the assertion is thus strengthened.
54 D. M. Lewis
An analysis of present-day English of course reveals two related discrete uses with
corresponding intonation contours: an “emphatic yes” use and a use that is often glossed
as ‘as expected’ or ‘naturally’. There is a clear unity of sense across tokens of the latter
type of of course. It acts as a relevance hedge, in that the speaker/writer anticipates the
hearer/reader’s expectations. In other words, it has become a marker of intersubjectivity,
lending support to Traugott’s (2003b) argument that there is a tendency for unidirectional
semantic change towards greater subjectivity and intersubjectivity of meaning. This
interpersonal role is much exploited by orators: of course is extremely common in
conversation and argumentation and is especially frequent in political speeches.16 Yet the
distribution of of course is not as random as this rather general sense of ‘naturally’ would
suggest. In fact there are at least four regular contexts of use, with corresponding nuances
of interpretation or “contextual modulations” (Cruse, 1986: 52-3). These contexts, i.e., the
roles played by the host of of course, include concession (19a), background in narrative
((19b) and (6)), topic shift (19c), and end of list (19d).
(19) a. Of course, the Sereny-Bell version may be the truth, | but we cannot, on
the basis of this book, make that assumption
(The Sunday Times, 10 May 1998
b. A: oh we did that a few times | when the .. the borders were closed with
(*** ) | came across the pontoon through .. | of course it’s a much more e-
.. elaborate pontoon now than it was then
B: I’m sure
A: that was er
B: oh it wasn’t very elaborate
(Conversation, recorded March 1992, BNC-KB0)
c. [current topic = NATO] ... | because creating a relationship of trust with
those outside NATO is just as important as enlarging NATO. [new topic
heading “The EU dimension”] | Of course change in NATO is only part of
the story of change in Europe.
(Speech by Malcolm Rifkind, UK Foreign Secretary, 10 March 1997)
d. We are already working with you ... on measures to counter the drugs
threat. | We also want to work towards a satisfactory banana regime ... |
And of course we are at one with you in resisting the objectionable extra-
territorial effects of the Helms-Burton legislation.
(Speech by Baroness Symons, Parliamentary Under Secretary of State,
UK Foreign & Commonwealth Office, 10 October 1997)
In a corpus of political speeches, three quarters of the tokens of of course are accounted
for by the above four contexts (37% concession, 6% background, 21% topic shift, 9%
end-of-list; n = 668). Each regular context imbues of course with a salient implicature. In
concessive contexts like (19a), of course implies “I grant that”; in narrative contexts like
(19b), “bear in mind that”. In (19c) and (19d) the discourse-organizational function is
more salient. In concession, backgrounding, and topic-shifting contexts, of course carries
a contrastive implicature.
Since the seventeenth century, then, the sense of of course has been shifting and
broadening by degrees. Over a relatively brief period of time, it has lost its external use
altogether (this sense can now only be expressed by the phrase as a matter of course). It
has acquired and then lost a causal relational sense as it has undergone increasing
subjectification and intersubjectification. It has greatly increased its frequency of

occurrence during its short life so far and now displays the kind of regular contextual
modulation that could herald further functional splitting. These modulations result from
the co-occurrence of of course with particular discourse constructions: familiar
constructions acquire default implicatures in addition to their compositional meanings.
The evidence from of course thus lends support to Levinson’s (1995, 2000) proposal for a
third level of meaning between the semantic and the pragmatic: the “utterance-type
meaning”. These intermediate meanings are stable default interpretations that accrue to a
construction and then extend to a component expression through typical use. Of course
may not be a full-fledged discourse marker by the definition of discourse marking given
above; it does, however, regularly carry discourse-marking implicatures.
If we assume that semantic shift is likely to occur across conceptually contiguous

relational meanings, we can make use of diachronic evidence, such as CONDITION >
CONCESSION shifts, to help build a picture of discourse-relational concepts. Relations of
CONDITION , CONTRAST , and CONCESSION are known to be close and to interact in
interesting ways (König and Siemund, 2000). Synchronically, the occurrence of a
discourse marker with different relations may reflect the conceptual closeness of those
relations. As seen above, for instance, REFORMULATION and RETREAT seem similar. And
for example not only occurs with the ELABORATION relation but can also be used to
introduce EVIDENCE, suggesting that these relations are close, too, or that one is a subtype
of the other. Sometimes, however, one form can indicate seemingly opposite types of
relation. These observations highlight the need to account for the polyfunctionality of
discourse markers in a panchronic perspective which recognizes discourse pragmatics as
the source of discourse markers, and, more generally, language use as the motor of
language change.
5. The discourse-pragmatic view
The contribution of discourse markers to discourse in this approach is discourse semantic

and information structural, as has been seen.
On the discourse-semantic level, discourse-marking status is attributed to those tokens of

a lexical or quasi-lexical expression that describe a speaker-determined relation between
two discourse segments (as opposed to a real-world relation between two events, states, or
individuals).17 The token acts as a predicate with two arguments. Discourse markers in
this framework, then, are discourse relational and speaker oriented. They are distinguished
from nonrelational speaker evaluations such as unfortunately, and from relational, non-
speaker-oriented connectives such as because or then as used to describe real-world
(external) causal or temporal relations.18 On the informational level, discourse relations
themselves are typically backgrounded. Discourse markers are often realised as
parentheticals. Nonparenthetical markers tend to occur early in the host clause. Moreover,
the discourse marker, together with the relation it expresses, frequently helps to define the
informational relation (relative salience) between the host segment and the related
discourse segment(s). Markers are described above as lexical or quasi-lexical. Each
discourse marking expression has a syntactic host such that the host and the marker can be
identified in the linguistic structure. The marker has scope over the host. Markers are thus
distinguished from self-standing comments such as those expressed in independent finite
clauses and from interjections. This formal constraint reflects how discourse marking is
56 D. M. Lewis
typically realized in English; for other languages, other formal boundaries may be
appropriate.
The proposed discourse marker category attempts to capture the co-occurrence of certain
types of speaker-attitudinal, relational meaning with certain types of coding. Whether this
is the most useful place to draw a category boundary remains to be shown by further
research. It corresponds to an intuition that inter-ideational relations, such as contrast or
elaboration, form some conceptual paradigm: some coherent area of conceptual space.
The relational meanings may well be universal, whereas the means of expressing them are
language specific. The definition here is based on English data and aims to cater for both
typical instances of discourse marking and more marginal ones. The term is not exclusive,
since it applies to expressions having simultaneous nonrelational properties too: in
English at least, discourse markers reveal much change in progress, with certain lexemes
slowly acquiring relational senses.
The semanticist’s view of discourse markers has been remarkably different from that of
the pragmaticist. The conventional (coded) meanings of discourse markers, not
contributing to truth-conditional meaning, have been allocated to the category of
conventional implicature and largely ignored by semanticists. But, as Lyons points out,
“the lexical and grammatical resources of a particular language can be adapted and
exploited to propositionalize what is not of its nature propositional” (1995: 274).
Discourse-relational meanings can be easily propositionalized, and according to some
semanticists may even be analysed truth-conditionally: “perhaps conventional
implications do make a contribution to truth conditions of a special context-dependent
kind that reflects only the speaker’s attitudes in a way analogous to certain uses of
modals” (Chierchia and McConnell-Ginet, 1990: 284).
Within pragmatics, by contrast, discourse markers have been a focus of attention.

Valuable qualitative work on individual expressions is being followed up by quantitative
analyses that can throw new light on the semantics/pragmatics interface. The distribution
of discourse-marking expressions reveals several regularities which any model of the
semantics/pragmatics interface should take account of. First, many—though not all—the
forms used for discourse marking have an external use as well as one or more speaker-
oriented uses. While some of these expressions plausibly have a single semantics in the
mental lexicon, for others it is hard to imagine a single sense rich enough to produce
adequate interpretations in context. This is explained by the diachronic development of
markers being visible in the synchronic state: advanced usage splits are revealed in forms
having two apparently unrelated or tenuously related senses, while incipient splits show
up as regular contextual modulation. Semanticization is thus a matter of degree. Instead of
choosing between a monosemy and a homonomy approach, we can accept that there are
degrees of relatedness and that these vary across speakers. (The diachronic dimension is,
of course, always present across speakers from older and younger generations and across
speakers of more conservative and more advanced regional or social varieties.) The
degree to which the relational sense of an expression is conventionalized can in principle
be measured by standard defeasibility tests. In practice, it is difficult, and perhaps not
useful, to establish a boundary. Second, interpretations of discourse marking expressions
tend to be influenced by more than one level of the rhetorical hierarchy at a time: that is,
by the host unit and the wider rhetorical context. Third, one-to-one mapping between
discourse marker and coherence relation is rare: the extensions of markers tend to overlap.
The way a language manages its discourse-marking functions also reveals information-
structural tendencies. Each segment of a coherent stretch of discourse is related to the rest
in such a way that it contributes to the goals of the speaker. The relation may be given a
high profile by a clause-level predicate such as I conclude from the foregoing that or What
I really meant to say was. The relation may be signalled less overtly by a conventional
discourse marker, such as so or I mean. In some languages certain relations may be
signalled by affixes. A relation may be implicated by a sentential modifier ostensibly
indicating time or manner or modality. Finally, it may be left for the hearer to infer a
relation from simple juxtaposition and the nature of the arguments. In English discourse,
and possibly in that of most languages, a minority of relations is overtly signalled (Mann
and Thompson, 1987), usually by informationally backgrounded means. The majority of
relations are implicated, despite the availability of expressions carrying clearly
semanticized relational meanings. Why might speakers prefer to implicate relations rather
than use these explicit, fully semanticized expressions?
Several possible explanations suggest themselves. One is politeness (see Fischer, this
volume; Weydt, this volume). The expression of speaker-oriented meanings such as
beliefs and evaluations involves face, and one strategy for managing face is to invite
inferences rather than be explicit. That is, to exploit relatively stable implicatures which,
because they still leave room for a retreat, are ideally suited to avoiding potential conflict.
Another possible explanation, related to politeness, is argumentation strategy. A message
that induces the hearer to draw his or her own conclusions to match those of the speaker
will be more powerful. A third is economy. Most relations need no clarification. Where
they do, the preference is for the idea and the comment on it to be accommodated in a
single clause, for relational predicates to be collapsed into single lexical items, and so on.
Speaker-attitudinal markers in general (discourse markers, hedges, evaluatives, etc.) tend
to be short forms: adverbs, particles, affixes. These are clearly speculative reasons. But
they may in the long term go some way towards explaining quite why discourse relational
meanings tend crosslinguistically to adopt the forms they do.
The relational categories we posit, the number of them, and the labels we use for them are
naturally grounded in our interpretations of discourses. Some types of relation seem
typically to be overtly marked while others are rarely marked. Some relations can be
expressed by a range of forms, while others seem to have only one or two exponents.
Many areas of discourse marking remain obscure. There is a pressing need to understand
better the various stress and intonation patterns of discourse markers and their hosts. The
relatively high turnover rate of discourse markers also merits further investigation, as does
the expression of discourse relations between subclausal elements and fragments. But by
beginning to describe the discourse-relational space we may find clues to these and other
discourse-marking puzzles.
Notes
1
The term discourse marker has been preferred to the many other possibilities on the
grounds that it has become the most frequently used term in discussions of English
data. Moreover, particle implies a grammatical category and seems at odds with the
58 D. M. Lewis
view taken here that the category of discourse marker is principally defined by
discourse function.
2
Discourse markers have suffered until recently from some neglect by syntacticians of
sentence-peripheral phenomena. There is distributional and phonological evidence that
some markers which do not look morphologically like adverbs have undergone, or are
undergoing, reanalysis as sentential adverbs (Thompson and Mulac, 1991). There may
be grounds for positing a separate syntactic category or subcategory for discourse
markers along the lines of “conjunctive speaker-oriented sentential adverb” (Bellert,
1977), but this category has yet to be fully explored. It is ignored, for example, by
Cinque’s major study of adverbs (Cinque, 1999: 11), though Traugott and Dasher
(2002: 187-188) propose extending Cinque’s hierarchy of clausal functional
projections to include Modal projections for discourse markers (specifically,
ModDMhedge for expressions such as well and ModDM for expressions such as then).
How far the formal evidence will allow increasingly fine-grained functional
projections to be correlated with discourse-semantic distinctions remains to be seen
(see also Rizzi, 1997). Meanwhile, the syntactic distribution of most of the
expressions described in the English discourse marker literature is perhaps best
described as that of functional adverbial adjuncts (Ernst, 1902: 9).
3
Relevant discourse segment boundaries are marked by |. The examples of spoken
language are based on punctuated transcripts. .. replaces a comma, ... a full-stop, and
[...] indicates that material is omitted.
4
JREI: ‘Joint Research Equipment Initiative’.
5
It is not easy to distinguish between a reference to the epistemic stance and a reference to
the illocution itself, i.e., whether it is the belief or the statement of the belief that is
being justified. The reference may be to both.
6
BNC = British National Corpus.
7
This constraint on the status of the after all host reflects Lyons’ distinction between
objective and subjective epistemic modality (1977: 797-800): compare “Don’t be cross
with her for being late. After all, she may have been held up” (objective) with “*After
all, perhaps she’s been held up?” (subjective).
8
There are many analyses of English well; see, for instance, Aijmer et al. (this volume)
and Schourup (2001). On anyway, see Ferrara (1997) and Lenk (1995).
9
The relations originally proposed for English by Mann and Thompson (1987) were the
outcome of analyses of several hundred written texts by several researchers. The RST
claim was that the relations were independent of any overt marking by elements such
as discourse markers. However, it seems clear that although much of the time markers
can be seen to strengthen an existing inferential relation, they often supply that
relation, as shown by the consequences of omitting or changing the markers.
Rhetorical relations are defined here more narrowly than in RST, where they include
not only speaker-oriented relations and discourse-organizational relations but also non-
speaker-oriented relations.
10
Interesting approaches to parameterization include Sanders et al. (1992b) and Louwerse
(2001).
11
These two domains have been variously named in work on coherence, cohesion, and
conjunction, for instance, external/internal (Halliday and Hasan, 1976),
propositional/illocutionary (Sanders and Spooren, 1999), subject-matter/presentational
(Mann and Thompson, 1987), semantic/pragmatic (Sanders et al., 1992b), content vs.
epistemic, and speech act domains (Sweetser, 1990).
12
See Kay (1992) for a discussion of the nonpredictability of senses of at least.
13
The term layering is due to Hopper (1991).
14
This diachronic account is based on an analysis of 20 seventeenth-century, 68
eighteenth-century, and 306 nineteenth-century examples of of course.
15
Of course in the eighteenth century was also used as a VP adverb in the sense of ‘by
turns’. The dialogic “emphatic yes” usage seems to have developed in the nineteenth
century.
16
In political speeches of course shows a frequency of 550 per million words, based on a
sample of 1.3 million words, compared with around 280 per million words in
conversation, based on a sample (the demographically sampled section of the British
National Corpus) of approximately 4 million words
17
A discourse segment is a sequence that expresses an idea unit. Although it often maps to
a clause or an intonation unit, it is neither a syntactic nor a prosodic category but an
information structural one. A single NP, for instance, may constitute a discourse
segment. On segmentation and idea units, see Chafe (1994), Croft (1995).
18
The distinctions among discourse markers, evaluatives, hedges, and connectives are by
no means clear-cut. It has been claimed, for instance, that evaluative sentence
adverbials can also function as relationals (Thompson and Zhou, 2000). Expressions
can function simultaneously as discourse marker and hedge; an example is well (see
example (3)). And while temporal connectivity seems objective enough, some
temporal connectives imply causality which arguably involves speaker attitude.
Chapter 4
The rise of discourse markers in Italian: a

specific type of language change
Richard Waltereit
1. Introduction
What are the specifics of the history of discourse markers? Even though a number of
studies have shed light on how they evolved in the course of time, I am not aware of a
coherent model on why discourse markers originated in the first place. Assuming that
discourse markers arise from lexemes belonging to other word classes, we may ask why
these lexemes turned into discourse markers and not into some other category. My
contribution is intended as a step towards such a model. An appropriate theory of the rise
of discourse markers helps us to understand to a great extent their synchronic properties as
well. Therefore the history of discourse markers is not just one problem among others in
the field but a problem whose clarification promises also to offer a full-fledged approach
to discourse particles.
1.1. Approach and methodology
My approach is inspired by recent developments in the theory of grammatical change

(Haspelmath, 1999; Detges, 2000a, 2000b, 2001). In earlier work on discourse markers
(DMs) it was often routinely assumed that their rise is an instance of grammaticalization,
and no significant distinction was made between the recruitment of discourse markers and
the rise of grammatical categories such as tense or case. More recent developments in
grammaticalization theory try to relate grammaticalization with specific speaker
motivations. The goal of this chapter is to show that the consideration of specific speaker
motivations can plausibly account for the typical properties of the diachronic recruitment
of discourse markers as well. The rise of discourse markers and grammaticalization proper
have many properties in common but also crucial differences. A careful consideration of
these differences will also help to explain some of the specific properties of discourse
markers.
When speaking of “speaker motivations” I do not wish to make reference to psychology.

Rather, I am invoking the Gricean pragmatic framework. It seems to me that speaker
motivation is an analytical category perfectly compatible with Gricean pragmatics for the
62 R. Waltereit
simple reason that the computation of implicatures relies on supposed communicative

intentions. Speaker motivations can be thought of as an abstraction of communicative
intentions.
1.2. Data
I will mainly consider Italian discourse markers. My data stem from the publicly
accessible Italian spoken language corpus LIP (Lessico dell’italiano parlato) as well as
from the equally publicly accessible Italian historical corpus LIZ (Letteratura italiana
Zanichelli). LIP (De Mauro et al., 1993) is an electronically searchable 540,000-word
corpus of contemporary spoken Italian. The composition of that corpus is balanced so as
to systematically reflect regional variation of Italian as well as various conversational
genres (telephone conversation, multiparty conversation, monologal speeches, etc.). LIZ,
also electronically searchable, contains 500 Italian texts (21,000,000 words) from the 13th
to the 20th century from a large variety of literary genres. For a diachronic approach it is
important to have data from earlier stages of the language. Even though these data are not
taken from natural conversation, they can be a valuable source of evidence for better
understanding the historical development of these markers.
1.3. Problem statement: what makes a nonmarker turn into a marker?
Many discourse markers have a homophonous counterpart which is not a discourse

marker. For example, the Italian form diciamo has a marker use (1) and also a counterpart
as a verb form (2):
(1) C: mh mh mh si' # quindi eh potrebbe fare questo_

questo primo capitolo piu' teorico in cui rientra magari qualche commento
sul sulla questione filosofica e poi addentrarsi di piu' nella questione
funzionale piu' nel merito DICIAMO della_ # dello specifico non so
G: si' si
C: DM DM yes hence DM you could do the following thing this first more
theoretical chapter, where you write some commentary on the philosophical
question and then you go into the more functional question, more on the
merits diciamo of the specific DM
G: yes yes
(2) B: parliamoci chiaro quanti giornalisti sportivi voi vedete in televisione a tu per
tu con un giocatore di calcio un campione no un ragazzino che arriva dalla
serie bi dirgli in faccia ma tu eri al cinema ma tu giochi quando hai voglia
ma anche tu eh no […]
A: si' ecco io io non e'
B: ahah ti va
riconosciuto questo merito ahah non e'_ no DICIAMOLO dai
A: no_ no_ va
be' io_ io dico ma io_ non tu perche'per esempio do del lei ai giocatori_ ma
lui che faceva_ cosi'
The rise of discourse markers 63
B: let’s talk point-blank how many journalists do you see on TV saying tu to a

soccer player a champion DM not a lad who arrives from the B league say
to him like that DM tu were at the cinema?1 Tu play when tu want but also
tu DM no […]
A: yes DM I I it’s not
B: DM ah you have this merit ah no it’s not, no shall we say it DM
In (1), diciamo occurs in the middle of a noun phrase (nel merito diciamo dello specifico
‘in the merit DM of the specific’). Most probably, it is a hesitation phenomenon (speech
management); a very typical discourse marker function. The speaker seems to be
momentarily unsure as to how to continue her current turn. In order to have some time for
formulation and to fill the pause, she utters the marker diciamo. In (2), however, diciamo
is not a discourse marker but a 1PL present subjunctive (i.e., imperative) of the verb dire
‘to say’. It is clearly a verb form, already because it governs a clitic object lo ‘it’.
According to Italian verbal morphology, it could be not only a subjunctive form but also
the 1PL present indicative. But the ensuing exhortative marker dai clearly suggests a
subjunctive interpretation.
Assuming that the verb form is diachronically prior to the discourse marker, a crucial
question arises: how, and why, did the verb form develop the marker use? This will be the
main concern of the present chapter. Previous studies on the history of discourse markers
have mainly studied their diachronic rise as an instance of grammaticalization and
subjectification. Subjectification is a form of semantic and/or grammatical change
whereby forms denoting “objective”, ideational meanings acquire more speaker-based,
subjective, attitudinal meanings in the course of time (Traugott, 1995b, 1999b). Onodera
(1995), Brinton (1996), and Traugott (1999b) investigate the history of discourse markers
in Japanese, Old and Middle English, and Modern English, respectively. Onodera studies
the Japanese discourse markers demo and dakedo, which originated from intrasentential
clause-combining devices. Today, demo and dakedo are discourse markers, both of which
can be used for point making, floor claiming, opening the conversation, and change of
topic. Brinton (1996) offers an in-depth study of a variety of discourse markers. For
example, the Middle English discourse marker gan stemmed from the Old English
aspectual (ingressive) verb onginnan ‘to begin’. Brinton (1996: 80-81) reports that in
Middle English this verb turned into a discourse marker with textual meaning marking
“new or important stages in the plot development”, which later underwent an additional
change to an “interpersonal (emphatic, intensive)” function. Traugott (1999b) and
Schwenter and Traugott (2000) study the history of in fact, which developed from
prepositional phrase via sentence adverbial into a discourse marker. All of these findings
confirm the overall tendency of subjectification, the development from propositional via
textual to attitudinal, subjective meanings. They describe a certain path of functional
change. Given that the same tendency of subjectification has also repeatedly been
observed in grammaticalization processes, many researchers were tempted to subsume the
rise of discourse markers under grammaticalization.
But this does not explain why speakers began to use forms such as the Italian subjunctive
diciamo in a new way, nor does it explain why some forms undergoing subjectification
ended up as discourse markers while others ended up as grammatical categories such as
tense or preposition. These problems are the starting point for the approach to discourse
markers presented here.
64 R. Waltereit
2. Definition
As explained in the introduction, I have chosen a diachronic approach to discourse

markers. Their very character depends upon the type of language change they have
undergone. In order to avoid a circular argument, however, it is necessary to have
synchronic criteria by which to identify them. This is the goal of the present section.
2.1. Discourse markers, a functional class
Intuitively, DMs are perceived as a homogeneous group of words. But they do not share
morphosyntactic or other formal features that would allow for a formal definition of this
word class (disregarding properties that are common to all types of markers, such as not
being subject to inflection). As argued by Hansen (1998a: 69-70) they do not form a
paradigm in the structuralist sense, where items carry meaning only by opposition to other
items of the same paradigm. For a more precise characterization of DMs, reference to
their function is required.
It also seems uncontroversial that DMs are nonpropositional. Their function lies outside
the ideational realm of language. They belong to both the textual and the interpersonal
language function. I would like to follow Hansen (1998a: 358) who argues that DMs
“function as instructions from speaker to hearer on how to integrate their host unit into a
coherent mental representation of the discourse”. That is, they situate their host unit with
respect to the surrounding discourse and with respect to the speaker-hearer relationship. (I
will address the definition of host unit in the next subsection). DMs can position their host
unit in many ways: they may present the state of affairs designated by the host unit as a
temporal or causal consequence of the preceding state of affairs, they may highlight their
host unit in various ways, or they may present the host unit as closing the current
discourse topic. In-depth studies of DMs have uncovered extremely subtle nuances in the
construction of textual and interpersonal meaning.
2.2. Scope variability
A particularly distinctive property of DMs is their scope variability. Scope variability has
two facets. For one, the scope of DMs as a word class can vary considerably. That is, they
can have scope over portions of discourse that range from intrasentential units to entire
turns composed of several sentences ( Hansen, 1998a: 73-74). This variability is a
remarkable property, but it is not an exclusive feature of DMs. For example, conjunctions
as a word class (and even some individual conjunctions as lexical items) can also have
variable scope, as with the conjunction and in the following example:
(3) a. Ed and Doris loved each other.

b. Ed worked at the barber’s, and Doris worked in a department store.
In (3) and has scope over two NPs; in (3) it has scope over two clauses. But importantly,
the scope of the conjunction and can always be determined in grammatical terms. It could
be defined as ranging over the two same-type constituents adjacent to and that recursively
make up a constituent of again the same type. The scope of DMs, however, cannot be
determined in grammatical terms. Consider the following examples of the Italian DM
insomma:
(4) B: insomma voler concentrare la mia attenzione di voler svolgere una ricerca
su un periodo particolare della linguistica tedesca del secondo dopoguerra e
eh insomma ho_ eh detto appunto che era un c'era insomma una temperie
culturale INSOMMA emergeva una certa insoddisfazione
DM to concentrate my attention, develop a study on a particular period of
German post-World War II linguistics DM DM I’ve said that it was DM a
cultural upheaval insomma a certain dissatisfaction was emerging.
(5) B: ed e' oddio non mi ricordo va be' non importa INSOMMA eh per cui mi
leggero' questo qui
and DM I don’t remember ok it doesn’t matter insomma that’s why I will
read this
Both examples have insomma in a highly similar syntactic environment. Insomma occurs
between two main clauses. In (4), it is the sequence ho detto appunto che era un c'era
insomma una temperie culturale INSOMMA emergeva una certa insoddisfazione. In (5) it is
the sequence non mi ricordo va be' non importa INSOMMA eh per cui mi leggero' questo
qui. However, the scope of the DMs does not seem to be the same in both cases. In (4),
insomma marks the subsequent clause as a near paraphrase, as a formulation alternative
for the preceding portion of discourse. In (5), insomma marks the preceding clause (non
importa) as a closing statement for the preceding portion of discourse. The following
clause (per cui mi leggerò questo qui) is anaphorically and thematically connected to the
preceding discourse, but it is not a formulation alternative. It makes a new point. This
means that in (5), the ensuing portion of discourse is not in the scope of insomma,
whereas in (4) it is. This difference in scope apparently does not covary with grammatical
features of the constructions they are part of. In both examples, the immediate syntactic
environment is analogous, but the scope of the markers is not. It seems therefore that the
scope of DMs makes reference to discourse, not to grammar. This is a highly distinctive
feature of that word class. Note that even modal particles, which otherwise share many
features of DMs, have a grammatically determined scope (the clause they are part of).
This is all the more so for conjunctions and sentence adverbs, which have also been
recognized by some scholars as word classes functionally akin to DMs.
Given that the scope of a DM is highly variable and subject to discourse considerations,
the syntagmatic sequence that can be considered their host unit cannot be determined in
grammatical terms. As argued convincingly by Hansen (1998a: 113-128), formal units
like the clause, the sentence, or even the turn are not the adequate levels of analysis for the
units of discourse. The plausible segmentation of discourse can be achieved only with
reference to function-based or even action-based units. It is at this level that discourse
markers operate, and it is at this level they are integrated in a host unit. As seen in
examples (4) and (5), discourse markers cannot be said to be integrated in their
surrounding syntactic structure. Of course, discourse markers do have a syntax in a broad
Peircean sense, i.e., their description must somehow make reference to the relation they
sustain with other linguistic signs in the flow of discourse. But they do not seem to belong
66 R. Waltereit
to the syntax as syntacticians would probably understand the term. That is, given their
scope variability and their apparent lack of sensitivity to grammatical host units, they do
not regularly form immediate constituents with some other syntactically defined co-
constituent.
3. Model: discourse markers arise as a consequence of speakers’ strategies related to

the structure of the discourse or the interaction
The main claim I would like to make in this chapter is that discourse markers are
historical relics of speakers’ strategies for manipulating the structure of the discourse or
the interaction. Speakers discover that some word forms may have a certain appeal for
textual and interpersonal purposes. They then start to employ these forms in
communicative contexts that do not properly justify their primary use. Hearers will then
discover that the form is being used “abusively”, thereby reanalyzing it as a discourse
marker. Many of the properties of this new discourse marker can be shown to be related to
the initial rhetorical strategies. In this section I will proceed to a detailed explanation of
this claim. The Italian marker diciamo will serve as an example.
Note however that the new discourse marker does not yet have the full functional
spectrum typically associated with discourse markers. Further steps of historical change
are necessary in order to enrich its functional range. I will deal with this problem in
section 4.
3.1. Stage 0: the rhetorical potential of the imperative diciamo still veiled
Before diciamo is recruited as a discourse marker, it exists for some time as a simple verb
form. It is, however, important to recognize that at that stage the form already has a
certain rhetorical potential whose effects may be observed in communication. Consider
example (6):
(6) B: dalle organizzazioni di protesta che cercavano di organizzare il il

movimento popolare per arrivare poi a rovesciare
A: DICIAMO che la protesta e' stata una protesta pacifica # del cinque
B sì
A: non pacifica la reazione dove?
B: of the organizations of protest who tried to organize the popular movement
in order to arrive and then to reverse
A: diciamo that the first protest was a peaceful protest the one of 1905
B: yes
A: where was the reaction not peaceful?
A and B are talking about the 1905 Russian Revolution. A begins her turn with diciamo,
which in this case is most likely to be a (matrix) verb form because it is followed by the
complementizer che ‘that’. B confirms A’s assertion with sì ‘yes’. The interesting point
here is that B’s confirmation is probably elicitated by A’s previous turn. By introducing
her turn with the plural verb form diciamo, A marks her turn as a joint formulation of
speaker and addressee. This has a reinforcing and at the same time attenuating effect. The
formulation with the imperative diciamo presupposes mutual consent of speaker and
addressee with respect to the propositional content of the utterance, which gives it a
reinforcing effect. Nonetheless, it was uttered by A alone. By marking her formulation as
joint work of speaker and addressee she indirectly invites the addressee to specify whether
he agrees or not with this monopolization and consequently with the assertion itself. (In
the example, the addressee does agree.) This can be characterized as an attenuating effect
attached to the imperative/subjunctive plural verb form diciamo.
Diciamo has hence a strong appeal both for the structure of the discourse (it highlights the
ensuing complement clause) and for the speaker-hearer relationship (it requests a
reaction). Note however that the implicatures underlying these effects are not attached to a
discourse marker. They are attached to tokens of diciamo that are in perfect compliance
with the grammar of the imperative/subjunctive verb form. Moreover, nothing entitles us
to assume that the speaker was specifically seeking these effects. One can assume that the
speaker introduced her utterance in that way maybe because she felt that she and the
addressee would in fact agree on the pacific character of the upheaval or that she was
simply resuming what she thought the other one wanted to say anyway.
3.2. Stage 1: initial overuse of the verb form diciamo
The history of the DM diciamo really starts off only when speakers begin to discover the
rhetorical potential inherent in the verb form and start saying it with the sole purpose of
attaining the rhetorical effect described in section 3.1. This is, then, an “abusive” usage of
diciamo. An example for this kind of context seems to be (7):
(7) B: ecco bene <?> lei era sul passaggio anche ? ecco pensiamo a questo
elettricista che non era del reparto # ma qualcuno lo ha avvisato # della
pericolosita'?
A: mah DICIAMO che li' era un passaggio dove_ si passava tutti passava anche
venti persone il giorno quando facevano_ eh i lavori
B: DM DM you were at that place too DM think of that electrician who didn’t
belong to the department DM did someone warn him about the danger?
A: DM diciamo it was a place everyone passed by even 20 persons the day they
did the works
A and B are discussing the working safety of a certain construction site where a serious
accident occurred. An electrician was hit by a falling load. B asks A whether the injured
worker had been previously informed about the danger of that particular place. A gives an
elusive answer to this critical question: she says that many people passed by there, thereby
suggesting that either the worker probably had been informed by one of the others or that
the place could not have been previously considered as dangerous. A introduces this
answer with diciamo che, which again complies with the grammar of the imperative/
subjunctive verb form because it is followed by a complement clause. But semantically
diciamo cannot be considered an imperative verb form because the turn it introduces is an
answer. An exhortation to jointly formulate some affirmation is not a felicitous answer to
a question. When A introduces her answer that way, she is probably targeting the
implicature conveyed by diciamo, namely, that the complement clause is presented as a
joint formulation of speaker and addressee, thereby reducing the speaker’s responsibility
68 R. Waltereit
for it and at the same time presupposing the addressee’s consent. In the particular context
of (7), this is visibly a desirable effect, given that the preceding question suggests that the
accident could have been prevented and that A might be responsible for this. The speaker
does not say diciamo because she wants to exhort the addressee to subscribe to some point
of view she is suggesting but because she wants to attain the secondary effect attached to
this imperative. In other words, she means something different to what she says. In gestalt
terms, this is a figure-ground shift, as illustrated in Table 1:
Table 1. Figure-ground shift for (7)
diciamo LET US SAY CONSENT / SHARED

RESPONSABILITY
(thing said) “let us say” figure ground

thing meant Consent / shared ground figure
responsibility
Contexts like those in (7) represent innovative uses of the imperative diciamo. These
“abusive” usages of the imperative are likely to happen. It does indeed frequently occur
that speakers wish to presuppose consent with respect to what they say and that they do
not wish to bear alone the responsibility for what they say. The imperative diciamo is a
simple and efficient means to accomplish this need. It has an inherent rhetorical potential.
Therefore it lends itself to be employed “abusively”, as in (7). By abusively I mean that
the speaker manifestly does not want to exhort the addressee to some joint formulation.
She does not want to say “let us say”. She utters diciamo not because the encoded
meaning of this form matches her communicative intentions but because the implicature
attached to the encoded meaning matches her communicative intentions.
Contexts like (7) do not, however, by themselves represent a language change. On the
contrary, the efficiency of the rhetorical strategy is guaranteed only if the addressee takes
diciamo at face value, i.e., if he understands it as a true imperative. A language change is
achieved only in a second step, which is dealt with in the next subsection.
3.3. Stage 2: reanalysis of the new usage as a discourse marker
The semantic change of diciamo is accomplished only when hearers correctly understand
that the imperative is not what the speaker wanted to convey. With increasing frequency
of the strategic use of the imperative, it becomes more and more likely that people (in
their role as addressees) will understand: “The speaker didn’t want to exhort me to
something, he or she simply wanted to share the responsibility for what he or she said”
(see also Hansen, this volume, section 5). This amounts to a reanalysis of the verb form,
which is reciprocal to the speaker’s strategy presented in the previous subsection:
Table 2. Verb-form reanalysis
Diciamo LET US SAY CONSENT / SHARED

RESPONSIBILITY
(thing heard) “let us say” figure ground

thing understood Consent / shared ground figure
responsibility
From now on, the form diciamo has a new meaning: “presupposing consent and sharing
responsibility for the ensuing portion of discourse, thereby attenuating one’s own
communicative act”. That is, what used to be an implicature attached to the old imperative
meaning is now itself conventionalized as a full-fledged meaning, according to the
conventionalization of conversational implicatures as described in Traugott and König
(1991). An immediate consequence of the reanalysis is that, in this particular usage,
diciamo cannot be understood as a verb form any more. The meaning of the reanalyzed
form as described in Table 2 is not of a format that can be encoded by verbs. It is difficult
to imagine this meaning being represented by forms that inflect for person, tense, mood,
and so on. It is not a propositional meaning but a meaning that directly makes reference to
the speaker-addressee relationship and indirectly to the structure of the discourse (by
attenuating the ensuing portion of discourse). An immediate consequence of the semantic
reanalysis of diciamo is that this form, in some contexts, acquires the status of a discourse
marker. Furthermore, the scope of diciamo changes. Whereas in an imperative/subjunctive
construction diciamo che + [subordinate clause] the scope of the matrix verb diciamo is
grammatically restricted to the subordinate clause, the scope of diciamo in (7) does not
seem to range only over the following clause li' era un passaggio dove si passava tutti but
also over the next unit passava anche venti persone il giorno quando facevano_ eh i
lavori, because this latter unit is simply a semantic specification of the previous clause.
Hence the scope of diciamo in (7) can be described only by making reference to the
structure of the discourse, not in grammatical terms.
Note that the reanalysis of the verb form does by no means eliminate the imperative use of
diciamo. Rather, after reanalysis, diciamo has acquired an additional function. Not only is
it a present indicative and a subjunctive/imperative form of the verb dire ‘to say’, but also
a discourse marker. It has become polysemous: one form has several functions that cannot
be mechanically derived from each other but which have to be separately noted in the
lexicon. Furthermore, this argumentation allows a straightforward account of the relation
between the discourse marker and its homophonous counterpart, the “particle lexeme”.
We now have an answer to the question as to why DMs arise in the first place. In order to
attain frequently occurring communicative goals, speakers employ words or sequences
which lend themselves to the achievement of these goals. However, they employ them in
contexts that are not adequate for the primary function of these words. For example, they
use the imperative diciamo without really wanting to exhort their addressee to some joint
formulation. Reanalysis by addressees of this “abusive” usage turns these words into
discourse markers. It is in this sense that discourse markers are relics of speakers’
strategies.
70 R. Waltereit
Assuming that discourse markers arise as sketched in this section, we still have not yet
accounted for the functional variety discourse markers typically display. The reanalysis
described here accounts for only one function of the marker diciamo. The variety of their
functional spectrum is the concern of the following section.
4. Functional spectrum
Discourse markers typically have a variety of functions rather than one single discourse-
related function. In the case of diciamo, at least the following three uses can be
functionally differentiated (Bazzanella, 1995: 250, her examples):
(8) Attenuation / politeness:

Visto che sei, DICIAMO, in pensione, ma continui a dare una mano, non so, a
persone che te lo chiedono ...
Given that you are diciamo retired, but that you still help out DM people who ask
you . . .
(9) Reformulation:
La funzione di magistrato dà, DICIAMO conferisce un certo potere che non deve
però entrare in conflitto con altri poteri …
The function of a judge gives diciamo confers a certain power that must not be in
conflict with other powers . . .
(10) Speech management (hesitation phenomenon):

Si possono eh DICIAMO avere molte varianti.
There can be DM diciamo many variants.
The interplay of overuse and reanalysis described in section 3 enabled us to account for
the change from the imperative verb form to a discourse marker with an attenuation
function. But now we are facing the following additional problems:
1. The DM diciamo may occur in distributional environments that are inaccessible to its
imperative reading: (8), (9), (10).
2. The DM has interpretations that cannot be subsumed under the attenuation function
described in section 3: (9), (10).
These problems are considered in the present section.
4.1. New distributional contexts
The imperative diciamo need not necessarily be followed by a subordinating che. Che
may be missing with direct speech (11) or with direct objects designating parole entities
(12):
(11) Diciamo no! ‘Let’s say no!’
(12) Diciamo la verità agli italiani! ‘Let’s tell the truth to the Italians!’
But in even in such cases, the imperative diciamo is always followed by a direct object.
Now, when reanalysis to a discourse marker occurs, the following sequence is no longer a
direct object (because diciamo is no longer a governor). Consequently, its distribution is
no longer commanded by a syntactic head-dependent relationship but by a relationship
which makes reference primarily to discourse-related and interpersonal communicative
intentions. This means that once reanalyzed as a discourse marker, it is no longer subject
to the distributional restrictions of verbs and it can show up in contexts that exclude a
verbal interpretation. As with other grammatical reanalyses, reanalyzed elements can be
used in new contexts that are suited to their new function but which exclude the previous
interpretation. This phenomenon is called actualization by Timberlake (1977). For
example, (8) shows a context where the attenuation function of diciamo applies to a
segment of discourse that would exclude an imperative interpretation. As the speaker
apparently wanted to attenuate only the phrase in pensione ‘retired’, she places diciamo
just before this phrase. This position is incompatible with the verbal interpretation. Given
that in principle any portion of discourse may constitute the scope of what the speaker
wants to attenuate, diciamo can, in principle, show up at any place in the utterance. This
side effect of the attenuation function is of great importance for the semantic enrichment
of the marker.
4.2. New functions
I would like to claim that the same mechanism of overuse and reanalysis which recruited
diciamo to the class of discourse markers is also applicable to the extension of its
functional range. That is, the discourse marker diciamo with an attenuation function can in
turn be employed “abusively” in order to meet other, frequently occurring communicative
needs. For example, a speaker who is unsure as to how to continue his or her current turn
may insert diciamo just in order to gain some time for formulation. The marker diciamo is
very useful for this because it can in principle be inserted at any place in the utterance.
When thinking of a word that may fill a pause, discourse markers like diciamo will
rapidly spring to mind because they can be inserted anywhere without constraining the
subsequent formulation. This means that the speaker will not say diciamo in order to
attenuate the following portion of discourse (the encoded meaning of this item) but just to
delay the formulation. The speaker means something different to what he or she says.
Hence, we are in the presence of the same overuse strategy that first gave rise to the
discourse marker use. As hearers uncover this strategy, they will also reanalyze the
attenuating discourse marker as a speech-management marker. Concomitantly, diciamo
will have acquired an additional function, which has to be encoded in its lexical
representation. It is important to stress that this must have happened later in the history of
Italian than the recruitment of diciamo to the class of discourse markers in the first place
(although not much time need have elapsed between the two diachronic steps).
There is a long-standing debate as to whether the functional spectrum of DMs can be

shown to derive from one and the same underlying semantic representation (monosemy)
or whether the observed functional variety corresponds to a variety of functions also at the
underlying level of lexical representation (polysemy). (See Hansen, 1998a: 85-90 and
Fischer, 2000a.) It follows from the preceding discussion that I am favoring a polysemy
approach. Only with a polysemy approach are we able to account for the rise of the
various functions in terms of discrete historical steps. Moreover, the model presented here
gives additional plausibility for a polysemy analysis. Under a monosemy account, the
72 R. Waltereit
semantic description would not be able to accommodate semantic changes of the type
sketched here. It would be very difficult to explain that some uses show up only later than
other ones, given that all of them should be equally derivable from the same underlying
semantic representation.
4.3. Clues for a relative chronology of the functions
Thus far, my historic argumentation has been supported solely by data from contemporary
Italian. Strictly speaking, the examples discussed are not able to prove language change in
the first place. Given that the initial stage (diciamo as 1PL imperative/subjunctive) is still
present in the Italian language, they can at best give a plausible picture of how things
might have changed. However, an inspection of diachronic data, the LIZ corpus for
example, might shed some light on the actual historical evolution. As explained in section
1, these data are from literary texts, and they have to be taken with all sorts of
methodological caution. For example, one would not expect to find examples of the very
oral speech-management use of diciamo in literary texts, except in very modern ones that
deliberately make use of simulated orality.
As expected, the imperative use of diciamo is the oldest one:2
(13) C. Goldoni, La famiglia dell’antiquario

\DORAL.\ Ma mia suocera cosa dirà?
\COL.\ Questo è il punto. Cosa dirà.
\DORAL.\ Trovaremo la maniera di farglielo sapere. Per oggi non le DICIAMO
nulla.
\COL.\ Benissimo, farò quello, che comanda Vusignoria illustrissima.
D: But what will my mother-in-law say?
C: That’s the point. What will she say.
D: We will find the right way to let her know. For today let’s not say
anything.
C: Very well, I’ll do what Your Highness commands.
The following example from an early 19th century text seems to be first one attested in
the corpus where diciamo has a DM use:3
(14) V.Cuoco, Saggio sulla rivoluzione a Napoli del 1799, Cap. 18 §6:
E fin qui il popolo francese fece sempre operazioni al livello, DICIAMO così, delle
sue idee.
Until then, the French people always carried out operations that were at the level
diciamo of their ideas.
This is an attenuating use, analogous to (8).
Examples of the formulation-related use of diciamo show up only in 20th century texts:
(15) I. Svevo, Racconti, Umbertino

Bisogna permettere a qualunque dolorante la soddisfazione, DICIAMO pure la gioia,
di esaltare il proprio dolore.
We have to allow to every suffering person the satisfaction, diciamo even the joy,
of celebrating their pain.
The following is an example of the speech-management use of diciamo in a dialogue

context. This function, too, shows up only in 20th century texts:
(16) L. Pirandello, L’esclusa, 1,11, §10

Non faccio . . . DICIAMO piacere, ma . . . la Giustizia comanda, noi portiamo il
gamellino.
I don’t cause diciamo pleasure, but . . . Justice commands, we bring the little
bowl.
I should not like to claim that the evidence presented here allows deciding upon the
relative chronology of the speech-management and the reformulation uses. But it does
enable us to say that the DM uses are more recent than the lexeme use of diciamo, and it
shows that the attenuation use is older than the formulation-related uses. The historic data
offer evidence to support the pragmatic argumentation presented in this and the preceding
section.
5.1. The rise of discourse markers is not grammaticalization
I would now like to address the relationship between the rise of discourse markers and
other types of language change, especially grammaticalization. In most recent work, it is
assumed that the rise of discourse markers is an instance of grammaticalization. As argued
in Waltereit (2002), however, a simple check of Lehmann’s grammaticalization
parameters (1985: 306) strongly suggests this cannot be the case:
Table 3. Grammaticalization parameters
paradigmatic syntagmatic
weight integrity scope
(attrition) (condensation)
cohesion paradigmaticity bondedness (coalescence)
(paradigmaticization)
variability paradigmatic variability syntagmatic variability (fixation)
(obligatorification)
Lehmann defines grammaticalization as the gradual loss of autonomy of a linguistic sign

as measured by these six parameters. I would now like to verify whether the rise of
discourse markers can be said to be grammaticalization in Lehmann’s sense.
74 R. Waltereit
Integrity: Discourse markers may indeed “suffer” phonological alterations and semantic
changes of the type Lehmann had in mind when describing grammaticalization processes,
i.e., they may undergo attrition. This is clearly shown in Brinton (1996) for Middle
English. Italian diciamo, however, is as yet not a case in point, but e.g., va be’ (from va
bene ‘O.K.’) is.
Paradigmaticization: As shown by Hansen (1998a: 69-70), markers do not enter a

paradigm in the classical sense of that term (see also section 2.1 above). Hence, one
cannot rightly speak of paradigmaticization.
Obligatorification: It is a crucial property of DMs that they do not become obligatory, i.e.,
their presence is never enforced by some other linguistic sign.
Condensation: As observed in section 2.2, the scope of DMs does not shrink. On the
contrary, they may have a much larger scope than the items they arise from. There is no
condensation.
Coalescence: DMs are always free morphemes with respect to their adjacent elements,
hence they do not undergo coalescence.
Fixation: DMs do not acquire fixed positions in the sentence. On the contrary, they can
occupy a large variety of syntagmatic slots.
So, attrition is the only parameter DMs comply with. This rather poor score should
exclude them from grammaticalization. They are recruited in a different, specific historic
process. Some authors have referred to this as “pragmaticalization” (see discussion in
Diewald, this volume). The point of this distinction is not only a mere taxonomic
classification of types of language change. Rather, it seems important to me to try to relate
the differences in diachronic recruitment between DMs and grammatical items to their
respective synchronic differences.
What the rise of DMs does however have in common with grammaticalization is that both
originate in speakers’ strategies (Detges, 2000a, 2000b). Whereas the source constructions
for discourse markers have an appeal for the structure of the discourse or for interaction,
the source constructions for grammaticalization have an appeal related to the proposition.
For example, in many languages, new future tenses arise from verbs meaning ‘to go’
(French, English, Spanish) (see Detges, 2001: 147-187, for the following argumentation).
The appeal of this verb lies in the fact that a speaker saying, e.g., I’m going to buy some
milk is able to underscore his or her willingness to do the future action in a far more
effective way than a speaker saying simply I will buy some milk. The use of the verb to go
implicates that the speaker is already about to carry out the future action, because it is
actually necessary to go somewhere else in order to buy the milk. If the future action does
not require some actual movement, the verb to go would represent an abusive form of this
strategy, which can then be reanalyzed by listeners as a future marker.
5.2. The “special status” of discourse markers reconsidered
Only quite recently have DMs been considered objects of study in their own right. They
seem to be a special word class which is in many respects very different from other word
classes. But under the model outlined in this chapter, many of their seemingly peculiar
properties can be related to independently motivated assumptions about discourse,
grammar, and the lexicon.
DMs are special in that they are nonpropositional, nonobligatory, and variable in scope.
All of these features can be related to their recruitment process as outlined in this paper.
As I have argued, I take discourse markers to be historical relics of speakers’ strategies
related to the structure of the discourse or to interaction. When trying to account in a
diachronic perspective for the features of DMs, we will at first tend to look for
predecessors of these features in the source constructions. In fact, many DMs stem from
words or constructions that already have some properties typical for DMs. For example,
many discourse markers stem from constructions that can constitute an utterance by
themselves. The imperative diciamo is, if followed by direct speech, a syntactically short
but complete sentence. The same holds for the imperative guarda ‘look’ (Waltereit, in
press). Their syntactic independence entails that their immediate environment is
controlled not by syntactic structure but by discourse structure. The “scope” of the source
construction cannot be controlled by syntactic structure, either. It corresponds to the size
of the portion of discourse which is the object of the speaker’s intention and is not
grammar related, but discourse related. This property is preserved after reanalysis of the
construction as a discourse marker.
The very fact that the source constructions are often syntactically independent furthermore
suggests that their diachronic offspring will be nonobligatory and nonpropositional.
Reanalysis of the source constructions as discourse markers makes the former lose their
status as a complete unit and integrates them into the host utterance of the by-now DM
item. But the syntactic connections with their environment remain as loose as before. The
DM will not be integrated into a structure of sentence grammar. On the formal level, this
implies nonobligatoriness; on the content level, this implies nonpropositionality.
Obligatoriness means that the presence of some element x is enforced by some other
element y. Such a relation can hold only in a structure of sentence grammar, which is by
definition excluded for discourse markers. Furthermore, their not being integrated in a
structure of sentence grammar implies that they cannot contribute to the semantic
structure of the proposition. This confines them to textual and interpersonal meanings.
Note, however, that the diachronic sources of DMs need not be syntactically independent.
For example, even full nouns can serve as a source for DMs. This is the case, for example,
with the French noun truc ‘thing’. The word truc (like its English counterpart) is virtually
devoid of semantic content. In fact, its only semantic feature seems to be [–animate]. As
such, it is typically used to refer to entities whose precise designation the speaker does not
know. In such cases, it is a full noun and plainly syntactically integrated. But it can also
be used as a speech-management signal:
(17) Corpus ELICOP

A: à Paris . toujours la même suite dans les questions . quel genre de pièces?
quels auteurs?
B: je sais pas mm . bon ben ça je euh en général je je regarde un peu sur les
journaux les critiques . oui oui les pièces horribles les pièces noires euh
TRUC euh . . .
B: oui . qu’est-ce que c’est pour vous une pièce horrible ?
76 R. Waltereit
A: In Paris. Always the same order in the questions: what kind of plays? what
authors?
B: I don’t know DM DM DM In general DM I look up the papers a little bit,
the reviews. Yes yes the horrible plays, the black plays. DM truc DM
A: And what is that for you, a horrible play?
Truc can serve as a DM because it has already in its full noun use the side effect of
delaying the formulation.4 It is a word that helps out when you don’t know the exact
designation, and as such it indicates that the speaker is unsure as to his or her formulation.
Speakers may take advantage of this implicature (as it were) and use that word when they
just want to gain some time for formulation, as in (17), without having an entity to refer
to, thereby using it as a DM.
The lesson to be drawn from DM uses like in (17) is that syntactic independence (or any
other property) of the source construction as such is not a requirement for recruitment to a
DM. Rather, it is the effect for the structure upon the discourse or the interaction enabled
by an item of any word class that paves the way towards its use as a DM. If anything, the
syntactic independence of the source construction may be useful for creating an
implicature related to the structure of the discourse or the interaction. Syntactic
independence of the source construction may favor reanalysis to a discourse marker, but it
is not a necessary condition for such a reanalysis. Once recruited as a discourse marker
the element in question is nonobligatory and nonpropositional no matter whether its
source construction was integrated in a structure of sentence grammar or not. The special
features of DMs are not direct traces of their diachronic ancestors. They are traces of the
strategic use speakers made of these ancestors in order to attain an effect related to
discourse structure.
Notes
1
Familiar address form roughly corresponding to French tu as opposed to the formal
address form Lei.
2
Example (13) is not the first example of imperative use. It was chosen for its dialogical
character.
3
An earlier offspring of the imperative is diciamo in the sense of ‘for example’, ‘say’, like
in Pensa tu stessi quali sono quegli, a fare per vil cosa ch'ella sia, DICIAMO così un
calzare, e' quali non cerchino tra quegli artefici sempre il miglior maestro “Think
yourself of those who, in order to make even the simplest thing, say a shoe, wouldn’t
look for the best craftsman available” (Leon Battista Alberti, Il libro della famiglia,
Lib. 2). It seems to me that this use does not represent a discourse marker but rather a
lexical semantic change of the source lexeme. The scope of this item can be measured
in grammatical terms (a conjunction of same-type constituents).
4
This effect, as well as the DM use of truc in the first place, was pointed out to me by
Wiltrud Mihatsch.
Chapter 5
A functional approach to the study of discourse

markers1
Salvador Pons Bordería
1. Introduction
It is now twenty-nine years since Halliday and Hasan’s (1976) Cohesion in English
highlighted the value of conjunctions as text-building devices and eighteen since
Schiffrin’s (1987) attempt to create a discourse model for the study of discourse markers.
Since then, an enormous flow of contributions has turned this field into a “growth
industry” (Fraser, 1999) which deserves a separate entry in the comprehensive Handbook
of Pragmatics by Verschueren, Östman, et al. (1995). Nevertheless, the study of discourse
markers does not constitute a unitary approach. Significant disagreements can be found
among scholars regarding the linguistic units which must be considered discourse
markers, the dimensions involved in their study, or even the distributional features of the
class. In this paper, I will outline some of the most interesting directions and problems in
the treatment of the word class of discourse markers (henceforth, DMs) and its hyponym,
connectives.2 I will also propose some guidelines for a future explanation of the functions
involved in the study of this pragmatic category.
1.1. Approach
The present approach deals with the functions that DMs perform in spoken language. It
relies upon two research programs. The first is the study of colloquial Spanish in the
German-Spanish-Italian tradition (Spitzer, 1922; Beinhauer, 1978: 1929; Sornicola, 1981;
Briz Gómez, 1998). In the eighties, this approach led to the questioning of the descriptive
syntax found in grammars (especially the definition of sentence or the classification of
subordinate sentences). The search for answers should not be based on received syntactic
concepts, because they cannot account for the structures of spoken, casual language; they
should hinge, instead, on a “new grammar” of spoken language, which is established as a
desideratum (Narbona Jiménez, 1989).
The second source is pragmatic and comprises a multitheoretical semantic-and-pragmatic

perspective that comprises Prototype or Relevance Theory as well as Conversational
Analysis or Text Linguistics. A multidimensional approach is a methodological necessity
78 S. Pons Bordería
in order to understand the polyfunctionality of DMs, as some authors have explicitly

stated (Schwenter, 1999: 245).
1.2. Methodology and data
My main interest has been the study of connectives in spoken, colloquial Spanish and is
basically corpus-driven.3 The research group Val.Es.Co (Valencia Colloquial Spanish), to
which I belong, has collected a corpus of colloquial conversations which provide the first
basis for a corpus study.4 Working with real data implies that, at least in the first instance,
an inductive methodology has been employed. The contact with real samples of language
is a healthy exercise to prevent research-based biases or artificial examples. However, one
should be cautious when determining the limits of the analysis: colloquial language is the
most productive register for nonspecific connectives (like Spanish pues, que, entonces,
etc.), but the researcher must not forget that some specialized uses of connectives cannot
be found in an informal register (see examples in Portolés Lázaro, 1998).
Description is only a first approximation. In a second phase, a quantitative analysis is a

useful tool to sharpen raw intuitions. Due to their multicharacter (multiperspective,
multifunctionality . . .), the study of connectives comprises many factors. It is unlikely
that a mere qualitative study can grasp all the relevant aspects of the association of
variables (e.g., what happens when a monologue token of an unstressed connective is in
utterance-initial position, thus conveying a topic shift, which can also be understood as a
signal of disagreement?). A multivariate statistical analysis provides us with methods to
perform dimensional reduction and clears the way to determine what (associations of)
variables are significant.
Finally, quantitative data must be subjected to a third, qualitative phase in order to place
the research in a broader theoretical framework. This was the procedure employed in Pons
Bordería (1998).
1.3. Problem statement
1.3.1. State-of-the-art
Twenty-nine years after Halliday and Hasan’s seminal work, we have learnt a great deal
about connectives. Comments on their puzzling nature, not uncommon twenty years ago,
have been replaced by explanations based on general principles and by myriads of
particular descriptions of single connectives in many languages. Among the findings of
this joint effort, I would like to highlight the (full) description of some (types of)
connectives in different, typologically unrelated languages. This implies the following:
1. The rejection of a sentential, grammatical paradigm as the basis for the description of
connectives. This also excludes formalist approaches from the mainstream and clears
the way for pragmatic, functional-based approaches, where connectives are no longer
facts of performance.
2. The recognition of distinctive functional features, making it possible to talk of textual
(Alcina Franch and Blecua, 1975; Halliday and Hasan, 1976; van Dijk, 1977; van
Dijk, 1980; Portolés Lázaro, 1998), argumentative (Ducrot et al., 1980; Ducrot, 1983;
A functional approach to the study of discourse markers 79
Briz Gómez, 1993; Anscombre and Ducrot, 1994; Ducrot, 1996), reformulative
(Gülich and Kotschi, 1983; Roulet, 1987; Rossari, 1994), or metadiscoursive markers
(Briz Gómez, 1998; Pons Bordería, 2000). This notwithstanding, there is still no
commonly-accepted grouping within the class.
The research has also helped to show what dimensions are needed to carry out a full
description of connectives. Text Linguistics highlighted the suprasentential values of
connectives and how they help to build coherence relations. Conversation Analysis
displayed the interactional aspects of their behaviour and Politeness Theory did the same
for the image of the self/others. Relevance Theory showed the cognitive role played by
connectives. French studies within and around Argumentation Theory discovered that
connectives do not merely display a relationship between propositions but also between
arguments and conclusions, explicit or otherwise. From a different perspective, neo-
Gricean developments insisted on what is encoded and what is not, what is said and what
is implicated. Finally, the diachronic carried out in grammaticalization studies established
universal clines of evolution linked to cognitive processes.
It is commonly accepted that connectives are inherently polyfunctional linguistic items.

Polyfunctionality must be read at two levels. First, at a type level, a connective is
polyfunctional if it conveys different values.5 For instance, the English conjunction but is
polyfunctional because it expresses contrast (mainly in monologic uses) and disagreement
(mainly in dialogic uses). A second reading of polyfunctionality is possible at the token
level. A token of a connective is polyfunctional if it displays different functions at
different discourse levels. For instance, in a given context, a token of English but can
express contrast at a sentence level and disagreement at an interactional level.
Although hundreds of connectives have been studied in the last twenty years, few
attempts have been made to achieve a global description of this class in a given language.
The importance of these attempts lies in the fact that a global description works as a
landmark for categorization; a full description will, in turn, help to achieve a better
understanding of how interaction among different discourse dimensions works.
Nevertheless, we only have three global descriptions: Bazzanella (1995) for Italian,
Martín Zorraquino and Portolés (1999) for Spanish, and Cuenca (2002) for Catalan, all
within comprehensive grammars. For other languages, global descriptions are not
available, although for English and French partial descriptions are available in numerous
works.
1.3.2. Problems
The current state-of-the art raises some questions regarding formal and functional aspects
of DMs. In this section, I will briefly outline the issues considered of primary interest.
Firstly is the name: the most basic disagreement among the scholars engaged in
describing DMs involves the name of the class. This is a subject of primary importance,
for it does not mean merely labeling a set of objects; it also involves a selection of items
which share a set of common properties. Discourse marker is perhaps the most
widespread concept in the literature but has been used in a broad and in a narrow sense.
Schiffrin’s well-known definition is given below:
I define [discourse] markers at a more theoretical level as members of a functional

class of verbal (and non verbal) devices which provide contextual coordinates for
ongoing talk (Schiffrin, 1987: 41).
Linguistic units such as y’know, well, or then will all belong to the same class. The range
of functions conveyed by these elements goes from the interpersonal to the textual or
modal; their scope also varies from sentence to paragraphs, from utterance segments to
turns or sequences; their syntactic or distributional features are also heterogeneous.
Nevertheless, a more restrictive definition of DMs is also possible. Fraser (this volume)
offers the following definition:
A DM is a lexical expression which signals the relationship of either

Contrast (John is fat but Mary is thin)
Implication (John is here so we can start the party); or
Elaboration (John went home. Furthermore, he took his toys.)
between the interpretation of S2 and the interpretation of S1.
In his version, the class of DMs comprises only linguistic items which serve to bind
discourse segments together. This will include sentential and textual markers like then or
so but not interpersonal markers like hear or you know. Hence, before proceeding, it is
necessary to make explicit the conception of DMs to which the researcher adheres.
The broad/narrow opposition can be formulated in the following way: are we aiming to
describe a broad functional word class whose only common feature is not to be included
in the syntactic and/or semantic structures of a sentence/proposition? If this is the case, we
will talk of discourse markers. If, on the contrary, we restrict ourselves to the study of a
subset within this group, namely, that of the items whose main function is to bind
elements together, we will talk of connectives. The first approach is commonly accepted
within U.S. linguistics. The second is more frequent in Europe, where the weight of
traditional grammars led scholars to focus on connection as the supra-, infra-, or
parasentential equivalent of sentential coordination and subordination.
In what follows, we shall distinguish two related but distinct concepts. On the one hand,
connectives are linguistic items whose primary (or prototypical) function is to convey
union relations between two linguistic items at the infra-, extra-, or intersentential level.
On the other hand, discourse marker is a global label for most elements without
propositional meaning, thus including connectives (like and, therefore, or but), modal
markers (like the German Modalpartikeln doch, eben, or mal) or interactional markers
(like say, hear, or phatic expressions) and some elements with propositional meaning (like
as a consequence or as a result). I consider that discourse markers are hyperonyms of
connectives; therefore, there is an entailment relationship between them whereby every
connective is a DM but not every DM is a connective.6 One of the tasks in the description
of DMs will thus be to draw an accurate picture of its hyponyms.7
A theory of connectives should be able to link grammatical with nongrammatical uses, in

order to provide a unitary description of each item. This is an important issue in Romance
languages, because it deals with three important syntactic problems (Narbona Jiménez,
1989a, 1990): the accuracy of grammatical descriptions for conjunctions, the definition of
sentence, and the classification of coordinate and subordinate sentences.
Although the formal features of a connective cannot be taken as a basis to predict a set of
functions, not even in morphologically related elements, (Rossari, 1994; Ruiz Gurillo and
Pons Bordería, 1996) there must be a limit to the number of functions that a connective
can convey.
An answer to the previous problem can be provided by a twofold cognitive approach.

Synchronically, the functions of a connective are limited by the family resemblances it
entertains with other categories. Diachronically, grammaticalization studies show which
functions a connective has effectively performed. Grammaticalization processes seem also
crucial because they provide answers to more general theoretical problems, like the
conventionalization of implicatures (Traugott, 1999a) and the adequacy of neo-Gricean
pragmatics vs. Relevance Theory (Levinson, 2000) (see section 4).
The most influential research on connectives has been carried out on written or intuitive
language samples (Halliday and Hasan, 1976; van Dijk, 1980; Ducrot et al., 1980;
Blakemore, 1987; Fraser, 1990).8 However, if one takes into account spoken language, the
relevance of some widely accepted features can, on occasion, be challenged. We are going
to exemplify this with the “parenthetical” feature. It has been repeatedly said that some
connectives are not integrated into the intonational contour of the utterance to their right.
This is represented in written texts through commas, so a conclusion which is drawn on
observing written language is that a parenthetical intonational contour is a feature of many
DMs. Nonetheless, after an analysis of spoken samples containing the Spanish DM bueno,
which is considered parenthetical (Martín Zorraquino and Portolés, 1999: 4064), the data
show (Hidalgo and Pons, 2001a) that bueno is prototypically followed not by a pause but
by a falling pitch. This relationship is significant, as shown below.
Table 1: Relationship between pause and pitch (Hidalgo and Pons, 2001a)
Pause No pause
Rising pitch 0.9% 1.2%
Falling pitch 94.0% 4.8%
p < .0001
The most evident division in connection with variation may perhaps be that between
written and spoken language. One can expect different discourse genres to shed light on
more subtile preferences of use.
The description of connectives has been pushed forward by qualitative, rather than
quantitative analyses. At this point, the introduction of statistical methods in qualitative
studies is highly desirable because it provides the researcher with validation methods to
determine significant correlations among qualitative features of connectives.
A disregarded aspect is the lexicographical description of discourse markers. The classical

lexicographical distinction between words with lexical versus grammatical meaning
leaves discourse markers in the second group, as words that can be described, not defined.
One of the few current projects, although far from finished, is the Spanish Diccionario de
partículas del español (Briz Gómez, Hidalgo Navarro, et al., forthcoming). For the class
of modal particles, some dictionaries are available, such as Helbig (1988) and Helbig and
Helbig (1990).
2. Definition9
2.1. The issues
Defining a connective is a rather controversial issue which raises at least the following
questions. Is the definition based on necessary and sufficient conditions, or does it adopt a
prototype-based approach? Similarly, is the definition of connectives based on form or
function? Very closely related to these questions is the issue of the monosemic,
polysemic, or homonymical meaning of connectives. All these concerns will be addressed
in this section.
The easiest way to define what constitutes the word class of connectives is in terms of
sufficient and necessary conditions. This procedure, which creates two disjunct sets,
connectives and nonconnectives, has traditionally been adopted and is perhaps the most
familiar for the reader/researcher. However, a definition in terms of necessary and
sufficient conditions poses certain problems in its application, because it relies on the idea
that there is a one-to-one relationship between form and function. X is a connective means
X is only a connective or, in the case of polysemy, X n is a connective = Xn is only a
connective. Nevertheless, what the description of connectives shows is that monosemy is
the exception and polysemy is the rule. Underlying each meaning there is a great deal of
functional variation. For example, when then functions in a temporal domain, it is either
an adverb or a conjunction; whereas when it works in a metalinguistic domain, it is
normally a conjunction or a connective. A definition in terms of necessary and sufficient
conditions forces the researcher to decide whether a given item is a connective or a
conjunction; therefore, it is not flexible enough to deal with this kind of case.
The alternative I propose is a prototype-based definition. In this approach, the category

has a core and a periphery. An item assumes different degrees of connectivity depending
on its closeness to the core or periphery of the prototype. Along these lines, the following
conclusions can be reached. A connective is a cluster of features. This can be defined in
terms of a prototype (Pons Bordería, 1998; Fraser, this volume; Hansen, this volume). The
degree to which certain features are present in a given unit will determine whether it is
close to the center or periphery of the category.
2.2. Guidelines for studying connectives
Adopting this theoretical approach enables us to present the following guidelines for the
study of connectives.
1. Connectives are a pragmatic category, that is to say, what all connectives have in
common is not the grammatical class to which they belong but their ability to signal a
relationship between two units.
2. One of the most striking features of connectives is that the core of the category is made
up of two clusters of features. Some authors argue that prototypical connectives are
stressed units with a parenthetical intonation contour and reduced lexical meaning.
Other authors hold that prototypical connectives are unstressed elements included in
the intonational contour of their host utterance, monosyllabic and semantically void of
any meaning (except for so-called morphological meaning). Both groups include
prototypical connectives: elements like therefore in the first case, units like and in the
second case. Rather than excluding any group from the core of the category, it is better
to consider connection as a category with two cores, so both groups can be studied as
prototypical connectives. This unexpected conclusion can be better justified if we
consider that continuative conjunctions (Spanish por tanto, por consiguiente)
constitute a more specialized means of displaying connection than other more basic
conjunctions. This implies that continuative conjunctions are more complex from a
morphological point of view, more specialized syntactically speaking, more
semantically restricted and poorer from a pragmatic perspective. A possible
explanation could be that continuative conjunctions were included later in the
grammar, through grammaticalization processes as suggested by empirical studies (see
Traugott, 1995a; Brinton, 1996).
3. Centrality within the category can be defined as the ability to link at the sentence/
utterance level, whereas peripherality is best defined as the lack of this ability. Thus,
the more peripheral a connective, the less likely it is to link at the sentence/utterance
level. However, the connective function is also found at a microsentential level or
formulation (see section 3.2.3) and at a macrosentential or textual level (which will be
defined as delimitation; see section 3.2.3). This indicates that there is a strong
tendency for peripheral connectives to be used for solving planning problems or
conveying topic shifts rather than joining utterances. Between these two opposite ends,
the core and the periphery, is an area where most connectives find their place. This
was the approach adopted in an empirical analysis of eleven connectives in spoken
Spanish in Pons Bordería (1998).
4. In Fig. 1 below, similarities and dissimilarities are transformed into distances; the
closer two connectives are to each other, the more alike they are.
5. The corollary of a prototype approach is that the researcher must not rely only on
formal features to decide if a given element is a connective or not nor expect a one-to-
one relationship between forms and functions. The connective value of an item should
first be functionally established in each occurrence, taking formal features as auxiliary
criteria. This is not to say that a semasiological approach has to be discarded; much
has been learnt through the study of single connectives. Instead, a new direction based
on functions is proposed, which contrasts with—but is not opposed to—the formal
approach. The consequences of this approach for the monosemy/polysemy/
homonymy distinction will be studied at the end of section 3.
Figure 1. Results of a correspondence analysis on eleven connectives (Pons 1998):

|--+---+----+------+------+-----+-----+-----+-----+-----|
.55+CLA +
| |
| |
| |
| BUENO |
.33+ O SEA +
| |
| |
| |
| |
.11+ +
| PER |
| Y |
| * O |
| QUE/PUES |
D -.11+ +
i | |
m | ENT |
e | |
n | |
s -.33+ +
i | |
o | |
n | |
| |
3 -.55+ +
| |
| |
| |
| MIRA |
-.77+ +
| |
| |
| |
| |
-.99+ OYE +
| |
| |
| |
| |
-1.20+ +
|--+---+----+-----+-----+----+----+----+-----+-----+---|
-1.47 -1.31 -1.15 -.99 -.83 -.67 -.51 -.36 -.20 -.04 .12 .28
Dimension 1
3. Functional spectrum: towards a function-based explanation
3.1. Relationship between form and function
Connectives maintain a special and difficult relationship between forms and functions. On
the one hand, connectives are elements which mostly belong to a word class and, on the
other, their grouping is based on their capacity to perform a common function, that of
connecting elements. Hence, the study of connectives can be carried out either from the
categorial perspective (description of a specific connective and the way it performs the
connective function) or from the functional perspective (description of the connective
function and of the particular instances which create it or display it). The picture is, in
fact, more complicated than an onomasiological/semasiological approach would suggest.
Let us consider the first path. Most connectives belong to well-established grammatical
categories, especially conjunctions and adverbs. When we say that then is a connective,
what we mean is that, besides being an adverb, some of its occurrences work as
connectives; then is polysemous. So when we say that entonces, alors, allora, etc. are also
connectives, what exactly do we mean? We mean that they belong simultaneously to two
or more categories—something not infrequent in sentence grammar, where double
categorial assignments are not uncommon. But where does the basis lie for this new word
class called connective? It is not found in its grammatical but in its functional or
pragmatic behaviour. Hence, the basis of the categorization process seems to rest on
functional grounds.
One could think of formal features as a valid resource for the categorization process.
However, formal features are risky and are better applied within a family resemblance
framework. Take first position for instance. Many of the contributors to this volume
would recognize that the prototypical position of DMs is the first position. But what does
first position mean? Is it the first word of an utterance, the first syntagmatic position of an
utterance, or the first slot outside the proposition within the utterance? Describing a DM
as utterance-initial may obscure some of its distributional properties if one cannot be more
explicit about its meaning.
Another disregarded problem in the description of connectives is that the initial purpose
of describing the connective uses of X sometimes turns into something more akin to
describing the uses of X. However, given the intrinsic multifunctionality of connectives,
their description occasionally obscures their connective values and two kinds of functions
are distinguished: grammatical functions and discourse marker functions. The latter may
not only express connection but other pragmatic functions, modality, for instance.
Hence, in the categorization process, a double reduction may occur in terms of the
type/token distinction pointed out above. When a semasiological approach is adopted,
caution must be taken in delimitating different pragmatic categories.
Turning to a second, functional alternative, we could imagine a function called

connection, displayed by elements from different origins; to the extent that they display
this function, they can be called connectives. The main point now is to characterize the
function as a gradient category: the more an element accomplishing the function of
connection occurs, the more central that connective is; the more prototypical the
functions, the greater its centrality. In this approach, no one-to-one relationship needs to
be assumed between categories/occurrences and functions. Nevertheless, this approach
does not overcome all the problems in characterizing connectives; it simply poses new
ones.
3.2. The issues reconsidered
The first problem with this new account lies in the fact that we do not have a clear picture
of the connective function of the cohyponyms, hyponyms, and hyperonyms nor clear
relationships among them. As a working hypothesis, consider the following point of
departure:
Figure 2. Functions involved in the treatment of DMs

Argumentative function
Delimitation
Formulation
Discourse Connection Meta- Structuring start
marked- discoursive Regulation progression
ness function close
Reformulation
Modalization
Interactive function
In the first instance, there is a macrofunction, called discourse markedness, which

comprises different nonsyntactic, nonpropositional functions. Discourse marker is the
name given to every linguistic item which prototypically performs this function. Within
discourse markedness, three dimensions can be provisionally distinguished: interactional,
modal, and connective.
3.2.1. The interactional dimension

A conversation is a social product, the result of an interaction between two or more
participants, who start talking in order to achieve a communicative goal.10 Social
interaction leaves a trace in conversation, where different strategic signals can be found.
The interactional function, then, covers the most “external” uses of language, including
politeness, face-saving, and face-threatening uses of DMs, as well as the regulative
function of connectives (Briz Gómez, 1998: 201-230), where phatic usages or turn-system
related occurrences of DMs link language with its participants.
(1) (C is talking about people ringing up late at night)

C: mi marido pegó un bote de la cama ¡no veas! se le cayó hasta el teléfono
//¿está Jesús? dice mi marido mire se ha equivocao //
se acuesta §
A: § ¿y volvieron a llamar otra [vez? ]
C: [¡oh que] si volvieron a llamar!
C: my husband jumped out of the bed/ my God! he even dropped the phone) // is
Jesús up there? my husband says look this is the wrong number// he goes back
to bed §
A: §and did they call [again?]
C: [oh sure] sure they did!11
In (1), mire (‘look’) minimizes a dispreferred answer. Its pragmatic value is in co-
occurrence with some formal features which cannot be found when mire is a perception
verb and which identify it as a DM (it cannot take a preverbal subject, the constituent to
its right cannot be considered a direct object—#mirelo—and so on).
3.2.2. Modalization
Modalization is a label which includes the expressions of the self in the propositional
content of an utterance. In this sense, the notion of modality is similar to that of the
German Partikelforschung. The parallelism is imperfect, though, because only a few of
the major Western languages have a word class with this value. What colloquial Spanish,
and perhaps other languages, shows is that modality spreads over elements belonging to
different grammatical categories, loading them with a modal content. Thus, when
modality is not grammatically fixed (i.e., in a word class), it is lexically or pragmatically
coded (e.g., in a semantic reading or in an implicature). Consider the following example:
(2) (V and S are arguing for and against the legality of a new law discussed in the
Spanish Parliament. V argues that illegality must be established with regard to a
particular interpretation of the Spanish Constitution)
V: [¿estás o no estás?]
S: [sí que lo había oído]// cuando hay §
V: § entonces§
V: [do you agree or not?]
S: [yes I knew it]// when there is §
V: § entonces (‘then’)
In (2), entonces ‘then’ is stressing the position of the speaker in a polemic exchange,
functioning as a means to express modality.
An important question is what processes count as modal in the conception outlined above.
For the moment, consider the following four values: two are prototypically monologic,
stressing and hedging, and the other two are prototypically dialogic, agreement and
disagreement. An important issue is the definition of such elusive concepts to delimit
them more precisely. A first consideration could be that a discourse marker hedges or
stresses a discourse segment when there is a contrast between the discourse segments with
and without a DM:
(3) a. A: ¿Vas a venir al cine esta noche?

A. Are you coming to the movies tonight?
b. B: Claro que voy a ir

clear that go-PRES-1S to go
B: Sure I’m going.
c. B: Ø Voy a ir.
go PRES-1S to to-go
B: Ø [yes] I’m going
In (3) above, a discourse segment falls under the scope of the DM claro (roughly
paraphrasable by ‘well’). In cases like this, two scales of the form <weak, neutral> and
<neutral, strong> can be devised.12 The first scale defines hedging and the second defines
stress. When a DM is the only constituent in the turn (like in (2) above), the neutral
element of the scale must be contextually established.
Agreement and disagreement can be explained in terms of argumentative orientation

(Anscombre and Ducrot, 1994). Agreement signals argumentative co-orientation with a
preceding discourse segment (4), and disagreement signals argumentative antiorientation
(5).
(4) (A is talking about telephone jokes)

A: no/ pero hay que aceptarlo porque yo lo veo por- por en esto del Primijuego /
solamente con decir sí seiscientas mil pelas […] entonces aunque te tome
alguien el pelo pues dices sí
[…]
D: [claro claro] / nunca [sabes si]
A: no/ but you have to play the game because I see it in—in the Primijuego [a TV
program] / if you say yes [you win] six hundred thousand pesetas […] so even
if someone fools you well you say yes
[…]
D: [right right]/ you never [know what]
A’s conclusion: you have to answer the phone
D’s conclusion: you have to answer the phone
(5) (A is talking about a law to fight against drug trafficking)

A: a mí [Corcuera me cae muy bien]
S: [sí/ claro/ y te pone una ley] que pegan una patá a la puerta y te entran
en casa de una patá§
A. I [find Corcuera is a nice guy]

S: [yes/ claro/ and he passes a law] where they [the police] kick the door and
come into your house
A’s conclusion: I like Corcuera
D’s conclusion: I don’t like Corcuera13
Stressing and hedging are processes whereby the speaker includes his or her point of view
in language: something has been stressed or hedged by someone to achieve a
communicative goal (Briz Gómez, 1998: 114). The same happens with (dis)agreement, a
category which cannot exist without reference to a participant, hence its modal value.
The relationship between the two groups can be established in the following way:
agreement and disagreement, which refer to a former utterance, often convey a stressing
value (repeating something is one of the clearest ways to stress it) but seldom a hedging
one. Both stressing and hedging are opposed to a neutral expression, as shown in Table 2
below.
Table 2. Relationship among modal values
Agreement Disagreement
Zero-degree sí no
yes no
Stressing claro que sí claro que no

of course (yes) of course (no)
Hedging bueno/ sí bueno/ no

well/ yes well/ no
3.2.3. Connection
Connection is the part of discourse markedness related to the linking of constituents.
Items whose main function is to bind together elements will be called connectives.
Connecting is a complex process, where different tasks are distinguished.
At a general level, connection is established between explicit or implicit constituents. An

implicit constituent can be an argument (Anscombre and Ducrot, 1994), a proposition
(Blakemore, 1987), an inference (Levinson, 2000) or, more generally, any unit in our
discursive memory (Berrendonner, 1983; Kay, 1990; Roulet et al., 2001). This is a distinct
aspect of connection: inferential function. Consider (6):
(6) S: ¿nunca te ha llamao mamá/ o qué?

L: a mí mi hijo tampoco me llamaba mamá
A: me desía Bárbara / me dise- Bár- me desía Bárbara y cuando se enfadaba
conmigo en algo asíi de casa me desía/ ¡Barbarita!/ ¡mira! (RISAS)
¡Barbarita! / ¡pumba! (RISAS)
S: pero a mí tam- a mí tampoco me gusta que llamen mamá
S: did she ever call you mom/ or what?

L: my son didn’t call me mom either
A: he called me Barbara / he says Bar- he called me Barbara and when he was
angry with me for something he called me/ Barbarita!/ see? (LAUGH)
Barbarita!
S. but I do- I don’t like to be called mom either
What pero (but) binds together is not two explicit propositions, but an utterance (I don’t
like to be called mom either) to S’s intended conclusion of the context proposition
(somebody doesn’t like to be called mom).14 If such an implicit element cannot be
recovered, the use of either would be odd (Schwenter, 2001).
The former function has to do with the construction of the message, be it planned or
unplanned (Ochs, 1979). This is a metadiscoursive function and comprises two distinct
dimensions, depending on the task performed by the speaker: organizing the linguistic
constituents of the message (structuring function) or taking a distance from previous

formulations (reformulative function).
As conversation proceeds, speakers shift from one position to another and construct their
speech accordingly, by rephrasing what has been previously said (by oneself or by the
others). In these cases the discourse shows a binary structure of the form A connective B,
where A and B are discourse constituents of variable length. Following a well-established
tradition (Antos, 1982; Gülich and Kotschi, 1983; Roulet, 1987; Rossari, 1994; Noren,
1999), I will call this process reformulation and I will distinguish paraphrastic
reformulation if the content of B merely rephrases A and nonparaphrastic reformulation
if the speaker takes the content of B as the only valid source for the continuation of
discourse. Examples (7) and (8) exemplify both concepts:
(7) (G. is talking of how a guy prepared his driving license exam)
G: yy después de haberse leído el libro en su casa o s(e)a en los ratos que tenía
libres/ fue al de la autoescuela y le dicee oye apúntame para examen
G: and after he read the book at home that is in his spare time/ he went to the
man in the driving school and tells him hey inscribe me for the exam
(8) (A, B and C talk about A’s stay in Belgium and the meals in that country)
C: ¿entonces/ en los bares qué hacen?
A: pues cosas raras// platos combinaos o cosas de esas
[…]
B: [o sea] allí no hay costumbre del bocadillo ni historias ¿eh?
C: So/ what do they do in bars?
A: well weird things// assorted dishes or things like that
[…]
B: [o sea] they don’t eat sandwiches or such stuff uh?
Structuring makes reference to the hierarchical, organizational aspects of connection, in

other words, to the way a speaker builds and structures a message. It has repeatedly been
noted that connectives display or create a relationship between linguistic constituents,
imposing a certain interpretation of how they are to be intended. For instance, in (9a),
where no ranking between the two sentences is made explicit, the choice of a connective
indicates that both constituents have the same status (9b), that the first is communicatively
subordinated to the second (9c), or that the second is communicatively subordinated to the
first (9d).15
(9) a. Europe is wonderful. Asia is impressive.

b. Europe is wonderful and Asia is impressive
c. Europe is wonderful. Nevertheless, Asia is impressive
d. Europe is wonderful, even if Asia is impressive
Connectives make explicit a relationship, which assumes distinct subfunctions according

to the size of constituents. Therefore, I will distinguish the following structuring aspects:
A. Delimitation: the connective structuring function is performed at a sequence or supra-

sentential level (Gili Gaya, 1983: 1943; Alcina Franch and Blecua, 1975; Halliday and
Hasan, 1976; van Dijk, 1977), as in (10):
(10) (B is talking about a clock she found in the street. B’s intervention adds a second
sequence to the story sequence that has developed over more than 200 lines).
A: no oye paa normalmente toos los días no se lo ponDRÁ/ pero asíi algún
día que see vista bien o algo §
C: §claro (( ))
B: ¡ah! y aún viene la otra noticia que también menuda semana han
tenido §
A: §luego han hecho fijo a mi marido
A: no listen usually- she’s not going to wear it every day but like that one
day if she dresses up or something §
C: §right
B: oh! and now comes the other news well what a week they had
A: now my husband has got a new contract.
B. Formulation: the connective structuring function is performed in the microsentential

level. Here emphasis is put on problems related to discourse planning; connectives reflect
the speaker’s effort to build his turn, leaving as traces hesitations, false starts, or
interruptions. Consider the following example:
(11) (The speaker is describing an elevator he saw in Mallorca)

S: claro// no el de allí también ¿eh? subía un piso o dos/ el dee-el de Mallorca
pero claro/ tu veías// veías la zo- la playa desde- desde l’ascensor ése por
eso sí que tenía muchoo ///(2.5”) a mí ese Pryca me gusta
S: right// no the one there too/ y’know? it went up one or two floors/ the-the one
in Mallorca but sure// you could see// you could see the ar- the beach from-
from that elevator that’s why it was very ///(2.5”) I like that Pryca [a
supermarket]
Hesitations and false starts (el dee- el de), as well as pauses (veías// veías) denote
problems in the construction of the intervention. Claro works as a filled pause indicating
the will of the speaker to hold the floor and to keep on talking.
Formulation is a frequent function in colloquial language, either in its spoken or written

form. It must not be confused with the function of reformulation above. While the latter
refers to the argumentative structure of a message and has to do with the rephrasal of
arguments, the former has to do primarily with the expression—not the content—and with
planning problems, which provoke “changes of project” (Sornicola, 1981), either syntactic
or semantic. Another difference lies in the scope of both operations; in its formulative use,
the scope of the connective is more limited than in its reformulative use (see section 5
below).
C. Regulative function: this is the prototypical function of the sentential level and consists
of displaying the beginning, the continuation, or the end of discourse constituents. While
the ideas of start, progression, and closing are common to delimitation and formulation
(e.g., a formulative connective indicates the speaker’s will to hold the floor, hence to
continue), the difference lies in the scope of constituents. This level is especially suited to
accounting for sentential relationships, and for noncanonical constructions in spoken

language (Narbona Jiménez, 1989a; 1989b; 1990). It establishes a link between
grammatical and nongrammatical uses. Although this is an open field, an outline of what
beginning, continuation, and closing could refer to might be that regulation at the
beginning of an utterance signals its initiative or reactive character, offering instructions
about its role with regard to the previous context.
(12) ?: [ve-vender] seguros// vendía yoo// una temporada/// no vendí ni uno y lo tuve
que dejar
A: (RISAS) pues por eso/ yo ni lo he cogío
?: [sel-selling] insurances// I sold// for some time// didn’t sell a damn insurance
and I quit
A: [LAUGHS] pues that’s why/ I didn’t take that job
In (12), pues ‘well’ at the beginning of A’s intervention signals its reactive nature, links
its interpretation to the preceding one, and consequently marks the existence of a
dialogical conversational unit.
Regulation as a mark of progression is found within utterances and indicates the

subordinate or coordinate character of constituents. In (13), pero ‘but’ regulates the
progression within the utterance, indicating a relationship between a subordinate and a
directive act.
(13) E: síi/ yo conozco gentee/ parezco muy liberal pero// la verdad es que soy muy
conservadora.
E: yes/ I know a lot of people/ I look like a liberal but// in fact I am very
conservative.
This explanation is compatible with both the adversative character of the conjunction but
and the argumentative value of the connective. The grammatical and pragmatic values of
connectives are thus related, the former perhaps being a functional specialisation of the
latter.
The last value of regulation is to indicate the end of the utterance. Many interactional
markers have this function:
(14) D: pasa Fanta// hay que beberse- aún queda un litro y medio ¿eh?
D: gimme some Fanta// we have to drink- there’s still half a gallon left uh?
Note that this functional conception of discourse markedness lets any occurrence of a
marker perform different functions at the same time, so there is nothing strange in
postulating that ¿eh? in (14) indicates an interactive function between participants and
marks at the same time the end of the constituent. Accordingly, pero in (13) signals an
inferential relationship between two arguments while marking the progression of the
utterance and a relationship between a subordinate and a directive unit.
The functional conception sketched here fits well with a prototype approach:
semasiologically, the different meanings of polysemous units are accounted for in terms
of family resemblance.16 This provides a link between connective uses and uses belonging
to other categories among the discourse-marker readings. Therefore, it is possible to

provide an unitary description for an expression. Onomasiologically, a DM performs
several functions simultaneously at different discourse levels. Functions need not be
interpreted exclusively. The link between both approaches can be established in terms of
preferences, either quantitative (a reading is considered more basic because it represents a
higher percentage of the global occurrences of a DM) or qualitative (the researcher
determines the importance of the different meanings on the basis of testable criteria).
Finally, it is possible that the reader may have observed a certain inconsistency in the use
of terms like sentence, utterance, or discourse constituent in this section. I have tried to
limit myself to these traditional labels, although in order to develop a functional
explanation theory of discourse units, more precision is needed.
4. Model (or the lack of it)17
Is it possible (or desirable) to construct a theory in order to account for the use of DMs?
This question can be construed in two different ways, depending on whether we want (i) a
theory which provides a place where DMs can be located or (ii) a theory whose only
purpose is to explain what DMs are. I think that, even if we confine ourselves to a more
restrictive term, connectives, the answers to these two questions are, respectively, yes and
no.
Considering what the study of DMs has represented in the last few years, two major
contributions to general linguistics can be highlighted. On the descriptive side, DMs have
provided a corpus of unanalyzed data deserving particular attention. On the theoretical
side, DMs have been a challenge for most theories within pragmatics. It is not by chance
that some of the first developments in Text Linguistics (van Dijk, 1977), Argumentation
Theory (Ducrot et al., 1980) or Relevance Theory (Blakemore, 1987), to quote some of
the best known theories today, have undertaken the study of DMs. Thanks to DMs, there
is a better understanding of the processes of inference, both synchronically and
diachronically (Argumentation Theory, Relevance Theory, Gricean pragmatics,
grammaticalization), of how speakers organize their messages, especially in spoken,
casual language (Conversation Analysis) but also in more formal registers (Geneva
School, Text Linguistics) and of how they manage their social images (Politeness
Theory). These approaches have all shown that the relationship between a pragmatic
theory and DMs is close and fertile.
A very different question is the creation of a theory to cope with DMs. This has not been
the method used by most scholars and the reason is that the ingredients needed to cook
DMs are already available in the pragmatic market. Even in the cases where a theory is
mainly concerned with DMs (like in Roulet et al., 1991), a careful reading shows us that
DMs are just a way to understand deeper underlying principles. DMs are eclectic by
nature and so are theoretical accounts of DMs.
In my view, the notion of model we should adhere to must be reduced to the more modest
but nevertheless broad purpose of putting together pragmatic principles, dimensions of
analysis, and linguistic features in order to provide coherent descriptions of sets of
connectives.
DMs are a melting pot of problems and perspectives, as the contributions in this volume
show. In this section, I am going to focus just on one of them, namely, the relationship
between a discourse unit theory and the description of connectives.
As mentioned in section 3, some of the functions subsumed under the concept of

connection are subservient to a vision of conversation as a structured whole. As
connectives take scope on discourse segments of variable length, a thorough segmentation
of conversation will help to determine what connection means. Perhaps the only attempt
to divide a whole conversation systematically into units has been performed by Eddy
Roulet and the Geneva School (Roulet et al., 1991; Roulet, 1991; Roulet et al., 2001;
Roulet, this volume). The Val.Es.Co research group is currently working on a system of
units suited to colloquial conversations (Briz Gómez and Val.Es.Co, 2003) which will be
developed in what follows.
The starting point for segmentation is the speaker’s turn. Conversational Analysis
considers that a change of speaker is a necessary and sufficient condition for recognizing
a structural unit: the turn. However, this idea can be challenged for two reasons. First, the
rest of the participants do not accept participants’ contributions every time they speak.
Sometimes, especially in spoken, colloquial language, what a speaker says has no
influence on the structural development of the conversation; his contribution is not
“picked up” by the others. This means that its contents are not rephrased, his point of view
is not taken into account, no lexical chain is created from their words, etc. So the
participant has not been recognized as a valid speaker and his or her contribution is
disregarded. In what is traditionally called turn, two types of units can be distinguished:
those which have been accepted by the others and make conversation proceed, and those
which haven’t been accepted and accordingly have no structural (i.e., thematic) weight.
The former will be called turns; the latter, interventions. The person who utters a turn is,
in our terminology, a speaker; the one who utters an intervention is a participant. Phatic
signals, for instance, are interventions, not turns; what they signal is the refusal of a
participant to become a speaker. 18
The recognition of a turn must be established with regard to the rest of the participants. It
is their acceptance of the speaker’s contribution that makes an intervention become a turn.
At this point, a (provisional) definition of turn is the following:
• Talk slot filled with informative interventions, recognized by participants through their
ostensive and simultaneous attention;
• Between interventions and turns there is an entailment relation. Every turn is an
intervention, but not every intervention is a turn
Second, if a turn (in the Conversation Analysis sense) is assigned every time a speaker
changes, we can lose sight of some interesting phenomena. In example (15), I3? does not
take B’s contribution into account (hence, A does not assign the role of speaker to B),
because A does not answer B’s question (I= intervention; T= turn).
(15) I1 A: siempre tienes laa- la desviación profesional la enfermedad

profesional
T1 I2 B: ¿el qué? ¿lo de ser filólogo?
I3? A: de observar a los demás y ahora es- sentirse observado ess
una sensación extraña
I1 A: You always have the occupational sickness the
occupational illness
T1 I2 B: What? Being a linguist?
I3? A: of watching other people and now it’s- feelin watched is a
weird sensation
Nevertheless, I3? does not seem to be a new intervention; on the contrary, it continues
both the topic and the syntactic structure of what was said in I1. A better account of (15)
would be to consider that I1 was interrupted by I2 and was continued later. The whole
exchange would constitute a single turn and the structure of (15) would be better
explained as a discontinuous intervention [T1 I1-I2-I1´ T1]. To account for these cases, a
definition of intervention in nonnegative terms is necessary. An intervention is delimited
when it is (a) the source of another’s speaking (in this case, it is an initiative intervention),
(b) a reaction to someone else’s intervention (reactive intervention), or (c) an initiative
and a reactive intervention.19 A plausible definition of intervention is the following:
Intervention: act or set of acts continuous or discontinuously uttered by a speaker.

An intervention is either a reaction to a previous intervention, a start to a new
intervention, or both, a reaction and a start.
In example (15) above, I1 is an initiative intervention, for it provokes B’s answer. I2 is a

reactive intervention but not an initiative one, because speaker A does not answer B’s
question. In turn, I1 and I1´ are recognized as a single unit because the next intervention
(not in the example) sí/ pero si me dices eso ya . . . (‘yes/ but if that’s what you’re saying
then’) contains an anaphoric expression (eso ‘that’) referred to I1 and I1´ as a whole.
The typical structure of a colloquial conversation is one in which an intervention is at the

same time reactive and initiative. Two adjacent interventions, one initiative and the other
reactive, constitute an exchange.
From our definition it follows that any intervention is made up of immediate constituents,
called acts. Acts can be provisionally defined in the following way:
Act: Immediate constituent of an intervention which is able to appear in isolation

in a given context.
An act can be the only constituent within an intervention, even if the other constituents of
the same intervention are cancelled. This is what in isolation means. Although the criteria
for delimiting acts are not totally clear, a practical example will make this concept clearer:
(16) a. A: quédate
A: stay IMP
A: ‘stay’
b. B: no/ me voy porque tengo prisa

B: no/ I go- PRES because have 1rst PERSON haste
B: ‘no/ I’m leaving because I’m in a hurry’
c. B: no
d. B: #no/me voy
B: #no/ I go- PRES
B: ‘#no/ I’m leaving
e. B: no/me voy
B: no/ I go- PRES
B: ‘no/ I’m leaving
f. B: # porque tengo prisa

B: because have 1rst PERSON haste
B: ‘#because I’m in a hurry’
g. B: me voy porque tengo prisa

B: I go- PRES because have 1rst PERSON haste
B: ‘I’m leaving because I’m in a hurry’
h. B: [A1no A1]/ [A2me voy porque tengo prisa A2]
(16a) is a turn, made up of two interventions. The first one is an offer. The second is a
refusal to that offer. No alone in (16c) answers the offer, hence is an act. No me voy in
(16d) gives rise to a pragmatically strange utterance, substantially improved by changing
the rising into a falling pitch (16e). Porque tengo prisa alone does not answer A’s turn
(16f), but me voy plus porque tengo prisa together do (16g). Hence, the immediate
constituents of B’s answer are the acts shown in (16h).
Syntactically, the second act in (16h) above is a sentence containing two clauses. These
two components are, in turn, immediate constituents of the act. We call them sub-acts.
The theoretical status of sub-acts remains to be determined, but two features should be
highlighted: they usually have their own intonational contour and perform a function
within the act (causal, conditional, etc.), although they are not independent in the sense
described above.
Between an exchange and a conversation there is another unit called sequence (in the
Conversational Analysis sense, for instance), but its status seems to be more semantic
than structural and will be the object of future research.
In short, the units of conversation distinguished in our proposal are summarized in Table
3.
Table 3. Units of conversation
LEVEL INNER STRUCTURE INTERACTIVE STRUCTURE
Conversation
Dialogic Exchange Turn changing
Intervention Turn
Monologic Act
Sub-act
Sequence
It is now possible to link some aspects of the functional schema in section 3 with the
theory of units devised here. First, it is possible to be more precise about the rank of DMs
in their different functions: the regulative function of connectives is developed in
interventions and exchanges; when it signals a start, this is usually a reaction to a previous
intervention (example (12)). When it is a signal of progression, we are dealing with two
acts or subacts within an intervention (example (13)). If what is signalled is a close, this is
usually at the end of the intervention (example (14)). A delimitation function is performed
between interventions or exchanges, not in lower level units. By contrast, formulation has
an upper limit in the intervention, but it functions prototypically in the act/subact level.
Secondly, it provides a clue to further distinguish connectives from subordinate
conjunctions. The latter takes the act as their upper limit (they bind together two subacts,
in our terminology), while the former take intervention and sequence as their lower and
upper limit, respectively.
Notes
1
I wish to thank Bruce Fraser, Santiago Posteguillo, Mila del Saz, and Sandra Cosham for
comments on previous drafts of this paper. I also want to thank Kerstin Fischer for her
enthusiasm and her willingness to question the contributor’s views. All shortcomings,
however, are mine.
2
See Section 1.4.2. for further details on this distinction.
3
Colloquial language (German Umgangssprache, Italian parlato) is a concept coined in
the above-mentioned tradition. It has to do with a twofold vision of registers, named
respectively formal and informal/colloquial (registro formal; registro
informal/coloquial). According to Briz et al. (1995), both registers are conceived as
prototypes, and their intersection gives rise to mixed registers (semiformal and semi-
informal). A conversation can shift from a formal to a more informal register through a
colloquialization process. A more detailed explanation of these concepts can be found
in Briz et. al. (1995).
4
A sample of this corpus has been published in Briz and Val.Es.Co (2002). This is a
subset of the general corpus, comprising more than 300 hours of recordings made in
Valencia (Spain), which has sociolinguistic validity. In order to ensure the
colloquiality of the recordings, a participant observer method has been adopted. The
transcripts follow the Jeffersonian method adapted to the requirements of theSpanish
language. See also http://www.uv.es/valesco. For other registers, the CREA corpus
(RAE, Spanish Academy of Language), provides a huge data base of 200 million
words. The diachronic information has been extracted from the CORDE corpus,
compiled by the R.A.E. Both corpora are available on-line at http://www.rae.es.
5
For the type/token distinction, see Lyons (1980).
6
This distinction is also found in Portolés (1998, 36-37) and Cuenca (2000a: 77-90).
7
Personally, I prefer the labels discourse markers or connective to discourse particle for
two main reasons. First of all, compared to discourse particle, discourse marker and
connective are two widespread and widely accepted terms used to refer to the category.
Secondly, particle has been used to refer to “an invariable part of speech” (Lázaro
Carreter, 1990: 315), and its use could lead to wrong assumptions about the lexical
units under investigation being invariable parts of speech.
8
Among the few exceptions are Roulet (1985) and Schiffrin (1987a).
9
In this section, I will not refer to DMs in general, only to a subset, that of connectives.
10
This goal can be transactional or nontransactional (the phatic communication of
colloquial language).
11
All excerpts come from Briz Gómez and Val. Es.Co (2002). Some of the transcription
conventions are as follows: /: short pause; //, ///: longer pauses. [ ]: overlap between
two speakers. §: no perceptible pause between speakers A and B. , : rising and
falling pitch. -: self interruption. italics: direct style. CAPS: emphatic stress. The full
system is described in Briz Gómez and Val. Es.Co (2002).
12
This conception is taken from Ducrot (1995) and from Levinson (2000).
13
Note that the same word (claro) is used for expressing both agreement and
disagreement.
14
Remark that the adversative connective pero has in this usage a somewhat additive
value: it introduces a new utterance to the topic of the conversation.
15
Of course, connectives are not the only source of information. The scale <wonderful,
impressive> or the implicatures associated to some connectives (e.g., and) are other
relevant sources. But this is not to say that connectives merely display a pre-existing
order (Schiffrin, 1987a), as shown in (9c, d), where the order of the constituents is
changed, leaving the connectives in the same place.
16
Hansen (1998a, this volume) has synthesized three solutions to this problem:
monosemy, homonymy, and polysemy and has consistently argued in favour of the
third. For a different point of view, see Fischer and Travis (this volume).
17
A lively discussion with the editor of this book made me aware that we understand the
word model in two different senses. I use the word model in the sense of model
theoretical semantics (Dowty, Wall, and Peters, 1981: 10-11; Cann, 1993: 39-41). In
this respect, to create a model for the explanation of DMs means that this is the only
object of that model. To me, this requirement is only found in Schiffrin (1987a), while
the rest of the explanations deal with issues of a more general scope (cognition in
Sperber and Wilson, the structure of discourse in Roulet, the argumentative component
of language in Anscombre and Ducrot, and so forth). A very different sense of the
world model involves the articulation of theoretical instruments in order to explain
what DMs are and how they work. In this second sense, all of the authors in this book,
including myself, have provided a model. My position in this section can only be
understood in relation to the first reading of the word model, as explained above.
18
This is not to say that phatic signals have no relevance at all. Phatic signals are relevant
for the social structure of conversation (they regulate the turn-taking system, assure the
speaker’s role, etc.).
19
Initiative and reactive are used in the same sense as Roulet et al. (1991).
Chapter 6
Pragmatic markers in translation: a

methodological proposal
Karin Aijmer, Ad Foolen, and Anne-Marie Simon-Vandenbergen
1. Introduction
Research on discourse particles faces two major challenges. On the theoretical side there
is a clear need for a model of communication which is rich enough to account for the
complexity of the functioning of discourse particles. It seems to us that there is no such
model and that research findings regarding various aspects of the functions of discourse
particles have not as yet been accommodated within an overarching framework. This
paper proposes one way in which such a model can be developed. In section 4 we discuss
how particles relate to the utterance, to the context and to the hearer. These three aspects,
which we refer to as reflexivity, indexicality, and heteroglossia respectively, seem to us
crucial to the understanding of discourse particles. The second challenge of particle
research is to deepen our insight into the multifunctionality of these elements in order to
arrive at a satisfactory account of meaning relations on the semantic and pragmatic level.
A variety of research methods are thus necessary so that findings arrived at by different
routes can be compared with each other to confirm or throw into doubt a given account of
multifunctionality. In section 6 we suggest that the translation method is a useful
empirical tool for contrasting particles in different languages. Not only does it provide
data on how languages deal with similar meanings (thus contributing to typological
insights), but, we argue, it also offers a way of dealing with the indeterminacy of the
meaning of discourse particles. Through translations, it becomes possible to map semantic
fields, which in turn give us a basis on which to establish basic underlying meanings or to
confirm an earlier hypothesised meaning. In section 3 we specify our own position with
regard to multifunctionality, polysemy and core meaning. This is further illustrated in
section 5. Before dealing with these issues we clarify our use of terminology in section 2.
2. Definition
Discourse particles belong, in our view, to the more general category of pragmatic
markers. The latter are defined here negatively: if a word or construction in an utterance
does not contribute to the propositional, truth-functional content, then we consider it to be
a pragmatic marker. We are in agreement in this respect with Fraser (1996), Foolen
102 K. Aijmer et al.
(2001), and Hansen (this volume), who also propose to start with pragmatic marker as the
most general functional term and then to subclassify the markers on the basis of more
detailed functional distinctions. Such functional subcategories include, for example,
politeness markers, hesitation markers, and discourse-organisational markers. In addition,
we propose to take formal criteria into account in this subclassification. Formal
subcategorisation distinguishes, for instance, between particles, adverbs, and pragmatic
expressions.
One of the difficulties in deciding whether a given form should be considered to be a

pragmatic marker is that a single form often fulfils in certain of its uses a function on the
propositional level and in other uses a function on the non-propositional level. Thus, if we
want to be precise, we should not ask whether a given form is a pragmatic marker or not,
but rather whether a given use of a given form can be considered a pragmatic marker.
While for some forms it is easy to distinguish uses as pragmatic markers from other uses
(for instance the pragmatic marker well as opposed to the manner adverb), for other forms
the line is less obvious (for instance, the pragmatic expression I think as opposed to the
mental process verb). One should also allow for fluidity and take a dynamic view on the
issue. For instance, many adverbs (including certainly, surely, of course) seem to be on
the boundary between modal adverb and pragmatic marker (see Hoye, 1997: 212; Lewis,
1999; Simon-Vandenbergen and Aijmer, forthcoming).
It should be emphasised here that we use the term marker in a purely technical sense.
Other authors (see Fischer, this volume, section 3) have felt that terms such as instruction
or marker leave too little room for the active roles played by speakers and hearers as
creators and interpreters of contextual aspects of the utterance. Roulet (this volume), for
example, prefers pointer or indicator. This accords with our own view that the character
of such words is typically indexical, in other words that these elements merely “indicate”,
leaving the hearer with a significant amount of interpretational work (see section 4.2). The
term marker for us does not imply either a passive speaker or a passive hearer. Its
usefulness as a technical term lies in its wide currency in a field in which, we believe, a
proliferation of terms must be avoided.
Even if research has not (as yet) come up with hard and fast criteria to determine whether
the use of a particular form is to be considered a pragmatic marker, it seems that there is a
sense in practice of what to include in the category. However, a concern with explicit
definitions should, in our view, remain on the agenda. How do we treat interjections,
focus particles, connectives like but and because? Do connectives have pragmatic uses in
addition to propositional ones? In Sweetser’s (1990) framework, which allows
connectives to function in different domains, this is indeed the case (see also Schiffrin,
this volume, on and).
Classification of the markers in formal terms is of a different order. We can subdivide

markers into segmental and suprasegmental elements, and into particles versus fuller
forms (a characterisation on the phonological-morphological level), and we can classify
the expressions in terms of syntactic form classes, based on the system in a given
language. Formal and functional classifications do not necessarily coincide, although,
assuming the existence of iconic forces, one would expect function-form correlations.
Pragmatic markers in translation 103
Discourse particles form a subclass of the more comprehensive functional class of

pragmatic markers. They are distinguished here on formal and functional grounds from
other types of pragmatic markers, such as modal or focus particles, pragmatic expressions
(a term we reserve for expressions such as I mean, I think, you know), or connectives. A
useful list of criteria is provided by Jucker and Ziv (1996), based on Brinton (1996). In
addition to the functional feature of multifunctionality, it includes phonological and
syntactic features: discourse particles are typically monosyllabic and are placed in the pre-
front field, cf. Auer (1996). A combination of functional, phonological, and syntactic
criteria would then classify such words as well or now as discourse particles, to be
distinguished from pragmatic expressions, as well as from elements such as German ja,
doch, eben; Dutch maar, toch, even; or Swedish ju, nog, väl. Such particles, which exist in
all Germanic languages except English, are sometimes called discourse particles, but they
actually share only some properties with the English-type discourse particles (Fillmore,
1984). Another common term for the German, Dutch, and Swedish particles just
mentioned is modal particles, which, although not entirely felicitous, has the advantage of
distinguishing the category from discourse particles (see also Jucker and Ziv (1996: 2) on
the terminological confusion in this area).
When we have made enough headway in determining what the pragmatic markers are in
different languages, a contrastive and typological step can be taken, guided by the
following questions:
• Which elements fulfil pragmatic functions in the languages of the world?

• How can these elements be classified in functional terms?
• How can these elements be classified in formal terms?
• How do languages differ with regard to these functional and formal classes?
On the basis of such inventories and comparisons, more interesting questions are possible:
can languages be classified into those which are “poor” in pragmatic markers versus those
that are “rich”? Are there correlations between this and other aspects of the language? It
will be some time before research on pragmatic markers has reached the level that such
questions can be answered. But if our aim is to make research in this area normal
linguistic business, then we should ask the same kinds of questions about pragmatic
markers as about any other kind of linguistic element.
3. The functional spectrum
In addition to the difficulty of differentiating pragmatic markers from the rest of the
linguistic system, two other problems seem prominent in this area of research, namely, the
elusiveness of the meaning of pragmatic markers and their polyfunctionality.
First, there is the problem of dealing with elements that lack propositional meaning.
Expressions that contribute to the propositional content of an utterance refer to objects,
properties, relations, and quantifications. Although the meaning of pragmatic markers
cannot be explained in similar terms, this does not imply that pragmatic markers are
meaningless.1 What kind of meanings is at issue then? Are there core-meanings and non-
core meanings? How many meanings is it useful to distinguish? Which meanings belong
to semantics and which ones to pragmatics? These are questions that have plagued
researchers for a long time (see Foolen and Van der Wouden, 2002), and while there may
be no right answer to these questions, we believe it is important to make one’s position on

them explicit.
We propose to adopt the notion of core meaning as a starting point. It should be noted that
this concept (also referred to as basic meaning) is not used consistently in the literature.
For some authors, it is the unifying semantic meaning that serves as input for pragmatic
differentiations. In this view, the core meaning is underspecified or undifferentiated and in
fact impossible to define except in terms of its multiple contextual uses. For others, core
meaning constitutes the central semantic meaning to which other coded meanings can be
related. According to this view, the core meaning may be identical with or coloured by the
original diachronic meaning.2
For us, the core meaning is the central, or underlying, meaning to which pragmatic
meanings can be related. This meaning is necessarily a fairly abstract notion, not
necessarily conforming to native speakers’ intuitions. We find evidence for it in the
diachronic development of the particle, that is, in the traces of its lexical origin which are
retained in its present-day usage. There are several advantages to positing a core meaning
defined by semantic components or by a paraphrase. First, it ensures that pragmatic uses
are explicable, since they cannot be random or arbitrary. Second, it allows us to specify
and explain overlapping and diverging functions between pragmatic markers in terms of
their different core meanings. Third, it serves as a hypothesis for a tertium comparationis
in contrastive research (see section 6). And finally, we do not need a separate account for
synchronic and diachronic phenomena, since the same principles explain extensions into
the pragmatic domains.
The second difficulty, the notorious polyfunctionality of pragmatic markers, is closely

related to the abstract nature of the core meaning and to the strategic uses that speakers
make of such markers in different contexts. It is evident that the contextual meanings are
the result of speakers’ tactical uses of elements that are semantically vague enough to
allow for multiple purposes.3 In general terms, we can say that pragmatic markers have
interpersonal and textual rather than ideational functions. Moreover, because of their
context dependency, they tend to have a rather large number of different functions in
these domains. The question that remains is how to explain polyfunctionality. In recent
years, different models have been proposed within cognitive semantics for explaining
polysemy, such as radial categories, prototypes, and core meanings. In principle, the
relative success of these models for content words and constructions on the propositional
level should be taken into consideration in adopting particular models, or aspects of them,
for the description of pragmatic markers. At the same time, pragmatic markers can serve
as an interesting testing ground for such models.
The difference between coded meanings and contextual implicatures (pragmatic

enrichments) should be considered in the analysis of polyfunctional phenomena. As a
guideline, we would advocate a methodological minimalism, in which coded meanings
are only assumed if they cannot be derived by processes of conversational implicature.
Translations can be of use here. In a translation corpus, some meanings are frequently
picked out by a special translation equivalent and recur in more than one language. These
are the coded, conventionalised meanings which should be distinguished from more
temporary, context-specific and creative meanings which can be processed by inferencing.
A model which is based on an underlying core meaning on the semantic level and
implicatures on the pragmatic level seems to be appropriate in most cases. Such a model
finds support in diachronic research. Grammaticalization processes seem to develop along
metonymic lines, as a result of rhetorical uses of particular items (Traugott, 1999b). In the
process, content meanings are bleached in favour of the types of meanings typically
expressed by pragmatic markers (see also section 4.3 below on the rhetorical function of
pragmatic markers).
The preceding discussion might give the impression that in our model, we want to allow
only a single core meaning and the contextual implicatures that can be derived from it.
But we are not opposed to polysemy networks for pragmatic markers as proposed, for
example, by Hansen (this volume, section 1.4.1). We simply want to stress that within the
polysemy network, one of the nodes often has a prototypical or core status, or,
alternatively, that an abstraction over the different nodes is possible, resulting in a core
meaning. Particularised and generalised implicatures, polysemy networks, prototypical
meanings and abstract core meanings, are all potentially relevant in a full description of
the meaning of a particle (see Aijmer, 2002a: 19-26, for further discussion).
4. The model
When we communicate we do not use language simply to convey a message. We use

certain linguistic elements metalinguistically to refer to the text or the utterance itself.
This relation is explained by our principle of reflexivity. We can also use pragmatic
markers to point to contextual or social phenomena outside the utterance or the text. This
relation between the speaker and the outer world is explained by a principle of
indexicality (Ochs, 1996). Finally, our utterances can be used strategically to take up
different positions vis-à-vis other people, and other opinions in a heteroglossic
perspective.
We will focus in the following sections on the notions of reflexivity, indexicality, and, in
particular, heteroglossia. These notions play a role in other contributions to this volume,
but not in combination with each other as we propose here. This account will be compared
with the relevance-theoretical account of pragmatic markers, in particular the idea that
they have procedural meaning.
4.1. Reflexivity
Pragmatic markers support the interpretation of more central informative aspects of the
utterance by commenting implicitly on the utterance or the text. From the point of view of
the utterance, the marker thus has a “meta” status, and an understanding of pragmatic
markers involves what is variously called their metalinguistic, metacommunicative, meta-
pragmatic, or metadiscursive character (Nyan, this volume). We refer to this property of
pragmatic markers as reflexivity.
That natural language is its own metalanguage has long been recognized in linguistic
philosophy. However, the pervasiveness of this phenomenon in language structure and
language use is now gradually becoming clear, thanks to theoretical frameworks in which
reflexivity is given a central place, such as Lucy (1993) and Clark (in press). As Lucy
states in his introduction (1993: 11), “speech is permeated by reflexive activity as

speakers remark on language, report utterances, index and describe aspects of the speech
event. . . . This reflexivity is so pervasive and essential that we can say that language is,
by nature, fundamentally reflexive”.
Clark (in press), within his theory on communication as joint activity, distinguishes two
systems of communication, the primary system which “represents the official business of
conversation, what people are primarily trying to do in speaking”, and what he calls the
collateral system. This latter system is “that part of language use in which people
coordinate, establish, or manage their primary language. It is not normally the official
business of conversation, or what the participants are primarily trying to do”. Within the
collateral system, Clark distinguishes different subsystems, among them side exchanges,
in which speaker and hearer explicitly take turns to clarify certain aspects of the ongoing
conversation, and asides, short signals like oh, uh, um, that speakers insert in their
utterance “to help explain features of the current performance”. In relation to asides, Clark
mentions that “expressions such as I mean, you know, excuse me, well, oh, like, ah, now,
uh, and um . . . have been classified under a plethora of names”. He summarizes several
schemes of classification, and concludes: “It remains to be seen how these schemes deal
with the contrast between primary and collateral language”.
We would suggest that the expressions that Clark mentions fit very well within his
collateral system. By using pragmatic markers, speakers try to prevent explicit, time-
consuming side exchanges that risk causing irritation or even conflict. From the point of
view of reflexivity, pragmatic markers are condensed, grammaticalized substitutes for
side exchanges. Speakers indicate that they are aware of certain potentially problematic
aspects of their utterances, while at the same time they propose implicitly to the hearer to
continue the “official” business of conversation.
4.2. Indexicality
What is the semiotic status of pragmatic markers? Do they have symbolic, iconic, or
deictic-indexical meaning? In recent years, several authors have argued that pragmatic
markers have deictic meaning: see, for example, Hentschel (1986) for German modal
particles, Wilkins (1992, 1995) for interjections, and Ochs (1996) and Schiffrin (this
volume) for various types of markers.4 As with polysemy (see section 3), the property of
indexicality is not exclusive to pragmatic markers. Some grammatical forms and
structures that participate in expressing propositional content are also deictic: pronouns,
tense forms, adverbs of time and place. They point to participants, time, and space as
present in the communicative situation, so their referential value depends on the
communicative situation. The same is true for pragmatic markers. The crucial question is,
however, what do they point to? Here we have to turn to aspects of the communicative
situation which are less concrete than participants, time, and place and which are not yet
understood in their full complexity. For example, pragmatic markers may index
positionings in relation to persons or to the proposition itself.
A number of authors have tried to explicate some of these aspects of the communicative
situation. Schiffrin (1987a) distinguishes five planes: the ideational structure, action
structure, exchange structure, information state, and participant framework.5 Ochs (1996)
also assumes a range of situational dimensions such as social acts and social activities
indexed either lexically or grammatically. However, only an explicit theory of

communicative interaction can form the basis for a reasoned distinction of the relevant
dimensions or planes particles can point to. Other authors in the present volume also
argue for the need for such a general model. Roulet (this volume, section 1), for example,
advocates “the elaboration of a global model of the complexity of the organization of
discourse, a model which provides a place to locate DM [Discourse Markers] within”.
Blakemore (1987) has analysed discourse markers within the framework of Relevance
Theory (Sperber and Wilson, 1986). She proposed the notion of procedural meaning to
characterise the type of contribution these elements make to the communicative process.
In Blakemore’s (1987) view, expressions like but, moreover, etc., are “instructions for
processing propositions” (see the reference to Blakemore (1996: 151) in the paper by Ler
Soon Lay, this volume). This view seems close to what we called the reflexive status of
discourse markers. If procedural meaning also implies the notion of indexicality, then
Blakemore’s formulation would encompass the two notions of reflexivity and
indexicality.
In recent years, however, Relevance Theory has made additional distinctions, with the
result that our distinctions can no longer be easily mapped onto Relevance Theory
distinctions. In addition to the distinction between truth-conditional and non-truth-
conditional, there are the distinctions between procedural and conceptual meaning,
between implicatures and explicatures, and between higher-level and normal explicatures.
The concept of procedural meaning is now applied to all elements that lack a clear
conceptual meaning (‘cat’, ‘happy’), so that indexicals like he and yesterday are
considered procedural elements (cf. Carston, 2002: 160-164). In this use of the term, it
comes close to, or might even be co-extensive with, what we term indexical meaning. As
is pointed out by those working within Relevance Theory, however, expressions like
frankly and in other words have conceptual meaning. Some of these expressions, such as
in other words, could very well be considered to be discourse markers. This would mean
that indexicality is not a defining property of discourse markers, but at most a prototypical
property. We would claim, however, in line with grammaticalization theory, that
discourse markers with conceptual meaning tend to develop diachronically into markers
with indexical meaning.
Relevance Theory does not distinguish a class of discourse markers per se, but a
distinction is made between elements that contribute to truth-conditional content and
those that do not. Within the latter group, some elements, such as frankly, contribute
higher-level explicatures, whereas others, such as but, contribute an implicature. This
suggests that propositional attitude markers and discourse markers involve different kinds
of meaning. Both, however, have a reflexive relation to the propositional content of the
utterance.
It follows from this comparison that indexicality and reflexivity are not properties that are
exclusive to discourse markers. However, all discourse markers are reflexive in relation to
the proposition, and most of them have indexical status. Further meta-theoretical research
seems to be necessary here (see Blakemore, 2002).
4.3. Heteroglossia
If we adopt the very general functional definition of a pragmatic marker proposed in

section 2, that is, as an item in the utterance which does not contribute to its truth-
functional content, we of course end up with a class of items with very diverse discourse
functions. Indeed, these functions are so diverse that it may be hard to see whether the
items share anything at all apart from the absence of a truth-functional contribution. This
is one reason why there is little agreement between authors on what to include in the class
and why authors make different decisions about categorization. Those authors who do not
take the most inclusive view tend to restrict the class either to items with a primarily
textual function (see especially Fraser, this volume) or to those with a primarily
interpersonal function (see for instance Andersen and Fretheim, 2000). If we adopt a
broad view, we must establish what, if anything, for example, a textual item such as
however shares with an interpersonal one such as surely. In addition, we must account for
the multifunctionality of individual pragmatic markers. For instance, the word well has,
judging by the various explanations which have been given in the extensive literature,
both a textual and an interpersonal function (see Aijmer and Simon-Vandenbergen, 2003)
for an overview). Fischer (this volume) refers to the range of functions which are “often
cumulatively” considered to be one of the essential characteristics of the class of
pragmatic markers, and it appears from her list that some of these functions are textual
(such as discourse structuring) while others are interpersonal (such as politeness).
We would like to propose one way of dealing with pragmatic markers which we believe
offers a framework that can accommodate their functional heterogeneity. Central to this
communication model is the notion of heteroglossia, a notion that originates with
Bakhtin. In what follows we will explain in what way the Bakhtinian concept will be
understood here and will then argue that pragmatic markers can be usefully integrated into
such a framework. We will also briefly refer to other work on pragmatic markers in which
a related approach is taken and will point to similarities and differences in focus.
Bakhtin (1981) uses the term heteroglossia to refer to the existence of different
“languages”, or world-views (see Björklund, 2000: 8):
By heteroglossia Bakhtin means the stratification of any language into different

sociological “languages” (Bakhtin, 1981: 271-272), which are forms for
conceptualizing specific world views (pp. 291-292).
This means that in a world dominated by heteroglossia, no utterance has meaning on its
own and that all texts reflect the existence of other texts and can only be understood in
these terms. It also means that monologue does not really exist, that all language use is
dialogue. Dialogue is a central notion in Bakhtin’s theory:
A word, discourse, language or culture undergoes “dialogization” when it

becomes relativized, de-privileged, aware of competing definitions for the same
things. Undialogized language is authoritative or absolute. (Bakhtin, 1981: 427,
editor’s note).
Our understanding and use of the concepts of heteroglossia and dialogization follows
White (1999, 2000), who adopts the Bakhtinian perspective to account for the options that
speakers have for positioning themselves intersubjectively.6 In White’s framework of

engagement, i.e., the meanings through which speakers express their alignments and
disalignments with other, previous propositions or expected propositions, the systems of
epistemic modality and evidentiality are brought together as different options taken by
speakers for different types of stancing.
It is clear that there is an affinity between White’s framework of engagement and other
models which see epistemic modality as part of a broader system. Chafe’s (1986) model
of evidentiality is an obvious precedent. Chafe, too, brings together epistemic modality
and expressions referring to the mode of knowing and the source of knowledge. Also
included are expressions signalling agreement or conflict with expectations, such as of
course, naturally, oddly enough, surprisingly. The main difference between Chafe’s and
White’s accounts lies in the way they explain the verbal choices language users make.
Chafe’s account is framed in terms of knowledge (degrees and modes), and choices are
seen as explicable from the extent to which speakers are certain that what they are saying
is actually true and from the ways in which they have acquired their knowledge about a
state-of-affairs. White’s account is not framed in terms of knowledge or degrees of
certainty, but rather in terms of rhetorical positioning. This entails that epistemic choices
(such as I think or may) do not always express lack of certainty but in fact are often used
for entirely different reasons, to do with interactional strategies. Although this difference
may be seen as one of focus rather than fundamental disagreement, it is the rhetorical
explanation which we believe allows us to account for both the existence of pragmatic
markers and their multifunctionality. For instance, work on I think (see especially Aijmer,
1997; Kärkkäinen, 2003; Simon-Vandenbergen, 2000) has shown that this expression,
while sometimes functioning as a marker of epistemic uncertainty, more often operates as
a conversational routine or even as a way of individualising one’s propositions in an
authoritative way. Aijmer (2002b) shows that surely typically conveys doubt rather than
certainty. Östman (1981) has demonstrated that you know is typically used when the
speaker knows that the hearer does not in fact know; hence, it appears to serve as a
rhetorical device to evoke solidarity by pretending that there is common ground. Thus, I
think, surely, and you know have a heteroglossic purpose and are options that allow
speakers to express heterogeneity of world-views and diverging stances.
Diverging stances are omnipresent in linguistic interaction and require communicative

remedying and problem-solving. But time-outs for metacommunicative interaction would
take up much of the time and energy invested in a conversation. In our view, pragmatic
markers exist to counteract metacommunicative interruptions by providing a way to
communicate implicitly, deictically, on possible diverging stances with regard to
particular communicative dimensions. This implicit communication runs parallel to the
uninterrupted explicit communication on the content level.
Diverging stances may be of different types. One type has to do with the relation between
interactants. In this case, pragmatic markers are ways of resolving the problem of open
conflict or disagreement. They thus serve as an acknowledgement of the speaker’s
awareness of their position or possible position and hence may function as politeness
markers. Another type of stance relates to the speaker’s own subjective reaction to either
an event or another proposition. In these cases, pragmatic markers signal to the addressee
the speaker’s interpretation of the position of the event or proposition in the larger
context. The choices which signal awareness of the dialogical process always assume a
divergence or possible divergence of some sort. In what follows, we shall illustrate these
points with uses of the discourse particle well.
The framework of heteroglossia and the view of pragmatic markers as dialogizing devices
which we propose here has parallels in other work, in particular in the concept of
polyphony as developed by Ducrot (1984). As pointed out by Roulet (1996), however,
Bakhtin and Ducrot “work within two very different frames, the sociology of verbal
interaction for Bakhtin and what he calls ‘ideal discourse structuralism’ for Ducrot”.
5. Illustration: well in a heteroglossic perspective
Well is notoriously vague and versatile (see Aijmer and Simon-Vandenbergen, 2003) and
a core meaning is therefore very difficult to define. However, it seems to us that the
communication model of heteroglossia makes it possible to capture the invariant meaning
as well as the multifunctionality of this particle. Consider the following examples from
the Oslo Multilingual Corpus:7
(1) “Oh,” she said. “You’re not married”

“Well, I am, but she’s . . . living elsewhere” (AT1)
(2) A great deal would depend, she knew, on whom she saw . . . luck . . . Well, she
had been lucky before. And besides, what she was suggesting was reasonable . . .
(DL2TNL)
(3) “Talk of the devil,” says Frank. “Well, here you are: Paul—Andrea.” (ABR1TNL)
(4) “Don’t worry,” said Beverley. “They don’t come in the houses. Well, not often.
They’ve got their own complex at the bottom of the gardens.” (ST1TNL)
In (1) the second speaker needs to resolve the divergence between A’s expectations and
his own. In example (2) the divergence is between the voice heard in the first proposition
(which raises a difficulty) and the one represented in the second sentence (which
dismisses the problem). In this case the two voices belong to the same person and they are
engaged in an interior dialogue. In (3) the divergence is between the speaker’s own
expectations and the actual situation he or she is confronted with. The pragmatic marker
signals to the hearer that what is to follow is somehow incongruous from the speaker’s
point of view. In (4) well signals that the previous utterance may lead to the wrong
expectations, which have to be corrected.
What well does in all four cases is signal something like “knowing this, accepting this as a
starting point, there is something else I want to say”. This is what we define as the core
meaning of well, namely, acceptance. It will be noted that this is very close to the
meaning ascribed to well by Fischer (2000a), who formulates it as “after all I know about
it I think this.” We want to go one step further and ask why speakers want to express this
meaning at particular points in the discourse. We have concluded from our research that
the multifunctionality of well arises from the fact that speakers may express acceptance of
a previous statement or of a situation for different purposes: to initiate a conclusion, a
disagreement, or a surprise, to name only a few. In all cases, however, the speakers attend
to heterogeneity in different forms: other possible expectations, other possible
conclusions, other possible reactions. Next to the core meaning (‘acceptance’) it is

therefore useful to posit a strategic function shared by all pragmatic uses within a
heteroglossic framework. This strategic function, encompassing the many more specific
context-bound uses, can be described as positioning the utterance vis-à-vis the hearer’s
real or suspected expectations. This general strategic function is what well shares with
other markers in the field, such as actually, in fact, as a matter of fact (see Smith and
Jucker, 2000, on the differences between actually and other markers with similar
functions).
6. The Translation Method
We agree with many authors that conversational analysis is the best basic methodology
for studying the function of pragmatic markers. However, it is generally difficult to
describe the pragmatic properties or functions of words and utterances. We would,
therefore, argue that additional methods of a more experimental type can yield deeper
insights or provide answers to questions that cannot be easily answered based on
conversation analysis alone. Fischer (2000a) has shown that varying the communicative
situation in an experimentally controlled way leads to differences in the occurrence and
use of certain pragmatic markers. This seems to provide a productive way to test
hypotheses about the dimensions of the communicative situation that the markers relate
to. In our contribution, we suggest an additional methodology that uses parallel translation
corpora of two or more languages.
The idea that translation corpora can be a valuable tool for exploring phenomena in a
source language has been convincingly defended in a number of publications (e.g., the
articles in Johansson and Oksefjell, 1998; Hasselgård and Oksefjell, 1999; and Aijmer and
Simon-Vandenbergen, forthcoming; see also Dyvik, 1998, 1999; Noël, 2003). The case is
especially strong in the study of pragmatic markers precisely because of their
underspecified core meaning and their polysemous nature. Translations of pragmatic
markers can therefore serve as a heuristic for discovering contextual dimensions or for
making more fine-grained divisions in these dimensions, because the translations force
one to account for the contextual factors that lead to particular choices. However, we
might ask whether we can rely on translations to reveal the cross-linguistic
correspondences. Translations are rarely literal renderings of the originals, but rather
reflect properties of either the source or target language. It is obvious that there are a
variety of reasons for a particular translation to be selected. Translators do not translate
words and constructions in isolation but rather choose a correspondence for a linguistic
element in a particular context. It follows that which words or constructions we regard as
correspondences between languages ultimately depends on the analyst’s own judgement.
Moreover, in the area of pragmatic markers it is particularly difficult to know what we

should consider a tertium comparationis, or common ground. We believe therefore that it
is important to start with a hypothesis of a core meaning, based on what we perceive to be
the similarities between the different translations of a discourse particle. In this way, the
translations become a heuristic for arriving at a tertium comparationis by allowing us to
check the hypothesized core meaning and, if necessary, to refine it (Altenberg and
Granger, 2002: 16). By looking at translations into a single language, we establish
unidirectional correspondences. Translations into multiple languages are interesting
because they show the closeness of target items to each other (see Fischer, 2000a) and
have what Dyvik (1998: 72) refers to as overlapping translational properties.
Aijmer and Simon-Vandenbergen (2003) used translations into Swedish and Dutch to
describe the polysemy of the English discourse particle well. They assumed that if a
particular meaning of well was mirrored in a translation (into Swedish or Dutch) it should
be explained by the monolingual analysis of the core meaning of well. The polysemous
nature of well was illustrated by the large number of translations into the target language
(50 translations into Swedish and 35 translations into Dutch; Aijmer and Simon-
Vandenbergen, 2003). For example, well was often translated as a response particle ja
(‘yes’), constituting 12% of the cases in the Sweden Parallel Corpus. Because of its
frequency, ja represents a prototypical translation of well. However, ja may also be used
in contexts where well could not be used. For instance, ja is also a backchannel item, and
it can be the answer to a yes/no question. Other functions are more specific and context-
dependent. In both the Swedish and the Dutch material, one could distinguish translations
focussing on emotion (tja, nja) rather than simple acceptance (ja ‘yes’).
Parallel translation corpora make it possible to reverse the translation process. A

translation equivalent of item X in language B can be studied in relation to its source
equivalents in language A. In other words, if X is translated by Y, we would expect Y to
have X as a potential source. The functional equivalents (bidirectional equivalents, since
they hold in both directions) can be assigned a correspondence value. In the unlikely case
that an item always corresponds to another item in the compared language, the
intertranslatability value would be 100% (Altenberg, 1999: 254). However, discourse
particles are not expected to reach a high degree of intertranslatability.
On the basis of the cross-linguistic method, we can also establish semantic fields in two or
more languages. A semantic field is defined by Dyvik (1998: 72) as “a large, vague
potential ‘sense’ which is not necessarily the sense of one sign, but rather the joint ‘sense’
of a set of semantically related signs”. In order to conclude that two items belong to the
same semantic field, it is not sufficient to look at translations in one direction only; one
must go back and forth from sources to targets. If item X in language A is translated by Y
and Z in language B, one can, by using B as source language, look for translation
equivalents of Y and Z (see also Ebeling, 2000: 17-18). This procedure will allow us to
show how the pragmatic marker X is related to other pragmatic markers, or to other
linguistic items such as modal particles or response words, in the same language.
A cross-linguistic analysis is needed to establish links between meanings in the languages

compared and to establish language-internal relations between elements in the source
language. The contrastive method also emphasises the need to explain how the core
meaning of pragmatic markers can explain the translations in the languages we have
contrasted with English. Fischer (this volume) points out that no matter how many
functions one finds it useful to distinguish, one has to be able to establish a link with their
core meaning. As Fischer notes, this entails that the range of functions carried by any
particular item is not arbitrary. Although some of the functions will also be fulfilled by
other pragmatic markers, one must be able to explain the difference between markers in
terms of their core meanings.
The frequent translations of well by response signals such as ja show that agreement is
very much part of the meaning of well. This confirms that the core meaning of well is
acceptance. On the other hand, it is clear that well has lost much of its meaning and can be
used to take stock of the situation when agreement must be sought in order to re-establish
common ground or when there is actual disagreement. The heteroglossic perspective
explains the fact that a marker of acceptance can be used strategically for different
interpersonal functions, including persuasion or politeness.
7. Conclusion
Particle research, which studies one of the most elusive aspects of natural language,
clearly struggles with a number of problems simultaneously: terminological, theoretical,
and methodological. Terminological work that is aware of the delicate balance between
functional and formal distinctions is still necessary. In this paper we have proposed to
take the functional view as a point of departure but to use both functional and formal
criteria for further subclassification. It is desirable, in order to arrive at a clearer picture of
this area of research, that researchers be explicit about the basis for their choice of terms.
On the theoretical level we need to create a framework that can accommodate the full
complexity of the function of pragmatic markers. We have proposed that reflexivity,
indexicality, and heteroglossia play a central role in a theoretical framework for pragmatic
markers. Reflexivity as a necessary property of pragmatic markers relates the item in
question to the utterance itself. Indexicality is optional but is a characteristic of the more
prototypical pragmatic markers. It relates the item to the context. Finally, all pragmatic
markers are heteroglossic in that they are strategies for positioning an utterance in the
context of other utterances, in either real or imagined discourse. Heteroglossia is thus the
characteristic which relates pragmatic markers to the hearer.
Finally, a broad range of methodologies is required to uncover the meaning and function
of markers. Both natural and experimental data are needed. Specifically, we believe that
comparing translations of a text in different languages can help to reveal the meaning of
markers which might be less accessible in a monolingual approach. The translation
method is useful not only in light of the need to move on to typological questions (how do
languages resemble and differ in this area?) but is also useful as a heuristic for
establishing contrastive semantic-pragmatic fields. Through closer study of such fields we
may arrive at insights into the question of multifunctionality and how it relates to
semantic and pragmatic polysemy. We have suggested in this paper that at least for some
pragmatic markers, it is desirable to define a core meaning and a general strategic
function in a heteroglossic framework. More empirical work on pragmatic markers in
different languages is needed, however, in order to test the explanatory power of the
model proposed here.
Notes
1
See Weydt, this volume, his thesis 4: “Every particle can be assigned a constant basic
meaning”.
2
On the problem of deciding whether a word or construction is polysemous or
underspecified, see Fabricius-Hansen (1999: 236).
3
Traugott (1999b) has shown how strategic uses, along with cognitive processes
constraining inferences, lead to semasiological change over time.
4
Weydt, this volume, denies that particles are deictic; see his section 2.
5
In her contribution to the present volume, Schiffrin shifts her terminology from planes to
domains.
6
We are aware that the terms polyphony, polyglossia, heteroglossia and dialogism have
been used in slightly different senses in different contexts (see the discussion on the
Bakhtin-Dialogism list bakhtin-dialogism@sheffield.ac.uk, May 2002). A clarification
of Bakhtin’s use of the term dialogism in its different senses is given by Roulet (1996).
Björklund (2000) points out that heteroglossia should not be confused with polyphony.
We follow White (1999) in using the term heteroglossia to refer to the speaker’s
explicit indication in his or her own utterance that different viewpoints exist.
Utterances in which no such markers are present are monoglossic.
7
The Oslo Multilingual Corpus is described on the website of the University of Oslo
(http://www.hf.uio.no/iba/prosjekt/). We wish to thank Stig Johansson for kindly
giving us access to the data for the purposes of this study. The explanation for the
source codes following the extracts can be found on the website.
Chapter 7
The description of text relation markers in the

Geneva model of discourse organization1
Eddy Roulet
1. Introduction
The problem of the description of the formal and functional range of discourse markers or
particles is so complex that I can treat it here only from a specific point of view, which is
slightly different from those of most papers in this volume, i.e., the elaboration of a global
model of the complexity of the organization of discourse, a model “which provides a
place to locate DMs within” (as opposed to models “whose only purpose is to explain
what DMs are”, to use Pons Borderia’s terms, this volume,). On the other hand, I will not
go deeply into the problem of the description of the core procedural meaning of discourse
markers or particles, which has been treated for French by Ducrot (Ducrot et al., 1980)
and Rossari (1994, 2000, this volume).
In my paper, after a brief presentation of the approach, methodology, and data, I will
propose (i) a definition of text structure and constituents, (ii) a definition of a subset of
discourse markers or particles which I call text relation markers (henceforth TRMs), (iii) a
description of a restricted set of generic text relations (henceforth TRs), and (iv) a
procedure for computing the specific TRs linking text constituents to information in
discourse memory according to the instructions given by TRMs. It seems to me that
points (i)) and (ii)) have been neglected too long in the study of discourse markers or
particles.
1.1. Approach
Our approach may be characterized as interactionist, top down, comprehensive, modular

and heuristic. It is interactionist and top-down following Bakhtin’s well-known
conception: “the methodologically based order of study of language ought to be: (1) the
forms and types of verbal interaction in connection with their concrete conditions; (2)
forms of particular utterances, of particular speech performances, as elements of closely
linked interaction—i.e., the genres of speech performance in human behavior and
ideological creativity as determined by verbal interaction; (3) a reexamination, on this
116 E. Roulet
new basis, of language forms in their actual linguistic presentation” (italics mine; in
Volosinov, 1973: 95-96).
Second, it is comprehensive as defined by Labov and Fanshel: “By comprehensive, we

mean that we have made ourselves accountable to an entire body of conversation,
attempting to account for the interpretations of all utterances and the coherent
sequencing between them” (italics mine; 1977: 354; see also Mann and Thompson, 1988:
243).
Third, it is modular, in the methodological sense defined by Simon (1962), Motsch

(1989), and Nølke (2000). As I said at the beginning of this paper, our research on
discourse markers is part of a larger project aiming at elaborating a general model of the
complexity of discourse organization (for the latest version, see Roulet, Filliettaz, and
Grobet (2001) and, for a brief presentation in English, Filliettaz and Roulet (2002)). Our
approach aims at capturing the complexity of discourse organization by combining
information referring to three distinct levels of analysis: the situational level, the textual
level, and the linguistic level. Presenting linguistic, textual, and situational factors as
distinct sources of information does not mean that we assume a clear-cut delimitation
between those components and that we subscribe to strict disciplinary boundaries such as
linguistics, pragmatics, or psychosociology. Discourse is rather defined essentially as a
combination of such phenomena and as a privileged locus for integrating the various
disciplines it belongs to. In order to do this, we adopt a modular methodology (see Nølke,
2000), which assumes that texts or talks may be analyzed in terms of different and
interrelated phenomena of varying complexity. We consider two distinct theoretical
categories, referring to two major steps in our analysis: (i) modules and (ii) organization
forms. Modules constitute the building blocks of a modular approach. Each module refers
to a restricted set of elementary information and circumscribes a specific domain of
discourse organization (the syntax, the lexicon, the textual hierarchy, the domain of
reference, the interactional materiality). Organization forms are complex units of analysis.
They result from what we call “couplings” between elementary information. For instance,
it is by combining lexical, hierarchical, and referential information that one can describe
the relational organization of a text or dialogue and, in particular, compute the specific
relation which is indicated by a discourse marker, as I will show later.
Finally, our approach is heuristic (such as the approach of Rhetorical Structure Theory;
see Mann and Thompson, 1988) and not formal (such as the approach of Segmented
Discourse Representation Theory; see Busquets et al., 2001), because I think that in our
present and still poor state of knowledge of the complexity of actual discourse, the main
priority is to find out how discourse is organized; it is premature to try to formalize this
organization.
1.2. Methodology
In our description of discourse markers, we combine three methodologies which have

been used by different approaches but appear to be complementary: (i)) the textual
analysis of relations between text constituents in actual discourses (see Roulet et al., 1991;
Mann and Thompson, 1988; and Marcu, 2000), (ii) the semantic analysis of the
instructions given by discourse markers or particles (see Ducrot et al., 1980; Blakemore,
1987; Schiffrin, 1987a; Rossari, 1994, 2000; Knott and Dale, 1994; Knott and Sanders,
Text relation markers in the Geneva model 117
1997), and (iii) the cognitive definition of the basic operations (and corresponding
relations) required by the joint construction of discourse (following initial suggestions by
Sanders, Spooren, and Noordman, 1992a, but reinterpreted in an interactionist approach;
see Clark, 1996); as has been shown by Bateman and Rondhuis (1997) and Knott and
Sanders (1998), the results of these different approaches are complementary and largely
convergent. In addition, I use a simple model of inference to compute the specific
relations indicated by discourse markers in actual discourse.
1.3. Data
We have been working for twenty years on all types and genres of authentic discourses,
monological and dialogical, oral and written, literary and nonliterary. By authentic, we
mean pieces of discourses which have not been fabricated or domesticated (to use
Schegloff’s terminology) to answer the analyst’s needs. See Roulet (1999); Roulet (2000);
Roulet, Filliettaz, and Grobet (2001); Kuyumcuyan (2001); Filliettaz (2002); Grobet
(2002); and Simon (2003) for comprehensive and detailed analyses of phone calls, service
encounters, interviews, debates, film and theater dialogues, narratives, letters, etc.
My paper presents an approach to discourse markers or particles centered on the following

issues, or challenges. First, if one admits that the description of discourse markers or
particles is only a small part of the description of the overall and very complex
organization of discourse, the study of discourse markers or particles must be integrated in
a global model of the complexity of the organization of monological and dialogical
discourse. Yet most of the models that have been developed so far are either restricted to
the description of discourse markers or particles or integrated into models of discourse,
such as Argumentation Theory, Conversational Analysis, Relevance Theory or
Argumentation Theory, that take into account only some types of discourse, monological
or dialogical, and some aspects of discourse organization.
Second, if one admits that discourse markers or particles apply to discourse constituents
or relate discourse constituents to information stored in discourse memory (as shown by
Berrendonner, 1983, 1990),2 it is not enough to give a precise definition of the class of
markers or particles, a recurrent and well-motivated stance in most studies, as a
preliminary to an adequate description. It is necessary to give also a precise definition of
the units and pieces of information to which discourse markers or particles apply or which
they relate at different levels,3 which implies a hierarchical model of discourse structure
and a concept of discourse memory. Yet this question is rarely treated explicitly (even in
the papers in this volume) and most studies are based either on units that are ill defined or
very difficult to define, such as utterance, turn constructional unit, text span, or units
which are clearly inappropriate such as clause, proposition, speech act (for a detailed
criticism, see Roulet, 2002a).
Finally, if one admits, with Roulet et al. (1991), Mann and Thompson (1988), and Marcu
(2000), that the description of general rhetorical relations is an important part of the
analysis and interpretation of a text, and, as in Relevance Theory, that there can be only a
unique and specific relation between a given text constituent and contextual information
118 E. Roulet
(see Blakemore, 1987), an adequate model for the description of discourse markers or
particles has to define, presumably on an interactionist cognitive basis, a set of generic
rhetorical relations (and, as a corollary, different classes of discourse markers) and to
provide an instrument to compute the specific relation linking each text constituent with
information in discourse memory (or contextual information, to use the terminology of
Relevance Theory).
2. Definitions
2.1. Definitions of text units
Our conception of text structure is based on the hypothesis that the construction of any
verbal interaction or written text reflects a process of negotiation in which
speakers/writers recursively initiate, react on, or ratify propositions by means of text
constituents belonging to various hierarchical levels: exchanges, moves, and acts. Thus,
our model postulates a recursive hierarchical structure which is governed by two distinct
completion principles. The principle of dialogical completion states that an exchange
comes to an end when both interactants agree about the closure of a negotiation process.
As for the principle of monological completion, it states that each move constituting an
exchange should provide sufficiently relevant information in order to function as an
adequate contribution to a negotiation process. This explains why moves are frequently
formulated by means of a complex sequence of acts, moves, and subordinate exchanges.
The hierarchical structure is defined by rules which specify how an exchange can be
analyzed into moves, and a move into a main act, possibly accompanied by exchanges,
moves, and acts which are subordinated to it (see Roulet, Filliettaz, and Grobet, 2001, ch.
3).
The minimal textual constituent, which we call text act, is not defined in linguistic but in
cognitive and interactionist terms as the smallest unit of a negotiation process; as such, it
is characterized by two features which have been described independently but which are
manifestly linked. First, the information given by a text act must be conceptually
independent (as shown by Schilperoord and Verhagen, 1998);4 for example, a main clause
that is followed by a restrictive relative or a complement clause is not conceptually
independent and is not a distinct text unit, whereas a main clause followed by an adverbial
clause is a distinct text unit; thus, the following sequence, which consists of three clauses
is analyzed in two text acts:
(1) Bien que ma voisine soit déjà riche, elle espère toujours qu’elle va gagner le gros
lot.
Although my neighbour is quite rich, she always hopes that she will win the
jackpot.
Second, a text act may be registered in discourse memory, which is attested by the
possibility of replacing by a definite expression in the following clause an anaphoric
pronoun that refers to an element in the first clause (as shown by Berrendonner, 1990).
Thus, in our example, one can replace elle in the second clause by la brave femme, which
confirms that the first clause is a text act; it is impossible for elle in the third clause:
(2) Bien que ma voisine soit déjà riche, elle (la brave femme) espère toujours qu’elle
(*la brave femme) va gagner le gros lot.
In order to explain the use of la brave femme, we must hypothesize that the information
given by the first clause has been stored in discourse memory before the uttering of the
second. If it is possible to substitute the first anaphoric pronoun by a definite expression
without losing coreference, it is not the case for the second anaphoric pronoun. This
confirms that this sequence consists of two text acts and that the border between them is
before the first anaphoric pronoun.
As the same test applies to left-dislocated noun phrases which are not complement of the
verb (as in (3)), they must be considered as distinct text acts, as shown by Auchlin (1993):
(3) Ma voisine, je ne veux plus entendre parler d’elle (de cette brave femme)
My neighbour, I don’t want to hear any longer about her (this brave lady).
It means that a text act should not be confused with the traditional speech act, as it does
not necessarily have an illocutionary function and need not coincide with a clause: a
subordinate act expressing a counterargument can be expressed by a prepositional phrase
(malgré la pluie ‘in spite of the rain’), which has no illocutionary function, as well as by a
clause (bien qu’il pleuve or il a beau pleuvoir ‘although it is raining’) (see Rubattel,
1987).
For a more detailed presentation and illustration of our methodology for text
segmentation, including other criteria, see Roulet (2002a); Simon (2003); and Roulet,
Filliettaz, and Grobet (2001, ch. 3).
2.2. A definition of text relations and text relation markers
I do not use the label particle, because it does not allow one to take into consideration all
the markers that play a role in the organization of discourse. It is not appropriate, even as
a generic label, because it usually relies on a formal linguistic criterion (i.e., uninflected
words), which excludes important text relation markers such as the three basic syntactic
structures (declarative, interrogative, and imperative as markers of illocutionary relations),
verbs (e.g., performatives such as demander or modals such as pouvez-vous as markers of
illocutionary relations, or avoir beau in French, as marker of a counterargumentative
relation). It seems more appropriate to use a functional characterization of the set of
markers which play a role in the organization of discourse.
Thus, as many other researchers, I use the functional label discourse marker or, more
precisely, discourse organization marker. But this category includes markers of different
aspects of discourse organization: interactional (particles such as tu sais), polyphonic
(speech verbs), topical (anaphoric pronouns), periodical (punctuation), etc. I therefore use
the more restricted notion of discourse relation marker. As I distinguish three types of
discourse relations, corresponding to the three basic components of our discourse model:
linguistic (syntactic and semantic relations), textual (textual relations), and situational
(praxeological relations) (see Roulet, 2002b), and as I intend to consider in this paper only
textual relations (TRs), I will use the label text relation marker (TRM). I do not use
120 E. Roulet
connecteur ‘connective’, which is common in French studies, because it designates only a

subset of TRMs that implicitly excludes verbs and syntactic structures.
The use of the term marker has given rise to long discussions between the authors of the
papers in this volume. As I will show later, the instructions given by the forms described
in this volume just help the interactants to compute a TR; it might thus be more
appropriate to use a term which is less strong than marker, something like indicator, to
follow Hossbach’s (1998) suggestion, or text relation pointer, but marker is so commonly
used in the literature that it is difficult to change the terminology.
Although this study does not take into consideration many discourse particles, such as tu
sais ‘you know’, that are analyzed in other papers in this volume and that might be
included in the hyperonymic category discourse marker under the label discourse
interactional or interpersonal markers, it should be considered as complementary to these
studies.
Referring to the hierarchical structure of text presented in 2.1, we distinguish two types of
TRs in the relational organization of text: illocutionary (or dialogical) relations that
concern exchange constituents (and not speech acts, as in speech act theory) and
interactive (or monological) relations that concern move constituents (see Roulet,
5
Filliettaz, and Grobet, 2001, ch. 6).
This does not mean that we define TRs, as many researchers do, as relations between text
constituents. It is not appropriate to restrict TR to relations between text segments, as is
shown by the use of TRMs at the beginning of a text or by the use of après tout in the
following example:
(4) Je n’irai pas au cinéma; après tout, je suis trop fatigué

I will not go to the pictures; after all, I am too tired.
In this move, après tout does not indicate the argumentative relation between the second
and the first act (a relation that could be marked by car; see je n’irai pas au cinéma car,
après tout, je suis trop fatigué); après tout indicates a relation of reformulation between
the second act and a piece of information in discourse memory such as “I thought I felt
well enough” (see Roulet, 1990). Therefore, following Berrendonner (1983), we define a
TR as a relation between a constituent of the hierarchical structure of text—act, move or
exchange—and a piece of information stored in discourse memory (this information may
have its origin in the preceding constituent, in the immediate cognitive environment, or in
our world knowledge). Thus a TRM can be defined as a linguistic form (lexical or
syntactic) which indicates an illocutionary or interactive relation between a text
constituent and a piece of information stored in discourse memory and which gives
instructions in order to facilitate the access to the relevant information. As shown by
Ducrot et al. (1980) and Blakemore (1987), TRMs have a procedural meaning, which has
to be described in the lexicon or syntax of any language.
3. A model for the description of the functional spectrum of text relations and text
relation markers
3.1. The definition of generic text relations
The definition I have given of a TRM requires a more precise characterization of the set
of TRs. In order to avoid the two risks of the proliferation of ad hoc relations and of the
mere listing of relations specific to a particular language (as a result of the analysis of
performative verbs and cue phrases),6 I hypothesize that it is possible to define a
restricted, finite, and universal set of generic TRs which is based on the basic operations
required by the satisfaction of the dialogical and monological completion constraints
mentioned above. We thus propose in Roulet, Filliettaz, and Grobet (2001, ch. 6) and in
Roulet (2002b) a system of ten generic TRs that is sufficient to describe the generic
relational organization of any text or dialogue.
At the exchange level, I distinguish two generic illocutionary relations, initiative and
reactive. The first move of an exchange is linked to the second by an initiative
illocutionary relation, the last move of an exchange is linked to the preceding one by a
reactive illocutionary relation, and each intermediate move is linked to the preceding one
by a reactive illocutionary relation and to the next one by an initiative illocutionary
relation. The category generic initiative illocutionary relation covers different specific
illocutionary relations as they have been described by speech act theory: question, request,
offer, promise, etc.
Most moves present a complex structure; in order to satisfy the monological completion
constraint, the speaker/writer may have to introduce a discourse object that will be the
topic of the following constituent to prepare his or her main act, to ground it (by
introducing arguments or rejecting counterarguments), to comment on it, or to reconsider
its formulation, which motivates the following generic interactive relations:
topicalization, preliminary, argument, counterargument, comment, reformulation.7 She or
he may have to link the successive events of a narration by the generic interactive relation
succession, and finally, if the addressee finds that the speaker/writer does not satisfy the
monological completion constraint, he or she may open a subordinate exchange linked to
the preceding move by the generic interactive relation clarification.
The categories I use are generic because each one covers a set of specific TRs, and the
labels chosen are partially arbitrary. For instance, argument is a generic category which
covers the following specific relations described by various researchers: cause (volitional
and nonvolitional), explanation, justification, motivation, evidence, consequence, purpose,
result (volitional and nonvolitional), argument, potential argument (as marked in French
by si), polyphonic argument (puisque), decisive argument (même), accessory argument
(d’ailleurs) or minimal argument (au moins), exemplification (par exemple, ainsi).
It is interesting to note that the generic categories resulting from an interactionist top-
down approach are quite different from those, such as cause, condition, concession, or
contrast, resulting from a traditional grammatical bottom-up approach, which are not quite
satisfying, since researchers have observed many overlappings between them (see
Couper-Kuhlen and Kortmann, 2000). In our classification, si, puisque, and bien, or mais,
122 E. Roulet
bien que, and même si indicate the same generic TR, respectively, argument and counter-
argument.
3.2. The description of generic TRs
The description of generic TRs in a monological or dialogical text is based on two

criteria:
1. the hierarchical structure, which defines text constituents and their dependency
relations;
2. the presence of (or the possibility of inserting) a TRM belonging to a certain class.
Generic initiative and reactive illocutionary relations can be identified by the position of
the move in the exchange structure. Generic interactive relations are for the most part
characterized by the presence (or the possibility of inserting) a TRM. Some relations, such
as preliminary, comment, and clarification, are characterized by the position of the
constituents in the hierarchical structure. The following list indicates the most common
markers in French for each relation:
1. argument : parce que, puisque, car, comme, même, d’ailleurs, si, alors, donc, par
conséquent, pour que, de sorte que, au moins, à cause de;
2. counterargument : bien que, quoique, quel . . . que, quelque que, même si, mais,
pourtant, néanmoins, cependant, quand même, quand bien même, seulement, avoir
beau, malgré (que);
3. reformulation : en fait, de fait, au fond, en tout cas, de toute façon, enfin, finalement,
après tout, en somme, somme toute;
4. topicalization : quant à, en ce qui concerne, or left dislocation;
5. succession : puis, ensuite, après que, dès que;
6. preliminary : no specific marker; the subordinate precedes the main constituent;
7. comment : no specific marker; the subordinate follows the main constituent;
8. clarification : no specific marker; it is the relation between a main constituent and the
following subordinate exchange, if it opens with an interrogation.
It is important to note that these classes bring together markers that indicate the same
generic relation regardless of their form (coordinate conjunction such as mais, subordinate
conjunction such as bien que, adverbial phrase such as quand même, verbal phrase such as
avoir beau, preposition such as malgré) and of the text constituent they mark (for
instance, bien que, quand bien même, or avoir beau indicate a counterargument relation in
the subordinate constituent, whereas mais, néanmoins, or quand même indicate the same
generic relation in the main constituent).
Using this classification, one can proceed to the description of the generic TRs that hold
between the constituents in a given text. Let’s take as an example the following extract of
a dialogue between a travel agent (T) and a customer (C) which is presented and analyzed
in Roulet, Filliettaz, and Grobet (2001, ch. 3); the travel agent has just offered a plane
ticket to Barcelona at a price that seems reasonable:
(4) T: est-ce que ça pourrait aller pour vous //

would you agree with this proposal
C: j’trouve
I think
(5) ça m’embête un peu comme ça /

it bothers me a little bit that way
parce que j’avais pris heu cet été un vol aller simple /
because I took last summer a one way flight
c’était cent nonante huit que j’avais payé \\
I paid one hundred and ninety eight
finalement c’est même plus que le double / .
finally it is even more than the double
c’est à dire plus que le double pour l’aller et retour /
that is more than the double for a return flight
(6) T: cent nonante huit //

one hundred and ninety eight
C: oui /
yes
j’avais payé ça . l’aller simple \\
that’s what I paid for a one way flight
I propose the hierarchical structure in Fig. 1 below (for a detailed explanation, see Roulet,
Filliettaz, and Grobet, 2001, ch. 3).8 I hypothesize that there always is a TR between two
constituents at the same level (or, more precisely, between the second constituent and a
piece of information in discourse memory which has its source in the preceding
constituent), be it marked by a TRM or not. It is convenient to represent the TRs on the
hierarchical structure (getting a representation which resembles the ones by RST: for a
precise comparison, see Roulet 2002c; for a detailed explanation, see Roulet, Filliettaz,
and Grobet, 2001, chap. 3).
However, as a TRM, such as finalement or même in (11), may indicate a relation between
a text constituent and a piece of information in discourse memory that does not have its
source in the preceding constituent, we have to go beyond the constituents in hierarchical
structure to be able to describe all TRs. In our example, we have to intuitively reconstruct
the piece of information in discourse memory reformulated by finalement c’est plus que le
double, something like “it seemed reasonable”, and the information in discourse memory
for which c’est même plus que le double is an argument, something like “it is very or it is
too expensive”. We might use a specific mode of representation for the overall relational
organisation of texts, using predicates and arguments, as I did in Roulet (2000), but it very
quickly becomes difficult to read when there are too many embeddings. That is the reason
why I prefer to adapt the hierarchical structure tree by introducing DM branches for
pieces of information in discourse memory that do not have their source in a text segment,
as shown in Fig. 2.
124 E. Roulet
Figure 1. Hierarchical structure A
M est-ce que ça pourrait aller pour vous
mA (7) j’trouve
E mM
sA (8) ça m’embête un peu
A (9) parce que j’avais pris euh cet été . . .

sM
M
mM A (10) c’était 198 que j’avais payé
A (11) finalement c’est même plus que le double

mM
A (12) c’est-à-dire plus que le double pour . . .

sM
I (13) 198
sE
mA (14) oui
I
sA (15) j’avais payé ça . . .

Figure 2. Hierarchical structure B
sA (7) j’trouve
pre
mM
mA (8) ça m’embête un peu comme ça
A (9) j’avais pris euh cet été …

M sM
arg
A (10) c’était 198 que j’avais payé
mM
DM (it looks cheap)
A A (11) finalement c’est même
arg plus que le double
mM A
SM ref
arg DM (it’s very, too expensive)
parce
que A (12) c’est-à-dire plus que le double pour …
M (13) 198
IN
sE
mA (14) oui
clar M
RE
sA (15) j’avais payé ça …

arg
/en effet/
We thus obtain a first global description of what I call the relational profile of this
exchange, which corresponds to one possible interpretation of this dialogue and gives
interesting information concerning dominant TRs. But we cannot satisfy ourselves with
this description, that resembles RST representations, for at least three reasons:
• it does not explain how we got to the proposed interpretation;

• in particular, it does not explain how the information in discourse memory was
selected;
• it does not account for the specific TRs due to the use of particular TRMs, giving
specific instructions.
126 E. Roulet
We thus have to complete this intuitive description of the generic TRs in this exchange by
showing how one can compute specific TRs, according to the instructions given by
TRMs.
3.3. The computation of specific TRs
The description that we have given so far does not allow for a precise description of either
the specific TRs (which may be indicated or not by TRMs) between a text constituent in a
given text or dialogue and information in discourse memory, or the selection of the
relevant piece of information in discourse memory. In order to do so, we have to appeal to
a third component, beside hierarchical and linguistic information: referential (or
contextual) information, i.e., information in discourse memory having its source in the
world knowledge, the immediate cognitive environment, or the preceding text.
In order to determine the specific relation linking a text constituent to information in

discourse memory, we have to use one of the following procedures, depending on the
presence or absence of a TRM. If there is no TRM between two text constituents, the
specific TR must be computed inferentially by combining information given by both
constituents on the basis of our world knowledge. For instance, we have hypothesized an
argument relation between the move formed of the coordinated acts (9)/(10) and act (11),
a hypothesis confirmed by the possibility of introducing a TRM as alors or donc. This
argumentative text relation is based on the referential (or contextual) knowledge that if the
customer paid 198 francs for a single flight in summer, then the ticket proposed by the
travel agent costs more than the double.
We can compute the relation by combining linguistic and contextual information in the
following way, using a simple model of inference that links premisses to a conclusion:
Table 1. Computing a specific relation
premise 1 linguistic information C says to T that C paid last

(enriched logical form) summer 198 for a one-way flight
premise 2 linguistic information C says to T that T proposes more

(enriched logical form) than 400 francs for a return flight
premise 3 easiest contextual information if a return flight costs 400 francs,

accessible in discourse memory then paid it is more than the
double of the price paid by C for a
single flight
conclusion interpretation C says to T that his proposal is
more than the double because C
paid 198 for a one way flight
(there is an argument relation
between both acts).
This presentation of the computation of a TR is informal. As I said at the beginning of this

paper, our approach is heuristic; it should allow us to better understand the processes
involved in the organization and interpretation of discourse. In agreement with RST and
in contrast with SDRT, I think that we do not yet have the instruments that would allow us
to formally compute the interpretation of authentic discourse segments, i.e., segments that
have not been fabricated by the analyst.
If a text constituent includes a TRM, we have to use the instructions delivered by this
marker as they are given in the lexicon, in order to (i) select the relevant information in
discourse memory to which the constituent may be linked, and (ii) compute the specific
TR between the constituent and this information in discourse memory. If the relevant
piece of information has its source in the preceding text constituent, one still has to
compute whether the relation concerns the content, the illocutionary act or the uttering act,
to refer to the distinction introduced by Ducrot as early as 1975 (see Groupe lamda-l,
1975).
We do not have a precise and unified description of the instructions delivered by TRMs in
any language. As far as French is concerned, we have descriptions for most argumentative
and counterargumentative connectives (see Ducrot et al., 1980; Roulet et al., 1991;
Moeschler, 1989; Rossari and Jayez, 1997; Rossari, 2000) and for some reformulative
connectives (see Roulet, 1990; Rossari, 1994; Hossbach, 1998; and Philippi, 1999), but
they are presented in different theoretical frameworks and different formats. So, in our
present state of knowledge, I have to content myself with an informal formulation of these
instructions.
As an example, I can now compute the TRs indicated by même and finalement in (11);
let’s first examine the contribution of même, following the description given by Ducrot in
Ducrot et al. (1980, 12-13):
Table 2. Computing a specific relation with même
premise 1 linguistic information C says to T that the price

(enriched logical form) proposed by T for the return flight
ticket is more than the double of
the price paid by C for a one way
ticket
premise 2 linguistic information if one says même x, it is in order
(instruction given by même ) to present x as a decisive
argument for a conclusion y
premise 3 easiest contextual information that the price is more than the
accessible in discourse memory double is a decisive argument for
saying it is too expensive
conclusion interpretation if C says (11) to T, it is as a
decisive argument for saying that
it is too expensive.
128 E. Roulet
Let’s now describe the contribution of finalement. It is more problematic, since finalement
has different uses, propositional and nonpropositional (like toujours, described by
Hansen, this volume), which have not been systematically described yet. Finalement may
be used in the following examples:
(16) Il a finalement réussi son examen or

Finalement, il a réussi son examen.
He finally passed his exam.
(temporal propositional use)
(17) D’abord, tu prends 1 litre de lait, puis une livre de farine, finalement tu mélanges
les deux.
First, you take 1 liter milk, then 1 pound flour, finally you mix both. (temporal
textual use, marking last text act)
(18) Finalement c’est plus que le double

Finally, it’s more than double.
(reformulative textual use).
It is interesting to observe that, in the first and the second examples but not in the third
one, finalement can be replaced by enfin without any change of meaning. For the time
being, as my aim here is not to give a systematic analysis of this form, I propose to
describe the instructions given by finalement in the following way, taking into account
suggestions by Luscher and Moeschler (1991) concerning enfin and by Rossari (1994: 30)
concerning the difference between enfin et finalement. Finalement, like enfin, indicates
either the last action of an event or the last act of a textual sequence; if neither is relevant,
it indicates the reformulation resulting from the last analysis of a problem (such as, in this
dialogue, the evaluation of the travel agent’s proposal); but, contrary to enfin, which
presents the reformulation as the result of a global reexamination of the problem,
finalement presents it as the result of the successive reexamination of the divergent
components of the problem (such as, in this dialogue, the travel agent’s proposal and the
price paid last summer for a single flight).
Following Hansen (this volume), I thus adopt a polysemy approach to the description of
TRMs, since the main uses of a lexical item such as finalement are defined by distinct but
related instructions: to indicate the last action of an event, the last act of a textual
sequence, or the last step of a reexamination process. The decision to apply one
instruction rather than another in actual discourse is determined by the principle of
optimal relevance.
We can now compute the specific text relation indicated by finalement as shown in Table
3 below. One can proceed in the same way to compute specific illocutionary relations, for
instance the illocutionary initiative question relation, indicated by est-ce que, that links the
first move to the second. This method allows the analyst to informally compute the whole
range of specific TRs in actual discourses by combining linguistic and contextual
information according to the instructions given by TRMs (if present). For instance, we
can compute the whole range of specific argument relations indicated by French TRMs:
decisive argument (même), accessory argument (d’ailleurs), potential argument (si),
polyphonic argument (puisque), minimal argument (au moins), purpose (pour que), result
(de sorte que) etc. We can use the same method to explain the multiple uses of mais (but)
described in the literature. Moreover, the method allows the analyst to select the piece of
information in discourse memory to which the text constituent is linked; it is important in
particular for the TRMs that indicate a simple argument, such as parce que or car,
because it explains the possibility to link up the constituent introduced by parce que or
car to the content, the illocutionary, or the uttering dimension of the last act stored in
discourse memory.
Table 3. Computing a specific relation with finalement
premise 1 linguistic information C says to T that the price

(enriched logical form) proposed by T for the return flight
ticket is more than the double of
the price paid by C for a one way
ticket
premise 2 linguistic information if one says finalement x, one
(instruction given by finalement) presents x either as the last action
of an event) or as the last act of a
text; if neither is relevant, x is
presented as a reformulation of
the speaker’s point of view that
results from the successive
reexamination of the divergent
components of a problem
premise 3 contextual information the content of x (it is more than
the double) does not express an
action
premise 4 contextual information it does not seem relevant to just
present x as the last act of the text
premise 5 contextual information C is confronted with divergent

aspects of an offer, that is
presented by T as advantageous,
but looks expensive compared to
C’s last flight experience.
conclusion interpretation C presents x as a reformulation
resulting from the successive
reexamination of the divergent
components of the offer.
130 E. Roulet
The study of TRMs in a global approach to discourse organization is a necessary

complement to lexical analyses of TRMs, such as those presented for French by Rossari
(2000, this volume), Philippi (1999), or Hansen (this volume). It can bring an important
contribution (i) to discourse analysis, mainly to a precise description of generic and
specific TRs and to an enrichment of the interpretation of texts and dialogues (for French,
see Ducrot et al., 1980; Roulet, Filliettaz, and Grobet, 2001; Kuyumcuyan, 2001); (ii) to
the description of the interaction between lexicon, syntax, and discourse structure, as the
same TR may be expressed by linguistic forms belonging to different syntactic categories,
for instance by a coordinate conjunction such as mais (il pleut, mais elle sort), a
subordinate conjunction such as bien que (bien qu’il pleuve, elle sort), an adverbial phrase
such as quand même or quand bien même (il pleut, elle sort quand même; quand bien
même il pleut, elle sort), a verbal phrase such as avoir beau (il a beau pleuvoir, elle sort),
or a preposition such as malgré (malgré la pluie, elle sort) that impose specific syntactic
constraints on the production of discourse. It constitutes thus a privileged field for the
study of grammaticalization processes. Finally, as is convincingly shown by Waltereit
(this volume), the study of discourse markers as “traces of the strategic use speakers make
of them in order to attain a certain effect related to discourse structure” offers a very
interesting frame for the diachronic description of the development of discourse markers.
Notes
1
I thank Laurent Filliettaz, Anne Grobet, and Corinne Rossari for their very helpful
comments on an earlier draft of this paper and Catherine Walther Green for a careful
proofreading.
2
Hansen (this volume) uses the terms developping model of the discourse or mental
discourse model under construction to translate mémoire discursive. Following
Berrendonner, she defines the discourse model under construction as containing
“information gleaned, among other things, from previous utterances, but also
information from the non-linguistic context, as well as contextually relevant
encyclopedic knowledge”.
3
This point is made quite explicit by Schilperoord and Verhagen (1998 : 142): “a
prerequisite for the analysis of discourse coherence is that the basic elements, or
discourse segments, between which coherence relations hold, are identified by the
analyst”. See also Pons Borderia (this volume).
4
“In brief, one clause is conceptually dependent upon another clause, if its semantics
cannot be conceptualized without essential reference to the conceptualization of
another clause” (p. 148).
5
The term interactive introduced in Roulet et al. (1985) to characterize monological
relations (considered as relations between acts) is surely unfortunate for at least two
reasons: first, relations are no longer defined as relations between acts; second, the
label interactive is commonly used today in a much broader sense, but we keep using
interactive as it is now common in French discourse studies. The term rhetorical used
by Rhetorical Structure Theory might be more appropriate.
6
The RST website (http://www.sil.org/linguistics/RST) currently gives a list of 30
monological relations and the recent extension of SDRT to dialogue analysis results in
an open list that already counts 15 relations (see Asher et al., 2001).
7
Counterargument is a technical term which covers the interactive relation between a
constituent introduced by bien que ‘although’, or preceded by mais ‘but’ and the main
constituent which follows; it should not be confounded with its common use, as a
synonym of objection.
8
E = exchange, M = move, A = text act, m = main, and s = subordinate.
Chapter 8
A dynamic approach to discourse particles
Henk Zeevat
1. Introduction
In this paper I discuss a formal semantic/pragmatic account of discourse particles. I will

deal only with a subclass of these particles and will limit the discussion to one possible
approach. It may well be that the approach can be applied to other particles as well or that
it can be applied to other expressive devices such as certain intonational patterns (e.g.,
contrastive stress), morphemes (past tense, agreement), words (pronouns), or
constructions (e.g., some uses of definite descriptions, clefts), but I will not try to show
that here.
The approach in this paper is a departure from my earlier treatment in Zeevat (2002) of
some particles within an optimality-theoretic presupposition theory (Blutner and Jaeger,
1999). That theory is a reconstruction of the standard dynamic accounts of presupposition
due to Heim (1983) and Van der Sandt (1992). My treatment develops an explanation
why particles like too, doch, and indeed do not give rise to the accommodation of their
presupposition, why they cannot be omitted in the utterances in which they occur, and
why their antecedents have an epistemic status that can be much weaker than being
components of the common ground between speaker and hearer.
The advantage of the treatment in this paper is not so much that it gives a better account
of the particles in question but that it generalizes better to other particles and that it is
more economical. There are more particles that can be seen as context markers than as
nonstandard presupposition triggers. More comprehensive treatments of particles would
be possible by developing the notion of a speech act marker within a framework of speech
act semantics. I sketch some of the issues involved in that in the final section.
1.2. Data and method
This paper does not present new empirical data. The old data that I use are some well-
known observations on the English particle too (see Kripke, undated), on the Dutch and
134 H, Zeevat
German particles (respectively) toch and doch (see Karagjosova, 2001), and related
particles (Zeevat, 2001).
In formal semantics and pragmatics, one tries to find formal linguistic and logical models
that explain the intuitively valid inferences between utterances. For discourse particles,
this implies that one wants to find an explanation of the inferences that they cause when
they are there and that are not there when they are absent. They constitute a special
problem, since there is agreement that many of them do not have a bearing on truth
conditions. This has meant that discourse particles have been studied as part of notions
like Farbe or Beleuchtung (a notion due to Frege that includes what we nowadays call
connotation), as conventional implicatures, or as elements that lead to special correctness
conditions of a pragmatic nature.
In Grice (1975), conventional implicatures are primarily a category where implicatures

should be put that are not conversational. Grice’s test for conversational implicature is
detachability: the implicature should not depend on the choice of a particular word or
construction. Conventional implicatures thereby must depend on the occurrence of
specific words or constructions. There is not however a substantial theory of why certain
words or constructions have conventional implicatures, not in Grice or anywhere else.
Karttunen and Peters (1979) make conventional implicatures identical to presuppositions.
It is clear however—and later on we will discuss examples—that there are other
conventional implicatures. Stalnaker’s (1978) idea of pragmatic correctness comes a bit
closer to a theory of what is going on with discourse particles and is close to the view I
am trying to express in this paper, but it also does not explain why natural language has
developed particles.
Even, too, also, doch, etc. can make the utterances in which they appear pragmatically
incorrect, but they can never make the sentence false. Dynamic semantics of the kind that
has been developed for the treatment of anaphora and presupposition seems to fare better,
since some of the particles have traditionally been described as presupposition triggers.1
In dynamic semantics, meaning becomes a function from an old information state (the
common ground, what speaker and hearer have already established to be common
knowledge between them) to a new information state (the new ground consisting of the
old information state together with the information conveyed by the current utterance).
Discourse particles then express conditions on the old information state. If the conditions
are met, the update is defined. If they are not met, the update is impossible or—in a more
liberal view—the update leads to an error message. But this is not an unproblematic view.
Why have devices that make a generally possible update impossible or (in the second
view) flawed? We need to say more about particles than just that they have this property.
The problem of discourse particles is therefore the characterization of their semantic or

pragmatic contribution to the utterances in which they occur. This is not just a puzzle in
pragmatics, but it is one that bears on the concept of pragmatics as such. I will conclude
that not even the dynamic notion of meaning is sufficient for explaining particles and that
the proper notion of the meaning of an utterance for characterizing particles is that of the
speech act analyzed in terms of conditions under which it can be carried out, the effects
that are achieved if the act is taken seriously by the hearer together with the effects that
the speaker intends to achieve. In this view, discourse particles are tools to indicate that
A dynamic approach to discourse particles 135
for the conditions or intended effects settings other than the default apply. The view is
similar to that taken in König and Requardt (1991).
2. Definitions
The particle too has occupied a central place in the presupposition literature, both before
and after Kripke (undated) on this particle. This argument is directed at the view of
Karttunen (1974) that a presupposition must be true in the context of an utterance of a
sentence that contains a presupposition trigger that triggers it if it is not filtered away or
stopped by a plug (filters are operators that let through some but not all of the
presuppositions of their arguments; plugs are operators that let none of them through).
This condition is always met by the simple context of the trigger, like the one in (1).
(1) John will have dinner in New York too.
What is the presupposition? If John carries so-called contrastive stress, it is the statement
that somebody different from John will have dinner in New York. Now, New York has
many inhabitants and most of them have dinner there every night. In addition, everybody
knows that. So in a normal context of utterance, Karttunen’s theory (and similar theories
like Gazdar, 1979; Heim, 1983; Stalnaker, 1973; and Van der Sandt, 1992, run into the
same problem) predicts that the particle too cannot change the felicity of the utterance,
because its presupposition is trivially met. But it does matter, as Kripke observes. The
sentence is infelicitous if the previous conversation has not mentioned another person who
will have dinner in New York.
One can try to escape from Kripke’s argument by assuming a different presupposition,
e.g., x is a person different from John who will have dinner in New York. This is an open
formula and can only be satisfied by finding a binder for the x in the context: it is very
much like a pronoun. This has been proposed by Geurts and Van der Sandt (2001) in the
context of a discourse representation theory and is compatible with Heim’s approach. A
problem is then that presupposition triggers in these theories generally allow the
possibility of accommodation and that the most natural way for applying accommodation
in this case leads to regaining the original problematic presupposition: there is somebody
apart from John who will have dinner in New York, too. Geurts and Van der Sandt
remedy this problem by treating the free variable as a proper pronoun and argue that since
pronouns do not accommodate “because they lack descriptive content”, this hidden
pronoun in the presupposition triggered by too does not accommodate, either. (Pronouns
indeed do not accommodate their antecedents and have little descriptive content. But not
less than the man or the woman, which, according to these authors, accommodate freely.)
This however still allows for partial accommodation: resolve the pronoun to some known
entity and accommodate that the person will have dinner in New York.2
(2) A man is walking in the park. John will have dinner in New York too.
(2) could (must, under the assumptions of Geurts and Van der Sandt) be treated by
resolving the pronoun from the presupposition triggered by too to the walking man in the
first sentence and by accommodating the remaining part of the presupposition, making it
equivalent to (3). This prediction is not correct.
136 H, Zeevat
(3) A man is walking in the park. He will have dinner in New York. John will have
dinner in New York too.
The assumption that pronouns do not accommodate because of a lack of descriptive

content leads to other problems as well. The particle indeed (or the Dutch immers, roughly
‘as you know’) presupposes the sentence in which it occurs and thus has arbitrary
amounts of descriptive content. But the presuppositions of these particles cannot be
accommodated anymore than the presupposition of too, and it seems even more artificial
to assume hidden pronouns in the presupposition of these particles.
In fact, it seems a general property of presupposing particles that their presupposition

cannot be accommodated. Again clearly has this property, like indeed, instead, German
and Dutch doch and toch, Dutch immers, and others. But they also have other properties
that make them unlike normal presupposition triggers. First of all, they are not optional in
the sense that if one finds them in a body of naturally occurring text or dialogue, they can
just as well be omitted. The dialogues in (4a, b) are examples, but one really needs to
consider many cases.3
(4) a. A: Bill will come tonight.

B: John will come (too).
b. A: Bill is ill.
B: He is (indeed).
Second, they have a rather minimal meaning apart from their presuppositional properties.
Again in (5) does not inform us of anything apart from the presence in the context of an
earlier occasion of failing on Mary’s part. The truth conditions are the same as the
sentence without the particle. It does not assert the existence of another occasion of
failing. For that, we have locutions like for the second time.
(5) Mary has failed again.
A third and even more puzzling characteristic is that the antecedents of some of these
particles can occur in contexts that are not accessible from the position of the trigger in
the sense of discourse representation theory.
(6) Mary dreamt that night that she would fail the exam and indeed she did.
None of the triggers that are central in the presupposition literature have these properties.
The only exception might be the obligatory nature of the trigger. Is the use of
presupposition triggers instead of nonpresupposing alternatives obligatory if the
presupposition is fulfilled? I think not, but the situation is not as clear as one would like.
Consider two examples.
(7) John believes/suspects that p.
If I say (7) when I know that p is the case, I am not pragmatically incorrect. I merely
suggest that John does not have the appropriate epistemic access to p to warrant the use of
know, so that using know is inappropriate. I would be violating Grice’s maxim of quantity
if John knows that p, but that is, by assumption, not the case here.
If we have discussed a new girl at the office, it is not incorrect for me to report that I saw
John with a girl in town, instead of saying that I saw John with the new girl at the office; I
may consider the connection irrelevant in the context. (I would only suggest that they are
different, if the hearer would think the identity would be relevant.) To the extent that the
standard triggers like know or the are obligatory, they are so because they are liable to
mislead the hearer. Not using them can be a transgression of Grice’s maxim of quantity.
The particles are different. They can only be used when the presupposition is there (since
they do not accommodate) and their absence cannot really mislead the hearer if the
presupposition is satisfied, since the presupposition is common knowledge already. Yet, it
is pragmatically incorrect not to use them when their presupposition is fulfilled or to use
them when the context does not contain their presupposition.
There are unclarities here, but it is obvious that know and the accommodate, have content,
and do not take inaccessible antecedents.
(8) John knew that Mary has failed.
(8) can be used to convey that Mary had failed. Knowledge is more than just belief with a
presupposition and so has independent content. The truth of the presupposition is
therefore not enough to make it necessary to use the word know. (9) is only acceptable
with the extra accommodation, that the content of the dream is true.
(9) Mary dreamt that she would fail the exam. Bill knows that she will.
Similar examples with the are given in (10a, b).
(10) a. I met the director of Peter’s school.

b. Mary dreamt there was a burglar in the house. The police captured the burglar
after a chase in the garden.
The sentence in (10a) can be used without Peter’s school having been mentioned before
and without it being known that it has a director. Example (10b) can (when it is not taken
as an elaboration on the contents of Mary’s dream) only be understood by the extra
assumption that the dream was true.
It is clear that if we want to analyze particles as presupposition triggers, we must be able

to modify our presupposition theories to make it possible that the particles come out as a
special case with special properties: no semantic content of their own, no accommodation,
the possibility of inaccessible antecedents, and the obligatory character of their use. The
particles we will discuss ideally have four properties:
• They do not contribute to the truth conditions of the sentences in which they occur.
• Their occurrence is not optional but obligatory: if they occur, they can normally not be
omitted.
138 H, Zeevat
• Their presuppositions cannot be accommodated, i.e., they cannot be used in contexts in

which they are not at least suggested or in contexts that can be reinterpreted as
suggesting.
• The antecedents of a particle presupposition can be much weaker than the antecedents
of other triggers, which require that the presupposition is true in the context of the
trigger.
Particles like again or immers do not meet the first and fourth condition, though they meet
the second and the third. The first and fourth conditions are systematically connected: if a
presupposing expression contributes to the truth conditions, the presupposition must hold
with respect to the worlds at which the clause in which it occurs is supposed to hold. A
weak antecedent does not guarantee that this is the case.
In Zeevat (2002), I propose three departures from standard presupposition theory. First,
weakly accessible antecedents are generally allowed (standard triggers do not allow them,
since a weak antecedent would lead to local undefinedness of the trigger). Second, non-
accommodation is explained by Blutner’s theorem (a bidirectional consequence of the
constraint Do not accommodate). Finally, I propose a number of marking constraints that
require properties like old (the content of the sentence is true or suggested in the context)
and other (there is another item of the same type in the context) to be marked.
The marking principles for presuppositional particles are additional: there is no way we
can derive them from an analysis that restricts itself to saying that they just presuppose
that particular presupposition or have that particular content (if they have any). A natural
strategy towards understanding particles better is therefore to turn the argument around
and investigate whether we can understand why they are like presupposition triggers if we
assume that they are markers of a relation of the content of the current sentence to the
context (or to another parameter of the utterance context) and can be there because of
either a functional necessity (if the relation in question is unmarked, wrong interpretations
result) or a universal principle that requires the marking of the relationship (according to
e.g., Haspelmath, 1999, and Bresnan and Aissen, 2002, OT constraints require a
functional grounding).
The kinds of relations for which it is plausible to assume a marking principle are given
below:
• The content is already part of the common ground. (OLD: indeed, immers, doch/toch
(unaccented), ja).
• The content has been suggested to be false in the context. (ADVERSATIVE: doch/toch,
proconcessives, concessives).
• The content was denied in the common ground. (CORRECTIVE: sondern, accented wel,
niet, doch, toch, do, didn’t).
• The topic has been addressed before but the content gives an expansion of the earlier
answer. (ADDITIVE: too, also, ook, auch)
• The topic has been addressed before, but this contribution needs to be replaced.
(REPLACING ADDITIVE: instead).
• The new content addresses the inversion in polarity of the old topic (CONTRASTIVE:
but, however, maar, aber).
Are these marking strategies universal? I do not know. There are many things unknown
about discourse particles and they are hard to understand even in a single well-studied
language. It suffices for our purposes to assume that there is a strong functional pressure
to have ways of expressing these relations. That assumption is necessary, since otherwise
it is not clear how we could have particles like the ones listed above or how they can
appear so often. And we can try to see what could go wrong in the interpretation process
if the particles (or other forms of marking) weren’t there. This is what I try now.
OLD marking
If an old element is not marked as old, it may be interpreted as new even if it is formally
identical to some old element (indefinites, tense). This will lead to copies in memory that
will be treated as distinct from each other. In addition, the original element is integrated
into the semantic representation by the original interpretation process, and the new
version will lack the connections that were constructed there.
ADVERSATIVE marking
If the presence of a suggestion to the contrary is not noticed, this means that the
suggestion to the contrary will be unchecked and can be the source of later errors. It is
possible to make a connection from the contrary suggestion to the new information that
makes the suggestion inactive.
CORRECTION marking
This should lead to the retraction of the corrected element. If this does not happen, the old
and wrong information may remain active. Like suggestions to the contrary, they should
be marked as corrected, since otherwise memory may give the wrong information later
on.
ADDITIVE marking
Additive marking finds an old topic and the way it was addressed before. Without the
additive marking, a different topic may be assumed. Without additive marking, the two
occasions of addressing the same topic remain unintegrated and can lead to wrong
information due to exhaustivity effects. If the one instance is noticed, it may be assumed
that that is all. Or the one instance may be noticed without the other one coming into
consciousness. The fact that we know more about the topic after the new information is
not exploited.
SUBSTITUTION marking
Here it is essential to make sure the two ways in which the topic is addressed are kept
distinct and that the two answers are not taken as a joint answer to the same topic. It is
related to correction and adversativity.
CONTRAST marking
I am assuming that contrastive marking indicates that a positive answer is given to the
negated current topic. If the polarity switch remains unmarked, it may be unnoticed,
which can lead to misinterpretation.
140 H, Zeevat
These motivations suggest that it is in the speaker’s interest to mark these relations:
without marking, he or she may well be misunderstood. And it is in the hearer’s interest to
pay attention to the marking particles: without that, confusion may result.
4. Marking in an Optimality Theoretic model
In optimality-theoretic syntax (see e.g., Bresnan, 2000, for an influential proposal), the set
of well-formed sentences is given as those sentences that are optimal for a possible
meaning in context, the input. Optimal for a certain input are the possible sentences that
best meet a system of ordered constraints with respect to the input. The constraints are
universal, and particular languages impose a linear ordering on the set of constraints. A
winner for a given input is one that does at least as well as any other candidate on the
strongest constraints up to the next strongest constraint C on which it does better. Here we
assume that everything else is captured by other constraints and that the only task that we
are facing is to compare sentences with and without a particle.
Let us assume that the convention for our particles is very simple: if the relation R obtains
between context parameters and the current utterance, add the particle P to the utterance.
(A more abstract version only asks for R to be marked somehow and so allows other
marking devices apart from P: other particles, lexical material, constructions, intonation).
This convention (a constraint max(R)) overrules a constraint against special devices (an
economy constraint *Particle). The combination of the two constraints guarantees that P
appears if and only if R holds between the content and the context parameter. From the
point of view of the interpreter of the utterance, an occurrence of P indicates that R holds.
Since the hearer now knows the content of the utterance and already knew the context
parameters, the hearer can make sure for him- or herself that R holds. This check of R will
force certain identifications, involving the current utterance, the common ground, and the
topic. The check is part of the interpreter’s task of reconstructing the intentions of the
speaker. It is also part of the interpreter’s task of integrating the new information within
the overall representation of the world and of doing so in an efficient way.
Can we now understand why there are similarities between presupposition triggers and a
class of particles? What we have so far is an explanation of two properties of our
particles: the fact that they do not accommodate (*Particle) and the fact that their
occurrence is not optional but obligatory (Max(R)). The other thing we need to explain is
the fact that they lead to a resolution process in which certain material is identified in the
context. The only assumption that we need to make is that R is checked with respect to the
local context. This will deal with examples like (11) where the relation OLD holds with
respect to the common ground to which the subordinate clause has been added.
(11) Falls du nach Berlin kommst, triffst du ihn ja.

In case you come to Berlin, you will meet him ja.
If we make the assumption that hearers can only be satisfied with an interpretation if they
would generate the same sentence from the interpretation as the speaker has done,4 it
follows that they must also check R with respect to the interpretation. This is enough to
establish that there is a resolution process prompted by discourse particles that identifies
material in the local context or the contexts around it like the common ground. Let us go
through each of the relations in detail.
OLD markers
OLD refers to the presence of the content of the sentence in the common ground. It may be
there directly, but it can also just be suggested: it is the opinion of somebody, the content
of a dream, of a suggestion or even an iteration of these things. is the content of the
current utterance, CG the common ground; OLD(CG, ) holds iff CG |= suggested( ). The
relation suggested ( ) can be defined by a recursive definition, using a set {O1…,On}
containing operators like dream, suggest, believe.
(12) suggested( ) <=> v O1 v … v On v suggested( )
Each of the particles does more than just mark R, almost by definition, in this case. Indeed
indicates the presence of better evidence for , immers makes a reason for assuming the
current discourse pivot (the discourse element to which the current utterance is related by
a discourse relation, normally the previous utterance), doch/toch without accent makes the
old information the subject of discussion again, ja presents it as common ground between
speaker and hearer (and allows further causal or other connections based on that). This
makes it hard for immers, ja, and unaccented toch/doch to have antecedents which are
merely suggested.5 What they have in common is that they mark the OLD relation. There
must however be a point to bringing up old material again, and the particles differ with
respect to the sort of point they allow.
ADVERSATIVE markers
Adversativity means that the content of the current utterance goes against material that
was already present in the common ground. This can happen in the weak sense because
material was there which normally has the negation of content as a consequent. It can also
be that the negation of the content is present in the sense that it is just an element of the
common ground or that the common ground suggests that material. ADVERSATIVE(CG, )
holds iff CG |= presumably(not- ) or CG |= suggested(not- ). The truth of presumably(p)
on an information state requires that CG |= 1, …, n and that 1, …, n together
constitute a reason for thinking that p, while at the same time the CG must not support a
similar argument for not-p.
The easiest case is that of full concessives. The complement of the concessive clause
gives the argument for not- and also chooses presumably instead of suggested. Since the
complement of the concessive connective is presupposed, it can be treated as part of the
common ground. Proconcessives (e.g., isolated though in English) indicate that the
complement is highly activated. Here the other branch based on suggested is necessary.
Consider (13):
(13) Mary dreamt that she failed the exam. She had passed though.
It seems impossible to construe dreams as arguments for the truth of its propositional
content. So this is really a nonconcessive adversative reading of though. If there is a
grammaticalization path here, one would expect it to go from proper concessives to the
vaguer adversative meanings.
142 H, Zeevat
Accented doch/toch is adversative. Partly, these are proconcessives with a normal stress
(like trotzdem, nevertheless, desondanks), and partly doch/toch has contrastive stress
contrasting with an activated negative version of the current sentence. The real puzzle
with doch and toch are the unaccented cases that can be proper OLD markers without the
slightest trace of adversativity (14a). These can probably be connected to affirmation
questions with a positive bias, elicited by an apparent opposite opinion of the interlocutor,
as in (14b).
(14) a. Ich bin nächste Woche doch verreist. Kannst du meinen Unterricht
übernehmen?
You know that next week I am away. Can you take over my teaching?
b. A: Ich werde es ihm nächste Woche sagen.
I will tell him next week.
B: Dann bist du doch verreist?
But you are away then, aren’t you?
Though doch is here appropriate because B seems to imply that what A said is false, it
also expresses that according to B the common ground is that A is abroad next week.
Reanalysis as an OLD marker is thereby possible. Another example of this use of
unaccented doch is given in (15).
(15) Wenn er doch hier ist, kannst du es ihm auch selbst fragen.
When he is here anyway, you can ask him yourself.
CORRECTIVE markers
Proper corrections are simple. They require that the content be false in the common
ground: correct(CG ) holds iff CG |= not- . The correction relation is an extreme case of
adversativity: the best reason for believing that is false is knowing that it is. At the same
time, unlike the weaker possibilities for adversativity, the current sentence is then not
consistent with the common ground. The intended change to the common ground is a
combination of retraction of not- and the addition of as a replacement. Doch/toch with
contrastive stress is one correction marker. Others are Dutch wel and niet, English do and
do not (all with contrastive stress). The German sondern (the correcting version of but) is
special by not requiring contrastive stress itself.
ADDITIVE markers
I will treat additivity as the reopening of an old topic to which new material addressing it
is added. The common ground must remember the topic and the propositions that
addressed it. The fact that the content of the current utterance is added to what we had
already and does not replace it is not a question of context marking but a separate
intention of the speaker. Additive(CG, ) is therefore a combination of a complex relation
to the common ground and a special intention.
The relation is between the common ground, a topic and a proposition. The topic must be
such that addresses it. The proposition must be the strongest to hold on the common
ground that addresses the topic, and the common ground must “remember” that the
proposition addressed the topic. This calls for a special predicate addressed( , T) that
finds a topic T that the current proposition is addressing and the proposition that
according to the common ground settles the topic.
(16) CG |= addressed( , T)
The predicate should entail CG |= , address( ,T), and address( ,T), and there should
not be a such that CG |= , |= but not |= , which also addresses T.
In a proper model of topic, addressing should be a formal relation between the formal
topic and the sentence. For example, in a model of topics where they are Hamblin-style
questions (i.e., the set of their possible answers), a proposition addresses a topic if it is
member of the topic. The intention of the speaker is that now the conjunction of and
becomes the information that the common ground has about the topic. That is,
addressed ( , T) will be false on the new common ground, and addressed ( & , T) will
be true. Close to ADDITIVE markers in functionality are other-markers like another in
Another girl walked in. If we think of the noun girl as a topic that is addressed by the
indefinite, their treatment is formally the same. But whether it makes no sense to think of
the noun as an additional topic, I do not know.
REPLACING ADDITIVE markers

Replacing additive markers like instead are only different in the intention. We here want
the effect that the proposition that is used to address our topic is replaced by the current
proposition , so that afterwards the common ground makes addressed( ,T) true and
addressed( ,T) false.
The choice between ADDITIVE and REPLACING ADDITIVE markers explains the relative
uncomfortability of antecedents that are only suggested for markers like too, also, and
instead. If in (17a) one uses too, the suggestion is that Sue is in Spain next to John; if in
(17b), one uses instead, one suggests that the dream is false. Leaving out the particle
completely, as in (17c), is not an improvement. We now no longer mark that the topic has
been addressed before.
(17) a. Mary dreamt that John is in Spain. (?) Sue is also in Spain.
b. Mary dreamt that John is in Spain. (?) Sue is in Spain instead.
c. Mary dreamt that John is in Spain. (?) Sue is in Spain.
In (18) we see how subtle this is. The situation (A and B are children in a secret phone
call) makes it clear that B’s parents do not know about the other child. And many people
find the example mildly anomalous.
(18) A: My parents think that I am in bed.

B: My parents think that I am also in bed.
I conclude that too and instead are not just context markers but also speech act markers
for the specialized speech act of adding to or replacing in an old topic. This would
explain the data in (18).
144 H, Zeevat
CONTRAST markers
The most complicated relation I consider here is CONTRAST . One can wonder whether
connectors like but are really particles. In German, aber ‘but’ also appears in later
positions in the sentence, and an extensive corpus study (Schoesler, 2002) reveals that
there is no essential difference in these uses, which are translatable by echter in Dutch or
by however in English. This suggests that like concessive connectors, they are particles.
My provisional analysis, derived from Umbach (2001), goes as follows, using the format
above.
Let be the discourse pivot (when a coordination but is the marker, this is just the first
conjunct) and let CG |= addressed( , T). is contrastive iff it directly or indirectly
addresses negate(T). Here, negate(T) is the topic that is addressed by the negation of any
formula that addresses T. For example, in the view of Hamblin (1973), we can obtain
negate(T) from T by replacing T’s elements by their negations. We can say that the
sentence S indirectly addresses a topic T iff the common ground updated with the
information that S settles its own topic T entails an element of the topic.
I illustrate the analysis by (19). In (19a), the second conjunct directly addresses the topic
of the first sentence: who was ill. I will assume that this is the topic of the first conjunct
also in (19b) and (19c). In (19b) we can construct the topic of the second conjunct as Who
was fit as a fiddle? or Was John as fit as a fiddle? In both cases, the answer entails that
John was not ill. In (19c), the topic of the second conjunct is something like What about
John? The fact that the answer John went to the party settles the topic entails that he was
not ill. The fact that the negation of the topic of the first conjunct must be addressed
implies that a weaker topic (e.g., Did John go to the party?) cannot be chosen.
(19) a. Mary was ill but John was not.

b. Mary was ill but John was as fit as a fiddle.
c. Mary was ill but John came to the party.
With Umbach, I hold that the concessive uses are derived.6 Example (20a) can be
rephrased by (20b).
(20) a. Although Mary was ill, John went to the party.

b. Mary was ill, but John went to the party.
Here but is reanalyzed as a proconcessive, taking its antecedent from the first conjunct.
This requires that the common ground makes Mary’s illness a reason for thinking that
John would not go to the party (it may be known that in such cases he feels his duty is at
home). My idea is that these readings find their origin in one way in which topics may
arise, by causal connection. If you know Mary and John, the fact that Mary goes to the
party makes it likely that John will go there as well. So in that case, the contrastive but in
(21) can relate to a causally related topic: did John and Mary go to the party. (A separate
adversative marker is not necessary anymore.)
(21) Mary went to the party, but John (however) did not.
A simple treatment of and along the same lines is to say that and forces the second
conjunct to at least indirectly address the same topic (this is consistent with the analysis of
Gomez-Txurruka, to appear).
5. Broader framework: speech act markers
Polysemy of discourse markers arises when the same discourse marker marks two
different relations of the content of the current sentence with the common ground. But this
is polysemy only if we consider an isolated sentence; in a particular common ground,
polysemy can only arise if more than one relation marked by the particle obtains between
content and common ground. In practice, polysemy should not be very important.
Example (14b) (repeated below) is interesting in this respect. It expresses both
disagreement with what the first speaker suggests (a function of accented doch) and the
fact that the absence of the first speaker is already common ground. Both relations seem
to be constructed, and it is not easy to find examples like (14a) where adversativity is not
constructible. Replacement of doch by ja in (14a) is possible.
(14) a. Ich bin nächste Woche doch verreist. Kannst du meinen Unterricht
übernehmen?
You know that next week I am away. Can you take over my teaching?
b. A: Ich werde es ihm nächste Woche sagen.
I will tell him next week.
B: Dann bist du doch verreist?
But you are away then, aren’t you?
I have discussed so far what context marking is if we assume that syntax tells us to mark
certain relations of the current utterance to context parameters like topic and common
ground and if the interpreter’s task is just to reconstruct the speaker intention. We have
assumed that the presence of context markers is largely explainable by the difficulties
facing the hearer in properly integrating the current utterance with the information that he
or she has already got. Particles in this view regulate the proper construction of
interpretations and, in particular, the proper integration of the interpretation in memory.
The presuppositional character of some of the particles is basically the reconstruction by

the hearer of the relation marked by the particle under which the utterance is made. This
forces the identification of a topic or a proposition in the common ground. There is no
accommodation because the parameters are overt: it makes no sense to warn the hearer
about a relation that does not obtain. Suggestions can open topics and address them. That
is enough to understand why OLD, ADVERSATIVE, and ADDITIVE markers can take indirect
antecedents.
It is therefore not necessary to invoke presupposition theory for the analysis of discourse
particles. In fact, one may wonder whether presupposition—or presupposition triggers—
must be considered to be a natural class in linguistics, since, after all, the triggers
normally considered in the presupposition literature fall into at least three classes: the
ones considered here, referential devices like definite descriptions, and the lexical
presupposition triggers, like bachelor, all three with a different projection behavior.
146 H, Zeevat
An attempt to understand particles as presupposition triggers also runs into the problem
that many are not. It is clearly the case that more particles can be analyzed as context
markers. But this should not fool us into thinking that context marking is all there is to
particles. Very obviously, many discourse particles mark other aspects of speech acts. The
clearest case are markers like Chinese ma that makes questions out of assertions as in
(23).
(23) Ni hao ma?

You good QUESTION-PARTICLE
Are you OK?
Or, take the unaccented wel in Dutch, as in (24).
(24) Het komt wel goed.

It’ll be fine.
The particle tones down the preconditions of normal assertion (the speaker has to believe
to know what he or she is telling the hearer) to mere undersupported belief. This, like a
repetition or a correction, is a specialization of the speech act of assertion.
A formal theory of speech acts is not the aim of this paper. I will sketch what it would
involve. We assume that there are at least three dimensions. The first dimension is the set
of preconditions for the speech act: what must be the case with the context of the
utterance for it to be possible to carry out the speech act. The second dimension is the aim
that the speaker wants to achieve with the speech act. The third is the effects that the
speaker achieves with the speech act independently of whether he or she reaches the full
aim or not, just by making it successfully. These are the minimal effects of the speech act
and the ones that are achieved if the act was successfully performed, even if the intended
response of the hearer did not take place.
Context marking affects the first dimension: the preconditions of the speech act and the
changes of the defaults assumed there. We have seen that the two varieties of additive
marking (too versus instead) also affect the second and third dimensions. With too, we
intend to bind an old topic question to a new value that is obtained by adding the value
specified in the sentence to the old value. With instead, we intend to replace the old value
by the value specified in the sentence. This also affects the third dimension: in the case of
too, the speaker endorses the old value of the topic in addition to the value specified in the
sentence, whereas in the case of instead, the speaker disagrees with the old value and only
expresses the belief of knowing that the value is as expressed in the sentence.
Default assertions have the following preconditions, intentions, and minimal effects. It is
tempting to think that all other speech acts derive from this notion of standard assertion by
overriding some of the default settings.
• Assertion: p
• Preconditions: the common ground contains no reason for thinking that p is true or
false.The hearer wants to know the answer to a new topic question Q; p settles Q.
• Intention: that it become common ground that p
• Minimal effect: to make it common ground that the speaker believes to know that p.
ADVERSATIVE, ADDITIVE and OLD marking can change the preconditions. We obtain
corrections and reconfirmations when the adverse or old information is in the common
ground itself. ADDITIVE markers make the topic question old. Other markers (wel, maybe,
schon) change the operator under which the new information enters the common ground
from the speaker believes to know to weaker ones such as the speaker thinks it is probable
that, the speaker thinks there is a chance that, or the speaker thinks that. The effect of an
accepted weakened assertion of this kind can also be different: it is probable that, there is
a chance that, speaker and hearer think that.
Particles, intonation, and syntax can be used to mark different kind of questions. This by
itself leaves the preconditions intact. (They can be changed by particles and in special
confirmation questions, e.g., ones that want confirmation of old material in the common
ground). Let us go through some examples.
(25) a. Du bist doch verreist?

b. Komt Bill wel?
c. He is ill, isn’t he?
d. Is Bill coming?
e. Isn’t Bill coming?
f. Is Bill coming or not?
g. Is Bill coming or Mary?
h. Who is coming?
In (25a), the preconditions are different. The hearer has suggested that p is false, the
speaker believes to know that p. The speaker intends that the common ground contain p,
but offers the hearer the option to give other information instead. In (25b), the speaker
expresses doubt about a common ground fact that p and intends that it be removed. In
(25c), the speaker intends to make it common ground that p, believes p, but does not
believe to know that p. In (25d), the speaker intends to make it common ground that p but
does not have any evidence for p. In (25e), the speaker proposes to make not-p common
ground without believing to know it. In (25f, g), the speaker proposes to make one of the
two alternatives into common ground, suggesting to know that one of them is the case.
(25h) can be seen as the same speaker’s proposal but for more alternatives, again
suggesting that one of them is true.
Yet other markers (performative verbs, please) turn the assertion into a promise or a
request. In promises and requests, the intention and the preconditions are the same (or can
be thought of as the same, since p becomes a fact in the common ground after the promise
or request is accepted). But the minimal effect is different: the speaker wants p to be true.
It should be clear that this last section especially is very sketchy. The concept of a speech
act semantics as a successor to dynamic semantics, however, seems well worth going for
as a framework for characterizing particles.
Notes
1
The discourse representation theory of Kamp and the file change semantics of Heim are
the first instances of dynamic semantics. This paper follows a slightly more abstract
version, the update semantics introduced in Veltman (1996).
148 H, Zeevat
2
I thank Nick Asher (personal communication) for this argument.
3
Corpus work by Tim Kliphuis and myself suggests that omitting them nearly always
leads to awkwardness or to differences in the implicatures.
4
This is the central assumption of bidirectional optimality theory, where both
interpretation and generator are constrained by optimality theoretic constraints, but the
only proper winners are ones where the interpretation wins for the given form and the
generation wins for the given interpretation.
5
This makes a proper account of them dependent on a constraint Defined requiring that
the semantic material is not undefined.
6
This can be doubted. Professor Asiatini noticed (personal communication) that in
Georgian the concessive and contrastive uses of but are lexicalized in a different way.
This shows at least that normal language users do not conflate the two uses and that
contrastive markers do not always allow concessive interpretations.
Chapter 9
A relevance-theoretic approach to discourse

particles in Singapore English
Ler Soon Lay Vivien
1. Introduction
In the literature, discourse particles are a heterogeneous class of items found in a variety
of European and non-European languages. Examples include but, so, after all, yeah, okay,
well, even, oh, and too (English); ja and doch (German); and ka, yo, da, ne, zo (Japanese).
The use of discourse particles is not intended to reflect a commitment to a class of
discourse particles. However, I will assume for the sake of argument that the items above
do form a class but will end up making a distinction between the class of discourse
markers (DMs) and discourse particles (DPs). In a contact variety of English in
Singapore, there is a group of discourse particles, such as lah, lor, meh, hah, and hor,
which give Singapore Colloquial English (SCE) its special flavor.1 These DPs belong to
the most frequent words used in conversations. The meanings and functions of these
particles have been the subject of previous discussion (e.g., Kwan-Terry, 1978; Platt,
1987; Gupta, 1992; Pakir, 1992). Examples of DPs in Singapore Colloquial English taken
from the recently completed lexical corpus of Singapore English (ICE-SIN) are:2
(1) a. Go to Chinatown lah. (ICE-SIN-S1A-007)

b. A: Aye when are you going to [do your] income tax.
B: Tomorrow lor, one to two. (ICE-SIN-S1A-077)
c. You mean at home cannot sleep with aircon on meh? (ICE-SIN-S1A-014)
d. A: The hand still shaking.

B: Hah. Damn funny. (ICE-SIN-S1A-014)
e. When? 5 to 6:30 pm. Come hor, alright. (ICE-SIN-S2B-043)
The frequency of the top-ten DPs in the spoken categories in ICE-SIN (a corpus of about
600,000 words) is shown in Table 1.
150 Ler Soon Lay Vivien
Table 1: A comparison of the top-ten SCE DPs in the spoken categories in ICE-SIN.
SCE particles Frequency

1 lah 1,742
2 ah 1,242
3 hah 256
4 what 224
5 lor 114
6 hor 63
7 nah 50
8 leh 43
9 ma 27
10 meh 20
1.1. Approach
The account argued here is a unitary one (like some accounts in this volume, for example,
Fischer, Nyan, Gupta, and Travis) in which inference plays a more considerable role than
those that are done in polyfunctional alternatives. The unitary account follows from
Grice’s Modified Occam’s Razor,3 that on the basis of cognitive economy concerning
storage of information, a unitary account is preferred over a polysemous one. I find that
relevance theory (Sperber and Wilson, 1995), with a view of utterance understanding in
which the consideration of cognitive economy extends to utterance interpretation,
provides an adequate framework for this task. My approach builds on work done by
Sperber and Wilson, Blakemore, and others.
1.2. Methodology
The methodology I am using is the cognitive approach based on the recently developed
pragmatic framework, Relevance Theory (Sperber and Wilson, 1995). This theory is a
general theory of communication based on cognitive principles. It looks at utterances as
inputs to inferential processes which affect the cognitive environment of the hearer. In this
account of communication, interpretation of utterances is not merely a matter of linguistic
decoding but relies heavily on inference. The process of utterance interpretation is
governed by the principle of relevance, that is, “every act of ostensive communication
communicates a presumption of its optimal relevance” (Sperber and Wilson, 1995: 176).
Optimal relevance can be simply expressed as the greatest cognitive effect for minimal
processing effort; while cognitive effect, also sometimes called contextual effect, is
achieved when newly presented information interacts with existing beliefs in one way or
another. According to Sperber and Wilson (1995), this kind of effect can be achieved
through three ways:
A relevance-theoretic approach to DPs 151
1. by strengthening an existing assumption;

2. by contradicting or eliminating an existing assumption;
3. by combining with existing information to yield a contextual implication, i.e., a logical
implication which can be derived from a combination of the context and the new
information.
Another important postulate is that expressions in language can be seen to encode not
only concepts but also procedures. Such expressions guide the hearer in the process of
utterance interpretation and contribute to relevance by reducing the processing effort
needed to reach the intended interpretation (Blakemore, 1987; Carston, 2002). How the
processing effort is reduced can be affected by “constraints on relevance” (Sperber and
Wilson, 1995), i.e., by making the context set smaller for the hearer. The following
analysis of examples (2) and (3) will give a better understanding of constraints on
relevance.
(2) Benjamin Bratt likes to please Julia Roberts.
(3) He loves Julia Roberts.
(2) and (3) can be construed as being in a variety of relations. For example, they could be
just two facts or beliefs, or one could be construed as giving evidence for the truth of the
other. In that case, one is the conclusion, while the other supports the conclusion. Because
either one can be the conclusion, there is the possibility of misinterpretation. In such
cases, argues Blakemore (1987), constraints on relevance play a vital role. Consider the
following:
(4) a. Benjamin Bratt likes to please Julia Roberts.

b. After all, he loves Julia Roberts.
(5) a. Benjamin Bratt likes to please Julia Roberts.

b. So, he loves Julia Roberts.
Although the two utterances remain in the same order in (4) and (5), they are related in
different ways. In (4), it is (4a) which is the conclusion, with (4b) providing the evidence.
In (5), it is (5b) that is the conclusion, with (5a) giving the evidence. In both cases the
speaker expects the hearer to have further contextual assumptions available, and they are
not the same for (4) as they are for (5). Thus, in (4), the speaker expects the hearer to
access assumption (4’):
(4’) If X loves someone then X likes to please this person.
In (5), however, the assumption is different:
(5’) If X likes to please someone then X loves this person.
Thus after all and so constrain the processing of the two utterances in different ways.
Blakemore argues, on the basis of examples such as (4) and (5), that words such as so and
after all do not contribute to the truth-conditional content of the utterance in which they
occur and that they do not encode conceptual meaning. Their role is to help “constrain the
hearer’s choice of context for its interpretation” (Blakemore, 1987:141).
1.3. Data
The data in my study are from two main sources. The first is the recently completed one-
million-word lexical corpus, ICE-SIN (Ni and Ler, 2000), which is by far the most
comprehensive collection of Singapore English. I have used the data from the spoken
component of ICE-SIN, which consists of about 600,000 words. The second source is
personal conversations or statements overheard at churches or canteens and recorded
either immediately or noted down in the immediate future. Occasionally, for clarity, I
have constructed examples.
Different approaches have been taken towards the study of DPs, for example,
conversation-analytic (Schiffrin, 1987a, 2001), contrastive-analytic (Fischer and Drescher,
1996), historical-pragmatic (Brinton, 1996), formal-pragmatic (Fraser, 1996), relevance-
theoretic (Blakemore, 1987, 2002; Jucker and Ziv, 1998; Andersen, 2000; Infantidou,
2001), discourse-pragmatic (Lewis, this volume), etc. Most researchers agree that DPs
are polyfunctional in nature. Nearly all the evidence provided in the literature so far seems
to indicate that DPs have different interpretations and perform different roles under
different contexts. But the view of polyfunctionality also raises a problem for
grammarians as it implies the existence of several distinct particles for each DP, each
displaying different semantic properties or pragmatic forces. Unfortunately, none of these
researchers could provide a clear explanation of why the same particle could always fulfil
different roles in different contexts.
In SCE, the particles have been discussed by various writers, but there is considerable
disagreement as to the functions of these particles. Generally, most previous accounts are
based on a few examples, except for Gupta (1992), who had a substantial body of data (18
hours of natural data) and Richards and Tay (1977), who included in their study a
fragment of a telephone conversation. Data in this study is mainly from the spoken
component (about 600,000 words) of ICE-SIN and involves different races (Chinese,
Indians, Malays, and others) and different ages. The question of how one particle can
perform many functions has preoccupied previous analysis. I believe previous accounts,
which are largely descriptions of the uses of these DPs, are inadequate. The important
point to consider is not the many functions that these DPs perform, but rather, how these
DPs are interpreted in actual discourses (Ler, 2001, 2002).
I will attempt to show that Relevance Theory offers tools which can provide a unitary
account for the DPs in SCE. My account will rest on the basic assumption that these DPs
are procedurally used by the speaker (see Sperber and Wilson, 1995) and that for each DP,
the various uses of the particle can justifiably be subsumed under a single description.
2. Definition
Interest in DPs has grown with the increased attention in the pragmatic and contextual
aspects of utterance interpretation. Numerous articles on DPs, both descriptive and
theoretical, have been written in the past decade. The study of DPs is a growth industry in
linguistics (Fraser, 1996). However, whether there is a class of DPs (also known by other
names, such as discourse markers and pragmatic markers) or not has not been ascertained.
Researchers are divided as to what constitutes this class. But, most researchers are in
agreement that these items are elements of discourse that are generally thought of as not
affecting the propositional content of the utterances in which they occur. They are non-
conceptual but encode procedural meaning (Sperber and Wilson, 1995; Blakemore,
1996).4
It appears that the term discourse is used to indicate that a description of these entities
needs to be at a level higher than the sentence, that is, at the discourse level (Blakemore,
2002: 1). This study adopts the view that makes a distinction between grammatical
systems and one that is outside grammar, namely, utterance interpretation. Thus,
discourse markers are linguistic constituents that are marginal to the syntax of the clause
and that can be left out without affecting the truth conditions of the sentence (Brinton,
1996: 267). They encode various functions, such as connectivity, turn taking, and stance
or propositional attitude.5
The term discourse particle refers to a morphologically rather different set of linguistic
expressions including grammatical phrases (e.g., good grief), formulaic clauses (e.g., I
think, you know), as well as monomorphemic words (e.g., oh, well). This study (like
Gupta, this volume) takes the view that particles must be single morphemes that occur
utterance finally. Thus, in this study, I propose the following provisional definition of
discourse particles. Discourse particles are monomorphemic items which occur utterance
finally and which encode propositional attitude. They are used in various discourse
marker functions and thus are a subset of discourse markers.
DPs in SCE are regarded as expressions (6a-d) which are syntactically optional in that
their omission will not affect the grammaticality of the sentence (6e-h).
(6) a. Don’t be shy lah.

b. He’s not asleep meh?
c. Buy this hor.
d. He also cries what.
e. Don’t be shy.
f. He’s not asleep?
g. Buy this.
h. He also cries.
Brinton, in her study on DPs, finds that omitting the DPs does not render the text
ungrammatical or unintelligible (1996: 267). Though syntactically optional, DPs are not
redundant, as they are semantically obligatory (as is the case of DPs in SCE). Fraser (this
volume) states that a DP (he calls it DM, discourse marker) is a lexical expression that
signals a relationship of contrast, implication, or elaboration between two segments. He
further states that he objects to the claim that DMs have procedural meaning but no
conceptual meaning (his claim is that DMs encode both procedural and conceptual
meaning). I will not deal with Fraser’s definition of DMs. I do think, however, that
Fraser’s definition of DMs would not be able to accommodate DPs such as meh and lah in
Singapore English (it is not his objective to do so, perhaps). This study takes the view that
DPs guide the hearer toward a particular interpretation (that is, they encode procedural
meaning) and will show that they do not encode conceptual meaning. Together with
Blakemore and others, I see DPs as expressions that are syntactically optional, not
contributing to the propositional content of the host utterance but contributing to the
procedural meaning, and constraining the relevance in the utterance (see section 4).
The following list is a summary of the basic features of DPs.
1. Syntactically optional
2. Multifunctional, operating on several linguistic levels
3. Generally non-truth-conditional
4. Expresses certain emotions/attitudes of the speaker
5. Contribute to procedural meaning
3. The functional spectrum
3.1. Polyfunctionality of DPs
One of the main characteristics of DPs in SCE, as in other languages, is that they are
polyfunctional. How is it possible that one particle can have many different functions?
Some of the descriptions given by past researchers of DPs in SCE are so varied (and
many) that one wonders if they are describing the same entity (see section 3.2). What is
puzzling is to establish how a hearer understands what is being conveyed by the particle.
It seems to me that a DP encodes certain cognitive information; in particular, it instructs
the hearer to proceed in a certain manner. A hearer is able to understand (process) an
utterance (containing a DP) through inference and the context. As such, a unified account
of DPs along these lines will help provide a clearer picture of the meaning of the DPs (see
section 4). In the next section, I give a description of the various functions of two DPs in
SCE, lah and meh. Lah and meh are chosen for various reasons. They are the first and
tenth items in the frequency table for DPs in SCE (Table 1). I choose lah, a particle that is
frequent, widely discussed but still hard to account for, and meh, a particle that is rather
less frequent and that has not been extensively discussed.
3.2. The DP lah
Lah has been discussed by many writers, but there is considerable disagreement as to its
uses and functions. Previous researchers say the following about the various functions of
lah. In the earliest published account of the lah particle, Tongue maintains that depending on
the way it is pronounced, lah can function as an “intensifying particle, as a marker of informal
style, as a signal of intimacy, for persuading, deriding, wheedling, rejecting and a host of
other purposes” (Tongue, 1974: 114). Tongue’s account of lah marks the beginning of
treating the particle as characteristic of SCE.
In a number of accounts that followed, in addition to treating lah as a marker of rapport or

solidarity (Tongue, 1974; Richards and Tay, 1977; Kwan-Terry, 1978; Bell and Ser, 1983;
Pakir, 1992), the particle has been described also as a marker of emphasis (Richards and
Tay, 1977; Platt et al., 1983; Loke and Low, 1988). Lah has also been seen as an entity
communicating a range of different attitudes, such as obviousness, persuasion, and
impatience. Other functions ascribed to the particle include an expression of friendliness
or of the opposite sentiments, such as hostility or annoyance. It can also be described as
an indicator of enthusiasm and assertion or as a word communicating the attitude of
objection. Lah is also described as belonging to the assertive group on a scale (expressing
varying degrees of commitment to an utterance) with three main categories: contradictory,
assertive, and tentative (Gupta, 1992). The list is not exhaustive. The main functions, as
described by past researchers, are shown in the following examples. (Unless indicated
otherwise, the examples are from my personal collection or from ICE-SIN).
A. Solidarity
Lah is first described as a marker of rapport or solidarity, comparable to the English
“filler” you know.
(7) Don’t be shy lah. [We are friends]
(8) No use trying to hide our roots lah. [We are Singaporeans]
The problem with this classification is that it cannot accommodate data such as example
(9), where no element of rapport or solidarity can be detected. It cannot be argued that in
this dialogue A is using lah to establish rapport with her daughter.
(9) Context: A mother (A) and her daughter (B) had a disagreement on who is to buy
Mandarin oranges. (It is customary for the Chinese to exchange Mandarin oranges
when visiting during the Chinese New Year).
A: Then after that it’s the Lunar New Year special lah.
B: So?
A: Ya lah, then during that period we can go what?
B: Cannot lah. Aiyah, when I wash my hair, I don’t want to go out.
Dirty my hair lah.
A: You bring one of them lah. (ICE-SIN-S1A-007)
B. Emphasis
According to previous researchers, what lah contributes to utterances such as (10) and
(11) is an element of emphasis.
(10) Do you want to go? I’m not going lah. [Emphasis] (Kwan-Terry, 1992: 69)
(11) Normal doctors lah who are on our medical panel. [not specialists] (ICE-SIN-
S1B-073)
C. Obviousness
The lah particle is often described as having the function of conveying the speaker’s
attitude of “obviousness”. There is also a note of impatience or annoyance in these cases.
(12) They generally don’t take beef lah. [It’s obvious; everybody knows that.] (ICE-
SIN-S1A-023)
(13) I mean of course it changes lah. (ICE-SIN-S1A-065)
D. Persuasion
Lah can also be used with a certain tone to persuade or to suggest.
(14) Come with us lah. [Won’t you?] (Oxford English Dictionary online, 2000)
(15) Go to Chinatown lah. [Why don’t you?] (ICE-SIN-S1A-007)
E. Friendliness
Lah is sometimes used when the speaker wants to be friendly.
(16) Okay, doesn’t matter lah. [It’s all right; we’re friends.] (ICE-SIN-S1A-091)
(17) Quite nice lah. [I’m your friend; consider my opinion.] (ICE-SIN-S1A-023)
F. Hostility
Sometimes, lah is described as conveying a sense of “hostility”.
(18) If you want then it should be after this week lah. [Not earlier!] (ICE-SIN-S1A-
091)
(19) I don’t want to eat lah. [Don’t force me!]
As has been mentioned, the list of functions ascribed to lah and illustrated by examples
(7) to (19) is not exhaustive. The inadequacy of looking at lah as performing multi-
functions can be seen when we compare lah-utterances with utterances without the
particle.
(20) a. I mean of course it changes lah! (ICE-SIN-S1A-065)

b. I mean of course it changes!
The “obviousness” in (20) is evident even without the particle. We see that it is present in
the semantics of the utterances, as indicated by of course in (20a, b). Hence,
“obviousness” cannot be characterized as an inherent part of lah. For “persuasion” (14,
15), “friendliness” (16, 17), and “hostility” (18, 19), we have to be careful that these
meanings do not come from other devices such as intonation or tone. These meanings (in
(14) through (19)) are preserved even when lah is omitted. Thus, these functions are not
inherent in the particle itself.
3.3. The DP meh
Treatment of the meh particle in SCE is scarce, as previous researchers concentrate on the
more popular particles such as lah and what. Gupta (1992) described it briefly, while
Wong (1994, 2000) concentrated on the invariant meaning of the particle. The various
functions of meh are as follows:
A. Questions a presupposition.
The DP meh serves to “question a presupposition” (Gupta, 1992: 43). The following is an
example from Gupta (MG = mother of the girls, EG = elder girl, YG = younger girl).
(21) Context: MG has asked YG where her color pencil is, twice.
EG: You don’t know meh?
YG: No, I don’t know. Didn’t see.
B. Expresses surprise.
Meh adds a sense of surprise at the question asked, as shown in (22).
(22) A rang B at 11:00 p.m. and was surprised to find that B is still not asleep.
A: You not asleep meh? [Surprise as B normally sleeps at 10:30 p.m.]
B: Got to finish some work. Urgent.
C. Meh means the opposite of what was thought to be true.

The speaker previously thought something (not P) was true, but someone tells him or her
or something shows him or her that the opposite (P) is true (Wong, 2000: 21).
(23) Context: Price tag on a pair of shoes indicates $109.

A: It’s forty dollars.
B: Not 109 meh? [Customer is surprised.]
A: Got discount. [A affirms proposition by giving reason.]
Speaker previously thought: Price is $109. (Not P)

Speaker now thinks: Price is $40 (i.e., price is not $109 = P)
Question: Not 109 meh? (P meh?)
What is striking about the descriptions of lah and meh is that they tell us different things
about each particle. In the case of lah, some of the descriptions contradict other
descriptions. For example, lah is said to show friendliness, which is the opposite of
another description, that of lah as a conveyor of hostility. I believe that the reason for the
varied descriptions could be that the descriptions are partial pictures of what the particle
is. It is like the traditional Chinese story of the three blind men describing what an
elephant is. The descriptions of the elephant include “like a wall” (body of the elephant),
“a snake” (tail) and “a fan” (ears).
Thus, while such findings shed some light on what lah and meh do, the problem of how
the varied uses of the particles are related to some general description of its properties
have not been dealt with. The relationship between the different functional interpretations
is also not accounted for. While looking at the inherent meaning of the particles, I am also
concerned with the meaning of the particles in different contexts. I will attempt to offer
an explanatory account of how lah and meh can be interpreted in various contexts within
the framework of Relevance Theory. By doing so, I would like to show that lah and meh
can be said to contribute to the relevance of utterances and that Relevance Theory can
provide an adequate set of tools for the description of DPs and illustrate its use in
communication.
3.4. DPs and syntactic positions
It appears that certain DPs can occur only in particular syntactic positions (Table 2).
Table 2. Syntactic distribution of DPs in SCE
Declarative Imperative Declarative Yes/No- Wh-

statement question question question
lah + + +
meh + +
hor + + +
hah + +
ah + + + +
leh + +
ma + +
nah +
what +
lor + +
In this study, I would like to concentrate on two DPs, lah and meh, and the syntactic
positions in which they occur or do not occur. I have observed that lah does not occur in
yes/no-questions and declarative questions (see section 4.2). It appears that meh is not
used with wh-questions. These restrictions are not in the literature, and an explanation
should be part of the full description of the particles.
4. Model
As mentioned earlier, DPs are polyfunctional. An adequate account of DPs would need to
account for the different readings of the DPs. In other words, how do we account for the
many varied functions of, for example, the DP lah? I believe that an understanding of how
the particle is interpreted in actual discourse should be of theoretical interest. For
example, it would address the problem of how the varied uses of particular particles are
related to some general description of their unique properties. It would also reveal ways in
which languages have evolved systematic means of coming to terms with problems that
communicators are faced with in conducting everyday practical affairs through
conversational interaction. In my model, I argue that although a linguistic expression,
such as a DP, performs various functions, it contains a certain cognitive meaning. An
adequate model would also need to account for the relationship between the different
readings and the context. These two points will be demonstrated through the discussion of
two DPs, lah and meh, in particular how they are interpreted in discourse. Wong (2003)
provides the inherent meaning but does not specify the context, nor does he relate
meaning to syntax, which is useful in our understanding of how the particles are
interpreted in discourse (Gupta, this volume). My account, like Gupta’s (this volume),
connects the meaning of DPs with the syntactic structures in which they occur.
In SCE, most studies of DPs have concentrated on the different uses of various particles,
but no general framework, except for Gupta (1992), is provided in which these can be
seen as parts in relation to a whole. Gupta attempted a unified account by placing the
eleven pragmatic particles of SCE (eight are mentioned in Gupta, this volume, Table 1)
expressing varying degrees of commitment to an utterance on a single scale of
assertiveness: contradictory, assertive, and tentative. Although this is a step forward in
describing DPs, I thought that it would be better to put aside arranging the particles on a
scale until we have a clearer understanding of the properties of the various individual
particles.
This study is an attempt to address the problem of how to describe the varied uses of
particles by providing a unified account of the various particles. I argue that we need to
have a unified meaning of the various particles in order to understand how the DPs are
interpreted in actual discourses. This can be done by using an approach that is concerned
with the cognitive processes involved in utterances. I explore the attitudinal functions of
various DPs in SCE using the relevance-theoretic framework. From the relevance-
theoretic point of view, DPs can be seen to facilitate inferential processes. For example,
Blakemore argues that DPs such as but, so, and after all constrain the derivation of
implicatures. There are other DPs that are “higher-level explicatures obtained by
embedding the proposition expressed under an appropriate speech-act or propositional-
attitude description” (Wilson and Sperber, 1993: 14). My study is a step in this direction
of linguistic enquiry. My concern is to examine the meanings of the DPs in terms of three
fundamental issues: the kind of meaning that the DP encodes (i.e., conceptual or
procedural), its effect on the truth conditions of the utterance it is attached to (i.e., truth-
conditional or non-truth-conditional), and the manner in which the DP constrains the
relevance of the host utterance.
4.1. A relevance-theoretic account of lah
This section offers a possible interpretation of the particle lah using the relevance-
theoretic approach. Consider (24):
(24) Context: A and B are discussing how the economic downturn has affected
business and as a consequence organizations have to be prudent to protect the
interests of shareholders.
A: So you know we are not spared lah. Okay we are not spared lah.
B: Uhm nice to know that I am not alone in all this.
A: You are not spared okay. (ICE-SIN-S1B-077)
(25) A’s contextual assumptions include:

Premise 1. The economic downturn has affected A’s business.
Premise 2. A knows that other businesses have been also affected by the
downturn.
Premise 3. A wants B to know that his business has been affected by the
downturn.
Premise 4. A wants to reassure B that he has his understanding.

Premise 5. A knows that he has to be prudent.
A’s intention in the utterance is not only to inform B that they are not spared the
consequences of the economic downturn but also to indicate the speaker’s desire for the
hearer to recognize the shared assumption behind the utterance, something such as (26),
which is similar to Premise 4.
(26) A wants to reassure B that he has his sympathy.
This shared assumption is recognized when B says that it is nice to know that he is not
alone in all this. B feels reassured. What lah does is to instruct the hearer about the
speaker’s desire for the hearer to recognize the shared assumption behind the utterance.
Let us now take a look at a lah-appended utterance involving an imperative.
(27) Context: Two friends (ladies), A and B are having a meal outside the home. B has
not been eating her food. They have to go off soon for an appointment. B knows
that she needs to eat her food quickly.
A: Eat your food lah.
B: Sorry.
In (27), without lah, Eat your food can sound impatient, given the right intonation. With
lah attached to Eat your food, A is not simply asking B to eat her food but is rather
signaling for the hearer to recognize an extra message, for example, that they do not have
much time. The speaker knows that B understands this message.
(28) A’s contextual assumptions may include:

1. A wants B to eat her food.
2. A and B have to go for an appointment.
3. A wants B to understand that they do not have much time.
4. A should have eaten the food some time ago.
What lah does is not to indicate impatience but to convey the speaker’s desire for the
hearer to recognize a shared assumption made manifest in the context. The assumption
made accessible is something like premise 3 (the extra message that the speaker would
like the hearer to recognize). In (28), B knows that A wants her to understand that they do
not have much time, and the speaker thinks that A knows it (shared assumption). In
attaching lah to her utterance, A shows that she expects more from B than just to carry out
an action. B’s reply, Sorry, indicates to A that she understands the urgency, thus showing
that she knows what A meant. Lah helps to convey the speaker’s desire for the hearer to
recognize the shared assumption made manifest in the context. In other words, the speaker
desires that her (informative) intention to make manifest the shared/common assumption
be fully recognized by the hearer.
Our analysis of lah can explain, I believe, all the communicative effects ascribed to lah in
the literature (as mentioned in section 2). It is able to account for the solidarity and rapport
felt between communicators. If I make known to you that there are common assumptions
between us, I am treating you as someone I can relate to, as a member of a certain
community which is also mine. In so doing, I create an impression of rapport between us.
The present account also explains why lah added to an imperative appears to be
persuading or pleading. In (14) “Come with us lah”, the speaker uses lah to signal to the
hearer her desire to have the hearer recognize the shared assumption behind the utterance.
In a context where the hearer appears unable to recognize the speaker’s intention to draw
on a shared assumption (e.g., it will be good for the hearer to go along with them), lah can
be interpreted as an attempt to persuade the hearer to accept the speaker’s point of view.
That lah appended to utterances adds an element of annoyance or impatience can also be
accounted for in our explanation of the particle. If the hearer appears not to recognize the
shared assumption as desired by the speaker, then the speaker’s insistence that he do so
may have a touch of annoyance or impatience. This does not mean that lah contains
annoyance or impatience in its semantic make-up.
The various functions of lah as described by previous researchers, such as obviousness,

friendliness, consultativeness, etc., can be subsumed under our description of the particle.
Lah encodes the speaker’s desire that her/his hearer recognize the shared (common)
assumption behind the utterance, which in turn, functions as an explicit guarantee of
relevance. As a consequence of such an explicit guarantee of relevance, the hearer is
encouraged to expand the contextual assumptions in order to obtain the intended
contextual effects.
I have shown so far that lah is used when the speaker intends to draw on the shared
assumptions behind the utterance. There are two ways in which linguistic meaning
contributes to the interpretation of utterances. It may encode conceptual meaning on the
one hand, or it may contain procedural information, that is, instructions for processing
propositions (Blakemore, 1996: 151). In the case of lah, the particle appears to belong to
the group of linguistic entities that encode procedural meaning. In my analysis, it can be
seen that lah does so by signaling the speaker’s intention to make the shared assumption
be fully recognized by the hearer, which in turn functions as an explicit guarantee of
relevance.
The distinction between truth-conditional and non-truth-conditional meaning does not

equal the distinction between conceptual meaning and procedural meaning. For example,
sentence adverbials such as frankly and seriously are treated as non-truth-conditional, but
they encode conceptual meaning (Wilson and Sperber, 1993; Blakemore, 1992). In the
case of lah in SCE, the particle can be omitted without affecting the truth-conditions of
the host utterance. Consider (29):
(29) a. They generally don’t take beef lah. (ICE-SIN-S1A-023)

b. They generally don’t take beef.
For both (29a) and (29b), the proposition is the same, namely, that there is a group of
people who generally do not eat beef. If the particle is omitted, as in (29b), there is no loss
in propositional meaning. In (29a), what the speaker wants to say is not “They generally
don’t take beef” but that the speaker need not spell this out; the hearer should be able to
gather that from the context. In this way, the presence of the particle allows the hearer to
process the utterance in the smallest context, thereby making for optimal relevance.
The present account of lah has several advantages. First, it explains the difference
between an utterance with lah and one without lah, which has not been adequately dealt
with in the previous studies. A relevance-theoretic account of lah explains the difference
in the following manner. By providing an overt guarantee of relevance, lah can guide the
hearer to explore assumptions implicitly communicated by an utterance with lah,
including contextual assumptions intended by the speaker. An utterance without lah does
not have this encouragement.
Next, I argue that lah is used procedurally by the speaker, and that the various uses of this
particle can justifiably be subsumed under a single description of a code marker
instructing the hearer to recognize the shared assumption behind the utterance.
4.2. A relevance-theoretic account of meh
In this section, we can see that meh can be similarly analysed using the above framework.
Consider the following exchange:
(30) Context: An overseas student having a conversation with someone in Singapore.

a. A: You got overdraft facility meh? (ICE-SIN-S1A-062)
B: Ya. Students have overdraft facilities there.
b. A: You got overdraft facility?
In (30a), the speaker is doubtful that B has overdraft facility. A previous assumption at
time T1, namely, that B does not have overdraft facility, is challenged by one that is
becoming manifest to her at the present time T2 (that B has overdraft facility). In other
words, there is an incongruity in A’s cognitive environment. The hearer processes the
utterance in a manner to help the speaker meet the challenges posed. By answering in the
affirmative and providing further verification (Students have overdraft facilities there), the
hearer provides the necessary auxiliary assumptions to help the speaker resolve the
incongruity.
What distinguishes (30a) from (30b)? The presence of meh restricts the possible
interpretations of (30b) by indicating that a previous assumption in the speaker’s cognitive
environment is challenged by one that is recently manifest in the situation. In (30b), the
fact that A asked the question gives the hearer grounds for assuming that he has
immediate access to some context in which the information the speaker is presenting is
relevant. However, the form of utterance (30b) does not suggest that the speaker expects
the hearer to have any particular idea of what exactly this context is. In other words, the
hearer is free to assess a whole host of meanings of the question including commenting,
keeping the conversation going, seeking confirmation, and seeking agreement. In the meh-
appended utterance (30a), the speaker expects the hearer to access a specific contextual
assumption. The m e h particle guides the hearer particularly towards a certain
interpretation to the exclusion of the other possible interpretations. In other words, the
addition of meh in (30a) constrains the relevance of the question, thus saving the hearer
processing effort. Meh indicates to the hearer that a previous assumption is challenged by
one that is recently manifest in the external environment.
In (30a), that the speaker knows the answer to the question she is asking is becoming
manifest in the external environment. If the answer is not manifest (at least partially), meh
is not used. Perhaps, this explains why meh does not occur in wh-interrogatives such as
“*Where are you meh?” and “*Which one is it meh?”, where the answers to the questions
are not manifest in the speaker’s cognitive environment.
Relevance in a meh-appended utterance, for example (30a), consists in resolving the

incongruity that is the result of the challenge posed by an assumption recently manifest
and a prior assumption in the speaker’s cognitive environment. We can predict, then, that
when there is no proposition expressed by the question being manifest in the external
environment, meh will not be used.
Let us consider another example:
(31) Context: B has just returned home with several shopping bags full of things and is
emptying them.
a. A: So many meh?
b. B: There was a 20% storewide discount.
This use of meh can be frequently heard in SCE. The speaker A utters (31a) when the
hearer is emptying her bags. A’s utterance suggests something like “I can’t believe it”
and shows that he cannot accept the proposition expressed by the question (she bought so
many things) in the given context. The use of meh indicates that the external environment
(the situation) poses a challenge to the contextual assumptions (32) manifest in his prior
cognitive environment.
(32) A’s contextual assumptions may include:

1. B is not working.
2. B does not have a lot of money.
3. Money can be put to better use.
4. B should not spend so much money on shopping.
Meh guides the hearer to access and consider assumptions explicitly communicated by an
utterance, assumptions that may or not be retrieved in the absence of the particle. The
assumption made accessible by the proposition in (31a) is something like assumption 4.
The hearer would choose the first interpretation coherent with the principle of relevance.
In a sense, A is inviting B to resolve the incongruity. B explains why she has bought so
many things. In other words, B tries to provide auxiliary assumptions to allow A to make
the proposition compatible with the context set (external environment). She makes the
utterance there was a 20% storewide discount, indicating that it is value for money. The
proposition made manifest by (31b) helps to resolve the incongruity between the
proposition made manifest by the external environment and the proposition as it would be
derived from the previous contextual assumptions in A’s cognitive environment.
On the other hand, let’s say B’s answer is:
(33) They are not mine. Juliana (sister staying with the couple) asked me to bring them
home for her.
Given the explanation, A will be able to accept the situation. In this case, the answer
given reinterprets the external evidence and has the effect of modifying the proposition
expressed in (31a).
Examples (21) by Gupta and (23) by Wong can be similarly explained. Let us take a look
at an example involving a rhetorical question.
(34) Context: A asks B if she had bought any classical video cassette discs (VCDs) at
the shop. A has forgotten that the shop did not carry any classical VCDs. B
reminds A of that with a rhetorical question.
A: Did you get any classical VCD?
B: They got classical VCD meh?
“They got / have classical VCD?”
(35) B’s contextual assumptions may include:

1. The shop sells VCDs.
2. A has bought some VCDs from the shop before.
3. B and A have looked for a classical VCD from the shop before.
4. The shop does not sell any classical VCDs.
5. A knows that the shop does not sell any classical VCDs.
However, the external environment (36) challenges B’s previous cognitive environment
(e.g., contextual assumption 4.):
(36) The shop sells classical VCDs.
B’s previous contextual assumption (e.g., The shop does not sell any classical VCD) is
challenged by one in the situation, resulting in an incongruity. B challenges A’s
implication that the shop sells classical VCDs. The incongruity can be possibly settled by
A saying something like What was I thinking about? Of course the shop does not sell
classical VCDs. B’s utterance serves as a reminder to A. In this case, the assumption that
the shop does not have any classical VCDs is strengthened.
On the other hand, A’s answer may cancel the proposition expressed, for instance, in (37).
(37) A: The shop got, but not the one I want.
Here, the answer provides clues for B to modify her prior cognitive environment in that it
cancels some previously manifest assumption (4: The shop does not sell classical VCDs)
and replaces it by another to make it compatible with the proposition.
In light of the above analysis, it is clear that meh signals to the hearer that an assumption
recently manifest in the external environment challenges an existing one in the cognitive
environment of the speaker. The resulting incongruity needs to be resolved. The speaker is
asking the hearer to provide the premises required to allow her to access the proposition
and to process it against a context with the aim of resolving the conflict. The DP meh does
not encode conceptual meaning. Its function is to impose guidance on relevance in virtue
of the inferential connections it expresses.
The study provides a way to account for the relationship between different functional
interpretations of DPs and the relationship between the different interpretations in various
contexts. DPs can be studied in terms of three fundamental issues: the kinds of meaning
the DPs encode (i.e., conceptual or procedural), their effect on the truth-conditions of the
host utterances (i.e., truth-conditional or non-truth-conditional), and the manner in which
the DPs constrain the relevance of the utterances they are appended to.
5. Broadening the picture
This study has given synchronic descriptions of individual DPs, lah and meh, within the
framework of Relevance Theory. Relevance Theory, a pragmatic theory of information
processing has provided a set of tools for the description of how the particles can be
interpreted in discourse. In my study, DPs can be observed as expressions which are
syntactically optional, non-truth-conditional, contain procedural meaning, and serve as a
guide in the inferential stage of the interpretation process.
This study calls to attention the two types of cognitive processes that are employed in
understanding utterances: decoding and inference. For example, in They generally don’t
take beef lah, lah signals the hearer to recognize the shared assumption behind the
utterance. What this shared assumption is will have to be recovered by taking the context
into consideration and by inference. In this case, it means “I shouldn’t have to spell this
out!” The DPs are linguistic entities that trigger a procedure or an instruction for the
hearer to follow. What the DP-appended utterance means can be specified not by
decoding alone but by taking into consideration the context and by inference. Thus, this
model can help shed some light on the how utterances are interpreted, that is, through
decoding and inference. I hope that this study can draw attention to these observations and
encourage further work on them in the goal towards a more refined understanding of what
DPs, their functions, and classifications are.
Acknowledgements
I am indebted to my informants, without whose cooperation this study would not have
been possible in its present form. I would also like to thank Vivienne Fong, Ni Yibin, and
Lionel Wee for their helpful comments on an earlier paper on the meh particle; and
Anthea Gupta and Desmond Allison for suggestions on content and wording of an earlier
draft of this paper. Any errors that remain are my own.
Notes
1
For a more detailed description of the contact variety, Singapore Colloquial English or
Singlish, as it is more popularly known, please refer to Gupta (this volume), Foley
(1998), Wee (1998), Brown (2000), and Lee-Wong (2001).
2
The lexical corpus of ICE-SIN was compiled at the Department of English Language and
Literature, the National University of Singapore. The initiative was supported through
a series of funded research projects. It was completed in April 2000 thanks to the
NUS-funded project A study of Definite Noun Phrases in Singaporean and British
Discourse (RP3982058). It is a one-million-word corpus, consisting of 500 texts (200
written and 300 spoken) of approximately 2000 words each. The data used in this
study is taken mainly from the spoken texts.
3
Grice’s Modified Occam’s Razor states: “Avoid multiplying senses (or polysemous
facets) without necessity”, cut back on a growing conflation of (linguistic) meaning
with use (Grice, 1989: 47). What Grice is saying is to avoid multiplying senses (or
polysemous facets) without necessity.
4
The distinction between conceptual and procedural meaning is a useful one, as it
expresses the intuition that there are different types of linguistically encoded
information. It is also useful in explaining expressions that do not affect the
propositional content of an utterance. Due to constraints of space in this paper, a
detailed account is not possible. I refer the reader to the literature (Blakemore, 1987,
1992, 2002; Wilson and Sperber, 1993).
5
I will capitalize on Andersen’s view on propositional attitude, that “we not only express
propositions, we also express different attitudes to them. That is, we communicate how
our mind entertains those propositions that we express” (2000: 3).
Chapter 10
From procedural meaning to processing

requirement
Thanh Nyan
1. Introduction
1.1. Approach
My approach to discourse markers (henceforth DMs) owes much to Anscombre and

Ducrot’s Argumentation Theory (AT) in the sense that I construe their meaning in
procedural terms, or in terms of constraints on interpretation which operate at specific
levels of analysis. My current position, however, differs from AT’s in one important
respect: where AT sees in DMs a means to achieve theoretical ends, my main concern is
to further my understanding of the type of procedural meaning they are and the role they
play in an overall processing strategy. For that reason, I am also taking on board
assumptions about language and the brain (Deacon, 1997); in particular:
1. Language and the brain are coevolved.

2. Language, as a latecomer, has co-opted strategies, systems, and structures already in
place for other purposes.
3. Just as there are brain adaptations to language, so there are language adaptations to
brain systems, of which the emergence of DMs is one.
Viewing DMs from the perspective of AT necessarily entails focusing on uses that have a
direct bearing on its development as a theory. Through such uses, however, DMs can be
functionally linked to a wider range of phenomena, known as argumentational, and
described by means of the same notions.
The need for a coevolutionary perspective arises in connection with procedural meaning
and the notion of topos1. A coevolutionary perspective allows the emergence of
procedural meaning (as exemplified by DMs) to be subsumed under a strategy evolved by
language in response to processing needs specific to the text level. It also provides support
for construing procedural meaning in terms of contextual constraints, as underlying means
do exist that subserve this type of encoding. Topoï, whose postulation has often been
called into question, can in fact be seen to have a direct connection with categorization
and the way factual knowledge is structured (Damasio, 2000: 179).
168 T. Nyan
For ease of reference I shall begin with an outline of AT, one which highlights aspects
having a direct bearing on DMs. A basic assumption of AT is that the core meaning of an
utterance must be construed in terms of its idealized context of utterance, including
speech participants, their stances towards viewpoints being expressed, and discursive
intentions. In other words, utterance meaning consists of linguistic constraints on the
interpretation of the cotext and the context of utterance. This leads to a notion of deep
structure (or phrase) with an argumentational component alongside an informational one.
The former, a set of instructions relating to how the utterance is to be interpreted, is
regarded as primary, the latter as derivative. The relationship between these two
components may be seen as AT’s central claim. All major conceptual changes AT
underwent over the years can be put down to its attempts to uphold this idea or to
radicalize it further. The details of what empirical evidence was brought to bear
(Anscombre, 1989: 15-7, 1995a: 17-21) are less relevant to our purposes than the way the
deep structure was first construed and then gradually evolved, thereby giving rise to
certain parameters, including levels of description, particularly suitable to DMs. As a
matter of fact, well into the late eighties the study of DMs provided AT with its most
crucial empirical input (Bruxelles, Ducrot, and Raccah, 1993 constitutes a major departure
in that respect.)
At an early stage the deep structure came to be described in polyphonic terms. Thus,
utterances consisted of two of more viewpoints (énonciateurs) imputable to the speaker
(S) and to the hearer (H) to varying degrees. The notion of viewpoint was subsequently
redefined to give way to those of topos and topical form (forme topique).
AT‘s ongoing theoretical agenda helps explain its long standing concern with connecteurs
and opérateurs argumentatifs (e.g., mais ‘but’, d’ailleurs ‘besides’, donc ‘so’, même
‘even’, justement ‘precisely’, and peu ‘little’, un peu ‘a little’, ne . . . que ‘only’, presque
‘almost’, à peine ‘hardly’, au moins ‘at least’, respectively). Argumentational connectors
(also known as connecteurs pragmatiques) are defined by Ducrot (1983: 9) to be “. . .
signes qui peuvent servir à relier deux ou plusieurs énoncés, en assignant à chacun un rôle
particulier dans une stratégie argumentative unique.” (“. . . markers whose function is to
relate two or more utterances by assigning to each a specific role within an overall
argumentative strategy”.)
Where argumentational connectors (ACs) impose an argumentational relationship

between two or more host utterances, argumentational operators (AOs),2 which have
single host utterances, are seen to directly affect the latter’s deep structures. Ducrot (1983:
10) provides the following definition of AOs:
Un morphème X est un opérateur argumentatif s’il y a au moins une phrase P telle

que l’introduction de X dans P produit une phrase P´ dont le potentiel d’utilisation
argumentatif est différent de celui de P. (“A morpheme X is an AO if there is at
least one deep structure P such that inserting X into P gives rise to a deep structure
P´, whose argumentational potential differs markedly from that of P.”)
Ducrot is careful to point out that this difference is not predictable on the basis of the
informational contents in P and P’: “. . . cette différence ne pouvant pas se déduire de la
différence entre la valeur informative des énoncés de P et de P.”
From procedural meaning to processing requirement 169
Consider the following example of AO (drawn from Anscombre, 1995b: 35-37):
(1) Il est huit heures. (It is eight o’cloc.k) (= P)
The deep structure assigned to P specifies the set of conclusions (r1, r2) that can be
derived from P:
(1a) r1: Il est tôt (It is early)

(1b) r2: Il est tard (It is late)
Which of the two is actually selected is determined in situ. In the absence of an operator,
both possibilities are open, as attested by:
(2) Il est tôt, il est huit heures.
(3) Il est tard, il est huit heures.
The presence of déjà ‘already’ or pas encore ‘not yet’ has the effect of modifying
—usually restricting—the range of conclusions that can be drawn from the resulting
argument (P´). Thus, déjà in (4) rules out r1, and (pas) encore in (7) has a similar effect
on r2:
(4) *Il est tôt, il est déjà huit heures.

*It is early, it is already eight.
(5) Il est tôt, il n’est pas encore huit heures.

It is early, it is not yet eight.
(6) Il est tard, il est déjà huit heures.

It is late, it is already eight.
(7) *Il est tard, il n’est encore que huit heures.

*It is late, it is not yet eight.
In light of what precedes, one would not expect the sequence in (8) to be possible:
(8) Dépêche-toi: il n’est que huit heures.

Hurry up: it is only eight.
Yet, a context can easily be found where it would make perfect sense. Thus (8) could
serve to urge someone to try and catch a train, who has previously assumed it was too late
to do so.
Counterexamples of this type alerted AT to an aspect of argumentational relationships it

had hitherto not given much thought to: linking an argument and a conclusion could
involve if not a circuitous then at least a less direct route than was formerly assumed. This
led to la théorie des topoï (in its standard version). Where previously, argumentational
relations were binary, they now include a further term in the form of topoï, whose
function is to orient the hearer to inferential routes (itinéraires) linking arguments and
170 T. Nyan
conclusions. Topoï are reasoning principles consisting of two predicates, an antecedent

and a consequent, each featured as a variable grade on a separate scale, with both scales
standing in a scalar relationship. Consider (9), borrowed from Ducrot (1988: 4):
(9) Il fait chaud. Allons à la plage.

P r
It is warm today. Let’s go to the beach
In uttering (9), S is said to be implicitly referring to a topos involving two predicates, one
to do with outside temperature, the other with the desirability of going to the beach
(schematically, TEMPERATURE and BEACH ). These two predicates stand in a scalar
relation, in the sense that each variation in temperature is correlated to a different level of
enjoyment to be derived from a day at the beach. Topoï can be invoked under different
aspects (or topical forms), as determined by the surface structure. In (9), the appropriate
topical form is [+TEMPERATURE, +BEACH]. In (10) it is [–TEMPERATURE, –BEACH]:
(10) Il ne fait pas chaud. N’allons pas à la plage.

It is not warm today. Let’s not go to the beach.
Other possible topical forms the same topos can assume—though not in (10)—include
[+TEMPERATURE, –BEACH], [–TEMPERATURE, + BEACH].
In light of the above considerations and changes, the role of these operators was redefined
as one of restricting the range of possible inferential routes, whereby the new found gap
between arguments and conclusions may be bridged. Consider again Il est huit heures and
Il n’est que huit heures (borrowed from Anscombre, 1995a: 41-2). Il est huit heures ‘It is
eight o’clock’ on its own leaves open the possibility of reaching r1, Dépêche-toi ‘Hurry
up’, or r2, Prends ton temps ‘Take your time’, via either of T1 or T2, in the former case,
and T3 or T4, in the latter. Consider (11):
(11) T1 = Plus le temps manque, plus on doit se dépêcher.

The less time one has in which to do something, the more reason to get on with it .
T2 = Plus on a de temps, plus on doit se dépêcher.

The more time one has in which to do something, the more reason to get on with
it.
T3 = Plus le temps manque, moins on doit se dépêcher.

The less time one has in which to do something, the less reason to get on with it.
T4 = Plus on a de temps, moins on doit se dépêcher.

The more time one has in which to do something, the less reason to get on with it.
By contrast, in the case of Il n’est que huit heures ‘It is only eight o’clock’, ne . . . que
rules out T1 and T4 as possible inferential routes to r1 and r2, respectively.
As can be expected, connectors also got redefined in terms of topoï and topical forms.
Consider this analysis Ducrot provides of mais:
(12) Il fait chaud, mais je suis fatigué.

P Q
It’s warm, but I’m not feeling well.
. . . le premier segment de cet énoncé montre un énonciateur qui applique à la

situation le FT “plus chaleur, plus agrément” (je suppose que cet énoncé global
est, par exemple, utilisé pour refuser une proposition d’aller à la plage, et le
locuteur lui donne son accord. Mais le locuteur met en scène un autre
énonciateur—auquel il s’oppose—exploitant cette FT pour suggérer une
conclusion (la baignade) réfutée par l’argumentation contenue dans le second
segment. (Ducrot, 1995: 89-90)
The gist of this is that assuming that (12) as a whole serves to turn down the suggestion of
a day at the beach, P features two énonciateurs: E1 and E2. E1 takes the view (= V1) that
the appropriate topical form to bring to bear on the situation is (TFa), /The higher the
temperature, the higher the level of enjoyment/. E2 takes V1 further in suggesting that
they should go to the beach (= V2). S’s position is defined in terms of going along with
V1, though not V2, and his view, as expressed in Q, is that they should not go to the
beach. Q can be similarly analyzed, in terms of two further énonciateurs (E3 and E4),
whose views coincide with those held by S. E3 sees the situation as requiring the
intervention of a topical form (TFb), /The less one feels up to something, the less reason
to do it/ (= V3). E4 takes V3 further in bringing TFb to bear on the situation at hand,
which amounts to turning down the proposed outing.
Further clarification of the role of ACs and AOs is found in Ducrot (1995: 98). As mots
vides (‘empty words’), ACs and AOs do not, as such, invoke topoï or topical forms, which
are associated with lexical content-words (e.g., courageux ‘courageous’). Rather, they
function to modify them in various ways. Thus justement ‘precisely’—an AC—will
change a topos or topical form into its converse. Consider the following examples
provided by Ducrot:
(13) A : Tu dois être content: il fait chaud.

You must be pleased, with this warm weather.
B: Justement.
Precisely.
Where the initial topos involves a correlation between warm weather and satisfaction, the
converse correlation, brought about by justement, concerns warm weather and
dissatisfaction.
(14) A: Pierre doit être content: je crois qu’il a réussi à son examen.
Pierre must be pleased: presumably he has passed his exam.
B: Justement il n’a pas réussi.
That’s precisely not the case.
In (14) the initial topos, a pairing of levels of satisfaction and levels of success, is left
untouched, while the initial topical form , + SUCCESS, + SATISFACTION, which is subject to
the inversion , gives way to – SUCCESS, – SATISFACTION.
172 T. Nyan
By contrast, the only change brought about by peu ‘little’, an AO, is to invert the topical
form. Consider the following argument against a job applicant:
(15) Il a peu d’experience.

He has little experience.
The topos associated with the predicate [avoir de l’experience] is one linking levels of
experience and degrees of suitability for a job. The aspect under which it is envisaged
corresponds to the topical form, +EXPERIENCE, +SUITABILITY. Peu has the effect of
changing this topical form into –EXPERIENCE , –SUITABILITY , leaving the topos itself
unchanged.
To the extent that the subsets of DMs under consideration require pre-existing topoï to act
on, something further needs to be said about them. As mentioned earlier, topoï are
associated with lexical items with a full conceptual content. What this association entails
lies in AT’s construal of word meaning, which is consistent with the way it defines
utterance meaning. Thus, Anscombre (1995a: 52) argues against an objective notion of
good weather. Inasmuch as our perception of what constitutes good weather can hardly be
dissociated from activities such as going for a swim or a walk, constructed topical
notions, he proposes, should be substituted for it. In brief, the meaning of a word is best
construed in terms of situations and scenarios featuring stances frequently taken towards
the referent and practices typically associated with it. Saying of someone that he is
wealthy evokes the idea that he can afford anything he fancies. This correlation between
wealth and buying power is an example of “intrinsic” topos.
As an integral part of word meaning, intrinsic topoï are apparently a source of preferred
continuations in context-free situations. Their postulation accounts for the contrast
between these two sequences (borrowed from Anscombre, 1995a: 57):
(16) Pierre est riche: il peut s’offrir n’importe quoi.

Pierre is wealthy: he can afford anything he fancies.
(17) Pierre est riche: il est donc avare.

Pierre is wealthy: he is therefore mean with his money.
The sequence in (16), which is based on an intrinsic topos, seems perfectly natural in a
context-free situation. (17), on the other hand, sounds odd in the absence of further
contextual assumptions, for instance in the form of an extrinsic topos linking wealth and
meanness with money.
1.2. Data and methodological issues
The examples I have been concerned with to date all come from contemporary Standard
French, as spoken and written by native speakers from France. Data, as I construe them
(i.e., in relation to my descriptive goals), cannot satisfactorily be characterized in terms of
the attested versus intuitive dichotomy. Even in the absence of a possible theoretical
agenda (as is the case with AT), the significance of attested data needs to be put into
perspective: what counts as data is often—albeit to varying degrees—determined by
parameters inherent in observational languages and research questions.
The next issue is whether marginal cases are worthy of description. This issue is linked to
how much of S’s internalized knowledge one sets out to describe (with or without
attendant ontological claims). Despite AT’s concern with “idealized discourse”, the data
on which its descriptions are based are assumed to exhibit a maximal degree of
acceptability. This said, to the extent that marginality can be dependent on practical
considerations on the part of researchers and the nature of framework, AT’s lucid take on
the matter is that there can be a fine line between acceptable and marginal cases and that
arbitrary decisions are a fact of life.3 My own view is that certain marginal cases should
be included in their own right, rather than as “naturalized” acceptable cases, for the
purpose of tightening one’s description. I have argued elsewhere (Nyan, 2000: 3-9) that
taking them into consideration would enable one to form a more specific idea of users’
internalized knowledge of DMs; thus, in addition to a core meaning, consisting of a
network of prototypical properties, there would be an extended inferential capacity to
account for users’ ability to arrive at positive judgements of cases that do not elicit a
spontaneous response. Such a capacity would also underlie the greater ease with which
further cases of a similar kind are subsequently handled. A core meaning thus construed
would be susceptible to recategorization.4
AT’s originality lies in its use of an “internal descriptive viewpoint” (Nyan, 1998: 21-26),
at a time when external viewpoints were still the norm, as attested by the prevalence of
distributional analysis and conditions of use. AT has been describing DMs in terms of
instructions, rather than merely syntactical and semantic environments, as early as the late
seventies (Ducrot et al., 1980). An immediate consequence of this decision is that a DM’s
environment need not always be verbalized. Consider the following occurrence of enfin:
(18) Enfin!
What on earth is going on! / What do you think you are doing?
(as an expression of shock and indignation on the part of a master stumbling on
pupils in clear breach of school regulations).
In saying (18), the master’s (expressed) intention is that his utterance should serve as a
reminder of those regulations, by way of clamping down on the pupils’ carryings-on. This
main intention is accompanied by a subintention, that of nipping in the bud any prepared
excuse the children may be about to produce. (Cadiot, Ducrot, Fradin, and Nguyen, 1985:
222). (For further reference, note that glosses such as the above appear to be linked to the
use of schematized situations for the purpose of capturing the meaning of DMs at the
interactive level.)
Viewing a DM’s environments from an internal point of view, rather than in terms of
(external) conditions of appropriateness, provides a basis for understanding why a non-
prototypical environment does not necessarily bring the interpretive process to a halt, but
merely slows it down (Nyan, 1998: 83-84).
From its very inception, AT has been concerned with idealized discourse. Inasmuch as
this notion overlaps with that of users’ linguistic knowledge—in particular their ability to
make acceptability judgements—this led to:
1. an account of DMs, whose orientation is predictive rather than merely descriptive;

174 T. Nyan
2. a view that environments whose occurrence has not been attested or can be ruled out
are also worthy of consideration.
This licences the use of constructed examples for testing, corroborating, and contrasting
purposes. As a matter of course, these various tasks are carried out with the help of native
speakers’ intuition. Some aspects of users’ knowledge, which remain largely untapped in
normal communicative situations, can thus be called upon to assess the acceptability and
appropriateness of constructed examples required to test descriptions of individual DMs
or to differentiate between DMs close in meaning (e.g., mais and simplement in Nyan,
1999a: 279).
Given AT’s resolutely noncognitive stance, it is hardly surprising that its descriptions of
DMs should have been based on classical categorization. It is unclear, however, how
widespread Rosch’s work on categorization was at the time. In any case, classical
categorization faces considerable difficulties in dealing with recalcitrant data. A case in
point in that of mais, whose various uses appear to have very few traits in common (Nyan,
1999b: 211-213).
Consider some of the functions mais can take on:
(19) Mais c’est Octavio!

Octavio!
(In a surprised tone at the unexpected sight of Octavio)
(20) Mais c’est Orlando (voyons)!

But it’s Orlando!
(In a reproving tone, as to a dog with an allegedly poor memory)
(21) Il pleut mais j’ai besoin d’exercice.

It’s raining but I need exercise.
(22) Cela ne fait pas une semaine, mais trois qu’elle est partie.
It has not been one but three weeks since she took off.
In (19) and (20) mais expresses a mental state. In (20), but not (19), in expressing this
mental state, S is performing a speech act (e.g., one of telling H off for failing to
recognize Orlando). This mental state is not found in (21) or (22), where the use of mais
gives rise to an act of refutation and an act of rectification, respectively. Furthermore, not
all uses appear to share the oppositive trait. Given the nature of the data, posing the
descriptive problem in terms of common traits may not be the best option imaginable; a
more promising approach may be to assume a common representation for all uses and to
ask what general processes are responsible for certain differences between the various
cases. Such an approach takes differences in meaning as indicative of processing
regularities (for further details, see Nyan, 1999b: 222-236). In this particular instance the
level of abstraction required to accommodate rarefied common properties was, indeed, so
high as to become virtually meaningless. Empirical assumptions aside, the way our
perceptual and memory subsystems operate should, ideally, also be taken into
consideration in choosing the mode of categorization.
1.3. Problem statements
The decision to describe DMs in procedural terms immediately prompts two questions:
1. What types of instruction does their core meaning consist of?

2. How do these instructions give rise to observed interpretations at the text and
communicative levels?
The former can satisfactorily be answered in linguistic terms; the latter, strictly speaking,
is about linguistic constraints on the processing system. To be amenable to an answer in
linguistic terms, the latter would have to be rephrased as follows:
2a. What levels of analysis should be postulated, which are concerned with the above
instructions?
2b. How does their interaction give rise to the various levels of interpretation?
2c. How do functional aspects correlate with the structures postulated?
2d. How do these same functional aspects relate to one another?
Questions which cannot fruitfully be posed within AT but which I would regard as worthy
of consideration include:
• How could the core meaning, as based on prototypical data, be altered to

accommodate speakers’ (extended) capacity to work out an interpretation for certain
marginal cases and the fact that their ability to handle similar cases improves in the
process?
• Is one’s construal of DMs compatible with what is known of brain systems (in
particular of the perceptual and memory systems involved in language processing)?
• Regarding the way DMs contribute to the meaning of their host utterances, what mode
of interpretation is involved? Bottom-up, or top-down? (Nyan, 2004: 24-25, 103-104)
• Can topoï be justified on cognitive grounds?
• Can the emergence of the various types of DMs (as determined by their paths of
historical development) be seen as exemplifying coding strategies specific to the
discourse and text level? And assuming this to be the case, in response to what
changes in the communicative situation and communicative goals (in the sense of
Givón, 1979: 285-305)?
2. Definition and function
2.1. Discourse markers
Discourse markers is used by different researchers to refer to overlapping groups of

lexical expressions (Hansen, 1998a). The only property these expressions appear to have
in common is that they do not contribute to the informational content of the utterances
they are a part of. For convenience’s sake I am, however, happy to go along with this
label, as long as (a) it does not rule out that the category to which it refers can be
construed in terms of instructions and (b) it leaves open the possibility that these
instructions are closely linked to constraints operating at the processing level.
176 T. Nyan
2.2. DMs as a functional category
From the standpoint of AT, DMs (or rather the subsets AT singles out) constitute a
functional category. More specifically, this function concerns context construction. As
pointed out by Rossari (this volume), the advantage of a functional analysis is that it
obviates the need for addressing the issue of syntactic and semantic nonintegration. But
equally significant is the fact that there should be such an issue in the first place. This,
indeed, raises the question as to whether linguistic logic, as inherent in our methodology
and categories, necessarily always keeps up with what we regard as linguistic phenomena.
DMs, as already mentioned, are also viewed by AT from an internal perspective. This use
of an internal viewpoint, however, goes well beyond a procedural conception of DMs;
indeed, it leads to a top-down hypothesis concerning the mode of interaction between
instructions and contents available from host utterances. In the event of a clash between
the two, this hypothesis correctly predicts that cases that do not elicit a spontaneous
judgement of acceptability are not automatically ruled out. Undecided cases can trigger a
further search for less obvious matches (Nyan, 2000: 1-3).
An interesting consequence of this construal of DMs, including their mode of interaction

with their host utterances, is that they can be characterized in terms of mental as well as
verbalized contexts (see (18)). In processing terms, this would fit in with the idea of a
working memory holding what has been previously said as well as surmised. (Discourse
memory, as used by Roulet (this volume) presumably refers to the same type of notion.)
2.3. Which DMs?
The subset of DMs I have been mainly investigating, which corresponds to ACs, can be
characterized in terms of three converging perspectives, as determined by the processing
system, the core meaning, and the users.
2.3.1. DMS as language adaptation

From a coevolutionary perspective, the emergence of DMs may be seen as an instance of
language adaptation to a requirement to which all processing systems are subject, that of
efficiency. Building on Givón (1979: 271-309), the following hypothesis may be put
forward: at the mutipropositional level, with its characteristic increase in the volume of
information to be processed and a communicative goal which tends to be more action-
oriented, the processing system would be unable to cope unless the level of
automatization is stepped up. One way this can be achieved is precisely by means of the
kind of lexical coding found in DMs. Thus, coded instructions relating to the way host
utterances are to be integrated into built-in contexts of relevance allow for a speedier
handling of larger chunks of discourse; such instructions simplify the task of accessing
and constructing mental contexts for interpretive purposes. In some cases, constraints on
instantiation are also available (as will be seen shortly). Similarly, constraints on
continuation of an argumentational kind and built-in contexts of evaluation, as provided
by topoï, can be taken as a further case of language adaptation, one that goes some way
towards securing an automatic response on H’s part (Nyan, 2004: 115-116,139-140).
In support of the above hypothesis, one can invoke a known feature of explicit memory
encoding (Kosslyn and Koenig, 1995: 317), which could have been co-opted to
implement the emergence of built-in contexts. This feature is one whereby perceptual
stimuli are encoded with their context of occurrence before they are stored as patterns of
activation.
Furthermore, the way DMs contribute to the interpretive process bears a strong
resemblance to a processing strategy in which the above patterns of activation participate,
Where a stimulus-response connection does not exist to ensure speed of processing, the
next best thing appears to lie in anticipating the outcome; adjustments, if called for, are
dealt with afterwards. “Anticipation and revision-on- the-fly” (Kosslyn and Koenig, 1995:
239-240) is less costly than computing outputs from scratch. The role played by patterns
of activation in this strategy is to provide perceptual hypotheses in identification tasks.
Built-in contexts and constraints inherent in DMs can be taken to serve a similar strategy
in the identification of S’s communicative intent. This is particularly obvious if one
envisages core meanings in terms of schematized situations, which provide a top-down
view of the interactive level. (For examples of schematized situations, see (18) and (37)
and (38).)
The type of core meaning AT assigns to DMs is consistent with what precedes. The
instructions it consists of appear to speed up H’s interpretive process by allowing various
forms of context constructions, which have a direct impact on cohesion.
Schematized situations, we have just seen, ensure cohesion on the interactive level by
providing a context of relevance for the act carried out through the use of the AC. Making
up this context is a sequence of reference/content pairs corresponding to various types of
stances S or H (as constructed by S) take vis-à-vis the contents at hand. These give rise to
a sequence of acts at a higher level of analysis.5
In addition to a context of relevance, each AC also imposes a network of semantic

relations. Some concern text constituents (e.g., illocutionary acts, acts of argumentation,
S’s or H’s commitment to a view). Note that some text constituents belonging to
schematized situations may not be verbalized; a case in point is the excuse the children in
(18) were supposedly about to blurt out, were it not for the master’s Enfin! Others hold
between text constituents—albeit by and large of lower order—and contextual
assumptions of various kinds. Thus au moins (Nyan, 1993: 2-3) designates pre-
determined hierarchies (e.g., diplomatic corps, church hierarchy, etc.) as its preferred
sources of instantiation for the scale of entities it presupposes. This accounts for the
contrast between (23) and (24):
(23) ?C’est au moins un manguier.

It’s at least a mango tree.
(24) C’est au moins un évêque.

He is at least a bishop.
Another case is simplement (Nyan, 1999b: 281), which requires that the reason it
accompanies should be of a different nature to the one previously rejected. (25), but not
(26), meets this requirement:
178 T. Nyan
(25) Ce n’est pas que ce soit gelé; simplement je n’ai pas faim.
It’s not that it is frozen; it’s just that I am not hungry.
(26) * Ce n’est pas que ce soit gelé; simplement c’est froid.

It’s not that it is frozen; it’s just that it’s cold.
The above types of search process should not be confused with those spontaneously
triggered during the interpretive process, when further contextual assumptions must be
accessed in order to identify appropriate contents. Alongside these constructed contexts,
which, by and large, cater to comprehension, AT postulates a further type, more to do
with evaluation.
As can be expected, contexts of evaluation arise from DMs acting on argumentational

relations, that is, relations between lower-order constituents (e.g., lexical items,
propositional contents) and topoï. Consider (27):
(27) A: Il a de l’argent, il se trouvera un bon hôtel.

He’s got money, he will find himself a good hotel.
B: En fait il a peu d’argent.
As a matter of fact, he has little money.
Here the initial topical form is +M O N E Y , +G O O D ACCOMMODATION , and its inverse,

brought about by peu, is –MONEY, –GOOD ACCOMMODATION.
Peu may be said to construct an evaluation context (i.e., one that justifies drawing the
conclusion that X will not be staying in a good hotel) because it provides an inferential
shortcut that enables A to reach the correct conclusion. This shortcut, as will be recalled,
involves a topos (or reasoning principle), which validates the conclusion in question.
What that evaluation context entails may be further clarified if complex categories is
substituted for topoï. A complex category (at its most basic) is a pairing of a category of
initial situations requiring a decision and a category of response options.6 According to
Damasio (2000: 179), having our factual knowledge structured in terms of such pairings
enables us to reach a quick decision the moment the initial situation has been properly
categorized.
With ACs, the way the evaluation context is brought about appears to be more
straightforward. Consider (28):
(28) Il est beau, mais il ne le sait pas.

He is handsome, but he is not aware of it.
Here we have a case of substitution of one topos for another. Assuming the initial topos to
be about beauty and vanity going hand in hand, mais amounts to a proposal that H should
base his decision on a topos corresponding to the assumption that if someone is not aware
of their beauty, they are not likely to be vain.
Before proceeding to our next point, two comments are in order:

• Though requiring instantiation by appeal to propositional contents and contextual

assumptions, topoï arguably provide an additional layer of constructed context: the
relations in which they stand to one another is specified by DMs.
• It is not uncommon for the relationships imposed by DMs to concern constituents
belonging to different levels of analysis. Argumentational relations thus hold between
elements of the propositional level and topoï, but also between segments of the cotext,
in the form of constraints on sequencing.
If DMs for the most part can be seen as increasing the level of cohesion, some appear to
have a markedly different role, that of enabling texts and discourse to accommodate a
certain amount of inconsistency without jeopardizing their overall cohesiveness. Consider
the following examples:
(29) Ce serait une excellente idée de la faire venir. Maintenant cela ne l’intéresse peut-
être pas.
It would be a great idea to fly her over. The problem is she may not be interested.
(30) Je veux bien l’inviter. Toujours est-il qu’il n’aime pas voyager.
I don’t mind inviting him. The fact remains that he hates travelling.
(31) Cet été je m’enferme à la campagne pour travailler, et je ne vois personne.

Ceci dit, si vous avez envie de venir un week-end, faites-moi un mot.
This summer I am going to bury myself in the country to do some work, and will
be incommunicado. This said, if you want to come down for a weekend, drop me
a line.
All three DMs trigger a context shift but with no adverse effects on cohesion. Maintenant
achieves this by merely drawing attention to the possibility of a new perspective, one
brought about by a new contextual element; toujours est-il, by pointing out how the initial
context of utterance would have been altered, had a certain factor not been overlooked;
ceci dit creates a parallel discourse space in which S can entertain a different view to one
previously set out, without incurring any charge of inconsistency. In all three cases, S can
be seen holding conflicting views while maintaining his initial stance (Nyan, 1998: 123-
147). An appropriate continuation to (29) would be Qu’est ce qu’on fait?(‘What do we
do?’), rather than Donc mieux vaut pas! (‘So we’d better not!’). In response to (30) H
would be entitled to say Ah, n’essaye pas de te désister! (‘Don’t try to get out of it!’) but
not Tu me fatigues à changer d’avis toutes les cinq minutes! (‘How tedious of you to
change your mind every five minutes!’) In (31) a potential inconsistency is averted
because the exception being made for H is conditional on whether the latter fancies a
weekend in the countryside.
2.3.2. User perspectives

DMs, finally, can be envisaged from the standpoint of users. If DMs’ core meanings are
about context constructions of the proposed kind, S’s goal in using a DM is to place
constraints on H’s interpretive process, with the view to leading him or her to the correct
interpretive decision. In other words, if S’s overall intention in producing a U1, DM, U2
sequence (where U1 and U2 stand for the host utterances) is to impose a certain
conclusion, the occurrence of a DM corresponds to a subintention aimed at providing the
180 T. Nyan
means whereby the overall intention is to be implemented. At the interactive level, this
translates into a manipulative intention.
The question that arises at this point is this: assuming that the use of a DM as a set of
encoded procedures successfully induces in H the formation of appropriate mental
contexts, including the correct topoï, on what grounds can H be expected to make the
correct interpretive decision (let alone accept the proposed conclusion)? The answer lies
in the generality requirement, as defined by Searle:
When I say, “That is a man”, I am committed to the claim that any entity exactly
like that in relevant respects is also correctly described as “a man” [. . .]
Furthermore, the generality requirement applies to other people. For if I am
committed to recognizing similar instances as also cases of men, my commitment
in a public language requires that I think other people ought to recognize this and
similar cases as cases of men. (Searle 2001: 159-160)
To accommodate an argumentational construal of lexical meaning, the above definition

can be amended as follows (changes are in italics):
When I say, “That is a man”, I am committed to the claim that any entity exactly
like that in relevant respects is also correctly described as “a man” [. . .], with
what that entails in terms of stances and practices our experience associates with
entities so described. Furthermore, the generality requirement applies to other
people. For if I am committed to recognizing similar instances as also cases of
men, my commitment in a public language requires that I think other people ought
to recognize this and similar cases as cases of men, with what that entails in terms
of stances and practices our experience associates with entities so described.
Much trickier, however, is how to get H to acknowledge that the entity in question—in
the event, the situation of reference—is correctly described as P, or rather is a member of
the category of situation that can be described as P.
This is where it would make a lot of sense to construe acts of argumentation in terms of
categorization. Whether the expression refers to an argumentative sequence (consisting of
an argument and a conclusion) or, by and large, to an argumentational sequence, I propose
that the act being carried out should really be an act of categorization. Consider (32):
(32) A: Je me demande comment Thomas va prendre la chose.

I wonder how Thomas will respond to that.
B: C’est un enfant (Il va probablement mal le prendre).
He is a kid. (It won’t go down well).
B is carrying out an act of categorization in the sense that he is proposing that Thomas is
correctly categorized as a kid. As a member of such a category, Thomas can be expected
to act accordingly, that is, to throw a tantrum, among other things. The assumption here is
that the reference is to a complex category involving a linkage between entities that can
be categorized as children and a typical range of behaviours children are prone to.
In light of what precedes, viewing ACs from the standpoint of users (i.e., in terms of what
S does in order to elicit a certain response in H, and the basis on which he or she can
expect H to respond appropriately) amounts to seeing their use as involving an attempt to
get H to categorize or recategorize (as the case may be) the situation of reference (SR) in a
way that is conducive to the expected response.
What about AOs, examples of which are presque and à peine?
(33) Tu as presque réussi.

You almost made it.
(34) Tu as à peine la moyenne.

It’s a bare pass that you’ve got.
In terms of categorization, they appear to provide the SR, as described in the propositional
content, with an assessment of its membership status within a certain category of
situations, which is associated with a certain category of responses. In (33) the SR is
assigned virtually all the membership criteria, which entails that it should be almost as
effective as Tu as réussi (‘You made it’) in eliciting whatever response the latter normally
elicits. (34) exemplifies the opposite situation: from the perspective introduced by à peine,
to have a bare pass hardly qualifies as a successful situation and does not exactly call for a
celebration.
Redefining acts of argumentation as acts of categorization clarifies the link between what
S does in terms of speech acts and the topoï he or she thereby invokes. Looking at DMs
from the standpoint of the core meaning alone does not provide such a link.
This brings us to our final point, one that further highlights the need for a user’s
perspective. It has to do with a type of manipulation which is not encoded in DMs as such
but depends on a lack of coincidence between the presentation given of the context of
utterance at large and what it in fact is (in terms of shared complex categories and actual
intentions on S’s part). If the complex categories introduced by a DM correspond to
representations in H’s working memory of categories he already possesses, then the DM
simply serves to guide H’s interpretive process. If, on the other hand, H neither possesses
those categories nor integrates them into his long term memory, then we are dealing with
a case of manipulation. Consider (35):
(35) Je veux bien appeler les pompiers. Toujours est-il que la dernière fois ils n’étaient
pas très contents.
Well, all right, I’ll ring the fire brigade. The fact still remains that last time they
weren’t exactly pleased.
Since toujours est-il que Q occurs following an agreement on S’s part to call the fire
brigade and does not result in a cancellation of that agreement, S can legitimately claim to
have meant no more than what toujours est-il says his intention is. Yet, to the extent that
he could, in fact, have intended for H to arrive at the conclusion that he would better not
make that call, toujours est-il is amenable to a strategic use.
182 T. Nyan
3. Model
This section is concerned with three issues:
• What does AT have to offer in the way of relevant levels of description?

• What kind of hypothesis does AT allow one to frame in respect of how the core
meaning interacts with certain elements of the host utterances and contextual
assumptions?
• What has been left insufficiently accounted for that could be better handled by an
extended model, one involving a coevolutionary perspective?
3.1. Relevant levels of description
Given that the main function of ACs is argumentational and that to be carried out it
requires topoï, the very first level they would be concerned with is the propositional level.
Topoï originate from this level (in the sense that they are accessed on the basis of its
constituents). Consider (36):
(36) Sa politique est peut-être cohérente, mais elle est pour la privatisation.
P Q
Her policy may be coherent, but she is in favour of privatization.
The situations described in P and Q give rise to two topoï: T1 (= +COHERENT POLICY, +/–.
. .), and T2 (= +PRO-PRIVATIZATION, +/–. . .). Mais indicates that the consequents, to be
instantiated by appeal to contextual assumptions, must feature opposite types of situation
(e.g., she will be elected vs. she will not be elected).
Inasmuch as mais attributes to H a preference for T1 and to S a preference for T2, and
furthermore ranks T2 over T1, this points to the polyphonic structure as the next level up.
This level, as will be recalled, may be construed in terms of reference point–
argumentational content pairs arranged in a fixed sequence; the latter is determined by the
schematized situation associated with each specific AC. (The reference points in this case
are H and S).
Schematized situations, as mentioned earlier, provide a context of relevance for the main
act carried out by S. This means that they also require a further support level, namely, the
interactive level. If we view the polyphonic structure for (36) in terms of this higher level,
we would have a prior act of H, whereby P is put forward in support of X’s election, to be
followed by a counterargument advanced by S for the opposite conclusion. (Note that
strictly speaking, we are dealing with the representation of a prior act, providing a context
of reference for the act carried out by S through an act of utterance.)
Why schematized situations should also include the lower level mentioned earlier is
because the viewpoints involved are not always construable in terms of past or current
acts. Consider (37), a much simplified (and more reader-friendly) version of a
schematized situation associated with this occurrence of toujours analyzed by Cadiot,
Ducrot, Nguyen, and Vicher (1985: 105-110):
(37) Allons au bistro, on y sera toujours au chaud.

Glossed to its last (cumbersome) nuances, (37) reads ‘Let’s go to a bar. This may not be
your idea of an evening out, but it will keep us out of the cold, which, under present
circumstances, is not to be sneezed at.’
Associated with this use of toujours is a schematized situation consisting of an ordered set
of viewpoints, a simplified version of which is given below:
(38) V1: Bars have much to recommend themselves, since they have central heating.
(= F)
V2: F is a good reason to go to a bar. (= conclusion r)
V3. I would not regard F as a particularly compelling reason to go to a bar.
V4: F does not provide a compelling reason to go to a bar. (= conclusion –r)
V5: Though admittedly not particularly compelling, F, under present
circumstances, is better than nothing.
Unlike the previous case, H cannot have said or thought anything remotely like V1 and
V2 (in response to S’s proposal, that is) before the latter actually came up with it. If V1
and V2 were to be left out (not to mention V3), we would end up with a somewhat
incomplete schematized situation. Another reason for retaining the lower level is that
there are such things as partial commitments to a view, and partial commitments can give
rise to strategic manoeuvres at the interactive level. Consider (36) again. Even though S’s
full commitment is to T2, he also acknowledges T1, though not as a proper basis on which
to draw a conclusion. In other words, this partial commitment has no argumentative
function whatever, hence the possibility that there may be some concessive strategy afoot,
one aiming to place H under an obligation to reciprocate. How much pressure is being
applied depends on whether attention is drawn to this partial commitment. The occurrence
of certes in (39) makes the obligation harder to ignore:
(39) Certes, c’est encore des “mice crispies”, mais le chat ne s’en plaint pas.
I know mice crispies are yet again on the menu, but the cat does not seem to mind.
Concerning the interactive level, should there be a single level for both illocutionary acts
(in the more traditional sense) and acts of argumentation or are two separate levels
required for the characterization of DMs?
Consider (40) (adapted from Nguyen, 1986: 202):
(40) A: On devrait s’en aller d’ici, le coin est infesté de crocodiles.

We should get out of here, the place is infested with crocodiles.
B: Comme tu voudras. Toujours est-il que c’est toi qui en as eu l’idée.
P Q
As you wish. I’ll point out, though, that you’re the one who came up with the
idea in the first place.
Here P marks B’s endorsement of A’s conclusion and Q a potential counterargument B

elects not to put forward. In AT’s terms, we have a case of visée argumentative which is
not taken further: on the basis of toujours est-il que Q, A cannot legitimately impute to B
the intention of arguing against his conclusion.
184 T. Nyan
While one can certainly leave it at that, toujours est-il, arguably, also conveys a mental
state—in the event, a critical attitude on S’s part—the expression of which gives rise to a
challenge or a provocation inviting a certain response. The same applies to toujours, as in
(41):
(41) A: Cette fois je pars demain!

This time, I am leaving tomorrow.
B: Tu n’as pas ton billet, toujours.
I don’t see that happening: you haven’t got your ticket.
A final argument in support of two levels is to do with the fact that the two acts under
consideration are fundamentally different. Consider (42):
(42) It is raining.
P
If (42) is intended as an assertion, the propositional content is viewed in a representational

capacity, and the direction of fit concerns words and world. If, on the other hand, it is
intended as an act of argumentation (i.e., of categorization), the propositional content is
envisaged in a categorizing capacity, and the direction of fit concerns words, (world,) and
mind. In the former case, S’s commitment is to the fact that P is a true representation of
the world; in the latter, this commitment is to the appropriateness of the category
judgment provided by P of the situation of reference. Viewing SR in terms of
categorization implies a reference to a complex category (involving a linkage between an
initial category of situations and a category of responses). Regarding SR as an instance of
bad weather will enable H, through the mediation of the complex category, to access the
appropriate set of responses. In this case, S’s commitment is to a words-to-(world-to-)
mind direction of fit, since the complex category he or she invokes is stored in our
memory system. A further difference between the two types of act lies in what is at issue
in each case. With the assertive, the issue is whether P is a true representation of the SR;
with the act of argumentation (or categorization), it is a matter of whether SR can be
categorized as P and seen as a member of a certain category of situations.
3.2. How does the core meaning interact with host utterances and the context?
AT’s construal of the core meaning in procedural terms and its consistent use of an
internal descriptive viewpoint lead to an overwhelmingly top-down mode of processing.
Such a hypothesis entails (a) interpretive hypotheses, in the form of networks of relations
and schematized situations, that are imposed from above on the cotext and context of
utterance; and (b) the prevalence of instructions over contents (immediately) available
from host utterances. As will be recalled, this correctly predicts that in the event of a clash
between the two, the interpretive process does not necessarily grind to a halt, with the
rejection of the sequence under consideration. A search for suitable matches can get
underway, which brings to bear fresh contextual assumptions. Consider (43):
(43) A: Est-ce parce qu’ils ont beaucoup d’enfants qu’ils n’ont pas d’argent?
Is it because they have a lot kids that they are penniless?
B: Ce n’est pas qu’ils aient beaucoup d’enfants, simplement ils en ont quatre.
It’s not that they have a lot kids, it’s just that they have four.
The clash is between the type of reason provided by Q and the requirement arising from
simplement, which is that this reason should not belong to the same category as P. Despite
such a clash, it is not unusual for interpreters to have another go at this kind of sequence
by bringing to bear relevant contextual assumptions. In the above example B’s reply
becomes amenable to a coherent interpretation in a context where parents are not eligible
for child benefits if they only have four children,
The prevalence of constraints over host utterances is also evident when the latter are left
uncompleted, as in (44), where there is no difficulty whatever to venture the guess that the
bit left out is in the nature of a reservation:
(44) Je veux bien lui envoyer mes excuses, ceci dit . . .

I don’t mind sending him my apologies, this said . . .
The fact that we are dealing with a top-down mode of processing does not entail that the
directionality is always the same as far as instantiation is concerned. Evaluation contexts
are a case in point. DMs, as will be recalled, have no direct access to topoï but have to
reach them through propositional contents. Finding out how the situation of reference has
been categorized is logically prior to the identification of the relevant complex categories.
In connection with (43) we saw one type of interaction between core meaning and
contextual assumptions. A different type is found in strategic cases such as (40) (repeated
below):
(40) A: On devrait s’en aller d’ici, le coin est infesté de crocodiles.

We should get out of here, the place is infested with crocodiles.
B: Comme tu voudras. Toujours est-il que c’est toi qui en as eu l’idée.
P Q
As you wish. I’ll point out, though, that you’re the one who came up with the
idea in the first place.
Here it will be assumed that B’s actual intention is strategic in that he wants A to reach
the conclusion he does not claim to be defending. To arrive at this interpretation, A cannot
rely on what B says alone but would have to access additional contextual assumptions.
Finally, DMs can also impose direct constraints on sources of instantiation, as we saw
earlier with simplement and au moins (see (23) to (26)).
3.3. Towards an extended model
The above model, which is closely based on AT, cannot handle certain issues whose
resolution would greatly further our understanding of DMs, especially where their
procedural nature is concerned. For that reason an extended model is called for. The one I
am working towards results from construing certain aspects of the current model in
coevolutionary terms, as set out earlier.
186 T. Nyan
Such a model would further the study of DMs in several ways:
• by clarifying their interaction with the processing system, in terms of how it fits in
with the existing processing strategy, as determined by the perceptual and memory
subsystems involved;
• by providing a coherent hypothesis about their emergence; as an instance of language
adaptation to processing requirements at the multipropositional level, they would have
to be compatible with other instances of language adaptation at that level not only in
terms of how they all work together to fill those requirements but also in terms of
coding strategies;
• by providing alternatives to the way the core meaning is to be construed, on the basis
of what is known of memory encoding, how various types of representations (modal
vs. amodal) are distributed and linked, how patterns of activation are subject to
revision, and what alternatives are available in respect of categorization.
Without going into too much detail, one issue such a model can be expected to throw
some light on concerns the status of topoï, which is clearly not linguistic; another is the
nature of the apparatus behind their interaction with lower levels.
Topoï, as suggested earlier, bear a strong resemblance to complex categories such as

found in our store of factual knowledge (Damasio, 2000). As pairings of categories of
initial situations and categories of response options, these complex categories arise from
individual and collective experiences and become encoded in memory as a result of a
routinization process. Frequent linkages between a certain type of situation and successful
ways of handling them lead, over time, to the creation of a direct link between the two
categories, thereby giving rise to a complex category. In Damasio’s theory, such
categories function to facilitate decision making. Thus, from past encounters with
unfriendly dogs we derive a better sense of what to do the next time round, one based on a
complex category involving initial situations featuring dogs and proven strategies to deal
with them. Such a category provides us with instant access to appropriate courses of
action as soon as the category membership of the initial situation has been established.
Topoï, as complex categories, would similarly function to bring H to a quick decision,
but, this time, regarding the upcoming text. Lexical items or propositional contents would
invoke topoï in virtue of the fact that they are linguistic representations of either initial or
target situations.
Viewing topoï as complex categories leads to a clarification of the acts that invoke them,
hence my proposal that acts of categorization should be substituted for acts of
argumentation. The aim of an act of categorization is to impose a certain category
judgement of the situation of reference by way of orienting H towards a quick interpretive
decision. How this decision can be speedily reached is via the complex category relative
to which the category judgement is made.
A clear advantage of thinking in terms of acts of categorization is that they allow for a
more detailed hypothesis of how lower and higher levels interact, one which draws on
pre-existing processes. The implication this has for DMs is that their operations can be
seen to have a direct connection with processes involved in decision making. This ties in
with what was said earlier regarding their emergence, namely, that it is a case of language
adaptation to processing requirements. As will be recalled, DMs contribute to processing
efficiency by imposing relational schemas of various sorts, the encoding of which may
have been subserved by mechanisms relating to explicit memory encoding. Another
aspect of this function now emerges: the fact that the constraints arising from DMs
concern the complex categories in presence and the extent to which they apply to the
situation of reference (as in the case of AOs) suggests that DMs intervene at various
known stages of decision making, namely the ranking and selection of complex categories
and the selection of response options within a designated category.
Further issues an extended model of the above kind can also be expected to handle more
successfully include the following:
• The basis for postulating the attentional shift inherent in the switch from the
representational aspect of a propositional content to its categorizing aspect. In the
absence of such a switch (the equivalent of AT’s ‘visée argumentative’) there can be
no act of argumentation or categorization.
• The basis for topical forms.
• The processes whereby DMs came to have the core meanings they do.
The study of DMs, as construed in this paper, is relevant to a number of general issues.
Among those yet to be explored in some depth, the following are, in my view, worthy of
consideration:
• How do linguistic markers in general, and DMs in particular, constrain processing?

• How does the postulation of semantic constraints on context construction and that of
search instructions affect one’s view of contextual intervention?
• What type of generalization do we want to capture? Are we interested in linguistic or
processing regularities?
• What type of rule and mode of categorization are suitable to our purposes?
• In what way would our methodology be affected if processing constraints were
brought to bear?
• Assuming the existence of language adaptations to processing requirements, what
coding strategies do we find at the multipropositional level?
Notes
1
A topos is a shared assumption, which functions as a reasoning principle. Thus the
argumentative sequence Soaring temperatures are expected, let’s head for the beach!
would be justified by reference to a topos, whereby good weather is conducive to
outings.
2
Argumentational seems preferable to argumentative, as argumentativité is not confined
to the relation exemplified by arguments and conclusions, as we shall see further on.
3
Ducrot, personal communication.
4
This is consistent with Edelman’s notion of memory, which is essentially recategorical.
Within his Theory of Neuronal Group Selection, the concepts stored in memory are
188 T. Nyan
susceptible to change under the influence of current categorizations (Edelman, 1989:

98-105, 141-148).
5
Acts in this instance subsumes both the one carried out by S through his or her act of
utterance and previous acts of S and H as featured in schematized situations
6
Response options themselves are linked to a set of consequences. See also Nyan, 2004:
94-96.
Chapter 11
Towards a theory of discourse markers
Bruce Fraser
1. Introduction
I start from the position that there is a class of lexical expressions in every language called
pragmatic markers (Fraser, 1996; see also Hansen, this volume; Foolen, 2001;
Schourup,1999). These expressions occur as part of a discourse segment but are not part
of the propositional content of the message conveyed, and they do not contribute to the
meaning of the proposition per se.1 Members of this class typically have the following
properties: they are free morphemes, are discourse-segment initial, signal a specific
message, and are classified not syntactically but in terms of their semantic/pragmatic
functions.
There are four types of pragmatic markers. The first, basic pragmatic markers, illustrated
by the bolded items in (1), signal the type of message (the illocutionary force) the speaker
intends to convey in the utterance of the segment.
(1) a. I promise that I will be on time.

b. Please, sit down. [a request but not a suggestion or an order]
c. My complaint is that you are always rude.
The second, commentary pragmatic markers, signal a message separate from but in the
nature of a comment on the basic message. The different types are illustrated in (2)
through (5).
(2) Assessment markers

a. We got lost almost immediately. Fortunately, a police officer happened to
pass by.
b. Mary hurried as fast as she could, but sadly, she arrived too late for the movie.
(3) Manner-of-speaking markers

a. A: Mark, you’ve got to do something. B: Frankly Harry, I don’t know what
to do.
b. You got yourself into this mess. Bluntly speaking, how are you going to get
out?
190 B. Fraser
(4) Evidential markers2

a. A: Will he go? B: Certainly, he will go.
b. I have great concerns over this. Conceivably, Tim is right.
(5) Hearsay markers

a. A: Is the game still on? B: Reportedly, the game was postponed because of
rain.
b. I won’t live in Boston. Allegedly, all the politicians are corrupt.
The third type, parallel pragmatic markers, signal a message separate from the basis
message.3 They are illustrated in (6).
(6) Deference markers

a. Sir, you must listen to me.
b. Your honor, can I help you?
(7) Conversational management markers

a. Now, where were we when we were interrupted?
b. Well, we could do it either of two ways.
c. Ok, what do we do now?
The fourth type of Pragmatic Markers are those I am calling discourse markers (DMs),
which signal a relation between the discourse segment which hosts them and the prior
discourse segment.4 These are illustrated in (8):
(8) a. A: I like him. B: So, you think you’ll ask him out then?
b. John can't go. And Mary can’t go either.
c. A: Harry is hurrying. B: But when do you think he will really get here?
d. I think it will fly. Anyway, let’s give it a chance.
e. Sue isn’t here, although she said she would be.
f. Donna left late. However, she arrived on time.
As most readers are aware, there is no general agreement on what to call these items.
They have been frequently referred to as discourse markers (Schiffrin, 1987a; Fraser,
1990, 1996, 1999), discourse connectives (Blakemore, 1987, 1992, 2002), discourse
operators (Redeker, 1991), and cue phrases (Knott, 2000; Knott and Sanders, 1997;
Sanders and Noordman, 2000). Other less frequent terms include discourse particles,
discourse signaling devices, indicating devices, phatic connectives, pragmatic
connectives, pragmatic expressions, pragmatic formatives, pragmatic operators, pragmatic
particles, semantic conjuncts, and sentence connectives.
Moreover, there is no agreement on what the class of DMs consists of. Schiffrin (1987a)
includes Oh!, Look!, Y’know, and certain unspecified nonverbal gestures, while Fraser
(1999) rejects these; Knot and Sanders (1997) include then again and admittedly . . . but,
whereas Schiffrin (1987a), Fraser (1999), Redeker (1991), and Blakemore (2002) do not.5
My purpose in this paper is to present a linguistic account of the class of English DMs. I
shall do so by first presenting a definition of DMs, which is deliberately restricted to those
A theory of discourse markers 191
lexical expressions which function as segment connectives and signal a semantic relation
of one of four types. Following this, I shall present the linguistic properties of DMs as I
see them, drawing on ideas from other research and researchers and using data from
research articles, corpora, and intuitions. I am concerned with the general properties of the
entire class, not the specific description of one or two particular DMs. Hence, the
frequency of occurrence of particular DMs, their privileged status in a dialect, or reader
disagreement with a judgment presented will not affect the general argument. I should
point out that my intent is not to capture how this class of expressions contributes to
discourse coherence (see Schiffrin, 1987a), what role it plays in argumentation
(Anscombre and Ducrot, 1989), or the way that a DM might contribute to the cognitive
interpretation of a discourse segment (Blakemore, 2002).
2. Definition
2.1. A canonical definition of a DM
I will begin with a canonical definition of a DM. Though it is too general to account for
every instance where a DM occurs (I will point out exceptions in the course of the paper),
it will serve as a focal point for what follows.
(9) For a sequence of discourse segments S1 – S2, each of which encodes a complete
message, a lexical expression LE functions as a discourse marker if, when it
occurs in S2-initial position (S1 – LE + S2), LE signals that a semantic
relationship holds between S2 and S1 which is one of:
a. elaboration;
b. contrast;
c. inference; or
d. temporality.
Let us explore this definition. First, the definition restricts a DM to only a lexical
expression, thereby excluding nonverbal gestures (suggested by Schiffrin, 1987a),
syntactic structures, and aspects of prosody such as intonation or stress.
Second, the definition specifies that S1 and S2 are single contiguous discourse segments.
While generally true, there are exceptions, as in (10), where the DM however relates non-
contiguous segments.
(10) A: I don’t want to go very much.

B: John said he would be there.
A: However, I do have an obligation to be there.
In addition, the segments need not consist of a single utterance, as the examples in (11)
illustrate:
(11) a. He drove the truck through the parking lot and onto the street. Then he almost
cut me off. After that, he ran a red light. However, these weren’t his worst
offenses.
192 B. Fraser
b. You want to know the truth? Essentially, John stayed away. Jane came but
didn’t participate. And Harry and Susan fought the entire evening.
In (11a), the however relates its host segment to the previous three segments, while in
(11b), essentially relates the following three segments to the previous one (see Lenk,
1998). The definition can be adapted easily to accommodate these relatively infrequent
cases but I will not do so.
Third, the definition requires that S1 and S2 encode a complete message. This is
illustrated by the examples in (12) and also in (13), although in these latter cases, the S2
has had part of the full discourse segment content elided.
(12) a. Water freezes at 32 degrees but boils at 212 degrees.

b. The movie is over, so we might as well go directly to the party.
c. A: Fred is a real gentleman. B: On the contrary, he’s a boor.
(13) a. A: Jack ate a hamburger. B: I did too.

b. Jane wants to leave, but not me.
c. A: I’m hungry. So (what should I make of that)?
d. She’ll go before John.
Compare these with the examples in (14), where the lexical expression which has the
potential to function as a DM is prevented from doing so because there is not a complete
message in S2 which can be related to S1.
(14) a. He has been home since 5 o’clock.

b. A: Who passed? B: All but one person passed the exam.
c. A: You won’t go. B: I will so.
d. I know how one scar occurred. But how did you get the scar on the other hand?
e. A: How is Sue going to get there. B: She’ll go however [in whatever way]
she can.
Fourth, the definition shows the LE occurring before the second segment, S2. However,
an alternative form of the DMs may be placed in the S1-initial position, often with
adjustments to the segment, as the examples in (15) show.
(15) a. He came back because he loved her.

He loved her. Because of that, he came back.
b. John didn’t take the letter. Instead, he left it.
Instead of taking the letter, John left it.
Finally, the definition specifies that every DM signals one of four types of relationships.
Whether or not there are other semantic relationships that hold between discourse
segments remains to be seen, but these four are intended to be exhaustive. Of course, as
we will see below, within each of the four major relationships DMs signal, there are
relatively more restrictive markers, for example, but vs. however/nevertheless/in contrast.
By using the term signal I mean that the DM marks a relationship between S2 and S1
which the speaker of S2 intends the hearer to recognize. This semantic relationship is
analogous to the additive relation, +, having a core meaning and, in this case, two
arguments, S1 and S2. A DM does not “create” a relationship between two successive
segments, since the relationship must already exist for the S1 – DM + S2 sequence to be
acceptable. For example, the but in (12a), repeated below, signals that a contrast exists
between S2 and S1 and that the hearer is to interpret the sequence while being aware of
this, while the so in (12b) signals that the conclusion conveyed in S2 is justified by the
message conveyed in S1, and the on the contrary in (12c) signals disagreement of the
second speaker with the message of the first.
(12) a. Water freezes at 32 degrees but boils at 212 degrees.

b. The movie is over, so we might as well go directly to the party.
c. A: Fred is a real gentleman. B: On the contrary, he’s a boor.
These relationships, and perhaps others, exist between the sequence of S2 and S1 in (12),
whether or not there is a DM present.
From the definition of a DM given in (9), certain things follow that have occasionally
been of interest. First, since a DM is a type of relationship, with its arguments being S2
and S1, it does not contribute to the semantic meaning of the proposition which hosts it,
S2. Second, because it does not contribute to propositional meaning, it plays no role in the
truth conditions of the S2 segment.
2.2. Non-definitional properties of DMs
Having defined a DM as a lexical expression that signals a relationship which exists

between adjacent discourse segments, let us now examine how DMs pattern on the
various linguistic levels.
2.2.1. Phonological properties

There do not seem to be any strong generalizations about the phonology associated with
DMs. They are not normally unstressed but they may be, especially when the DM is
monosyllabic, for example, but, so, and and, and, where the sequence consists of one
sentence, such as S1 + DM + S2 in (12a) above. When the DM is in initial position, as in
(16), and there is emphasis on the second segment, the DM is often followed by a pause.
(16) a. Child: There was a big puddle. Parent: So - you had to jump right in?
b. You will have to take the chairs. However - don’t touch those chairs over by
the wall.
c. A: John is at home. B: But - I just saw him at the mall.
And when the messages conveyed by S1 and S2 involve other than propositional meaning
(see Section 3), there may be a pause before the DM, as illustrated in (17):
(17) a. John was hungry – so he ate a sandwich.

b. John was hungry – so he must have been really grouchy.
194 B. Fraser
2.2.2. Morphological properties

Here, also, there is little to say. While many DMs are monosyllabic (e.g., but, so, and, and
thus), there are those which are polysyllabic (e.g., furthermore, consequently,
nevertheless, and before) and others which consist of an entire phrase (e.g., as a
consequence, I mean, and that is to say).
2.2.3. Syntactic properties

Although the class of DMs is defined functionally as those lexical expressions which
signal a relationship between adjacent messages, all are members of one of five syntactic
categories: coordinate conjunction, subordinate conjunction, preposition, prepositional
phrase, adverb. The examples in (18) are illustrative though not exhaustive.
(18) a. Coordinate conjunctions: and, but, or, nor, so, yet, . . .6

b. Subordinate conjunctions: after, although, as, as far as, as if, as long as,
assuming that, because, before, but that, directly, except that, given that,
granting that, if, in case, in order that, in that, in the event that, inasmuch as,
insofar that, like, once, provided that, save that, since, such that, though,
unless, until, when(ever), whereas, whereupon, wherever, while . . .
c. Adverbials: anyway, besides, consequently, furthermore, still, however,
then, . . .
d. Prepositions: despite, in spite of, instead of, rather than, . . .
e. Prepositional phrases: above all, after all, as a consequence (of that),as a
conclusion, as a result (of that), because of that, besides that, by the same
token, contrary to that, for example, for that reason, in addition (to that), in
any case/event, in comparison (with that), in contrast (to that), in fact, in
general, in particular, in that case/instance, instead of that, of course, on that
condition, on that basis, on the contrary, on the other hand, on top of it all, in
other words, rather than that, regardless of that, . . .
For the prepositional phrases, there are three variations, shown in (19), and there are
synonymous DMs but of different morphological form which fall into two syntactic
categories, depending on whether it is placed with S1 or S2, as shown in (16), repeated
here.
(19) a. Fixed Form: above all, after all, as a conclusion, . . .

b. PREP + that (where that refers to S1) despite that, in spite of that, in addition
to that, . . .
c. DM + of this/that (where that refers to S1): as a result of that, because of that,
instead of (doing that), rather (than do/that)
(16) a. Child: There was a big puddle. Parent: So - you had to jump right in?
b. You will have to take the chairs. However - don’t touch those chairs over by
the wall.
c. A: John is at home. B: But - I just saw him at the mall.
It is of course the syntactic category of each DM that determines where it may occur in
S2. All DMs, with the possible exception of though, occur in S2-initial position; for
coordinate and subordinate conjunctions, the S2-initial position is the only place they may
occur, due to the syntactic constraints placed on conjunctions. The other three categories
(prepositions, prepositional phrases, and adverbials) have a much greater latitude

syntactically, some occurring in S2-final position, with others occurring in both the final
and medial position. I have found no DMs which can occur in S2-medial position but not
S2-final position.
(20) a. A: You must go today. B: But I (*but) don’t want to go (*but).

b. We started late. However, we (however) arrived on time (however).
c. The trip was tiring. Despite that, he (*despite that) remained cheerful
(despite that).
d. A: The movie is over. B: Then we (*then) should head for home (then).
Leaving aside subordinate conjunctions such as although, since, and because, which must
be retained for syntactic reasons, the presence of a DM is optional in cases such as (21a,
b), while in other cases, such as (21c, d), it must be present for an acceptable sequence to
occur.
(21) a. A: We started late. B: (But) we arrived on time.

b. He didn’t pick up the letter on the table. (Instead/Rather,) he left it lying
there.
c. Fred a gentleman? On the contrary, he is a boor.
d. Harry didn’t arrive on time. In any event, the meeting was late in starting.
However, as I stated above, in no case does the DM create the relationship between S2
and S1. Whatever the relationship, it is present due to the linguistic interpretation of the
segments, taken together with the discourse context, and the DM merely makes clear what
relationship the speaker intends.
Of course, this does not mean that DMs are redundant. Whereas the sequence in (22a)
enjoys all of the DM relationships indicated in (22b-f), it is doubtful that all the
relationships would be recognized, absent the presence of a DM forcing recognition that a
specific relationship is present and intended.
(22) a. This flight takes 5 hours. There’s a stop-over in Paris.

b. This flight takes 5 hours, and there’s a stop-over in Paris.
c. This flight takes 5 hours, because there’s a stop-over in Paris.
d. This flight takes 5 hours. So, there’s a stop-over in Paris.
e. This flight takes 5 hours, but there’s a stop-over in Paris.
f. This flight takes 5 hours. After all, there’s a stop-over in Paris.
Another variable aspect of DMs is the manner in which the DM signals that there is a
relationship between S2 and S1. In (23a), where the DMs are conjunctions, the syntactic
properties of the DMs require that there be two discourse segments. On the other hand, in
(23b), the anaphoric that, which is often elided, indicates that there is a previous segment
which serves as the S1 for the relationship, while in (23c), the relationship between S2
and S1 is implied by the meaning of the DM.
(23) a. Syntactic requirement: and, although, but, or, since, so, while, whereas
b. Anaphoric expression: as a consequence (of that), as a result (of that), as a
result (of this/that), because (of this/that), besides that, contrary to that,
196 B. Fraser
despite that, for that reason, in addition(to that), in comparison (with that), in
spite of that, in that case, instead (of this/that), on that basis, on that
condition, rather (than this/that), regardless (of that)
c. Implied by meaning of the DM: above all, accordingly, after all, all things
considered, also, alternatively, analogously, as a conclusion, besides, by the
same token, consequently, contrariwise, conversely, correspondingly, equally,
further(more), hence, however, in particular, likewise, more accurately, more
importantly, more to the point, moreover, nevertheless, nonetheless, on the
contrary, on the other hand, on top of it all, otherwise, similarly, still, then,
therefore, thus, what is more, yet
The following sequences reflect the possible syntactic arrangements of DMs in sequences,
ignoring the initial/medial/final option discussed above.
(24) a. S1, DM+S2.

Coordinate Conjunction: John left late, but he arrived on time.
Subordinate Conjunction: John was sick because he had eaten spoiled fish.
b. S1. DM+ S2
Coordinate conjunction: John left late. But he arrived on time.
Adverbial: John left late. However, he arrived on time.
Prepositional phrase: John came late. After all, he’s the boss.
Preposition: John left late. Despite that, he arrived on time.
c. DM+S1, S2
Preposition: Despite the fact that John left late, he arrived on time.
3. Model
Although there are over 100 DMs in English, I have found only four basic semantic
relationships reflected in their use, with subclassifications within each of these basis
relations. The resulting classes of DMs (not intended to be exhaustive) is an elaboration of
Fraser (1999), where I have represented what I consider to be the primary DM of each
class in bold, with the others being ordinary members.7
(25) a. Contrastive markers (CDMs): but, alternatively, although, contrariwise,

contrary to expectations, conversely, despite (this/that), even so, however, in
spite of (this/that), in comparison (with this/that), in contrast (to this/that),
instead (of this/that), nevertheless, nonetheless, (this/that point),
notwithstanding, on the other hand, on the contrary, rather (than this/that),
regardless (of this/that), still, though, whereas, yet
b. Elaborative markers (EDMs): and, above all, also, alternatively, analogously,
besides, by the same token, correspondingly, equally, for example, for
instance, further(more), in addition, in other words, in particular, likewise,
more accurately, more importantly, more precisely, more to the point,
moreover, on that basis, on top of it all, or, otherwise, rather, similarly, that is
(to say)
c. Inferential markers (IDMs): so, after all, all things considered, as a
conclusion, as a consequence (of this/that), as a result (of this/that), because
(of this/that), consequently, for this/that reason, hence, it follows that,
accordingly, in this/that/any case, on this/that condition, on these/those

grounds, then, therefore, thus
d. Temporal markers (TDMs): then, after, as soon as, before, eventually, finally,
first, immediately afterwards, meantime, meanwhile, originally, second,
subsequently, when
One property associated with some DMs is the fact that while no DM requires a single
speaker, some require two speakers for the felicitous use. This is illustrated in (26).8
(26) a. A: Fred is a nice guy. *On the contrary, he is a boor/B: On the contrary, he
is a boor.
b. A: I want to go home. *Then, go/B: Then go.
Another property is the fact that with the exception of reformulation markers such as that
is to say, for example, and more precisely (de Saz, 2003), and temporal markers, most
DMs may occur without the presence of the initial S1, just in case the nonlinguistic
context provides a suitable message. The examples in (27) indicate the possibilities:
(27) a. Context: Joel, on seeing his bike being taken by a stranger.

Joel: But that’s my bike!
b. Context: John, on seeing his roommate walk in smiling.
John: So, you aced the exam.
c. Context: Father, after a teenage boy has just left the dinner table in a huff.
Father: And where do you think you’re going, young man?
d. Context: Walking through a garden, Harry holds up a bunch of withered
flowers.
Harry: In comparison [with these flowers], my flowers look spectacular.
e. Context: For the third time, the glue failed to hold the pieces of wood
together.
Max: You could try rubber bands as an alternative [to that glue].
This fact does not bear on the definition of a DM. It simply reflects that in a discourse,
messages may be conveyed by nonlinguistic means.
In terms of an individual meaning of a DM, I take a polysemous approach, similar to that

articulated by Hansen (1998a). Specifically, I take each DM to have a core meaning of a
general nature (for example, for but, the meaning is ‘simple contrast’), with various
meaning nuances triggered as a function of (i) the core meaning of the specific DM, (ii)
the interpretations of S2 and S1, and (iii) the context, linguistic and otherwise.9 Thus, I
take the interpretation of the sequences in (28), to emanate from the same core meaning of
but.
(28) a. Water boils at 212 degrees but mercury boils at a much higher temperature.
b. Mary is thin. But she still weighs more than me.
c. A: John is right here. B: But I just saw him on TV.
d. John died. But he was ill.
e. A: The flowers are beautiful. B: But they’re plastic.
f. A: We had a very nice meal. B: But did you ask him about the money he owes
us?
198 B. Fraser
For example, the interpretation of (28a) is one of direct contrast of S2 and S1, that of
(28b) is one of contrast and rejection of an inference drawn from S1, that of (28c) is of
contrast and challenge of an inference drawn from S1, and so forth. Other contrastive
DMs have a more specific meaning. For example, the meaning of however is that the
message of S2 is contrary to expectations raised by the message of S1, while the meaning
of nevertheless is that the message of S2 is valid, despite the facts conveyed by S1.
Similarly, in (29), each of the segment pairs has an inferential relationship, with the
various nuances being derived.
(29) a. Susan is married. So, I guess she is no longer available.

b. John was tired. So, he left early.
c. The beach is empty, so where do we go from here?
d. Teenage son: The Celtics are playing tonight. Father: So?
e. The train is late again. So don’t wait up for me.
f. A: Wash the dishes right away. B: So give me the soap.
Whether or not this claim of polysemy of DMs meaning can be supported will have to be
assessed through further research. I have made a first attempt at this in Fraser
(forthcoming).10
The position espoused above should not be confused with a polyfunctional analysis of
DMs, or for particles in general. In the polyfunctional approach, the but in the following
examples, (in which but is a DM only in (29a)), are combined into one “class,” and the
task is to explain why this one morpheme (assuming that all instances of but are, in fact,
the same morpheme), can fulfill so many different functions.
(30) a. I like you but I can’t go out with you.

b. Everyone but John was here.
c. I have but a moment.
d. You may think I’m crazy, but where is the dog?
e. He has all but clinched the championship.
f. I can’t help but obey her.
g. I will get you but good.
I see these two efforts to be separate though not in conflict. In any event, I am not
considering the latter approach.
There is, however, a different polyfunctional analysis of certain English DMs, which was
proposed, among others, by van Dijk (1979), who suggested that there were two types of
connectives. He wrote:
We assume that each connective has a certain (minimal) meaning which may be
further specified depending on its semantic or pragmatic use. At the same time,
semantic conditions may underlie conditions of pragmatic appropriateness. Thus,
denoted facts may be normal conditions for the possible execution of subsequent
speech acts . . . we assume that each connective has a certain (minimal) meaning
which may be further specified depending on its semantic or pragmatic use (p.
449).
More recently, others (e.g., Knott and Sanders, 1998) consider DMs to operate in these
same two domains.
Sweetser (1990) elaborated on this notion by proposing that many DMs which are
syntactically conjunctions or adverbials are pragmatically ambiguous, and that there are,
in fact, three pragmatic domains to which DMs apply, as shown in (31).
(31) a. Propositional domain: John is very hungry, so he is eating a sandwich.

[The facts of S1 are the cause of the facts in S2]
b. Epistemic domain: John is very hungry, so he must be very grouchy.
The knowledge of S1 justifies the conclusion stated in S2]
c. Speech act domain: John is very hungry, so go get him some food, please.
[The report of S1 justifies the request stated in S2]
Here the assumption is that some DMs have a single semantic core meaning but may
function on more than one pragmatic domain. Sweetser writes that the “correct”
interpretation of the sequence depends not on form, but “on a pragmatically motivated
choice between viewing the conjoined clauses as representing content units, logical
entities, or speech acts” (p. 78). This proposal deserves a closer scrutiny than it has yet
received. Among potential problems is the fact that in Sweetser’s view, DMs operating in
the propositional domain signal a relationship between two propositions, not messages
(illocutionary acts), which runs counter to the view that all utterances convey one or more
illocutionary acts.
In terms of ambiguity, DMs fall into three groups. The first group consists of those DMs
which have the same meaning as both a DM and a homophonous form functioning as a
content lexical item. These are illustrated in (32). DMs which have a homophonous form
but with a different semantic meaning are shown in (33), while lexical expression which
occur only as DMs are shown in (34).
(32) The DM meaning is the same as the homophonous form (e.g., as a result,
similarly, as a conclusion, in addition)
a. He didn’t brush his teeth. As a result, he got cavities.
b. The substance hardened. This wasn’t the outcome we wanted as a result.
(33) The DM meaning is different from the homophonous form (e.g., then, however,
but, while, on the other hand, so, since)
a. We started late. However, we arrived on time.
b. We started late. Therefore, John will have to get here however he can.
(34) The form functions only as a DM (nevertheless, on the contrary, moreover,

conversely, consequently, whereas, in comparison, although)
Interestingly, I have found no cases of a DM and its homophonous form occurring in the
same linguistic environment. In the examples in (35), it is very clear that the first
200 B. Fraser
however/after all is functioning as a DM while the second is functioning as an adverbial,

and these roles may not be changed.11
(35) a. I expect him to come. However, he will have to get there however he can.
b. Give it to me. After all, it belongs to me after all.
The specific meaning of some DMs is relatively opaque, with the primary members of the
subclasses (and, but, so, then) by far the least transparent. For example, but signals only
that S2 is in “simple contrast” with some aspect of S1 (e.g., the explicit message
conveyed, an implied message, a presupposed proposition, a felicity condition of S1, etc.),
with nothing further specified; the work of interpretation of this fine-grained reading must
be done primarily by the recipient relying on the interpretation of S2 and S1 and the
contextual information. On the other hand, the meaning of in comparison, for example, is
relatively transparent, and the relationship signaled can be read off the meanings of the
words. This range of semantic opaqueness is hardly surprising.
Relevance theorists, notably Blakemore (1987, 1992, 2002), argue that there are two
mutually exclusive types of meaning of linguistic forms (lexical expressions and syntactic
structures): procedural meaning, which specifies instructions of how to manipulate
conceptual interpretations, and conceptual meaning, which specifies substantive concepts.
In their view no linguistic expression can have both types of meaning. Since DMs must
encode procedures for relating S2 and S1, they cannot encode conceptual meaning as
well. And, since many expressions which are considered DMs unarguably encode
conceptual meaning (in addition, as a consequence, as a result, contrary to expectations),
they cannot be DMs. Her conclusion: DMs are not a coherent class worthy of study.
I take issue with this position in Fraser (forthcoming b). For the moment, I will make only
two comments. First, given the way relevance theory views language, this may be a valid
conclusion within that theory. However, since most of the world does not embrace the
relevance theory view, it is premature to write off DMs as unworthy of study. Second,
while it is plausible (though not by any means provable) that but or so do not have a core
meaning which might be thought of as “conceptual meaning,” it is strange that however
and contrary to expectations, for example, both DMs according to the definition given
here in (9), function exactly alike, occur in the same linguistic environments, and indeed,
have the same meaning “contrary to expectations.” Yet according to relevance theorists,
only the former can be a DM, for, to include the other would destroy the mutual
exclusivity of their two types of meaning, a hallmark within the theory. There are other
similar cases. In fact, as I argue in Fraser (forthcoming c), the procedural/conceptual
distinction as an either/or option is simply incorrect, with all lexical expressions having
both a conceptual and procedural meaning.
Sequencing of DMs is a relatively unexplored semantic area, yet it occurs quite often, as a
brief scrutiny of the British National Corpus will convince you. A sequence of DMs
typically occurs when two DMs occur as a part of S2, as in (36), where the first DM in the
sequence is one of the primary DMs (and, but, so, then), and the second DM (not
necessarily following directly as in (30e)) is one of the other members of the subclass
(e.g., for the contrast class: but vs. however/nevertheless/on the contrary, instead, rather,
in comparison, despite that, . . .).
(36) a. John went swimming, but, in contrast, Mary went sailing.

b. John went swimming and, in addition, he rode his bicycle.
c. John went swimming, so as a result, he won’t be home for dinner.
d. John went swimming; then, afterwards, he went sailing.
e. We started late. But, we arrived on time nevertheless.
Subordinate conjunctions may not be a second DM in a sequence, and two ordinary

members of the same subclass (e.g., the subclass of contrastive discourse markers)
typically do not occur in a sequence, although some combinations, when said a few times,
don’t seem too bad.
(37) a. John started late. ?Still/?However, he arrived on time, ?nevertheless.

b. We ordered ham. ?Moreover, we asked for it with lemon sauce,
?furthermore.
A second case is where a primary DM of one subclass occurs as the first in a sequence
with an ordinary member of another subclass (e.g., but [CDM] + as a result [IDM]). The
first DM (always a coordinate conjunction) signals the major relationship between S2 and
S1 (contrast, elaboration, inference, temporality), while the second DM signals a more
specific relation not within the first relation’s domain.12
(38) a. He walked to town but, as a result, he caught a cold.

b. He was sick and thus he was unable to work.
c. He was home, and yet he hadn’t spoken to his wife.
d. She criticized him. Then, for that reason, he left her.
Note that when there are different subclasses present in the DM sequence, each class
condition must be satisfied.
(39) a. He walked to town. But, as a result, he caught a cold.

b. He walked to town. But, *as a result, he didn’t want to visit with Mary.
c. He walked to town. *But, as a result, he was the winner.
From an initial examination, I have found the following general possibilities of sequences:
Table 1. Possible DM sequences
Primary DM Reg.-CDM Reg.-EDM Reg.-IDM Reg.-TDM

CDM (but) All* All All All
EDM (and) Some All Nearly All All
IDM (so) None Few All All
TDM (then) Some Some Some Some
For example, all regular DMs from every subclass will combine with but (except
however), but no regular CDM will combine with so, while some will combine with
and.13
202 B. Fraser
4. Some final thoughts
In the foregoing I have sketched out a view of DMs based on data from English. Of
course it is possible that data from other languages will cause revision of parts of the
theory presented here or may cause it to be rejected as a general theory entirely. I certainly
hope not.
One area which requires more investigation is the extent to which there is polysemy
throughout the class of DMs as opposed to only in a few, top-level markers such as but,
so, and and. Another area is the extent to which all DMs operate in three domains. While
most DMs appear to, there are some such as for example, as a result, in contrast, that is to
say, and moreover for which I cannot find sequences in which they function in the
epistemic and/or speech act domain.
A third area worth looking at is the extent to which at least the primary DMs (and, but, so,
then) have the same nuanced interpretations across languages. I have conducted a
preliminary investigation into this matter with but using sequences which favored the
different interpretations of but as a DM in English, such as the examples in (40).
(40) a. John is fat, ____ Mary is thin.

b. John is not tall ____ short.
c. Take an orange, ____ leave the apples alone.
d. John is a cop, ____ he’s also a carpenter.
e. I could give you this book, ____ frankly I don’t want to.
f. I’m a nurse. ____ my husband won’t let me work.
g. A: It’s warm in here. B: ____ turn up the heat anyway.
h. A: What time is it? B: ____ why do you want to know?
i. Now you know all the facts. B: ____ I’m still not convinced he is guilty.
j. The flower was beautiful, ____ it was plastic.
k. A: All the boys left. B: ____ there were only two boys to start with.
l. A: I realize that John is sick. B: ____ John isn’t sick.
m. John died yesterday, ____ he had been ill.
n. A: John is home. B: ____ I just saw him at the store.
I sent the questionnaire to one or more speakers of Arabic, Catalan, Danish, Estonian,
Finnish, French, German, Greek, Korean, Sinhala, Spanish, and Vietnamese. Quite
surprisingly, the respondents found that the but equivalent in their language (e.g., pero in
Spanish, aber in German, mais in French) could occur in nearly all (for some languages, it
was all) of the sequences. The one systematic difference occurred with (40b), where the
but was not permitted but another form was required, for example, sino in Spanish and
sondern in German.
Quite understandably, when given sequences such as those in (41), where the but in all
cases is not a DM and in some cases is idiomatic (e.g., (41i)), there was no general
agreement.
(41) a. Everyone ____ John was here.

b. I have ____ a moment.
c. You may think I’m crazy, ____ where is the dog?
d. I’m not sure if this is relevant, ____ isn’t that bag leaking?
e. A: Is it finished? B: ____ of course it’s done.
f. He has all ____ clinched the championship.
g. I can’t help ____ obey her.
h. Thanks are due to John and Mary. ____ above all, I want to thank Harry.
i. I will get you ____ good.
These by no means exhaust the areas that should be examined before concluding that we
are well on the way to developing the theory of discourse markers. I only hope that what I
have provided is a good start.
Notes
1
I am using discourse segment rather than utterance in an effort to avoid controversy
associated with the latter term.
2
Ifantidou-Trouki (1993) claims that evidential and hearsay markers contribute to part of
the propositional meaning. While I have serious doubts about this, if it is true, then
these are not included in the class of commentary pragmatic markers.
3
Prosodic features, attendant to the production of the segment, may contribute additional
parallel messages. But they will not concern us here.
4
There are exceptions which I will address below.
5
Exclamation particles, such as Wow! Gosh!, Damn!, and Yippee!, are not part of a host
utterance and are pragmatic idioms, not pragmatic markers. Also, modal particles such
as indeed in English and doch in German, and focus particles, such as just, even and
only in English, contribute to propositional meaning and are thus excluded from the
ranks of pragmatic markers.
6
Where there is an anaphoric form (this, that, these, those) possible, that has been chosen
arbitrarily.
7
There are relatively few DMs which fall into more than one semantic relationship,
rather and then being two that do so. Blakemore (1992: 138-141) posits an analogous
four way division in which information conveyed by an utterance can affect relevance.
8
In dialogic discourse, when a subordinate conjunction is used by speaker B to initiate his
contribution, the contribution of speaker A is assumed to have been elided:
A: Jim: Leave! [said to a third person, John.]
B: Mark: (Leave!,) Because we hate you.
In DM uses of then, which always requires two speakers, an if-clause is assumed
elided from the second speaker.
A: The movie is over.
B: (If the movie is over,) then there is no point in going over there.
9
The notion of prototype may be useful here as the sense of core meaning is investigated.
Whether the mean relationships for a given DM are expressed through a meaning
chain or a network of interconnected nodes is left open here. See Aijmer et al. (this
volume) for a discussion of types of core meaning.
10
The polysemy of DMs varies greatly, with the primary DMs in each subclass having
considerably more than most of the ordinary members of the subclass.
204 B. Fraser
11
However, Hansen (this volume) suggests that at least for French, there are ambiguous
sequences, for example, “Jean a tiré. Alors, Pierre s’est écroulé” (‘Jean fired. Then/So
Pierre fell down’).
12
Of the four paradigm DMs, only and + so and and + then occur among the combination
possibilities.
13
An explanation for these facts is presented in Fraser (forthcoming b).
Chapter 12
What are particles good for?
Harald Weydt
1. Introduction
Since the sixties, particles have aroused the interest of linguists far and wide. The various
classifications into parts of speech or word classes that particles have undergone in the
recent years will not be discussed here. Most of what will be discussed will be illustrated
with German Abtönungspartikeln (~ modal particles?), and, for reasons of economy, with
German aber, but it should be clear from the beginning that it will hold true for particles
of other classes as well and not only for German but also for other languages.
Instead of discussing particle classifications and the respective criteria,1 I would like to
raise the basic question of what Abtönungspartikeln are good for, or, why one uses
Abtönungspartikeln. This question has not really been considered in this direct way and
that is why I think it is particularly suitable for a new look at an old subject.
2. Problems of definition
When speaking of particles in a general way, we have good reasons to avoid too
restrictive a definition. Instead, we will enumerate some denominations in German,
French, and English. They roughly circumscribe the range of the words that this article
will be about, and they will already reveal to a certain degree what the authors think they
are good for.2 The particles that I am going to treat are the following:
1. In German: Abtönungspartikeln, Modalpartikeln, Satzpartikeln, Gliederungssignale,

illokutive Partikeln, diskursorientierende Partikeln, Einstellungspartikeln, diskursive
Partikeln, Gesprächswörter.
2. In French: modalisateurs, particules expressives, appréciatifs, argumentatifs,
particules illocutoires, adverbes de phrase, connecteurs, mots de la communication,
charnières du discours (de discours), particules énonciatives, marqueurs, adverbes
explétifs.
3. In English: pragmatic markers, discourse particles (as in this volume), discourse
markers, interpersonal markers, argumentative markers, presentative particles,
parentheticality markers, modal particles, adverbial connectives, connectives, modal
206 H. Weydt
discourse particles, elusive particles, particles of truth, contrastive and set-evoking

particles, sentence-structure particles, down-toners.
Let me provide some general remarks concerning the definition of particles:
In principle, I consider the category particle to be a crosslinguistic one. Contrary to

widespread ideas (as in Sasse, 1993: 682, “These subcategories are too language-specific
to justify crosslinguistic treatment”), particle is a category and has as such universal
validity, just like other linguistic terms such as verb and noun. When defining noun, we
do not define the French noun nor the Japanese one, but we have an idea of noun and, in a
second step, we look at a certain language, observing whether there is a noun and if so,
what qualities it has (does it inflect, does it mark the plural, does it bear articles, etc.) The
same holds for particles.3 By particle I understand a word class. That means that only
single words, not clauses, are considered to be particles. Word groups like English I mean,
you know, French après tout, au fond, au total, c’est-à-dire, tout compte fait, je dirai que,
Spanish por este motivo, and Portuguese de maneira que are not particles, in spite of the
undisputed fact that they can occupy the place of a particle, replace it, and be replaced by
it. An analogy may help to point out what I mean and justify my claim. If linguists speak
of nouns, they only mean nouns and not elements which can stand in the place of a noun.
So, that you come is not a noun, even though it can stand for one. In the sentence I know
the answer, the answer can be replaced by that you come or by it. Neither that you come
nor it, however, is a noun. A constitutive feature for the definition of particles is that they
do have (synsemantic) meaning. Nonetheless they do not refer to sections of the
extralinguistic reality (they have no lexical meaning), nor do they position anything
relative to the ego, the speaking person (they have no deictic meaning), and they do not
have word class meaning (as pronouns do, which are the empty forms of nouns,
adjectives, or adverbs). Interjections, having no synsemantic meanings, are not particles.
To sum up, particles are (single) words, which have no dissecting (lexical), deictic, or
word class meaning, but they do have semantic content which they deploy in connection
with other elements of the utterance (for further details see Hentschel and Weydt, 1989:
6).
3. Functions: why we use Abtönungspartikeln
I recall the results of an experiment presented in the preface of the Kleine deutsche
Partikellehre (Weydt et al., 1983: chapter 0, 11–12). Two dialogues, held between two
young people, were presented to our informants. The first, dialogue A, contained a
relatively large number of Abtönungspartikeln. The second one, dialogue B, was identical
to A except that all particles had been removed from it. It is still grammatically correct. A
teacher could not find any grammatical mistakes.
A section of this dialogue is given in Table 1. Both dialogues were presented to

informants who were asked to read them and judge them relative to a given matrix which
contained the features natural, rejecting, warm, wooden, smooth, authentic, difficult to
make contact with, friendly. The results of the analysis are shown in Fig. 1.
What are particles good for? 207
Table 1. Dialogues
Dialogue A Dialogue B
[. . .] [. . .]
X: Ja, das gibt’s doch gar nicht! Was X: Ja, das gibt’s gar nicht! Was machst
machst Du denn hier! Ich denk’ Du bist in du hier? Ich denk’ Du bist in England!
England! Y: War ich auch, aber jetzt wohn’ ich in
Y: War ich auch, aber jetzt wohn’ ich in Berlin. Bin gerade auf dem Rückweg.
Berlin. Bin gerade auf dem Rückweg. X: Ist toll, ich fahr nämlich auch nach
X: Ist ja toll, ich fahr’ nämlich auch nach Berlin, aber nur übers Wochenende.
Berlin, aber nur übers Wochenende. Y: Gut, dann können wir während der
Y: Gut, dann können wir ja während der Fahrt ein bißchen über die alten Zeiten
Fahrt ein bißchen über die alten Zeiten quatschen.
quatschen. X: Ja, aber sag’, wo fährt der 9.30 Uhr-
X: Ja eben, aber sag’ mal, wo fährt denn Zug ab?
der 9.30 Uhr-Zug eigentlich ab? [. . .]
[. . .]
Figure 1: Comparison of dialogues: average results of 82 German native speakers4

Dialog A Dialog B
0 1 2 3 4 5 6 0 1 2 3 4 5 6
natürlich (natural) 5,7 2,8
abweisend (rejecting) 1,7 3,3
warm (warm) 4,4 2,7
hölzern (wooden) 1,4 5,0
flüssig (smooth) 6,0 2,7
echt (authentic) 5,7 2,7
kontaktschwach 3 2,4 4,0
freundlich (friendly) 5,7 3,7
We have repeated the experiment often, with native German speakers as well as with non-
native speakers of German, and every time the results were very similar. The differences
in values assigned to the dialogues A and B must be explained by the difference in the
presence or absence of particles.
How should this be interpreted? One can see that the matrix contains two different
qualities. The first is a complex value, which may be labeled authenticity. It answers the
question Do Germans really speak like this? The respective values are natürlich ‘natural’,
flüssig ‘smooth’, echt ‘authentic’. The answer for dialogue A is yes; in such a situation,
they do speak with and not without particles. Dialogue A is much more authentic than
dialogue B. The second value is a social one and shows up in the other features. It can be
labeled friendliness. Compared to B, dialogue A is conceived of as friendly and warm,
neither rejecting nor stiff, nor unsociable.
208 H. Weydt
A conclusion that could be drawn and a first answer to the title question could then be that
the speakers use particles in their speech when and because they want to be friendly, and
if they don’t use them, their particleless speech is strange. It will be the task of this article
to ask whether this is an acceptable statement.
However, this conclusion leads to a number of difficulties. The first is the phenomenon of
oversummativity. The overall impact of particles does not coincide with the meaning of
only one of them. Neither eigentlich, nor doch, nor denn, nor any of the other particles are
friendly in and of themselves. If one does not want to give up this first conclusion, then
one would have to try seriously to explain how particles, which are not friendly (the
meaning of which is neutral in this respect), bring about friendliness. The second problem
is that while particles may have friendly effects, they frequently do not. They can appear
in utterances meant to hurt the partner. Was hast Du denn jetzt schon wieder gemacht?!
“What did you do this time?!” and Haben Sie überhaupt einen Führerschein? “Do you
even have a driver’s license?” can be very aggressive.
4. The semantics of modal particles
In order to find out how particles act in speech and to answer the title question of what
Abtönungspartikeln are good for, it is useful to observe how the meanings of the particles
function and how they act in conversation.
4.1. Seven general theses on particle meaning
1. Every particle has a meaning. It is present in every occurrence.

This thesis opposes the idea that particles are meaningless, at least in certain contexts
(mots de remplissage, vides/explétifs, palabras vacías), or as a weaker claim, semantically
reduced (bleached), just fulfilling expletive functions. In that view, particles have lost
their semantic content and are only used in their context for euphonic reasons. If this
statement is not just based on a terminological difference (that meaning is only used for
‘lexical meaning’), then we must disagree with it. We know of no case where a particle is
used without its meaning, and we don’t think that one can be found.
2. It is not the function of particles to down-tone the utterance in oral discourse.

This claim corrects a frequent idea about particles, above all on German
Abtönungspartikeln. For example, James (1983) calls eigentlich “one of the voluntary
markers of imprecision”. Similarly, Grevisse (1993: §920) describes the function of
particles like bien, donc, un peu, and voir as explétifs: “Ils servent seulement à renforcer
ou à atténuer l’expression. Certains participent aussi à la fonction phatique”. It is very
commonly said that particles fulfill the function of down toners and that they are used in a
general way and without having a specific meaning in order to make the utterance in
which they occur imprecise and vague.5 This in turn is said to be one way a speaker can
take the sharpness from utterances, in order to prevent a so-called FTA (face threatening
act) or at least to make it less threatening. In reality, however, eigentlich and the other
Abtönungspartikeln do not express weakening (or extenuation), but they have a precise
meaning which is not extenuating. Eigentlich indicates that the utterance is true in a
deeper sense. This holds for every single occurrence. This meaning does not in itself
contain any extenuating, down-toning element. And no particle is in and of itself friendly
(see section 4.2). As we have seen in the earlier discussion, it is true that
Abtönungspartikeln (and comparable elements in other languages) can make dialogues
friendly, social, and natural, but not via down toning. Particles show that the actual
speaker takes into account his partner’s perspective on the subject, that he cooperates.
This is why his speech is conceived as amiable. This effect can be achieved by various
means and in any linguistic society. For more details, see Hentschel and Weydt (1983),
Weydt et al. (1983), and Weydt (2003).
3. Particles, even if used as Gliederungssignale (‘discourse markers’?), conserve their

primary meaning.
The term Gliederungssignal was created by Gülich (1970). It referred to particles (such as
French et, alors, mais, puis, enfin) in the wider sense, particles which take on specific
functions in spoken dialogues or in spoken speech. In Gülich’s conception,
Gliederungssignale form a homogeoneous word class in the grammar of spoken language,
as opposed to traditional grammar, which, she writes, represents above all the system of
the written language. In the grammar of the written, but not the spoken, language, the
particles belong to different word classes. Among the Gliederungssignale one finds
syntagms consisting of more than one traditional word, like vous savez, n’est-ce pas, tu
comprends, et bien, you know, I mean, which we have already excluded from our
definition of particles and one-word Gliederungssignale, as, for instance et, alors, mais,
puis, quoi, hein, enfin.
A widespread opinion says that it is their task to structure the dialogue (Gülich, 1970:
270) and to help the hearer interpret the other’s turn. When these words are
Gliederungssignale, they lose their original meaning. Following Gülich´s proposal, mais,
when used as G l i e d e r u n g s s i g n a l (or, more exactly, as an opening signal,
Eröffnungssignal), no longer expresses any contrast, and in the same way, puis and alors
do not contain temporal relationships (Gülich, 1970: 297). We disagree with this theory.
In deference to the space it would require to argue against it in detail, let it suffice to say
briefly that we could not find any example where a particle is a Gliederungssignal and
did not preserve its original meaning at the same time. Instead, a particle can fulfill the
function of a Gliederungssignal because of the fact that it maintains its original meaning.
It is able to structure the dialogue because it means something. Let’s take again the
example of the French conjunction mais. According to Gülich (1970: 77), French mais
(and the same holds for English but, German aber, and probably for Dutch maar) used as
Gliederungssignal carries no longer any adversative meaning: “. . . hat es seine
ursprüngliche lexikalische Bedeutung in vielen Fällen aufgegeben zugunsten seiner
Eröffnungsfunktion”. If its only function were to open up a turn, it could be replaced,
without changing the context conditions, by any other opening signal. This is, however,
not the case. Mais can only occur in cases which are compatible with its original, constant
meaning. It demands a context which fits its meaning, and that explains why we can
predict the context conditions for every occurrence.
4. Every particle can be assigned a constant basic meaning, which appears in every
occurrence of that particle. This meaning may be conceived of as a set of semantic
features.
In the process of discovering the meaning of the particle, the semanticist first establishes a
hypothesis about the meaning of the particle and then tries to corroborate it, exposing it to
as many occurrences of the particle as possible. According to thesis 4, no particle can
210 H. Weydt
occur in a usage which is incompatible with its meaning. Otherwise, the semantic
description is falsified. As soon as a usage appears which is not covered by the
hypothesis, this hypothesis must be given up or modified in such a way that a new
hypothesis covers every known occurrence, including the noncomplying case. The
opposite is not necessary: one does not demand that this particle be really used if its use is
possible and in agreement with its meaning. From John is tall, but Jim is small, but can
easily be omitted. Consider again French mais. The basic meaning of mais, as well as of
Dutch maar, English but, or German aber, is this pattern: “It would be wrong to continue
the preceding thought in the expected direction. One has to change the direction of the
thought”. The idea may be illustrated by Fig. 2.
Figure 2. The semantic structure of mais (aber, but)
a a aa ‘mais’
(aber, but) b‘
This diagram may be condensed into the formula “against the expectation”. This explains
the usage of mais as an adversative, coordinating conjunction: French il est grand, mais
faible; English he is big, but weak; German er ist groß, aber schwach; Dutch hej is groot
maar dapper. There are many more examples in Foolen (1993). The speaker assumes that
the hearer concludes that in general, the one who is big is also strong. By using mais, he
warns the hearer not to think in the anticipated direction. Thesis 4 admits restrictions of
usage. In some languages, the respective particle may appear in the imperative: in a scene
where a victim of an accident lies on the ground covered with blood, one of the bystanders
may say in French, Mais occupez-vous de cet homme!, which is not possible in German
*Aber helfen Sie dem Mann. The underlying idea which justifies the usage of mais is
“You don’t seem inclined to help. However, you should help”. On the other hand, there
are restrictions on but and mais in their respective languages which do not exist in
German. In German, one can combine aber with oder (‘or’), which is not possible in
English or French: German Sie brauchen jetzt neue Reifen, oder aber sie geben die Reise
auf, versus French *Il vous faut de nouveaux pneus, ou mais vous abandonnez le voyage
and English *You need new tires, or but you give up going on your trip. The basic
meaning also explains the usage of aber as Abtönungspartikel in German and of maar in
Dutch, and both as Gliederungspartikel: but structures while signaling a contrast to the
expectation.
When checking to see if all the empirically appearing occurrences of a particle are
compatible with the hypothesis about its meaning, one has to consider carefully the
reasons for the apparent semantic deviance. Sometimes one can find reasons which seem
to present counterevidence to the assumed meaning. One of these is irony. Aber used as
Abtönungspartikel in an affirmative clause expresses not only surprise but a surprise that
the content of the sentence is given excessively. The speaker it astonished not about the
fact, the that, but about its extent, the how much. An apparent counterexample could be
seen in the following case: someone sees a small man and says Guck mal, der ist aber
groß! “Look, that man is aber tall!” The use of aber can be explained by irony, where the
designated reality is the contrary of what is said.
5. The basic meaning can be diversified according to its context. More sophisticated
meanings remain compatible with the basic meaning.
In actual use, one finds more semantic rules than are contained in the common overall
meaning of the respective particle. Take again the example of German aber. The overall
meaning is, as pointed out, ‘against expectation’. A closer look, and the opposition with
vielleicht, reveal a more subtle semantic structure. It is compatible with but richer than the
overall meaning:
• The astonishment concerns the how much of the surprising fact, not the that of the fact.
• aber is used in order to express surprise only if the hearer knows the fact. It
accompanies a comment about a known fact or shared knowledge. It does not tell the
fact itself but presupposes that it is known to both of the interlocutors.
An example will help to make this subtle detail clear. Hans tells his friend about his last
holidays in the mountains and mentions the fantastic view from the terrace of his hotel.
He may then say Da hatten wir vielleicht eine schöne Aussicht! “There we had vielleicht a
beautiful view!” but he cannot say *Da hatten wir aber eine schöne Aussicht! “There we
had aber a beautiful view!” because of the fact that his partner does not see the view
himself. If, however, he shows a photo of the view, the other can say Da hattet ihr aber
eine schöne Aussicht! “There you had aber a beautiful view!” due to the fact that they
both see it at that moment.
The description, therefore, must be given on two levels. The overall meaning, which holds
for all occurrences, is relatively abstract. It explains the coherence of all uses and
guarantees its identity. The specific meaning is richer and deploys more subtle semantic
rules which control differentiated usages. The specific rules often appear in the lexicon as
numbered variants.6
6. It is necessary to distinguish these basic meanings of participles from their pragmatic

“meanings”.
Basic meanings of particles and pragmatic effects that can be brought about by using
particles have to be distinguished from each other. One example is the friendliness that
may accompany the use of particles. Another effect is a compliment which is expressed
by particles. Du kannst aber kochen! “(How) you can aber cook!” expresses the surprise
about how (good) the host can cook and can express a compliment. Du kannst ja kochen!
“You can ja cook!” would indicate that the guest had had a very low expectation, which is
hardly a basis for a compliment.
7. Interchangeable particles may (and normally do) differ in meaning.

It has often been observed that in a given context, particles are interchangeable. For
example, in a normal situation, one may either say: Wie heißt du denn? or Wie heißt du
eigentlich? or Wie heißt du? “What is your name?” This is compatible with the fact that
these particles differ in meaning, and it is a general phenomenon in semantics. So, in a
given context, one may refer to the same person saying: the man over there, the guy over
there, the gentleman over there, or the traveler over there. Obviously, person, man, guy,
212 H. Weydt
and traveler differ in meaning. These words may, however, designate the same thing in
certain contexts, because the designated things belongs to different classes.
Interchangeability of particles has the same explanation. If the situation carries features of
eigentlich and of denn, the speaker may freely choose either one, a combination (denn
eigentlich), or none.
4.2. The content of particle meanings
4.2.1. How to present the meanings?

Before discussing in more detail a few particle meanings, I shall allow myself some
general comments on the presentability of particle semantics. I support the idea that so-
called different meanings can be reduced to a single semantic nucleus. This monosemantic
approach is discussed in various places, from Weydt (1969) on, and it has been well
shown, for example, in Foolen (1993). An additional problem, however, is how to express
this meaning, how to describe it in an understandable form. When looking for a way to
present the semantic definition, it would be ideal to find a form which is easily
comprehensible, ideally in such a way that it can be understood by lay people, such as
teachers and students of German as a second language.
Widely used ways of presentation, so far, are either notations in symbolic logic or
paraphrases. Logical notation has the disadvantage of causing additional work. First, one
has to discover the semantics and then, in a second step, one has to translate it into the
respective logical language. The paraphrasing of particle meaning is a comment on the
level of metalanguage. But as such, it does not preserve the original impact. Being
linguistic descriptions, the sentences which express the particle meaning do not serve the
purpose of intersubjective communication. The meaning of the German particle denn can
be correctly paraphrased by “I ask this because the situation makes me think that you
know the answer”. In normal contexts a wh-question with denn sounds more amiable than
without a particle or with, for example, the particle eigentlich. Wie heißt du denn? is a bit
more friendly than Wie heißt du eigentlich?, and much more friendly than Wie heißt du?
This impact can be explained on the basis of its meaning. The semantic paraphrase,
however, does not have the same effect. Ich frage Dich, wie Du heißt, weil etwas in der
Situation mich motiviert, die Frage zu stellen “I’m asking you your name because
something within the situation motivates me to ask this question” is just awkward and not
amiable at all. The reason is that semantic definitions serve other functions than the object
language expressions they paraphrase. While Can you pass me the salt is polite, the
paraphrase is not polite, but, rather, very strange: Is the adressed person (paraphrase of
you) able (lexical meaning of can) in the moment in which the speaker speaks (paraphrase
of present tense) to move from there to here (attempt to paraphrase pass) the salt to the
person which speaks (paraphrase of me)?
4.2.2. Some examples

The following examples attempt to demonstrate the overall meaning and show that they
do not contain friendly elements within themselves. They may help to better explain the
monosemantic approach.
Denn ‘for, then’ appears as a coordinative causal conjunction (e.g., Er aß ein Wurstbrot,
denn er war sehr hungrig ‘He ate a sandwich for he was very hungry’) and as a stressed
or unstressed Abtönungspartikel in wh-questions: Wie hEIßt Du denn? (unstressed denn)
“What is your name?” and Wie heißt Du dEnn? (stressed denn), the latter being used if the
name previously given was wrong. It also appears in yes/no-questions: Warst Du denn
noch nie wirklich verliebt? ‘Were you really never in love?’ (expressing surprise). All of
these usages of denn may be reduced to one single semantic nucleus, illustrated by Fig. 3.
Figure 3. Illustration of the meaning structure of denn
Denn indicates that the content of the sentence in which it appears points back to
something that can be found in the preceding context (for a more detailed explanation see
Weydt and Hentschel, 1983, and Hentschel and Weydt, 1983).
Auch ’also, too’ appears in sentences such as Günter will nächste Woche auch in Ferien
fahren ‘Günter also wants to go on vacation next week’, and as an Abtönungspartikel in
rhetorical questions: Warum auch? may be translated by ‘Why should I?’
The function of auch is to join two ideas, a and b, in such a way that they are no longer
seen as isolated but are subsumed under a common denominator (c.d. in Fig. 4). One of
the two elements may be implicit (symbolized by parentheses) and must be reconstructed
from the context. Fig. 4 may help to make the function clear.
Figure 4. Illustration of the semantic structure of auch.
c.d.
+
a b
Eh, ohnehin, and sowieso ‘anyway’ are cognitive synonyms and differ only in speech
register, for example, A: Wir können nächste Woche nicht an den Strand fahren. B: Macht
nichts, ich muss sowieso/ohnehin/eh fürs Examen arbeiten ‘A: We cannot go to the beach
next week. B: That does not matter, I have sowieso/ohnehin/eh (’anyway’) to work for my
exam’. Fig. 5 may explain the semantic structure. There are two potential reasons, a and
a’, for b (not going to the beach). Reason a’ is not the decisive one, because a alone
causes b (“that you want to cancel the trip is not the important point, but the fact that I
have no time”).
214 H. Weydt
Figure 5. Illustration of the meaning structure of eh (ohnehin, sowieso).
a b
(a‘)
Jedenfalls ’in any event’ is used as in the following example: Ich weiß nicht, ob Maria
intelligent ist, jedenfalls habe ich noch nichts Kluges von ihr gehört ‘I don’t know if
Maria is intelligent, jedenfalls (in any event) so far I have not heard anything intelligent
from her’. Jedenfalls can be explained by Fig. 6. It reads as follows: “there is a relatively
far-reaching claim”. The dotted circle symbolizes it here. The speaker can not fully affirm
it (here, he cannot deny Maria’s intelligence). But he can guarantee a weaker claim which
is part of it (the fact that he never had a corroboration of her intelligence), symbolized by
the full circle. Both claims stand in an inclusive opposition.
Figure 6. Illustration of the semantic structure of jedenfalls.
Immerhin ‘after all’ appears in affirmative sentences, for example, Hans hat ziemlich
viele Fehler in der Übersetzung, aber immerhin hat er das Examen bestanden ‘Hans made
quite a few mistakes in his translation, but he immerhin passed the test’. The particle
positions the content of the sentence in which it stands between two expectations. As can
be seen in Fig. 7 below, a first high expectation (a) is not met and as a consequence, one
might form a very low expectation (b). Compared to this second expectation, the reality of
the actual occurrence is still higher.
4.3. Why we use Abtönungspartikeln
Let’s return to the question posited in section 3 of how an amiable effect is brought about.
You see that dialogue A is full of elements which position the utterance in the context.
We have indicators like the following:
• denn: “I say this and I know that you know the answer. It’s precisely your preceding
utterance which made me ask you this.”
Figure 7. Illustration of the meaning structure of immerhin.
high initial
expectation
reality
Low
intermediat
ee
• ja (summary of shared idea): “I say this, knowing that you know it already, and that
you agree.”
• aber: “I say this and I am surprised about the extension of the fact which is expressed
in the sentence.”
• vielleicht: “Saying this I presume that you will be surprised about what I’m saying
because it is new for you.”
These particles have in common that they create a network of relationships between the
actual hearer and the actual speaker. They transform the dialogue into a common speech,
make it become more than a simple sequence of I say/you say. The actual speaker, A,
expresses that he or she not only makes his or her contribution in an authentic way but
models it in such a way that it takes into account the other’s (B’s) perspective. Instead of
making an independent statement, A continues B’s idea. A is aware of what B thinks and
believes, and A bases his or her contribution on B’s assumed state of mind. Therefore, the
dialogue, instead of being an exchange of independent turns, let alone of mutual feedings
with bits of information (“I give you some information, and then you give me some
information”, etc.) becomes a cooperative process of both interlocutors. They each in turn
express—by use of Abtönungspartikeln—that they respect and/or consider the other’s
view; each utterance is based on the preceding one. In a certain sense, the dialogue, even
in its individual steps, is joint work, a creation of both partners.
When such a dialogue occurs, it conveys to the partners a feeling of profound satisfaction.
The feeling exists, even if they disagree in content, because they realize that the one
understands (or at least tries to understand) the other. They cooperate in their effort to
understand each other and try to make each other understood. It is this very feeling which
is an important factor in bringing about the features of friendliness and amiability.
In summary, Abtönungspartikeln and related linguistic elements are used as specific

instruments for the partners’ cooperation. They help them to make the actual intention of
an utterance clear and to assign it its function in the developing interplay. Speakers who
express—by using particles—the fact that they seriously try to cooperate, are perceived of
as friendly, sociable, amiable, and able to make contact.
216 H. Weydt
5.1. Politeness versus friendliness
Friendliness and sociability are not to be identified as equal to politeness. Though

dialogue A is friendlier than dialogue B, it can hardly be said to be politer. In our
explanation of verbal friendliness, we don’t use the patterns of face-threatening acts
(FTAs) and of FTA avoidance, which are common in politeness research. There is no
danger of FTA in the speakers’ utterances and the particles don’t serve the purpose of
neutralizing FTAs. An interpretation of particle effects, based on FTA conception, would
miss the essential point.7
5.2. The question of equivalence and language comparison
There are languages which have a large inventory of Abtönungspartikeln and their
speakers use them frequently (languages like German, Dutch, White Mountain Apache,
Guaraní, Toura, Kera) and others which don’t (like English and the Romance languages).
No one, however, would claim that only German-, Dutch-, Apache-, Toura-, Kera-, and
Guaraní-speakers are able to be friendly and amiable when speaking and conversing with
each other, while Anglophones and the French are not. This raises an interesting new
problem for comparative linguistics, namely, how to deal with the tertium comparationis
“friendliness” in language comparison. If someone has to judge a translation of a literary
work, where, for example in the German original a friendly speaker shows his good
intentions by using Abtönungspartikeln, then he has to ask how the same effect is brought
about in the target language. It may be reached by means which are specific for that
particular language.
Without going into details, let me just mention two specific means in other languages
which are candidates for this purpose.8 In French (the hexagonal variety, not the
Canadian), using a form of reference to the partner is much more common than in other
languages. It is rather unpersonal and unfriendly to say au revoir ‘good bye’ instead of au
revoir, Madame. One may even omit au revoir or bonjour and just greet: Madame! or
Monsieur! French mothers correct their children, ironically repeating the phrase au revoir
qui? ‘good bye, who?’. The social process is essentially the same. By using the partner’s
name or title, the speaker shows that he takes his specific presence into account. In
Spanish, the threefold pronoun system esto, eso, aquello which expresses whether the
thing belongs to one’s own sphere, to the partner’s, or to neither of them may serve the
same purpose.
Notes
1
For the criteria and problems of classification, see Hentschel and Weydt (2002).
2
For a cross linguistic definition of particle see Hentschel and Weydt (1995) and Weydt
(2001).
3
For a discussion of this problem and the potential criteria of definition see Weydt (2001:
2.1 “Zum Problem der Einzelsprachlichkeit des Begriffs Partikel”).
4
Translation: kontaktschwach ‘difficult to make contact with’
5
I don’t know where the origin of the term down toner as a linguistic term lies. It may be
a calque, a loan translation, of the German Ab-tönung (ab ‘down’, Tönung ‘toner’). If
that were the case, it would be a rather bad translation. German ab- here does not have
the meaning of ‘down’ nor tönen the meaning of ‘to tone’. The expression abtönen
was taken from the art of painting, where abtönende Farben means ‘shadowing
colors’. These paints lend the painting a certain nuance.
6
As an example see Métrich (1993: 398 ff.).
7
The fact that particles often appear without polite effects is pointed out in Berger (1998)
and Hentschel (forthcoming).
8
A third is analyzed in Hentschel (1991).
Chapter 13
The Natural Semantic Metalanguage approach

to discourse markers
Catherine E. Travis
1. Introduction
This paper presents an analysis of discourse markers based within the framework of the
Natural Semantic Metalanguage (Wierzbicka, 1996; and references therein). It argues that
discourse markers can only be fully understood if the meaning(s) they carry when used in
different contexts are exhaustively defined. Within this framework, discourse markers are
treated as polysemous, having a range of different meanings all of which share some
element in common. The shared element of meaning can be considered a “partial semantic
invariant” (Goddard, 2000: 144; Wierzbicka, 1988: 344), and it is this that ties the uses of
the marker together, while other components of meaning that differ account for the
variation across the range of use. Such an analysis makes a clear distinction between what
is encoded in the semantics of the marker and what is encoded in its pragmatics of use. I
will illustrate how this can be done through an analysis of the Spanish discourse marker
bueno (‘well’, ‘good’, ‘right’), based on a corpus of conversational Colombian Spanish.
I am using the term discourse marker here in a broad sense to refer to the heterogeneous
group of linguistic items that act on (or “mark”) segments of discourse and function to
indicate how those segments are to be understood in the context of the surrounding
discourse. As in the other papers in this volume, the class is thus given a functional rather
than a formal definition. There are, however, a number of formal features that are
typically associated with discourse markers, such as prosodic, syntactic, and semantic
independence. I will discuss in some detail the functional and formal features of discourse
markers in an attempt to help clarify what constitutes this class and how discourse
markers can be distinguished from other related elements.
1.1. Approach
My approach to the study of discourse markers involves three key aspects: the
environments in which the marker occurs, the functions it plays, and the meaning(s) it
carries. Understanding each of these is essential to a full understanding of discourse
marker use.
220 C. E. Travis
The environment of occurrence refers to the structural position in which the marker is
found, including: position in the turn (initial, medial, final), the kind of syntactic and/or
discourse unit it follows and/or precedes, whether it acts on preceding or upcoming
material, and so on. Included here is the prosody of the marker. I deal with prosody in a
very general sense, in terms of the intonation contour with which the marker occurs:
continuing (with a slight rise in pitch), final (with a fall to low pitch), or rising (with a
high rise in pitch).1 In some cases, the prosody of the surrounding material may also be
relevant–for example, whether the intonation unit prior to the marker occurs with
continuing, final, or rising intonation, or is left incomplete. Discourse markers often occur
in a range of different positions and with a range of intonation contours, and as we will
see in the analysis of bueno, these must be taken into account when considering marker
meaning.
Discourse markers are highly multifunctional. For example, they can be used to mitigate
an utterance, to highlight an utterance, to seek listener agreement, to move on to a new
topic, to close a topic, to reformulate an utterance, and so on. The one function may be
found in a range of discourse environments (e.g., utterances can be mitigated at any point)
or it may be specific to one environment (e.g., a marker that introduces a new topic may
be more likely to occur turn-initially). Thus, function needs to be considered
independently from environment of occurrence.
Function also needs to be considered independently of semantics because not every

function of use represents a distinct meaning. For example, a marker can be used with the
same meaning to carry out related functions, such as seeking agreement and closing a
topic, or seeking agreement and mitigating an utterance (as is the case for bueno). In order
to recognize whether a marker is being used with the same meaning in different contexts,
we need to be able to distinguish between what is inherent to the meaning of the marker
and what is contextually induced, and thereby identify the semantic core of the marker.
A description based solely on environment of occurrence and function fails to capture

much important information about the marker. In particular, it does not capture what it is
that that is shared across the range of use, how one marker differs from others that can be
used with similar functions and in similar environments, nor what the relationship is
between the marker and homonymic forms in other word classes. The semantics of the
marker represents “the unifying principle behind its apparently diverse use” and “the logic
which controls the native speaker’s use of it” (Wierzbicka, 1976: 333). A full analysis of a
discourse marker must therefore include an outline of its semantics, alongside discussion
of its environments of occurrence and functions.
In the analysis of bueno presented below I will attempt to show how the environments of
occurrence can be distinguished from its functions and how, from this, we can extrapolate
its meanings. Before doing this, I will outline the basic tenets behind the methodology to
be applied in the semantic analysis.
1.2. Methodology
The analysis of discourse markers to be presented here represents an extension into the
area of discourse of a semantic theory that has been widely applied to lexical semantics,
NSM Approach to Discourse Markers 221
namely Natural Semantic Metalanguage (NSM) (Goddard and Wierzbicka, 1994, 2002;
Wierzbicka, 1996; among others). This approach is a way of defining words and concepts
via reductive paraphrase. It involves using a limited set of simple words and simple
syntactic patterns as the language of definition to present a paraphrase of the meaning
under consideration, that is, “an equivalent expression composed exclusively of simpler
meanings than the original” (Goddard, 2000: 129).
NSM theory is based on the premise that although semantic systems of different
languages are unique and most words do not have direct translations in other languages,
there is a small set of words, called primitives, that are found in all languages. NSM
theorists maintain that the meaning of each of these words is so basic that they cannot be
defined and that they can be used to define all other words and concepts. The list of
primitives proposed in Goddard and Wierzbicka (2002) numbers around 60 and is given
in Appendix A.
NSM theorists also maintain that there is a small set of universal rules regarding the ways
in which these words can be combined to form sentences. It is claimed that these patterns
of cooccurrence represent a kind of “universal grammar” (cf. papers in Goddard and
Wierzbicka, 2002; Travis, 2003; Wierzbicka, 1996: 112ff.). Thus, SAY , for example, is
argued to occur universally with a person subject and an object complement (e.g., “I say
something”), with a topic of speech (“I say something about something”), with an
addressee (“I say something to someone”), and to introduce direct speech (“I say: I want
you to know something”). Although these notions may be expressed with different
syntactic structures in different languages, the theory maintains that there is no language
in which SAY cannot occur in these environments and that these basic sentences are
directly translatable across languages.
Within the NSM approach, word meaning is understood to be made up of a set of

components. The set of components presented in a definition represents the core meaning,
or the semantic invariant, that is, those elements of meaning that are shared across all
contexts of use. This involves making a strict distinction between pragmatics and
semantics: any elements of meaning that are attributable to the environment of occurrence
are understood to be pragmatic and should not be included in the definition. Thus, unlike
many other semantic approaches, NSM argues that the meaning of words is determinate
and can be exhaustively explicated.
If it is not possible to capture all uses of a word in the one definition, that is, where there
are differences in meaning across the range of use that cannot be attributed to the context,
then the word is polysemous. The traditional definition of polysemy is the existence of
two or more related meanings (Lyons, 1977: 550ff.). This definition is adopted by NSM,
but the nature of the approach allows us to be more specific about what is meant by
related: a word is polysemous if it can be shown to have two (or more) distinct meanings
each of which shares at least one of their components. Components here does not refer to
some abstract notion (e.g., the fact that the two meanings of port, ‘harbor’ and ‘liquor’,
both refer to concrete entities) but rather to a specific element of meaning that can be
expressed via reductive paraphrase and that would need to be included in an exhaustive
definition of the item under consideration.
222 C. E. Travis
This definition of polysemy means that the range of definitions of a polysemous item will
share a semantic core, or a “partial semantic invariant” (partial because on its own it does
not account for the full meaning, but nevertheless invariant because it is found across the
range of use of the item; Goddard, 2000: 144; Wierzbicka, 1988: 344). Where there is no
partial semantic invariant the meanings are not related, and we are dealing with
homonymy, not polysemy. In the analysis of bueno presented below I will argue that it
has four different but related meanings, and I will illustrate how its polysemy is captured
in the set of definitions proposed through both shared and distinct components of
meaning. (See Fischer, this volume, for a similar account of discourse marker meaning).
The notion that discourse markers are polysemous with a partial semantic invariant shared
across the range of use offers great insight into their meaning and use. Extracting shared
components of meaning highlights the link between different uses of the markers as well
as the link between marker and lexical uses of the same forms (where such forms exist),
for example, in the case of bueno, which functions both as a marker and as an adjective
meaning ‘good’. Identifying the range of meanings a particular marker has also helps to
explain the broad functional spectrum in which it is used.
NSM has not been widely applied to the analysis of discourse markers. Where it has been
applied, however, it has proved an extremely useful tool for capturing marker meanings
(Fischer, 1998; Goddard, 1994; Rieschild, 1996; Travis, 1998, 2005; Wierzbicka, 1976,
1994). This paper (and that by Fischer, this volume) represents an important advancement
of the application of the theory into the area of discourse where its merits remain to be
exploited.
1.3. Data
This study is based on a corpus of conversational Colombian Spanish recorded in the city
of Cali, Colombia, in 1997. The corpus consists of five and a half hours of conversation
(roughly 60,000 words), and contains 130 tokens of the marker bueno, to be discussed in
detail below. The data were collected by two native speakers, who recorded spontaneous
conversations between themselves and their family and friends over a period of two
months. I believe the data are as natural as is possible in a situation where participants are
aware they are being recorded.2
The data have been transcribed in accordance with the method developed at the University
of California, Santa Barbara (Chafe, 1993; Du Bois, Schuetze-Coburn, Paolino, and
Cumming, 1992). The transcription conventions are given in Appendix B. Central to this
method is the concept of the intonation unit, which is defined as “a stretch of speech
uttered under a single, coherent intonation contour” (Du Bois et al., 1992: 17). The
perception of a stretch of speech as being uttered under a coherent contour is largely
determined by the boundaries of that contour, and intonation unit boundaries often co-
occur with any or all of the following: pausing, pitch reset, accelerated speech at the
beginning of a new intonation unit (anacrusis), and lengthening of the syllable or syllables
at the end of an intonation unit (Chafe, 1993: 34ff., 1994: 57ff.; Du Bois et al., 1992:
100ff.). This is significant to us here, as discourse markers typically occur in an intonation
on their own, as will be discussed below.
The three biggest issues surrounding discourse marker research in recent years and
discussed at length in the papers in this volume are the following:
• how to define the class and what to call it,

• how to capture discourse marker meaning and whether it is best characterized as
semantic or pragmatic, and
• how the polyfunctionality of markers and the relationship between the different
functions can be accounted for.
The lack of agreement on a name and a definition for the class is partly due to the
different theoretical approaches within which discourse markers have been studied, but it
is also partly an artifact of the heterogeneity of the class (see discussion in Fischer, 2000a:
13ff.; Jucker, 1993: 436; and Jucker and Ziv, 1998: 1ff.). The only area of consensus
regarding discourse markers, even for the authors in this volume, appears to be that they
are heterogeneous and represent a functional class made up of items playing some kind of
pragmatic role in discourse management. In an attempt to better understand what makes
up the class of discourse markers, in the following section, I will discuss some functional
and formal characteristics that are typically associated with them.
Given the lack of agreement about what discourse markers are it is not surprising that
there is no widely accepted label for such items. In order to bring about some degree of
standardization in the terminology, a maximally broad term that can cover as wide a range
as possible of items generally considered to fall into this class is desirable. I believe the
term discourse marker meets these criteria. This term has been used in the past to include
a range of related items, including connectives, (discourse and modal) particles,
interjections, tag questions, and so on. The term also has the advantage of accurately
capturing the idea that these elements “mark” segments of discourse, given their role of
indicating how chunks of discourse are to be understood in the broader context of use.
In terms of discourse marker meaning, I am drawing a sharp distinction between

semantics and pragmatics, based on what elements of meaning are inherent in the marker
itself and what are pragmatically induced from the context. In the analysis of bueno
below, I will show that it adds an identifiable and specifiable meaning to the contexts in
which it is used and that this can be distinguished from further implications that may
become attached to bueno in particular contexts (what Hansen, this volume, refers to as
“side-effects”).
In accordance with this understanding, I believe that discourse markers can be defined
using the same methodology as is applied in lexical semantics. Thus, I am in agreement
with Fraser (this volume) that the distinction between “procedural” and “conceptual”
meaning is not tenable (see discussion of this distinction in Nyan, this volume; Ler, this
volume; and Blakemore, 1987, 1996). The main problems with treating discourse markers
as carrying a different kind of meaning from other lexical items are first that it is often
unclear whether a given item is functioning as a discourse marker or not (as noted by
Hansen, this volume), and second that it obscures the relationship between the marker and
lexical uses of homonymic forms. The relationship between the marker bueno and the
224 C. E. Travis
adjective meaning ‘good’ will be discussed in the analysis presented below, and it will be
shown that the meaning of the adjective forms part of the meaning of the marker.
This leads us to the final problem noted above, related to the polyfunctionality of
discourse markers. The NSM framework employed here neatly captures the relationship
between the different uses of the marker in the identification of a partial semantic
invariant that is shared across the range of use. At the same time, it shows how marker
uses relate to the lexical use of the same item, as there is likely to be some element of
meaning in common given that markers typically derive from their lexical counterparts
(see Waltereit, this volume; also Brinton, 1996; Onodera, 1995; Traugott, 1995a, 2003).
The polyfunctionality is not problematic in this analysis, in which (i) markers are
recognized as commonly being polysemous, so it is accepted that they may also have a
range of functions and (ii) each function does not equate with a distinct meaning, as
discussed above, so we do not have a random assignment of multiple meanings.
In Sections 3 and 4, in which I present the analysis of the Spanish marker bueno, I will
demonstrate how polyfunctionality and polysemy interrelate and how the notion of the
partial semantic invariant ties the different uses together. Before doing this, we will first
consider in more detail some of the features of discourse markers that can help delimit the
class.
2. Definition
I noted above that discourse markers represent a heterogeneous class. Because of this,
there appears to be no set of criteria that can definitively distinguish discourse markers
from other related linguistic elements and exhaustively delimit the class (as is also noted
in many other papers in this volume). There are, however, certain features that are
typically associated with discourse markers that can help us identify whether a given
element is functioning as a marker in any one context. These features are both functional
and formal. Functionally, discourse markers can be described as playing both a
contextualizing and an interactional role, being used to indicate how an upcoming or prior
utterance is to be understood in the context of the surrounding discourse, and to indicate
something about the speaker’s attitude to the message content and to the addressee.
Formally, discourse markers tend to be intonationally, syntactically, and semantically
independent from the surrounding discourse. I will now briefly discuss each of these
features and will show how they can help distinguish discourse marker uses from non-
discourse marker uses of the same forms (see Travis, 2005, for more detailed discussion
of these features). It will be seen that these features are not categorical, but taken as a set
they provide a useful guide as to whether or not a given item is functioning as a discourse
marker in a certain environment.
2.1. Contextualizing and interactional role of discourse markers
One of the key roles of discourse markers is to indicate how an upcoming or prior
utterance is to be understood in relation to the discourse context in which it occurs
(Blakemore, 1987: 105; Briz, 1993a: 147; Fischer, 2000a: 26; Fuentes Rodríguez, 1993c:
71; Hansen, 1998a: 75; Lenk, 1998: 246; Redeker, 1991: 1168; Schiffrin, 1987a: 327). In
this sense, markers function as a kind of metamessage to the addressee, guiding them
towards the interpretation intended by the speaker. In doing this they also play an
interactional role, indicating something about the speaker’s attitude to the hearer and/or
the discourse content (Briz, 1993b: 50ff.; Fischer, 2000a: 26; Fuentes Rodríguez, 1993c:
71; Lenk, 1998: 247; Martín Zorraquino, 1991: 225; Östman, 1981:39ff,; Schiffrin,
1987a: 322ff.).
The following examples present some widely used discourse markers in Colombian
Spanish and illustrate the contextualizing and interactional role they play as compared
with their corresponding nondiscourse marker forms. Examples (1) and (2) illustrate,
respectively, the marker use of bueno, meaning something similar to English ‘well’ and
its lexical use, meaning ‘good’. In (1), Milena is discussing work done by environmental
organizations in the small town where she lives and is saying that in the light of all there
is to be done, there is not enough money. Rosario accepts this with bueno but adds that
even small projects, when timely, can make a difference.
(1) M: . . No es suficiente.
not be-3SG sufficient
R: Sí.
~~ Yes
~~ . . . Bueno,
~~ pero a veces cositas pequeñitas,
~~ but PREP times things-DIM small
~~ no?3
~~ not
~~ Hay proyeticos así=,
~~ there-is projects-DIM like.this
~~ Como=,
~~ like
~~ . . puntuales,
~~ ~ timely
~~ . . No?
~~ ~ not
MILENA: It’s not enough.
ROSARIO: Yes. . . . Bueno, but sometimes little things, don’t you think? There
are projects like that, kind of, timely, don’t you think?
contamination (1262-1269)4
The role of bueno here is to indicate that Rosario accepts what Milena has said but that
her acceptance is only partial and some modification is required. Note that it occurs with
continuing intonation, indicating that the speaker has something further to say.
Interactionally, it mitigates this upcoming modification. We can readily see how this
differs from the lexical use of bueno illustrated in the following example.
(2) P: . . Es que,
be-3SG COMP
~~ ese diccionario es muy bueno.
~~ that dictionary be-3SG very bueno
226 C. E. Travis
~~ Muy completo.
~~ very complete
PATRICIA: It’s that that dictionary is very bueno [‘good’]. Very thorough.
dictionary (635-637)
Here, bueno is modifying the noun dictionary. It is not contextualizing a prior utterance,
and it is not playing an interactional role.
Bueno can also occur on its own to form a turn. In this environment, it expresses the
speaker’s acceptance of a prior utterance. It is clearly not functioning as an adjective, as it
is not modifying a noun, but its status as a discourse marker is not clear-cut, as it does not
contextualize an utterance (because it occurs on its own). This is illustrated in the
following example where Angela asks her aunt, Celia, to fill in the consent form for this
study, and Celia agrees with bueno.
(3) A: Me haga el <@ favor~@>.

1SG.DAT do-3SG.SJV the ~ favor
C: Yo?
~~ I
~~ Bueno.
ANGELA: If you’d be so kind.
CELIA: Me? Bueno.
almuerzo (1277-1279)
The role of bueno here is primarily interactional in that it encodes a notion of acceptance
similar to that encoded in (1) (although here it indicates full, as opposed to partial,
acceptance). In this sense, it can be treated as a kind of discourse marker and should be
analyzed alongside the use presented in (1). The differences in these two uses serve to
demonstrate the heterogeneity of the class and the difficulty that sometimes arises in
classifying different uses.
Examples (4) and (5) present, respectively, the use of o sea as a marker, meaning
something like English ‘I mean’, and its lexical use, meaning ‘or be it’. In (4), Sara is
explaining different payment options available in an insurance policy. She states that six-
monthly payments are made to a cashier and then, following o sea, specifies the mode of
payment this involves.
(4) S: .. Cuando se paga semestral,

~~ ~ when 3REFL pay-3SG six-monthly
~~ el pago es po=r caja.
~~ the payment be-3SG for cashier
~~ O.sea,
~~ en efectivo o en cheque.
~~ in cash or in cheque
SARA: When it’s paid six-monthly, it’s paid to a cashier. O sea, by cash or by
cheque.
insurance (130-133)
In its lexical use, o sea introduces one noun phrase as an alternative for another. Literally,
o sea is made up of ‘or’ and the third person singular present subjunctive form of the verb
ser ‘to be’. There are no examples of the lexical use of o sea in the conversational
database, and this use is very rare in casual conversation. I will therefore illustrate this use
with an example drawn from a Colombian novel. The writer here is discussing rally
driving and the fact that it is hard for Colombian drivers to get sponsorship, because the
companies that make the car parts sponsor people from their own country and there are no
Colombian companies involved in manufacturing such parts.
(5) Allí pesan los intereses de los fabricantes de motores, los

there weigh-3PL the interests of the manufacturers of engines those
de las llantas, los de no sé quién, para que el escogido
of the tyres those of not know-1SG who for COMP the chose-PART
sea un alemán o sea un francés o sea
be-3SG.SJV one German or be-3SG.SJV one French or be-3SG.SJV
un japonés~. . .
one Japanese
There what counts are the interests of the manufacturers of the engines, of the
tyres, of I don’t know who, so that the chosen [driver] be a German o sea (‘or be’)
a French person o sea (‘or be’) a Japanese . . .
(Castro Caycedo, 1999: 205)
The lexical use of o sea could be described as a more formal way of saying o ‘or’, in that
the verb in the preceding clause is repeated. The reason the verb ser ‘to be’ is used in (5)
to introduce the alternative is simply because that is the verb used in the preceding clause.
The role of o sea in this literal sense is neither contextualizing nor interactional, and it can
readily be distinguished from the marker use in (4).
The following two examples illustrate respectively the use of verdad ‘true’ as a tag
question and as an adjective. Example (6) comes from a conversation in a restaurant, and
Angela is urging her interlocutors to decide what to order. Verdad here does not mark the
truth of her utterance (which is a question and therefore not verifiable) but highlights its
importance, indicating to her interlocutors that they should pay attention to her.
(6) A: Ve,5
hey
qué vamos a pedir,
what go-1PL PREP order-INF
~~ verdad.
ANGELA: Hey, what are we going to order, verdad.
pizza (340-342)
In (7), verdad is used in its literal sense meaning ‘true’. Here it does not act on another
utterance to contextualize it, nor is it playing an interactional role. Rather, it is used to
verify the truth of a prior statement.
228 C. E. Travis
(7) A: yo no te estoy diciendo que no, que mi mamá

I not 2SG.DAT be-1SG say-GER COMP no comp my mum
no sea así, porque sí.
not be-3sg.sjv like.this because yes
Sí es verdad.
yes be-3SG verdad
ANGELA: I’m not saying that no, my mum isn’t like that, because yes, it is
verdad (‘true’).
restaurant (627-630)
These few examples serve to illustrate the functional role markers play and how this
distinguishes them from related forms in other word classes. They show how it is that
markers function as metamessages about how the speaker wants an utterance to be
understood in the context of the surrounding discourse. As is also argued by Nemo (this
volume) and Nyan (this volume), the metamessage the marker gives represents its core
meaning. Discourse markers can therefore be defined in terms of the comment they make
about the utterance(s) they mark, and this can be captured in NSM definitions with the
component “I say: . . .”.
Another point to make here is that markers indicate that an utterance is noninitial. That is,
they mark an utterance as responding to some prior aspect of the discourse, or, in some
cases, to an extralinguistic situation (see Martín Zorraquino, 1994: 410, for further
discussion of this point). This allows us to specify further the format of marker definitions
proposed within the NSM framework, which should include the notion that discourse
markers function as a metalinguistic comment on prior discourse or on an extralinguistic
situation. I propose that marker definitions be presented in the following format:
someone said something / something happened

I say: . . . .
“Someone said something” captures the use of a marker to respond to prior discourse and
“something happened” to respond to an extralinguistic situation. The metalinguistic
comment expressed by the discourse marker is given following “I say”. It should be
understood that this format is not intended to define discourse markers as turn-taking
devices. Fischer (2000b) conclusively shows that discourse markers do not signal turn
taking as such, but that their use in turn taking is contextually derived. That is, they
indicate that the speaker is responding to prior discourse, and in some contexts, this may
involve taking the floor (“I want to say something”) or handing over the floor (“I don’t
want to say any more”), but any such information is derived from the context of use in
conjunction with the semantics of the marker rather than being explicitly encoded in the
marker. We will see how this applies to the different meanings of bueno in section 4
below.
2.2. Independence of discourse markers
2.2.1. Prosodic independence

Discourse markers tend to be set off intonationally from the surrounding discourse, as has
been widely recognized in the literature (Chafe, 1993: 37; Du Bois et al., 1992: 103;
Hansen, 1998a: 66; Martín Zorraquino, 1991: 255; Redeker, 1991: 1166).
It is important to note that prosodic independence is not determined solely on the basis of
pausing, for, as Pons (this volume) states, discourse markers are not necessarily
surrounded by pauses. As can be seen above, bueno in (1) is preceded but not followed by
a pause, there is no pausing surrounding o sea in (4), and verdad in (6) follows directly on
from the prior material. Prosodic independence as used here refers to discourse markers
occurring in an intonation unit on their own, and, as noted above, pausing is just one of a
set of features to identify intonation units. In the database, the majority of the discourse
markers occur in intonation units on their own, and when they do not, they occur with
minimal accompanying material, such as ah ‘oh’ (e.g., ah bueno ‘oh, OK’), or y ‘and’
(e.g., y entonces ‘and so’).
The notion of prosodic independence can help distinguish discourse marker uses from
lexical uses. In the examples given above, we can see that bueno in (1) and (3), o sea in
(4), and verdad in (6) occur in an intonation unit on their own when they are used as
markers, but they are prosodically integrated into the surrounding material in (2), (5), and
(7), when they are used in their lexical sense.
2.2.2. Syntactic independence

Discourse markers are syntactically independent in that they are not part of the core
syntactic structure of the utterance in which they occur. This means that the utterance
would remain syntactically intact if they were removed. This is presented in most studies
as one of the defining features of discourse markers (see, for example, Fischer, 2000a: 26;
Fraser, 1988: 27; Hansen, 1998a: 75; Schiffrin, 1987a: 328; among others for discussion,
as well as the other papers in this volume).
This can again be illustrated with the examples seen above. Consider the use of bueno in
(1) and (2): in (1), bueno need not be stated, while in (2), the removal of bueno results in a
syntactically incomplete utterance (ese diccionario es muy ‘that dictionary is very’). In the
case of o sea, in (4), the utterance remains intact without its use, while in (5) a different
meaning is expressed. Without o sea (para que el escogido sea un alemán, un francés, un
japonés ‘so that the chosen one be a German, a French person, a Japanese’) the utterance
reads as a list of different possibilities, rather than specifically presenting ‘French’ and
‘Japanese’ as alternatives for ‘German’. Verdad in (6) can also be removed, while its
removal in (7) would imply an elided adjective (leaving sí es ‘it is’).
As is stressed by Hansen (this volume) and Fraser (this volume), the fact that the utterance
remains syntactically intact without the use of the marker does not mean that the same
meaning is expressed. Removal of the marker affects the pragmatics of the interaction, as
the utterance is no longer contextualized in the same way and the interactional meaning is
lost.
230 C. E. Travis
2.2.3. Semantic independence

Semantic independence is used here analogously with prosodic and syntactic
independence to refer to the fact that discourse markers are not inherently tied to any
specific element of the discourse, that is, their scope is indeterminate.
In (1), for example, we cannot know how much of Rosario’s turn bueno is intended to
introduce, whether it is it just marking ‘but sometimes little things’, or whether it is also
marking what Rosario goes on to say, ‘there are projects that are timely’. In (2), on the
other hand, where bueno is used as an adjective, it is marking the preceding noun,
‘dictionary’. In (3), bueno responds both to Angela’s request and Celia’s confirmation that
it is she who needs to sign the form. Here bueno is referring to a general notion that has
been established in the prior discourse rather than to any one specific contribution.
In the case of o sea, we can see that in (5) it is marking the noun phrases which it presents
as alternatives (German, French, Japanese), while in (4) it could be understood as
specifying what is meant by the preceding intonation unit ‘it’s paid to a cashier’, or by the
two preceding units ‘when it’s paid six-monthly, it’s paid to a cashier’.
Likewise, in (6) it is impossible to state how much of the preceding material is marked by
verdad (i.e. ‘what are we going to order’ or ‘hey, what are we going to order’). The lexical
use of verdad in (7), however, also has broad scope, as it is referring to what has been said
about Angela’s mother in the prior conversation. Discourse markers are of course not the
only elements that can be used with indeterminate scope, and this feature on its own is
therefore of little use. It is significant only when seen as part of a set with the other
features outlined here.
The features in this set are interrelated: the prosodic independence of markers is related to
the fact that they are not integrated into the core syntactic structure of the utterance, and it
is this that facilitates their marking large, often indeterminate, segments of text. This in
turn is related to their functional role of contextualizing those segments in the surrounding
discourse.
3. Functional Spectrum
As discussed above, discourse markers fulfill a multitude of functions, and a single

discourse marker can also have a very broad range of use. Polyfunctionality has been of
great concern to those working with discourse markers, and because of their wide range of
use, a certain vagueness or indefinability in the semantics appears to be assumed.
However, it seems that the vagueness of the semantics of markers is taken as a given,
rather than treated as a testable hypothesis. In the remainder of this paper, I will argue
against this notion and will show that, with an appropriate semantic methodology,
discourse markers can be analyzed with the same method as other lexical items, and their
meaning(s) can be identified and exhaustively defined.
In order to do this, we will consider the marker bueno, looking further at its range of use,
and outlining the meanings these uses carry.
3.1. Discourse environments of bueno6
The marker bueno occurs in a number of different environments. We have seen above, in
(1) and (3), that it can occur in response to a contribution made by an interlocutor. I
classify all such uses as turn initial, although, as can be seen in these two examples, it can
be preceded by other minimal material. Turn-initial bueno occurs with both continuing, as
in (1), and final intonation, as in (3), and can be uttered in response to both comments, as
in (1), and questions, or other material requesting a response, as in (3).
Bueno also occurs turn medially, responding to the same speaker’s speech, and again, it
occurs with both continuing and final intonation. As well as the general distinction
between turn-medial bueno with continuing and final intonation, we can also note its use
in one specific environment, which is prefacing direct speech. This use is clearly
functionally distinct from the others, and it is also structurally distinct in that it occurs
either accompanied by the verb decir ‘to say’, or with marked syntax indicating that it is
introducing direct speech (e.g., a subject without a verb, y yo, bueno. Ojalá. ‘And I,
bueno, I hope so.’).
Bueno does not occur turn finally in the database. We saw above that it can constitute a
turn on its own, but I have classified such use as turn initial to highlight the similarity
between this use and its turn-initial use prefacing a response. Bueno is not used with rising
intonation in Colombian Spanish, though such use is heard in other dialects of Spanish.
3.2. Functions of bueno
The discourse environments in which bueno occurs map onto a range of different
functions. These are laid out in Table 1 below. As this table shows, there is a very close
correlation between the structural position in which the marker occurs, its prosody, and its
function. The correlation breaks down for the use of bueno to mark acceptance and pre-
closing, which share the same structural and intonational properties. This is because these
two functions share the same meaning. Bueno marking a reorientation occurs in both turn-
initial and turn-medial position, which is evidence of its more generalized semantics. And
bueno marking direct speech occurs with both final and continuing intonation, which is
due to the fact that in this function, bueno serves to “quote” one of the other functions
identified, and its variable intonation is a feature of the function being “quoted”.
In the following discussion, I will address the different functions individually, and in
section 4, I will outline the different meanings of the marker.
3.2.1. Acceptance
The use of bueno to encode acceptance was seen above, in example (3), in which Celia
used bueno to agree to Angela’s request to sign a form. This function has been widely
recognized in the literature (Bauhr, 1994: 92ff.; Beinhauer, 1968: 352ff.; Cortés
Rodríguez, 1991: 112; Fuentes Rodríguez, 1993a: 208ff.; Martín Zorraquino, 1991: 263;
Ocampo, 2002), although it is often described as encoding a notion of concession or
resignation, as though it were a kind of reluctant acceptance (Bauhr, 1994: 92ff.;
Beinhauer, 1968: 353, 355; Fuentes Rodríguez, 1993a: 208ff., 1993b: 194). There appears
to be no implication of reluctance in (3), and I would argue that rather than registering
232 C. E. Travis
“concession” or “resignation”, bueno is simply a neutral acceptance device which can be

contrasted with elements such as con mucho gusto ‘with pleasure’ and muy bien ‘very
good’, which could be seen as more enthusiastic ways of expressing acceptance. Because
of its neutrality, it can occur with reluctant responses, but the reluctance is not encoded by
bueno itself but by other material with which it may co-occur.7
Table 1. Functions of bueno
Structural position Intonation contour8 Function

acceptance
final
Turn initial preclosing
continuing mitigation
Turn initial/medial final reorientation
continuing correction
Turn medial
final/continuing direct speech
3.2.2. Preclosing
Bueno is often used prior to the final closing of a conversation as a way of moving into
the leave-taking phase. This use is very similar to that of bueno marking acceptance, but
in this case, bueno expresses acceptance of what has been said in the conversation or that
the conversation has taken place. This use has been discussed in the literature, and it is
generally described as playing the role of seeking agreement to end the conversation
(Bauhr, 1994: 111ff.; Gregori Signes, 1996: 168; Martín Zorraquino, 1991: 263;
Placencia, 1997: 59). The following example comes from a telephone conversation of
which only one side was recorded (hence the long pauses, which are when the other
interlocutor is talking), but the preclosing role of bueno can clearly be seen.
(8) M: Sí.
yes
. . . (2.0) Ah hah.
Ah hah
. . . Bueno.
. . . Bue=no.
Hasta luego,
until later
chao.
ciao
MILENA: Yes. . . . (2.0) Ah hah. . . .Bueno. . . . Bue=no. See you later, ciao.
campaign (206-211)
This use appears to have become a somewhat ritualized part of the leave-taking process,
as interlocutors mark their general acceptance of the preceding conversation in an attempt
to undertake a smooth farewell.
3.2.3. Mitigation
The use of bueno to mitigate an upcoming modification of a contribution by an
interlocutor was seen in (1), where it introduces a response that partially disagrees with
the prior contribution by another interlocutor. This use is also found when bueno prefaces
an answer to a question, in which case it occurs with answers that are not straightforward.
This has been noted in the literature, and it is argued that bueno is used to soften
responses that are not what would be expected (or desired) from the surrounding context,
such as comments that do not concord with prior discourse, statements that disagree with
what someone else has said (as in (1)), or answers that do not fully respond to a question
(see Bauhr, 1994: 120; Cortés Rodríguez, 1991: 108; Fuentes Rodríguez, 1993a: 219;
Gregori Signes, 1996: 161ff.; Ocampo, 2002; Serrano, 1999: 121ff.). I am therefore using
mitigation in a broad sense here to refer to the use of bueno marking dispreferred
responses (Levinson, 1983: 332ff.).
An example of the use of bueno prefacing an answer to a question is given below. Here,
Angela asks her interlocutor about the difference in price between various options
available in an insurance policy she is taking out. Sara does not immediately answer this
question: she first outlines one very important element of the policy (not reproduced
here), and only then does she turn to answer Angela’s question. Thus, the response
immediately following the question does not in itself constitute an appropriate answer,
although Sara does go on to answer Angela’s question following this. Bueno is used in
this context to acknowledge the validity of the question and to indicate at the same time
that it is not going to be answered without some further information being given.
(9) A: Y cuál es la diferencia,

and which be-3SG the difference
O.sea,
I mean
En plata.
in money
@@
S: .. Bueno,
Y .. entonces,
and so
Eso también es super importante.9
that also be-3SG super important
ANGELA: And what’s the difference, I mean, in money. @@
SARA: Bueno, and so, that is also very important.
insurance (377–383)
The mitigating role of bueno (when it prefaces a response) and its role marking
acceptance (when it constitutes a response on its own) are similar, as both involve saying
something positive about some aspect of the prior discourse. However, when bueno
234 C. E. Travis
prefaces a response, it encodes only partial acceptance, and indicates that a modification is
upcoming. This is also in accordance with the prosodic difference between these two uses,
with bueno encoding acceptance occurring with final intonation (implying that nothing
more is forthcoming) and bueno encoding partial acceptance occurring with continuing
intonation (implying that the speaker is going to say something more). In this sense, we
have prosody interacting with the semantics of the marker, as will be discussed further in
section 4 below.
3.2.4. Reorientation
We will now consider the use of bueno to mark a reorientation in topic. This includes
introducing a new topic, closing a topic, prefacing a digression from the main topic,
returning to a prior topic following a digression, and moving on to the key point of a
topic, having outlined background information (see Bauhr, 1994: 106; Beinhauer, 1968:
352ff.; Cortés Rodríguez, 1991: 105ff.; Fuentes Rodríguez, 1993a: 210ff.; Gregori Signes,
1996: 167; Martín Zorraquino, 1991: 261, 1994: 411; Ocampo, 2002; Serrano, 1999:
120). In accordance with this broad range of use, bueno can occur in turn-initial position,
responding to a comment by another interlocutor, or in turn-medial position, in response
to a contribution by the same speaker. In the following example, bueno indicates that the
speaker is moving from introductory information of the story she is telling to the main
point. She is talking about a man in the restaurant where she and her husband are having
lunch, and having stated that he appears to have some kind of problem, she goes on to
present an example of this, that he has been talking to himself.
(10) A: Mira que,

look-2SG.IMP COMP
.. este señor,
this man
. . . (2.0) Hm,
Hm
Parece que tiene es como~--
seem-3SG COMP have-3SG be-3SG like
.. un problema,
one problem
o yo no sé.
or I not know-1SG
Bueno.
. . . Estaba ahorita=~--
be-3SG.IMPF now-DIM
así como,
like.this like
hablando . . solo?
speak-GER alone
ANGELA: Look, this man, . . . (2.0) Hm, he seems to have kind of a problem, or
something. Bueno. . . . Just now he was kind of talking to himself.
restaurant (911–920)
As with the other uses of bueno we have seen, its role here is to mark acceptance of what
Angela has said. In this case, her acceptance of this information allows her to move on;
she does not provisionally accept what she has just said in order to modify it but in order
to continue with the main point of what she wants to say. Note that bueno occurs with
final intonation, giving the prior discourse a tone of completion, with the implication
being that having completed one aspect of the discourse the speaker can move on to
another.
3.2.5. Correction
When bueno occurs turn medially with continuing intonation, it functions to introduce a
correction or a modification of what the same speaker had said in prior discourse.
Specifically, having made a comment, the speaker comes to be aware of a possible
objection to what they have said, and they use bueno to acknowledge this possible
(unspoken) objection and to introduce a modification of what they have said. This
function has also been discussed in the literature (Bauhr, 1994: 101ff.; Cortés Rodríguez,
1991: 106ff.; Fuentes Rodríguez, 1993a: 175; Gregori Signes, 1996: 162ff.; Martín
Zorraquino, 1991: 263, 1994: 409). This is illustrated in the following example, where
Omar begins to ask how a certain problem has been resolved but then cuts himself off to
acknowledge that the problem has not been fully resolved. He introduces this correction
with bueno.
(11) O: Cómo solucionaron~--

how solve-3PL.PRET
.. Bueno,
claro que,
of.course COMP
me imagino que no se ha solucionado.
1SG.REFL imagine COMP not 3REFL have-3SG solve-PART
. . . El problema de~--
the problem of
de~--
of
Presupuesto para=~--
budget for
OMAR: How did they resolve--Bueno, of course, I imagine that it hasn’t been
resolved. . . . The problem of the budget for --
Tumaco (1825–1831)
Bueno here indicates that the corrected material is not entirely wrong, that is, it can be
partially accepted, but that some minor modification is needed. Note that this use is
similar to that described above as functioning to mitigate an upcoming modification of
another’s utterance: both imply some kind of acceptance of prior discourse before going
on to say something slightly different.
3.2.6. Direct speech

The final function of bueno identified in the corpus is its use introducing direct speech.
When used in this way, bueno highlights the upcoming quoted material by overtly
236 C. E. Travis
marking the transition to direct speech (see Bauhr, 1994: 112; Gregori Signes, 1996: 164).
At the same time, bueno serves to contextualize the quote. In this environment, bueno is
interpreted as part of the quoted material and appears to “quote” one of the functions
discussed above. That is, it indicates that the upcoming material (the quote) was produced
as an acceptance, a preclosing, a mitigation of an upcoming comment, a reorientation, or a
correction in the conversation from which it is drawn (i.e., in the quoted conversation).
Bueno introducing direct speech therefore occurs with both continuing and final
intonation, as the intonation of the function being quoted determines the intonation with
which bueno occurs. In the following example, bueno occurs with continuing intonation
and appears to be playing the role of mitigating an upcoming response. Here, Santi is
describing the fact that his wife’s cousin talks about others behind their backs and that he
draws the conclusion on this basis that she must also talk about him behind his back.
(12) S: Y yo digo,
and I say-1SG
bueno,
si es así,
if be-3SG like.this
.. Entonces,
then
.. de mí también tiene que ser así.
of me also have-3SG COMP be-INF like.this
SANTI: And I say, bueno, if she’s like that, then she must also be like that about
me.
restaurant (574–578)
It is interesting to note that other Spanish markers commonly occur introducing direct
speech, in particular pues ‘well, then’ (Travis, 2005) and no ‘no’. Redeker (1990: 374) has
noted that English discourse markers are also commonly used in this way. This may be a
rhetorical device used to help contextualize direct speech and deserves further attention.
The above discussion has illustrated the multifunctional nature of bueno. In the following
section, I will outline how these different uses can be accounted for in a semantic model.
4. Model
In this section, I will show how the meanings of bueno map onto the discourse
environments in which it occurs and the functions it has. The aim of the semantic analysis
is to determine if any elements of meaning are shared across the range of use of a given
discourse marker, and if so, what those shared elements are; whether they relate to the
lexical source; and whether additional elements of meaning are encoded in the different
functions.
From the discussion of the different uses of bueno presented above, it is apparent that
these uses cannot be accounted for by just one definition but that they do not require
entirely distinct definitions either, since shared across the range of use is a notion of
positive evaluation of an aspect of the prior discourse. We have seen bueno used to
encode acceptance, to move towards a conversational closing by accepting the
conversation up to that point, to mitigate an upcoming utterance (which it does by

encoding partial acceptance), to mark reorientation in topic (which it does by accepting
what preceded and thereby paving the way to move on), and to introduce a correction
(where it functions to acknowledge partial validity of the corrected item). Thus, these
functions represent a set of related meanings centered around the notion of positive
evaluation. We can therefore analyze bueno as polysemous, with the notion of positive
evaluation as its semantic core or partial semantic invariant. This notion can be captured
in NSM with the following component:
I say: “this is good”
In this way we can see that the marker bueno contains the adjective bueno ‘good’ as part
of its meaning. Thus, the semantics of the lexical source have been retained in the
semantics of the discourse marker, but with the interactional and contextualizing role of
the marker, its meaning has extended beyond that of the adjective. In this sense, the
marker and the adjective can be seen to be polysemous, sharing the notion of ‘good’ but
differing in the discourse role each plays.
I propose that the functions of the discourse marker bueno discussed here represent four
related meanings, which are defined and discussed further below. These include its use to
encode acceptance, including acceptance of the conversation in general to move towards a
closing (which I shall term bueno1); to mitigate a response (bueno2); to mark a
reorientation (bueno3); and to mark a correction (bueno4). As discussed above, its
occurrence with direct speech is best treated as a subfunction of these others, where it
serves to contextualize the quote, marking it as an acceptance, a response, a reorientation,
or a correction.
The four definitions I propose for bueno are given below. Note that all definitions begin
with information about the material to which bueno responds. Variations of the basic form
presented above (“someone said something”) are given where further specification of the
environment is needed. This will be discussed further below.
bueno1 (acceptance, preclosing)

1. You said something to me now
2. I think that you want me to say something now
3. I say: “this is good”
bueno2 (mitigation)
1. You said something to me now
2. I know that you want me to say something now
3. I say: “this is good,
4. I want to say something more about this”
bueno3 (reorientation)
1. Someone here said something
3. Someone here can say something else now”
238 C. E. Travis
bueno4 (correction)
1. I said something now
2. I think that someone can say: “I don’t think the same”
4. I want to say something more about this now”
In each use, bueno responds to some aspect of the prior discourse, but the uses differ in
terms of whose speech it responds to. When encoding acceptance and mitigating a
response, bueno occurs in response to an immediately prior comment by another
interlocutor (therefore, “you said something to me now”). Furthermore, in these functions,
bueno indicates that the speaker either thinks that the addressee expects a response (for
bueno1) or knows they expect a response (for bueno2), which captures well the use of this
marker to preface answers to questions.
When marking a reorientation, it can respond to the same speaker’s speech, as seen in
(10), or to someone else’s speech, as seen in (8). The crucial point is that it marks a
response to a comment from a discourse participant and therefore is defined as responding
to “someone here”. And when introducing a correction, it always responds to the same
speaker’s immediately prior speech, and therefore is defined as “I said something now”.
All uses of bueno encode the notion “I say: this is good”. In the case of bueno1, no further
information is given: the speaker merely accepts what has been said, without indicating
anything more. In all other cases, bueno indicates that something more needs to be said,
which often gives rise to the implication that the acceptance is only partial. Bueno2
indicates that the speaker wishes to say something more about what has been said, which
may be a modification of a prior comment or the presentation of further information
before going on to answer a question. In this sense, it functions as a mitigatory device.
Bueno3 does not indicate that the same speaker will continue but that the discourse can
continue (by the same speaker or by another discourse participant) because of what has
been established, and it allows for the conversation to move in a new direction. For this
reason I have specified that the speaker says that someone can say “something else” as
opposed to “something more” (as was proposed for bueno2). Bueno4 includes an
additional component which involves recognition of a possible objection to what the
speaker has just said and the desire to go on and say something more in response to this,
that is, to modify their preceding utterance.
It is important to note that these meanings correspond to the intonation contour with
which bueno occurs: when the speaker wishes to say something more, bueno occurs with
continuing intonation, and when they wish to indicate that they have completed what they
want to say, it occurs with final intonation. This does not mean that the meaning can be
attributed to the intonation contour alone but that we have an interaction of different
linguistic features giving rise to different meanings. Note that a similar notion is proposed
by Yang (this volume) and by Ferrara (1997) in her study of the prosody of anyway.
These definitions demonstrate that despite the wide range of contexts in which bueno
occurs and the many functions it carries, its meaning in each of these environments can be
accounted for. The definitions may not be fully predictive, due to the syntactically
optional nature of discourse markers, but they do show which meaning is being employed
in any one context, and they capture precisely the meaning encoded by bueno each time it
is used.
This paper has discussed the pragmatic role discourse markers play in conversation.
Assigning markers a pragmatic role, however, does not mean that they must be treated as
having purely pragmatic meaning. I have attempted to show that while a discourse-
functional analysis based on identifying environments of occurrence and corresponding
functions is highly insightful into discourse marker use, a semantic analysis is essential
for a full understanding of the meanings motivating the functional role. I have argued that
the semantics and pragmatics of discourse markers can be distinguished in the same way
they are distinguished for other lexical items. And I have argued on this basis that
discourse markers can be defined with a semantic methodology of wide application and
do not require any specific semantic treatment. This allows us to account for marker and
nonmarker uses within the same model and thereby allows direct comparisons to be
drawn between such uses. This is particularly important when considering the
development of discourse markers, given that they tend to develop as extensions from the
lexical meaning. In relation to this, I have also discussed in some detail the notion of
polysemy and how it can be rigorously accounted for in a semantic model through the
notion of the partial semantic invariant. In the case of bueno, I have argued that the partial
semantic invariant is “I say: this is good”, and that this component of meaning has been
retained from the lexical meaning of bueno ‘good’ from which the marker derives.10
Much remains to be done in the study of discourse markers. This paper has addressed
some of the key issues of the field, namely, what markers are, how to account for their
meaning, and how to deal with their polysemous and polyfunctional relations. It is hoped
that this contribution, with the other papers in this volume, will help heighten our
understanding of discourse markers and pave the way for more unified research in this
area.
Appendix A: Proposed semantic primes (Goddard and Wierzbicka, 2002: 14)
Substantives: I, you, someone, people/person, something/thing, body

Determiners: this, the same, other
Quantifiers: one, two, some, all, many/much
Evaluators: good, bad
Descriptors: big, small
Mental predicates: think, know, want, feel, see, hear
Speech: say, word, true
Actions, events, movement: do, happen, move
Existence and possession: there is, have
Life and death: live, die
Time: when/time, now, before, after, a long time, a short time,
for some time
Space: where/place, here, above, below, far, near, side, inside
“Logical” concepts: not, maybe, can, because, if
240 C. E. Travis
Intensifier, Augmentor: very, more

Taxonomy, partonomy: kind of, part of
Similarity: like
Appendix B: Transcription conventions (Du Bois et al., 1992)
. final intonation contour .. short pause (about 0.5 secs)

, continuing intonation contour ... medium pause (> 0.7 secs)
? appeal intonation contour . . . (N) long pause (of N seconds)
-- truncated intonation contour @ one syllable of laughter
= lengthened syllable <@ @> speech while laughing
Appendix C: Gloss-line abbreviations
NB: all verbs are in the present indicative, unless otherwise indicated.
COMP complementizer INF infinitive 1 1st person

DAT dative pronoun PART participle 2 2nd person
DIM diminutive PREP preposition (a) 3 3rd person
GER gerund PRET preterite SG singular
IMP imperative REFL reflexive PL plural
IMPF imperfect SJV subjunctive
Notes
I would like to thank Alan Baxter, Hilary Chappell, and Timothy Curnow for their
insightful comments on an earlier version of the analysis presented here; María Elena
Rendón and Marianne Dieck for their help with data collection; and Kerstin Fischer for
her valuable feedback on this paper.
1
See Yang, this volume, for a more detailed account of discourse marker prosody.
2
For more details about the corpus, see Travis (2005).
3
The tag question ¿no? used here and in the intonation unit below would also be
classified as a discourse marker according to this analysis.
4
The information given here represents the name of the conversation from which the
example is drawn and the line numbers of the excerpt.
5
Ve is a discourse marker which literally means ‘look’ and is used as an attention-getting
device.
6
See Travis (1998, 2005) for more detailed analyses and discussion of bueno, though it
should be noted that the analysis presented in the 1998 work differs slightly from that
presented in these more recent works.
7
Other material that can be used to indicate reluctance are lexical items, pauses, tone of
voice, etc.
8
It should be noted that those uses I have specified as occurring with final intonation may
occur with continuing intonation in some marked contexts, for example when followed
by a vocative (e.g., Bueno, mi amor. ‘Bueno, my love.’).
9
Eso ‘that’ here refers to another element of the policy that Sara had been describing
when Angela asked her about the cost, and not to the cost itself.
10
For discussion of the relationship between the different uses of bueno, see Ocampo
(2002).
Chapter 15
Integrating prosodic and contextual cues in the

interpretation of discourse markers
Li-chiung Yang
1. Introduction
The importance of discourse markers arises because of their ability to provide immediate
indicators of the coherence of a communicative exchange in an effective and efficient way.
A key focus of the current volume is on how the meaning and functions of discourse
markers are conveyed through structures of syntax, semantics, pragmatics, and discourse
context. As both prosody and discourse markers act to mediate relationships of discourse
on each of these different structural levels, they share the common function of bringing
coherence to communication, and this shared characteristic provides a strong motivation
for exploring the link between prosodic form and discourse marker meaning. Prosody is
critical in spoken language because it provides an additional dimension that
communicates subtle and finely differentiated layers of meanings, and in some cases,
prosody is the sole cue for disambiguating discourse structure and discourse interpretation.
In this chapter I report on my on-going research on prosody and discourse markers,

focusing on how the prosody of discourse markers reflects cognitive and discourse
phenomena of uncertainty and certainty, intensity of emotional response, and interactive
signals of knowledge state. By examining the variability of form and function of prosody
and discourse markers, I hope in the current chapter to contribute to an understanding of
how levels of linguistic structure and discourse context work together with prosody to
provide a more complete characterization of discourse markers and their role in discourse.
1.1. Approach
In this chapter, I present my results on prosody and discourse markers based on a corpus
of spontaneous conversations. I show how prosody contributes to the communication and
interpretation of the multilevel meanings expressed in discourse markers and suggest how
the prosody of discourse markers fits within an overall system of discourse. My approach
is to abstract both from the surface shapes and from knowledge of the specific emotional
and cognitive processes the specific discourse interactions and the topic structure, to form
a theory linking the surface prosodic shapes to these other contextual or discourse
266 L.-C. Yang
elements. My approach differs from previous research in that I take an integrated

approach of combining detailed qualitative discourse analysis with a quantitative corpus-
based approach, utilizing both acoustic and discourse data.
1.2. Data and methodology
My data consist of two sets of spontaneous conversations in Mandarin Chinese that I

collected and annotated. There were two participants in each conversation, for a total of
four speakers and six hours of speech. Speech data were digitized using ESPS Waves+
software. These conversational data were analyzed and annotated for discourse relations,
topic structure, discourse markers, and speaker turns, and discourse markers that were of
interest were extracted. The acoustic measurements of pitch (f0), amplitude, and duration
were then correlated with the specific characteristics of the discourse markers extracted
from the corpus and examined within the larger discourse context.
Table 1 gives the total numbers of instances for each of the following discourse markers
tabulated in my data.
Table 1. Discourse markers extracted from our conversational corpus
Marker dui oh ranhou wa ey umhum um

Data1 368 148 92 4 17 385 311
Data2 170 133 70 12 13 115 54
Total 538 281 162 16 30 500 365
1.3. Data Presentation
We present our data in both textual and graphical forms because of the importance of
prosodic shape to function. Presenting data visually and encapsulating the tokens in one
frame also gives us a useful contextual frame of reference for interpreting contrasts and
generalizing patterns in the data, as well as showing a particular token’s place in the total
system.
The results presented are based on a large number of instances in different discourse
situations and derived from a very detailed analysis based on text, discourse context, and
acoustic measures of prosody. Therefore, our case-by-case analysis of each instance will
not be presented in every case in this chapter, but rather, our focus will be on synthesis of
our analysis to show the comparisons and major subtypes of situations encountered, so
that a systematic framework showing the role of prosody in explaining the functional
spectrum of discourse markers can be presented.
Enumeration of discourse markers is very rich, and a great amount of progress has been
accomplished in exploring and in extensive description of the functions of discourse
Integrating prosodic and contextual cues 267
markers and discourse particles by researchers in the field (e.g., Schiffrin, 1987a; Chafe,
1992; Redeker, 1991; Fischer, 1998; Aijmer, 2002; among others). On the other hand, the
prosodic functional correlates of discourse markers are not well-known or categorized,
and discourse markers are underutilized in automatic speech processing, where their
relationship signaling function shows great promise. In addition, the contribution of
prosody to the differentiation of multifunctional discourse markers has not been estimated
in terms of the relative contribution as compared with lexical meaning, syntax, and
context. A key question in a prosodic approach to these problems is, what role does
prosody play in distinguishing the lexical homophones and how does prosody map
different discourse markers to the same function? In addressing these problems, it is
important to take a multidisciplinary viewpoint, as valuable information has been obtained
from many different approaches, and to incorporate the best elements from each in an
integrated optimal way.
2. Definition: what are discourse markers?
2.1. Definition of discourse markers
In discourse, topic focus and the interest of participants are constantly changing, and
many aspects of speech are involved in cooperatively coordinating the flow of topics and
interests between participants. The complex topic relationships and the ongoing dynamic
interactions contribute to a unique rapidly changing mix of patterns that are commonly
signaled by discourse markers at key junctures.
Discourse markers are words or phrases which are commonly used to signal the
relationship between discourse units. They signal the relative status of discourse units by
linking together phrases in a characteristic way. They include connectives and adverbials
such as so, because, and, but, or, then, anyway, and now; feedback acknowledgments
such as yes, yeah, umhum, and right; as well as particles such as ah, oh, well, hmm, each
marking the particular status of an utterance in the discourse (Schiffrin, 1987a; Hirschberg
and Litman, 1992; Heeman, 1999; Fischer, 2000a).
Schiffrin (1987a: 31) defines discourse markers operationally as “sequentially dependent

elements which bracket units of talk” and theoretically as “members of a functional class
of verbal and nonverbal devices which provide contextual coordinates for ongoing talk”
(1987a: 326). In this paper, we suggest the following definition of discourse markers:
Discourse markers are short lexical items that signal the relationship between utterances
or the status of utterances and that bring coherence to sequential utterances in terms of
participants’ exchange of information states and judgments of new information.
We believe that using the term discourse particle to refer to particle or mainly sentence-
final particles (as proposed by Gupta) is too restrictive because it would exclude a whole
class of items that share a certain set of characteristic features, e.g., a phrase-linking
function, and also would exclude many languages that do not have sentence-final particles
but instead rely on other features that are critical to discourse structure and not just
syntactic structure. In our view, the class under study must be delineated on primarily
functional grounds, as different languages provide contradictory criteria for the language-
268 L.-C. Yang
specific class of particles, e.g., final particles of Chinese and Japanese vs. (initial)
particles of Western languages. Although restricting the term discourse particle to
“morphemes belonging to a specific word class” (see Gupta, this volume) may be
grammatically clearer, we would lose sight of many interesting phenomena which are
conceptually unified and which characterize important properties of discourse structure.
In traditional Chinese grammar, particles and interjections are treated as separate

categories: particles are always bound and always occur in phrase- or sentence-final
position, whereas interjections are always free, occurring in phrase- or sentence-initial
position (Chao, 1968). Discourse particles in the usual sense are not a subclass of the
particle class in Chinese and in fact are often loosely associated with the items which fall
under the term discourse marker. In Fischer (2000a: 15), particles are seen as the
superclass containing discourse particles and modal particles, with segmentation markers,
hesitation markers, and interjections as subclasses of discourse particles. While this may
be valid for Western languages, it creates some confusion for Chinese and other Asian
languages, as interjections and particles are separate nonoverlapping classes in Chinese
and Japanese. With these considerations, we therefore choose to use the less ambiguous
term discourse marker in this chapter, with particles (including initial, medial, and final
positions) and interjections as a subset in that class.
In our definition, we try to capture Schiffrin’s sequential dependency by the idea of

relationship or status and the bracketing of units of talk by the idea that these relationships
or status indicators apply between units of a discourse, and we emphasize the linking of
phrases through interactive signals of relationship and interested or presumed status in the
discourse environment. We also feel that the interactive and communicative nature of the
item should be part of the definition, as we view the function of discourse markers as a
running commentary on the underlying text that helps to make the text coherent in the
context of the particular conversation, i.e., a context which takes into account the
knowledge states and reactions of the participants in that specific conversation.
2.2. Criteria
What elements of language can be used as discourse markers and how do we decide if an
item is or is not a discourse marker? Building on Schiffrin (1987a: 328), we propose the
following characteristics of the word class that can be used as inclusion criteria for the
class of discourse markers (for more extensive discussions of this issue, see Fraser, this
volume, and Fischer, this volume):
1. An item in this word class signals a relationship or status between units of a

discourse.
2. The item is syntactically independent from the utterances; this facilitates a
predominantly discourse function for the item.
3. If it is content-full, the item should have status or relational lexical meaning, e.g.,
then lexically relates two units in a sequential or causative relationship meaning.
4. If the item is lexically less constrained, then it must be able to take on status or
relational meaning. For example, oh has limited inherent lexical meaning, but its
discourse meaning is derived from the context and prosody.
5. The item brings salience and coherence to sequential utterances in terms of
participants’ information states and judgments of new information.
6. It brings coherence to discourse over both local and global scope, i.e., the overall
scope of the coherence brought about is not necessarily restricted to immediately
adjacent units.
7. The item integrates interactive, expressive, cognitive, and transmittal of
information functions.
8. Differentiation of the status or relational meaning arises from the syntactic,
contextual, discourse, and prosodic environment.
3. Functional spectrum of discourse markers
3.1. Theoretical discussions
What makes discourse markers have the range of functions that they do? And what are the
characteristics that allow discourse markers as a class to function as relational markers
that bring coherence to discourse? I think that a good way to account for disambiguation
of many possible readings of a particular discourse marker in a given context is the idea
that the positioning of discourse markers in a multitiered grid of discourse relationships of
exchange, action, ideas, participation, and information gives a specific relational status
(Schiffrin, 1987a). Schiffrin proposes that the specific position of the discourse marker in
the contextual coordinate space allows it to take on different functions. It is this
positioning that gives uniqueness to the function when it actually appears in a discourse.
On the whole, we agree that this is a very useful criterion for accounting for the ability of
discourse markers to disambiguate meanings.
In addition, discourse markers also embody an expressive element that further refines the
relationship or status marking, and this expressive element is commonly mirrored in the
prosody of the discourse marker. This expressive element is not independent of the
interactive space, idea, etc. but anchors the discourse context in the speaker’s own
viewpoint. For example, in our data, the marker dui ‘right, yeah’ almost always signals
agreement with a proposition presented by the other speaker, but the speaker’s different
degrees of agreement, such as quick acknowledgement, sympathy, or enthusiastic,
emphatic approval are distinguished predominantly by changes in the prosodic shape of
the discourse marker, with emphatic approval having greater pitch height and movement,
while simple acknowledgment has moderate pitch height and light pitch fall.
In this work, we concentrate on three discourse markers: ranhou, dui, and oh, each
highlighting a set of specific functions: ranhou ‘then’ provides coherence through phrase
linking, dui ‘right’ expresses agreement and support, and oh is an interjection marking a
range of uncertainty-based states.
3.2. Multilevel meaning of discourse markers: how are they disambiguated?
3.2.1. Contextual linking of phrasal meaning: the marker ranhou ‘then’

Analysis of our data shows that discourse markers such as ranhou ‘then’ perform at least
three principle functions. They signal the flow of topic by acting as indicators of
relationships between phrases and subtopics. They also simultaneously serve as
270 L.-C. Yang
interactional conversational devices which are used to convey information about the
mutual knowledge of events held by participants and the presumed or intended status to
be given a discourse description. In this way, they guide the conversation along so it can
proceed in a more unified, smooth manner. Thirdly, because they occur at critical
transition points of speaker intention and reaction, they often function as expressions of
cognitive-affective states. Because of these considerations, discourse markers tend to be
highly expressive and therefore carry a lot of intonation. The prosodic patterns of ranhou
play a key role in distinguishing cognitive planning processes such as certainty and
uncertainty and in signaling topic direction.
In our corpus, the discourse marker ranhou ‘then’ was used quite frequently (occurring
162 times) and often occurs in clusters, especially in more narrative sections, to connect
both temporal sequences and event sequences, i.e., narratives which follow a natural or
logical development. Our data show that ranhou with a larger pitch range tends to signal a
change in topic or a return to a previous topic after an intervening subtopic. In contrast, a
narrow pitch range for ranhou usually indicates continuing topic development from an
immediately preceding phrase, and in these cases, ranhou often has a more gradual and
smoother contour, as well. The following shows the discourse text for the corresponding
ranhous (y9-y11) of Fig. 1, which exemplifies these points.
(1) [Speaker Y was talking to a friend about her research experience at a Lab]
Y: jiu shi nayang. It’s just like that.

y9 Ranhou women jiu qing ren lai Then we just asked people to come
luyin ma in to record
y10 ranhou luyinle yihou. then after the recording.
y11 ranhou jiu ba nage luyindai – then we just took the tape –
Oh wo xiang xiang kan Oh let me think
ta shi shi zenmeyang? what happened then?
Fanzheng fangzheng — In any case, in any case —
wo wang le zenmeyang, I forget what happened,
In y9, ranhou has a well-defined rise-fall pitch pattern with a large pitch range, making it
perceptually prominent; this prominence signals the break from the immediately
preceding topic and effectively returns the dialogue to a previous topic. Conversely, the
following ranhou in y10 introduces a phrase that is a natural continuation of the preceding
phrase, and this is reflected in the moderate pitch height and narrow pitch range of this
expression. In these cases, the degree of certainty and uncertainty also has a critical effect
on pitch shape: the slightly rising contour in y10 expresses both continuation and some
uncertainty, as the speaker tries to recall information. By the following phrase, y11, the
speaker has successfully retrieved one relevant piece of information, and is more
confident of what to say. This is reflected in the downward pitch slopes in ranhou,
although the level ending suggests that some uncertainty is still present.
Figures 1 and 2. Shape variations of the discourse marker ranhou ‘then’ of Speaker Y in
different contexts showing topic marking, floor negotiation, uncertainty, and emphasis.
Ranhou 9-11 - Y Ranhou 12-15 -Y
400 400
350 350
300 300
y 12
y9
250 250 y 13
y 10
y 14
200 200
y 11 y 15
150 150
100 100
50 50
1 11 21 31 41 1 11 21 31 41
Ranhou also acts as signal to control floor-negotiations such as floor holding and turn
taking and as a strategy to gain time to recall or organize what to say next, particularly
under the conditions of uncertainty which are inherent in discourse interactions. Fig. 2
presents paired instances of ranhou occurring in such interactive competitive floor
negotiation reactions between participants, with salient attention markers and subsequent
lower pitched repetition in each pair.
(2) [Speaker Y continues to elaborate on the above topic and reacts to potential
interruptions]
W: Oh, zheyangzi Oh, is that so

Y: Dui. Right
y12-13 Ranhou ranhou women zaiyibu Then then at the next step it’s already
lai de shihou jiushi yijing digitized
digitized || umhum || umhum
Y: jiu yijing cun zai diannao it’s already stored on the computer
limian le.
W: umhum Umhum
W: Wo zhe ci hui – This time I went back –
y14-15 Y: Ranhou ranhou ni jiu Then then you can just -
ni jiu keyi qu — you can just go —
In y12 and y13, one speaker tries to interrupt, and the main speaker immediately reacts by
repeating ranhou two times with a high pitch level, loud amplitude, longer duration and
expanded pitch range, with a systematic pitch lowering of about 50Hz between the first
and second instances.1 A similar pattern was repeated in y14 and y15. The drop in pitch
that occurs between instances in both pairs of ranhou represents the resolution of
uncertainty and the normalization from the initial immediate reaction.
272 L.-C. Yang
(3) [Speaker W was impressed by the speech demo she had seen]
W: Ta hui you – you yige biao rang ni He’ll have – have a list for you to
nian read
Y: Dui dui dui Right right right
w4 W: Ranhou ni nian wuan zhihou, then after you’ve read it
w5 ranhou ni jiu keyi shuo: then you can just say:
Wo – shi – Zhong – guo – ren I – am – a – Chinese
Y: umhum Umhum
w6 W: ranhou ni jiushi duizhe maigefeng then you can just talk into the
jiang: wo microphone and say: I
Y: umhum Umhum
W: t - ta nage nage diannao de yingmu The word ‘I’ will show up on the -
shang |jiu chuxian – wo the computer screen
Y: |Jiu hui da chu zhongwen de The same Chinese character will
nage zi, umhum just show up, umhum
W: Zhongwen de wo the Chinese character for ‘I’
Y: Dui, umhum Right, umhum
W: Dui Right
w7 ranhou ni zai nian ‘shi’ then when you say ‘am’
ta jiu hui tachu ‘shi ‘ the character for ‘am’ will show up
Y: umhum umhum umhum umhum
W: Zhong – guo – ren A – Chinese
For Speaker W, ranhou also acted as a phrase-to-phrase link that reflected topic
development, degree of certainty, floor negotiations, and emphasis. As with Speaker Y,
less prominent shapes of ranhou occur when the linked phrases are closely connected or
when the succeeding phrase develops within the same topic point. Speaker W’s ranhous
in w4-7 in Fig. 3, for example, are all short in duration, moderate in pitch height, and
generally have a well-defined rise-fall shape. These shapes reflect the very orderly step-
by-step succession within the current topic in this section. The sequential pairs w4-5 and
w6-7 also introduce two different specific points within the topic, and the pitch height
pattern for this cluster of ranhous reflects that relationship: the ranhou of w4 introduces
the new subtopic (that describes the process of inputting live data into the system) and has
the highest pitch level for ranhou in this cluster. W5 follows up on the same point, and
ranhou occurs at a lower pitch level than w4. W6 and w7 repeat and expand on the
sequence of activities involved, with w7 acting as the next logical step in the process. In
each pair, the initial part of the first instance of the pair rises more steeply than the second,
and each pair follows the same overall pitch lowering within topic idea.
By comparison, the pitch contours of w2, w8, and w13 in Fig. 4 are very long, with
striking shapes that highlight the contrastive emphasis that ranhou is used for in these
instances. In each case, the duration is much longer than in w4-7, and the pitch changes
are correspondingly much greater. The duration and pitch changes in these cases present a
very prominent shape that emphasizes the distinctiveness of the subsequent phrase content:
in both w2 and w8 the prominent signal emphasizes a point of contrast, i.e., “output” as
opposed to “input”, while in w8, the speaker is simultaneously taking time to reflect, and
the strong emphasis is combined with hesitation, resulting in the exceptionally long
duration and ending rise in pitch. In w13, the similarly prominent shaped contrastive
emphasis is used to attract attention to a turn in topic.
Figures 3 and 4. Shape variations of ranhou ‘then’ for Speaker W. The short moderate
shapes in Figure 3 show orderly topic succession, while the striking extended forms
highlight the contrastive emphasis in Figure 4.
Ranhou 4-7 - W Ranhou 2, 8, 13 - W
350 350
300 300
250 250
w4
w2
w5
200 200 w8
w6
w13
w7
150 150
100 100
50 50
1 11 21 31 41 1 11 21 31 41 51 61 71
(4) [In these sections Speaker W is talking about the demos she had seen]
W: Ranhou ta jiushi eh – demonstrate Then he just – eh demonstrates the

zhege speech recognition speech recognition system
Y: umhum umhum umhum umhum umhum, umhum
W: Ta jiushi nege voicing input || Y: It uses like voicing input || Y:
ummm? ummm?
w2 W: ranhou – Chinese character output then – get Chinese character output
Y: umhum umhum
W: jiushi zheyangzi || Y: umhum it’s just like that || Y: umhum
(5) [Speaker W continues her description of another demo shown at the conference]
W: Tade nage machine translation shi His machine translation system is

nage eh . . . Chinese character input that – eh . . .Chinese character input
Y: umhum Umhum
w8 W: ranhou shi English output || Y: then you get English output || Y:
umhum umhum
shi paraphrase || Y: umhum It’s paraphrase || Y: umhum
shi zheyangzi de it’s just like that
274 L.-C. Yang
(6) [In a later section, the topic changed to a new book that Speaker W had just
translated]
W: Ta nege shi yige jiaopai – shi He belongs to a sect – it’s

dasheng fuojiao Mahayana Buddhism
|| Y: oh! uhuh uhuh uhuh uhuh || Y: oh! uhuh uhuh uhhuh uhuh
shi cong Yindu laide dasheng it came from India, it’s one of the
fuojiao de yipai Mahayana Buddhism sects
w13 W: Ranhou jiushuo eh – yao nexie Then it’s just eh – to make those
heshang – monks –
jiushi nengguou zuodao nexie just so that they can do like those
hoaxiang jingsi de nexie gongfu monk meditation practices
3.2.2. Cognitive status, understanding, and degrees of agreement: the marker dui
In discourse, speakers often explicitly search for some indication of mutual agreement and
understanding, and the hearer provides corresponding feedback or backchannels as
expressions of understanding and interest. Both the speaker’s prompting and the hearer’s
feedback are often done through the use of specific discourse markers, such as um,
umhum, uhuh, yeah, and right, which are associated with specific pragmatic distinctions,
marking new or old information, certainty and uncertainty, degree of comprehension, and
emotional response.2 Such markers perform a specific signaling and monitoring function
in the cooperative give-and-take of conversation, and signal different nuances of
participants’ intentions, as well as information on their reactions to the conversation flow.
One of the most important feedback markers in Mandarin discourse is the word dui,
meaning ‘right’ or ‘correct’ (dui occurred 538 times in the data we analyzed, followed by
umhum which occurred 500 times). In conversation, dui frequently acts as a signifier of
agreement and support, showing the speaker’s approval, judgment, and interest in the
current topic. The expressive nuances that dui assumes are directly related to the specific
interactive context, emotional intensity, and the simultaneous expression of other
emotions that are consistent with dui’s underlying function.3 Dui is the most frequently
occurring feedback word in our conversational corpus because it functions both as a signal
of cooperation and as a marker of information receipt, two functions especially critical to
a successful discourse.
Our data show that the speaker’s cognitive status and level of involvement are commonly
manifested in intonation.4 Looking at all the duis on our plots (Figures 5 and 6), we can
see that although there is a general falling pattern, the slope of fourth-tone dui varies
greatly from very steep to nearly flat. There is a clear variation in shape depending upon
the degree of intensity and the definiteness of the agreement. The degree of emotional
intensity is signaled by pitch range and pitch height.5 As shown in Fig. 5 (as well as the
subplot Fig. 5a), in the consecutive responses w-dui8, w-dui9, and w-dui10, the speaker is
getting progressively more involved and her duis follow a corresponding progressively
higher pitch pattern, from the gentle agreement of w-dui8 to the extreme level of
emotional involvement and exaggerated emphasis of w-dui10. The intensity variation is
also systematically indicated by the uniform stepwise increments in pitch level. Similarly,
in Fig. 6, both w-dui22 and w-dui23 are intense expressions but w-dui23 has a larger pitch
range, exemplifying the speaker’s higher degree of emotional intensity. In contrast, w-
dui24, an immediate follow-up confirmation of w-dui23, has a more gradual convex shape
and a moderate pitch level and pitch range because of the speaker’s more normalized state,
as shown in the subsection plot in Fig. 6a.
Figures 5 and 6 . Thirty instances of emotional intensity and variations in expressions of
dui of one speaker in different pitch shapes indicating varying degrees of understanding
and involvement. The similar pattern of forms for dui in both figures from different
dialogue sections suggests the representativeness of the distribution.6
450 450
400 400
w-dui1
w-dui2 w-dui17
350 w-dui3 350 w-dui18
w-dui5 w-dui19
300 w-dui7 300 w-dui21
w-dui8 w-dui22
250 w-dui9 250 w-dui23
w-dui10 w-dui24
w-dui11 w-dui25
200 200
w-dui12 w-dui28
w-dui13 w-dui29
150 w-dui14 150 w-dui30
w-dui15 w-dui31
100 100
50 50
1 11 21 31 41 緱緱緱緹緱縈緱縑緱
(7) [Both speakers are talking about proposal writing in an involved manner]
W: Ayo wo juede sanxinliangyi zui Ayo I think being indecisive is the

buhao le worst thing
Y: Timu ding le zui hao bu yao zai After a topic’s decided, better not to
gai change it
w8 W: Dui Right
Y: Dui yinwei yao buran Right ’cause otherwise all the work
iangongjinqi yao zai congxin done is wasted and you have to start
kaishi over again
Y: buguo wo juede proposal ta ye but I think it’s not necessary to write
buyiding yao xie tai chang ma that long for the proposal
w9 W: Dui! Right!
Y: yinwei zhao xuexiao nege biaoge ’cause according to the form from
zhi you yidiandian school there’s only a little space
w10 W: DUI! um RIGHT! Um
(8) [The speakers are discussing the whereabouts of a friend]
Y: Ta ban le nege difang ta - tamen - She moved - that place she - they -
w23 W: Dui || Y: ban zou le Right || Y: moved
Y: Dui Right
276 L.-C. Yang
w4 W: Dui Right
Figures 5a and 5b: Shape variations of dui: subsectional plots of duis from Figure 5.
450 450
400 400
350 350
300 300
w8 w13
250 250
w9 w14
200 200
w10 w15
150 150
100 100
50 50
0 0
1 11 21 31 41 1 11 21 31 41
Figures 6a and 6b. Shape variations of dui: subsectional plots of duis from Figure 6.
450 450
400 400
350 350
300 300
250 w23 250 w30
200 w24 200 w31
150 150
100 100
50 50
0 0
1 11 21 31 41 1 11 21 31 41
Pitch shape characteristics such as pitch slope, concavity, and convexity are very
important features in distinguishing intonational meaning and cognitive states, with more
moderate slopes associated with more tentativeness, and sharper falling slopes with
greater definiteness. The concavity or convexity of slope is also critically related to the
perceived degree of harshness or softness of the utterance, and these shape characteristics
reflect the underlying expressive states that often arise from the discourse process itself.
For example, w-dui13, w-dui14, and w-dui15 (see both Fig. 5 and Fig. 5b) are shorter and
have sharper slopes than w-dui1 and w-dui2. In this sequence, the speaker first started to
express her opinion in w-dui13 but was interrupted, and in w-dui14 she restarts, so her
pitch level is higher. In w-dui15, the speaker was just providing further confirmation after
an explanation, so the dui here is low-pitched with a convex shape, corresponding to the
more gentle agreement. W-dui17, w-dui18, and w-dui19 show a similar pattern of
intensity variation, with a higher-pitched sharper slope dui perceived as more definite.
(9) [Speaker W is clarifying a point raised by Speaker Y]
Y: Oh ni gen ta – nege zuoguo Oh you were – you worked for him

shibushi? right?
w13 W: Dui wo – Right I –
Y: Ni zuoguo ta de nege TA shibushi? You worked for him as a TA, right?
w14 W: Dui wo – Right I –
w15 W: Dui wo – wo bang ta zuoguo Right I – I worked for him for a
yixiaoduan while
Y: Cengjin bang ta daiguo ho Helped him with the classes huh
W: Dui yixiaoduan shijian Right for a short while
(10) [The speakers are discussing some changes occurring in school policy]
Y: Xuexiao you gai le a? The school changed it?

W: Xuexiao you gai le jiushishuo – The school changed it, it’s just –
Y: You gai le a? bu xiwang shuo zai They changed it? They don’t want
zhao – you to ask –
w18 W: Dui (more definite) Right (more definite)
Y: tuixiu de laoshi retired professors
w19 W: Dui dui Right right
Y: Oh Oh
While most of the high pitched instances of dui in Figures 5 and 6 are intense and have a
concave shape, the remaining duis in the mid and low pitch ranges exhibit mostly convex
shapes. For example, w-dui2, w-dui3, w-dui5, and w-dui15 have a gradual pitch slope
with a convex shape, expressing the speaker’s sympathetic understanding and
confirmation. W-dui11, w-dui12 and w-dui31 are at the other extreme from harshness and
definiteness and have the flattest slopes of all the instances. In these cases, the speaker
reconfirms the hearer’s correct receipt of matter-of-fact information, and the neutral
quality of these expressions is represented in the insignificant pitch range, low amplitude
and pitch level, as well as the mild shape of these duis.
The contrast between a more definite agreement and a matter-of-fact reconfirmation is

signaled clearly by prosodic shape, captured in Fig. 6b. W-dui30 is a definite response to
a prompt and w-dui31 is a follow-up response where the speaker is repeating the whole
phrase to reconfirm a point. In contrast to w-dui30, which has a sharper slope and louder
amplitude, w-dui31 here is very flat and mild.
The pervasive use of feedback markers such as dui in our data and the wide range of
prosodic shapes that they can take on in the different conversational discourse contexts
presented here support the view that discourse is a cooperative process and that feedback
markers play an essential role in the mutually influencing interactions of discourse
participants, signaling cognitive, emotional, and cooperative information flow. At each
point in the conversation, the hearer’s utterances provide important clues about their own
evolving information states, their views of the topic under discussion, and the specific
278 L.-C. Yang
cognitive-emotional status, ensuring that the direction of the discourse flows in synchrony
with both speaker’s states.
3.2.3. Dimensions of meaning: surprise to dawning realization to acknowledgement—the

marker oh
Prosody expresses fine gradations in meaning even when lexical information is largely
absent, as in the case of the marker oh ‘oh’ and its variant ah ‘ah’, which frequently
function as an expression of cognitive reorientation (Schiffrin, 1987a; Heritage, 1984) and
communicates a range of uncertainty-based states, including doubt, surprise, acceptance,
acknowledgement, and registering of information. Upon encountering new information,
participants often have both a cognitive reaction to the unexpectedness of the event, as
well as an emotional reaction to the specific content of the new information. The specific
prosodic shape of oh is highly expressive in providing such types of information in an
immediate transparent way, and this gives oh great interactional significance as a
discourse marker.
Our view of the basic function of oh is in agreement with the viewpoint of Schiffrin
(1987a) and Heritage (1984, 1998) that oh expresses cognitive reorientation or
transformation, and that may be its basic or invariant meaning. We also agree with
Heritage (1984: 309) that the production of an oh marker is not necessarily a function of
the degree of unexpectedness of information, as in our data, oh expresses both simple
acknowledgment as well as great astonishment. What a prosodic viewpoint adds to this is
that it gives an account of the varying functions of oh by placing the concepts of
unexpectedness and uncertainty in an acoustic gradient framework. Thus, an instance of
oh that expresses simple acknowledgment will exhibit a prosodic shape that distinguishes
it from unexpectedness, surprise, and other gradiently related uncertainty-based
phenomena.
Three basic patterns for oh are evident in the plot containing 28 instances of oh of one
speaker in Fig. 7. (Oh occurred 281 times in our corpus. As the speech situations of ohs
presented here are very diverse, we include the discourse text for each at the end of this
chapter.) As shown, oh often expresses surprise in a rise-fall shape with an arched and
extended concave pattern communicating different intensities of dawning realization. It is
the differences in shape, height, and duration that communicate the degree of uncertainty
or certainty with respect to the speaker’s knowledge state, the intensity of emotion, and
the effects of other co-occurring emotions.
Figure 7. Expressive meaning of twenty-eight instances of the interjection oh of one

speaker (speaker S).
s-oh1
s-oh2
s-oh3
450
s-oh4
28 1 s-oh5
400 s-oh6
s-oh7
4 20 s-oh9
350 s-oh10
13
s-oh11
s-oh12
300
Frequency
s-oh13
s-oh14
18
250 s-oh15
22 11 s-oh16
6 s-oh17
200 s-oh18
10 s-oh19
17
150 s-oh20
s-oh21
s-oh22
100 s-oh23
s-oh24
s-oh25
50
s-oh26
Time s-oh27
s-oh28
Figure 8. Twenty-two instances of the expressive use of the interjection oh of another

speaker (speaker K).
400 ky1k1
ky1k2
ky1k3
350 ky1k4
ky1k5
ky1k6
300 ky1k7
ky1k8
Frequency
ky1k9
250
ky1k10
ky1k11
ky1k12
200
ky1k13
ky1k14
150 ky1k15
ky1k16
ky1k17
100 ky1k18
ky1k19
ky1k20
50 ky1k21
ky1k22
Time
280 L.-C. Yang
Our data show that the high degree of uncertainty inherent in intense surprise causes a
high rise in pitch, as in s-oh2 and s-oh5, whereas a sharper and narrower arch shape
indicates the presence of surprise with co-occurring emotions, as in the great amazement
of s-oh4 and the horror expressed in s-oh3. A lower pitch range often reflects acceptance
and registering of information, with a lesser degree of surprise, as in s-oh11, s-oh13, and
s-oh22, and a matter-of-fact acceptance of information that offers little challenge to the
speaker’s knowledge state causes the pattern of nearly flat pitch slopes in s-oh10 and s-
oh17. Emotions that are closely related to acceptance, such as sympathy and approval,
also tend to be expressed in a low pitch level.
S-oh1 and s-oh18 are at the other extreme of uncertainty, with rapid rises in pitch within a
short time frame exemplifying incomprehension, alertness, and a need for further
information, in contrast to the completely realized acceptance of information
accompanying more extended duration pitch shapes. The uncertainty in s-oh1 in particular
stands out because of the convex steep rise to at a high pitch level with nearly no
subsequent fall, reinforcing the final incomprehension, while the moderate pitch and
gradual rise of s-oh27 expresses the speaker’s doubt and heightened interest. By contrast,
concavity of pitch shape is associated with greater comprehension, as in s-12, s-24, and s-
28 which express surprise, interest, and quick recognition upon encountering unexpected
new information.7
The functions of an interjection are closely related to the nature of discourse and
individual speaking style. In contrast to the highly varying expressive ohs of the previous
speaker (speaker S), speaker K’s ohs in Fig. 8 seem much more subdued and are much
shorter on average, as they function mostly as quick acknowledgement or quick
recollection responses. Due to the particular nature of the discourse, there are many
instances of reorientation, recalling, acknowledging, and sudden occurrence of ideas by
the speaker, and this is why these particular shapes dominate, in contrast to the varied
reactions to new information experienced by speaker S.
It is worth noting that even within all these short expressions of ohs, there still exist finer
variations in shape, height, range, direction, duration, and intensity, and these variations
are systematic and are related to interpolation, status of information, and emotional state.
For example, there is the short falling type as in 3 and 16; slightly longer falling type as in
2, 6, 10, and 20; the curvy twist type as in 14, 17, and 22; the arch type as in 4, 9, and 11;
and the more intense emotional type, characterized by wider pitch movement and longer
duration, as seen in 1, 5, 7, and 18, for expressing the speaker’s disgust, sudden
remembrance of an important or exciting event, and appreciative acknowledgment,
respectively. How expressive a speaker is in a particular conversation also depends on the
degree of topic relevance. This is exemplified in the striking extended oh of 8 expressing
the speaker’s mock terror and protesting emotion on encountering an unexpected event.
4. A model of prosody and discourse markers
Because of the many questions still unanswered in the field, it is difficult to propose a
complete model for discourse markers and prosody. However, based on consideration of
the data from both analytical and experimental results of my research (Yang, 1995, 2001,
2005), my proposed model at present would be constructed to include the following
elements:
1. Identification of the elements of the multifunctionality of discourse markers, that

is, a classification of the major categories of functions performed by discourse
markers. These include at least the following:
a. Specific phrase relationships, as in the sense of “This is a consequence of

that” or “This is a restatement of that”, or “The following phrase negates the
preceding phrase”.
b. Functions of the discourse marker when considered in its interactive role or as
an attention marker, such as when signaling “Hey, this is important” or “This
is related to something we both know”, or “I’m saying something in terms of
what I know you already know”.
c. Specific cognitive relationship to the speaker, such as different degrees of
certainty or uncertainty held by the speaker with respect to what is being said,
or signaling of degree of newness of the information.
d. Specific emotional relationship of the phrase or phrases to the speaker, such as
angry, happy or unhappy, worried, etc.
2. Identification of the factors that lead to a correct interpretation of the discourse

marker. The model includes at least the following:
a. Discourse context. The discourse context refers to content of the conversation

thus far, knowledge states of the participants, etc.
b. Position in phrase.
c. Lexical meaning or conventional range of meaning usages for the specific
discourse marker.
d. Prosody, including pitch, amplitude, duration, and voice quality variations.
e. Prosodic context, i.e., the knowledge of how the prosodic form in question fits
into the prosodic context established in the conversation up to that point, as
well as familiarity with prosodic patterns of individual speakers.
3. Identification of the domain and extent of applicability of each of the factors in

item 2, that is, how much each factor in item 2 contributes to a correct
interpretation of each category of item 1.
Based upon analysis of our data so far, we postulate that discourse context, position in
phrase, lexical meaning, and prosody all contribute to the correct interpretation of item 1a,
the specific phrase relationship, and similarly for item 1b, but prosody plays a more
prominent role in the latter case. For both items 1c and 1d, the correct interpretation of the
specific cognitive or emotional relationship to the speaker, prosody plays the predominant
role in our model, although lexical meaning and context will also help, and the position in
phrase may also have an important effect.
The three main parts of the model are graphically shown in the following representation:
282 L.-C. Yang
Figure 9. Model representation of contextual determination of discourse marker meaning

in discourse. In this model, combinations of factors in the left column lead to the correct
disambiguation of discourse marker functionality (center column). Postulated proportions
of factors for each functional category comprise the right-hand column.
Interpretation Factors Elements of Multi-functionality Relative Contribution
2.a Discourse Context (DC)

1.a Phrase Relationship DC PP LM P
2.b Phrase Position (PP)

1.b Interactive Relationship DC PP LM P
2.c Lexical Meaning (LM) 1.c Cognitive Relationship DC PP LM P
1.d Emotional Relationship DC PP LM P

2.d Prosody (P)
2.e Prosodic Context (PC)
In this model, prosodic context works through prosody by helping to disambiguate the
meaning of the local prosodic form. We also connect the output boxes to represent that the
different functional elements often co-occur and are simultaneously expressed. In
positioning the expression and interpretation of discourse markers in the multitiered
relationships of the model, we expect to provide a more complete picture of the variety of
elements that make discourse markers so important in language.
5.1. Issues of speaker dependency and discourse context
One issue that inevitably comes up when dealing with large multispeaker corpora is the
issue of speaker independence. Our observation is that in some cases it can be difficult to
determine whether differences of prosodic shape reflect primarily speaker characteristics.
This mainly because a given speaker may assume different roles or “attitudes” in different
conversations according to the social or informational context, and these roles may persist
for some time in a given conversation. Therefore, it may be difficult to judge from given
speech samples whether they represent a speaker characteristic or a context role. In
practice however, our research so far has found more similarities than differences across
speakers, and this can be seen from our examples here as well. For instance, the patterns
of ranhous in Figures 1 through 4 show similar patterns for two different speakers.
We have found that the general pattern of ohs across all speakers in our data shows
similar patterns for similar degrees of emotional expressiveness. The contrast presented in
Figures 7 and 8 in this chapter were chosen specifically to illustrate the point of the
discourse context: it is the different roles with respect to information state taken by each
of the speakers in the particular conversation that is causing the differences in the general
pattern of oh contours. This could be misinterpreted as speaker variation if one didn’t take
this context into account. Only an adequate amount of data from many different speech
situations can determine the speaker-dependency of specific contours; however, our data
in general support the view that the effect of speaker dependency is of lesser extent as
compared to the similarities across speakers.
5.2. The interaction of tone and intonation
A fascinating aspect of discourse markers in Mandarin Chinese is the remarkable

interplay between the Chinese tonal system and prosodic expression that occurs
throughout spontaneous conversational speech. 8 How can prosody be used to
disambiguate the meanings of discourse markers, given the presence of the lexical tones
(final particles and interjections in Mandarin Chinese are traditionally designated to have
the status of neutral tone, i.e., toneless, whereas other discourse markers may have defined
lexical tones)? In my research (see Yang, 1995, for more information), I have found that
the characteristics of the particular tone are parameters that are actively used for
expressive effect. The degree of conformity to the lexically defined tonal shape is often
utilized for emphasis, and nonconformity of shape is usually the result of expressive
intensity. For example, dui ‘right’ has a fourth-tone falling lexical shape, and the majority
of the dui cases in my data preserve this falling contour, perhaps because of the synchrony
of shape with the falling-pitch emphasis used for agreeing that may occur in all languages.
The degree and contour of the fall, though, is very expressive of different states. Ranhou
‘then’ has a lexical rise-fall shape, and the production of a contour with a very
pronounced rise-fall shape usually indicates emphasis of some type. In the case of ranhou
used as a continuation marker, the overall shape often assumes a rising contour, is short in
duration, and is often interpolated as well, as shown previously. The same general
principles apply to cases of the other tones, but my research indicates that there is a
complicated but systematic accommodation and utilization of lexical tone for expressive
effect through prosody, exemplifying the view that expressive prosody is a crosslinguistic
phenomenon with a universal basis in cognition, emotion, and physiology.
5.3. Conclusion and general implications
As elements located at the natural intersection of several different but complimentary

relationship dimensions, discourse markers offer an intriguing and informative window
from which to view discourse as a whole, where relationships among concepts, speakers,
and referents are a predominant motivating force for discourse development.
Relationships are expressed throughout discourse, whether through explicit linking of
concepts to external referents, through differential weighting of words by emphasis, or by
signaling of expressive or phrasal relationships. In our research on discourse, we
consistently find that these relationships are expressed systematically by prosody.
Because of their brevity and lexically limited character, discourse markers represent
general prosodic principles in a crystalline and compact form, and as locus points for the
signaling and interpretation of multilevel relationships, discourse markers provide an
environment where the critical role and forcefulness of prosody in distinguishing and
communicating meaning is fully achieved.
284 L.-C. Yang
Appendix A: Discourse Texts
1. Discourse text for Speaker S’s ohs
K: zheshi lingwai yige difang nege This is a different place. It’s called
wenquan Bath.
s-oh1 S: oh? Oh?
K: zhe bushi Oxford zheshi Bath, This is not Oxford, it’s Bath,
B-A-T-H, nege difang jiao Bath B-A-T-H, that place is called Bath
s-oh2 S: oh! hen youming, dui dui dui Oh! Very famous. Right, right,
right.
K: ba nege ren na chulai ye he – They took the mummies out!
heihei xiang nege shenme xiang Black, black just like – like dried
nege rougan meat
s-oh3 S: oh-wa Oh-wa!
K: rougan (laugh) S: rengan Dried meat (laugh) || Dried people
K: dui, rengan. ta toufa dou hai you Right, dried people. The hair is
ye! still there!
S: nage – nega daxue zenmeyang, That – how was that university,

Oxford? Oxford?
K: Oxford hen hen piaoliang Oxford’s really beautiful
S: hen piaoliang ho Really beautiful huh
K: buguo ta limian you sanshi jige But it has more than 30 colleges
college inside
s-oh4 S: oh hen da ma ho Oh it’s really big huh
K: hen da Really big
K: ranhou zai University of Munich Then there’s more than one

you yibai duo ge zhongguo hundred Chinese students at
xuesheng University of Munich
S: wa hen duo ho Wow! That many!
K: yi liang bai ge One or two hundred
K: ta shuo you yiban shi dalu lai de She said half came from China
s-oh5 S: oh Oh
W: wo qu le zhihou cai zhidao Only after I went there did I find

Kengding haiyou yige jiushi – out there’s another one at Kenting
s-oh6 S: oh haiyou jie xialai de . . . Oh another one afterwards. . .
S: Computational de hen duo ma? Lots of talks on computational?

W: Yeah computational hen duo o Yeah, lots of computational stuff!
s-oh7 S: oh Oh
W: tingshuo tamen zhege fangzi shi I heard that they bought their
gen nege TD mai de house from TD
S: zhende a?! Really?!
W: shibushi gen ta mai de fanzheng Whether they really bought it
jiushi touguo ta de from her, anyway it’s through her
S: oh || W: zheyangzi Oh || That’s the case.
W: um um
s-oh9 S: oh zheyangzi a um Oh is that the case um
W: umhum umhum
W: wo – ta – wo tingshuo ta qu I – she – I heard that she went to

Princeton shibushi Princeton, isn’t it?
s-oh10 S: oh Oh
W: qu nali jiaoshu a She went there to teach
S: ta qu nabian jiaoshu a ? She went there to teach?
s-oh11 W: dui S: oh (ah) Yeah || Oh (ah)
W: wo jiushi wo kao wan nege oral I – it’s just after I took the oral
zhihou wo qu le xiaweiyi exam I went to Hawaii
S: oh ni qu xiaweiyi a oh oh you went to Hawaii oh
W: nege shihou wo meimei zai My sister was in Hawaii at that
xiaweiyi time
s-oh12 S: oh ni meimei zai xiaweiyi Oh your sister was in Hawaii
W: ta huiqu le She went back already
s-oh13 S: oh ta zai na nianshu ma? Oh she’s there studying?
W: ta zai tai – ta – ta yijing nian wan When she was in Tai – she already
shu le ta zai Taiwan yijing nadao finished her study. She already got
yige shuoshi le an MA in Taiwan
s-oh14 S: oh(ah) Oh(ah)
S: ta shi zhongyi haishi xiyi ne? Is it Chinese or western medicine?

W: ta shi shuyu nege – eh She studied – eh – something like
weishengjiaoyu zhilei de health education, that type
s-oh15 S: oh Oh
W: jiushi gonggong weisheng It’s just public health
S: umhum umhum umhum umhum
W: jiushi yiwu renyuan yao gen just like medical staff they have to
pirushuo zuo zhege gonggong go with the public health people,
weisheng zhege de ren yao gen they have to go, or the nurses have
zhe qu huoshi huzhi ye genzhe qu to go along
na ta shi shuyu community And she belongs to community
health, health
W: suoyi ta jiu gen zhe qu So she just went with them
s-oh16 S: oh (ah) Oh (ah)
W: zheyangzi That’s it
286 L.-C. Yang
S: meiyou la! buhui la. Ta ta shi No, you didn’t! Your sister’s the
dajie ma. Ta shi zuei da de oldest, isn’t that the case?
duibudui? bushi a? tashi No? she’s the 2nd oldest? The 2nd
laoer? Dierge? one?
s-oh17 oh umhum Oh umhum
W: yingwei wo shangci – wo Because last time I – last time

shangci qu wo jiuyuechu de when I went to Chinatown in the
shihou qu Chinatown shuaile beginning of September, I slipped
yijiao and fell
s-oh18 S: oh zhende a Oh really
W: um zhege difang shuaile yijiao um I fell and hurt (my ankle) here.
jieguo wo meiyou rou yinwei wo I didn’t massage it. I was afraid of
pa tong wo jiu meiyou rou the pain, so I didn’t rub it
W: wo yige gege haiyou wo meimei, One of my brothers and my sister

wo nege gege shi xue yi de, ta shi – my brother studied medicine,
shuyu xinzhang neike he’s a heart specialist
s-oh19 S: oh Oh
W: na wo meimei shi shuyu and my sister studied public health
gonggong weisheng de zheyangzi that’s the case
W: ta zai tai – ta – ta yijing nian wan When she was in Tai - she already
shu le. Ta zai Taiwan yijing finished her study in Taiwan. She
nadao yige shuoshi le got an MA already
S: oh(ah) Oh(ah)
W: ranhou zai chulai jinxiu then she came abroad to study
s-oh20 S: oh jinxiu yizhengzi Oh to study for a while
K: ranhou hao duo dagai jiqian ge Then many maybe thousands of

mummies ranhou zhengzhengpai mummies and rows of them
s-oh21 S: oh-wa queshi hen jinren Oh-wa! that’s really amazing
K: zhi neng da sige yue hai jige yue You can only work for 4 months
ne wo nege tongxue zai yige or a few months. And my friend
insurance company was with an insurance company
s-oh22 S: oh Oh
K: ta jiushi zhuo nege filing a she’s just doing some filing stuff
K: wo yige tongxue zai munihei One of my classmates is studying

dushu ta shuo women liangge zai in Munich, she said let’s meet in
London meet London
S: oint – oh zheyangzi a oh Oint – oh is that so. Oh.
K: dui yinwei ta ganghao you yige Right. Because she happened to
libai de vacation wo ye vacation have a 1-week vacation and I was
also on vacation
s-oh23 S: oh zheyang Oh is that so
K: ta shuo women gancui jiu yue zai She said let’s just arrange to meet
London Meet in London.
S: oh zhe jiushi ni de wenti a? D: Oh so that was your question? D:

dui (laugh) right
S: wo hai yiwei ni you shenme – I thought you had something –
D: bushi nege jiushi wo yiging No, that was the one I already
wen le asked.
s-oh24 S: oh ni wen le nege campus de Oh you asked about the campus
nege //D: dui //(right)
weishenme ni name haoqi? why are you so curious?
S: ey tingshuo tamen tushuguan ye Ey! I’ve heard that their library’s

bu bu kaifang ho ye buneng jinqu not open to the public.
kan de ye You can’t go in to look either.
K: shibushi keyibukeyi || tushuguan Whether it is – or can it – || The
zhiyou zhiyou kaifang waimian library is only only open to the
yidiandian public in the front.
s-oh25 S: oh zhiyou kan dao – oh Oh you can just see – oh
K: tushuguan wo you zhao I took some pictures of the library.
S: oh zheyang K: dui Oh is that so K: right
K: zhe tamen weiyide yige college - This is their only college -

s-oh26 S: oh keyi jinqu || K: King’s haishi Oh you can go in || King’s or
Queen’s Queen’s
K: oh zheshi Yingguo zuei da de Oh this is the biggest department

baihuo gongsi store in England
s-oh27 S: oh? ey, zenmeyang, Yingguo Oh ey how was it? how are the
nege dian zenmeyang? shops in England?
K: bi meiguo gui It’s more expensive than the U.S.

erqie wo ne shihou – wo gen wo Also, at that time I – Me and my
tongxue ding nege B&B be – bed friend reserved a B&B be – bed
and breakfast and breakfast
s-oh28 S: oh dui dui dui dui umhum Oh right right right right umhum
K: yinwei ta shuo gui de hen duo because she said there’s lots of
pianyi de hen nanzhao expensive ones, but really hard to
find cheap ones.
2. Discourse text for Speaker K’s ohs
/oh4/kyohk001
K: women jiu hua le yitian zai limian We spent the whole day inside
kan nege aiji de mummies looking at the Egyptian mummies
S: umhum umhum umhum umhum
K: oh hen ke – hai wo wanshang Oh terrible – I couldn’t get to sleep
sheibuzhao jiao ranhou zhende ba that night and they really put them
nege ren nachulai ye out!
/oh4/kyohk002
288 L.-C. Yang
K: hen duo ne zhengge dixiashi A lot! The whole basement – oh we

oh kan le yige xiawu kan buwan tai spent one afternoon looking, can’t
lei le finish – too tired
/oh4/kyohk003
S: jibenshang hai bu zhidao nege Cam – basically they didn’t even know
nege Cambridge dou meiyou yige whether Cam – whether Cambridge
campus has a campus or not
K: oh Cam – Cambridge wo meiyou qu Oh Cam – Cambridge I didn’t go
/oh4/kyohk004
S: xian nadao degree meiyou? nian Now did she get her degree yet? She’s
buoshi xuewei studying for her doctorate
K: umhum umhum
S: nian wenxue fangmian de Studying literature
K: oh buguo tamen nebian nian Ph.D. Oh but it takes you 10 years to get a
yiding yao shinian Ph.D. there
S: zhendea? Really?
/oh4/kyohk005
K: hen nanchi (laugh) erqie erqie you It tastes terrible! (laugh) also – also it
hen you hen gui was really expensive
oh nege shenme Deguo de nege Oh that – the McDonald’s in Germany
McDonald hen gui are very expensive
S: umhum umhum
/oh4/kyohk006
S: ey ne Deguo ne cai haobuhao chi Hey, so does German food taste good?
K: oh ne || S: you mei qu chi deguo cai|| Oh that - ||did you have German food||
K: you a xiangchang zhujiao... Yeah, I did. Sausage, pig’s feet . . .
/oh4/kyohk007
S: na haobuhao chi K: hen hao chi Do they taste good? || Delicious!
S: hen hao chi a Delicious?
K: umhum keshi ta nege beer hall hen Umhum but the beer halls were really
cao noisy
oh wo yao qu – wo yao huilai qian Oh the day before I went – I came
yitian shi tamen Octoberfest back – it was their Octoberfest
/oh4/kyohk008
S: yinwei you shihou hui fasheng Because sometimes it may not work,
guzhang ruguo meiyou lu jinqu de and if it’s not recorded then the talk
hua na women budou baijiang le ma? would be all wasted, right? (joking,
(joking, laugh) laugh)
K: oh-aaaa Oh-aaaa
/oh4/kyohk009
S: ey ni na le meiyou? zhe zenme Hey did you take it? How come
zhiyou yizhang a? there’s only one (picture) here?
K: oh yinwei wo wo you yizhang ji gei Oh because I – I sent one to her

ta le already
/oh4/kyohk010
K: haiyou hua // S: umhum// K: gankuai Lots of flowers still ||umhum|| hurry
zhao up and take pictures ||
S: umhum S: umhum
K: yeah oh zheshi Oxford Yeah Oh this is Oxford
S: dui zhe hen piaoliang ye Right it’s really beautiful
/oh4/kyohk011
S: jiu yingguo hen youming jiu England’s very famous for this type of
zhezhong garden ho || K: dui gardens huh || right
ranhou tamen nege hua ranhou ba ta Then the flowers – then they just
anpai gesi geyang nege yingguo hen arrange them in all different ways.
youming ye zhe yi fangmian England’s very famous for this.
K: oh zheshi tamen library zheshi tamen Oh this is their library. This is their
library library.
/oh4/kyohk012
S: dui (talking about caps and gowns) Right (talking about caps and gowns)
K: zhende a || S: chuan neyang Really || they have to dress like that
K: oh wo qu de shihou – shi suoyou de Oh the time I went – all the students?
xuesheng? || dou yao daizhe nege || they all have to carry their gowns
paozi ho
/oh4/kyohk013
S: ni de hei paozi Your black gown
K: oh zhende a || S: dui umhum Oh really || yeah umhum
/oh4/kyohk014
K: yinwei wo qu de shihou tamen shi Because it was summer when I went
summer ||vacation meiyou|| S:umhum there ||vacation…didn’t ...|| S: umhum
K: oh zheshi tamen weiyi yige yige Oh this is their only only college || oh
college || oh
/oh4/kyohk015
K: oh zheshi – S: dui zhe hen hao Oh this is – || yeah that’s really nice
/oh4/kyohk016
K: oh zheshi lingwai yige difang nege Oh this is another place – that’s Bath
wenchuan
/oh4/kyohk017
K: oh zheshi || S: ey zhe zhao de bucuo a Oh this is || hey this is a good shot ||
|| Bath limian de yige wenchuan it’s a hot spring in Bath
/oh4/kyohk018
S: ni meiyu xi wenchuan (laugh) You didn’t go to the hotspring?
290 L.-C. Yang
K: um meiyou ta lian ta zhiyou gei ni um no they only let you take a look,
kan yi kan, buneng rang ni muo they don’t let you touch it
S: zhe zhao de hen hao ye K: oh thank This is a really good picture! / Oh
you thank you
/oh4/kyohk019
K: oh zhe ye shi Bath Oh this is also in Bath
S: yinggai ba ta xiyixi ba ta fangda ma You should develop it – get it
enlarged
/oh4/kyohk020
K: ta ye you nege shenme President They also have President Bush
Bush keshi ta zuo de buxiang wo (statue) but it didn’t look like him so I
meiyou gen ta zhaoxiang (laugh) didn’t take a picture with him (laugh)
oh zheshi tamen nege royal family Oh this is the royal family
/oh4/kyohk021
K: oh zheshi Yingguo zuei da de baihuo Oh this is the biggest department store
gongsi in England
/oh4/kyohk022
S: ey zenmeyang Yingguo nega dian Hey, how is it? How are the stores in
zenmeyang? England?
K: oh dongxi hen gui// S: shangdian Oh things were really expensive! // the
shops?
3. Discourse text for Speaker W’s duis
/dui2/neduiw001
Y: danshi ta shi congzhong chuanzhen But she was helping in between
W: dui right
/dui2/neduiw002
Y: yinxian zheyang bangmang || W: dui in between helping like this || right ||
|| ho huh
/dui2/neduiw003
Y: wo benlai hen xiang wenwen ta de, Originally I was planning to ask her
yinwei women dou yao you jinguo about that, because we all have to go
tongyang de licheng through the same process
W: dui ia right
Y: zhuo yige cankao serve as a reference
W: shi a exactly
/dui2/neduiw004
Y: gankuai huiqu zamen dou huiqu Hurry up, let’s go back, let’s all go
(laugh) back (laugh)
W: dui ia zamen huiqu right let’s all go back
Y: dui ia dui ia right right
/dui2/neduiw005
Y: buyao zai tuo hen jiu Don’t drag on for too long
W: dui ia right
Y: umhum wo shi juede haishi – haishi umhum I think it’s better – better
/dui2/neduiw006
Y: yinwei henduo dongxi jianglai zai Because there’s lots of things that you
neli gonzuo de shihou hai keyi zai can still learn on the job
xue ma
W: shi a dui ia right exactly
/dui2/neduiw007
W: erqie ni jianglai gongzuo de shihou – Also when you are on the job –
m– m–
Y: faner tiaojian gen hao actually the conditions are a lot better
W: dui right
/dui2/neduiw008
W: ayo wo juede sanxinliangyi zui Ayo I think being indecisive is the
buhao le worst thing
Y: timu ding le zui hao buyao zai gai after a topic is decided, better not to
change it
W: dui right
Y: dui yinwei yaoburan qiangongjinqi right ‘cause otherwise all the work
yao zai congxin kaishi done is wasted and you have to start
over again
/dui2/neduiw009
Y: buguo wo juede proposal ta ye bu but I think it’s not necessary to write
yiding yao xie tai chang ma that long for the proposal
W: dui right
/dui2/neduiw010
Y: yinwei zhao xuexiao nege biaoge zhi Because according to the form from
you yidiandian school, there’s only a little bit of space
W: dui m right m
/dui2/neduiw011
Y: ni xian de mentor shi shei? Who’s your mentor now?
W: Professor N Professor N
Y: oh mentor shi Professor N oh your mentor’s Professor N
W: dui right
Y: hai you shei ne? who else?
292 L.-C. Yang
/dui2/neduiw013
Y: oh ni gen ta – nege zuoguo shibushi? oh you – you worked for him right?
W: dui wo – right I –
/dui2/neduiw014
Y: ni zuoguo ta de nege TA shibushi? you worked for him as a TA, right?
W: dui wo – right I –
/dui2/neduiw015
W: dui wo wo bang ta zuoguo right I – I worked for him for a short
yixiaoduan while
Y: cengjin bang ta daiguo ho helped him with the classes huh
W: dui yixiaoduan shijian right for a short while
/dui2/neduiw016
Y: ta ta shi xue nayi fanmian de? He – what area did he study?
W: ta shi xue linguistics de he studied linguistics
Y: oh ta shi xue linguistics de oh he did study linguistics
W: dui (laugh) right (laugh)
/dui2/neduiw017
W: ta dui wo bucuo He’s very good to me
Y: ta dui ni man hao huh he’s very good to you huh
W: dui right
/dui2/neduiw018
Y: xuexiao you gai le a The school changed the policy?
W: xuexiao you gai le jiushishuo – the school changed it, it’s saying –
Y: you gai le a? bu xiwang shuo zai they changed it? They don’t want you
zhao – to ask –
W: dui right
/dui2/neduiw019
Y: bu xiwang shuo zai zhao tuixiu de they don’t want you to ask retired
laoshi professors
W: dui dui right right
Y: oh oh
/dui2/neduiw020
W: dui || Y: oh (overlap) right || oh (overlap)
/dui2/neduiw021
W: ta dong zhongwen, ta shi Chinese He knows Chinese, he’s a native
native speaker, duibudui ? speaker of Chinese, right?
W: ranhou ta yingwen fangmian ye And his English is also pretty good
bucuo
/dui2/neduiw022
Y: buguo houlai wo ting ta jiang guoyu But later I heard him speaking
jiang de bucuo Chinese – pretty good pronunciation
W: ta guoyu jiang de bucuo His speaks Chinese very well
Y: umhum bi dabufen de nege umhum much better than most of the
guangdongren yao hao Cantonese speakers
W: dui jiang de hao duo le Right his is much better
/dui2/neduiw023
Y: ta ban le nege difang ta – tamen – She moved, that place she – they –
W: dui right
Y: ban zou le moved
/dui2/neduiw024
Y: ban zou le dui They moved right
W: dui right
/dui2/neduiw025
W: ta keneng liang – Maybe she’s both –
Y: keneng you you jihua you gaibian maybe there’s there’s a change of plan
W: dui jihua gaibian m right change of plans m
Y: dui right
/dui2/neduiw026
W: yinwei daxue shi – eryue duo kaixue ‘cause universities start sometime in
duibudui February right?
Y: keneng ba yinwei taiwan yao bi maybe, because Taiwan’s one month
zhebian wan yige yue dui later than here, right
/dui2/neduiw027
W: zheyangzi That’s the case
yinwei ta ye huiqu ma duibudui because she also went back, right?
Y: umhum umhum
W: ta hui gaoshu ta she will tell her
wo laide shihou wo cai zhidao shi ta I hear she went back to the university
shi hui zhengda only when I was about to come back
/dui2/neduiw028
Y: nege ta you ta ziji nege xueshu de And she has her own professional
shiyie ma academic career
W: umhum umhum
Y: ne zheyang de hua jiu – then in that case it’s –
W: zheyang jiu hen hao it’s pretty good
Y: ranhou quanjiaren zai yiqi na dangran then the whole family’s together of
hen kuaile course they are all happy
W: dui ia ||Y: dui ia right || right
294 L.-C. Yang
/dui2/neduiw029
Y: yinwei women zhexie ren dou – wo Because all of us are all – I think we
xiang women buhui hen jijiao won’t care very much about the
gongzuo nezhong hen – workload, whether it’s very –
W: dui right
Y: nezhong nezhong qinzhong duibudui whether it’s very light or heavy, right?
ho hum
W: umhum umhum
/dui2/neduiw030
Y: ta jiu yizhi – yitian daowan baoyuan He just kept – complaining all the
ey time
W: ta dui yuyanxue meiyou xingqu he’s not interested in linguistics
Y: ni huibuhui juede? do you feel that way?
W: dui right
/dui2/neduiw031
W: ta dui yuyanxue meiyou xingqu dui He’s not interested in linguistics, right
Y: ni you meiyou faxian? did you find that out?
Notes
Portions of this work were done while I was a visiting researcher at Advanced
Telecommunications Research Institute International (ATR), Japan. I would like to thank
the Japan Science and Technology Corporation for support of this research under the
Expressive Speech Processing Project in the research area of “Information Processing for
an Advanced Media Society”, and the National Science Council of Taiwan and Tunghai
University for their support. In particular, I would like to thank Wallace Chafe for his very
generous support and encouragement of this research as well as valuable discussions on
the issues presented. I would also like to thank the volume editor, Kerstin Fischer, for
very stimulating and helpful comments on this chapter.
1
In our work we define an interruption as a situation in which one person intends to
continue speaking, but is forced by the other person to stop speaking, at least
temporarily, or the continuity or regularity of that person’s speech is disrupted (Yang,
1995). Interruptions are therefore seen as consisting of three essential ingredients:
intention of the main speaker to continue, entrance of the other person into the
conversation, and disruption or stopping of the main speaker. The interruption can be
simultaneous with the start of the main speaker’s utterance, it can be in the middle of
the main speaker’s speech, or it can even occur when the main speaker has completed
an utterance but is not finished with the idea or story. Therefore, it is the degree of
disruption to the intended continuation of the main speaker which is the critical element,
and the degree of competitiveness or cooperation is determined by the actions and
intentions of both speakers (see Yang, 1995, for more detailed discussion). In this
example, the discourse context clearly indicates the disruption of the main speaker’s
topic development and the speaker’s wish to continue and therefore is treated as an
interruption. Alternatively, this instance can be considered as a transition relevance

place, as the other speaker may have interpreted this as the end of the main speaker’s
turn.
2
Emotion and cognition are often interrelated and hard to separate. The view that they are
not to be considered as completely autonomous entities and that emotion is inevitably
involved in cognition has also been gaining support from recent research in
neuroscience. Based on the observation that a specific type of brain damage leaves
knowledge, attention, memory, logic, and the ability to calculate intact, but leaves both
reasoning and feelings impaired, Damasio claims that feeling is in fact an integral
component of reason (Damasio, 1994; 159-162):
First, it is apparent that emotion is played out under the control of both subcortical
and neocortical structures. Second, and perhaps more important, feelings are just
as cognitive as any other perceptual image, and just as dependent on cerebral-
cortex processing as any other image. To be sure, feelings are about something
different. But what makes them different is that they are first and foremost about
the body, that they offer us the cognition of our visceral and musculoskeletal state
as it becomes affected by preorganized mechanisms and by the cognitive
structures we have developed under their influence.
Because of the interrelatedness of emotion and cognition, I treat the term emotion in a
broad sense and include both cognition and emotion in our analysis in cases when they
are difficult to distinguish.
3
It is not a necessary condition of the definition of discourse markers that they express
emotion, nor is this a mandatory feature for inclusion in the class of discourse markers.
However, it is imperative to consider the emotions that can be concurrently expressed
by discourse markers to give a more complete account of their functional polysymy.
Researchers differ in their account of the range of functions performed by discourse
markers. For example, Schiffrin (2001: 54) points out that discourse markers function
in cognitive, expressive, social, and textual domains; my viewpoint is in agreement
with that position.
4
The judgements of the cognitive-affective states presented in my analysis were made by
reference to evidence from a number of different knowledge sources. Inferences were
made based on semantic content, overt linguistic markers, and discourse context such
as information derived from the preceding and following conversational sequences. As
analysts, we have extra knowledge gained from the direct access to the total digitized
speech data and thus have a deeper understanding and familiarity with the participants’
states and speech styles. Knowledge of discourse context, analyst expertise and
sensitivity, and familiarity with the data all contribute to deriving the interpretation.
Independent verification was also obtained through a perceptual experiment on
identification of cognitive and emotional states (see Yang, 2001). The results
demonstrated that there are clearly identifiable patterns among observers in the
recognition of cognitive-emotional states, in agreement with previous research on
emotion recognition in the field.
5
The central role of prosody in expressing cognitive and emotional states has been
pointed out by numerous researchers, including Pierrehumbert and Hirschberg (1990),
Hirschberg and Ward (1992), Chafe (2001a), and Levelt (1989), and some universal
296 L.-C. Yang
prosodic principles have been proposed (e.g., Bolinger, 1987; Ohala, 1984; Scherer, 2000;
Gussenhoven, 2002). In particular, it is well established that pitch level and pitch range
are correlated with the degree of intensity, i.e., a higher pitch level and a wider pitch
range signal a higher degree of emphasis, involvement, and intensity (Bolinger, 1987;
Scherer, 2000; Chafe, 2001b; Liberman and Pierrehumbert, 1984; among others). Our
perceptual experiment also supports that viewpoint (for a detailed account, see Yang,
1995).
6
The fact that pitch shapes drawn from two independent subsections of the conversation
have a similar spectrum of shapes suggest that the variability of forms represented here
is likely to be a typical feature of conversational exchanges of the type that are
exemplified here.
7
Sample analysis for s-oh1, s-oh2, and s-oh4 are presented here to show how the
interpretations in our examples are derived based on both discourse contextual cues
and prosodic features. The corresponding discourse texts are presented in Appendix A.
Examples of s-oh1 and s-oh2: In this section, the speaker (speaker K) was describing a
place she visited on her trip abroad, referring to the place as wenquan, and speaker S
responds with a high rising pitch oh (s-oh1) indicating her incomprehension. Speaker
K’s grasp of this puzzlement is evidenced in the subsequent response, as she first
clarifies that it is not Oxford, and then provides further specification by spelling out
the name and emphasizing that Bath wenquan is the name of the place. Speaker S’s
following oh with a prominent rise and fall shape indicates her resulting realization
and understanding of the intended meaning. This comprehending oh is reinforced by
her subsequent comment to show that she recognizes that place, and is further
supported by the confirmatory “right right right” sequence.
Example s-oh4: Here, speaker S introduces a topic of interest to her by prompting
speaker K with a question about Oxford University. Speaker K first presents her
assessment, then elaborates by bringing in a relevant but contrastive key point with
strong emphasis on sanshi ‘thirty’, highlighting this point as noteworthy. Speaker S
responds with a sharp pitch rise and fall oh, capturing her surprised and impressed
state at receiving the new and unexpected information. Her recognition and agreement
of speaker K’s intended emphasis were indicated by her restatement of the fact and
this is in turn reconfirmed by speaker K. (See Hirschberg and Ward on their work on
prosody of surprise and incredulity.)
8
In Chinese, pitch variation is used to distinguish the lexical meanings of syllables in a
semantic sense. Every syllable is associated with a specified tone and each tone has a
distinct shape. Tone 1 is high and level, tone 2 is rising, tone 3 is fall-rise, and tone 4 is
falling. There is also a neutral tone for weakly stressed syllables. The traditional notation
and pitch value for these tones in citation form are as follows (Chao, 1968):
Tone Shape Description Pitch Value
Tone 1 - high level 55
Tone 2 ´ rising 35
Tone 3 ˇ fall-rise 214
Tone 4 ` falling 51
Tone 0 variable
See Yang (1995) for a more complete account of the interaction of tone and intonation
in Mandarin Chinese.
Chapter 16
Formal properties of a subset of discourse

markers: connectives1
Corinne Rossari
1. Introduction
In this paper I discuss the semantic constraints involved by the use of a subclass of
discourse markers (DMs) generally designated by the term connectives. I consider them to
be a DM subclass because they do not contribute to the truth value of the proposition in
which they occur, they are polyfunctional items, and they do not belong to one particular
grammatical category. The discrepancy between grammatical function and discourse
function has been taken for granted since Ducrot’s (1975) seminal paper on car, parce
que, and puisque; coordinating and subordinating conjunctions, as well as adverbial
conjunctions, may play the role of discourse markers. Nevertheless, all these items differ
from other DMs in that they may give indications on how the discourse unit in which they
occur relates to the particular piece of information to which they are attached. My aim is
to show that this property can be captured formally by identifying the semantic constraints
imposed by the connective on the left as well as by the context on the right.
Polyfunctionality will be treated in relation to the type of discourse configurations in

which they can occur and the type of operation they perform on the left context. A
specific item may assume various semantic values while performing one and the same
operation on the left context. This view corresponds to a narrow version of the
monosemic approach; it assumes that a certain item having a particular function always
performs one and the same operation, which may give rise to different semantic values
depending on the configurations in which the marker occurs. However, this does not
amount to saying that one marker may not have different functions. If this is the case, a
polysemic approach should be used in order to put these functions in relation to each
other. For instance, the French marker donc has a connective value (The weather is nice,
DONC I’ll go out for a walk) and an exclamatory one (How nice this place DONC is). These
two values correspond to two different functions. Therefore, they will not be treated as
performing the same operation. I will only deal with the first case, in which one item
having a connective function may convey various semantic values.
300 C. Rossari
1.1. Approach, methodology, and data
The present approach to this particular DM class is related to a type of lexical semantics
dealing with an item’s conditions of use. We adopt the idea that the items analysed
convey constraints determining the semantic profile of the entities they connect. These
constraints belong to the connective’s semantic component. In other words, the lexical
semantics we are dealing with consists in sorting out the factors determining the
compatibility of a marker with specific linguistic structures. It does not consist, as one
might think, in seeking the coded meaning of an item by analysing the possible
interpretations of the utterances in which it may occur. This particular standpoint is
responsible for at least four parameters of our analysis.
• We focus our attention only on structures where each information unit connected by
the marker corresponds to an utterance.
• We consider the constraints in relation to one particular function assumed by the
marker. We do not try to describe them in relation to all the functions a marker may
possibly assume. For instance, the constraints valid for the French inferential donc are
not necessarily relevant for the other functions of this marker.
• The issue of polyfunctionality is addressed in relation to the type of discourse
configurations in which the markers may occur and the type of operation they perform
on the left context (see section 4). There are three main types of contexts where a
connective may occur and in each one it performs a particular operation (see section
3).
• We consider “bad” and “good” uses of an item to be equally meaningful data.
We use a classical distributive methodology. It consists in a controlled variation of the

linguistic contexts where a specific item may occur. We investigate the factors to which
the marker is sensitive and use them in order to identify the semantic type of the entities
connected. A general characteristic of connectives is their capacity to impose constraints
not only on the semantic nature of the discourse unit they introduce but also on the
semantic type of the preceding discourse unit. This feature may be captured when we
have a direct link between discourse units and the information units connected.
Our study is based on standard written French data. We use constructed examples as well
as corpus examples. We acknowledge the notion of norm. This means that (i) other things
being equal, the marker’s occurrence in one particular linguistic context can be considered
as less natural than its occurrence in another particular linguistic context and (ii) this
difference in acceptability is coded in the marker’s conditions of use; it does not depend
on the pragmatic nonappropriateness of the situation.
1.2.1. State of the art: some general assumptions shared by various approaches to
discourse markers
Analyses specifically focused on connectives as a DM subtype have paid a particular
attention to the scope issue. The notion of scope addresses many aspects involved in a
connection process, which are mainly related to two problems: (i) the delimitation of the
DMs and semantic constraints 301
linguistic material involved in the relation expressed by the connective and (ii) the kind of
entities that the connective takes as arguments. Approaches interested in the former
problem attempt to find criteria for delimiting the linguistic sequence concerned by the
connection. When they are conversation oriented, they use notions such as act,
intervention, turn taking, or dialogic unit to delineate these sequences. Those focused on
written text resort to notions such as clause, sentence, paragraph, full stop.
Approaches interested in the latter problem are aimed at representing a schema generally
applicable to various forms of connection realised by a connective. They have to deal with
the problematic definition of the kind of entities being connected. Are they utterances,
propositions, inferences, or something else? The analyses using the notion of utterance
have to determine the level (i.e., illocutionary, epistemic, or content) at which the
connection is established. Those resorting to the notion of proposition have to specify the
type of framework adopted (i.e., modal or truth conditional). As for the linguists who
think that connectives deal with inferences, they have to conceive of a model meant to
determine how these inferences are obtained. Certain approaches adopt yet other
perspectives to describe the entities taken as arguments. For instance, Berrendonner
(1990) opts for a strictly cognitive perspective where these units are conceived as abstract
information stored in discourse memory, the latter being continuously enriched by
discourse flow and context. In contrast, Carel and Ducrot (Carel and Ducrot, 1999; Carel,
2002) adopt a strictly lexical point of view. They assign an intrinsic argumentative power
to every lexical item and explore it by using two basic argumentative clauses, namely,
donc-clauses and pourtant-clauses. These clauses are assumed to be interdependent: the
one cannot be conceived without the other (Carel and Ducrot, 1999).
There are not many approaches accepting the combination of distinct perspectives.
Analyses based on formal semantics adopt the proposition as the relevant unit for the
connectives’ description, textual approaches focus on the utterance level, and instructional
or procedural analyses use inferences. The multidimensional approach adopted by the
Geneva school (Roulet, Filliettaz, and Grobet, 2001) is compatible with a combination of
these different representations. Due to a modular framework, they may be displayed at
various analysis levels. Hence, discourse organization is considered to be the product of
various organizational systems, the main ones being the linguistic one, the textual one,
and the situational one. The connectives’ scope is analysed in relation to the
organizational system taken into consideration. The notion of proposition will be relevant
at the linguistic level and that of inference at the textual one. As far as the situational level
is concerned, it enables connectives to be analyzed as using praxeological structures.
Nevertheless, such an approach does not need specific tools in order to represent a general
schema capable of accounting for the different forms a connection may assume. Its
specific aim is to provide an overview of discourse organization.
1.2.2. Problems
The main problem that any theory about DMs has to face is their intrinsic polysemy. This
feature characterizes many connectives. One relevant example might be the French DM
donc, which, as we have already mentioned, may assume different functions. It may be
analysed as:
• A deductive connective in (1):

302 C. Rossari
(1) Max est petit, donc il fait encore des bêtises.

Max is young, DONC he still does silly things.
• A rephrasing marker in (2):
(2) Vous pensez donc qu’il vaut mieux renoncer à cette affaire.
You think DONC that it’s better to give up this matter.
• A phatic particle in (3):
(3) Dis-donc, comment tu parles à ta mère?

Hey DONC, how do you speak to your mother?
• An exclamatory marker in (4):
(4) Que Paul est donc gentil!

Paul is DONC so nice!
Moreover, these different functions are not always so easy to determine. There are many
ambiguous cases, where the analyst may hesitate between several interpretations, as in
(5):
(5) Les baleines allaitent leurs petits.

Whales suckle their offspring.
Ce sont donc des mammifères.

They are DONC mammals.
In such a context, donc may be interpreted as a reformulation as well as a deduction

marker.
To put it shortly, any classification has to cope with the diversity of uses, the numerous
ambiguous cases and the factors responsible for these ambiguities (are the markers
intrinsically ambiguous or does ambiguity arise when they occur in some specific
context?). Since these items are so polyvalent, the notion of connective itself is pretty
difficult to delineate. The function characterizing all the members of the class cannot be
but generic. This does not allow one to put clear limits on the lexical items that can be
used as connectives and those that cannot.
The formal approach I will explore will make it possible to highlight some features
characterizing connectives as a DMs subclass. The basic idea is that their capacity of
putting information together should bear on their conditions of use. Therefore I shall focus
exclusively on the constraints they impose on their linguistic environment.
2. Definition
Connectives are a DM subclass that imposes restrictions on the formulation of the right
and left linguistic context when the entities taken as arguments correspond to linguistic
sequences. This claim is to be derived from two facts:
• A discourse relation connecting two discourse units does not impose the same
constraints when a connective is used to express it and when no connective is used.
• Each connective determines the semantic profile of the entities it takes as arguments.
The first fact is illustrated by the analysis of counterfactual utterances proposed in

Akatsuka and Strauss (2000). This study points out that the use of a counterfactual
utterance requires an explicit desirability assessment from the speaker in the preceding
utterance. The result of this assessment can be positive if the state of affairs is desired by
the speaker or negative if this is not the case. Akatsuka and Strauss examine a series of
examples borrowed from Fauconnier (1984), who does not take this parameter into
account in his analysis.
(6) Fortunately, the fire did not cross the highway. My house would have been
destroyed.
(7) Luckily, the fire was prevented from crossing the highway. My home would have
been destroyed.
They show that the use of the counterfactual is clearly less natural if the attitudinal
adverbs expressing the speaker's stance are not present:
(8) ??The fire did not cross the highway. My house would have been destroyed.
(9) ??The fire was prevented from crossing the highway. My home would have been
destroyed.
At first glance, (8) and (9) might seem less natural than (6) and (7). However, it is not the
expression of the speaker’s stance per se that impinges on the adequate use of the
counterfactual but some general principle related to the possibility of recovering a specific
relation attaching any discourse constituent to its context. In (6) and (7), the
counterfactual is interpreted as a justification of the speaker’s positive stance. Examples
(8) and (9) may also be seen as natural if the counterfactual utterance is interpreted as a
justification of the validity of the state of affairs expressed in the first unit. The use of
some epistemic indicator might facilitate this interpretation, which could seem less
immediate for (8) and (9).
(10) Probably the fire did not cross the highway. My house would have been
destroyed.
(11) Probably the fire was prevented from crossing the highway. My home would
have been destroyed.
However, when a connective is used to express the same type of coherence relation, its
very presence places constraints on the way the state of affairs is presented.
(12) Il y a des souris chez Marie. J’en ai vu dans sa cuisine.

Mary has got mice in her house. I have seen some in her kitchen.
304 C. Rossari
None of the discourse markers that are commonly used to convey such a relation is
acceptable in such a sequence:
(13) Il y a des souris chez Marie. ??Car j’en ai vu dans sa cuisine.

Mary has got mice in her house. CAR I have seen some in her kitchen.
(14) Il y a des souris chez Marie. ??En effet, j’en ai vu dans sa cuisine.
Mary has got mice in her house. EN EFFET I have seen some in her kitchen.
(15) Il y a des souris chez Marie. ??Effectivement, j’en ai vu dans sa cuisine.

Mary has got mice in her house. EFFECTIVEMENT I have seen some in her kitchen.
To make these connections more natural, it is necessary to add an epistemic clue in (13)
and (14) and a viewpoint clue in (15).
(16) Il y a probablement des souris chez Marie. Car j’en ai vu dans sa cuisine.
Mary has probably got mice in her house. CAR I have seen some in her kitchen.
(17) Il y a probablement des souris chez Marie. En effet, j’en ai vu dans sa cuisine.
Mary has probably got mice in her house. EN EFFET I have seen some in her
kitchen.
(18) Selon Luc, il y a des souris chez Marie. Effectivement, j’en ai vu dans sa cuisine.
According to Luc, Mary has got mice in her house. EFFECTIVEMENT I have seen
some in her kitchen.
Each of these linguistic clues assigns a specific semantic type to the proposition (namely a
modal type for the first two and an evidential one for the third), rendering it compatible
with the marker.
These data show the differences between the constraints at work behind a simple
discourse relation and a relation signalled by a discourse marker. The construction of an
adequate discourse relation is not dependent on a particular linguistic clue. It rather rests
on pragmatic principles allowing the construction of an adequate context in which each
discourse constituent may be attached to a particular information unit. This is not the case
with connectives. They may require some special linguistic clue, especially for their left
linguistic context. This is one of the features that distinguishes them from other DMs.
The second fact, i.e., that every connective determines the semantic profile of the entities
it takes as arguments, can be illustrated by their property of calling for different clues in
the left clause. As mentioned above, en effet and effectivement require an epistemic clue
and a viewpoint clue, respectively. Autrement, which also conveys a justification relation
(Inkova-Manzotti, 2002), places a constraint on the right clause, which has to be
compatible with the accommodation of a counterfactual derived from the left clause.2
(19) Va voir Marie! Autrement elle se fâchera.

Go and see Mary! AUTREMENT she will get angry.
(20) Va voir Marie! ??Autrement elle se serait fâchée.

Go and see Mary! AUTREMENT she would have got angry.
(20) is ill-formed because the past conditional used in the right utterance is incompatible
with the proposition accommodated from the imperative in the left utterance, i.e., if you
don’t go and see Mary.
On a more general level, accommodated propositions are not always accepted as

antecedents by connectives. À ce moment-là may assume an inferential value; it conveys
then the same type of relation as donc or alors. In such uses it cannot always be replaced
by donc. Their interchangeability depends on the semantic type of the proposition used in
the left utterance. If it is an accommodated proposition, i.e., an if clause, donc is
inappropriate, while alors and à ce moment-là are natural (Choueiri, 2002).
(21) Tu dois aller voir Marie. À ce moment-là tu sauras ce qui s’est passé.
You must go and see Mary. À CE MOMENT-LÀ, you will find out what has
happened.
(22) Tu dois aller voir Marie. Alors tu sauras ce qui s’est passé.
You must go and see Mary. ALORS, you will find out what has happened.
(23) Tu dois aller voir Marie. ??Donc tu sauras ce qui s’est passé.
You must go and see Mary. DONC, you will find out what has happened.
By contrast, à ce moment-là is not appropriate when the left proposition cannot be

interpreted as being an accommodated if-clause. As for alors, it accepts both
accommodated and nonaccommodated propositions as its antecedents.
(24) Tu es allé voir Marie. ?? À ce moment-là tu sais ce qui s’est passé.

You went to see Mary. À CE MOMENT-LÀ, you know what has happened.
(25) Tu es allé voir Marie. Alors tu sais ce qui s’est passé.

You went to see Mary. ALORS, you know what has happened.
(26) Tu es allé voir Marie. Donc tu sais ce qui s’est passé.

You went to see Mary. DONC, you know what has happened.
A final group of items that I will use to illustrate the way in which connectives determine
the semantic type of their arguments conveys a completely different type of relation. It
provides information about the topical organization of discourse and includes items such
as à ce propos and à propos de. They are studied in Beaulieu-Masson (2002), where the
difference between the antecedent required by à ce propos and the one required by à
propos de is pointed out.
(27) J’ai vu de très belles églises à Naples. À propos d’églises, es-tu bien allé à la
messe hier?
I saw some beautiful churches in Naples. À PROPOS DE churches, did you attend
the mass yesterday?
306 C. Rossari
(28) J’ai vu de très belles églises à Naples. ??À ce propos, es-tu bien allé à la messe
hier?
I saw some beautiful churches in Naples. À CE PROPOS, did you attend the mass
yesterday?
(29) J’ai vu de très belles églises à Naples. À ce propos, es-tu allé visiter la cathédrale
de Chartres?
I saw some beautiful churches in Naples. À CE PROPOS, have you ever visited the
Chartres Cathedral?
These examples are used to indicate that à propos de may take a generic object as
antecedent, while à ce propos requires a proposition denoting the attribution of a property
to an object, i.e., the property of being beautiful as applied to certain churches.
Within this approach, the functional spectrum is determined in relation to the type of
context where a connective may occur. Three main context types may be distinguished.
• Inferential contexts: the relation is motivated by an inferential link (i.e., deduction or

abduction).
• Corrective contexts: the relation is motivated by a left-context modification.
• Topical contexts: the relation is motivated by the use of some topic accessible from the
left context.
The connective performs a different operation in each context type. The same operation
may be applied to various discourse configurations. For instance, in an inferential context
we may have argumentative/counter-argumentative or illocutionary/content-based
configurations (see section 4). The interaction between the operation and the discourse
configuration itself is responsible for the connective’s polyfunctionality. The operations
corresponding to these three context types can be represented in a dynamic semantics
framework inspired from Veltman’s (1996) Update Model.
Within such an approach, discourse flow is represented as a series of successive

eliminative updates on information states. An information state consists of a set of worlds.
Each world stands for a complete set of propositions. Updating an information state with a
proposition means eliminating all the worlds where that proposition is false. The update is
considered successful if it does not result in the empty set. Connectives are conceived of
as devices signalling relations between updates. The notation system we adopt uses X and
Y as labels for the utterances corresponding to the left and right contexts, respectively.
Each operation is determined by the way in which the connective interacts with the
information state.
• The connective uses a certain information state in order to guarantee the success of an
update on another information state.
• The connective modifies the information state provided by the left context.
• The connective uses the information state provided by the left context as a frame for
highlighting a parallelism or a contrast.
Connectives commonly considered as causal (i.e., therefore, because, since) use an

information state as a guarantee to ensure the success of an update on another information
state; several inferential processes are involved in this mechanism. Nevertheless, certain
authors show how a concessive relation may be considered to be the negative counterpart
of a causal one.3 The example given below is taken from König and Siemund (2000):
(30) a. / The house is no less comfortable because it dispenses with air-conditioning. /

b. The house is no LESS comfortable / although it dispenses with air-
conditioning./
[/ stands for a pause; capitals stand for emphasis on less]
(König and Siemund, 2000: 354)
The authors of this paper base their demonstration on the existence of a presupposed
logical implication noted P Q, corresponding to the expression of a law derived from
the propositions connected. This implication is valid for a negated causality as well as for
a concessive relation. Using this expedient, they establish the semantic equivalence
between the construction ~(because p, q) and the construction although p, ~q as follows:
(i) ~(because p, q) although p, ~q

(ii) P Q; p (presuppositions) P Q; p (presupp.)
(iii) ~(p & q)
(iv) p & ~q (since p is a presupp.) = p & ~q
Since the negation of a causal construction does not affect its presuppositions,
these go through unchanged (cf. line (ii)). The negation relating to the whole
causal construction in line (i) can therefore only relate to the assertive part of its
meaning, i.e., the conjunction p & q (cf. line (iii)). Since we also assume that the
causal clause represents presupposed material, the negation can only affect the
main clause, as is indicated in line (iv). The external negation of a causal
construction is therefore shown to be equivalent to the internal one of a
concessive construction. (König and Siemund, 2000: 354).
This enables us to see that causal and concessive connectives share the property of using
information states in order to guarantee the success of an update on another information
state.
The above-mentioned implication indicates that the connective uses the current
information state as a guarantee. It bases the relation on some general law derived from
this current state. This law represented by an implication between two quantified
propositions enables the connective to use one information state in order to ensure the
success of an update on another information state. The term guarantee means that the
connective does not only refer to the previous information state, but it also uses it in an
inferential procedure.
Another situation occurs when the connective modifies the information state provided by
the left context. Some connectives have the capacity of cancelling updates on some
previous information state. Let us look at the following examples provided by
Razgouliaeva (2002).
308 C. Rossari
(31) La Bovary traînotte toujours, mais enfin avance. J’espère d’ici à quinze jours
avoir fait un grand pas.
(Flaubert, Correspondance, 1853, p. 131).
Mrs. Bovary is still dawdling, MAIS ENFIN she is moving on. I hope I will manage
to take a big step forward in the next fifteen days.
(32) Vous êtes encore très bien, et tout le monde dit que vous ne paraissez pas votre
âge; et quand vous sortez avec moi, mes anciennes camarades de classe vous
prennent pour mon amoureux . . . mais enfin, vous avez changé (Fonson,
Wicheler) (Razgouliaeva, 2002).
You’re still very good looking, and everybody thinks that you don’t look your
age; and when you go out with me, my former classmates believe that you are my
lover . . . MAIS ENFIN you are no longer the same.
This paper shows that the suppression of enfin makes the contrast stronger. Leaving aside
the problems raised by the concatenation of two connectives, we may say that the
difference in contrast strength sheds light on the way enfin acts on the information state
provided by the left context. Using these examples as a starting point, we may construct
some simpler structures where the difference in contrast is quite clear.
(33) Les journées passent lentement en prison, mais enfin elles passent.
Days are very long when you’re in jail, MAIS ENFIN they pass by.
(34) Les journées passent lentement en prison, mais elles passent.

Days are very long when you’re in jail, MAIS ENFIN they pass by.
(35) Vous n’êtes pas vieux, mais enfin vous n’êtes pas tout jeune.
You are not old, MAIS ENFIN you are not very young.
(36) Vous n’êtes pas vieux, mais vous n’êtes pas tout jeune.
You are not old, MAIS you are not very young.
The stronger contrast in the examples where enfin is absent comes from the fact that the
right utterance seems to contradict the left one by denying a conventional implicature
attached to it (the days do not pass by in (34) and you are young in (36)). According to
Razgouliaeva, enfin has the capacity of cancelling such an implicature, thus making the
contrast weaker. The ability to delete some information conveyed by the left context can
be tested by resorting to examples where enfin is not accompanied by mais. As assumed
in Rossari (2000, ch. 3), in some contexts enfin may cancel the illocutionary goal of a
speech act.
(37) Où étais-tu hier soir? Enfin tu n’es pas obligé de répondre.

Where were you yesterday night? ENFIN you don’t have to answer my question.
(38) Va voir Marie! Enfin fais ce que tu veux.

Go and see Mary! ENFIN do as you want.
In such structures the use of mais is impossible because the connective does not eliminate
the contradiction and causes an illocutionary suicide.
(39) Où étais-tu hier soir? ?? Mais tu n’es pas obligé de répondre.

Where were you yesterday night? MAIS you do not have to answer my question.
(40) Va voir Marie! ?? Mais fais ce que tu veux.

Go and see Mary! MAIS do as you want.
This difference in acceptability indicates that enfin succeeds in eliminating the

contradiction by cancelling the conventional implicature which consists of the
illocutionary goal attached to any question or request (i.e., the desire of getting an answer
or the desire of getting a certain action performed by the interlocutor).
A final situation occurs when the connective uses the information state provided by the
left context as a frame for highlighting a parallelism or a contrast. Some connectives (such
as à l’instar, de même, à ce propos) underline a parallelism and others (such as à
l’inverse, à l’opposé, au contraire) a contrast. However, both types have to resort to the
left information state in order to ensure the link with the information state they update
with utterance Y; parallelism or contrast exist only if the two situations are conceived
together, in relation to each other. Because of this, the connective may hardly be
suppressed, as shown in the following examples:
(41) J’ai vu de très belles églises à Naples. À ce propos, es-tu allé visiter la cathédrale
de Chartres?
I saw some beautiful churches in Naples. À CE PROPOS, have you ever visited the
Chartres Cathedral?
(42) ??J’ai vu de très belles églises à Naples. Es-tu allé visiter la cathédrale de
Chartres?
I saw some beautiful churches in Naples. Have you ever visited the Chartres
Cathedral?
(43) Marie parle anglais couramment. De même ses frères et sœurs sont très doués
pour les langues.
Mary speaks English fluently. DE MÊME her brothers and sisters have a real gift for
languages.
(44) ??Marie parle anglais couramment. Ses frères et sœurs sont très doués pour les
langues.
Mary speaks English fluently. Her brothers and sisters have a real gift for
languages.
However, these structures become natural if parallelism is emphasized in utterance Y,

because the absence of the connective is compensated for by an analogy marker.
310 C. Rossari
(45) J’ai vu de très belles églises à Naples. Es-tu allé visiter la cathédrale de Chartres,
qui est aussi très belle?
I saw some beautiful churches in Naples. Have you ever visited Chartres
Cathedral, which is also very beautiful?
(46) Marie parle anglais couramment. Ses frères et sœurs ont également appris
l’anglais très jeunes.
Mary speaks English fluently. Her brothers and sisters also learned English when
they were very young.
These connectives’ capacity of using a particular information state in order to highlight

parallelism or contrast is not derived from the contextual property attaching any discourse
unit to a particular piece of information. It is rather due to a lexical property as such, as
shown by the difference between the examples where the connective is being used and
those where it is not.
4. Model
The various operations mentioned in the preceding section are represented within
Veltman’s (1996) framework. We will use two connectives, alors and après tout, to
illustrate the way we may correlate the parameters governing their use with general
devices in update operations.
4.1. Alors
It has often been noted that some connectives may act not only on the content level, but
also on the speech act level (see Sweetser, 1990, among others). In its nontemporal use,
alors seems to have this property, since it can be employed to introduce a question after
an assertion on which the former is based.
(47) Tu mourais de faim tout à l’heure. Alors pourquoi ne manges-tu pas maintenant?
You were starving a moment ago. ALORS why aren’t you eating right now?
It may thus appear as being able to assume both semantic values, since it either connects
speech acts or contents. However, it cannot be used in a speech act configuration such as
the following:
(48) Je n’ai pas pu assister au dernier cours de maths. ??Alors pourquoi est-ce que les
triangles sont isocèles?
I wasn’t able to attend the last maths class. ALORS why are these triangles
isosceles?
Structures similar to (47) are analysed in Jayez (2002), where it is shown that, because of
the connective, the question is entailed by the propositional content of the preceding
assertion. Jayez uses the neologism impliquestion to designate such configurations.
Specifically, his analysis examines yes/no-questions and wh-questions. He proposes to
represent the consequence link in relation to the propositional content of the impliquestion
and not in relation to its denotation (i.e., the possible answers it might have). The
assumption he makes about yes/no-impliquestions goes as follows: “Dans un contexte où

un contenu propositionnel p n’est pas tenu pour certain ou pour faux, une impliquestion
oui-non de contenu p est légitimée par le fait de renforcer ou d’affaiblir p (et non pas les
réponses possibles à la question)” (Jayez, 2002: 149).
The structure in (47) corresponds to a wh-impliquestion. It is quite difficult if not

impossible to represent its propositional content. Hence, I will use the question denotation
in order to represent the link with the preceding assertion. This link determines the
connective’s appropriate use. The difference between (47) and (48) is accounted for by
the way we interpret the wh-question. In (47) it is interpreted as having a negative answer
(the interlocutor has no reason for not eating), while in (48) it is interpreted as having a
causal answer (these triangles are isosceles because they have two equal sides). Since a
question can be assigned no truth value, there is no update performed but merely a check
on the information state.4 This check guarantees that such an update would not result in
the empty set. That amounts to saying that there is at least one world where the
propositions the interlocutor has no reason for not eating and these triangles are isosceles
because they have two equal sides are respectively true. The difference between (47) and
(48) may be accounted for by taking into consideration these two answers.
The relation established by the connective between utterance X and utterance Y could be
schematically represented as follows. In (47), the check on the information state resulting
from Y cannot fail since the general law updated by alors (i.e., when someone is starving,
he has no reason for not eating) ensures the existence of p [i.e., the interlocutor has no
reason for not eating] in at least one world. In (48), the check on the information state
resulting from Y fails because the general law when someone does not attend a class, he
has to catch up with what has been taught does not ensure the existence of p [i.e., these
triangles are isosceles because they have two equal sides] in at least one world.
This representation accounts for the constraints affecting the polyfunctionality of a

connective occurring in inferential contexts. A marker such as alors will always be
propositional-content sensitive, even in structures where the connection seems to concern
the speech act level. Thus, the operation it performs stays the same, even if the relation
may hold either at the speech act level or the content level.
4.2. Après tout
The use of après tout is appropriate when some revision process is operational. This
process has two stages:
• Suppression: we may say that an update is being suppressed if a connective renders the
proposition updated by utterance X neither true nor false in the subsequent information
state.
• Substitution: after the suppression procedure is performed, the converse proposition is
updated. Thus the initial proposition is replaced by its converse in the resulting
information state.
The revision process has been tested on French rephrasing connectives such as enfin,
disons, de toute façon, quoi qu’il en soit (Rossari, 2000, ch. 3), and en réalité (Rossari,
2002). Here I will resort to it so as to capture some possibilities of use of après tout. As it
312 C. Rossari
has been emphasized by various approaches, this connective is particularly

polyfunctional. Philippi (1999) has pointed out the extreme diversity of its functions; for
instance, it has been described both as a reformulation connective (Roulet, 1990) and as
an argumentation marker (Blakemore, 1987). These divergent perspectives are not
surprising, since après tout seems adaptable to any discourse configuration. For instance,
it may be used in argumentative configurations, where its host utterance can be interpreted
as justifying the preceding one:
(49) Cette voiture est trop chère. Après tout elle a déjà plus de 100.000 kilomètres.
This car is too expensive. APRÈS TOUT it has run more than 100,000 km.
It may occur in counterargumentative configurations, where the host utterance introduces

some sort of counterexpectation:
(50) Cette voiture est trop chère. Après tout elle est vraiment belle et j’ai absolument
besoin d’une nouvelle voiture.
This car is too expensive. APRÈS TOUT it is really beautiful and I definitely need a
new car.
It may also occur in illocutionary configurations, where it attenuates the illocutionary

suicide in the same way as enfin does.
(51) Où étais-tu hier soir? Après tout tu n’es pas obligé de répondre.
Where were you yesterday night? APRÈS TOUT you don’t have to answer my
question.
(52) Va voir Marie! Après tout fais ce que tu veux.

Go and see Mary! APRÈS TOUT do as you want.
Nevertheless, it may also be used in contexts where enfin is unnatural.
(53) Marie est enceinte. Après tout/??Enfin tu es peut-être déjà au courant.

Mary is pregnant. APRÈS TOUT/Enfin perhaps you already know it.
(54) Avec qui étais-tu hier soir ? Après tout/??Enfin je le sais.

Who did you go out with yesterday night? APRÈS TOUT/Enfin I know who that
was.
(55) Arrête de boire ! Après tout/??Enfin je ne suis pas ta mère.

Stop drinking! APRÈS TOUT/Enfin I am not your mother.
The marker’s use in the above-mentioned configurations may be accounted for by the
underlying suppression mechanism. The proposition being suppressed corresponds to a
semantic entity derived from the speech act performed. It may either be one of its felicity
conditions or its presumed relevance. More specifically, in (53) après tout enables the
suppression of one of the felicity conditions of the assertion, i.e., the hearer does not
already know the information conveyed by the proposition. As far as questions are
concerned, the suppressed proposition may correspond to the felicity condition stipulating
that the speaker does not already know the answer (54). As for orders, the felicity
condition being suppressed may run as follows: the speaker has the appropriate social
status (55). The illocutionary suicide provoked by utterance Y in these three
configurations is thus cancelled.
However, après tout may also be used in configurations where utterance Y, instead of
giving rise to a possible illocutionary suicide, reinforces the speech act conveyed by
utterance X.
(56) Marie est enceinte. Après tout tu ne pouvais pas le deviner.

Mary is pregnant. APRÈS TOUT you could not have guessed it.
(57) Avec qui étais-tu hier soir? Après tout je n’en ai aucune idée.
Who did you go out with yesterday night? APRÈS TOUT I have no idea who that
was.
(58) Arrête de boire! Après tout je suis ta mère.

Stop drinking! APRÈS TOUT I am your mother.
In such contexts, utterance Y reasserts the appropriateness of the act conveyed by

utterance X. In other words, the speaker acts as if the felicity conditions attached to the
latter hadn’t been fulfilled.
The difference between the structures where enfin is possible (namely, (51) and (52)) and
those where it is unnatural (namely, (53), (54), and (55)) is that in the latter, the
suppression operation affects the relevance of the speech act, whereas in the former this
very speech act remains apposite: the speaker merely communicates that the hearer
doesn’t have to perform the action required. When après tout occurs in argumentative
configurations, the suppression mechanism bears on the presumed relevance of the speech
act conveyed by utterance X. In (49), après tout suppresses a proposition of the type X is
a good argument for Z (Z = Not to buy the car). In counterargumentative configurations
(50), the suppression concerns the same type of proposition, but it is motivated by the
introduction of a counterargument to X, while in argumentative configurations it is
motivated by the introduction of a better argument than X.
This revision process accounts for the polyfunctionality of après tout. By suppressing an
update, this connective is able to occur in various rhetorical configurations where it
displays a wide range of semantic nuances (i.e., argumentative, counterargumentative, and
corrective).
5. A broader perspective
The broader perspective in which our research may find some relevant extension is
twofold. First, it provides for confrontation with theories on discourse relations. One of
the main features emphasized by such an approach is the radical discrepancy between the
constraints governing discourse relations in general and the parameters governing the use
of a connective. Even if there are cases where connectives do signal a particular
coherence relation, they should never be considered as the linguistic counterparts of
discourse relations. Although they do not contribute to the semantic content of the
proposition in which they occur, they impose semantic constraints on the entities they
314 C. Rossari
connect. In contrast, the appropriateness of a discourse relation depends exclusively on

the possibility of attaching an information unit to a certain discourse constituent (see
Asher, 1993, among others). Therefore, such relations are based on principles that are not
adaptable to constraints conveyed by connectives. An approach that would aim to predict
both types of constraints might be able to take a big step forward in the study of the
parameters governing discourse well-formedness.
Second, the formal properties of connectives could be used as a starting point for
proposing some distinctions within the DM class. It might prove useful to adopt a
dynamic semantics framework in order to highlight the properties of other members of the
class, such as interjections (i.e., regarde) or modal particles (i.e., vraiment). The fact of
placing constraints on the left context could be, for instance, one of the properties that
might differentiate connectives from other DMs. By contrast, items such as regarde and
vraiment are focused on the utterance in which they occur; they could perhaps be
described in relation to the kind of indication they give on updates.
Notes
I thank Eddy Roulet for his very helpful comments on an earlier version of this paper, and
I wish to express my gratitude to Corina Cojocariu for the assistance provided in the
writing process. I also thank Peter Machonis and Antoinnette Renouf for their very
careful proofreading.
1
The present paper represents a contribution to the project Semantic Typology and
Classification of French Connectives, funded by the Fonds National Suisse de la
Recherche Scientifique (request no. 610-062821). An initial stage of this research was
developed in Cahiers de linguistique française, 24.
2
The notion of accommodation is borrowed from Lewis (1979). Roberts (1996) uses it to
show how a particular modal discourse concatenation works. Schwenter (2001)
employs it to account for the dialogical use of Spanish additive connectives.
3
Carel and Ducrot adopt such a view from a different perspective, when they assume that
the so-called donc-clauses and pourtant-clauses are interdependent.
4
A different hypothesis is put forward in Jayez (2002). Since this analysis uses the
propositional content of the question, it is possible to envisage a real update performed
on at least one of the worlds pertaining to the information state under consideration. I
resort here to the representation adopted in Jayez and Rossari (1998), where the
answers are used to describe the way the link works.
Chapter 17
Discourse marker research and theory:

revisiting and
Deborah Schiffrin
1. Introduction
The study of what Robert Longacre (1976) aptly called “mystery particles” has
proliferated over the past twenty years. Words such as well, and, like, now, and y’know
have been studied by scholars from virtually all branches of linguistics (e.g., applied,
formal, computational, sociolinguistic, psycholinguistic, historical, developmental) and
have kept pace with the development of new approaches for the analysis of discourse
(e.g., corpus linguistics) and new paradigms in both semantics (e.g., cognitive semantics)
and pragmatics (e.g., Relevance Theory). The range of languages in which such terms
have been examined is typologically diverse, including, for example, Chinese, Danish,
French, Hebrew, Indonesian, Latin, and Mayan. Attention has been focused on both
synchronic patterns (within/across speech situations and language contact situations) and
diachronic change (first and second language acquisition, grammaticalization). Given so
wide a range of theoretical and analytical diversity, perhaps it should not be surprising
that there has not yet emerged a consensus in some of the basic tenets of discourse marker
research or theory.
After introducing my approach (section 1.1), methodology (section 1.2), use of data in
this study (section 1.3), and problem to be addressed (section 1.4), I present my definition
of discourse markers (section 2). I then use a brief analysis of and to explore one of the
central problems in discourse marker research—their functional spectrum (section 3). My
conclusion fits the analysis into a more general model of discourse markers (section 4)
and links both analysis and model to broader issues in the study of language (section 5).
1.1. Approach
The basic components of my approach are meaning (semantic, pragmatic), discourse, and
function. Although all of these are common terms in linguistics, each is itself polysemous,
evoking a range of meanings that are embedded in what are sometimes very different
approaches to, and goals of, linguistic inquiry.
316 D. Schiffrin
In his two volume reference book Semantics, Lyons (1977) begins by illustrating over a
dozen meanings of the word meaning. Only two are usually taken as falling within the
scope of linguistic theory and analysis: sense and reference. The sense of a word is rooted
in linguistic knowledge and stems from relations among words themselves. It is generally
assumed that we share the sense of words through our knowledge of the networks of
meaning in which words are embedded and through membership in a speech community.
For example, we would share knowledge of the semantic relationship that links fruit and
pear (hyponomy). Reference is a relation between language and something in the world.
In keeping with Morris’ (1938) view of semantics as the study of how signs are related to
the objects to which they are applicable, the study of reference also falls within the
linguistic subfield of semantics. Much current work in semantic theory links the study of
sense and reference by focusing on truth-conditional meaning, i.e., formally specifying the
conditions that would have to hold for a proposition to be true.
Some markers (e.g., and) are homonymous with words whose semantic meaning is based
on their logical properties, hence, on their contributions to the conditions under which
propositions would be true. Although this allows a small group of markers to be
incorporated as discourse operators into formal models of discourse processing (e.g.,
Polanyi, 2001), other markers contribute meaning in ways other than through truth
functions. Markers may contribute semantic meaning to discourse through metaphorical
extensions (e.g., now and then; Schiffrin, 1990). Markers of speaker stance may develop
through loss of literal meaning (e.g., Brinton, 2003, on I mean; Karkinnen, 2004, on I
think) and may be classified as markers of pragmatic commentary rather than discourse
markers per se (e.g., Fraser, 1990).
Unlike sense and reference that remain relatively stable across speaker and situation,
pragmatic meanings vary across speakers and situations. This dependence can be captured
by defining pragmatics as the study of the use of context to make inferences about
meaning (Fasold, 1990: 119; see Levinson, 1983: ch. 2; Leech, 1983; and Mey, 2001).
Added to the contextual dependency is an inferential component: pragmatic meanings are
derived from a small set of interpretive principles (Grice, 1975) that use information from
a wide and varied range of possible contexts (e.g., text, common knowledge,
interpersonal relationship, social situation) to allow hearers to draw inferences about
speaker’s communicative intentions (i.e., Grice’s meaning-nn; Schiffrin, 1994a: ch. 4).
Because pragmatics presupposes context, it privileges the study of actual samples of

language use rather than the study of hypothetical examples of language use. Thus the
study of pragmatic meanings turns attention away from language as an abstract
representational system to concrete instantiations of language in utterances, i.e.,
verbalizations/inscriptions by a speaker/writer for a hearer/reader in a context.
The entry of context into the study of meaning also leads us to the analysis of discourse:
we do not produce utterances in isolation from other utterances. This shift to a larger unit
of analysis creates other challenges, stemming primarily from the scope and diversity of
discourse theories, approaches, and methodologies (Schiffrin, 1994a; Schiffrin, Tannen,
and Hamilton, 2001). It is thus helpful to separate discourse analysis into three separate
(but interrelated) foci of inquiry:
Discourse marker research and theory: Revisiting and 317
1. Within a sentence (or other syntactic, informational and/or prosodic unit): how “core”
parts of grammar (including morphemes, lexemes, phrases, clauses, sentences) and
“marginal” parts of grammar (e.g., intonation, prosody, information structure) are
related to the larger textual units in which they occur and the context that those texts
co-constitute.1
2. Texts: how sequences of sentences (or informational and/or prosodic units) are linearly
and hierarchically structured; what makes them cohesive and coherent; how types of
texts (e.g., stories, descriptions, lists, arguments) are differentiated and what their
defining characteristics are.
3. Contexts: how language is part of larger systems of meaning and practice, including
those embedded in concrete situations of face-to-face interaction, social gatherings,
societies, ideologies, and cultures—the “work” that language and other semiotic
systems accomplish in all areas of our lives.
The three foci are related: whereas the first focuses on sentence-level units (words,
phrases, and so on), the second and third move to larger units within which the smaller are
embedded. Thus, beginning at any one focal point requires attention to the others.
The final part of my approach is function. Although functions usually reflect recurrent
use, they are not the same as use per se. Whereas there is no inherent relationship between
one use and another, functions are related to one another: they are located within a system
or organization in which they connect to one another and to the larger system. Thus words
(at the lowest level of discourse analysis) are related to each other through their position
in a network of meanings and through their recurrence in the larger systems to which they
contribute (texts and contexts). What is thus at issue is an abstract system in which
utterances (or parts of utterances) are related to one another within a speaker/hearer-based
system of text/context that enables the production and reception of meaning(s).
In my model of discourse (Schiffrin, 1987a: ch. 1), I proposed several domains within
such a system.2 What is within these domains, as well as the relations between them,
provides the system within which markers function. An information state concerns what
speaker and hearer know: their organization and management of knowledge and meta-
knowledge. A participation framework focuses on the more social side of speaker and
hearer: their identities, alignments, relationships to each other and to what they are saying.
Acts also relate speaker and hearer. However, because they require structured knowledge
about what counts as a particular act and have somewhat constrained sequential
contingencies, I separate an act structure from both information state and participation
framework. Likewise, I consider an exchange structure—the organization of turns at
talk—to involve interactional contingencies that are at least partially unique to the
distribution of speaking/hearing rights. Finally is an idea structure—the most semantic
structure—involving not only propositions but also topic/comment and information status.
Relationships within these domains, and between them provide the system within which
markers function as indexicals (see section 5).
The functions of markers are very similar to their pragmatic meanings. Both are
embedded within, and dependent on, text/context as sources of their systematic
contribution to the structure, significance, and coherence of discourse. What differs is that
functions are relational: they relate units within domains to each other and relate domains
themselves to each other (Schiffrin, 1987a: ch. 10). Pragmatic meaning is based upon
318 D. Schiffrin
speakers’ recurrent use of a marker to convey a communicative meaning that depends

upon the relational functions of markers in text/context.
Now that I have described some basic features and terms of my approach, I turn to my
methodology—the more empirical side of the study of discourse markers.
1.2. Methodology
My analyses of discourse markers require, first and foremost, attention to actual uses of
markers in spoken discourse. After choosing a corpus, I identify all occurrences of the
lexical item that are potential appearances of the discourse marker and then decide which
are discourse markers. For analysis of and, for example, I exclude and only when it
appears within one continuous intonation unit, thus including all other occurrences. Once
I have identified the tokens to be examined, I analyze the discourse sequences in which
the tokens appear, as well as other occurrences of the marker, in order to balance
sequential and distributional accountability (Schiffrin, 1987a: 69-71).
Sequential accountability is an attempt to account for the occurrence of a marker within

an ongoing emergent discourse. Sequences (of idea units, intonation units, utterances,
turns, clauses, sentences, and so on) arise from a complex interplay among different levels
of structure and significance that come to be taken as “ongoing discourse.” As we talk to
one another, each contribution to a sequence is both anaphoric and cataphoric: it reflects
prior context and helps create upcoming context. Utterance-initial items have an even
more analytically and functionally privileged position in this co-constitutive view of
utterance and context. As argued by a variety of scholars (e.g., members of the Prague
school of functional grammar (Firbas), systemic-functional grammarians (e.g., Halliday)
the first part of a sentence often plays a critical role in conveying (or creating) a
relationship or revealing a dependency on prior text (see also Ward and Birner, 2001).
Line by line analysis of the progression of a sequence thus provides a means through
which to understand the marker’s contribution to the structure and significance of a
particular discourse.
Distributional accountability is an attempt to explain all occurrences of a marker within a

corpus, or more modestly, a set of markers within a subset of sites within the corpus.
Ideally, such an explanation would account not just for occurrences of a marker, but also
for its variable appearance (or nonappearance) in expected sites. Because a marker is
typically not limited to one particular type of sequence (let alone to the particularities of
just one sequence itself), distributional analyses also help us avoid elevating a particular
use of a marker to the status of a general function. Examining a given marker wherever it
occurs thus balances the specificity of a sequential analysis with generality: an
explanation of why a marker occurs in one slot should be related to an explanation for
why it occurs in another slot and why it does not occur in other slots.
The interdependence between distributional and sequential analyses also flows in the
other direction. Finding a distributional pattern depends on line-by-line analyses and then
on the classification of the results of such analysis, i.e., sequential environments, as
“slots” (the sites referred to above) in discourse. For example, when I was interested in the
relationship between markers and turn taking (Schiffrin, 1987a), I examined the presence
and absence of markers in two sequentially defined slots: turn-initial and turn-final
position. My interest narrowed to the use of markers in turn-initial position, specifically,

to the interplay between the semantic/pragmatic meaning of different markers (e.g., but
vs. well) and to whether the onset of next-speaker turn transition overlapped with prior
speaker, and if so, where in prior speaker’s turn such overlap began. I then refined my
coding to include how turn-onset fit into prior turn-transition spaces.
In sum, identifying slots in which markers occur (or do not occur) is crucial to my
approach. Yet these slots must be based on sequential analyses of markers throughout a
corpus. Given the diversity of words that serve as markers and the many different levels
of structure and significance that emerge in ongoing discourse, however, it should hardly
be surprising that the identification of slots varies for different markers. The multiplicity
of potentially relevant slots not only highlights the importance of balancing sequential and
distributional accountability. It also suggests that a model of discourse markers should
always remain heuristic (section 4); although it can point us toward the general parameters
of discourse slots, a model must remain adaptable enough to incorporate different
meaning(s), use(s), and function(s) that become apparent through sequential and
distributional analyses.
1.3. Data
Closely linked to both approach and methodology are data: what are the texts and contexts
in which a marker is analyzed? how does the data provide different slots and sequences in
which markers can occur?
Data for my analysis are interviews from two sources: a sociolinguistic interview in a
Philadelphia neighborhood and an oral history interview with a Holocaust survivor.3 The
purposes of these two types of interviews differ. The research goals of sociolinguistic
interviews are to understand linguistic change and variation in a speech community; this
requires a large body of stylistically varied speech from people whose social
characteristics vary (Labov, 1984; Schiffrin, 1987a: ch. 2). Although oral histories are
also research interviews (they provide data for historians, sociologists, anthropologists,
and psychologists), they have additional goals of autobiography and public
commemoration (Schiffrin, 2002). Despite their differences, both types of interviews rely
on relatively open-ended agendas in which interviewers pose not only questions that elicit
specific information but also questions that open a topic of talk and prompt respondents to
expand on their own topics.
Interviews contain a variety of text types, e.g., stories, descriptions, explanations,

assessments, arguments. I chose lists as the initial data for analysis because both their
meanings and their structures—both of which are crucial to the analysis of and, the
marker in which I am interested—are relatively transparent. An operational definition
includes the following (Schiffrin, 1994b): spoken or written texts in which (i) the parts of
a list are “items,” either entities or acts that are (ii) members of a larger set. Enumeration
of the set members (iii) typically occupies an extended turn at talk from one speaker (iv)
in which the main coherence source is the semantic connection among the items as set
members. This connection can be conveyed through (v) the use of repetition, ellipsis,
parallelisms, and the recurrent use of and and then. Thus, the central coherence relation
(Knott and Sanders, 1998) of lists is membership in a set; the central structure is
320 D. Schiffrin
coordination of subunits as equal level branches of a larger overarching unit (see Polanyi,
2001).
Since the lists occupy a number of different turn spaces (e.g., they can be projected so as
to occupy a single turn at talk, their continuity can be prompted), I comment upon and in
turn initiation and turn continuation. Likewise, since the two lists focused upon provide
answers to questions, I also analyze and in interview questions themselves. In keeping
with the methodology outlined above (section 1.2), I draw upon sequential analyses of
and in specific lists and distributional analyses of and within lists, turns, and questions.
As explained in previous sections, my approach to the functional spectrum of markers

depends upon analysis of the interplay among meaning, function, and discourse
(text/context). I explore this interplay through an analysis of one marker, and, in three
discourse sites: lists, turns at talk, and questions.
Prior research suggests that a n d connects structurally coordinate units. Yet the
coordinating function of and appears in a range of discourse environments and with a
variety of units, as noted not only in my own work (Schiffrin, 1987a: 128- 152) but also in
diachronic studies of and (Cotter, 1996) and studies of language development. According
to the latter (e.g., Sprott, 1992), the first site in which and appears is during exchange
structures in children’s (2;7 to 3;6) disputes; added later are act and idea (first local, then
global) functions. By the time children at later stages of language development (3;6 to 9;6,
Peterson and McCabe, 1991) tell stories, and has gained a textual use that parallels adult
patterns (Segal et al., 1991): and links narrative events with each other more frequently
than with information tangential to the narrative plot.
Although the coordinating role of and predicts that units at the same structural level will
be connected by and, it does not tell us what kind(s) of units are coordinated in what
domain(s). Thus, one problem to be addressed is the identity of the units coordinated by
and. Other questions follow: where are those coordinate parts found? are they always
adjacent? if not, how are they identified?4
Prior research also agrees that and has meaning. However, attempts to assign meaning or
meanings to and have “been a moot point since antiquity” (Dik, 1968: 25; Posner, 1980).
The wide range of meanings that can be inferred when and connects propositions has led
many to reject a meaning-maximalist view of and in which multiple senses are housed in
one lexeme. Instead, what has become more prevalent is a meaning-minimalist view in
which a very simple meaning is supplemented by pragmatic inference (e.g., a maxim of
manner to provide an interpretation of temporal sequence).
The problem of meaning is exacerbated when and connects a range of units other than
propositions in units larger than and/or different from grammatial units like clauses. In
earlier work (Schiffrin, 1987a), I adapted Halliday and Hasan’s (1976) view that and has
an additive meaning that works like other cohesive devices (such as repetition and lexical
collocations) to presuppose a prior text. Yet unlike other cohesive devices, and can also
be used metaphorically, to convey not only “external” relationships warranted by
referential meaning but also “internal” relationships perceived or attributed by the speaker
(Halliday and Hasan, 1976). Included in the latter are relationships between units other
than propositions. But given the wide range of possibilities that can be “added” together,
how do we know what earlier part of discourse is to be the first part of the “sum”
presupposed by the and-prefaced utterance?
The structural and interpretive problems raised by and are actually two different facets of
one problem: what is the textual anchor for an and-prefaced utterance? Resolving this
problem requires finding a way to identify units that can be added together, presumably,
because they share some quality whose combination is important to the coherence of the
discourse.
2. Definition
The different labels for what I am calling discourse markers are not just alternative words
for the same thing: they reflect different ways of thinking about the organization of what
ends up being different sets of words and expressions. For some, the underlying unity is
pragmatic function; for others, it is role in discourse. Some scholars find the term markers
to presuppose a preexistent meaning that is linguistically indexed, suggesting instead that
the term particles allows for words that create meanings to be added to the utterances. The
consequences of different labels are thus both practical and theoretical. On a concrete
level, very different items can end up being accounted for within an analysis. At a
theoretical level, the inclusion of different expressions can represent a reliance on
different unifying principles, some formal, some functional.
In my initial work on discourse markers (Schiffrin, 1987a), I defined discourse markers in

two ways: an operational definition that allowed identification and (if possible)
measurement (e.g., at the relatively low level of “present” or “absent”) and a theoretical
definition that located markers in a conceptual framework. At an operational level, I
defined discourse markers as sequentially dependent elements that bracket units of talk,
i.e., nonobligatory utterance-initial items that function in relation to ongoing talk and text
(p. 31). I proposed that discourse markers comprised a set of linguistic expressions from
word classes as varied as conjunctions (e.g., and, but, or), interjections (oh), adverbs
(now, then) and lexicalized phrases (y’know, I mean). I also proposed a heuristic discourse
model with different domains: a participation framework, information state, idea structure,
act structure, exchange structure (see section 1.1).
Although I had initiated and developed my analysis of markers with an operational

definition, I concluded with a more theoretical definition. First, I tried to specify the
conditions that would allow a word to be used as a discourse marker: syntactically
detachable, initial position, range of prosodic contours, operate at both local and global
levels, operate on different planes of discourse (Schiffrin, 1987a: 328; see Jucker and Ziv,
1998: 1-4). Second, I suggested that discourse markers varied in terms of their
propositional meanings: whereas the functions of some markers (e.g., I mean, y’know)
were based on their propositional meanings (some involving metaphorical extension over
time and across domains, i.e., well (Jucker, 1997), other markers (e.g., oh) had no
propositional meaning.
Finally, I suggested that discourse markers were comparable to indexicals (Schiffrin,

1987a: 322- 325; cf. Levinson’s, 1983: ch. 2, on discourse deictics), or in a broader
322 D. Schiffrin
sociolinguistic framework, contextualization cues (Schiffrin, 1987b). Like other

indexicals whose “pointing” function delimits a contextual realm (i.e., space, time, or
person), I claimed that markers have primary domains within which they establish
coordinates. My analyses also showed that markers could connect utterances either within
a single domain or across different domains. The use of markers in ongoing discourse—in
which there is always more than level of structure and significance—expands their
domains and creates what appears to be multifunctionality, i.e., one marker contributes to
more than one discourse domain. I suggested that this functional range—establishing
coordinates in different domains of discourse—helps to integrate the many different
simultaneous processes underlying the construction of discourse and thus helps to create
coherence. Subsequent work that focused specifically on then (Schiffrin, 1992) expanded
upon the indexical property of markers by showing how the deictic meaning of then
provides a template not only for meanings within discourse (successive, epistemic), but
also for grammatical (aspectual) meaning. In section 5, I pursue the indexical properties of
markers more fully and suggest that markers are a subclass of indexicals.
3. Analysis of functional spectrum
In this section, I illustrate how my approach, methodology, and model fit together by
analyzing and in two lists, in the exchange of turns, and with two types of questions. The
analysis suggests that and has one meaning (additive) that is essential to interpretation of
why prior and current utterances can be treated cumulatively. The additive meaning of
and combines with its structural status (as a coordinating conjunction) to have one basic
function (“continue a cumulative set”). This function contributes to the process of
constructing discursive sequences whose smaller parts combine to form a larger structure.
Different uses of and (e.g., continue a list, continue a topic from a prior answer) are
grounded in the specific sites in which and appears and is interpreted.
My analysis begins with two lists. Lists typically have a clear semantic structure in which
items have both coordinate links (e.g., between X1 and X2) and hierarchical links (e.g.,
between X1 and X1A). Thus we can use lists to learn more about the semantic and
structural bases of and. Like most discourse units, however, lists do not appear on their
own: they emerge in concert with other means of text/context organization, e.g., within a
turn-taking system and within sequential structures including (but not limited to)
adjacency pairs. By attending to the emergence of lists within turn exchanges and
adjacency pairs, we can examine the role of and not just in idea structures (the semantic
relationships between items (set members) in a list) but also in exchange structures (the
management of turns at talk) and act structures (the asking and answering of questions).
This is important because when and does appear at the intersection of simultaneously
emerging structures, the constraints impacting it do not always converge or complement
one another. Instead, they may diverge or even conflict with one another.5
I analyze and in three different subsections. In section 3.1, I present a monologic list to
show the role of and with coordinate list items; the use of and in turn taking is briefly
mentioned. In section 3.2, I turn to a more dialogic list in which an interviewer’s
questions co-construct (and alter the structure of) the interviewee’s list and introduce
additional turn-taking environments. In section 3.3, I turn to the use of and with
interviewer questions. Since each subsection raises issues to be pursued in the next, the
analysis not only tells us about and but also illustrates methodological and theoretical
aspects of my approach to markers, especially their close interdependency and mutual

reliance on data.
3.1. And in a list of race tracks around here
In this section, I discuss a list in which Kay presents the race tracks near her house in
response to a tag question from Anne (a sociolinguistic interviewer) about the popularity
of racing. Five features of the list are relevant to analysis: (i) the list covers a closed set of
items (RACE TRACKS AROUND HERE), (ii) its linear order matches its hierarchical semantic
organization, (iii) it follows a depth-before-breadth order, (iv) it is relatively monologic,
and (v) it occupies an extended turn at talk. The Roman numerals and letters on the left of
Kay’s list indicate the organization of items in the list. These help us see the depth-before-
breadth structure: a superordinate item [X1] is presented and expanded with subordinate
items ([X1A], [X1B] . . .) before the next superordinate list item is presented. I use ____
to indicate a list item not prefaced by and.6 We will examine how idea structure and turn
taking are reflected in the distribution of and.
(1) RACE TRACKS AROUND HERE

X LOCAL RACE TRACKS Anne: (a) Racing’s big around here, isn’t it?
Kay: (b) Yeh.
Anne (c) Yeh.
X1 RACE TRACKS IN NJ Kay: (d) Well, you got uh, Jersey.
X1a (e) ___You got . . . Monmouth
X1b (f) and you got Garden State.
X1c (g) ____Y’got Atlantic City.
Anne: (h) Mmhmm.
X2/X2a RACE TRACK IN PA Kay: (i) And then uh here you got Liberty Bell.
X2b (j) And they’re building a new one up
in Neshaminy.
Anne: (k) That’s right. [I’ve never seen that, =
X3/X3a RACE TRACK IN DE Kay: (l) [And uh... you got=
Anne: =[though.
Kay: =[Delaware.
X4 RACE TRACK IN NY (m) And of course, if you want to re-
be- really go at it you can go up to New
York.
Anne: (n) Mmhmm.
X4a Kay: (o) =____You got Aquaduct
X4b (p) and you got Saratoga
X4c (q) and you have that Belmont, y’know.
Table 1 summarizes the use of and in the list. The subordinate column includes the two
list-items MONMOUTH [X1A] (e) and AQUADUCT [X4A] (o) that branch from (and
sequentially follow) their superordinate list items JERSEY [X1] (d) and NEW YORK [X4]
(m). The coordinate column includes list items at the same level, either upper-level items
(X1, X2 . . .) or lower-level items (X1A, X1B . . .).
324 D. Schiffrin
Table 1. And in list (1)
Subordinate Coordinate Total

and 0 7 7
“zero” 2 2 4
Total 2 9 11
The distribution of and in the list is largely predicted by its semantic structure. 78% (7/9)
of the same-level items are and-prefaced. Neither of the two different-level list items in
the Subordinate column is and-prefaced.
Although the figures in Table 1 provide distributional evidence of the importance of

semantic structure, we can also explore the importance of this domain in more sequential
terms. Let us start by considering the absence of and with the items branching from
upper-level nodes of the list: MONMOUTH [X1A], AQUADUCT [X4A].
Note first that since the items in the list are place names, an understanding of their set
relationships requires geographical knowledge that may not be available to everyone
hearing or reading the list. This means that their conceptual relationship is not as
semantically or lexically explicit, as the listing of generally familiar categories (e.g.,
family members). As suggested by Pons (this volume), if ranking between sentences is
not made semantically or lexically explicit, then and indicates that both constituents have
the same status. Notice, then, that if MONMOUTH were and-prefaced, we would interpret
JERSEY as a default textual anchor:
(1d) Well, you got uh, Jersey.

(1e) (?and) you got . . . Monmouth.
What and would thus convey is that MONMOUTH is a same-level item in the list as JERSEY
(i.e., another state), hence not a subordinate member of the set RACE TRACKS IN JERSEY
[X1] at all.
The next slot in which and does not occur brings up the important issue of multiple (and
thus convergent or divergent) constraints from different discourse domains. When
introducing AQUADUCT [X4A] (in You got Aquaduct (o)) after Anne’s turn continuer
(mmhmm (n)), Kay does not use and. Since AQUADUCT is the first lower-level item of
[X4], it is not surprising to find that You got Aquaduct is not and-prefaced. This absence
becomes more analytically interesting, however, when we consider its turn-taking
environment.
Anne recurrently follows a strategy common in sociolinguistic interviews: her mmhmm

and that’s right work as turn continuers that pass responsibility for the floor back to Kay
(cf. Jucker and Smith’s (1998) view of turn continuers as reception markers). Yet Kay
does use and after Anne’s other turn continuers: And then uh here you got Liberty Bell (i),
And uh . . . you got Delaware (l), And of course, if you want to re- be- really go at it you
can go up to New York (m). If Kay is continuing her turn in all four cases, why does she
use and only in the latter three? This puzzle is solved by returning to the semantic
structure of the list. Although the turn space occupied by You got Aquaduct is consistent
with the use of and (Schiffrin, 1987a: 143-146), the idea structure is not. It is only when
the list items following Anne’s turn continuers are at the same structural/semantic level of
the list ([X1] and [X2], [X2] and [X3], [X3] and [X4]) that they are and-prefaced.
I will use one more example from the list to support the importance of semantic structure
for and-prefacing in lists: the absence of and with the same-level list item ATLANTIC CITY
[XIC]. Although ATLANTIC CITY is certainly a part of the subset [X1] TRACKS IN NEW
JERSEY, it has a different status. Whereas [X1A] and [X1B] are New Jersey race tracks
that are not named after towns, [X1C] shares its name with the well known resort near
which it is located. So the switch from and to “zero” is a textual switch that iconically
reflects the different way that ATLANTIC CITY fits into the overall set. (See Schiffrin,
1994b, 1987a: 129-32, for other examples of this; see also Maschler, 1998, on discourse
structuring devices that mark more difficult transitions).
In sum, the general distribution of and in the list about race tracks reflects its semantic
structure. This distribution suggests that and-prefacing of list items builds upon the
grammatical role of and as a coordinating conjunction (Schiffrin, 1987a: 182-190).
Distributional and sequential observations support this conclusion. And is used in the very
same turn-taking environment (after the other’s turn continuer) only if the list item in that
next turn is semantically coordinate with the list item from the prior turn. And and “zero”
create sequential contrasts that differentiate typical from atypical list members.
3.2. And in a list of something about yourself now
In this section, I discuss a second list, from the opening of a video taped oral history
interview with a Holocaust survivor, Susan Beer (SB).7 This list is longer and more
complex than the list about race tracks. Its topic is potentially broader and open ended: the
interviewee (SB) is asked to tell the interviewer (IVer) something about yourself now. The
IVer’s first question provides a breadth-before-depth structure in which lower-level items
([X1A], [X1B] . . .) that are part of one upper-level item [X1] are expanded before
opening the next upper-level item ([X2]). Later questions continue to build upon this
structure to co-construct SB’s list. Because the list analyzed here is more dialogic, sites of
participant co-construction create mismatches between sequential presentation and
hierarchical structure.
(2) SOMETHING ABOUT YOURSELF NOW
1. IVer: [LOOKS AT CAMERA] I’m Dr. Donald Freidheim.

2. [LOOKS AT CLIPBOARD, UP AND DOWN]
3. [LOOKS AT CAMERA]
4. This afternoon I’m interviewing Mrs. Susan Beer.
5. Mrs. Beer is a survivor from Czechoslovakia,
6. and we’re privileged to hear her story today.
7. [TURNS TO FACE SB] Hello. [How are y’ Nice to see you
8. SB: Hi. How [are you uh:
9. IVer: ___I’d like you to tell me a li- something about yourself now. [slow]
10. Your . . . family and . . .
11. SB: Mmhmm.
326 D. Schiffrin
12. Uh I’ve been living in Cleveland for the last 36 years.

13. IVer: mmhmm
14. ___ I uh at the present time uh I am a housewife,
15. and uh uh occupy myself uh uh sometimes helping my husband, with his
office, when needed [IN BREATH]
16. IVer: What does he do?
17. SB: He’s a podiatrist.
18. IVer: uhhuh
19. And uh other times, I pursue, uh really uh . . . um . . . things that I enjo
20. um going to the museum,
21. and swimming,
22. and uh visiting ill people,
23. and uh um spending time uh decorating my home,=
24. IVer: mmhmm
25. SB: =and that’s about . . . [NODS]
26. IVer: ____May I ask how old you are?
27. SB: Yes, I’m sixty years old.
28. IVer: Mmhmm. Sixty.
29. SB: Mmhmm.
30. IVer: What I’d like to do first,
31. oh- d- lemme ask- if you have-you have children?
32. SB: Yes, I have two children.=
33. IVer: mmhmm mmhmm
34. SB: =I have a son,
35. who is thirty three.=
36. IVer: mmhmm
37. SB: =And I have a daughter,
38. who is twenty seven.
39. She’s married
40. and lives in New York.=
41. Iver: I see.
42. SB: =And um she uh studied journalism
43. but uh works as a public relation person.
44. IVer: mmhmm.
45. And what does your son do?
46. SB: Uh he’s a-
47. in . . . in Wooster
48. and um . . . doesn’t do very much really.
49. IVer: mmhmm
50. ____D- does he have a family?
51. SB: No.
52. IVer: ____Lemme ask you to go back to, the years, before the war,=
53. SB: okay
54. IVer: =in the- in the 1930’s, let’s say [about the mid 1930’s,=
SB: [Right. okay
55. IVer; =and um describe a little bit about your experiences then, [what-=
56. SB: o[kay
57. IVer: where you live:d,
58. and something about your family.
SB’s list structure is co-constructed by her own orientation to personal information and
the overarching structure presented by the IVer and reinstated by his questions. These two
participant structures create both semantic and turn-taking environments for and. These
participant structures create an interactional sequence in which the hierarchical list
structure in Fig. 1 is co-constructed.
The IVer’s initiating question establishes the first part of a binary distinction between
NOW [X1] (post WWII current life) and THEN [X2]. Consistent with the breadth-before-
depth structure, two subtopics of NOW are also introduced: YOURSELF [X1A] and YOUR
FAMILY [X1B]. SB speaks about herself [X1A] in lines (12) to (16) and lines (20) to (26).
[X1A] branches ([X1AA] to [X1AD]) to include where SB L I V E S , HOW LONG,
OCCUPATION (housewife), and ACTIVITIES . The latter, ACTIVITIES [X1AD], branches
further to SOMETIMES and OTHER TIMES. The SOMETIMES activity is not expanded beyond
helping my husband (which is also the first introduction of a family member [X1B]). The
OTHER TIMES activity [X1ADB] is specified as THINGS I ENJOY. This list item branches
further into four subtypes: GOING TO MUSEUMS, SWIMMING , VISITING ILL PEOPLE,
DECORATING HOME.
Although I have thus far been describing the list as constructed by SB alone, the IVer asks
five questions during SB’s list that bring up topics from levels in the list structure higher
than the items in just-prior talk. It is for this reason that we need to examine the presence
or absence of and not only in SB’s list but also in the IVer’s questions (in lines 16, 26, 30,
44, 49; see section 3.3). We will also examine how SB returns to her own list structure.
Although some of the questions (lines 16, 44, 49) build on what SB has just said, they all
bring up topics from levels higher in the list structure than the adjacent items. SB answers
the IVer’s questions, and in doing so, deviates from her own list expansion. But she then
returns to her own emergent list structure. For example, after answering What does he
do? with He’s a podiatrist (line 18), SB returns to the distinction between SOMETIMES
[X1ADA] and OTHER TIMES [X1ADB] that she had been expanding. Her return is prefaced
by and: And uh other times, I pursue, uh really uh . . . um . . . things that I enjoy (line 19).
Thus SB’s use of and to return to her own list structure illustrates an intersection between
idea structure and turn taking similar to that seen in Kay’s list about race tracks. Like Kay,
SB used and after the IVer’s turn continuers (and question) only when she was returning
to a coordinate item in her own list structure.
Although we have noted occurrences of and in passing, let us now examine how the co-
construction of SB’s list has a bearing on the use of and. Table 2 summarizes the presence
and absence of and in the list.8 Since the IVer’s questions helped co-construct SB’s list, I
include them with SB’s list items in Table 2. However, I then separate them and discuss
them on their own.
Table 2 shows that the overall distribution of and in the oral history interview list is
similar to its distribution in the sociolinguistic interview list. For example, the absence of
and when shifting to a lower-level item (0 cases of and in the Subordinate column)
reflects semantic structure. Consider the change in interpretation if and were present in
“zero” in this environment (shown in (3) below):
328
Figure 1. List structure: From sequence to hierarchy
X tell me something
X1 about now X2 about then
X1a you X1b family X2a you X2b family
X1aa where... ab how long... ac job... ad activities... ae age X 1ba husband X1bb two children
X1bb job
X1ada X1adb X1bba X1bbb
sometimes other times son daughter
X1adaa X1adbb X1bbaa X1bbab X1bbac X1bbad X1bbba X1 bbbb X1bbbc X1bbbd X1 bbbe
help husband enjoy things age job where children age married where edc job
X1adbba X1adbbb X1adbbc X1adbbd

go to swim visit ill decorate home
museum
D. Schiffrin
Table 2. And in co-construction of list (2)
Subordinate Coordinate Total

and 0 9 9
“zero” 6 7 13
Total 6 16 22
(3) line 19 of (2)

And uh other times, I pursue,
uh really uh . . . um . . . things that I enjoy
[?and] um going to the museum,
Without and, we use our knowledge of the world to correctly infer SB’s intended
relationship between “things I enjoy” [X1ADBB] and “museum” [X1ADBBA]; likewise,
for “children” [X1BAB] and “son” [X1BABB]. And would disallow these readings and
provide radically alternative, and confusing, readings: we would infer that going to the
museum is not something that SB enjoys (or even more dramatically that a son is not one
of SB’s two children (line 34)).
When we examine more closely the use of and with same-level items from SB’s list, we
find a surprisingly slim majority (56% (9/16)) in the Coordinate column. As we see in
Table 3, however, the coordinate-level uses of and are heavily skewed toward SB’s list
items, with only one appearing as a preface for an IVer question.
Table 3. And in different participant sites in list (2)
SB’s list items IVer’s Q items Total

and 8 1 9
“zero” 2 5 7
Total 10 6 16
Whereas SB uses and to preface 80% (8/10) of the same level list-items that she presents,
the IVer uses and to preface only 17% (1/6) of the questions (and what does you son do?)
that open a slot for list items.
This difference highlights the different orientations that SB and the IVer have to the list
and its role in the interview. Whereas SB is organizing and providing autobiographical
information to answer a question, the IVer is eliciting another’s biographical information
in order to fulfill the goals of an interview. Thus each participant is working from a
different information state: SB from the facts of her own life, the IVer from a general
interview template. SB and the IVer also occupy different positions in the act and
exchange structure: the IVer’s turns are focused on asking questions; SB’s turns, on
answering questions.
330 D. Schiffrin
These intersections of the two different participant orientations create two different sites
to examine and. First is the IVer’s question. SB’s list items evoked two questions from
the IVer: about husband’s job [X1BAA], and about son’s job [X1BBBB]. It was only
when the question built upon SB’s most recent list item to seek information that it was
and-prefaced. We discuss this in section 3.3. Second (in the exchange structure) is when
SB returns to the floor after either an embedded question/answer sequence or the IVer’s
turn continuers. Like Kay’s use of and in her list, and prefaced a next-turn SB’s list item
only when that item was coordinate with the item from her prior turn (compare line 34
with lines 19, 37, and 42).
In sum, the oral history interview list has allowed us to examine how two different
participants in an interaction orient to the construction of one list. Whereas the IVer asks
questions that elicit different parts of the list from SB’s personal biography, SB organizes
and provides autobiographical information within a framework partially evoked by the
IVer’s questions but also attendant to her own schema. These different participant
orientations stem partially from the information state from which each began but also
become interwoven with the emergent semantic structures and the organization of turn
taking. Although this creation of more complex discourse sites complicates the use of
and, we have seen, again, the crucial impact of semantic structure on the use of and in
lists.
3.3. And with questions
In prior sections, we analyzed and in idea structures (of lists) and exchange structures
(turn taking). In this section, analysis of and in the IVer’s questions allows us to adds
another domain—act structure—to our analysis.9 We examine two types of questions.
Local/dependent questions depend on topics of adjacent talk: the IVer pursues a topic
from SB’s answer to a prior question, either by expanding SB’s topic or creating a step-
wise transition to a new topic. Examples (from section 3.2) are What does he do (line 17)
about SB’s husband and And what does your son do? (line 44) about SB’s son. Global/
independent questions need not be tied to adjacent talk. Rather, they are related to the
overall goals and guidelines of the interview, thus providing an overarching and higher-
level organization for the interview. An example is Lemme ask you to go back to, the
years, before the war, in the- in the 1930’s, let’s say about the mid 1930’s, and um
describe a little bit about your experiences then (line 59).
Figure 2 below shows the location of and in the two question types. Located in the Act
Structure are questions and answers. At the Global level, and connects the same speech
acts (questions). This relationship is similar to the connection between higher-level
coordinate list items. At the Local level, and connects two different speech acts (answer
and question) from different speakers. This relationship is reminiscent of a connection
between two different-level list items—which does not occur. It is thus the Local
connection that is puzzling: how can two structurally different units be connected by and?
Figure 2. And in two domains
Act structure
Global Local
IVer: Question-1 IVer: Question-1
IVee: Answer-1 IVee: Answer-1
IVer: and Question-2 IVer: and [Question-2 [Answer 1]]
Idea structure
Indepedent Dependent
IVer: Incomplete Proposition-1 IVer: Incomplete Proposition-1
IVee: Complete Proposition-1 IVee: Complete Proposition-1
IVer: and Incomplete Proposition-2 IVer: and [Incomplete Proposition-
2 [Proposition-1]
The answer lies in the Idea Structure. In both Independent and Dependent sequences, the
IVer presents an incomplete proposition; the IVee completes that proposition. What
differentiates the two idea sequences is what happens next. In the Independent sequence,
the IVer presents an incomplete proposition unconnected to what the IVee has just stated.
In the Dependent sequence, the IVer presents an incomplete proposition based on the
IVee’s completion of the former incomplete proposition. Thus what is connected by and
prefacing a question about a prior answer is not speech acts: it is propositions.
We are now ready to examine the distribution of and-prefaced questions in Table 4.
Table 4. And in questions in the oral history interview
Local/Dependent Global/Independent Total

and 18 1 19
other markers 20 3 23
“zero” 69 11 80
Total 107 15 122
And appears slightly more with local/dependent questions (17% (18/107)) than
global/independent questions (7% (1/15)). Although this seems to show a slight
dominance of idea (over act) structure, it contrasts with Heritage and Sorjonen’s (1994)
finding of a preference for and with agenda questions (cf. global/independent) in clinical
consultation interviews.
332 D. Schiffrin
When we examine the one an and-prefaced global/independent question in the interview,

however, we see that it blends features of the two question types. In (4), SB has been
answering the IVer’s question about THEN by describing her hometown and family life in
that town. She concludes with it was a small town life (line 96):
(4) Example of global/independent question
96. SB: And it was a comfortable life, it’s a- it was a small town life.
97. IVer: Who- how many did you have in your family? I [m-
98. SB: [Just myself.
99. IVer: You were the: only, [child.
100. SB: [Yes. Yeh.
101. IVer: And uh when were you born.
102. SB: I was born in 1924.
After SB closes the description of her town (line 96), the IVer’s question (line 97) returns
to an item from an earlier agenda question (describe . . . something about your family (line
59)). SB answers the question (line 98) and clarifies her answer (lines 99-100). The IVer
then asks another basic demographic question (And uh when were you born (line 101))
whose lack of connection with prior topic defines it as a global/independent question.
The sequential position of And uh when were you born, however, is reminiscent of a
second location for and-prefaced questions (Heritage and Sorjonon, 1994): a strategic use
of and to normalize contingent questions or problematic issues. This use of and extends
Halliday and Hasan’s (1976) external (or metaphorical) meaning: rather than being
manifest in text, what provides the additive relationship is the speaker him- or herself.
Notice, then, that asking about date of birth is not itself problematic. But asking about it
after SB has already been talking about her later life disturbs the usual chronological order
not only of life itself but also of oral history versions of life. Thus it is the IVer who
establishes an additive connection through and to routinize a question that is out of place
in the overarching organization of the interview.
We saw in earlier sections that idea structure was the most important constraint on and.
Analysis of the questions in the Heritage and Sorjonen (1994) corpus by Matsumoto
(1999) suggests the relevance of idea structure in that corpus: and-prefacing was more
frequent with questions linguistically oriented toward affirmative polarity. This skewed
distribution of and suggests an emergent idea sequence in which different speakers would
end up presenting congruent information.
To test this idea, I separated the local/dependent questions into two groups. Interrogatives
included yes/no- and wh-questions, both of which verbalize a choice between (or among)
options. Declaratives are statements with optional rising question intonation that requests
confirmation. Table 5 shows the presence and absence of and with these two forms of
local/dependent questions.
Table 5 shows that and prefaces 22% (7/31) of the declarative forms and 7% (11/76%) of
the interrogative forms. This preference suggests that and appears more when congruence
of ideas is sequentially implicated (and expected) across speakers.
Table 5. And with two forms of local/dependent questions
Declaratives Interrogatives Total

and 7 11 18
other markers 6 14 20
“zero” 18 51 69
Total 31 76 107
In sum, understanding the use of and with the interviewer’s questions in the oral history
interview has required that we separate two domains of discourse: idea and act. Separating
these domains has led us to ask the same questions asked in analyses of and in lists: how
can we account for multiple constraints? what happens when constraints converge or
compete with one another? The answer is also the same. Although we have added
question/answer pairs to the potential range of discourse sites in which and can appear,
the pattern of and-prefacing with questions in this oral history interview again highlights
the importance of idea structure.
3.4. Conclusion
In this section, I have analyzed the meaning(s), use(s), and function(s) of and in lists,
different turn-taking environments, and types of questions. The weakest pattern was with
IVer questions: why? The size of the structures is large (what’s being connected is
distant)—the link is hard to find. Also, speech acts are less semantically organized than
list items.
The analysis suggests that and has one meaning (additive) that combines with its
structural status (as a coordinating conjunction) to have one basic function (“continue a
cumulative set”). The sets can vary (ideas, turns, acts), as can the specific uses (e.g., add
an upper-level list item, continue a topic from a prior answer) that are produced when
different features combine to create highly specific discourse sites. Nevertheless, when the
domains underlying the discourse sites create divergent units that provide potential
“parts” of a set, the idea domain continues to constrain the use of and.
4. Model
In this section, I integrate the general results of the analysis into the model reviewed in
section 1.1. Recall that the model contained different domains. Relationships appeared at
two levels: units within each domain could be related locally and globally, and domains
themselves could also be related. The analysis of and in section 3 showed that local and
global relationships within one domain at a time can be indexed by and. These domains
can differ: and can link propositions at an idea level, questions at a speech act level, and
turns at an exchange level. Although and can occur at the intersection of different
domains, one domain was prioritized. The idea domain was most pertinent for the use of
and—a fitting outcome given the coexistence of and as a sentential conjunction.
334 D. Schiffrin
In my earlier problem statement (section 1.4), I noted that the structural and interpretive
problems raised by and are actually two facets of one problem: what is the textual anchor
for an and-prefaced utterance? Resolving this problem requires identifying units that can
be added together, presumably because they share some quality whose combination is
important to the coherence of the discourse. Although part of the problem has been
resolved—the default unit providing a textual anchor is idea—still not completely
resolved is which idea units, out of the vast number of those being put forth in discourse,
combine into a cumulative set.
Here I suggest that two pragmatic principles help account for which units in a single
domain can be related to one another and in what way. First is an Adjacency Principle.
Although this principle can be derived from the Gricean maxim of Manner or Quantity
(Grice, 1975), I base it on the Conversation Analytic (CA) injunction that utterances are
both context reflecting and context creating. Sacks (1973, Lecture 4: 11-12) describes the
importance of next-position for coherence:
There is one generic place where you need not include information as to which
utterance you’re intending to relate an utterance to . . . and that is if you are in
Next Position to an utterance. Which is to say that for adjacently placed
utterances, where next intends to relate to a last, no other means than positioning
are necessary in order to locate which utterance you’re intending to deal with.
The Adjacency Principle leads hearers to try to define a coherence relation between
adjacent utterances. Because Utterance-1 immediately precedes Utterance-2, it creates a
context for Utterance-2. Utterance-1 is thus the default location from which to define a
coherence relation with Utterance-2.
An Informativeness Principle helps define the relationship between idea units. This
principle, adapted from Levinson’s (1983: 146) maxim to “read as much into an utterance
as is consistent with what you know about the world,” helps us choose among different
possible coherence relations between utterances. Our knowledge of the world can warrant
increasingly strong relationships between propositions, even if those relationships are not
explicitly encoded. When these relationships are specified through words that encode
increasingly informative relations, the coherence relation between utterances is narrowed
down from the range of possibilities opened up by our world knowledge.
And appears in the following scale of informativeness, in which each item to the right in
(5a) provides more information (specified in (5b)), than the items on the left, about how to
connect two propositions:
(5) a. “zero” and then so

b. relevance addition succession consequence
The Principle of Informativeness allows an inference of succession without then, but the
then-prefacing of Utterance-2 encodes a ‘successive’ meaning. Likewise, going back in
the scale, we might infer ‘addition’ by mere adjacency (i.e., at the “zero” at the far left)
through world knowledge, but the use of and would encode ‘addition.’ The Principle of
Informativeness thus allows the inference of possible relationships between propositions
without discourse markers. What markers thus do is select a meaning from among those
potential relationships.
The role of pragmatic principles within the model also recalls the similarity between
pragmatics and functions noted in earlier discussion of my approach (section 1.1). The
pragmatic meaning of and is akin to its function: both are based on its semantic meaning
(‘additive’) in combination with its structural status (as a coordinating conjunction) to
mark the speaker’s communicative intention to “continue a cumulative set.” Both
pragmatic meaning and function are embedded within, and dependent on, the emergence
of text/context and the systematic ways in which parts therein relate to one another to
form more macrolevel structures and meanings. Pragmatic meaning thus contributes to the
indexing of relations within text/context.
5. Relevance
Analysis of and is relevant to two broader issues in discourse marker research and theory:
multiple functions of markers and indexicality. I have suggested that and has one
semantic meaning, many uses, and one pragmatic meaning/function. But this allocation
may differ for different markers whose sources are in different word classes or whose
text/context distribution differs. Thus we must include the possibility of input variance
among markers: the impact of meaning and discourse can vary across types of markers
and across individual tokens of those types. And this means that the functional spectrum
of markers can itself vary. Multiplicity may appear at lexical levels if a single discourse
marker has more than one meaning or function. Alternatively, if all markers have
single—but different—functions and it is only markers in toto that perform multiple
functions, multiplicity may appear only at the word class level.
Although multiple possibilities for multiplicity may seem unnecessarily complicated,

these functional layers make sense once we pursue more seriously the larger class within
which markers are situated: markers are a subclass of indexicals. The advantage of
viewing discourse markers as indexicals is that many of the features that seem so
worrisome—including, but not limited to multiplicity—are actually regular features of
deictic expressions.
Deictics provide indices to different aspects of context, most centrally to space, time, and
person. Yet this does not mean that a particular deictic expression cannot extend its reach
to another domain. Spatial indices, for example, commonly acquire temporal
interpretations. If we speak of “moving up” or “moving back” a meeting, we do not
literally mean that the meeting is a physical object to be moved vertically or horizontally
in space: it is a situation with a temporal onset that will now shift in linear time. The
indexical range of discourse markers is similar. Markers may have default contextual
“homes” in the particular domains of discourse to which they point. For example, some
markers point to an information state, others to an act structure, and still others to the
organization of ideas. But this does not mean that they cannot extend their reach as
different domains come into simultaneous play during a discourse or as the marker itself
is metaphorically extended over time (Schiffrin, 1987a: ch. 10; Sweetser, 1990, on see).
Another similarity is that deictics and markers both vary along a scale of proximity and
distance from a symbolic center (the unmarked version is the situation of speaking) in
336 D. Schiffrin
which the utterance is located. The proximal end of the axis for deictics is usually me in
the here and now of a physical world; the distal is you/they, there, and then. Proximity and
distance can also be located in a textual world. The common opposition between local and
global scope of markers (local includes adjacent utterances, global, distant utterances)
mirrors the proximal/distal axis.
It is quite possible that there are some default understandings of what counts as close to or
distant from a me in the here and now of a physical world (despite great cross-cultural
variation (Hanks, 1993; Levinson, 1983)) as well as comparable default parameters in a
textual world. Yet speakers and hearers can nevertheless manipulate what they take to be
proximal or distal in order to bring a broad range of entities into—or move them out
of—their perspective. For example, the personal proximal pronoun we can index a small
two-person with (Goffman, 1971) or an entire nation. Similarly, the marker oh can index a
change in information state evoked either by retrieval of a momentarily forgotten word or
by the understanding of a new, long, and complex algorithm. In both physical and textual
worlds, then, the problem of fixing the scope is the same.
Another similarity is that both traditional deictics and discourse markers can be treated
not only as an open class that allows temporary members but also as a class whose
members vary in their degree of core or peripheral membership. For example, here is
always a deictic, but he is not always a deictic; well is always a marker, but and is not
always a marker. For some scholars, nouns like neighbor have a deictic component
simply because they evoke someone who lives close to one’s home base. Likewise, north
and south are orthogonally fixed to one another, but whether we drive north to get to
Boston or south depends upon our starting point: is this deictic? Comparable questions
can be asked about the expressions that are like markers (e.g., I think) in some dimensions
but not in others (Karkkainen, 2004). The variability in terms of core and peripheral
status, then, suggests that both deictics and markers are porous: context can “leak” into
their meanings, their uses, and their functions in different degrees.
Finally, viewing markers as indexicals provides a way of breaking down two of the key
barriers in the definitional divide between markers and particles. First is the difference
between displaying (markers) and creating (particles) meaning; second is whether markers
(or particles) portray speaker stance and attitude.
The term marker often implies that a linguistic item is displaying an already existent
meaning; the term particle often implies that a meaning not otherwise available is being
added into the discourse. Yet deictics have a more complex relationship with context than
the one way path implied by either verb (display or add) used above: they select among
possible coordinates and possible ‘”centers” (points of reference) for those coordinates. If
I say, for example, I live here, the word here doesn’t tell you exactly what here I mean:
the room, the house, the neighborhood, the city, the country. The specific here depends on
many factors, including what we have been talking about before. Nevertheless, the
proximal meaning of here does fix one coordinate: if you know where I am physically
situated at the moment of speaking, you know that this place is located within the physical
parameters of where I live. Notice, however, that this whole set of assumptions can be
completely overridden if the deictic center shifts from the utterance to a map: I may say I
live here when pointing to a city (or street, or country) on a map even if I am not
physically situated at that place when speaking.
Like deictics, discourse markers can also select contextual coordinates from a range of
possibilities in their world—the text/contextual world—by shifting their center, i.e., their
domain. The distal meaning of then can convey temporal succession across episodes in a
narrative or succession of items in a list, both between adjacent utterances (local) or non-
adjacent utterances (global), as well as between a single utterance or multiple utterances.
Describing the principles by which a speaker chooses, and a hearer interprets, those
textual coordinates raises analytical problems parallel to the selection of a location in I
live here.
As noted above, markers—like deictics—can switch their “center.” For example, different
domains can serve as centers for production and interpretation of the same marker: so can
mark a transition from a warrant (information state) or from a turn (exchange structure);
okay can mark approval of an idea (participation framework) or agreement with a
proposed activity (act structure). Rather than display or create meanings, then, it may be
more accurate to say that, as indexicals, both deictics and markers select from a range of
possible meanings that depend on the domain and its point of reference. Speakers use both
deictics and markers to “display” their selection of a meaning from a possible range of
meanings. Because the verbalization of that deictic/marker makes explicit what had
previously been only one possibility from a range of possibilities, it can appear to be a
newly added or “created” meaning.
Another parameter on which markers differ from particles concerns the reliance of the
former on sequential units of discourse. Here I want to suggest that if we conceive of
discourse as sequences of utterances, i.e., text/context pairings (Schiffrin, 1994a: ch. 2),
then we can include not only relationships between units (e.g., acts, turns, propositions)
that typically appear in sequences but also relationships between aspects of text and
context. For example, self and other are clearly part of a context: they can have
relationships of solidarity, distance, and so on. The way a speaker is committed to (or
detached from) a belief is a relationship between self and content of talk. Of course the
self-other, and self-content relationships, are not necessarily sequentially organized parts
of discourse.10 But once we realize the centrality of self, other, and content to text and
context, what is said to be marked by particles—speaker/hearer alignment, stance—can be
said instead to be marked as relationships between parts of a discourse. To do so requires
recognizing self, other, and content as units of discourse—not utterances themselves, but
certainly part of the context that creates an utterance—and thus open to the same indexical
marking as other aspects of utterances.
Notes
1
I use the core/marginal distinction (but see Gumperz’, 1982, challenge) to contrast
features of language clearly tied to sentence structure from those that are not. Of
course core and marginal meanings are both central to communication and pragmatic
meaning.
2
My initial use of planes suggested to some readers different levels arranged in a
hierarchy. To avoid this reading, I will speak here of domains instead.
338 D. Schiffrin
3
Although the Holocaust survivor is not a native speaker of English, her level of L2
competence is high. She spoke English as a child and has been living in the United
States and speaking English as her primary language since 1947. However, one might
predict that markers that are more pragmatically nuanced than and (e.g., y’know,
anyway) would be less fully integrated into nonnative discourse.
4
It is sometimes difficult to decide whether and is a conjunction or marker. At one end of
the continuum is and conjoining parts of a clause (e.g., verbs, nouns); at the other end
is and linking turns across speakers. But other uses of and blend syntactic properties of
conjunctions and intonational cues of markers. Although we can sometimes assign and
one or the other identity based on the function of the units in the larger discourse, and
can also combine both conjunctive and marker functions.
5
The possibilities created by divergent or conflicting constraints are familiar to social
scientists that use multivariate statistics to evaluate the relative weight of different
independent variables (e.g., age, social class, gender, ethnicity) on a dependent
variable (e.g., political affiliation). Quantitative sociolinguists have relied upon similar
ways of thinking when analyzing phonological variation. See Pons Bordería (this
volume), who points out that the mulifunctionality of connectives makes it unlikely
that a qualitative study could capture all the different ways that variables might be
associated with one another. Although I will not pursue a statistical analysis here, the
logic of the analysis is similar.
6
Two points are relevant here. First, the race tracks are grouped by states (the upper-level
item): NJ is New Jersey, PA is Pennsylvania (the location of Anne and Kay), DE is
Delaware, NY is New York. Second, I assign a dual status to the list items in lines (i)
and (l) because they are presented in one syntactic unit. Each is counted only once in
Table 1.
7
I am grateful to the United States Holocaust Memorial Museum (Washington, D.C.) for
making this oral history available to me, as well as to the National Alliance of Jewish
Women (Cleveland Branch) for interviewing Mrs. Beer and allowing USHMM to act
as a national repository for the oral history. For further work on oral histories, see
Schiffrin (2001, 2002).
8
I only counted list items that were syntactically compatible with and (an issue that didn’t
arise in Kay’s list). For example, although a daughter is a subcategory of two children
and twenty-seven is a subcategory of daughter, it is syntactically anomalous to conjoin
a relative clause to its head noun phrase (*a daughter and who is twenty-seven).
9
I speak interchangeably of questions and requests (cf. Schiffrin, 1994a: ch. 3).
10
Fetzer and Meierkord (2002: 12), in their new collection on sequentiality, however,
argue that even intentions can be sequentially ordered within discourse, simply
because “intentions manifest themselves in the performance and interpretation of
speech acts or communicative acts.”
Chapter 18
Discourse markers as attentional cues at

discourse transitions
Gisela Redeker
1. Introduction
Discourse particles and discourse markers (henceforth DMs) are intriguing objects of
study, as they promise the researcher ready access to the very fabric of talk-in-progress.
They can occur as lexical equivalents or complements of more elusive gestural or
intonational cues that subtly guide and modulate the participants’ understanding, or they
can saliently signal relations between utterances or larger discourse units. My own
fascination with DMs originated in my interest in developing a model of discourse
coherence, which I will sketch in section 4. Accordingly, my approach is a radically
functionalist one, as I will explain in the present section and in section 2, where I discuss
my definition of coherence-oriented DMs. In section 3, I narrow my focus to one type of
discourse-structuring function, that of marking transitions between segments. I illustrate
how these transitions can be identified without depending on the presence of a DM. I then
report an experiment that tested the hypothesis that DMs at discourse transitions guide the
listener’s attention.
1.1. Approach
Most studies of DMs focus on one or more lexical items (words or lexicalized phrases)
and investigate their functions in various contexts. Against that background and in the
context of developing a theory of language, it is natural to strive for a definition of the
word class of DMs (see Kerstin Fischer’s introduction to this volume). The difficulties in
delineating such a class and its syntactic and semantic properties are discussed in
Schourup (1999).
For my investigations of DMs in the framework of developing a theory of discourse, that

is, a theory of language use as social action, I advocate a radically functional approach in
which the objects of investigation are not lexical items but contextually situated uses of
expressions (Redeker, 1991a: 1168-1169; 2000: 250-251).
340 G. Redeker
The discourse-as-action approach has distinctive implications concerning the research

questions to be asked, the methods to be used (section 1.2), and the data to be considered
(section 1.3). First of all, aiming for a coherent functional category strongly suggests a
clear distinction between DMs that modulate the interpretation of utterances and those
with discourse-structuring functions, in spite of the fact that various items can fulfill either
function. I call the former uses discourse particles and the latter discourse operators (see
section 2).
Functionally divergent uses of lexical items, which in a more semasiologically oriented

approach will raise the question of polysemy (how can the same word have these different
functions?), suggest for a discourse-theoretic approach that the functions under
consideration may be related in some systematic way (why can these different functions
sometimes be realized by the same expressions?). Consider the case of lexical items that
can serve as intrautterance particles (like hedging, focusing, editing, hesitations, etc.) and
also occur as coherence-oriented discourse markers (i.e., as discourse operators). Instead
of asking how the disparate meanings of an item in those uses can come about, the
discourse approach will try to identify functional commonalities between those two
classes of uses. A good candidate here is Clark’s (1996: ch. 8) concept of grounding,
which he conceives of as a collateral activity on a separate “track” of the interaction
realized mostly by nonverbal or paralinguistic means but also signaled with lexical
devices such as editing expressions in repairs and other disfluencies (pp. 262-264, 273-
274) and discourse markers at discourse transitions (pp. 345-346). Similarly within the
more narrowly defined area of coherence-related functions, the flexibility with which
many DMs can fulfill a variety of such functions across instances and within a single
instance (see, e.g., Redeker, 1991a; Schiffrin, 1987a, 2001, this volume) suggests that the
different planes, domains or, as I call them, components should be considered as
interrelated parts of one coherence paradigm (see Redeker, 2000, and section 4 below).
1.2. Methodology
It is widely acknowledged that the data available to the discourse analyst are not the
discourse activities proper but rather “frozen” records of talk or writing. Sequential and
distributional analyses of those material traces can undeniably yield solid evidence for
particular hypotheses about the functions of an expression or a construction (see, e.g.,
Schiffrin, 1987a, this volume), especially when the analyses include speech prosody (e.g.,
as in Ferrara, 1997; Hirschberg and Litman, 1993; and Horne et al., 2001). But the process
itself remains elusive, however meticulously we reconstruct what must have happened.
Multiple data sources and research methods are thus called for to provide converging
evidence for the empirical claims of discourse theory. Functional claims derived from
descriptive and distributional analyses of naturally occurring talk should be tested with the
help of participants (for instance, by using reader judgments as control variables; see
Redeker, 1992, 2000: 253-257), with experimentally elicited spoken and written discourse
(e.g., Redeker, 1986, 1990), and with psycholinguistic experiments to test processing
hypotheses (e.g., Bestgen, 1998; Bestgen and Vonk, 1995, 2000; Tree and Schrock, 1999;
Redeker, 1992; Sanders, Schilperoord, and Spooren, 2001; see also section 3 below).
Discourse transitions 341
1.3. Data
My main source of data are various collections of spoken and written discourse:
experimentally elicited oral narratives in American English (Redeker, 1986, 1990), a
corpus of descriptive, explanatory, argumentative, and entertaining texts from a Dutch
news magazine (Redeker, 1992, 2000), and videotaped interview and discussion programs
from Dutch television (Redeker, 1992).
In my analyses I have focused mainly on monologic stretches of discourse, as they can be

found in and compared across all genres. However, I often use data from interactive
genres, as my goal is a discourse theory that will do justice to the fact that all discourse
genres can ultimately be seen as originating in face-to-face interaction (see Redeker,
2000).
To test some of my processing hypotheses, I have measured naming latencies in a cross-

modal priming experiment based on original and manipulated versions of fragments of
Dutch television talk (see below).
2. Definition of coherence-oriented discourse markers
As I explained in section 1.1 above, I will not attempt to give a definition of a word class
of discourse particles or discourse markers. Instead, I define what I consider a functionally
sufficiently homogeneous class of coherence-oriented marker uses, which I have called
discourse operators (Redeker, 1991a, 2000: 1168; the definition is similar to that of
discourse connectives in Relevance Theory, e.g., Blakemore, 1987: 72-141).
A discourse operator is any expression that is used with the primary function of
bringing to the listener’s attention a particular kind of relation between the
discourse unit it introduces and the immediate discourse context.
This definition leaves the nature of the discourse context unspecified: it need not be a
linguistic context. This allows for inclusion of discourse-initial discourse operators, which
mark a relation with the nonlinguistic context (for instance, a situationally-given event or
state of affairs).1
The scope of the discourse operator is only partly specified in this definition. The minimal
unit under consideration is an intonationally and structurally bounded, usually clausal idea
unit (Chafe, 1980: 14). The stipulation that the relation marked by the operator has to
involve “the discourse unit it introduces” allows for more global coherence functions like,
for instance, the marking of list items, which can involve items of paragraph or chapter
length (see Lenk, 1998). It excludes expressions like focus particles or intrasentential
interjections (e.g., ohh), whose scope does not exhaust an idea unit. This also excludes
anaphoric pronouns and noun phrases. While they may have coherence-relevant functions
(especially at episode boundaries; see Vonk et al., 1992), they do not primarily function as
coherence signals, nor do they suggest a particular kind of relation.
342 G. Redeker
Note that my definition of discourse operators is wider than many current definitions of
DMs (see, e.g., this volume and Schourup’s, 1999, review). Most importantly, discourse
operators need not be optional, need not be syntactically or intonationally independent,
and may add truth-conditional content. Included are all intersentential connectives and
(contrary to, for instance, Georgakopoulou and Goutsos, 1998) also all clause-combining
uses of coordinating and subordinating conjunctions, ranging from simple subordinators
and relative pronouns to semantically rich connectives introducing adverbial clauses
(because, although, and so forth). The class of discourse operators further includes
contrastive uses of indexicals (e.g., now, here, there, today) and other adverbial expressions,
for instance, the next day; [today . . .] tomorrow, that create temporal or spatial chains.
3. Markers of discourse transitions
Within the class of discourse operators, I distinguish ideational and rhetorical connectives
from markers of transitions between discourse segments. I will discuss this model in
section 4 below. In the present section, I will discuss the role of discourse operators in the
management of attention during the processing of sequentially and hierarchically
organized discourse units. Most if not all of the functions I will discuss here have been
described in the literature but usually not as examples of the signaling of discourse
transitions. I will first discuss an example that illustrates the segmented structure of
spontaneous talk. In subsection 3.1, I will present a wider range of segment transitions
with examples from spoken and written genres. I will argue that DMs that signal segment
transitions function as cues to direct the listener’s attention. An empirical test of this claim
will be presented in subsection 3.2.
Consider example (1) from an experimentally elicited description of a short fragment

from a silent movie (Redeker, 1986), presented here in a structured format that shows my
analysis of the segmental structure.
(1) [SEsmj]2
1 So- the fraulein went into the room where the pilots were
2 and- one of the pilots –
3 who had/ who
4 after- the/ they had made accusations,
5 the one who was feeling very down,
6 was uh v/ uh upset by the whole thing,
7 well, the other pilot was trying to comfort him
8 saying
9 there’s more girls in Vienna,
10 and we can go out to the nightclub tonight
11 and- kind of drink your- your sorrows away.
12 And uh and HE was kind of leading him out the door
13 to make sure he wasn’t feeling so bad.
14 Well, as the fraulein encountered the two men, [. . .]
In my analysis of this fragment, lines 2 through 13 present a parenthetically inserted

narrative episode that in turn contains two inserted segments, one (lines 3 through 6)
providing extra information to identify the reference to a character in the narrative, and
one (lines 9 through 11) presenting direct speech of a character.
Some, but not all of the transitions into and out of these segments are signaled by DMs.
So in line 1 marks the beginning of this episode, while well in lines 7 and 14 signals
returns after parenthetical discourse units. Note that the inserted episode in lines 2 through
13 is not initially marked as a parenthetical unit. In fact, the temporal position of those
events remains uncertain until the utterance in line 14, when it finally becomes evident
that the fraulein was not present during the pilots’ conversation (she meets them as they
come out of the room).
Another case in this example where a parenthetical segment is not introduced with a DM
is the reported speech in lines 9 through 11. The speaker here could have used an
“enquoting device” (Schourup, 1985), for instance, oh, well or you know. The function of
this segment as a quotation (implying, as I will argue in 3.1 below, its parenthetical
status), is announced explicitly with the predicate saying in line 8. The speaker returns to
the story line with And uh in line 12.
The parenthetical segment in lines 3 through 6 is especially interesting, as it illustrates

how speakers can introduce background information—here a lengthy disambiguation of
the reference “one of the pilots”—with minimal disruption of the story line (Polanyi,
1978, called this kind of sequence a “true-start”). Note that the use of this construction
need not be motivated by floor-holding concerns: this example comes from a true
monolog (speaker and listener knew that the listener’s microphone was shut off after they
had made initial voice contact, and the example occurred at the end of the story).
The discussion of this example illustrates my approach to the analysis of discourse

structure: I visualize idea units and segmental structure and then determine the most
salient relations between the units. This forces me to consider all possible instances of
transitions as well as ideational and rhetorical relations between the units (see section 4),
regardless of the presence of a DM.
3.1. Paratactic and hypotactic discourse transitions
Units of discourse above the levels of clause, sentence, and turn-at-talk (for a
comprehensive recent discussion see Johnstone, 2002: ch. 3) have been widely
recognized, for instance in conversation analysis (noted already by Sacks, 1992; see also
Houtkoop and Mazeland, 1985, and various contributions in Ochs, Schegloff, and
Thompson, 1996), in sociolinguistics (e.g., Labov, 1972; Schiffrin, 1980, 1987a; Polanyi,
1989), in text linguistics (e.g., Longacre, 1983; Mann and Thompson, 1988), in text
comprehension research (van Dijk and Kintsch, 1983), in semantic and computational
approaches to discourse (e.g., Polanyi and Scha, 1983; Grosz and Sidner, 1986), and in
the cognitive approaches of Chafe (e.g., 1980), who discusses intonational chunking in
spoken discourse in analogy to the typographical paragraph structure in writing, and Clark
(1996), who conceives of interaction as “joint projects” which can contain subprojects and
digressions.
Segment transitions and their signaling have mainly been studied for topic shifts or
episodes in narratives, focusing on discontinuities in time, space, or personal reference
(for a recent example and references to earlier work see Bestgen and Vonk, 2000). In
computational linguistics, research on task-oriented dialogues led Grosz and Sidner
344 G. Redeker
(1986) to the postulation of an “intentional structure” with transitions between subtasks.

In contrast to narrative discourse, where transitions between topics or events are most
often sequential, task-oriented or procedural discourse often involves subordination
following from hierarchical relations between subtasks. Dialogs, moreover, also contain
clarification sequences which interrupt the ongoing discourse and suspend the current
“discourse purpose”. Grosz and Sidner (1986) model the intentional structure as a kind of
push-down stack, where only the current discourse unit is accessible, unless an inserted
unit gets “popped” off the stack to allow resumption of a previously incomplete unit (see
Reichman, 1978, and Polanyi and Scha, 1983, for similar proposals). I will return to these
models in subsection 3.2.
Drawing on this earlier research and on the structures I encountered in various genres of
spoken and written discourse, I have developed a preliminary classification of discourse
segment transitions (see Table 1). I distinguish paratactic transitions between segments
that follow each other at the same level, that is, one segment is completed and followed by
the next one, and hypotactic transitions involving interruption or suspension of an
incomplete unit with parenthetical material. Examples of paratactic transitions are lists of
topics or subtopics, actions, agenda points, and so forth. DMs can signal these transitions
prospectively, marking the beginning of a new segment, or retrospectively, closing off the
current segment (cf. Schiffrin’s 1987a: 322-325 “textual coordinates”). I call these next-
segment markers and end-of-segment markers, respectively (see Table 1 below).
Parenthetical discourse units, which are bracketed by hypotactic transitions (see Table 1),
can vary widely in function and in disruptiveness. The clearest cases of parenthetical
segments are digressions and interruptions, as they are defined in terms of their effect of
suspending or disrupting the topic or purpose of the discourse unit that is in progress.3
Specifications and paraphrases elaborate on (parts of) an utterance and thus will show some
referential overlap with it, while explication, clarification, background, and meta-
communicative or epistemic comment segments can contain any information that might help
the listener’s understanding or acceptance of some previously presented material. Similar to
comments, but referring indexically to an element of “trouble” in the ongoing talk, are
repairs, defined here as emendations of (part of) an utterance that suggest that the speaker
found the initial formulation incorrect or inappropriate (this latter stipulation distinguishes
repairs from paraphrases). Quotations are also considered as parenthetical discourse units, as
they shift the deictic center of the discourse (Bühler’s 1934 “origo”) and refer indexically to
the context in which the quoted speech was uttered or is imagined to be uttered. I include
here fictive quotes like “I was like ‘I don’t believe this!’”, “[. . .] as if to say ‘. . .’”, and
generic or collective ones like “Everybody says ‘. . .’” or “We all went ‘AH!’” (Clark and
Gerrig, 1990; Redeker, 1986: 72-78, 1991b).
Table 1: Turn-internal discourse segment transitions in spontaneous talk
Segment Transitions Typical Discourse Markers

Id Description English Dutch
es end of segment okay?, you know, so hé?, weet je wel, dus
ns next segment okay, so, but, now, well, and nou, dus, maar, en
digression, trouwens, overigens,
di by the way, you know
interruption dus
specification,
sp that is, you know, well namelijk, dus
definition
pp paraphrase I mean, you know, that is ik bedoel, dus
explication,
ex because, you know, I mean want, dus, ik bedoel
clarification
background
bg because, see, well want, namelijk
information
dus, trouwens,
cm comment you know, I think, I guess
overigens
correction,
co oh, or, I mean oh, of, ik bedoel
emendation
qu quote you know, like, well, oh nou, ja, ach, oh
maar (eh), dus, nou,
pop return but (anyway), so, now, well
wel
Example (2) below, from a talk-show interview with a famous Dutch writer of children’s
books, contains a particularly rich variety of transitions (codes preceding the lines indicate
the type of transition as introduced in Table 1). DMs at segment transitions (set in bold)
include paratactic segmentation signals, “push”-markers signaling the beginning of a
parenthetical segment, and “pop”-markers signaling the return from a parenthetical
segment. Note here that this return need not actually lead to a continuation of a previous
segment. Often the previous context is only briefly reinstated with a repetition, sometimes
accompanied by an end-of-segment marker, after which the speaker moves on to the next
unit.
(2) [from interview with Annie M.G. Schmidt, Hoe later op de avond 14 Feb 1989]
Dutch original English gloss

ns Maar! we hadden een huisnaaister. BUT! we had a seamstress.
En die noemden we Mietje. And we were calling her Mietje.
cm Maar we noemden geloof But I think we were calling
346 G. Redeker

ik iedereen Mietje, toen in everyone Mietje, back in
die tijd hoor, those days you know,
cm waarom dat weet ik niet. I don’t know why,
pop Maar goed, but anyway,
es dat was ook een Mietje dus. 4
so that was also a Mietje.
ns En ehm die kwam uit Belgie. And uh- she was from Belgium.
En d’r waren- dat was een And there were- she was a
Belgische vluchtelinge, Belgian refugee,
bg want in in de oorlog, because in in the war,
sp in de eerste wereldoorlog in the First World War
pop kwamen er allemaal all those refugees were coming
vluchtelingen uit Belgie en die g/ from Belgium, and they were g/
kwamen dan in Zeeland en coming to Zealand looking for
zochten daar werk. work.
pop En ZIJ was dus huisnaaister bij ons. And so SHE was our seamstress.
Note that the pop-marker dus (translated here as ‘so’) occurs in noninitial position in the
original utterance (“En ZIJ was dus huisnaaister bij ons”), illustrating that discourse
operators in Dutch need not be utterance-initial or utterance-final. The return function is
further marked by the heavily stressed pronominal reference ZIJ (SHE), indicating that a
previously introduced referent has not been active in the immediate context and must be
retrieved from the discourse record.
In both, examples (1) and (2), storytelling is the speaker’s main purpose. Segmentation
and parenthetical embeddings are not limited to narration, however, as the next examples
will show.
Example (3), from a discussion of ethical questions about journalists’ paying their sources
for news, contains parenthetically introduced background information (bg), a quotation
(qu) introduced by an eh ‘uh’ that might be functioning as a push-marker, and a
commentary (cm) that is parenthetically inserted inside an ongoing clause.
(3) [from the Dutch television news magazine Nieuwslijn 1 March 1989]

Er broeide duidelijk iets, There was clearly something cooking,
bg he, de Surinaamse oud-president you know, the Surinamese ex-
Chin A Sen had een paar president Chin A Sen had said a
maanden few
qu daarvoor gezegd eh months earlier uh
we staan voor kerstmis in we will be in Paramaribo by
Paramaribo, en dan hebben we Christmas, and then we will have
die Bouterse . . . verjaagd. . . . chased away Bouterse.
pop Nou, tegen die achtergrond kregen Now, against that background we got
wij informatie over een groep zwaar information about a group of heavily
bewapende huurlingen, die aan de armed mercenaries who were going to
kant

van Brunswijk eh zou gaan vechten. fight for Brunswijk.
Tegen betaling, For money,
cm want kennelijk- vechten ze niet because apparently- they not only
alleen maar eh- voor geld, maar fight only- for money, they also
praten ze ook alleen voor geld, talk only for money,
pop ehm, hebben ze dat verhaal verteld. uhm did they tell that story.
Note that want ‘because’ here does express a semantic causal relation (a reason). Its more
salient function, however, is its push-marker function, as it most clearly addresses the
question of contextual relevance of the comment (“why this here?”) by signaling its
parenthetical status.
Example (4) contains more instances of lexically marked and unmarked segment
transitions. The beginning of this fragment also illustrates a rather frequent phenomenon
in (media) interviews: the speaker starts with a comment or with a qualification of the
question (analyzed as a parenthetical segment) before beginning his response (marked
here with the pop-marker maar ‘but’).
(4) [from interview with police inspector in Rondom Tien 4 March 1989]

I: Meneer Blaauw, (. . .) er was veel I: Mister Blaauw, (. . .) there was a
kritiek op ‘t eh optreden van de Duitse lot of criticism of the uh actions of
politie, zou u ‘t anders gedaan hebben? the German police,would you have
done it differently?
co B: Da’s heel erg makkelijk eh B: That is very easy uh
gezegd, to say
cm als ik dat al zou zeggen, if I were to say that at all,
pop zeker als je zoals hier op de certainly when you’re in the
voorste rij zit en de beelden vanaf front row like this watching the
’n afstand ziet. pictures from a distance.
ns Ehm kijk, wanneer ooit ’t eh ehm Uhm look, if ever uh uhm the
‘t lugubere schaakspel om disgusting chess game with
onschuldige mensen, innocent people
sp >want daar gaat ‘t om,< >’cause that’s what it’s
about<
pop eh in met al z’n facetten en ook al uh has been shown to the public
z’n lafheid aan de volke vertoond in all its facets and also in all its
is, dan is ‘t wel in deze zaak cowardice, it was in this case.
geweest.
pop Maar d’r is natuurlijk meer aan de But there is of course more going on.
hand. Want waar het om gAAt is het Because what it is abOUt is the most
meest cruciale moment crucial moment
sp in overigens ieder ontvoering of in by the way any kidnapping or
gijzeling, hostage taking,
pop namelijk ‘t antwoord op de vraag that is the answer to the question
qu moet je nu toeslaan ja of nee. should you attack now or not
348 G. Redeker

bg Let wel, wanneer de ontvoerde, Mind you, while the kidnapped,
pp de gegijzelde the hostage,
pop nog in handen van de ontvoerders is still in the hands of the
is.= kidnappers.=
I: =ja. I: =Yes
pop B: Immers, bij ontvoeringen en bij B: After all, in kidnaps and in
gijzelingen is het uitgangspunt dat het hostage situations, the basic rule is
leven van de ontvoerde vóór alles gaat. that the kidnapped’s life has the
highest priority.
ns Maar d’r kan zich natuurlijk een But of course a situation may
situatie ontwikkelen, {kucht} waarbij, develop{cough} where, if you keep
wanneer je dat standpunt following that view,
sp dat uitgangspunt dat absoluut is, that principle that is absolute,
pop blijft volgen, het weleens noodzakelijk it may become necessary that
zou kunnen zijn dat dat dan wel gaat that will then cost the life
ten koste van het leven van degene die of the one who is threatening the
de ontvoerde bedreigt, kidnapped, …
cm waar ik ‘t overigens mooi al niet, which by the way I really don’t-
eh HElemaal niet uh ABsolutely don’t have any
moeilijk mee heb. problems with.
pop De vraag is dan, The question then is,
sp > da’s meer een tactische vraag > that’s more of a tactical
dan nog wat anders, < question than anything else, <
pop ehm op welk moment doe je dat. uhm at what moment do you do that.
Segment transitions abound in spontaneous talk; but they can also be found in edited
writing, as the following two fragments show. Example (5), from a long political
background article in a Dutch weekly news magazine, illustrates a variety of grammatical,
typographic, and lexical means, including DMs, for signaling segmentation in written
discourse. Note that Dutch ook (here translated as postposed too) occurs in sentence-initial
position in the original.
(5) [translated from Vrij Nederland, 27 June 1987]

De man die de kroon op deze The man who perfected this
ontwikkeling zette was Richard development was Richard Nixon.
Nixon.
Hij nam als minister van Buitenlandse He took as Foreign Minister his
Zaken zijn vriend William Rogers, friend William Rogers,
bg een tamelijk zwakke figuur die a rather weak figure who knew
weinig van buitenlandse politiek little of foreign politics,
wist,
pop en benoemde Kissinger tot National and assigned Kissinger as National
Security Adviser, en gaf hem zelfs Security Adviser, and even gave him
opdracht orders

cm (dat heeft Kissinger tenminste (at least that’s what Kissinger has
altijd beweerd) always claimed)
pop die Veiligheidsdienst, to strengthen the Security Service,
bg die zijn zetel in het Witte Huis which has its seat in the White
heeft, House,
pop zoveel mogelijk te versterken. as much as possible.
Kissinger onderhield persoonlijk de Kissinger personally maintained the
betrekkingen met de Russische relations with the Russian
ambassadeur, Dobrynin, en bereidde ambassador, Dobrynin, and also
ook zijnbezoek aan China in het prepared his visit to China in secrecy,
geheim voor,
sp dus zonder de minister van that is, without informing the
Buitenlandse Zaken erin te kennen. Foreign Minister.
pop [. . .] [. . .]
ns Ook Jimmy Carter wilde een eigen Jimmy Carter, too, wanted to play his
rol in de buitenlandse politiek spelen own role in foreign affairs and took
en liet zich leiden door zijn Security guidance from his Security Adviser,
Adviser, Zbigniew Brzezinski, die Zbigniew Brzezinski, who took his
zijn taak serieus opvatte. task seriously.
Example (6), finally, illustrates maar ‘but’ as a paratactic segmentation signal. Note that
there is no semantic or rhetorical contrast here that would combine the segments to form a
higher level unit. Maar (but) simply marks the shift to a new discourse segment (also
signalled by a paragraph boundary).
(6) [from a book on Dutch culture5]

[…] Rebelleer niet, maar vergader. […] […] Don’t protest, but have a meeting.
‘In Nederland is geen sprake van […] “In the Netherlands there is no
participatie, er is sprake van overparticipation, there is overparticipation.”
participatie.’ We vergaderen wat af. [end We keep having meetings. [end of
of paragraph] paragraph]
Maar waar komt die hang naar But where does this propensity for
collectieve lulkoekerij vandaan? collective chitchat come from?
3.2. Transition markers as attentional cues in processing6
Discourse segment boundaries are usually marked prosodically by the intonation contour
and shifts in speed, pitch, and/or volume (Hirschberg and Litman, 1993; Horne et al.,
2001). Upon encountering such a boundary, listeners will initially assume that the
previous focus space is closed and a new one opened (Reichman, 1978); that is, they will
assume a paratactic transition by default. Discourse markers, the hypothesis claims, can
cue the listener as to the kind of transition: paratactic transition, embedding, or return.
Embedding will only suspend the current focus space without closing it off.
350 G. Redeker
This hypothesis was tested with a cross-modal priming experiment. The rationale was as
follows. If a focus space is closed, activation of the referents associated with it will cease.
If it is only suspended, the referents should remain in the periphery of the listener’s
consciousness with a lower but sustained level of activation. This activation can be tested
by presenting the listener with a word that is semantically related to one of the referents.
The reaction will be faster if the word is preactivated. This “priming effect”—activation
of one word yielding faster reactions to a semantically associated one—can be explained
by the well-supported assumption that activation of a lemma in the semantic lexicon
spreads to the lemma’s associates.
To avoid interference with the ongoing aural presentation of the recorded talk, the test
word is presented visually on a computer screen (cross-modal priming). The listener’s
task is to read the test word aloud as soon as it appears. The time between the appearance
of the word and the voice onset (the “naming latency”) is shorter if the semantic field of
the test word has been preactivated.
In the experiment, the test words were related to the previous segment. The hypothesis
predicts shorter naming latencies for test words in recognized parenthetical segments than
for test words appearing after a (putative) paratactic transition. That is, a priming effect
should occur when a hypotactic transition is marked with a DM (as compared to the same
transition without the marker). Furthermore, it is expected that the presence of a pop-
marker will help to reactivate the suspended focus space and should thus also show a
priming effect. No priming is predicted for markers of paratactic transitions.
As the test words occur within several hundred milliseconds of the boundary, there might
still be some residual activation of the concepts from the previous focus space. When the
marker is present, the distance between test and boundary is increased, allowing a further
decrease of the activation. Replacing the marker with a pause in the “no-marker”
condition (as in two of the experiments on oh reported in Tree and Schrock, 1999) would
have disrupted the flow of talk, as the markers were often quite long. Note however that
the direction of any effect from residual activation would go against the hypothesis, as it
would decrease the reaction time in the no-marker condition and thus decrease the
predicted priming effect.
3.2.1. Method
Paratactic and hypotactic transitions between discourse segments were identified in two
hours of Dutch television talk from political discussions, talk shows, and news magazines.
The choice for media talk as opposed to private conversations was a deliberate one. Media
talk is designed for the media audience; the listeners in the experiment were thus intended
overhearers or auditors of the talk (for general discussions of audience roles see Clark and
Carlson, 1982, and Bell, 1991: 90-95, who adds the intermediate category of “auditor”
between the categories of participants and overhearers).
Two parallel versions of each sound track were created by digitally removing and adding
discourse markers. Each version contained marked and unmarked transitions and original
as well as manipulated instances; that is, some DMs had been present in the original and
some were inserted digitally, and idem ditto for transitions without a marker. The
insertion of DMs copied from elsewhere in the same sound track was a very time-
consuming effort, and many carefully tuned instances still had to be excluded because
they were detected by listeners in a control study, who were instructed to find anything
that sounded strange or manipulated. Using both manipulations was important, though, as
it ensured that any differences between presentations with and without the DM would not
simply be due to the fact that some manipulation had occurred in one condition.
Twelve fragments of 5 to 13 minutes each, preceded by a short training fragment, were

presented to 30 listeners in individual sessions. The listeners had two tasks: listening
(checked by occasional comprehension questions) and responding to visually presented
words by reading them out aloud as fast as possible (naming task). The test words
appeared at varying intervals after segment transitions, far enough into the new segment
to allow for some referential processing before the test, yet not so far that the contents
would too clearly give away the function of the segment (parenthetical or paratactic).
Example (7) illustrates this (placement of the test word is indicated by [*]).
(7) [from interview with Annie M.G. Schmidt, repeated from example (2)]

Maar! we hadden een huisnaaister. En die BUT! we had a seamstress. And we were
noemden we Mietje. Maar we noemden calling her Mietje. But I think we were
geloof ik iedereen Mietje, toen in die tijd calling everyone Mietje, back in those
hoor, waarom dat weet ik niet. Maar days you know, I don’t know why, but
goed, dat was [*] ook een Mietje dus. En anyway, so that was [*] also a Mietje.
ehm die kwam uit Belgie. En d’r waren- And uh- she was from Belgium. And there
dat was een Belgische vluchtelinge, want were- she was a Belgian refugee, because
in [*] in de oorlog, in de eerste in [*] in the war, in the First World War
wereldoorlog kwamen er allemaal all those refugees were coming from
vluchtelingen uit Belgie en die g/ kwamen Belgium, and they were g/ coming to
dan in Zeeland en zochten daar werk. En Zealand then and were looking for work
ZIJ was dus huisnaaister bij ons. there. And so SHE was our seamstress.
The test words in these cases were naaister ‘seamstress’ and onderdak ‘shelter’. The pop-
marker maar uh ‘but anyway’ and the push-marker want ‘because’ were present in one
version and absent (cut out) in the other. Example (8) from the same interview contains a
push-marker dus ‘you know’ that did not occur in the original but was spliced in for one of
the experimental versions. The test word, appearing on screen at the time marked here by
[*], was onbekend ‘unknown’.
(8) [from interview with Annie M.G. Schmidt]

Nou, ik weet wel dat ik eh, in vroegere Now, I know that I uh, at an earlier time,
tijd, ik toen ik een jaar of 30 was, [dus] I/ when I was about 30, [you know]
nog vóór [*] de oorlog, of misschien ‘t before [*] the war, or maybe at the
begin van de oorlog, dat ik wel hele beginning of the war, that I did make
mooie- probeerde hele mooie verzen te very beautiful- try to make very beautiful
schrijven, verses,
352 G. Redeker
Example (9) contains a spliced-in end-of-segment marker dus ‘so’, which was followed
on the screen by the test word taboe ‘taboo’, and an added push-marker dus ‘you know’,
followed by the test word aantallen ‘numbers’.
(9) [from Dutch television talk show Op leven en dood 29 July 1989]

En waar men toen over viel was, dat men And what people objected to then was
in die tijd niet gewend was om te praten that they were not used to talking about
over fouten binnen de professie. Als d’r al professional mistakes. If it was talked
over gepraat werd dan mocht ‘t alleen about at all, it was only allowed behind
binnenskamers, met collegae, bij congres- closed doors, with colleagues, at
sen en symposia, maar niet ten aanhore van conferences and symposia, but not to be
een groot algemeen publiek. [Dus] dat was heard by a large general public. [So] that
[*] punt een, en ‘t tweede punt was, dat er was [*] point one, and the second point
over die mortaliteit door mij niet gesproken was that I did not talk about mortality in
werd in termen van percentages, van terms of percentages, like so many
zoveel procent gaat er fout, maar in termen percent goes wrong, but in terms of
van concrete getallen. [Dus] zo veel [*] concrete figures. [You know,] so many
honderd mensen gingen d’r dood. En dat [*] hundred people died. And that was
kwam veel harder aan dan alleen maar ‘t much harder to take than just giving
noemen van percentages. percentages.
Overall, there were 180 test points in the twelve fragments, including 52 sites where one
version contained a push marker, 39 pop-marker sites, 67 paratactic transitions, and 22
fillers where a test word occurred after an ideational or rhetorical discourse operator.
3.2.2. Results and discussion

Figure 1 shows the naming latencies for the test words at paratactic and hypotactic
transitions with and without a marker. As expected, paratactic markers did not facilitate
naming; the naming latencies were even slightly longer when the marker was present (678
ms for the versions without the marker and 688 ms with the marker). At hypotactic
transition points, however, the presence of a push- or pop-marker did produce facilitation,
that is, presumably caused the relevant previous concepts to remain active (in the case of
push-markers) or to be reinstated more easily (for pop-markers). For the 52 push-markers,
the average facilitation effect is 36 ms; the mean naming latencies are 693 ms without and
657 with the marker. This effect is statistically significant in the item analysis (t(51) = 2.8,
p < .01) and in the subject analysis (t(29) = 2.9, p < .01). For the 39 pop-markers, the
facilitation effect is smaller (28 ms) and the difference is statistically significant only in
the item analysis, not in subject analysis.
Figure 1: Naming latencies in cross-modal priming experiment
marker absent marker present

710
700
naming latency (ms)
690
680
670
660
650
640
630
paratactic markers push-markers pop-markers
Note that the presence of DMs at paratactic transitions did not produce a facilitation
effect, although the lexical items used are often the same as those occurring as push- or
pop-markers. The alternative explanation that the attentional effect of those hypotactic
markers might have been a direct consequence of their lexical semantics can therefore be
ruled out in favor of the discourse-functional explanation proposed here.
4. The parallel-components model of discourse coherence
4.1. A sketch of the model
Transitions between discourse segments are one of three components in the model of
discourse coherence I have developed (see Redeker, 1990, 1991a, 2000). I assume that
each new utterance has to be evaluated with respect to its relation to (i) the current
contents of the ‘text world’ (to borrow Werth’s 1999 terminology), (ii) the current
discourse purpose, and (iii) its sequential relation to the current context space or (in
Clark’s 1996 terms) the ongoing “joint project”. The first two of these components are
widely acknowledged in the literature and correspond to the semantic/pragmatic or
representational/procedural distinction (e.g., van Dijk, 1979; Sanders, Spooren, and
Noordman, 1992b; Blakemore, 1987; and various contributions in this volume).7 Not
usually discussed in the literature (but see Kroon, 1998; González, 2004, 2005) is the third
component, which can be seen as a generalized version of Schiffrin’s (1987a) exchange
structure—generalized because the sequential structure component encompasses
monologic and dialogic segmentation.
Table 2 presents the units and relations in each of the three component structures in a
terminology that is inspired by systemic functional linguistics (Bazzanella, this volume).
354 G. Redeker
Table 2: Components of discourse coherence
Ideational Structure
units: propositions
relations: semantic relations that hold in the world described by the discourse
Rhetorical Structure
units: illocutions
relations: reinforcement or support of one unit by the other
Sequential Structure
units: discourse segments
relations: transitions to next unit or to/from parenthetically embedded unit
The model asserts that each utterance will update the discourse in all three components,
though I have found in my analyses that there is usually one relation that is most salient in
the overall context.8 Not surprisingly, the relations in the three components tend to be
parallel in nature, for instance, a supporting argument (evidence relation) in the rhetorical
structure requires an underlying cause, result, or reason relation to hold (to be implied by
the speaker) in the ideational structure, and a concession or antithesis presupposes some
kind of implicit semantic contrast. Table 3 sketches some of the correspondences between
relations in the three components (for a more detailed list and discussion of relations and
their correspondences see Redeker, 2000: 248-249).
Table 3: Selected coherence relations in the three parallel components
Semantic Relations Rhetorical Relations Sequential Relations

(addition of information)* (next speech act)* (segment continues)*
temporal relations contingent act (response) next segment
elaboration, enablement motivation, encouragement comment, digression
alternative, contrast concession, antithesis paraphrase; return
cause, reason justification, support background segment
result, effect conclusion end of segment
*Simple additive relations are taken to be the default whenever no more specific relation holds in the
component in question.
The model has been tested empirically with various quantitative distributional analyses of
discourse markers and with the experiment reported above. The distributional evidence
shows that using markers with the most salient function in one of the components tends to
reduce the need for more explicit marking in the other components. If narrators use many
pragmatic markers (that is, markers of rhetorical relations and of discourse transitions),
they will tend to use fewer markers of semantic (temporal, causal, and so forth) relations;
and if the semantic and pragmatic complexity of a range of discourse types is controlled
statistically, the partial correlation between the density of ideational and pragmatic
marking is significantly negative, indicating again a trade-off between the components
(Redeker, 1990, 1992, 2000). I have taken this as support for the treatment of those three
components as constituting one identifiable paradigm of discourse coherence.
4.2. Discourse structure and dialogue structure: bridging the gap
My discussion here has focused mainly on turn-internal or monologic transitions, but the
parallel-components model is intended to also apply to relations across turn boundaries.9
Interactional (cross-speaker) realizations of end-of-segment markers are end-of-turn
markers like okay?, right?, no?, whatever, tag questions, trailing off, and so forth. Turn-
initial discourse operators like so can signal the beginning of a new segment (paratactic
transition), introduce a parenthetical segment (e.g., a clarification or repair sequence), or can
pop back to a previous topic (e.g., with but).
Let me illustrate the parallelism of turn-internal (monologic) and turn-initial (interactional)

uses for the paratactic and hypotactic uses of but (Dutch maar). The monologic use as next-
segment marker was illustrated in examples (4) and (6) above. Example (10) shows a
paratactic transition where the new segment is produced by a second speaker. Note that B’s
turn is not a speaker return, nor a rejection of S’s contribution. B has been agreeing with S’s
point before and is now introducing a new aspect into the discussion.
(10) [from Dutch television discussion program Het Capitool, 26 February 1989]

S: [. . .] blijkt dus dat er ongeveer 50% S: [. . .] it becomes evident that there are
weigeraars zijn en het is zeer wel about 50% refusals and it is very well
mogelijk dat bij die 50% nou juist de possible that those 50% do include the
groep zit die seropositief is. group that is HIV positive.
B: Maar niemand stelt zich de vraag B: But nobody is asking the question
waarom er zoveel mensen zich why so many people are withdrawing
terugtrekken [. . .] [. . .]
The parallelism between monologic and interactional occurrences is also evident in

Schiffrin’s (1987a: 152-177) discussion of but as a marker of “speaker return” within as
well as across turns, which is quite similar to the monologic pop-marker use illustrated in
example (2) above. Example (11) contains an interesting partially interactive occurrence:
356 G. Redeker
(11) [from talk show Op leven en dood 29 July 1989; G has just explained that he has to
expose healthy persons to anti-depressant drugs in his clinical trial]

P: Maar d’r zijn bijwerking en natuurlijk . P: But there are side effects of course.
Een gezonde student van 20, die eh A healthy student, twenty years old, he
uh
di nou ja, hij eet verkeerd, well his diet is bad
omdat ie student is, because he’s a student,
maar verder is ie gezond. but apart from that he’s healthy,
G: Ja nou, ‘t ligt niet zozeer aan ‘t G: Yes, well, it’s not so much the
eten, meestal meer aan ‘t drin ken, food, more often the drinking, but
maar hij eet eh hij eet vaak to ch wel he’ll he’ll often eat reasonably
redelijk tegenwoordig, dankzij d e well these days, thanks to the
mensa’s. canteens.
pop Maar ‘t eh, nee ehm de mensen hebben But it uh, no uhm the people don’t
geen kwaal en worden toch met i ets have an illness and still they get treated
behandeld tegen een kwaal en on der- with something against an illness and
vinden dan de bijwerkingen die o ok een will experience side effects that a
patient zou kunnen gaan ondervinden. patient might also experience.
Mazeland and Huiskes (2001) present detailed sequential analyses of several truly
interactional (turn-initial) occurrences of Dutch maar ‘but’ as a resumption marker. They
identify two “prototypical environments in which resumptions occur” (p. 141): after repair
sequences and after competing topics. The functions they discuss would be analyzed as
pop-marker functions in my model.
Turn-initial and turn-internal uses of resumptive anyway (which corresponds to Dutch maar
or maar goed) in narration are discussed by Ferrara (1997). She distinguishes resumptions
after “teller-triggered” and after “listener-triggered” digressions and finds the former to
account for 74% of the cases. In particular, teller-triggered digressions in her corpus
occurred mostly turn-internally, but also across turns (e.g., after word completion by the
listener), and listener-triggered cases often involved (listener-initiated) joint laughter, again
suggesting close parallels between monologic and interactional uses of the resumptive
marker.
5. Concluding remarks
The model of discourse coherence I have discussed here provides a conceptual framework
for developing and testing theories about the role of discourse operators in spoken and
written, monologic and interactive discourse. I have exemplified this for the subcategory
of segment transition markers, which are shown to affect listeners’ attentional processing.
The methodology employed in this study combines systematic corpus analysis with
psycholinguistic experimentation.
Discourse markers in this approach are considered mainly with respect to the effects on
the processing and representation of discourse arising from the presence of a marker in a
particular type of context. With Schiffrin (1987a), I assume that discourse markers select
coherence options. In my model, these options are organized in three parallel components.
Each discourse unit is considered to contribute to the ideational, the rhetorical, and the
sequential structure, one of which is usually the most salient on any particular occasion.
Distributional and sequential analyses of particular markers in my own research and in

many studies by others have been essential in refining the model (Redeker, 1990, 1992,
and 2000). My focus, however, has been on the fully contextualized interpretation of
marker tokens in their sequential context with the purpose of identifying generalized types
of functions in the three components and not on the modeling of individual marker
meanings as arising from the interaction of the marker’s semantics with the discourse
context. This is why I chose not to address the question of polysemy here and refer the
reader to the other contributions in this volume.
Notes
1
I will not discuss discourse-initial uses in this paper but point this out here to prevent or
correct a misunderstanding evident, for instance, in Schourup (1999: 237, 239), who
apparently assumes I meant discourse segment where I wrote discourse context
(Redeker, 1991: 1168). In my usage of the term discourse, language is a necessary
ingredient, but context is an equally inalienable part.
2
Examples are presented in a structured format, roughly one idea unit per line (sometimes
line breaks become necessary within a long unit) and with minimal intonational
annotations. Syllables with exceptional stress are capitalized. Notable differences in
speed of talk are marked by enclosing fast speech in angle brackets (> fast speech <).
Other punctuation symbolizes: stops (/), lengthening of final syllables (word–), clause-
final (comma) and sentence-final (period) intonation, and pauses (. . .). Indentation
shows my interpretation of the discourse segment structure.
3
My description of these categories is rather informal, as their differentiation is not
crucial to my main purpose of identifying paratactic and hypotactic discourse
transitions. The categorization is intended as a partition (i.e., as exhaustive and non-
overlapping). Note that the category of digression, for instance, is therefore taken more
narrowly than elsewhere in the literature, e.g., in Ferrara (1997), who includes among
digressions “orientational detail” (here, background information) and epistemic
comments (here classified as comment segments).
4
Note that the end-of-segment marker dus (English thus), here translated by so for
idiomaticity, appears in utterance-final position in the Dutch original (ending the
segment after the utterance).
5
Willem Pijffers (1992). Nederland in 20 seconden. Dubbele bodems in de Hollandse
moraal [The Netherlands in 15 seconds. Double standards in the Dutch moral].
Aramith, Bloemendaal, The Netherlands, p. 15.
6
The research reported in this section was supported by a grant from the Royal
Netherlands Academy of Sciences and was carried out in the speech laboratory of the
Max Planck Institut für Psycholinguistik in Nijmegen.
7
Compare also Schiffrin’s (1987a) ideational structure and action structure and Sweetser
(1990), whose epistemic and speech act uses of conjunctions would both fall in the
second component of my model. In the dynamic semantic account of discourse
358 G. Redeker
relations in SDRT, Asher and Lascarides (2003) distinguish three kinds of relations
that would fall in the pragmatic category: cognitive-level relations, divergent relations,
and meta-talk.
8
“Overall” is no gratuitous specification here. In testing the reliability of my
classifications, I found that even rather long fragments allowed multiple interpretations
of a particular relation much more readily than a complete text. Only analyses that are
informed by the full context yield satisfactory intercoder reliabilities (see Redeker,
1992, 2000).
9
Kroon (1998) makes a similar point when she discusses monological (within-turn) and
dialogical (across-turns) uses of Latin connective particles, arguing that the within-turn
occurrences of what she identified as interactional functions are licensed by diaphony.
Chapter 19
A dynamic-interactional approach to discourse

markers
Barbara Frank-Job
1. Introduction
Research on discourse markers (DMs) in various languages faces the following recurrent
analytical problems:
• the semantic polyvalence of syntagmas and word forms used in a discourse-organizing

function,
• the broad functional range that DMs cover, and consequently,
• the difficulty of defining discourse markers as members of a semantically, formally,
and pragmatically coherent and homogeneous word class.
The present paper suggests that in order to resolve these problems, it is necessary to take
into account the dynamism inherent in the development and interactional functioning of
DMs.
1.2. Approach
This paper focuses on both the interactive processes that lead to the use of certain
linguistic items as DMs (thereby assigning a polysemic status to them) and the
mechanisms of discourse processing that underlie the interactional functioning of DMs
(and that contribute to a better understanding of their multifunctionality).
DMs evolve out of processes of “pragmaticalization”. At the beginning of such a process,

we find lexical items (nouns, adjectives, adverbs, and verbal syntagms) with propositional
meanings which are used in a metacommunicative way. Through processes of
habitualization and automatization, metacommunicative use creates a variant of the
original item, whose main function is interactional (see section 3 below).
360 B. Frank-Job
Meanwhile, in their interactional functioning, DMs fulfil important tasks for the discourse
processing activities of the participants. It is because discourse processing works
simultaneously at different levels that some DMs are multifunctional (see section 4
below).
My approach to DMs can therefore be considered as polysemic (see Hansen, this volume,
section 1.4.1.) in two different ways: first, we are dealing with a polysemy resulting from
a diachronic process in which additional metacommunicative meanings appear. Second,
we are dealing with a polysemy that consists of several pragmatic meanings working
simultaneously on different levels of discourse processing.
Given its twofold orientation, this paper has recourse to different research traditions: the
first part, which looks at the development of DMs, deals with research on
grammaticalization (Hopper and Traugott, 1993; Lehmann, 1995; Traugott, 1995b;
Hagège, 2001). The second part, which looks at DM-functions in the discourse
processing, is based on conversation analysis (Schegloff, 1972; Sacks, Schegloff, and
Jefferson, 1974; Bergmann, 1981; Gülich, 1991, 1999; Gülich and Mondada, 2001;
Mondada, 2001) and on text/discourse processing research (van Dijk, 1980; van Dijk and
Kintsch, 1978, 1983).
1.3. Pragmatic meaning and the study of DMs
In real-life conversation, we can distinguish three coexisting types of meaning: a lexical

or propositional meaning relating to nonlinguistic entities, a grammatical meaning relating
to the syntagmatic functions of linguistic entities,1 and a pragmatic meaning revealing the
relation between persons participating in a conversation as well as their intended and
actual behavior.
Whereas lexical and grammatical meaning can be described without respect to actual
communication, pragmatic meaning is essentially tied to the context in which utterances
are produced:
Semantics is primarily concerned with meanings that are relatively stable out of
context, typically arbitrary, and analyzable in terms of the logical conditions under
which they would be true. Pragmatics, by contrast, is primarily concerned with the
beliefs and inferences about the nature of the assumptions made by participants
and the purposes for which utterances are used in the context of communicative
language use. It concerns both speakers’ indirect meaning, beyond what is said,
and also hearers’ interpretations, which tend to enrich what is said in order to
interpret it as relevant to the context of discourse. (Hopper and Traugott, 1993:
69).
The study of pragmatic meaning belongs therefore to the study of discourse, whereas the
study of lexical and grammatical meaning belongs to the study of language as “historical
techniques” (Coseriu, 1981; Coseriu, 1982: 7).2
This is fundamental for the linguistic status of DMs as object of research. DMs only
function in real communicative contexts. Within these contexts, the actual function and
meaning of a given DM are not ambiguous; hearers are usually able to choose its correct
A dynamic-interactional approach 361
meaning. It is only when the necessary context information is lacking that a hearer can
misunderstand the meaning of a DM. That is why, in the analysis of DMs, the use of
constructed data should be excluded and illustrating examples should be taken from real
verbal interactions.
1.4. The data
Most of the data presented in my paper are taken from the Italian spoken language corpus
LIP and from the German spoken language corpus DGD.3,4 Additionally, some examples
of real conversation in Italian, French, Spanish, German, and English are taken from
conversational analysis research (Schiffrin, 1980, 1987a; Bazzanella, 1990; Chodorowska,
1997).
2. Definition
DMs constitute a formally heterogeneous, open class of linguistic items (particles,

adverbs, substantives, verbal syntagms, etc.) that have undergone a linguistic change by
ways of regular metacommunicative use and the regular fulfilment of discourse-
interactional functions.5 This process can be described as pragmaticalization.
Pragmaticalization is the process by which a syntagma or word form, in a given context,

changes its propositional meaning in favor of an essentially metacommunicative,
discourse interactional meaning. In this regard, pragmaticalization functions like
grammaticalization as described by Hagège (2001):
En d'autres termes, si un sens est perdu, un autre est acquis. Il n’est pas vrai que,
comme on aime à le répéter, la grammaticalisation aboutisse à des unités figées ou
sans contenu. Il serait plus vrai de dire qu’elle aboutit à des unités spécialisées.
(Hagège, 2001: 1612)6
The same is true for the words and phrases that undergo the process of pragmaticalization.
The main function of grammaticalization and pragmaticalization processes is to facilitate

communication. Recurrent communicative problems both on the level of message
structuring (grammar) and on the level of discourse structuring (discourse pragmatics)
tend to be resolved by speech communities in a durable way, i.e., in routinized techniques
which can be used in a merely automatic manner.7
In the case of pragmaticalization, the routinization and functional specialization affects

the discourse organizing function of words: instead of contributing to the propositional
content of the interaction, the pragmaticalized linguistic item operates on the level of
discourse organization.
In the synchrony of a historical language, this process leads to polysemy between the
pragmaticalized word form and its propositional origin.8
362 B. Frank-Job
3. The development of DMs in interaction
The following examples are all taken from the same conversation. They show three
instances of the German adverb gut (English ‘well’) demonstrating a successive loss of
propositional meaning in favour of a metacommunicative meaning.
In example (1) the adverb is used with its propositional meaning attributing the value
‘well’ to a verbal phrase:
(1) S2: im vorigen Jahrhundert waren vielleicht Ehen noch mehr gefährdet, da die
Menschen doch sich zusammenfanden, weil vielleicht die Höfe zusammenpassten
oder die Fabriken gut sich gegenseitig gebrauchen konnten
S2: in the last century marriages were perhaps even more endagered, since people
got together because their farms went well together or because their factories
could well profit from each other
(DGD FR 030, 14)
Example (2) shows the same adverb still used with a propositional meaning, but in this
instance the adverb refers to the preceding utterances of S1 (das is . . . ein Extrem in the
first instance and vielleicht ist es doch ein bißchen übertrieben in the second instance of
gut):
(2) S1: entschuldige aber das is sagen wir ein Extrem. aber du hast richtig vorher
gesagt nicht wahr also an diesen Extremen an diesen Extremen kristallisieren
sich
S2: ja gut schön ja
S1: die also die die Standpunkte irgendwie ganz deutlich vielleicht is es doch ein
bißchen übertrieben. aber trotzdem
S2: es ist ein Aspekt
S1: ja
S2: gut. das geb ich zu.
(DGD, FR 030, 23)
S1: sorry, but this is, let’s say, an extreme. But previously you were right to say,
well, its with these extremes, with these extremes that
S2: yes fine, okay fine
S1: the points of view come out very clearly in a way perhaps it is a little bit
exaggerated after all, but still
S2: it is one aspect
S1: yes
S2: okay, I admit you are right, but again I would now have to . . .
Example (3) shows the metacommunicative use of the same adverb, which has lost all of
its propositional meaning (there is no aspect in the preceding utterances of S3 that is
positively evaluated by the German gut of S2). In a metacommunicative way it indicates
the explicit ending of the preceding turn of S1 and thus prepares the beginning of the turn
by S2:
(3) S3: ich brauch da nich noch großartig eine eine eine Bestätigung von vom Pastor
und so
S2: nein paß mal auf gut
S3: daß sie daß sie mir gehört ja das das dazu brauch....
S2: ja ja also die Frau gehört dir das is überhaupt mal die Frage
(DGD-Fr 030, 11)
S3: I don’t need a a a special confirmation from, from the pastor and so
S2: no, listen well
S3: that she, that she belongs to me, yes that that for that I need
S2: yes, yes, so the woman belongs to you but that is in fact the question....
The frequent metacommunicative use of the adverb gut in actual colloquial German as
seen in example (3) has led for the German gut to develop into a discourse marker.
3.1. The pragmaticalization process9
The starting point in the development of DMs are linguistic units (words and expressions)
which refer to the physical referential environment of a conversation (the Zeigfeld of
Bühler, 1934). Particularly central within this framework are signals for reception and
action (Rezeptionssignale and Aktionssignale in Bühler's terminology10)—expressions that
symbolize the physically perceptible entities that are part of the direct speech context: the
persons involved in a conversation and their physical behaviour (English listen, look;
Italian senti, guarda; French écoute, regarde; German hör, schau; etc.), evaluative
reference to the directly preceding part(s) of the interaction (English well, okay; French
bon/bien; German gut), as well as local (English here; Italian ecco, qui; French voilà,
voici; German hier) and temporal (English now, Italian adesso, French maintenant,
German nun) features of the situation the conversation is embedded in.
In order to fulfil their communicative needs, speakers use these signal words in a
metacommunicative way, no longer referring to the features of the situation but to the
linguistic act itself.
Thus, in the following example, the temporal deictic adesso (‘now’), which originally
refers to the actual moment of conversation, now refers to the utterance that follows:
(4) A: […] comunque se vuole far la terapia magari

B: mh
A: eh chiama_ insomma adesso ci pensa un attimo e vede un pochino
B: si'
A: […] well, if you want to do the therapy perhaps

B: mh
A: eh call_ okay, now, think about it for a moment and look it over a little bit
B: yes (LIP RB 13)
As DM, adesso prepares the next utterance (ci pensa un attimo e vede un pochino) and
guides the attention of the participant in that it functions as initial marker for a new
interactional unit (a new thematical sequence in this instance).
364 B. Frank-Job
In a way which is similar to the processes of lexicalization and grammaticalization,

pragmaticalization functions by means of routinization and frequency. This leads to some
formally detectable features of DMs.
3.2. Clues to pragmaticalization processes
The formal (phonetic, morphologic, syntactic, and textual) features which accompany the
pragmaticalization of a lexical item or an expression into a DM and which point to the
fact that pragmaticalization processes have occurred are the following:
• frequency
• phonetic reduction
• syntactic isolation
• co-occurence in contiguity
• deletion test
3.2.1. Frequency
In real-life conversations DMs appear strikingly often. Thus, in English everyday
conversation the particle well is used every 150 words on average (Svartvik, 1980: 169).
In the LIP-Corpus there is evidence of a regular, frequent use of DMs throughout all types
of conversation. During a radio call-in quiz (LIP FB14) nearly one word in ten was a
DM.11
Another interesting aspect in this context is the frequent co-occurrence of several DMs.
The LIP-Corpus gives many examples of this phenomenon. In the following case, each
pair of DMs fulfils one single communicative function:
(5) B: okay_ va be'

A: allora diciamo che_ cominciamo da queste pagine qui perche'_ sono quelle
che ho preparato
B: okay_ alright
A: then, let's say that_ we start with these pages there because_ these are the
ones that I have prepared
(LIP MA27)
In other examples, co-occurring DMs fulfil different communicative functions which

complement one another. In the following example, both types of combination appear in a
sequence of five DMs. This indicates an important transitional point within the
conversation:
(6) P: a me serve un altro giorno io studio filosofia

N: ah ho capito va be' allora senti (incomprensibile) comunque cerchi #
(incomprensibile) nel [catalogo] ...
P: serve me another day, I am studying philosophy
N: ah, I see, okay, so listen (incomprehensible) anyway, look #
(incomprehensible) it up in [the catalogue] ...
(LIP MA21)
Whereas the first two DMs confirm and close the preceding turn, the next three initiate a
new turn and simultaneously signal the end of a thematic sequence in conversation.
As Gülich (1999) argues, there is a clear correlation between the amount of DMs
combined and the structural importance of their place in a given discourse:
Une étude systématique de ces combinations peut montrer que plus il y a de

marqueurs, plus le changement thématique est important. Pour signaler la
discontinuité le locuteur fait plus d’efforts que pour signaler la continuité, et ces
efforts laissent des traces plus explicites que quand il s’agit de continuité.
(Gülich, 1999: 34)
3.2.2. Phonetic reduction

The more often DMs are used in actual speech, the more reduced tends to be their
phonetic material:
There is a link between frequency of use and phonetic bulk such that more
frequently used material, whether grammatical or lexical, tends to be shorter
(phonetically reduced) relative to less often used material. (Bybee et al., 1994: 20)
Consequently, with many DMs we find phonetically reduced variants, such as Italian va
be' instead of va bene (see (5)), Italian di' instead of dimmi (see (11)) or French ben
instead of bien.12
Further studies may investigate whether or not the use of the reduced variant differs from
that of the complete expression, as one could assume upon examining (7). In this
example, the same participant uses both the long and the reduced variant of va bene in one
and the same context:
(7) F: la voce l l’unica cosa che non va in offerta e' la voce m

E: okay va be' quindi tutto tranne_
F: ... tutto abbiamo preso tutto siamo stati molto buoni
E: va bene va bene <?> senti ...
F: lot l. The only one that will not be put on sale is lot l
E: okay, okay, so, all but_
F: ... all, we have gone through all, we have been very good
E: okay, okay <?> listen ...
(LIP NB5)
The first instance (reduced form) is used by E in order to conclude the preceding turn and
to take over. Since this goal is not achieved, E repeats the term, this time in its full form.
This may lead to the assumption that the “complete” form functions simply as a
reinforced variant of the reduced form.
During the process of pragmaticalization, expressions also tend to amalgamate into fixed
units. This development is closely linked to the reduction of phonetic material as we can
366 B. Frank-Job
observe in the Italian expression va be' forming a unit in which no other word can be
inserted.
The fusing of elements within the DM can be seen in the long term development of
languages. The French and Italian DMs that developed from the Latin temporal
expression ad illa(m) hora(m) (‘at that time’) have merged into one single word—French
(a)lors, Italian allora—as they became DMs.13
3.2.3. Syntactic isolation

DMs are syntactically isolated. In (8), for instance, Italian guarda is accompanied by an
accusative complement (guarda questo) while in (9) it is not:
(8) B: guarda questo e' il eh quello che m'ha ril<asciato> rilasciato l'architetto
B: Look this is the one which the architect issued to me

(LIP FA10)
(9) A: no poi soprattutto io_ dico_ guarda quando sono stato in Croazia_ per
esempio # io son andato a far una storia par<ziale> specifica cioe' bambini_
ammazzati eccetera ...
A: No, then, above all, I_ say_ look when I was in Croatia_ for example # I was
making a special story like killed_ children etc. . . .
(LIP MB8):
In (10), Spanish entiendes governs a subordinate (conditional) clause (me entiendes

cuando . . .), while in (11) it is syntactically isolated:
(10) ¿Me entiendes cuando te hablo muy rápidamente?

Do you understand me when I speak very quickly?
(Chodorowska, 1997: 356)
(11) Si yo, a mí eso me da igual ¿me entiendes? ... Era más o menos para saberlo tú.
Well me, it doesn't matter to me, you know, ... I just wanted you to know it.
(Chodorowska, 1997: 356)
3.2.4. Co-occurence in contiguity

As DMs lose their original lexical meaning, it becomes possible for other items in the
direct linguistic context to express that original lexical meaning, as can be seen in the
following example:
(12) Allora, ero assai giovane allora.

Now, at that time he was quite young.
(Bustorf, 1974: 24)
3.2.5. Deletion test

Finally, there is a deletion test that works with DMs. As Bazzanella (1990) points out, the
content of an utterance is not altered if the DM is removed:
(13) La via dove abito, sai, è cosi rumorosa.

The street where I live, you know, is so noisy.
(Bazzanella, 1990: 632).
4. Multifunctionality of DMs in discourse processing
4.1. Levels of discourse processing
Consider the telephone conversation in Table 1.
Table 1. Telephone conversation
1 A: pronto? Hello?
2 B: <?> c’e’ Paolo? <?> Is Paolo there?
3 A: eh no Paolo e' uscito ha detto che Ah, no, Paolo’s gone. He said that he
4 B: tornava verso le sei would be back around six
5 A: va bene grazie Okay, thank you
6 B: cosa devo dire_? Do you want to leave a message?
7 A: sono Tiziana magari I’m Tiziana perhaps I'll call again later, at
richiamo_verso_ le sei e mezzo about half past six
8 B:
ah va bene Ah, okay
9 A:
grazie Thank you
niente arrivederci You are welcome, good bye (LIP FB2)
The core part of this short conversation consists of two lines only (ll. 5-7). It is embedded
in opening and closing routines. Opening, core, and closing parts of the conversation
together form the global structure of a conversation, each of them being organized by
pairs of speech acts or turns. In longer conversations, moreover, the core can be organized
by different conversational subjects.
As has been shown in van Dijk (1980) and van Dijk and Kintsch (1978 and 1983), these
organizational levels of conversation correspond to levels of discourse processing.
Persons involved in a dialog perceive and produce the conversational interaction on these
three levels of conversation structuring, illustrated in Fig. 1.
As research in conversational analysis has shown (Sacks, Schegloff, and Jefferson, 1974;
Gülich and Mondada, 2001; etc.), the first and basic structural instance of real-life
conversation is the organization of turn taking. In order to guarantee correct functioning
of a conversation, the participants have to deal with the two basic interactive problems of
turn taking organization: they have to identify the moments in conversation when a
change in turn is possible, and they have to manage the changes of the turn.
368 B. Frank-Job
14
Figure 1. Levels of conversation structuring
Turn a1
Turn a2 ... initial sequence(s) Opening of conversation
Turn b1
Turn b2 ...
core sequence Center of conversation
Turn c1 final sequence(s) Closing of conversation
Turn c2 ...
Turn-taking Macrostructure Superstructure

System
In order to minimize organizational problems, participants constantly and systematically

look for possible moments of transition of the turn (Mondada, 2001; Gülich and
Mondada, 2001). On this basic level of conversational interaction, we find the first and
fundamental functions of new appearing DMs: they indicate the moments where a change
in turn is possible (see examples in part 2).
Since participants methodically use DMs in a first instance to manage turn-taking

problems, the process of pragmaticalization of a given item starts out at this basic level. It
is then up to the participants in the dynamics of the verbal interaction whether the DMs
are used (and interpreted) to refer to other levels of discourse processing.
Analytically, the distinction of different levels of conversation enables us to classify

different functions of DMs according to the conversational level on which they operate.
Thus, we can easily categorize the different functions of the Italian particle va bene in
(14).
V a b e n e (‘okay’) is an example of positive backchannel behavior. Here, it

metacommunicatively refers to the preceding speech act of the dialog partner. In making
explicit that the message has been correctly understood and that there are no objections to
its contents, va bene also serves as a closing signal to the turn. As such, it can very well
serve as a signal to close the conversation as a whole, which indeed it does in line 7. The
example of va bene shows that at least one aspect of the multifunctionality of DMs is
systematic. There is a principle behind the use of va bene which the dialog partners do
indeed follow. The necessary condition for this broad functional use of va bene is given
by its primary function on the basic level of conversation. This initial function remains an
inherent feature in all the secondary functions that the particle may take on in a discourse
setting.15
4.2. Types of DMs and their function at different levels of conversation
The role of the turn-taking system as the basic level of conversation becomes particularly
clear when considering the fact that in many languages metacommunicative expressions
are frequently used as DMs. As can be seen in (14), such expressions verbalize the basic
communicative problems to be handled in turn taking:
(14) A: si'_
B: pronto_ sono <?>
A: ah dimmi ciao
B: di' un po'
A: dimmi
B: c'e' questa storia che a Kuwait city stanno spegnendo l'ultimo pozzo . . .
A: Yes_
B: Hallo_ its me <?>
A: ah tell me, hi
B: tell me a little . . .
A: tell me
B: there is that story that in Kuwait City they are about to extinguish the last
bore-hole . . .
(LIP MB6)
(14) is the beginning of a phone conversation. The DM di/dimmi functions on all three
levels of discourse structuring: on the level of turn taking, it is used by both participants in
order for them to yield the turn to the partner. In the fifth line, dimmi is placed at the end
of the initial sequence and thus leads directly to the following first thematic sequence. In
this instance dimmi also initiates the center of the conversation.
As they appear very frequently, these metacommunicative expressions are used in an

automaticized way:
(15) A: quando le mandi le lettere a Mario?

B: quando tu non ci rompi le palle
A: senti eh
B: ah dimmi
A: eh # Claudio non c'e' Gianni Oletta c'e'?
A: when will you send them, the letters, to Mario

B: when you’ve finished getting on my nerves
A: Listen
B: Yes?
A: Ah # Claudio, is Gianni Oletta, isn’t he there?
(LIP NB2)
370 B. Frank-Job
Table 2 shows examples for this type of DMs in several languages.
Table 2. Basic communicative needs on the turn-taking level and their

metacommunicative equivalents16
verbalized speech act verbalized act of reception

X wants to voglio dire... (LIP MA4) senta una cosa (LIP FA 12)
take (keep) let me tell you – I’ll tell you yeah but listen to me ... (Schegloff,
the turn something ... (Schiffrin, 1980: 207) 1972: 353)
je voulais seulement dire que ... je écoutez mon cher ... (Meyer-
voudrais simplement dire ... Hermann, 1978: 134)
(Meyer-Hermann, 1978: 131/139)
ich wollte sagen, ... darf ich hier
mal einhaken (Schwitalla, 1976:
83)
X wants to Patrizia dimmi una cosa (LIP FB fammi sentire (LIP FB5)
quit (stay 5) fatemi sentire (LIP NA3)
out of) the say, can you lend me a dime?
turn fammi sapere (LIP MB3)
(Schiffrin, 1987a: 328)
alors, dites euh c’est pas loin ... ?
(Koch and Oesterreicher, 1990: 57)
Another type of DM makes a metacommunicative comment on the preceeding turn. As

we have seen in (13), expressions like Italian va bene function on the basic level of
discourse structuring as signals for the turn holder that the reception process has been
successful and, therefore, that he can stay on the turn. In making explicit that the turn is
closed, they can then be used as cues for those points in the conversation where a change
of turn becomes possible. All types of back-channel expressions can be used in this way,
i.e., to mark the end of a turn. In fact, they very often serve the hearer to prepare his
taking over of the turn:
(16) C: il discorso di fondo e' diverso

A: si' va bene ma voglio dire
...
C: no non e'riprovevole e' che fa schifo
A: va be' per<r> per me voleva dire ...
C: the basic discourse is different

A: yes, okay, but I want to say
...
C: No, it’s not something to disapprove of, it’s simply disgusting

A: okay, for, for me I wanted to say ...
(LIP RA4)
(17) N: ah ho capito va be' allora senti ...

N: Ah, I see, okay, now listen ...
(LIP MA21)
A third type of DMs comments on the structure of the conversation itself. These
expressions present the discourse as temporal or local movement and mark salient points
in it (English now, Italian allora, French alors, German nun; English then, Italian poi,
French puis; Italian ecco, quindi, French voilà); they present the discourse as the
development of an argumentative chain (English but, Italian ma, French mais, German
aber; English however, Italian invece, French par contre, German dagegen); or they sum
up the communication (Italian insomma, French enfin, German also).
On the basic level of conversation these DMs are used to guide the attention of the
participants towards the following speech act and, in doing so, eventually to the following
turn, thus preparing its beginning. On the macrostructural level, they can be used to
introduce a new thematic sequence:
(18) B: questo anzi e' uno simpatico

A: vabbe' eh
B: e tu come stai invece?
A: niente io sto_ sto molto bene
sono un po'_ cosi' un po'_ # ...
B: he’s a nice guy, as well
A: okay
B: and you, for your part, how are you?
A: nothing, I feel_feel very good I’m a little_ ah a little_ # ...
(LIP RA1)
On the level of superstructure, they can function as initial signal for the center part of the
conversation:
(19) B: ahah Giovanna insomma ci si sente

A: va bene
B: va bene
A: okay
B: ciao ciao
A: ciao
B: ahah Giovanna, so, we will speak again
A: okay
B: okay
A: okay
B: bye, bye
A: bye
(LIP FB1)
372 B. Frank-Job
4.3. How do the participants handle the multifunctionality of DMs?
The question remains as to how the participants understand the correct meaning of DMs
and react adequately to it in conversation. We have seen that the first and basic function of
DMs lies on the level of the succession of turns. This very fact shows that it is up to the
participant in the ad hoc situation to decide upon the value of a given DM. From the
perspective of the hearer, a DM that closes a turn, for example, presents a choice. He is
free to take the turn and continue the thematic sequence, he can start a new theme, or he
can start the routines to end the conversation.
Of course, the range of possible and adequate reactions is not completely open. It is
determined on the one side by the basic function of the DM—to close, to open, to prepare
—and on the other side by the three levels of conversation processing: a turn, a thematic
sequence, the conversation.
Thus, the multifunctionality of DMs evolves in response to the dynamics of free spoken
conversation where each turn, each new theme, and each start or end of conversation as a
whole has to be negotiated spontaneously. The processes whereby mutual meta-
communication indicates possible moments of turn change are necessarily open ones, to
be determined by the negotiations of the participants.
5. Perspectives
As we have seen, at least a great part of the initially mentioned problems in the analysis of
DMs as a functional class are due to the fact that DMs belong to the linguistic domain of
discourse, a domain which is essentially determined by the dynamics of ongoing
interactional processes. In discourse, meanings are not predefined and stable but
constantly negotiated, altered, innovated, and attributed to the different levels of discourse
processing. In the study of DMs, this basic value of any element of the analytic level of
discourse has to be taken seriously.
Notes
1
Cf. Croft (2000) for the definition of lexical vs. grammatical meaning.
2
For Coseriu, discourse or “text” represents an autonomous linguistic level.
3
Lessico di frequenza dell’italiano parlato a cura di Tullio de Mauro, Federico Mancini,
Massimo Vedovelli, Miriam Voghera, Etaslibri, Fondazione IBM Italia (Milano) 1993
(LIP).
4
Datenbank Gesprochenes Deutsch of the Institut für Deutsche Sprache, Mannheim:
http://dsav-oeff.ids-mannheim.de/DSAv/.
5
This definition excludes modal particles from the class of DMs.
6
As a result of this process however, grammaticalization and pragmaticalization differ
considerably. Thus, grammaticalization results in the formation of new grammatical
items:
[grammaticalization is] le processus par lequel une unité lexicale d’une langue se
développe, au cours du temps, en unité grammaticale, ou une unité grammaticale
en unité plus grammaticale encore. . . . ce qui est en cause, c’est l’évolution

morphogénétique par laquelle les langues spécifient leur grammaire. (Hagège,
2001: 1608-9)
The relations between grammaticalization and pragmaticalization, though, will not
form the center of my attention. In my eyes, the main difference between the two
processes consists in the linguistic status of the domain in which the new (pragmatic)
meaning functions. In the case of grammaticalization, the domain to which the new
meanings belong is that of the grammar of a historical language. In the case of DM,
the domain is that of discourse (Coseriu, 1982; see note 3 above). With this distinction
in mind, I also refer to Oesterreicher (1997) and to Koch (forthcoming). In addition to
Coseriu, Koch and Oesterreicher distinguish on the historical level between
Einzelsprache (“historical language”) on the one hand and Diskurstradition
(“discourse traditions”) on the other. The turn-taking rules that fulfil DMs in spoken
conversations belong to the second level, whereas the grammatical norms and rules
belong to the first one.
7
For the problem-solving and routinization aspect of grammar see Lüdtke (1988) and
Hagège (2001).
8
It is characteristic for all long term processes in language change that for a relatively
long period of time, new meanings and functions coexist with older meanings and
functions.
9
Another view on the same processes is given by Waltereit (1999, this volume). For the
parallel processes of grammaticalization of German modal particles see Diewald (this
volume). For studies on diachronic aspects of DMs, see Stein (1985), Brinton (1996),
Onodera (1995), Manoliu (2000), Schwenter and Traugott (2000), and Auer and
Günthner (2003). Only a few studies exist on spoken discourse in past stages of
Romance languages; see Spitzer (1922), Schlieben-Lange (1983), Koch (1995,
forthcoming).
10
“Das sprachtheoretische Axiom, daß alle Sprachzeichen Symbole derselben Art sein
müssen, ist zu eng; denn einige darunter wie die Zeigwörter erweisen sich als Signale.
Und von einem Signal darf man nicht dasselbe verlangen wie von einem (reinen)
Symbol, weil zwischen beiden ein sematologischer Unterschied besteht. Die
Zeigwörter sind eine eigene Klasse von Signalen, nämlich Rezeptionssignale
(verschieden von den Aktionssignalen, zu denen der Imperativ gehört). Ein dér oder
ich löst eine bestimmte Blickwendung u. dgl. und in ihrem Gefolge eine Rezeption
aus.” (Bühler, 1934: 52-57)
11
The occurring DMs are the following (in the order of their frequency): ma (176
occurrences), ciao (122), ecco (97), pronto (77), va bene (68, among them 22 va be'),
senti (55), allora (44), dimmi (23), sentiamo (13), ho capito (12), vedi (12), guarda
(9), insomma (9), niente (7), volevo dire (2); the total number of words is 7739.
12
See Moeschler (1996: 191): “bon ben oui – mais là c’en était pis voilà”.
13
For the analysis of typical prosodic features belonging to DMs, see Bazzanella (this
volume, section 2.2.3).
14
See Bergmann (1981), van Dijk (1980), Fritz (1994), Henne and Rehbock (1982),
Mondada (2001: 6), Gülich and Mondada (2002: 206ff.).
374 B. Frank-Job
15
In most of the cases, this basic discourse-marking meaning conserves central features of
the propositional meaning of the same word or expression. See Bazzanella (this
volume, section 2.2.2); Fischer (2000a, this volume) for the concept of a “core”
meaning which the DM and its propositional origin have in common.
16
See also Schwitalla (1976: 82-83) and Bazzanella (1990: 640). In many languages,
DMs have been pragmaticalized out of these direct verbalizations of turn-taking
devices.
Chapter 20
Discourse particles as morphemes and as

constructions
François Nemo
1. Introduction
Discourse particles (DPs) are lexical items with puzzling semantics, puzzling pragmatics,
puzzling syntax, and interesting morphological diversity. If they have for long been
considered as a minor word class, somehow marginal to core linguistic mechanisms, their
study since the late sixties has become the cradle of contemporary linguistic semantics,
and a unique window onto both the complexity of language construction and
interpretation and the understanding of what meaning is about. Presenting possible
solutions to some of these well-documented puzzles is the aim of this paper.
1.1. Approach
I shall be using the four-layered model of semantic (and pragmatic) interpretation

developed in Nemo (1999, 2001a) which distinguishes between (i) contributions, (ii)
utterances, (iii) constructions, and (iv) morphemes. I shall first distinguish between
morphemes which may or may not be used in connective position and the connective
constructions into which these morphemes are inserted and assume that we need
morphemic semantics to describe the former and constructional semantics to describe the
latter. By doing so, the model proposed differs from all models that accept lexemes, such
as the connective but, the adverb but (only, just), the preposition but (except), as input to
semantic models, and assumes that lexemes are the output of a process in which
morphemic information, constructional interpretation, and contextual specifications are
combined by the use of a noncategorial morpheme in various positions and contexts.1 I
shall assume that producing lexemes, as well as nonmemorised interpretations, is the goal
of a semantic-pragmatic theory.
Similarly, I shall distinguish between utterances and contributions, showing that the
conflation of the two levels in Grice’s work has led to important difficulties in our
understanding of discourse and pragmatics and that a distinction between these two levels
is necessary for defining two subclasses of DPs, namely speech particles (SPs), whose
376 F. Nemo
scope are individual utterances, and discourse connectives (DCs), whose scope are
contributions.
I shall assume that there are specific types of constraints associated with each of these
levels and that accounting for all the issues at stake in the description of DPs (see section
1.5.) requires understanding all of them.
1.1.1. Contributions: the missing link

Contributions are sets of utterances bound by joined attention—which implies that they
must be considered together—and are individual or collective definitions of what has to
be taken into account at a given moment about a given topic.
The criteria for distinguishing contributions from utterances—apart from size since
contributions usually consist of a group of utterances—is the nature of the pragmatics
constraints at work (Nemo, 1999). The maxim of quantity for instance is a pragmatic
constraint whose scope is a contribution, since being informative enough implies most of
the time saying more than one utterance, whereas the scope of modal and scalar
constraints of relevance described in Nemo (1992, 1999) is every single utterance.2
Although contributions may sometimes be reduced to a single utterance,3 it is important to
separate the mechanisms at work and hence to allow a full prediction of interpretation;
relying on communicational-attentional constraints (Grice, 1975; Sperber and Wilson,
1986) and overlooking all the framing machinery at work is as problematic as focusing on
framing mechanisms (Anscombre and Ducrot, 1983; Nemo, 1992; Levinson, 2000) and
presupposing attentional constraints without providing an account for them.
Distinguishing between utterances and contributions is necessary in order to understand

the variety of uses of specific DPs:
1. DCs, but not SPs, are contribution builders (ex ante) or contribution modifiers (ex
post);
2. connecting utterances (or segments) within a contribution, and connecting
contributions are different tasks, and many DCs have both intra-contributional uses (in
which they are contribution builders), inter-contributional uses (in which they are
discourse builders), and extra-contributional uses (in which they build separate
discourses).4
What do we know about contributions? Surprisingly, quite a lot, if we assume with both
Argumentation in Language Theory (ALT) and Relevance Theory (RT) that speaking is
basically a matter of attracting the hearer’s attention to something and asking him or her
to take it into account, which is what ALT calls argumentative orientation and what RT
calls ostensive-inferential communication. This characterisation is probably the most
general result of contemporary pragmatics: controlling somebody else’s attention and/or
building a common attentional ground is a very old game that can be played also without
language.5
Beyond attracting attention, the issue at stake at the contributional level is the definition
of what has to be taken into account.6 The pre-Gricean, Gricean and post-Gricean idea
that there are norms that regulate what can or should be focused on should thus be
integrated in a more general model of attention, including manipulative (and goal-
Discourse Particles as Morphemes and as Constructions 377
oriented) descriptions of controlled attention.7 This would enlarge our understanding of

what is at stake in verbal communication: describing uninformative contributions requires
explicating their attentional value and explaining, for instance, why a fifteen year old son
may tell his parents I am not three years old, why a mom may tell her fifteen year old kid
I am not your dad, or why a dad may say to his son I am your dad, either to tell him
“don’t talk to me like that” or to tell him “you know you can talk to me”. Such
contributions, formed here of a single utterance, are indeed embedded in attentional
frames such as I want to attract your attention to the fact that I am not three years old (or
not your dad, etc.) and I am asking you to take it into account. The use of DCs in It’s a
very interesting proposition but I am exhausted those days will be described along similar
lines as I am taking (have taken) into consideration the fact that it is a very interesting
proposition but the fact that I am exhausted those days must also be taken into
consideration.
Similarly, noncontrastive uses of but such as Peter is tall but so is John, when their
attentional frame is made explicit, are described as: we must take into consideration the
fact that Peter is tall but we must also take into consideration the fact that John is tall
(too).8
1.1.2. Utterances
Apart from their contributional values, utterances are related to constraints which are
specific to them. Utterance-type semantics (Nemo, 1986, 1999; Levinson, 2000; Nemo,
2003b) distinguishes between the proposition p and the utterance (p); furthermore,
utterance-type semantics does not describe this difference as a matter of an encounter of p
with a set of previously stored utterances (called common knowledge or cognitive
environment; Sperber, 1974; Sperber and Wilson, 1986). Instead, utterance-type
semantics assumes a difference between the semantic content of proposition p and the
semantic content of utterance (p), i.e., of p when it is said. In other words, I assume that
• uttering a sentence presupposes that its context is appropriate (Keenan, 1971),

• the conditions of appropriateness of an utterance are an important part of its semantic
content since they are elements of meaning for the hearer (Verschueren, 1980: 281),
• and a large part of this semantic content is built up from the morphemes used, i.e.,
morphemes are modal frame builders (Ducrot et al., 1980; Nemo, 1992).9
These joined hypotheses imply a clear distinction between two kinds of implicit semantic
content. Semantic frames (SFs) are implicit comparisons (upstream implicit), whereas
scalar values (SVs) are the result of these comparisons (downstream implicit), which are
the pragmatic heads of utterance interpretation but are semantically less important than
the upstream implicit.10
Describing utterances thus implies considering them as an association of an image of what

is the case with an image of what is/has been possible (Nemo, 1988, 1992, 1996, 1999;
Krifka, 1999; Levinson, 2000). Thus, utterances are not propositions, since the
proposition Bill Clinton is alive describes any moment in which Clinton is alive as true,
whereas the utterance “Bill Clinton is alive” describes only a moment in which something
has happened (car accident, heart attack, killing attempt, pretzels) that renders it relevant
to consider whether he may or may not have been alive (Nemo, 1988). Consider the
following examples:
378 F. Nemo
(1) C’était loin donc j’ai pris la bicyclette. (It was far so I took the bicycle.)
(2) C’était pas loin donc j’ai pris la bicyclette. (It wasn’t far so I took the bicycle.)
Interpreting (1) requires associating with proposition 1a and 1b the framing alternative
(bicycle or foot), whereas (2) introduces the framing alternative (car or bicycle). All
utterances are thus described in terms of the comparison set they introduce, in terms of the
possibilities which are simply ignored (since they don’t belong to this comparison set),
and finally in terms of what is the case in all the alternatives which are introduced.11 The
result of the comparison introduced is called the scalar value of the utterance.
This scalar value—i.e., the answer to the question “what difference does it make?” (for
instance I shouldn’t be treated like that for I am not three years old)—is thus the
equivalent of ALT’s conclusions and RT’s inferences. However, scalar values are
differences, not inferences, and thus not the result of the encounter of a new proposition
with a pre-existing set of propositions (the cognitive environment) but the result of the
comparison of an existing world with alternative worlds, i.e., of what is the case with
what could have been the case (see Nemo, 1996, 1999).
I shall also assume that since describing utterances (Nemo and Cadiot, 1997b) implies
considering simultaneously all the dimensions of the space in which they occur, namely,
the interpersonal (i.e., faces, power, politeness, etc.), interlocutive (i.e., definition of the
addressee, turns, exchange status), scalar (i.e., the “so what?” question), epistemic (i.e.,
the image of what is possible and the image of what is the case), and dynamic (i.e., the
given/new status of all this information), a linear account of utterance interpretation is
impossible and codetermination of the different values (interpersonal, interlocutive,
scalar, and epistemic) is the rule and not the exception. As for the study of DPs, I shall
assume that this multidimensional and dynamic nature of utterances is responsible for a
large part of their polyfunctionality and for the failure of all those attempts that reduce
DPs to any specific dimension. The linguist’s dream of linking specific DPs with specific
speech or discourse moves, so that DPs would express, make explicit, or mark such
moves, is indeed a dead end.12 It can only be partially rescued by restricting the
description to certain prototypes.13
1.1.3. Morphemes and constructions

I shall assume that there are on the one hand form-meaning pairs, called constructions
(Goldberg, 1995), that exist independently of the specific semantic units used, and on the
other hand form-meaning pairs, called morphemes, that exist independently of the specific
constructions into which they are inserted. Morphemes thus are defined as semantic, but
not syntactic, units. The main consequence of this distinction is that the semantics of the
morpheme is never assumed to be a prefiguration of the construction in which it is
inserted. This allows the morpheme to be inserted in completely different constructions
and constructional positions: describing p but q and describing but are hence completely
different tasks.14
The semantic treatment of morphemes that I shall use here is indicational-indexical.

Morphemes are supposed to encode indications—for instance, the “there are X and Y”
type of information. The interpretation of morphemes is based on a single indexical
procedure associated with the indication, namely, “look for X and Y”. I have shown
elsewhere (Nemo, 2001b) that since Ducrot’s notion of instruction includes both elements
(indication + indexical principle) and implies a process of contextual unification (finding
something that may unify with X), renaming these instructions procedures (as in Wilson
and Sperber, 1990; Roulet, this volume) is misleading: instructions have more in common
with clauses of a declarative language such as Prolog than with any step-by-step
procedural process (see Luscher, 1994, for a coherent procedural treatment of DPs within
RT and Nemo (2001b) for a discussion of it). However, the NSM description proposed in
Travis (this volume), for instance, is compatible with the basic assumptions of
Instructional-Indexical Semantics (IIS).
1.2. Methodology
The methodology used is basically distribution-based (D-methodology), whose goal is to

generate all uses and explain all constructions occurring. Thus, this D-methodology does
not assume that some uses are more central than others and that, for instance, describing
Peter is tall but John is small would be any more important than describing Peter is tall
but so is John. It is also assumed that the most important issue in semantic description is
to separate encoded indications from constructional interpretations. Accounting for all of
the uses is of course hard, but it is the only scientific task: there is no way semantics and
pragmatics can ever pretend to produce knowledge if their claims are only
commonsensical overgeneralisations based on chosen fragments. The opposite
methodology and strategy is thus an explicativist approach according to which
considering all constructions in which a morpheme is inserted at the same time is the only
way to avoid projecting constructional semantic features into the morpheme’s encoded
meaning. Semantic descriptions therefore have to start from the observation of apparent
paradoxes and the falsification of intuition.15
D-methodology is a broadly applicable semantic methodology that has been used to

account for all kinds of semantic units, from nouns (Cadiot, 1993; Nemo, 2002b) to
affixes (Nemo, 2002a), from discourse words to adjectives (Bouchard, 1995) and
prepositions. It has also been tested on polycategorial units (Nemo, 1998, 1999).
1.3. Data
All the data used in this study have been extracted through data mining. Data mining is
achieved on different types of corpora: newspapers, books, movies, talk shows,
conversations. Three sets of uses are defined: a narrow set (around 200 to 400 uses) that
must be completely accounted for, a wider and open set that is explored in order to find
rare uses that are not present in the original set, and an ex-post testing set whose role is to
quantify the predictive capacity of the descriptions proposed.
Picking out or selecting examples is forbidden for the first and third sets and allowed only
for the second set, whose aim is to increase the probability of including less frequent uses
in what has to be accounted for. In that sense, a distinction can be made between the data
whose description (sets 1 and 3) has got to be exhaustive and the wider data whose
treatment is restricted to interesting and problematic examples.
380 F. Nemo
Creating examples (or interpretations) is allowed only in order to test (and falsify)
hypotheses, i.e., to produce negative evidence. The acceptability of the examples must be
uncontroversial.
1.4. Studying DPs: development of the field and issues at stake
During the past thirty years, the study of discourse particles (or mots de discours,
discourse connectives, argumentative connectives, discourse markers, etc.) has become a
field in itself. It has also, and more importantly, turned into a pioneer front for (general)
semantics and (general) pragmatics. Consequently, descriptions and observations are now
available for a wide range of languages, often for a wide range of particles within each
language and—sometimes for the very same particles—in the most diverse semantic and
pragmatic frameworks. Starting in the late sixties and early seventies, this development
has been gradual, although with clearly distinct periods and perspectives. Since new
directions and dimensions have been explored successively, and almost all the general
models in semantics and pragmatics (from Argumentation Theory and Relevance Theory
to Conversational Analysis and DRT) have been used and tested in the description of DPs,
it is clear that what is most needed is an integrative perspective, in order to (i) clarify what
any theory of DPs should be able to account for, (ii) spell out what existing models agree
on, and (iii) identify the dividing lines among the existing models in the description of
DPs.
1.4.1. Dividing lines: DPs as markers or DPs as modifiers

The most important dividing line between current descriptions of DPs lies between
models defending what should be called the Discourse-Marking Hypothesis (DMH) and
models which favour what should be called the Modifier Hypothesis (MH). As for the
description of a discourse formed of two utterances (or discourse segments) U1 and U2
joined by a discourse connective C, the difference between the two hypotheses is that
within the DMH, C is believed to mark explicitly the relationship between U1 and U2,
whereas within the MH, C is believed to deal with (or make explicit) the relationship
between [U1] and [U1U2].16 In other words, the DMH approach considers that C deals
with the relationships between subsequent utterances (or discourse segments of any size)
whereas according to the MH approach C deals with alternative discourses and the nature
of this alternative. Saying “A” and saying “B” on the one hand is different from saying
“A” and saying “AB” on the other hand (see Ducrot et al.’s descriptions of mais, enfin,
and toujours; see Luscher’s (1994) procedural description of mais, d’ailleurs, etc.).
While the DMH is probably as old as there are dictionaries, the first descriptions which
are coherent with the MH can probably be found in Anscombre and Ducrot’s pioneering
work (1976, 1983; Ducrot et al., 1980) on discourse words. Both hypotheses may be
found nowadays in linguistic literature, some authors actually adopting both of them
simultaneously.
1.4.2. Dividing lines: FF-methodology versus D-methodology

As for methodology, another dividing line must be traced between those who favour a
paradigmatic, systemic, or functional family (FF) approach, according to which DPs
marking the same kind of relationship must be studied contrastively,17 and those who
consider the distribution of isolated DPs. Consequently, crucial issues at this
methodological crossroads are the problem of polyfunctionality and the question of how
to account for the semantics of DPs.
Within the FF methodology, three positions may be defended regarding the polyfunctional
semantics of DPs. First, one can assume (implicitly) the existence of sense-enumerative
lexicons in which the various uses of a DP are described as different DPs,18 so that, for
instance, deductive donc (i.e., therefore, thus; Jayez and Rossari, 1998, 2000) or
reformulative enfin will be described independently of the other uses of donc or enfin
(Rossari, 1994). Second, one can adopt prototypicality and assume that even though a
given DP may be used in very different ways, it is possible to consider some uses as more
central than others and to postulate desemantisation and resemantisation processes (see
Fraser, 1996, 1998). Third, one can assume that whenever DPs are somehow
substitutable, we need to postulate the existence of a construction (i.e., of a form-meaning
pair which exists independently of the specific units involved (Goldberg, 1995)) and to
consider polyfunctionality as a result of the insertion of the same units in different
constructions.
1.4.3. Dividing lines: dealing with semantic continuity

As for the tenants of the distributional method, things are a little more complex. The study
of the distribution of a given semantic unit may well lead beyond class boundaries, and
thus force the linguist to postulate (or not) semantic continuity between the different uses,
and adopt (or not) a noncategorial framework.
A typical example of the problem linguists must face when adopting a D-methodology is
the distribution of English but across word classes. Since but may be used simultaneously
in connective and nonconnective positions (i.e., with the “meanings” of almost, without,
except, only, etc.), linguists must decide whether they are actually dealing with the same
unit or not, and if the answer is yes, they must be able to find the meaning encoded by the
morpheme.19
As for the semantics of DPs, there are three positions which are compatible with a D-
methodology, namely:
1. accepting some kind of polysemy and thus describing the semantics of DPs in terms of
more or less shared procedural information (Luscher, 1994);
2. adopting a single semantic description for all the (categorial) uses of a given DP
(Anscombre and Ducrot, 1977; Ducrot et al., 1980; Cadiot et al., 1985);
3. adopting a single semantic description for a (categorially unspecified) morpheme, the
DP uses of the morpheme being described as the insertion of a morpheme in
connective position, i.e., being described as morpheme-construction pairs.
I shall defend and illustrate here the third position, which contrary to the first two
positions, challenges the classical distinction as the conceptual/procedural distinction or
the idea (Wilson and Sperber, 1990) that linguistic items provide either information about
the world or about how to process such information.
The possibility of providing type 2 semantic descriptions of DPs was probably the most
important contribution of DP studies to general semantics and was Ducrot’s most
important legacy in all kinds of frameworks. Nevertheless, it must be said that from the
382 F. Nemo
very beginning, Ducrot and Anscombre’s descriptions were not actually unitary
descriptions of the encoded meaning of DPs such as mais, but rather a demonstration of
semantic continuity between all of its uses.20 It is also worth noticing that even though
Ducrot’s notion of semantic instruction (Ducrot et al., 1980; Ducrot, 1987) has later been
used in all fields of linguistic semantics and is thus not restricted to the description of
DPs, Ducrot himself has never tried to enlarge the scope of the study to nonconnective
uses and would probably not treat but for X, Y in the same movement as X but Y.
1.4.4. Dividing lines: DPs and discourse theories

A final dividing line between the different approaches of DPs addresses the relationship
between studying discourse (i.e., discourse or conversation theory) and studying discourse
particles. Roughly, three positions may be distinguished:
• the heuristic of discourse acts (Fraser, 1996), according to which we can study DPs
directly by studying what we do when we use DPs, without any more general theory of
discourse;
• the heuristic of discourse theory, according to which the right metalanguage to
describe DPs is the metalanguage coined by a theory of discourse (or conversation);
• the noncategorial heuristic, which I shall defend, according to which we cannot
understand how DPs are interpreted without understanding what discourse,
conversation, or utterances deal with (and what connecting utterances is about), but
according to which the morphemes used as DPs do not encode any information about
connecting discourse.
Tenants of the second position may be considered mainstream, even though the diversity
of the models of discourse/conversation/language which are used (and consequently the
diversity of the metalanguages used) makes it a very divided mainstream. All descriptions
which take for granted that the same terms can be used to describe what language,
discourse, or conversation is (and how it works) and what the meanings of DPs are belong
to this second kind of approach, irrespective of whether their metalanguage is inherited
from argumentation theory, relevance theory, conversational analysis, conversation theory
(i.e., the Geneva School), coherence theory, Culliolian linguistics, etc.
The third position I shall defend is a consequence of the noncategorial, distributional

approach previously described. If studying But for X, Y, studying P but Q, and describing
All but x Verb is in fact studying a single morpheme and three constructions, constructions
in which the indications encoded by the morpheme are interpreted in (at least) three
different ways, then but does not forcefully encode a description of the relation between P
and Q in P but Q.21
1.4.5. Issues at stake

Comparing approaches to discourse particles is not an end in itself, even if it helps clarify
some important issues and allows us to predict what each approach will favour, defend,
and ignore, as well as the nature of the criticisms they would make to each other. It is
surely more important to wonder whether these approaches are to be thought of as
alternatives (and in case they are, to find a way of testing them) or as complementing each
other.
I will assume that but for the approaches in terms of prototypicality, the current diversity
of approaches of DPs allows us to spell out what the issues at stake are, which any
description of DPs should fulfil. These are, namely,
• to account for the (functional) diversity of uses of DPs, both as DPs and as anything
else (i.e., to account for the semantics of DPs), which supposes being able to separate
encoded meaning (if any) from local interpretations (even when these interpretations
are conventional);
• to account for the pragmatics of DPs (i.e., to be able to describe and explain the
pragmatic diversity of the uses of DPs), and specifically to account for the non-
metacommunicational uses of DPs, for instance, interjective uses;
• to account for the semantic continuity of the uses of linguistic items as DPs and in
other syntactic positions (adverbs, prepositions, etc.);
• to account for the possibility of exchanging DPs for each other, i.e., for the fact that
they seem to share a part of their semantic content and thus constitute paradigms, but
also to account for the differences between them in terms of the syntactic, semantic or
pragmatic nature of the arguments they require ;
• to account for the semantics of paradigmatic and functional families.
Until now, none of the models available has been able to meet such specifications or has
even accepted all these requirements. Consequently, it is easy to compare their respective
agenda with the general agenda provided here. As for myself, I shall attempt to show that
accounting for these questions requires adopting a morpheme/construction distinction,
i.e., adopting a frame in which a clear-cut distinction is made between
• morphemes as form-meaning pairs that exist independently of the constructions in

which they are inserted, whose study is distributional and semantic;
• constructions as form-meaning pairs that exist independently of the specific semantic
material which is used;22
• lexemes as morpheme/construction pairs that combine (encoded) morphemic
information and inherited constructional information (see Fischer, this volume).
This allows me to follow simultaneously (i) a D-methodology in the treatment of

morphemes, (ii) a paradigmatic perspective on constructions, and finally (iii) to account
for the fact that since the same morpheme may be used in different constructions, many
lexemes may be sharing a single phonological form and a single encoded morphemic
meaning and at the same time be completely different units as far as combinatorial
properties and inherited constructional meaning are concerned.23
1.5. Mission statement
In my view, answering the questions stated above implies considering the semantic
description of DPs as the main problem we have to solve: it is clearly the case that
describing each use of a DP separately is possible within all frameworks.24
As mentioned earlier, the central paradox we must face is that even though we would like
to think that DPs guide the interpretation of the cotext (and context), the diversity of
interpretations they receive from one context to another shows all too clearly that they are
very context-sensitive. Models that ignore polyfunctionality or assume the existence of
384 F. Nemo
sense-enumerative lexicons without accounting for the various senses cannot thus be
considered explanatory.
As for here, I shall thus consider that we need to test which of the following statements is
correct:
• the polyfunctionality of DPs is an example of Wittgenstein’s “language is use”

hypothesis, according to which semantic units have no stable semantic content but are
embedded in language games in which they take sense (thesis A);
• the polyfunctionality of DPs is a matter of contextual desemantisation and bleaching
(thesis B);
• the polyfunctionality of DPs is an example of diachronic fossilisation of local
pragmatic interpretations, in a Benveniste-Traugott way (Benveniste, 1966, 1974;
Traugott and König, 1991) (thesis C);
• the polyfunctionality of DPs is linked with polyprocedural semantic content, as in
Luscher (1994) (thesis D);
• the various uses and interpretations of DPs are all based on the same initial semantic
instruction, in Anscombre and Ducrot’s way (thesis E);
• the polyfunctionality of DPs is not a matter of semantics since DPs do not encode
functions but stable nonfunctional meanings, in the IIS way. In other words, what we
need is to describe the relationship between functional interpretations and encoded
meanings (thesis F).
In order to do so, I have to formulate an illustration of thesis F about English but, which
could be tested in contrast with existing accounts of but by proponents of theses A to E.25
2. Definition
Describing word classes in general, and describing discourse particles in particular, may
easily turn into an impossible mission when it comes to producing crosslinguistically
valid generalisations (Vogel and Comrie, 2000; Fernandez-Vest, 1994). As for DPs—and
this is one of the reasons why the labelling of the class has remained so problematic—
there is indeed enough typological variation in all the aspects of what has to be described
and defined (i.e., morphological, syntactic, semantic, or pragmatic properties) to falsify
most of the claims we could be tempted to make about them, for instance, by focusing on
supposedly prototypical properties.
A recurrent problem is that defining a word class implies two parallel but independent
postulates: (i) the possibility of identifing testable formal properties defining class
membership, and (ii) the possibility of associating these formal properties with a lexical
class, i.e., a (possibly open) list of class members.
Experience shows (Broschart, 1997; Amfinn, 2000; Anward, 2000; Bhat, 2000) that the
existence of word classes is problematic whenever the same semantic material is used
(indifferently) in different “classes”, i.e., whenever the lexical class associated with a
formal definition overlaps the lexical class associated with another set of formal
properties. So that although the linguist’s dream would be to find something like that in
Table 1, what we are faced with is often something like that shown in Table 2 (Broschart,
2000; Dhat, 2000):
Table 1. Lexical classes and formal properties
Formal definition 1 Formal definition 2 Formal definition 3
Lexical class 1 Lexical class 2 Lexical class 3

items 1, 11, 21, 31, items 2, 12, 22, 32, items 3, 13, 23, 33,
etc. etc. etc.
Table 2. Lexical classes and formal properties
Formal definition 1 Formal definition 2 Formal definition 3
Lexical class 1 Lexical class 2 Lexical class 3

items 1, 11, 21, 31, items 1, 11, 21, 31, items 1, 11, 21, 31,
etc. etc. etc.
The lack of a one-to-one correspondence between lexical sets and formal definition is
what we are facing when describing DPs; identifying DPs (especially DCs) is not very
difficult in most languages, but any definition of DPs will have to deal with the fact that
this definition does not allow the prediction of the actual uses of most of the items
involved, which are routinely used in other functions as well.
2.1. Defining DPs
To avoid most of the acknowledged shortcomings of classical definitions of DPs,

discourse particles will be defined here either as lexicalised (or grammaticalised)
utterance modifiers called speech particles (SPs), or as lexicalised contribution modifiers
called discourse connectives (DCs), just as adverbs are verb modifiers and adjectives are
noun modifiers (Hengeveld, 1992).26
The distinction between lexicalised metalinguistic devices and nonlexicalised

metalinguistic comments is essential for defining DPs as a lexical class and not in terms
of function, since the same function may be fulfilled by all kinds of linguistic devices.27,28
It is also essential to understand that the study of DPs is forcefully included in a more
general account of the pragmatics of contributions and the semantics and pragmatics of
utterance frames, i.e., in the study of how “viewpoints are integrated into structure”
(Nølke, 1993). DPs thus comprise all linguistic devices that introduce meta-
communicational information, as long as the linguistic expressions involved behave like
lexical items and range from grammaticalised lexical units (e.g., evidentials) to rather
“free” particles.
386 F. Nemo
This definition has several advantages; it avoids syntactic overspecification,

constructional overspecification, and semantic overspecification and forces the linguist to
explain the uses of DPs that do not match with this definition.
2.2. Constructional underspecification
Defining DPs (SPs and DCs) as modifiers allows us not to consider them as markers of
relations between specific types of linguistic elements, for instance, as markers of
discourse relations between utterances, whose function would be to ensure cohesion of
discourse or to make explicit discourse relations. Thus, we can deal with the indisputable
syntactic indetermination of DPs and the limits of classical grammatical distinctions
between conjunctions, phrasal adverbs, sentence adverbs, etc. These distinctions overlook
that some DCs do not introduce any linguistic element and that most DCs can introduce
discourse segments and discourse fragments ranging from adverbs, or phrases completing
a prior utterance, to new predications, metalinguistic comments, etc. So that we have
simultaneously:
(3) Il ira jusqu’à Dijon. Et encore! (He shall reach Dijon, if even!)
(4) Ça a été n’importe quoi! Enfin! (This is nonsense! Well forget it!)
(5) Ça a été n’importe quoi! Enfin, on va faire avec. (This is nonsense! No matter,
we’ll manage.)
(6) Il est arrivé en retard. (He arrived late)
(7) Il est arrivé. En retard! (He arrived. Late!)
(8) Il est arrivé. Mais en retard! (He arrived, but late!)
(9) Il est arrivé. Mais lentement. (He arrived, but slowly.)
(10) Il est arrivé. Mais à Paris. (He arrived, but in Paris.)
(11) Il est arrivé, avec Paul d’ailleurs. (He arrived, with Paul actually.)
(12) C’est un ami, de longue date d’ailleurs. (It’s a friend, a very old friend in fact)
(13) Ici, le port de la cravate est une tradition, et encore pas n’importe laquelle.
(Here, wearing a tie is a tradition, and furthermore not any kind (of tie).)
The classical idea that DCs connect two utterances, U1 and U2, may thus be considered as
an oversimplification of the most simple data, since the hypothetised U2 is very
frequently a simple modification of an element of U1 and in such cases DCs would be
better described as ex-post introducers of modifiers and modifications than as connecting
two utterances. In other words, what data mining shows very clearly is that connecting
two distinct utterances U1 and U2 is only one end of the spectrum of what DCs can do
syntactically, and that we have to admit that:
• they sometimes introduce no linguistic element at all, as in examples (3) and (4);
• they often introduce elements (fragments) which either complement or replace
elements of the previous utterance, as in examples (6) through (12), and thus could be
syntactically integrated to this utterance (I-fragments);
• they sometimes introduce elements which can be articulated syntactically with (but not
integrated in) the previous sentence, but which cannot stand alone syntactically (A-
fragments), as in example (13);
• they sometimes introduce elements that cannot be integrated syntactically in the
previous utterance but can syntactically stand on their own, as in example (5).
Such syntactic behavior cannot be explained within a theory of discourse or discourse

relations, since both only consider the last case and explicitly deal with the description of
relations between syntactically autonomous elements.
In contrast, defining DCs as contribution modifiers allows us to account for this syntactic
behavior. If DCs do not rule out syntactic fragments for instance, or zero-segments, as
arguments, it is because they require only ex-post attentional modifications, no matter
what exactly is being modified. This is what Nølke (1993) has called the integration of
viewpoints into structure and what Fillmore et al. (1988) have described as multifocused
constructions. The indisputable continuum between the four types of syntactic arguments
DCs accept (zero, I-fragments, A-fragments, utterances) is a consequence of the fact that
the only constraint is contribution modification.29
2.3. Semantic underspecification
Defining DPs (SPs and DCs) as modifiers also allows us not to consider them as
forcefully encoding procedural information:
• it is obvious that lexicalized modifiers may perfectly well communicate conceptual

content; the French very successful intercontributional DC sans transition (‘without
any transition’) cannot be said to encode any procedural meaning;
• it can be shown that morphemes used as DPs may perfectly well encode declarative
(and not procedural) information, i.e., indications, with the indexical instruction to find
their cotextual and contextual counterparts;
• some DPs, such as et encore in example (3), clearly contribute to the truth-conditional
semantic content of what is said;
• the distribution of many of the morphemes used as DPs or in discourse-modifying
expressions overlaps the conceptual/procedural distinction, leaving us with little
choice but to either consider the various uses as mere homonymies or to postulate a
semantic continuity between all the uses that would abolish the conceptual/procedural
distinction.
Many of the usual claims according to which DPs forcefully encode procedural content
must thus be considered as at least controversial, especially when such general claims are
not actually backed by any procedural description of specific DPs or are simply based on
a strange conflation of the idea that language either communicates information about the
world or about language itself and the idea that language encodes either conceptual
information or procedural information. On the contrary, it must be acknowledged that
many DPs have explicit semantic content (e.g., sans transition) and that the
388 F. Nemo
polyfunctionality of the others shows clearly that they cannot be said to encode such or
such function or procedure. Thus the linguistic encoded meaning of such morphemes, if
any, is forcefully functionally unspecified.
2.4. Pragmatic underspecification
Defining DPs (SPs and DCs) as modifiers allows us to consider their nonmetalinguistic
uses in a more integrated way. As mentioned earlier, one of the most puzzling difficulties
in the description and definition of DPs is the fact that many of the lexical items used as
DPs may also be used in other ways which are not consistent with our definition of DPs,
namely, indexical (and interjective) uses on the one hand (as described in Ducrot et al.,
1980; Rouchota, 1996), and mind-reading uses on the other hand (as described in Ducrot,
1980).
In order to clarify this issue, we need to understand that attracting somebody’s attention to
something and asking him or her to take it into account (in the Ducrot-Anscombre-
Sperber-Wilson way) is exactly what indexical (interjective) uses do, whereas mind-
reading uses consist in following somebody’s attention shifting from one thing to another.
There is a pragmatic continuity between all these types of uses: attentional shifts may be
described (hence the mind-reading interpretation) or provoked; in the latter case, either by
directly guiding real-life attention (hence the indexical uses, Mais arrêtez-les! ‘But stop
them!’) or by redirecting discursive attention, as in all the classical Ducrotian
argumentative uses. This explains why the linguistic items used as DPs—mais, bon, bien,
sans transition, etc.—may also be used in nondiscursive contexts: connective
constructions are the linguistic form of a much more general attentional competence
(Nemo, 1999, 2001a) which includes (manipulatory) ostensive-inferential communication
and mind reading. The main consequence is that by no means could we ever describe the
lexical items used as DPs as encoding discourse relations (or discourse cohesion devices)
or as subcategorising such or such types of arguments. Items like but or mais may be used
alone in a situation, with no prior and no further discourse, and will remain perfectly
interpretable because of the indicational-indexical meaning they encode.
The functional spectrum of DPs is often as puzzling as their non-DP uses, especially for
models which postulate a stable relationship between function and meaning. Since
semantic continuity may nevertheless be considered to be the rule, we need to understand
how the same DPs may have such different functional interpretations and why, for
instance, the same French DC enfin can express simultaneously feelings like irritation and
relief, “forget it”, or self-correction (or reformulation). Despite the fact that any corpus
study of (most) DPs leads to classifying their different uses in various existing categories
and also in nonexisting categories, not much can be said about polyfunctionality in a
discourse-marking perspective, No other explanation than (an unexplained) functional
drift is provided for such a capacity to “mark” the most diverse discourse relations and for
the fact that the number of interpretative patterns associated with a DP appears to be
unlimited.30 Accounting for such polysemy is nevertheless possible and is indeed what the
development of IIS has dealt with (Nemo, 1998, 1999, 2001b). IIS descriptions of French
morphemes toujours, enfin, and encore in both their connective and nonconnective uses
have been proposed. All of these descriptions show that the diversity of their referential
and functional interpretations was predictable from the indications encoded by these
morphemes. Without reproducing these analyses here, and because I shall illustrate in
detail (on English but) how the same morpheme may be associated with so many different
interpretations, I shall confine myself to explaining the four uses of enfin (relief, irritation,
“forget it”, reformulation) introduced earlier.31 According to the IIS description of enfin, it
encodes the double indication that “there is/was a problem” and that “this problem has
been (is, should be) solved”. Since such an indication does not specify anything about the
moment in which the first step takes place nor about the nature of the problem, it will be
possible to associate/unify the indication with (i) a situation in which there was a problem
and it has been solved (i.e., with a constative interpretation of the two indications that
allows the expression of relief), (ii) a situation in which there is a problem and it must be
solved (i.e., with a constative interpretation of the first indication and a directive
interpretation of the second interpretation that allows the expression of reproval and
irritation), (iii) a situation in which there was a problem and the problem is performatively
declared to be over (the forget-it use), and iv) a situation in which a wrong wording takes
place and is corrected (the reformulative use). Accounting for polyfunctionality thus is
exactly like accounting for polysemy: it requires looking for an underlying semantic
equation instead of making a list of all points and describing the coordinates of each point
(Nemo, 2003a). This is precisely what IIS is about.
4. Describing DPs: a case study
Mais, but, pero, aber are probably the most studied discourse connectives (Ducrot et al.,
1976; Anscombre and Ducrot, 1977; Cadiot et al., 1979; Ducrot et al., 1980; Ducrot,
1980; Schiffrin, 1987a; Blakemore, 1989, 1992; Luscher, 1994; Fraser, 1998, Nemo,
1999; Fischer and Nemo, 2000; Schwenter, 2002), and hence a good test to illustrate the
originality and efficiency of the approach I am proposing. Since English but occurs in
various constructions, it will be chosen here to illustrate simultaneously the morpheme-
lexeme-construction distinction and the role of contextual unification in the interpretation
of the indications provided by a morpheme. But, indeed, may occur in various syntactic
positions:
(14) Broward had all but finished counting on Tuesday night, and the other two
counties both said they could finish in time for the Monday deadline.
(15) These results are all but absolute. In twenty years, only four observations of the
opposite are recorded.
(16) On Monday, [. . .] Zoran Djindjic indicated Yugoslavia’s new management was

out to rebuild ties to the West which were all but destroyed under Milosevic.
(17) This phenomenal rate cannot but have some effect on the children.
(18) She couldn’t eat anything but cucumbers.
(19) All but one of the thieves were caught.
(20) But for Peter, I would be dead.

390 F. Nemo
(21) Some of the government supporters will say things might have been worse but for
the free trade agreement.
(22) When Alcan takes all but 14 percent of the water of the Nechako, we need to
know not just what the effect of taking that water out will be in running down the
Kemano River.
(23) With the Texas governor and his deep-pockets GOP allies stretching Gore thin,
the Democratic vice president has reduced by more than two thirds his ad
campaign in the battleground state of Ohio—all but conceding those 21 electoral
votes.
(24) She had planned to make the certification of a winner on Saturday, but after the
court ruling, said she would follow the court’s decision.
(25) Vedrine—like President Clinton and dozens of other Western leaders—were

sentenced in absentia to 20 years in jail by Milosevic’s court last month for
ordering the NATO airstrikes. But Kostunica and other pro-democracy leaders
here no longer consider Milosevic’s decisions valid.
(26) Secretary of State Madeleine Albright has spoken of “accountability” and respect
for “the rule of law” but not of a Milosevic extradition.
(27) The Israeli army had announced Monday that the three soldiers were wounded in
the ambush but were probably alive.
(28) Defense Secretary William Cohen said he knew of no other specific threats
against American forces in the region, but said he ordered an increased alert level
for all U.S. forces around the world, including those in the United States.
(29) Albright said she would discuss with Barak and Arafat creating a mechanism to
allow the sides to implement agreements on a cease-fire and return to peace talks,
but gave no details.
(30) Arafat and Ivanov met in the Gaza Strip, and Arafat said afterward the two
“discussed in detail everything to save the peace process, and how to protect it in
spite of all the challenges we are facing.” But he renewed accusations that Israel
has used excessive force against rioters.
(31) It’s nearly three o’clock but the heat seems not to want to go down.
In these examples, we can first separate examples (14) to (23), in which but is not in a
connective position and is not a discourse connective, from examples (25) to (31) in
which but does connect two utterances. We can further separate the nonconnective
examples depending on the kind of construction or context in which but occurs and the
semantic interpretation it takes, namely, the without, except, almost, no other option and
only interpretations. If we were to consider semantics at a lexical level, it would be easy to
postulate semantic discontinuity between all of these uses, at least because in some cases
but indisputably contributes to the determination of the truth-conditional content of its

host utterance, whereas in other cases it does not. As stated earlier, I shall actually show
the opposite, namely that all these uses derive from a single morpheme but, the various
lexemes but being local (and memorised) interpretations of the indication it
provides/encodes.
4.1. The meaning of but
Whenever but is used, it indicates that:
Something is / has been / could have been / should be)32 stopped.
In other words, the presence of but indicates the existence of a stopping factor, which may
be successful or not either in a descriptive/constative way or in a prescriptive/
performative/self-fulfilling way. Because this indication has its own presupposition,
namely that something that was going on is (has been, could have been, should have been)
stopped, the simplest way to describe the indication is graphical. Whenever but is used
then, the hearer has to find in the context of use the two following steps:
Figure 1. Successful stopping factor

t0 t+1
As in any unification process, contextual unification does not mean that any use of but
must look like this (i.e., in utterances such as I tried to phone but the line was busy); what
it says is only that any use of but must include this pattern, so that unification may be a
success. Hence, when the stopping factor is not successful, we may have something as
illustrated in Fig. 2, in which something is initiated, encounters difficulties, but is finally
completed.
Figure 2. Unsuccessful stopping factor
ti t i+1 t i+2
( )
Before explaining how this indication can generate in each case the interpretations that
may be observed in examples (14) to (31) and many others, it must be noticed that if this
description is correct, the information encoded by but is more aspectual than
metacommunicational, as it deals with the fact that something is, or is not, completed. As
we shall see, even if the chosen formulation tries to avoid being too specific in terms of
what it may apply to, it does impose some aspectual limits to the scenario encoded by but.
Consequently, it will apply better to some examples than to others, and this situation can
be predicted.
392 F. Nemo
In order to show how contextual unification works, we must remember that according to
this description, but does not say what is stopped or what the stopping factor is but just
urges us to look for one. As a result, describing the uses of but means describing how in a
context, the content of the indication is related to contextual elements. As we shall see,
contextual unification is both a creative process and a highly routinised process. As for
this last case, and whenever a stable relationship is established between the content of the
indication and a specific contextual element, we shall from now on speak about the
existence of a semantic construction, i.e., about the existence of a fixed pattern of
interpretation of the indication. We must thus distinguish between the notions of semantic
construction and syntactic construction; various semantic constructions may be associated
with the same syntactic context. For instance, nobody would distinguish two syntactic
constructions in The weather is nice but I am tired and I am tired but the weather is nice,
but as far as unification with the indication encoded by but is concerned, a distinction
must be made, since the stopping factor precedes but in the second case and follows but in
the first one, which of course is very important for the interpretation of the utterances.
Now, let us describe some, and only some, of the most frequent semantic constructions
associated with but, before considering later on more open unification processes. In order
to do so, we shall start with the connective constructions before considering the other uses
of but (Nemo, 2002c). Most of the data used was collected from the Internet (for instance
Excite news).
4.1.1. The SFAB construction

The SFAB construction is a construction in which but occurs in a connective position and
in which the Stopping Factor occurs After But. Quite frequently, it is associated with a
successful stop, but unification only requires the existence of a stopping factor. Typical
examples of SFAB constructions are examples (24) to (25) and the following:
(32) This flat is beautiful but expensive.
(33) Bush has a 930-vote lead after machine counts of the 6 million ballots cast in the
state [. . .] and the Republican secretary of state in Florida—Katherine Harris—
had wanted to certify him as victor on Nov. 18. But the Supreme Court blocked
that plan pending its consideration of the case this week.
(34) “The secretary of state may ignore such late filed returns,” ruled Leon County
Circuit Court Judge Terry Lewis, “but may not do so arbitrarily, rather, only by
the proper exercise of discretion after consideration of all appropriate facts and
circumstances.”
(35) “We all want finality, but not at the expense of fairness,” Daschle said [. . .].
(36) This is very kind of you but I will be out of town on Saturday.
(37) The WFL is a pretty good deal. Good salary, good hours, good-working
conditions. But the NFL is everybody’s dream and few make it here (Fraser,
1998).
It must be noticed that a SFAB construction does not say anything about what is before
but. What is blocked may be p in p but q, as in (24) or (33), where it is the certification
which is blocked; or a move p supports, as in (32) and Ducrot’s description of mais; or

simply something else, as in (36), where p is the answer to an invitation that cannot be
accepted because of the SF q. The important point is that the status of p and the relation of
p and q are not vital at all for unification to be a success; this allows for a great variety of
such status and sometimes for the necessity to discuss the precise nature of the connective
status of but.
4.1.2. The SFBB construction

The SFBB construction is a construction in which but occurs in a connective position and
in which the Stopping Factor occurs Before But. In most cases, it is associated with an
unsuccessful stopping factor. Typical examples of SFBB constructions are the following:
(38) “This ugly cycle must end, it will leave scars that are not simple but it is within
our power together to overcome them.” (where SF = scars)
(39) Israeli Prime Minister Ehud Barak said Tuesday it was too soon to tell if the
violence in the West Bank and Gaza Strip was coming to an end, but he said if it
was, Israel would act accordingly. (where SF = too soon to say)
(40) It has been tough but we did it!
4.1.3. The NI construction

The NI construction is a variant of a SFBB p but q construction whose reading is the
following: p did Not Impede q (i.e., p is a SF for q to be the case). It concerns examples in
which p should/could impede q and does not. Typical examples of SFBB constructions
are example (28) and the following:
(41) He is Republican but honest.
(42) She is ten years old, but brilliant already.
(43) The top hole was for Bush, who was listed at top left; the second hole was for
Buchanan, listed at top right, and the third hole was for Gore, listed under Bush on
the left. Arrows linked the names with the proper hole, but some voters feared
they had missed the arrows and punched the wrong hole.
(44) It is yellow but I like it.
If we consider example (43), for instance, what is said is that although arrows linked the
names with the proper hole should have prevented (stopped) voters from choosing the
wrong hole, it may nevertheless not have stopped them, so that the SF was no SF. Hence,
there is both a SF before but and a claim immediately after but that this SF was actually
no SF. This construction is of course directly related to the classical denial of expectation
description: p does not impede q.
4.1.4. The BT and NBT constructions

The BT (Beyond That) and NBT (Not Beyond That) constructions are p but q connective
constructions for which the question is not to know whether a first movement was stopped
or not but to know how far something is going. Consequently, they deal with gradable
394 F. Nemo
processes or moves, and the question is to know up to where or when the move may go
on. NBT constructions are the most frequent, asserting that things did not go further than
p. Typical examples of NBT constructions are examples (26), (27), (29) and the
following:
(45) In Delaware, state House majority leader Wayne A. Smith, a Republican, said
Bush holds “higher ground and more legitimacy” than Gore because the Florida
tally so far gives the state to Bush. But, Wayne added, “neither of them deserve to
be placed on a pedestal in this matter.”
(46) Israel has said it would consider less than full control, but would not accept
Palestinian rule over the compound, former home of the biblical Jewish Temple,
the most sacred shrine of Judaism.
(47) The Royal Swedish Academy of Sciences, which also chooses the chemistry and
economics winners, invited nominations from previous recipients and experts in
the fields before whittling down its choices, but deliberations are conducted in
strict privacy.
(48) The weather was fresh, the sky was grey, a few drops but no showers.
(49) The crippled warship USS Cole was listing but still afloat in a Yemeni port as
investigators tried to find who planned its apparent bombing.
Typical examples of BT constructions are the following:
(50) I cannot see that Hezbollah was to be content with its May victory, but it will take
the fight into Israel proper.
(51) I love relaxing, walking but above all reading.
(52) Harry plays tennis but he (also) plays golf (Fraser, 1998).
The frequent uses of mais (and but) in non seulement . . . mais encore (‘not only . . . but
(also)’) constructions are closely related to BT constructions:
(53) But for the most part, in more than 100 interviews conducted by Associated Press
reporters across the country, citizens seemed exhilarated not only by the
excitement of this particular race, but by the chance to watch and discuss
American history as it unfolded with each passing moment.
4.1.5. The NY construction

The NY (Not Yet) construction is a p but q connective construction which shares with
NBT the fact that p and q are often different steps of a co-oriented process. But adds the
idea that something has not been completed yet; in other words, something has been
engaged but not accomplished:
(54) Bush spokesman Andrew Malcolm said the campaign was studying the close New
Mexico result but would not make a decision soon about requesting a recount.
(55) A voice announces the “arrival of president Bush” in twenty minutes. But twenty
minutes later, he is still not there.
4.1.6. FMWM constructions

The FMWM (First Move Wrong Move) construction is a p but q connective construction,
in which p and q are different alternatives (or in which q shows that a first move is
problematic and should be abandoned), the first move (idea, explanation, hypothesis,
plan) being abandoned because of q or something included in q. This kind of use is
different from the one we have seen so far because we are in contexts in which we have
no go and stop or stop and go, as in SFAB or SFBB, or discrete levels of completeness, as
in NBT, but rather alternatives, i.e., moves for which each move means the end of its
alternative. Graphically, we have something like Fig. 3:
Figure 3. FMWM
t0 t+1
(56) This curious flight pattern has commonly been thought to be a communal defense
system: [. . .]. But Eshel suggested a different explanation.
(57) I don’t know. It’s a very—a lot of people do think so. That at one time it’s just
gonna be one. But I really don’t know.
(58) That idea has certain plausibility. But years of observation raised so many
puzzling questions that we finally dared ask ourselves whether the calls were
indeed meant as warnings.
(59) One could possibly understand the first bark as a warning to the group, and
perhaps the second and the third as efforts to make all members of the group
aware of the danger. But what is the point of repeating the calls after the entire
group has already taken cover?
(60) He is not fifteen but sixteen.
Another illustration of a little more complex FMWM interpretation is given in (61):
(61) Then you must turn to the left but at the second traffic lights.
If we compare (61) with and without but, we can graphically express what the use of but
adds to the sentence, showing what the speaker wants to change in the initial plan of his or
her addressee:
396 F. Nemo
Figure 4. A more complex FMWM
ti ti+1
It is with such examples that the something must be stopped indication and the way it
unifies contextually may be shown to be more effective than a strictly aspectual
description of but.
4.1.7. Some NSFBB constructions
Given the little space available here, I shall now turn to a few examples that illustrate the
fact that even though there are some interpretative patterns which account for a
statistically significant part of the uses of but, interpretation is not forcefully guided by
such interpretative patterns and that unification can be purely local. Even if it is indeed
difficult to estimate what years of experience in interpreting uses of but may lead to in
terms of memorisation of semantic constructions, it is nevertheless clear that a great
number of uses unify very well contextually without fitting precisely into prior models:
interpretation is by definition an open process, the only real constraint being contextual
unification. Examples like (62) and (63) clearly show that unification is an open process
and that constructions may interfere.
(62) “I have no reason to think this was anything but a senseless act of terrorism” said
Adm. Vern Clark, the chief of naval operations.
(63) —Je vais m’en aller. (—I am going to leave)

—Mais personne ne te retient. (—But nobody is stopping you) (Ducrot et al.,
1980).
Indeed, (62) does state that there is no SF or S-alternative that could prevent the speaker
from speaking of senseless terrorism while but itself receives the nothing else than
interpretation it would have in this is but a senseless act of terrorism. As for (63),
unification goes even one step further: the but answer implies that saying I am going to
leave was a way of fishing for an answer such as please don’t!, which would have stopped
the speaker from leaving, and this answer asserts that nothing like that will come. Hence,
since the please don’t! SF has remained virtual all along the exchange, the whole
interpretation thus depends on the pragmatic possibilities of saying something in order to
force people to prevent you from doing something. This is a good way of understanding
what instructions, indications, indexicality, and contextual unification are all about: while
morphemes tell us very precisely what has to be looked for, they say nothing about what
we will find nor much about where to look.
4.2. Nonconnective uses
Now, having dealt with only a small portion of the connective uses of but, let us move
next to the nonconnective ones.
4.2.1. Nonconnective uses: the But For construction

Translated into French as sans (‘without’), but occurring in but for constructions is very
easy to account for with the indication proposed earlier. In but for X, Y constructions, X is
the factor that has blocked (or impeded) a process whose result was Y. A typical example
is (20) But for Peter, I would be dead. What (20) means, indeed, is that there was a
process that would have resulted in my death if it had not been interrupted before
completion by something or here, specifically, by someone: Peter. Other illustrations of
this interpretative pattern are found in (64) and (65):
(64) I am sure but for his larger duties he wished he could have been there on the
convention floor in the heat and excitement of that occasion.
(65) Some of the government supporters will say things might have been worse but
for the free trade agreement.
Whereas the following example illustrates the flexibility of contextual unification by

unifying SF not with X but with Y or something implicit in Y:
(66) But for his tenacity, he would never have succeeded.
4.2.2. Nonconnective uses: quantificational but

Translated into French as sauf or excepté (‘except’), this use of but is quantificational;
completeness hence is not aspectual but quantitative. With all but X Y or everyone but X Y
etc., X is what prevents one from saying that Y is the case. An overwhelming tendency
does not come to its end.
Figure 5. Quantificational but
0% 100% 0% X Y
Typical cases are examples (18), (19), and the following:
(67) The response from all but the most vociferous critics was that yes, probably any
bill was better than no bill.
4.2.3. Nonconnective uses: the only (NOO) construction

Translated into French as que or seulement (‘only’), this use of but is interesting because
it shows that the something is blocked indication encoded by but may unify not with what
is the case but with alternatives to something: there are indeed No Other Options (NOO),
and hence what is blocked are these options. It is also interesting because its only
interpretation shows that only may refer to a quantitative or qualitative limitation (both
readings being accessible with all but). A typical example is example (17).
4.2.4. Nonconnective uses: the only (‘no more than’) interpretation
Another interpretation of but may remind us of the NBT construction, namely the (non-
NOO) only interpretation. Associated with numbers, it receives a ‘not more than n’
interpretation, with various pragmatic interpretations (minimising or not) itself.
398 F. Nemo
(68) Of these processes, regular sound change is but one, even in phonology.
(69) These are but a few in-between claims in the article.
Specifically, it is often used in a post-Gricean argumentative way to say that not

everything has been said, that more could be said but will not be.
4.2.5. Nonconnective uses: The almost interpretation

Nonconnective uses of but contribute to the propositional content of the utterance in
which they are included. Saying this bird species has disappeared and saying this bird
species has but disappeared clearly communicates a different propositional content, for it
has different truth conditions. Specifically, the rather direct relationship between the idea
that something has not gone to its end and the meaning of almost, makes it easy to
understand how in but P (in which P is the predicate) constructions, but may indicate that
P is not completely the case. Typical illustrations include examples (14) to (16).
4.2.6. Nonconnective uses: conclusion

But, as we have just seen, may be used as a conjunction, as a preposition, or as an adverb.
These constructional meanings, however, show an undisputable family resemblance with
the connective uses, since they all indicate that the predicates, quantifiers, or quantity it
associates with either are not fully completed (or true) or represent only a part of a larger
set that could be spelled out completely but is not. Hence, the indication that something is
stopped before completion is direct in such uses and advocates for the existence of a
single morpheme whose indexical meaing is completely independent of the syntactical
positions it may take.
Such an observation falsifies thus both the idea that but could have a procedural meaning
in some of its uses and a conceptual meaning in the other and the idea that its meaning
could be predicted from its syntactic/functional status of (contrastive). Far from marking a
specific kind of relationship between two connected segments, but, as toujours or encore
(Nemo, 1998; 1999), encodes the same indicational-indexical information in all of its
nonconnective uses.
4.3. Describing but: conclusion
The approach which has been presented here has provided us with a description of but
which is radically different from the classical descriptions mentioned earlier. This is
basically because it is not a constructional approach nor an interpretative approach, but
above all because it can claim to fulfil all the missions stated in section 1.5. It explains the
distribution of but and all its local interpretations without postulating a sense-enumerative
lexicon, without postulating homonymic degrouping, and without relying on intuition to
decide that such or such specific (and local) interpretation(s) should be considered as
semantically more central than any others.
What I have mainly tried to demonstrate here is that DPs may be described with the same
tools as any other kind of linguistic sign. Such a demonstration does not, however, imply,
for instance, that studying discourse relations, discourse coherence, conversational
routines, conversation structure, etc. would be in any sense of no interest. What it shows
is only that we cannot expect such studies to account for the diversity of uses, the
diversity of functions of the semantic items used as DPs, nor for their syntactic scope or
polycategorial character. It also shows that we must refrain from the idea that since an
item plays an important role in the process we are studying, it must be designed to play
such a role. This claim is as old as the study of DPs (Ducrot, 1972) itself and it has been
repeated by most authors, including myself (Nemo, 1999), but it overlooks the distinction
between encoded meaning and inherited meaning. As for today, I had to limit myself to
the illustration of what is possibly the most controversial aspect of what I am proposing,
namely that the puzzling distribution of semantic items is compatible with the idea that
they convey a very stable meaning and that the puzzling diversity of pragmatic variation
in the uses of these items must ultimately be grounded in the attentional dimension of
language use. But such a practical limitation is not a denial of the existence of
constructions that deserve to be studied in considerable detail,33 only a reminder that no
constructions will ever fix or limit what can be done with the semantic material they are
using. If we refuse the idea that language’s core mechanism is the combination of
lexemes, then a large part of the discrepancies between what we know about DPs from
observation and what we are able to predict will disappear.
Notes
1
So that what is to be described (Gasiglia, Nemo, and Cadiot, 2001) is a function
f(m,cstr,ctxt) = i, in which m stands for the morpheme used (which encodes semantic
indications), cstr for the information inherited from the construction, and ctxt for
contextual specifications. The result of the interpretative process is i, an interpretation
which is memorised whenever the use becomes somehow conventional.
2
The scalar constraint of relevance is a constraint which says that if p is said then it must
make a difference whether p is the case or not (Nemo, 1988, 1992, 1999).
3
One of the most confusing legacies of Grice’s work on conversational maxims is the
scope of such maxims. Maxims like the maxim of quantity use the undefined term
contribution, while Grice himself always gives examples of contributions reduced to a
single utterance. Consequently, no distinction is ever made between mechanisms
whose scope is an individual utterance (Nemo, 1986, 1988; Nemo, 1999: 393-412) and
mechanisms whose scope is a contribution (see also Clark and Schaefer, 1989).
4
Extracontributional uses are frequently observed with DPs/DCs such as French
d’ailleurs, whenever the comment is not addressed to the actual addressee but, for
instance, to oneself, thus marking a change of focus of the speaker himself and not
promoting a change of the addressee, as in the Ducrotian argumentative uses.
5
Which is built on the cognitive bricks described by cognitive ethologists,
primatologists, and psycholinguists (Tomasello, 1995): mutual attention, deictic
attention, joined attention, and controlled attention. This game may be deeply
400 F. Nemo
manipulative and it is the recognition of this Machiavellian dimension that separates

ALT and RT.
6
This issue is easily overlooked as long as the utterances are considered one by one but
plays a central role both in discourse and conversation or dialogue, as it deals with the
questions “Why did (s)he mention that?” or “Why did (s)he mention only that?” or
“Why didn’t (s)he mention anything (about that)?”, etc. .DPs/DCs, are routinely
attentional shifters.
7
Pre-Gricean definitions of the notion of relevance, notably Furberg (1963), were already
focussing on the fact that attracting somebody’s attention to something presupposes
that the thing was of some interest to him/her (Ducrot, 1972, 9), and that not saying
something that is fairly relevant would be considered non-cooperative (and possibly
manipulative). The main difference between ALT and RT is that the former assumes
that attracting attention is always manipulative whereas the latter assumes that it is
basically a cooperative phenomenon. Social uses of controlled attention include both
aspects.
8
The difference between utterances and contributions here is obvious, since there are two
utterances in each case and only one contribution in each case. In that sense, DCs may
be described as attentional binders.
9
This claim is what has distinguishes ALT (or ALT-friendly models) from other
frameworks. It seems however to be widely acknowledged—from Levinson (2000) to
Krifka (1999, 2000) and from Modal Frames Semantics (Nemo, 1988, 1992, 1996) to
Alternative Semantics—that describing the interpretation of utterances supposes
describing the specific alternative sets that words introduce.
10
This statement is coherent with the idea that upstream context is chosen and not given
(Sperber and Wilson, 1986: 120), but assumes the semantic content of linguistic items
(and not the principle of relevance) to drive the building of this chosen context.
11
What is shared by all the alternatives within the comparison set (and which is thus
shared by (p), (not p) and (p?)) is what is usually called the presupposition of an
utterance.
12
No association of a DP with a discourse relation has ever been established (Sanders,
personal communication).
13
This is the case, for instance, in an explanatory relation (Peter didn’t call but he had
been extremely sick that day) and in fact in relation with each and every pragmatic
constraint we know of, for instance, the Gricean maxim of quantity, as in the answer
But when? to the initial utterance He came yesterday.
14
The idea that describing the meaning of but and describing the (prototypical) meaning
of the construction p but q, as defended by Fraser (1998), is thus abandoned for its
various self-formulated shortcomings, notably the fact that the initial claim (“the core
meaning of but is to signal simple contrast, nothing more, and the speaker will select it
when intending to highlight a contrast”) leads to further recognition of fuzziness (“I
can offer no precise definition of what qualifies as a Contrastive Discourse Marker”),
intuitiveness (“I call for your intuition that each of the [listed] CDMs signals a
contrastive relationship between the [segment] S2 they introduce and a foregoing S1”),
unfalsifiability (“Even if one cannot find two specific areas of contrast between the
direct S2 and S1 messages, the messages may nevertheless be contrasted in one of
several ways”), and linguistic indeterminacy (“I am not treating other uses of but such
as found in All but one left today, There was no doubt but that he won, it has not
sooner started but it stopped, He was but a poor man, I may be wrong but I think you
are beautiful. Whether or not they could be included under my analysis is left open.”).
15
Since we have intuitions about senses (lexemes) and not about encoded meaning
(morphemes).
16
In categorial terms, the DMH describes discourse connectives as a D/(U1,U2)
category, requiring two utterances (or discourse units) and forming a discourse D,
whereas the MH, coherent with the categorial treatment of any class of modifiers (such
as adverbs, adjectives) describes them as a C/C category, whose output is of the same
categorial nature (contribution) as its input. Within a MH perspective, a distinction
must be drawn between U/U and C/C modifiers, i.e., between utterance modifiers
(Nølke, 1993), which integrate viewpoints at the level of utterances, and contribution
modifiers, which integrate viewpoints at a the level of contributions.
17
Within such a view, one has to study discourse relations such as reformulation (Rossari,
1994), deduction, elaboration, or contrast (Fraser, 1998), and, contrastively, all the
lexical items which can be used to mark such relations.
18
Not doing so would lead to the recognition of the fact that DPs are actually not devices
guiding the interpretation but signs whose interpretation is itself very sensitive and
dependent of the context.
19
It has become conventional in DP studies with a D-methodology to use the term
morpheme to refer to a unit regardless of the specific positions in which it may be
observed, so that the morpheme still may be distinguished from both the adverb still
and the DP still. It follows from this convention that morphemes are purely semantic
units (free of any combinatory information) and consequently not the result of
morphological decomposition (they are not combinatory units) but the result of
distributional analysis. In such a view, ever and every or able and -able may be said to
be different uses of a single morpheme if and only if it can be proven that they share
(but are not limited to) a single piece of information encoded by the morpheme.
20
What Ducrot (1980) has shown in great detail is that even when the argumentative
description does not seem to work, all the elements included in that description may
indeed be found in the apparent counterexamples, so that without being directly
adapted to all contexts of use, it still captures something which is to be found in all
cases.
21
It cannot thus encode the classical denial of expectation meaning, according to which it
should follow from P that not Q, nor can it encode the idea that P and Q are opposed.
22
Within such a definition, constructions are not defined as primarily syntactic,
morphological, semantic, or pragmatic. The succession of turns, for instance, is no less
syntactic than pragmatic; succession is a matter or order of elements (here, turns), and
if a stable relationship may be identified in terms of succession of turns independently
of the specific utterances performed, then we are dealing with constructions.
23
Such a combined approach, however, leaves unsolved the question of the competition
between the various elements which may be used in the same construction. The idea
that distinctiveness within a FF would be central to the semantics and syntax of DPs
402 F. Nemo
(as in Rossari, 1994; Rossari and Jayez, 1996) is in a way incompatible with the idea
that meaning is either encoded or inherited.
24
Even if we can indefinitely measure each use of a DP in terms of any relevant
dimension of speech and discourse what should actually be the goal of any account of
DPs is the diversity of their uses itself and the instability of their local interpretations.
25
In some cases, the theses A to E are in fact just scenarios and are not associated with
testable theories of bleaching, language games, or polyprocedurality.
26
In both cases, they may be used either ex ante (or online), to build the contribution or
the utterance’s frame, ex post to modify a contribution or an utterance, or together to
introduce a contributional shift.
27
Metalinguistic covers all aspects of verbal communication, from turns to addressee
status, from scalar (i.e., argumentative or inferential) value to evidentiality.
28
Such devices include explicit sentences or phrases that never lexicalize, as well as all
kinds of lexicalized items, ranging from affixes to multilexical “frozen” expressions,
and even sometimes real constructions (e.g., the French encore + verb + il connective
pattern). In a language like French, for instance, the average DC is a rather heavy (and
not so mobile) multilexical expression, not easy to label particle. The morphological
diversity of DPs within a language and crosslinguistically must thus be considered
before making any general claim about their nature and behavior.
29
The existence of modifiers implies the existence of something to modify, so that both
SPs’ and DCs’ scope is an utterance or contribution. Explaining the nonmetalinguistic
(narrative, interjective, indexical, etc.) uses of the lexical items used as DCs
(Rouchota, 1996), is another issue, whose explanation will be provided in section 2.4.
30
From a statistical perspective, it seems that most of the time, a limited number of
patterns account for most of the data, which are thus described as the senses of the DP,
but that nevertheless the number of possible patterns itself is not limited, some of them
being associated with single uses.
31
For a more complete IIS account of the polyfunctionality of the morphemes toujours,
encore, et enfin, see Nemo (1999).
32
The fact that the indication may receive a directive reading and various temporal
unification, is not specific of but. In some uses of mais, which could be named Gricean
mais, such as:
—Devine ce que je mange —Des bonbons —Gagné! Mais à quoi?
—Guess what I am eating —Candies —Right!! But which flavour?
it is the issue of not telling which flavours, and hence the maxim of quantity, which is
at stake in the use of but, whose interpretation thus is close to “your answer is correct
but not complete, don’t stop, you must make precise the flavour”.
33
As mentioned earlier, I am using the notion of construction as defined by Goldberg
(1995), i.e., as form-meaning pairs which exist independently of the specific semantic
material which is being used. This allows any aspect of meaning (interactive functions,
turns, ratification, etc.) to be constructionally inherited.
Chapter 21
Discourse particles and modal particles as

grammatical elements
Gabriele Diewald
1. Introduction
1.1. Approach, methodology, and data
Present-day German (PDG) is notorious for its extensive use of particles of all kinds.
Most of these particles have “homonyms” in other word classes, such as conjunctions,
adverbs, and adjectives, and, moreover, even in their particle use, they are highly
polysemous and have functions on different layers of the speech event, which can be
arranged on a scale reaching from more lexical to more grammatical and/or pragmatical.
This fact, beyond its synchronic relevance, also reflects the diachronic development of the
linguistic elements in question: most particles of PDG are diachronically derived from
conjunctions, scalar particles, and focus particles, which themselves can very often be
traced to adverbs, adjectives, or syntagmatic constructions (to name only a few sources of
particles).
The particles of German are therefore ideally suited to explore the questions this volume
focuses on, namely:
1. the question of word class membership, which in the case of particles coincides with
the homonymy problem;
2. the attribution of functional domains to different particle uses;
3. the question of inherent semantic content, polysemy, and context dependence.
The approach taken here is based on grammaticalization theory (Lehmann, 1995 [1982];
Hopper and Traugott, 1993; Bybee, Perkins, and Pagliuca, 1994; Diewald, 1997),
combined with conversation analytic concepts (following Fischer, 2000a and Diewald and
Fischer, 1998), and an adaptation of Wierzbicka’s framework for the description of
pragmatic meaning (Wierzbicka, 1986, 1991, 1996).
404 G. Diewald
The subject is treated in the form of a qualitative study with data from PDG and its
diachronic stages. The diachronic data are taken from historical dictionaries and
grammars; the data for PDG are taken from various corpora on spoken and written
language, from dictionaries and from grammars. For explanatory purposes, made-up
examples are used as well.
1.2. State of the art and problems
There have been numerous studies on the diachronic development of discourse

markers/particles from lexical items in various languages (Traugott, 1995a, 1999b; Tabor
and Traugott, 1998; Aijmer, 1997; Vasko and Fretheim, 1997; Barth and Couper-Kuhlen,
2002; Laitinen, 2002; Lima, 2002), as well as on the development of modal and discourse
particles in German (Diewald, 1997, 1999a; Gohl and Günther, 1999; Günthner, 1999;
Günther, 2000; Autenrieth, 2000; Wegener, 2002; Molnar, 2002).
These studies have established various grammaticalization paths for particles and have
provided new insights into the cognitive and pragmatic mechanisms involved in this type
of linguistic change. Furthermore, because uses and functions dating from different
historical ages are usually retained side by side in one synchronic stage, these studies have
shed light on the possible range of coexisting synchronic functions of a given linguistic
element or construction.
However, at least two problems have not yet been answered satisfactorily. The first
problem is the definition and delineation of the functional domains of particles and
discourse markers. Although it is generally agreed that these items do not operate in the
referential (or truth-functional) domain, i.e., they do not display lexical semantics in the
narrow sense and therefore cannot be used to denote elements of the propositional content
of the sentence, no conclusive answer has yet been found to the question of whether there
is a common and constitutive function that distinguishes the use of an item as discourse
marker/particle from all its other uses.
Many grammaticalization studies on discourse markers/particles have more or less

explicitly treated this problem by asking whether the development of those particles from
other elements should be subsumed under the heading of grammaticalization, or whether
it should be treated as a separate process, which is usually dubbed pragmaticalization or
subjectification. In her work on the development of discourse functions of the English
expression I think, Aijmer (1997) suggests that a sharp line ought to be drawn between
grammaticalization on the one hand and pragmaticalization on the other. In Aijmer’s
view, the former process (i.e., grammaticalization) “is concerned with the derivation of
grammatical forms and constructions (mood, aspect, tense, etc.) from words and lexical
structure”, whereas pragmaticalized items, i.e., items having undergone a process of
pragmaticalization, involve a “speaker's attitude to the hearer” (Aijmer, 1997: 2). It is,
however, difficult to conceive how a separate “cline of pragmaticalization” (Aijmer, 1997:
6 ff.) should be established in parallel to grammaticalization scales, since the descriptions
which Aijmer gives of both supposedly distinct processes are not mutually exclusive and
thus may equally apply to one item. Günthner (1999) and Gohl and Günther (1999), who
investigate the rise of discourse functions from several conjunctional uses of the German
subjunctions obwohl and weil, likewise consider this problem. Günthner (1999: 437)
considers it plausible to treat the fact that the functional domain in the development of
Discourse particles and modal particles as grammatical elements 405
obwohl shifts from purely grammatical functions to conversational functions as an

argument in favor of a distinct process of pragmaticalization, but goes on to point out that
the development of discourse particles is indiscernible in many formal and semantic
aspects from proper grammaticalization processes, defined in terms of Lehmann's
grammaticalization parameters (Lehmann, 1985), so that the distinction between
pragmaticalization and grammaticalization becomes minimal.
In a similar line of reasoning, Barth and Couper-Kuhlen (2002), who treat the
development of discourse functions in final t h o u g h in English, suggest that
pragmaticalization ought to be subsumed as a specific subtype under the broad heading of
grammaticalization, which deviates in some aspects from prototypical cases of
grammaticalization but is too similar to it to be treated as “a separate, independently
definable process” (Barth and Couper-Kuhlen, 2002: 357; see also Lima, 2002).
The solution suggested here begins by turning things around and proposes that the
pragmatic functions of discourse markers/particles are genuine grammatical functions
which are indispensable for the organization and structuring of spoken dialogic discourse.
Thus, the question of whether their diachronic development should best be called
pragmaticalization or grammaticalization can be abandoned in favor of more fruitful
questions such as:
• What types of grammatical functions do these elements fulfill?

• How can they be distinguished from one another and from other grammatical functions
(e.g., the functions of conjunctions)?
• To what degree are their respective functions based on the inherent semantic features
of each lexical item?
The second problem, which has not yet been convincingly solved, is closely related to the
first one and concerns the polyfunctionality of particles and the impact of contextual
factors on their respective interpretations. Must we assume that from a purely synchronic
perspective, the different uses to which an item is put are unrelated homonyms belonging
to different word classes and having independent meanings? Or should we postulate one
core meaning per linguistic item with several contextually triggered interpretations? Or
should we treat the items in question as strongly polysemous and set ourselves to the task
of delineating each meaning and its connections to neighboring meanings?
The solution suggested here again derives from observations of grammaticalization

processes and semantic change that are summarized in section 3. It postulates a core
meaning inherent in a lexical item and found in all its uses. The synchronic
polyfunctionality of the particle lexemes is due to the reinterpretation of the basic semantic
template, which—depending on frequency, stereotypicality, distance of semantic domains,
and functional specifications—can result either in a distinction of word classes (i.e., different
‘heterosemes’), or in polysemy within a particular word class, or in contextually triggered
pragmatic interpretations. Of course, because diachronic change continually leads to a
semanticization of stereotypical context-induced interpretations (Traugott and König, 1991),
the decision whether to treat a particular reading as part of the semantic content, i.e., as one
meaning of a polysemous item, or as a pragmatic function legitimated by, but not part of its
lexical meaning, can only be made individually for each item and each diachronic period,
because it is certain to change.
406 G. Diewald
To account for the interaction of the semantic template and contextual factors, a tentative
model is developed that spells out the respective contextual impact of each use with
reference to Wierzbicka’s work on the description of pragmatic meaning.
2. Definition
This section offers an initial definition of the elements treated in the following section.
The justification of the distinctions drawn here will became clear in the course of the
paper.
Among non-inflecting linguistic items, membership in a specific word class is primarily

defined via functional criteria, with concomitant morphosyntactic features providing
additional criteria. Accordingly, and following suggestions in Diewald and Fischer
(1998), discourse particles (DPs), as well as modal particles (MPs) and conjunctions, are
here considered to be functionally constituted classes of grammatical markers which
operate outside the syntactic confines of the clause, relating two items through an
indexical procedure, i.e., through a process of linguistic pointing (see section 4 for an
explanation). The three classes differ in the functional domains they apply to and in the
type of elements they link.
DPs relate items of discourse to other items of discourse, i.e., they operate as indexical
elements in those domains that are fundamental for spoken dialogic communication. This
includes the organization of the turn-taking system, the thematic structure, speech
management, etc. From this, it follows that the term discourse particle is construed here
in its broad sense, which encompasses response signals, segmentation signals,
interjections, hesitation markers, etc. Thus, ja, obwohl, and aber are discourse markers in
the following examples:
(1) und dann kommt der Querflügel, ja?

and then, the crossbeam comes next, JA?
(Sagerer et al., 1994; quoted from Diewald and Fischer, 1998)
(2) Glaubst du, daß er das Spiel gewinnen wird? Obwohl – mir kann's ja egal sein.
Do you think he’ll win the game? OBWOHL – I don't care.
(Zifonun, Hoffmann and Strecker, 1997: 2316)
(3) K: und das wird dann da so seitlich draufgeschraubt oder?

and that’s going to be screwed there to the side this way, isn’t it?
I: ja genau, aber mach das erstmal so.

yes exactly, ABER do it this way first.
Turn-final ja in (1) asks for agreement and initiates the transition of the turn from the
present speaker to the hearer. It belongs to the class of turn-taking signals. Obwohl in (2)
serves as a correction signal by which the speaker tries to withdraw the illocutionary force
of the previous utterance and introduce a justification for this withdrawal. Aber in (3)
operates on the thematic plane of discourse. After the first interlocutor (K) has asked for the
next step in constructing a toy airplane, the second interlocutor (I) uses aber to relate his
utterance to the preceding utterance of the partner and simultaneously to indicate that he or
she wants to change the topic (Diewald and Fischer, 1998: 87).
As can be derived from the description above, DPs have no referential content in the strict
sense, i.e., they do not denote anything that is part of the proposition, the “spoken of”. The
entities they denote are features of the discourse and the communicative situation (which,
of course, can be made the topic of a conversation, i.e., the “spoken of”. This, however,
cannot be achieved by discourse particles).
Similar to DPs, MPs are indexical devices of spoken discourse. MPs differ from DPs in
that they apply to propositions and speech-act alternatives, which is to say that they have
propositional or speech-act scope, while DPs have scope over non-propositional discourse
elements of various sizes.
The function an MP fulfills is indexical insofar as it points backwards from the linguistic unit
in which the MP appears and relates the utterance to a proposition or speech-act alternative
which the speaker regards as relevant and given. In referring ‘back’ to something that is
treated as communicatively given, albeit unexpressed, the MP marks the utterance which
contains it as non-initial (see section 4). Ja and aber are used as MPs in the following
examples:
(4) Das ist aber keine gute Idee.

That is ABER not a good idea.
(5) Es soll ja auch schwimmen.

It is JA meant to float.
In addition to their class constitutive indexical meaning, each modal particle has a
diachronically motivated, lexeme-specific semantic feature. The specific semantic content
of aber is ‘adversative’, that of ja is ‘affirmative’.
While MPs point to linguistically unexpressed propositional or illocutionary entities, con-

junctions, on the other hand, typically relate two propositions, both of which are
linguistically expressed, i.e., they connect textual elements. Obwohl and aber are used as
conjunctions in the following examples.
(6) Obwohl es schon spät war, machte sie sich zu einem Spaziergang auf.
Although it was already late, she set out on a walk.
(7) Sie wollte telefonieren, aber sie hatte kein Kleingeld.

She wanted to make a call, but she didn't have any change.
The functional criteria discussed thus far are essential for the definition of the respective
word classes, i.e., for DPs, MPs, and conjunctions. In addition to them, there are formal
and structural features that correlate with these distinctions.
The most prominent structural feature is the degree of syntactic integration of a particle
lexeme in a specific use. Elements functioning as DPs are not integrated into the syntactic
408 G. Diewald
structure of the sentence. Many of them can be attached to different host structures and
appear in various positions in relation to their host (Diewald and Fischer, 1998). On the
other hand, both MPs and conjunctional elements are syntactically integrated into the
sentence: MPs are confined to the middle field of the sentence, i.e., they appear after the finite
verb and before the right “sentence bracket” in a verb-second clause in German; conjunctions
are restricted to a left peripheral position, i.e., they appear to the left of the first constituent of
the sentence.
There is a further formal, or more precisely, morphological feature of DPs/MPs in

German. Prototypically, these items are non-inflected monosyllabic linguistic units that
have segmental status and can be isolated as such (as opposed to clitics or inflections). For
this reason, the term particle is preferred here to the term marker.
While the functional criteria mentioned above are believed to be cross-linguistically valid
for DPs/MPs, the syntactic and morphological features may well be language specific and
thus apply only to German and related languages.
To sum up, DPs/MPs (as well as conjunctions) are defined here as word classes that are
constituted by the clustering of specific formal, structural, and functional features,
whereby the functional parameter is essential, while the others are concomitant. In a
simplified manner, the realization of features for each class can be given as follows:
Discourse particles:
function and domain: relate non-propositional discourse elements which
are not textually expressed
syntactic integration: syntactically non-integrated, i.e., no syntactically
fixed position, no constituent value
morphology: particle
Modal particles:
function and domain: relate propositions and speech-act alternatives,
one of which is not textually expressed but treated
as given
syntactic integration: syntactically integrated, appear only in the middle
field of the sentence, no constituent value
morphology: particle
Conjunctions:
function and domain: relate propositional elements, both of which are
textually expressed
syntactic integration: syntactically integrated, i.e., fixed position at the
left periphery, no constituent value
morphology: particle (complex conjunctions are used as if they
were particles)
Because large portions of the following deal with the modal particles of German, while
the focus of this project is on DPs, a brief justification may be in place here. The German
modal particles form a prominent and clear-cut functional class which has undergone an
extensive process of grammaticalization in the history of the German language and
thereby acquired several new members.1 The reason to concentrate on MPs in this paper is
the hypothesis that MPs, which are an important grammatical device of contemporary
spoken discourse, cover an intermediate domain between the functions of text-connecting
elements such as conjunctions and conjunctional adverbs on the one hand and discourse-
structuring elements such as turn signals, hesitation markers, etc. on the other. That is to
say, modal particles are treated here as the link between strictly textual functions and
strictly discourse-relational functions. Taking into account that languages like English,
which have been the object of extensive research concerning their discourse-marking
devices, do not have a functional class comparable to MPs in German, the latter might
even be called the missing link to deepen our understanding of the interrelations between
text-connecting and discourse-marking elements.
3. The functional spectrum and its diachronic motivation
As pointed out, most particle items have functionally-defined “heterosemes” in other

word classes. This has been illustrated by aber, which can be used as discourse particle,
modal particle, and conjunction, and thus neatly illustrates the polyfunctionality of
particle items in PDG (see examples (3), (4), and (7) in the last section). 2 Furthermore,
even within one word class, an item may have several distinct uses (see the discourse
functions of the DP ja discussed in Diewald and Fischer, 1998), which is the point where
the problem of polysemy and context dependence comes in.
From the perspective of grammaticalization theory, multiple membership in several word

classes, polysemy, and extensive context-sensitivity are by no means exceptional
phenomena or disturbances of the normal functional and structural makeup of a language.
Instead, they are regarded as the natural outcome of constant and pervasive language
change, of which grammaticalization is a subtype that among other things, is
characterized by unidirectional developmental clines from more lexical to more
grammatical functions. Given the fact that in the course of the development of new, more
grammatical functions, the older, more lexical functions may persist, so that on a
synchronic level several distinct functions of a linguistic item coexist side by side, a look
at the diachronic development of an item may well help to clarify the problem of
discriminating among different functions on a synchronic level.
This section applies these general reflections to the particles, whereby the development of
aber serves as an illustration of the principle (Diewald, 1999a). Because linguistic change
that amounts to grammaticalization is composed of several changes on different linguistic
levels (e.g., morphonological, semantic, syntactic, and functional change), for the
investigation of a specific grammaticalization process, it is necessary to single out each
layer and describe it according to the general tendencies which have thus far been
discovered to be relevant for the grammaticalization path in question. For the purpose of
this paper, it may suffice to restrict our attention to the following three developmental
tendencies:
• the metaphorization scale formulated by Heine, Claudi, and Hünnmeyer (1991)

• the tendencies of scheme retention elaborated by Sweetser (1988, 1990)
• and the model of successive semantic-functional stages suggested by Traugott (1989)
and Traugott and Dasher (2002).
410 G. Diewald
As Heine, Claudi, and Hünnemeyer (1991) have shown, a large percentage of the
semantic changes accompanying grammaticalization processes can be described in terms
of successive metaphorization procedures which, in accordance with general cognitive
principles, unidirectionally evolve from local to temporal to more abstract meanings. The
following is a simplified metaphorization scale encompassing the basic semantic-
cognitive domains:
local > temporal > abstract (e.g., causal, adversative, copulative)
The semantic development of aber aptly confirms this scale. Its origin lies in the
comparative form of an adverb meaning ‘farther way’, which, in addition to its use as a
local preposition meaning ‘behind’ in Gothic (afar), also developed the temporal meaning
‘after that’, ‘later’ (DWB Neubearb., vol. 1, col. 175ff.; Paul/Henne, 1992).
In Old High German (OHG), the adverb afur/abur had the temporal meaning of ‘after
that’, ‘later’, and later developed the more abstract, iterative meaning of ‘again’, ‘once
more’. In the following example from the ninth century, the adverbial abur is ambiguous
between the temporal and the iterative reading and also supports a reading that merges the
temporal and iterative meaning to ‘later again’:
(8) giuuelîh dê dar trinkit fon uuazzare thesemo, thurstit inan abur.
Latin: omnis qui bibet ex aqua hac sitiet iterum.
Whoever drinks from this water will be thirsty again later.
(Tatian 87,4)
The iterative meanings of aber, which developed an attenuated copulative meaning,

persisted up to the eighteenth century. Today, it is found in fossilized and idiomatic uses
such as aber und abermals, tausend und abertausend.
The next step, which already started in OHG, was the rise of an adversative meaning
(‘however’, ‘yet’). The adverb abur in (9) exemplifies this:
(9) Ther geist giuueso funs ist, thaz fleisc ist abur ummahtîc.
Lat.: Spiritus quidem promptus est, caro autem infirma.
The spirit indeed is ready, the flesh, however, is weak.
(Tatian 181,6)
This adversative meaning is the dominant meaning of aber in present-day German. It is

present in all its uses, but due to functional differentiation (and contextual influences), it
appears attenuated or modified towards only mildly adversative or even copulative
meanings in many instances.
Thus, even in this very abbreviated diachronic sketch, it can be seen that the semantic
development of aber is a specification of the metaphorization scale given above. It may be
summarized as follows:
local > temporal > abstract

‘farther away/behind’ ‘later/again’ ‘adversative/juxtaposition/copulative’
Sweetser (1988) and (1990), who also works with metaphorization chains, claims that in
grammaticalization processes, we can make out a basic semantic template (Sweetser calls
it image scheme) that is retained in its fundamental relational structure through all
developmental stages. In the course of semantic change, this relational template is
successively transferred to different semantic domains, which are characterized on the one
hand by increasing abstraction, but on the other hand also provide new semantic features
that had not been present before.
Though the details of the semantic development of aber cannot be traced here, it should
be noted at least in passing that Sweetser’s postulate is relevant as well. We observe the
retention of a relational template consisting of a vector between a source, a path, and a
goal that originally applied to the concrete local domain and was later transferred to
successively more abstract domains. The basic relational structure present in the source
lexeme (i.e., the local adverb meaning ‘farther away’, which must be ascribed to the
compositional effect of an adverb plus comparative morphology), pointed from one local
element to another. It was retained in later diachronic stages, where it came to express
temporal, iterative, copulative, and adversative relations, whereby the entities being
related came to belong to increasingly abstract domains.
For the functional development that co-occurs with the two processes described so far,
Traugott (1989: 34-35) postulates three diachronic tendencies, which she describes as
follows:
Tendency I:
Meanings based in the external described situation > meanings based in the
internal (evaluative/perceptual/cognitive) described situation.
Tendency II:
Meanings based in the external or internal described situation > meanings based in
the textual and metalinguistic situation.
Tendency III:
Meanings tend to become increasingly based in the speaker’s subjective belief
state/attitude toward the proposition.
Beyond the change in the semantic domains involved (which neatly fits into the
metaphorization scale), these tendencies focus on a functional change concerning the
semiotic status of the item in question. It develops from mainly referential to mainly
textual and connective functions, and further evolves to indexical grammatical functions.
In subsequent studies (Traugott, 1995a, 1999b; Traugott and Dasher, 2002) this model
was expanded, refined, and generalized in order to account for as many instances of
semantic change as possible. The modifications concern first the notion of
subjectification, which is complemented by intersubjectification in order to account for
social deixis (Traugott and Dasher, 2002: 23-24), and second the integration of possibly
concomitant structural changes, especially the change of scope and topological features in
the development of discourse markers (Traugott, 1999b: 177ff.; Traugott, 1995a; see
below). The chronological ordering of the three stages as well as the description of the
respective semantic and cognitive domains has been retained in its essence. With some
412 G. Diewald
minor shifts of focus and perspective, Traugott’s findings on general tendencies of

semantic change accompanying grammaticalization are compatible with the approach
taken here and may be adapted and summarized as follows:
referential function > text-connective function > indexical-grammatical function3
Describing the development of English discourse particles such as indeed, in fact, and
besides, Traugott (1995a) and (1999b) shows that in the case of particles, this functional
shift is accompanied by changes in scope, topological position, and word class. On the
first stage, the referential source lexemes/source structures have adverbial functions as
verb phrase adverbials. On the second stage, they acquire the syntactic status of sentence
adverbials and fulfill text-connective functions, and finally, as discourse markers with
speaker-based indexical functions, they extend to utterance scope.
The functional development of German aber is parallel to that of the English discourse
markers described by Traugott. From its origin as an adverb denoting local, temporal, and
iterative relations, where it had the status of a verb phrase adverbial (see example (8)), it
developed conjunctional uses in Late OHG. Although there was no clear-cut distinction
between the functional class of sentence adverbials and the functional class of
conjunctions as long as word order had not been fixed the way it is now in German (and
as can be seen in example (9)), we can safely assume that abir in examples like the
following is a conjunction:
(10) der eine [ging] in sîn dorf, abir der andere zuo sîme gewerbe.
One of them [went] to his village, ABIR the other to his trade.
(Evangelienb. Behaim 51 B. (1343), DWB Neubearb. 1, 181)
As this example shows, even the conjunctional use need not be strictly adversative, but
can express mere juxtaposition or even a copulative relation between the two
propositions. As a conjunctional element, abir clearly has text-connective functions,
expressing an adversative/ copulative relation between two propositions that are textually
present. The rise of this function corresponds to tendency II in Traugott’s model.
Aber reaches the third stage of Traugott’s model as soon as it is used as a modal particle.
This new function is first documented in the eighteenth century.4 It is illustrated by the
following example:
(11) Ihr müßt aber hier jämmerliche Langeweile haben.

You must ABER be terribly bored here.
(GoeWb, quoted after Paul/Henne, 1992)
Here, aber shows neither referential nor text-connective/conjunctional functions. Instead,

it relates the utterance to its pragmatic pretext (see section 4), marking it as non-initial and
expressing a slightly adversative meaning. It has reached the stage of the PDG modal
particles as illustrated in example (4).
The stages leading from a referential adverb to the modal particle, via a text-connecting
conjunction, confirm the tendencies postulated by Traugott. Thus, taking into account the
semantic, functional, and syntactic (i.e., word-class) features, the development of aber can
be summarized as follows:
functional level:
referential function > text-connective function > discourse function
word class:
adverb (with comparative morphology) > conjunction/conjunctional adverb >
modal particle/discourse particle
semantics:
local > temporal/iterative > adversative/juxtaposition
The observation that the diachronic development of German modal particles such as aber
is analogous to the development of English discourse markers such as indeed supports the
claim that German modal particles and English discourse markers, though syntactically
and categorially distinct, do indeed have parallel functions that place them together in a
broader group of discourse markers. Therefore, it seems justified to claim that the MPs in
German are a syntactically integrated, specific subtype of discourse-marking devices.
However, there is a remarkable difference between the developmental clines of discourse

elements in English and German: while English only developed discourse particles in the
strict sense, German has at least two distinct grammaticalization paths for discourse
elements and developed two clearly distinct classes, modal particles and discourse
particles.
In this development, one source item may give rise to both particle types simultaneously,
which can, again, be exemplified by aber. As shown in (3), German aber, beyond its
function as a modal particle, is also used as a prototypical discourse marker in the
narrower sense. Because of topological and syntactic restrictions, this use cannot have
evolved from the use as an MP as in (11), but must have arisen diachronically from the
use of aber as a conjunction. To the best of my knowledge, this last development has not
yet been investigated for aber.
There are, however, studies on the rise of other DPs from conjunctional elements in PDG
which can support this claim. Günthner (2000) and Gohl and Günthner (1999) present data on
the gradual evolution of discourse functions from the conjunctional use of weil. As a
subordinating conjunction, weil indicates a causal relation between two propositions. In its
discourse functional use, weil is a speaker signal indicating that the speaker wants to keep the
floor, i.e., it is used as a discourse particle in the interpersonal or interactive domain of
discourse organization. Between these functionally and categorially distinct uses, the authors
discern several intermediate stages which document the gradual syntactic and functional shift
and which, according to Gohl and Günthner (1999: 70), can be summarized in the following
scale:
weil (subordinating conjunction) > weil (coordinating conjunction) > weil (discourse marker)
Similar data are known for obwohl, which is on the verge of developing from a concessive
conjunction into a corrective discourse marker in present-day spoken German, as we can
see in example (2) (cf. Günthner, 1999, 2000, de Groodt, 2001).
414 G. Diewald
In summary, we have seen that the diachronic development of old particles, as well as
newly emerging discourse functions in conjunctional elements, show that the functional
spectrum of particles is by no means arbitrary or random, but can be ascribed to the
interplay of general grammaticalization scales concerning the semantic development, the
functional development, and the inherent semantic core structure of the items in question.
The core structure remains stable in its essence, but is reinterpreted for its broad semantic
or functional domains, which become more and more abstract, less referential, and more
grammatical. This semantic change, or more precisely this shift of core structure to
various domains, is accompanied by a functional change corresponding to the general
tendencies set up by Traugott, which transcends word-class boundaries. The result is the
coexistence of several heterosemes whose functional spectra, though seemingly unrelated
in a purely synchronic perspective, retain the successive gradient steps of regular
grammaticalization.
4. Model
This section elaborates the main thesis of this paper, i.e., the assertion that DPs and MPs
are grammatical markers of spoken dialogical language use which are analogous to
grammatical markers of written monological language use. The concluding portion of this
section offers a brief outline of the descriptional format for MPs and shows how the
question of polysemy and contextual variation is treated in this model.
4.1. Definition of grammatical signs
One of the defining features of grammatical signs is their indexical potential. They link
linguistic elements of varying size and function to one another or to some relevant non-
linguistic entity.5 It is postulated here that such an indexical relation is also found in the
inherent semantic structure of MPs and DPs, which thus qualify as grammatical markers.
Grammatical functions differ in their specific relational structures and according to the
cognitive and pragmatic domains to which they pertain. This can be illustrated by some
examples that lead to the grammatical function of modal particles. A large number of
grammatical functions are deictic in the strict sense of the term, i.e., they localize the
linguistic entity they apply to with respect to the coordinates of the speaker, the deictic
origo (Bühler, 1934). This deictic function is most obvious for the verbal categories of
tense and mood. The grammatical category of preterite, for example, achieves the
temporal perspectivization or localization of the scene described with respect to utterance
time and, in addition to this purely relational function, also denotes a specific past value
that constitutes the opposition to the present and future tense, as in (12):
(12) Sie schrieb einen Brief.

She wrote a letter.
A different realization of the basic relational structure of grammatical items is found in

anaphoric pronouns. Rather than representing a deictic relation between the speaker origo
and the uttered proposition, they represent a relation between elements of different,
successively uttered propositions. The anaphoric pronoun sie in (13) refers back to the
reference point, i.e., the noun phrase die Katze, whose semantic content is indirectly taken
up by the pronoun.
(13) Die Katze wollte ins Haus zurück. Sie sprang auf das Fensterbrett und drückte
sich an die Scheibe.
The cat wanted to get back into the house. It jumped onto the window sill and
pressed itself against the pane.
A third case, which brings us close to the modal particles, are conjunctions. The basic
semantic content of conjunctions quite obviously contains a relational structure which
usually serves to link clauses. The conjunction aber in (14), for example, points back to
the preceding clause and relates it to the following one.
(14) Sie wollte telefonieren, aber sie hatte kein Kleingeld.

She wanted to make a call, but she didn’t have any change.
This relational component of grammatical signs can be described as a vector, i.e., a

directed relation between a source, via a path, to a goal, as in the following diagram:
(15) point of reference (grammatical sign & unit modified by grammatical sign)
This common semantic template, which consists of the grammatical element relating the
linguistic entity it modifies to some other element, can be specified for the different
grammatical functions that have been described above. For tense markers it can be given
as (16):
(16) utterance time (tense marker & proposition)
For anaphoric pronouns like sie in (13), it can be given as follows:
(17) preceding noun phrase (pronoun & syntactic function)
For conjunctions like aber in (14), the relational structure is illustrated in (18):
(18) proposition 1 (conjunction & proposition 2)
The next section deals with the following question: to what degree can modal particles be
said to have the same relational structure and thus to qualify as grammatical elements?
4.2. Modal particles as grammatical elements
The indexical meaning of MPs has long been recognized. Hentschel (1986: 31), for
example, regards the “metacommunicative deixis” of the MPs as their constitutive
meaning. Franck (1979: 8) holds that the MPs have a sort of anaphoric function and goes
on to specify the element the MP is pointing to as the “specific premises [. . .] about the
argumentative and interactional context of the utterance” [my translation].6 In the present
volume, a similar view is expressed by Aijmer, Foolen, and Simon-Vandenbergen, as well
as by Fischer. Thus, there is general agreement on the relational, respectively indexical
416 G. Diewald
nature of MPs. However, because modal particles display a highly abstract meaning that
is connected to the structuring of discourse, the difficulty lies in exactly stating the
domain to which the relational structure of the MPs applies. It is not directly comparable
to any of the relational functions illustrated in the last section. It applies neither to the
location of the proposition with respect to the speaker origo (as genuinely deictic signs
such as tense markers do) nor does it connect linguistic material that is explicitly
expressed (as pronouns and conjunctions do). In short, the difficulty lies in naming the
reference point of the MP. Here, it is useful to resort to the method of contrasting minimal
pairs, which in this case consist of two utterances differing only insofar as one contains an
MP and the other does not, as in the following examples:
(19) Deutsch ist schwer.

German is difficult
(20) Deutsch ist eben schwer.

German is EBEN difficult
Sentence (19), which has no MP, represents an unmodalized statement that is maximally
independent of or neutral to its communicative context. It does not make any reference to
some other linguistic or non-linguistic entity (ignoring the referential function of the NPs
and the deictic elements of the inflected verb, which of course must be present in any
finite sentence).
The modal particle eben in sentence (20), on the other hand, adds the information that the
speaker regards the proposition Deutsch ist schwer as given, as communicatively
understood. In addition to this process of pointing backwards to something given, the
particle expresses its distinctive lexical meaning, which can be glossed for eben as
“indicating or pointing to the fact that the speaker has held the opinion expressed by the
proposition before and is now iterating it” (see Ickler, 1994: 391). That is, beside its class-
constitutive relational meaning, eben has a lexeme-specific meaning that can be
abbreviated as ‘iterative’. The speaker refers to the pragmatically given unit and modifies
it according to the semantic feature inherent in the MP.
In sum, the function of the MP consists of relating the utterance to a proposition which the
speaker regards as relevant and given and to which he or she relates the actual utterance.
From this observation, we can abstract the central function of all MPs: by referring back
to something communicatively given, the MP marks the utterance containing it as non-
initial.
Because the interchange of initial and responsive turns is the constitutive feature of spoken
interaction, this relational function of the MPs clearly is an indispensable grammatical device
for structuring discourse. With the help of MPs, the speaker marks the turn as non-initial and
responsive, and thus is able to manipulate and modify the ongoing interchange.7 Let us now
turn our attention to the details of this discourse-linking device.
Usually, the proposition to which the modalized utterance refers is not explicitly expressed in
the preceding linguistic material. It is this feature that—beside syntactic differences—sharply
separates MPs from conjunctions. MPs are not text connectives in the strict sense.
This feature is linked to another characteristic which is typical of MPs, namely the fact that
the content of the proposition which the speaker treats as given and refers back to is expressed
in the modalized utterance for the first time. Its propositional core is essentially identical with
the core of the modalized utterance; through the MP, it is modified in some communicatively
relevant way.8
This can be illustrated by the above example: both the proposition that is indexed by the
speaker as pragmatically given and the modalized proposition that is explicitly uttered contain
the proposition Deutsch ist schwer as their core proposition. The modalized utterance
simultaneously achieves two objectives. First, it marks the proposition Deutsch ist schwer as
the one which the speaker thinks is relevant at this stage of the discourse. Second, via the
specific semantic content of the MP, the speaker expresses a modification of the given
proposition. In the case of the MP eben, this distinctive semantic feature consists in specifying
the relation as an iterative one, i.e., it states a repetition of the pragmatically given proposition
in the present scene.
To sum up, MPs refer to a proposition that the speaker treats as pragmatically given, but that
has not yet been explicitly expressed before. This relational structure accounts for the non-
initial status of the modalized utterance. It is the basic indexical relation common to all MPs.
The indexical structure of MPs is given in diagram (21), which highlights the analogy to the
other indexical grammatical categories mentioned before:
(21) point of reference (MP & utterance in the scope of the MP)
Although the pragmatically given unit frequently is a proposition (as in the case of eben
illustrated above), it can include more than a proposition, e.g., a proposition and a specific
type of speech act. Therefore, in a generalized relational scheme for MPs, the
pragmatically given entity is called the pragmatically given unit, as in (22):9
(22) pragmatically given unit (MP & utterance in the scope of the MP)
At this point, it is possible to summarize the differences between the indexical function of
conjunctions and MPs. First, while conjunctions refer to a linguistically expressed entity
that is explicitly stated in the preceding text (and sometimes also in the following text),
MPs refer back to a proposition that has not been linguistically expressed before but is
nevertheless treated as pragmatically given. Second, conjunctions typically represent the
relation between two propositions with different denotative contents, i.e., two clauses. For
(14), this can be spelled out as follows:
(14a) relational scheme for (14):

proposition 1 (conjunction proposition 2)
Sie wollte telefonieren, aber sie hatte kein Kleingeld.
She wanted to make a call, but she didn’t have any change.
MPs, on the other hand, relate two essentially identical propositions and express their
givenness and modification. This is spelled out for sentence (20) by (20a):
418 G. Diewald
(20a) Relational scheme for (20):

pragmatically given proposition 1 (MP & modified proposition 1´)
(Deutsch ist schwer) Deutsch ist eben schwer
German is EBEN difficult
In this section, we have shown that MPs have a grammatical function that is parallel to the
function fulfilled by conjunctions. Thus, MPs qualify as genuine grammatical signs. They
constitute a missing link between text-relational functions and strictly discourse-relational
functions.
4.3. A model for describing the relational meaning of particles
As shown in the foregoing, MPs—in addition to their common, class-constitutive

relational structure—contain additional semantic features which are specific to each
particular lexeme and which comprise the distinctive values of the paradigm of MPs.
These lexeme-specific meanings, which of course are ultimately derived from the
diachronic sources of the particle lexemes, specify how the relation between the relevant
situation and the pragmatically given unit is conceptualized by the speaker. Their
semantic content is analogous to the meaning of conjunctions and conjunctional adverbs,
e.g., eben indicates an iterative connection, aber is adversative, ja affirmative, auch
augmentative, and we can form paradigmatic oppositions as in (20b) that highlight the
distinctive semantic content of each particle in opposition to its neighboring class
members.
(20b) Deutsch ist eben/aber/ja/auch schwer.
Because these distinctive semantic features are inherent in the lexical item and
independent of contextual factors, they will not concern us here any longer.
Instead, the rest of this section is concerned with developing a scheme for the description
of the relational meaning of MPs that permits a systematic treatment of contextual
variation. This model takes up elements of Wierzbicka’s work on complex
communicative meaning (Wierzbicka, 1986, 1991), which is based on a natural semantic
metalanguage (NSM), i.e., a small set of natural-language expressions that are defined as
semantic primes. These semantic primes are used to construct paraphrases (usually
consisting of several clauses), which denote illocutive and situational meaning
components in a uniform way in the same formula as lexical and propositional meaning.
It is not possible to discuss the details of the application of this model to the German
particles here, and it should be mentioned that the following schemes are the first attempt
to formalize the meaning and function of MPs. The basic scheme for MPs, which
specifies the relational endpoints of the indexical relation (i.e., the pragmatically given
unit and the relevant situation) is given in (23):
(23) Basic semantic scheme for MPs:

pragmatically given unit (PGU): proposition (and, if applicable, speech-act type)
relevant situation: speaker’s ‘attitude’ towards PGU
utterance: modified proposition with MP
The first line formulates the pragmatically given unit; the second line notes the relevant
situation; and the third line quotes the modalized utterance, which relates the relevant
situation to the pragmatically given unit and thus gets marked as a non-initial turn or
utterance.
The basic idea behind this model is that specific functions, meanings, and contextual
variations of a particle can be represented by controlled modification of the entries in the
first two lines of the basic relational template in (23). Not only does this apply to different
usages of a modal particle, it also applies to its heterosemes, e.g., discourse particles and
conjunctions. The model thus provides a general template for describing the meaning of
particles in a systematic and unified way. Because the application of the scheme to
conjunctions and discourse particles has been explored in Diewald and Fischer (1998), the
following will focus on specific usages of modal particles, i.e., on the problem of
polysemy and contextual variation. I begin with two prototypical uses of the modal
particle aber, then discuss some highly context-bound interpretations of the modal particle
denn in questions.
As mentioned above, the general scheme must be specified for the different uses of a particle
lexeme. The first example is the usage of the modal particle aber in verb-second statements
with a more or less prominent exclamatory overtone, as in (4), repeated here:
(4) Das ist aber keine gute Idee.

That is ABER not a good idea.
For this very frequent, context-independent usage, the basic scheme gets specified as
follows:
(24) Semantic scheme for MP aber in (4):

pragmatically given unit: das ist eine gute Idee
relevant situation: I think: das ist keine gute Idee
utterance: Das ist aber keine gute Idee!
As can be seen from this, the description of a specific use requires the insertion of the core
proposition in every line. Beside this, the first line (the pragmatically given unit), as well
as the second line (the relevant situation), can be filled up with more specific descriptions.
In (24), the first line contains nothing but the pragmatically given proposition without any
further information. This is the most neutral and most general formulation of the
pragmatically given unit. It signals that there is no specification as to, for example, which
person or which situational factors are the cause of the pragmatic givenness. In this case,
the speaker refers to default situations. If, however (merely to mention a second
possibility), the pragmatically given unit is seen as originating in the discourse partner,
i.e., if a genuinely dialogic structure is implied, then this specific origin must be noted in
the first line as part of the pragmatically given unit. This is achieved by formulations like
‘from what you have said, it follows that PROPOSITION’.
The second line, i.e., the line describing the relevant situation, is likewise open for
modifications and amplifications. They concern the type of illocution which the speaker
chooses in the relevant situation. Different speech-act types are noted by expressions such
420 G. Diewald
as ‘I think’, ‘I want’, ‘I say’, i.e., by an introductory matrix clause indicating the

illocutionary potential that the speaker intends. In example (24), the matrix clause of line
two ‘I think’ notes that the relevant situation is such that the speaker performs an assertive
illocutionary act, i.e., he or she makes a statement.
If, however, aber appears in directive speech acts, which is the second type of
conventionalized usage of aber in German, as in (25), then the intention of the speaker that
determines the relevant situation is indicated in line two by the introduction ‘I want’.10
(25) Das ist echt alles ein klappriger Kram. Mußt aber gucken, ob da auch zwei
Gewinde aneinander sind.
That’s really shaky stuff. You got - ABER - to check whether there are two
threads close to each other.
(Sagerer et al., 1994, quoted from Diewald and Fischer, 1998)
The scheme for this sentence is given in (26):
(26) Semantic scheme for MP aber in directive speech acts such as (25):
pragmatically given unit: daß du nicht gucken mußt, ob da auch zwei Gewinde
aneinander sind
relevant situation: I want: du mußt gucken, ob da auch zwei Gewinde
aneinander sind
utterance Mußt aber gucken, ob da auch zwei Gewinde
aneinander sind
These two uses of aber are highly conventionalized in PDG and, except for the
specifications given in the schemes, do not call for additional context information. In the
case of denn, which serves as the second example here, there is more room for
contextually triggered enrichment (Diewald, 1999b). Furthermore, because the modal
particle denn is exclusively used in questions,11 the dialogic function of marking the
utterance as non-initial, which is taken here to be the common function of all modal
particles, is particularly prominent in this particle.
A typical example for the modal particle denn is given in (27):
(27) Kommst du denn mit?

Are you - DENN - coming along?
Questions with the MP denn can be seen as one of the standard expressions of
interrogative speech acts in German. By using the MP denn, the question is marked as
being merely a consequence of the communicative interaction that precedes it. The
question appears to be motivated externally, instead of being asked arbitrarily or
intrusively by the speaker (see Ickler, 1994: 383). In short, denn indicates a consecutive
relation between the pragmatically given unit and the relevant situation.
The pragmatically given unit towards which denn points back, in addition to containing
the propositional core, also contains the specification of the eligible speech-act type, i.e.,
the interrogative type. Because this component is an obligatory ingredient of the
pragmatically given unit, it must be noted in the first line of the general scheme for denn-
questions, which is given in (28):
(28) Semantic scheme for the MP denn:

pragmatically given unit: I ask: proposition?
relevant situation: I ask: proposition?
utterance: denn & proposition?
For the analysis of individual occurrences of denn in spoken discourse, this basic scheme
must be modified in various ways. I shall now show this for some of the more frequent
variations of denn-questions in spoken discourse. The transcriptions are taken from a
volume edited by Redder and Ehlich (1994).
In the first type, of which (29) is an example, the denn-question is directly connected to
the preceding turn of the interlocutor, which is to say that the pragmatically given unit can
be derived from the preceding utterance.
(29) Zeuge: Und wir fahren, nich wahr, und dann mit der Geschwindigkeit
achtzig bis hundert.
Witness: And we are driving, you know, and then at a speed of eighty to one-
hundred.
Richter: Warum fahren Sie denn so schnell?
Judge: Why are you DENN driving that fast?
(Hoffmann, 1994: 61, 53ff.)
In using denn, the judge marks his question as directly triggered by the turn of his
interlocutor, and in so doing, although introducing a new sequence, at the same time
performs a reactive turn. Thus, the function of this type of denn-question is to link a new
adjacency pair (a question-answer sequence) to the preceding turn. The scheme for denn
as it is used in (29) is noted in (30):
(30) Contextually specified scheme for denn in (29):

pragmatically given unit: from what you said follows: I ask: Warum fahren Sie
so schnell?
relevant situation: I ask: Warum fahren Sie so schnell?
utterance: Warum fahren Sie denn so schnell?
A second type of denn-question resembles the first type insofar as it directly relates to the
preceding turn of the partner. It differs, however, from the first type in that it presupposes
a negative answer, as in (31):
(31) Richter: Und ich weiß jetzt nich, warum Sie versuchen, die Sache jetzt äh
so etwas . schwierig darzustellen. Ob Sie Angst haben oder weil
Sies nich mehr wissen, ich weiß es nich.
Judge: And I don’t know right now why you are trying to present the
case now ugh as somewhat difficult. Whether you are afraid or
whether you don’t remember, I don’t know.
422 G. Diewald
Zeuge: Wieso? Wovor/warum soll ich denn Angst haben, und dann [. . .]
Witness: Why? What/Why should I DENN be afraid, and then . . .
(Hoffmann, 1994: 85, 10ff.)
The witness rejects the judge’s presumption that he, the witness, could be afraid. Because
this type of denn-question clearly displays a preference for a negative answer, it falls
among the group of tendentious questions (see Franck, 1979: 4). The communicative
purposes of these tendentious denn-questions encompass irony, defense, counter-attack,
and others. Usually, the interlocutor—to whom a certain attitude is attributed—is
challenged to give a justification for its content.
In this usage of denn questions, the pragmatically given unit consists of the propositional
core plus speech act type (which is common to any use of denn questions, see example
(28)), and, in addition, of some reference to the attitude of the interlocutor as well as to
the opposing attitude of the speaker. These highly context-dependent factors are
represented in the first line of the scheme as a specification of the pragmatically given
unit. Diagram (32) renders the full scheme for the denn question in (31):

pragmatically given unit: (from what you said follows) youi think: ichj
habe Angst;
Ij think ichj habe keine Angst;
from this follows: I ask: Warum soll ich
Angst haben?
relevant situation: I ask: Warum soll ich Angst haben?
utterance: Warum soll ich denn Angst haben?
This tendentious or rhetoric use of denn questions is a fine example for conversational
implicatures on their way to becoming conventionalized. There is no explicit linguistic
marker to indicate the tendentious or rhetoric sense,12 but this usage of denn questions is
so frequent and stereotypical that it is listed as one of the meanings of the modal particle
denn in the Lexikon deutscher Partikeln by Helbig (1994: 105-106). In the approach
suggested here, there is no need to treat this contextually-induced meaning as part of the
semantic content of the particle itself. Instead, it is treated as a specification of the basic
semantic scheme of denn, the particular contextual conditions of which are spelled out as
part of the pragmatically given context.
In a third type of denn-questions, the pragmatically given unit contains a default

presupposition which the speaker regards as pertaining to the ongoing interaction. The
denn-question is marked as being motivated by this default presupposition.
Unlike the two types of denn-questions treated before, in this third type, the pragmatically
given unit does not point back to the preceding turn. Because this type of denn-question is
very frequently found as the opening turn of a conversation, there usually is no preceding
turn to begin with. Example (33) gives the discourse initial question of an official in a
communal advisory service:
(33) Sachbearbeiter: So, was kann ich denn für Sie tun?
Official: Well, what can I DENN do for you?
(Becker-Mrotzek and Fickermann, 1994: 110, 1)
Here, the pragmatically given unit consists of routine knowledge about different types of
situational frames and discourse structures. By using a denn-question in this type of
communicative situation, the speaker aims to facilitate or mediate the difficulty of starting
a communicative exchange. Because he relates his opening question to a supposedly
given unit, his initial turn does not appear to be initial, but pretends to be non-initial and
reactive. The opening question becomes less intrusive, more polite and partner-oriented.
Number (34) shows the basic scheme for this case:

pragmatically given unit: from what we know about this situation follows: I
ask: Was kann ich für Sie tun?
relevant situation: I ask: Was kann ich für Sie tun?
utterance: Was kann ich denn für Sie tun?
In this way, the basic scheme for the description of MPs can be adapted to more detailed
analyses of individual utterances and thus provides a consistent and homogeneous
description of the varied functions which individual MPs can perform in discourse.
The problem of polysemy and context dependence is solved here not by deciding once and for
all which usage falls under the heading of polysemy and which under the heading of
contextually triggered enrichment, but by providing a unified descriptive template that
incorporates all information relevant for a particular interpretation.13 This seems to be a
reasonable solution in the light of the following facts, some of which have been mentioned
before:
• because the meaning of particle items is non-referential and highly abstract, it is a

priori difficult to discern several lexeme-inherent distinct meanings;
• the answer to the question of whether one usage is merely contextually triggered or
should instead be regarded as an independent and distinct meaning inherent in the
semantic make-up of the lexeme depends, among other factors, on frequency,
stereotypicality, and register, which have to be explored for every item via more
comprehensive empirical data;
• because diachronic change is gradient and pervasive, and because it continually leads
to a semanticization of stereotypical context-induced interpretations, there may not be
a clear cut-off point between contextual variation and polysemy.
5. Conclusion and broader perspective
By applying the frame of grammaticalization theory to the particles of German, this paper
has worked out the following suggestions that can help solve the problems of the
description of discourse elements:
• Discourse particles and modal particles are two distinct classes in German. These
classes are, however, both related to discourse, so that it is reasonable to treat them
424 G. Diewald
together. Having developed a still-expanding class of modal particles, in addition to

discourse particles in the strict sense and conjunctions, German proves to be especially
instructive because the class of modal particles provides a kind of missing link
between grammatical core devices (such as subordinating conjunctions) and
pragmatical devices (such as discourse particles).
• The functional continuum of particles that runs from propositional to pragmatic
functions, and even intermingles them, finds a natural explanation in their diachronic
development, which has been described as a grammaticalization process deriving
discourse particles and modal particles from lexical elements via various intermediate
stages in grammatical word classes.
• The central functions of discourse particles and modal particles are genuine gramma-
tical functions that are indispensable in spoken dialogic interaction, i.e., these items
belong to the broad class of grammatical markers in German.
• As is also the case for grammatical functions in general, the essential function of DPs
and MPs is indexical. Therefore, it is possible to postulate a common semantic
template which is relational and can be described as a vector, i.e., a directed relation
between a source, via a path, to a goal.
• The polyfunctionality of the particle lexemes is due to the reinterpretation of the basic
semantic template, which, depending on contextual and functional specifications, can
result either in a distinction of word classes or in polysemy within a word class or in
contextually-triggered interpretations.
• The functions of MPs/DPs are described via a unified scheme that specifies the
endpoints of the relational structure and allows for the systematic integration of
specific and contextual information.
In a broader perspective, the suggestions put forward in this paper aim to extend the
notion of grammatical device into the pragmatic sphere in order to incorporate items that
organize spoken dialogic language use. In their function and structure, these items are
parallel to grammatical devices which organize written monological language use.
Because the main focus of studies on grammatical markers has traditionally concentrated
on the latter devices, the model developed here may call attention to the continuum
between these modes of linguistic production, and may thus lead to a more comprehensive
notion of grammar.
Notes
1
The core of this class consists of the following fifteen extremely frequent items: aber,
auch, bloß, denn, doch, eben, eigentlich, etwa, halt, ja, mal, nur, schon, vielleicht,
wohl (Gelhaus, 1995: 371; Helbig and Buscha, 1986: 487ff.). For an extensive
treatment of MPs, see Diewald (1997) and Diewald and Fischer (1998).
2
In addition to this, aber is also sometimes used as an adverb, as in the following
example:
Und zwei Löcher überlappen sich, zeigen aber in die gleiche Richtung.
And two holes overlap, however, they point into the same direction.
Here, aber has constituent value, appears in the middle field after the finite verb, and can
be substituted by the adverb jedoch ‘however’.
3
For a discussion of the notions of indexicality and subjectivity see Diewald (1999b: 14-
16).
4
The details of this development cannot be discussed here; see Diewald (1999a) and
Diewald (1997) for an extensive treatment.
5
On indexicality as a defining feature of grammatical signs, see also Jakobson (1971
[1957]), Leiss (1994), Diewald (1997).
6
Many other theorists likewise argue in this vein. See, for example, Doherty (1985: 15),
Ickler (1994: 377), and Petric (1995).
7
Ickler (1994: 377) also hints at the dialogic function of modal particles when he states
that “modal particles fit certain utterances into larger textual entities, which, in
principle, are to be interpreted as dialogic” [my translation].
8
This view is similar to Foolen’s description of the class-constitutive meaning of modal
particles: “Als Klassenbedeutung für Modalpartikeln gilt, daß sie immer auf eine
implizite, im Kontext relevante Proposition hinweisen. Diese implizite Proposition ist
immer eine logische Variante der explizit ausgedrückten Proposition” (Foolen, 1989:
312-13).
9
In Diewald and Fischer (1998) the pragmatically given unit is called the pragmatic
pretext; see also Fischer (this volume).
10
In this example, the directive speech act is encoded indirectly via a statement with an
elliptical subject. However, there are true morphological imperatives with aber in
German.
11
Denn has a heteroseme as a causal conjunction as in:
Er schreibt alle Beobachtungen auf, denn er will Detektiv werden.
He makes notes of all his observations, because he wants to become a private
detective.
12
Structural sentence type is not distinctive here. Both wh-questions and yes-no questions
with denn may receive a “regular”, non-tendentious interpretation as well as a
tendentious interpretation.
13
It seems relevant here to repeat that the details of the descriptive format and its
uniformity still need more elaboration, including further empirical research. I hope,
however, that the above discussion has explained the principle and demonstrated the
usefulness of this model.
Chapter 22
Frames, constructions, and invariant

meanings: the functional polysemy of discourse
particles
Kerstin Fischer
1. Introduction
Discourse particles are items that are notoriously difficult to describe regarding all
linguistic levels involved (Wilkins, 1992: 155), and it is not even clear whether they
constitute a class. The main problem in the description of discourse particles, however, is
taken to be their functional polysemy.
1.1. Approach
The functional spectrum each discourse particle lexeme can fulfil is considerable, and
their interpretation can vary very much depending on the situation in which they are used.
Still, interactants seem to be able to interpret occurrences of discourse particles, and many
previous studies show that discourse particles are not distributed randomly and that they
cannot be used interchangably (e.g., Fox Tree and Schrock, 2002). The best starting point
for an analysis therefore seems to take the discourse particle lexemes as the anchor point,
and to investigate the different variables that determine their interpretation and use one by
one. The approach taken here is therefore semasiological. That is, the perspective is to
start out from particular linguistic forms, to describe their functional spectrum, and on this
basis to develop a model of pragmatic interpretation.
One factor that influences the interpretation of a discourse particle occurrence seems to be
the structural context in which it occurs (e.g., Heritage, 1984). Investigating the structural
contexts of discourse particles provides us with information on the role of their position
with respect to turn and utterance, of their prosodic properties, and on their role in the
verbal and extralinguistic action taking place. In order to get an idea about the possible
variability of their functional spectrum, we then have to compare the results from our
analysis with data from corpora recorded in different situations. Only by observing the
way discourse particles can be employed in a variety of situations can we hope to
understand how the contribution of the properties attached to each particle morpheme, the
428 K. Fischer
word class, the structural context, and the situation contribute to the interpretation of its
uses.
The approach taken here proposes three concepts to account for the contributions of the
discourse situation, the structural context, and the morpheme respectively: frames,
constructions, and invariant meanings. Thus, it is believed that the fact that not every
discourse particle can mean everything is best accounted for by the postulation of
invariant meaning components. In order to account for the extreme variability of the
meanings and functions of discourse particles, a mechanism is needed by means of which
it is explained how discourse particles get their particular functional spectrum. This is
achieved in this approach by the postulation of a number of communicative domains to
which speakers orient and which constitute a communicative background frame, i.e., a
model of the situation attended to. Finally, in order to account for the interpretation of
particular discourse particle occurrences in context, a number of constructions, i.e.,
general form-meaning pairs, are proposed.
1.2. Methodology
There are a number of methodological premises which have guided this investigation. For
instance, it is assumed that intuition is not reliable if we want to determine the meanings
and functions of discourse particles. Instead, it is necessary to base each analysis on the
close examination of corpora of spoken interaction, since, to use Sacks’ (1984: 25) words,
“from close looking at the world we can find things that we could not, by imagination,
assert were there”. Similarly, it is held that we have to be very cautious about which
explanatory concepts to use in our analyses; that is, the descriptive categories employed
should be those to which the speakers themselves can be shown to attend. Thirdly, if we
classify the data beyond the structural contexts apparent in the sequential structure of
dialogues, we have to ensure the explicitness and intersubjectivity of the classification
criteria.
The method mainly used is conversation analysis (CA). For instance, the different
possible interpretations of the discourse particle lexemes are identified by means of CA
methods. Using CA commits us to the use of attested corpus data and the focus on speaker
categories. However, a consequence of the focus on the interactional relevance of the
items under consideration is the close connection between particular uses of discourse
particles and their interactional contexts. Such a methodology does not provide the means
to address the connections between the different uses. Thus, other methods have to be
used for the development of a model of the polysemy of discourse particles. Here, CA can
be again very useful by providing methods for the identification of the components of the
model of the communicative situation as domains attended to by the speakers. Especially
in the comparison of different corpora, these domains become apparent as speaker
categories.
Comparing different corpora is crucial to the analysis here proposed because only if we
understand how the mechanisms work that allow speakers to use discourse particles for so
many different purposes depending on the situational requirements can we arrive at a
model that can deal with the flexibility observable. Such a model, it is assumed here, has
to be based on the speakers’ needs, and these cannot be assumed from an outside
perspective but have to be categories to which speakers are really oriented. A comparative
Frames, Constructions, and Invariant Meanings 429
CA analysis can provide the different communicative tasks attended to differently in the
different corpora as speaker categories.
For describing the invariant meaning components of each discourse particle morpheme,
Natural Semantic Metalanguage (NSM; Wierzbicka, 1986, 1996) is taken to be, in spite of
a number of methodological problems, the most useful descriptive language available.
Core meaning descriptions, however, need to be validated. The formulations in NSM
proposed can be tested on corpus instances by means of tests outlined in Fischer (1998).
These tests rely on judgements of oddity, based on Cruse’s (1986) test frames. While
there are a number of problems with tests of core meaning components (see Schourup,
1999: 255-7), the only alternative is to argue for the plausibility of the analyses.
For the representation of generalized structural contexts, the notion of construction

(Goldberg, 1995; Fillmore and Kay, 1999) is proposed, and the influence of the
communicative situation will be accounted for in a frame-based representation (e.g.,
Fillmore and Atkins, 1992).
1.3. Data
The functions identifiable for a given discourse particle crucially depend on the
communicative purposes relevant in the situation in which it is used. Therefore, the
corpus studied is of great importance. To use conversation as a corpus is of course the
most natural starting point, yet some functions of discourse particles only become
apparent when speakers are simultaneously engaged in particular non-linguistic actions.
For example, in task-oriented dialogues in which one speaker instructs another how to
construct a part of an air compressor (Grosz, 1982: 154), the English discourse particle
okay may be used to signal that the constructor of the compressor has finished a
construction phase. In the following example, the expert (E) provides the apprentice (A)
with an instruction. After a partial repletion of the expert’s instruction, the apprentice uses
okay to signal the completion of the task, as is apparent from the expert’s reaction, which
describes the consequences of the successful completion of the working step:
(1) E: Open the top of the valve and let the water out. Just open the faucet up on top.
Just like you were going to turn the water on.
A: Oh, like I’m going to turn the water on. O.K.
E: Now, that’ll relieve the pressure.
It is unlikely, though not impossible, that such a reading of okay could be found in
conversation as well, because the task, to signal the completion of an action requested,
will be rarely relevant in conversation. However, such uses are likely in naturally
occurring service encounters (see Merritt, 1980).
The first set of data I will investigate here are American English appointment scheduling
dialogues in unrestricted settings between two adult speakers, recorded at Carnegie
Mellon University in the framework of the Verbmobil project (VM Database, 1995). The
participants’ tasks is to schedule appointments, and the calendars they use are not their
own but have been set up by those collecting the data. For this reason, participants do not
freely construct their topics. Yet, turn length, turn distribution, and turn content are free to
vary and thus the interactive processes between the participants are still close to those
430 K. Fischer
observable in conversation. Using task-oriented data has not only the advantage that we
can investigate how discourse particles may be used to fulfil functions regarding the extra-
linguistic context, but it also allows the controlled investigation of particular variables that
may influence their usage.
The second set of data are correspondingly also appointment scheduling dialogues, yet
they are held between humans and a (simulated) computer, that is, the participants
believed to be talking to a computer while the computer output was actually simulated by
a human “wizard”. The dialogues were recorded in order to study how speakers react to
communicative failure in dialogues with automatic speech processing systems, and thus,
after a short cooperative phase, the computer output often exhibits misunderstanding,
failed understanding, or synthesis errors. For details see Batliner et al. (2003).
The focus of this paper is on the functional polysemy of discourse particles. While their
functional interpretations can vary immensely depending on their contexts of usage, it is
assumed here that there are some basic observations that, if accounted for, allow a
sufficient explanation of the functional spectrum observable as well as further possible
readings they may get in different situations. Thus, the model of the meanings and
functions of discourse particles proposed here attempts to account at least for the
following observations:
• The particular lexical item contributes to its interpretation; that is, although there is a
huge range of possible interpretations for each discourse particle lexeme, not
everything is possible. Instead, the use and interpretation of individual discourse
particles seems to be learnable, and the interpretation of the individual occurrences is
obviously possible. There is thus some morpheme-specific contribution by each
discourse particle.
• It is a particular, rather than an arbitrary, range of functions discourse particles can
fulfil. Although scholars disagree much about the functions of discourse particles,
there is a range of functions that is commonly, and often cumulatively, attributed to
them. This spectrum includes functions with respect to the turn-taking system, the
indication of discourse relations, discourse structuring, the regulation of interpersonal
relationships, speech management, or politeness. Most of the functions proposed are
also fulfilled by items of other word classes, and thus only the accumulation of the
functions proposed in the literature seems to be specific to discourse particles. This
functional spectrum, however, is directly related to the tasks to which speakers can be
shown to attend in the communicative situation.
• Discourse particles do not occur anywhere in a turn but in particular structural slots,
and their position and intonational realization (see Ehlich, 1986) may influence their
interpretation.
In order to account for these three observations, a tri-partite model of the functional
polysemy of discourse particles will be presented below. Such a model may also allow a
definition of the word class, and it will have particular consequences for several
distinctions of general linguistic interest.
2. Definition
For the items discussed in this paper, the term discourse particle is used, rather than, for
instance, discourse connective or discourse marker. The term discourse connective, for
instance, would associate the whole class with a particular function, which would not do
justice to the broad range of functions discourse particles are held to fulfil. The functional
spectrum discourse particles can fulfil is indeed so broad that a definition on functional
grounds does not seem justifiable. A functional definition indeed requires an independent
model of the functional spectrum and the role of discourse particles therein. Without such
a model, it appears as if the only thing that seems to connect the different functions is the
fact that discourse particles can fulfil them. While indeed a general function can be
assumed (see section 5.1), this general function does not explain the individual functional
interpretations. The functional spectrum discourse particles can fulfil is only motivated by
the communicative tasks to which the speakers attend. Thus, the definition of the
functional spectrum requires an explication of what the communicative situation consists
of. This definition can of course vary from situation to situation. The functional spectrum
of discourse particles can thus not be determined once and for all; just the mechanism by
means of which they get their functional interpretations in the various contexts can be
specified. It seems therefore necessary to take a particular, formally definable class as the
starting point for the investigation, rather than attempt a functional definition.
Of course it is true that a number of discourse particles are not particles, that is: small,
syntactically, semantically, and often prosodically unintegrated, uninflected words. For
instance, you know is formally complex and, like look and listen, it can be held to be
inflected. Yet they occur invariantly in this form and can thus be regarded as lexicalized.
Therefore, they are both formally, regarding their unintegratedness into the utterances
they may occur with, and functionally similar to discourse particles. The most typical
members of the class under consideration (in English) are thus indeed small, uninflected
words, yet the list of objects to be studied should include other lexicalized items with
inflexible formal aspects. The word particle is thus used as a cover term because it calls
up conveniently the association of not-integrated items, accounting for the fact that
discourse particles are generally not part of utterances. More correctly, however, the class
under consideration should be understood as unintegrated, lexicalized, idiomatic items,
the prototype being particles.
Discourse particles also display a characteristic semantic structure. They are not believed
to mark anything. Instead they are regarded as lexical items, i.e., lexicalized form-
meaning pairs, whose meanings are underspecified. In this way, they are similar to
linguistic signs like plural morphemes, word order, or tense markers. Their semantic
content consists in claims of ongoing mental processes, specified by reference to aspects
of the communicative situation. Common to all discourse particles are thus not only their
underspecified invariant meanings, which is why, for instance, Hentschel and Weydt
(1989) and Weydt (this volume) have described them as synsemantica, but also the nature
of their lexical meanings, which all report on the speaker’s mental state. This does not
mean that they really express the speaker’s cognitive processes, but that they are claims
(Schegloff, 1982) of ongoing mental processes that can also be employed strategically
(Fischer, 1999). These underspecified signals of mental processes, for example, as a
“change-of-state token” (Heritage, 1984), are contextually specified by reference to a
432 K. Fischer
certain communicative domain on the basis of a model of the communicative tasks to

which speakers attend.
Discourse particles can be distinguished from their “homonyms” in other word classes by
the domains to which they refer. For instance, answer particles directly refer to the
propositional information conveyed, while discourse particles refer to particular aspects of
the speech situation (the communicative background frame, see section 4). Likewise,
modal particles can be distinguished from discourse particles in that the former refer to
the pragmatic pretext, to some kind of information presented to be at hand (Diewald and
Fischer, 1998; Diewald, this volume), or, in Ducrot’s words, the “different points of view
which are expressed within an utterance” (Ducrot, 1996: 68-69). Connectives (see Pons
Bordería, this volume) or text relation markers (Roulet, this volume) refer to two
elements: to a host utterance and to an aspect of the pragmatic pretext, the discourse
situation (see example (2) in Hansen, this volume), or what Roulet (this volume) calls
discourse memory. Therefore, unlike discourse particles, connectives are not prosodically
and structurally unintegrated but connected to some host utterance (see Lewis, this
volume, and my introduction to this volume).
By attributing the word class-specific properties of particles to different domains of

reference, the model explains how the same particle morpheme may function in different
word classes (Abraham, 1991). Similar to Weydt (this volume), it is thus assumed here
that “a certain particle can fulfil the function of a Gliederungssignal [a discourse particle,
K.F.], because of the fact that it maintains its original meaning.” The same morpheme
may thus function in different word classes, from which different lexemes, i.e., discourse
particles, modal particles, connectives, etc., may result. The contribution of the respective
word classes, besides specifying the syntagmatic properties of the item under
consideration, consists in specifying particular reference elements (see also Diewald, this
volume).
Now, discourse particles are semantically not an entirely homogeneous group. The
different subclasses can be distinguished with regard to the kinds of mental processes
signaled: interjections, such as English oh, ah, and oops, all display the sudden
recognition of some kind of information. Describing their invariant meaning components
by using NSM, it can be found that all interjections contain an “I now”-component
(Fischer, 2000a). Hesitation markers, such as uh and um, in contrast, do not display a
sudden change in the speakers’ mental states but instead indicate a current process: “I am
thinking” (Fischer, 1999, 2000a). Segmentation markers, such as well, yes, or okay, which
actually fulfil many more functions than just segmenting discourse, can be divided into
two different kinds: those whose meanings directly involve the communication partner
(e.g., for English yes: “I think that you and I think the same”; Fischer, 2000a: 253-8), and
those which state the results of a cognitive process, such as English well (“after I have
thought about all I know about it I say this”; Fischer, 2000a: 245-9). Thus, the mental
processes displayed allow a classification of discourse particles regarding the subclasses
interjection, hesitation marker, and segmentation marker, while the domains to which
particles may refer and which specify their underspecified meanings serve to distinguish
discourse particles from other word classes.
The functional spectrum of discourse particles will be exemplified here by an analysis of

the discourse particle okay in the appointment scheduling dialogues described in section
1.3. Okay shares many functions with discourse particles signaling agreement, like yes or
yeah, with topic structuring discourse particles like well, uh, hm, or oh, and with tag
questions. The morpheme okay can furthermore function as an adjective or adverb, as in
“mjcb_5_04: Wednesday’s okay” or in “fkcf_5_05: would one thirty be okay with you?”
Since okay participates in two word classes, it is a useful example for illustrating the
contribution not only of word-specific but also of word class-specific meaning aspects and
their interaction.
In the human-to-human appointment scheduling dialogues investigated, besides the

adjectival and adverbial uses of okay, we can find okay
• turn-initially;
• turn-medially, but in utterance-initial or second position (after another discourse
particle); and
• turn-finally.
The first example of okay is turn-initial after a question. After the occurrence of okay,
speaker fcaw_4 repeats the date and time previously proposed by her communication
partner, summing up the information agreed upon. We can thus assume that okay
constitutes a response to a proposal here, stating agreement regarding the date and time
proposed:1
(2) mdkr_4_06: well, it looks like our only choice is gonna be Thursday, um I’m in a
meeting till ten, and then I’ve got the rest of the day free right now,
um so do you wanna say early afternoon? around one?
fcaw_4_07: okay. Thursday afternoon around, uh one o’clock, on June third’ll
work for me. why don't we make it then.
In the following example, there are two further turn-initial occurrences of okay. The first
follows a suggestion for a date when to meet. Thus, okay occurs in a position in which
agreement may be relevant, namely directly after a proposal, as we have seen in the
previous example. However, okay is followed then by an occurrence of uh and a
clarification question concerning which Tuesday her partner had meant in her proposal.
The partner’s reaction is correspondingly a clarification of the day meant, followed by a
question containing the same proposal as previously made. Thus, the speaker does not
take the previous occurrence of okay to mean acceptance of the proposal. In the next turn,
speaker ffmv_7 utters a second okay, followed by an explicit statement that the date
suggested suits her well. Thus, in the second case, okay can be taken as a signal of
agreement regarding the content of the utterance, whereas in the first occurrence okay has
not been interpreted as such by her communication partner:
(3) menm_7_05: well, could you come in at eight o’clock on Tuesday, then we could
do it from, eight to ten on, Tuesday.
ffmw_7_06: okay, uh do you mean Tuesday the twenty third?
434 K. Fischer
menm_7_07: yes I do, Tuesday. November, twenty third. eight to, ten AM. how’s
that sound to you.
ffmw_7_08: okay, that’s fine, I'll see you Tuesday November twenty third then.
#paper_rustle# thanks.
Also in the next example, okay occurs turn-initially; however, it is not preceded by a
proposal for an appointment, as the previous instances of okay investigated were. Instead,
it occurs at the very beginning of the dialogue. Yet, the speakers do not greet each other
because they have previously carried out another appointment scheduling dialogue. It thus
occurs at a thematic break, the transition from one appointment scheduling phase to the
next, possibly mediated by new instructions or the exchange of calendars by the person
supervising the recordings. That is, okay marks here the beginning of a new phase in the
interaction:
(4) fmjm_3_01: okay Danny, now that this meeting’s over, we need to schedule
another one so we continue, to get our work done on this project,
and, the times that, I would have free, coming up, would be on the
twenty fifth, in the morning, twenty seventh in the afternoon,
eh February third, in the afternoon, and that’s about it, do you have
any of those times free?
Consider two more turn-initial uses of okay. The first use of okay in the following
example occurs after the communication partner has uttered you know, which seeks for a
reaction of accordance from the partner (see Schiffrin, 1987a:275; Östman, 1983). So
okay may be a reaction to the tag question, signalling understanding. However, the
utterance following okay does not refer to the information that the event will be a joint
thing. Instead, okay and the statement that speaker fdlw_2 will bring something occur in
reaction to the partner’s request to bring something to drink, formulated as a matter of fact
at the beginning of the turn. Thus, okay may do many different things in this utterance:
return to the previous topic, signal understanding regarding the preceding utterance, and
indicate an agreement with the whole scenario outlined by her communication partner:
(5) mggd_2_07: alright, I’ll bring the donuts, and um you bring something to drink.
maybe some coffee. or maybe some uh some coca cola. something, I
dunno. it’s gotta be like it’s gotta be a joint thing. you know?
fdlw_2_08: okay, I’ll do it, I’ll bring something, I’ll see you then, bye,
mggd_2_09: okay, bye,
The second okay in the example occurs at the end of the interaction, after the
communication partner has concluded both the appointment scheduling phase and the
interaction itself by summing up and uttering bye. Thus, okay is used here to conclude the
conversation, which is then ended by the second pair part of the closing bye (see
Schegloff and Sacks, 1973).
The following example shows okay in second position within the turn, that is, it is
preceded by another discourse particle, here um. The previous turn ended in a question
proposing two possible dates for an appointment. After the hesitation marker um, okay is
used with falling intonation and a short pause, as indicated by the full stop in the
transcript. The speaker continues by stating her restrictions with respect to times when to
meet on those two days suggested. Thus, the speaker takes up the proposal and continues
with relevant information, elaborating on her communication partner’s proposal. So
although acceptance of the proposal made is relevant here, okay may display here only
successful perception and understanding to the communication partner, indicating that
what is to come is related to her partner’s turn. Thus it may be a signal of acceptance of
the communication partner’s contribution (Clark and Schaefer, 1989):
(6) fmjm_1_01: hi Danny, /glottal/ now that we’ve finished our last meeting, we
need to, arrange another one #begin_drawer_noise# within the next
two weeks, and I’m looking at my schedule, and the days that I
would have free, so we could meet and get #end_drawer_noise# it
over with at a reasonable time would be the sixteenth, or the
nineteenth. what do you think of that,
mdrd_1_02: um okay. the sixteenth, I’m, busy from nine to twelve, and the
nineteenth I'm busy from eight to five, so anytime around there
would be, hip with me.
Okay can also occur turn-finally:
(7) mjay_6_01: #paper_rustle# hi Arthur, before we go, I think we should schedule

a meeting sometime in the next two weeks, for at least two hours. o/
okay?
maem_6_02: sounds good. um let’s see, on, Monday I have something from one
to four, can you meet in the morning?
In this example, okay occurs at the end of the turn with rising intonation. The
communication partner reacts to this with a signal of agreement: sounds good. Similarly,
in the following example, okay is used turn-finally at the end of a statement summing up
the joint plan. Also in this example, the communication partner reacts with a statement of
agreement, in this case even with another instance of okay with falling intonation,
followed by a repetition of the communication partner’s previous words: see you then. We
can conclude that the function of okay in turn-final position and with rising intonation is
to ask for agreement.
(8) fdlw_1_14: sounds great. don’t know about you but, um /begin_laughter/ it’s
been a hell of a week so, /end_laughter/ I’m ready to, uh to uh go
tie one on, or /begin_laughter/ whatever you wanna call it.
/end_laughter/ so, I’ll see you then, okay?
mggd_1_15: okay, um, see you then,
The agreement asked for is provided in this example by another instance of turn-initial
okay, followed by a repetition of the communication partner’s linguistic material. The
exchange serves furthermore to bring the conversation to a close.
To sum up, okay has been found in turn-initial position to signal agreement after
proposals (example (2)), even if the proposal was followed by further explanations as in
example (5). It was found to mark the beginning of a new appointment scheduling phase
(example (4)). Furthermore, it has been found in a similar structural context as in example
(2) to be used to continue relevantly with the same topic, yet without signaling agreement
436 K. Fischer
with the previous proposal (examples (3) and (6)). In example (8), it occurred as part of a
closing sequence (see Schegloff and Sacks, 1973). Finally, it has been found turn-finally
with rising intonation in checking function (example (7)).
Previous analyses have identified a similar functional spectrum for okay. For instance,
Stenström (1994: 67), investigating English conversation, holds okay to work as appealer,
that is, as a turn-final checking signal, as an answer and acknowledgement signal, and as a
framer, re-opening (1994: 124) or terminating topics (1994: 154). Regarding the framing
function, example (4) in our data shows that the topic needs not be previously dealt with.
Similarly, examples from Condon (1986: 80) illustrate that okay can be used at the
beginning of a new decision sequence. Thus, resuming or reopening the topic would not
be an accurate description of these examples. However, although the topics in example (4)
and in Condon (1986: 80) have not been previously discussed, they have already been
agreed upon by the participants. Condon (2001: 492) therefore argues that okay marks
“transitions to some expected sequence of talk.” This could explain that we have also
found okay at the beginning of turns in which the speaker elaborates on the proposal
previously made (example (6)) or in which she asks for further clarification (example (3)),
thus continuing on the previous topic. Similarly, Bangerter and Clark (2003) and
Bangerter et al. (2003) argue that okay is primarily used for vertical navigation of
conversation, that is, to enter and exit joint projects. In particular, okay is used for
signaling consent to beginning or ending a topic. To conclude the discussion of the topic
function of okay, we can summarize that okay can introduce new topics that have been
jointly established before, it can continue on, reopen and terminate topics.
Grosz (1982: 153) identifies four functions of okay in task-oriented dialogues, which do
not include the framing function. Instead, the function she identifies comprise signaling:
“I heard you,” “I heard you and I understand,” “I heard you, I understand, and I am now
doing what you said”, and “I’m finished (O.K. what’s next?)”. Similarly, Merritt (1980),
in her conversation analytic study of okay in service encounters, shows that okay, in
contrast to yes-like items which usually occur in response to requests for information, is
generally used in response to requests for action. She therefore holds okay to fulfil a
bridging function, namely to anchor non-verbal action in what has gone on verbally and
to provide explanations of what follows (1980: 165). However, she argues that it is not
always nonverbal action following, but that with okay the speaker may also acknowledge
that it is her turn to take some action. A consequence is that okay, in signaling
acknowledgement that it is one’s own turn to act, releases the communication partner
from further obligations. This is why, in her view, it can also occur after rejections (1980:
166).
Merritt’s uses of okay in reaction to requests can be compared to examples (2) and (5)
above in which okay occurs in reaction to proposals. In these examples, the speaker
provides the communication partner with some behavior that is in accordance with the
communication partner’s wishes or suggestions. The point therefore does not seem to be
whether the action the speaker commits herself to is nonverbal or verbal. Condon (1986:
87) argues that the behavior requested just cannot be the provision of information. Thus,
in her view, okay cannot be used as a response to information questions such as “are your
eyes blue?”. Correspondingly, in spite of its form as a yes/no-question, “Do you wanna
say early afternoon” in example (2), which is replied to by means of okay, is treated as a
proposal by the communication partner, not as an information question. Stenström (1994)
supports the restriction of possible responses after questions that exclude okay as an
answer to a question. Okay therefore seems to be restricted to particular illocutionary acts.
This relates to Helbig’s (1988) suggestion that discourse particles function as
illocutionary indicators. Yet, the question is whether speakers really attend to such
information as informatory cues. It seems more likely that the inherent properties of okay
constrain its usage.
Condon (1986) argues that the two functions identified by Merritt in service encounters,
to acknowledge and to link two stages of an encounter (1980:168), can also be found in
other types of discourse, such as the task-oriented family interactions she analyses. She
identifies many instances in which okay is used to signal agreement in decision making
sequences. The turn final response eliciting use of okay has also been found by Condon
but was discarded from her analyses that focus on the framing function of okay. Condon
concludes that okay is a “virtually contentless particle” (1986: 98), but argues for a strong
correlation with the structural organization of the interaction. The organization function of
okay which in her view consists in reflecting “the complex, difficult-to-identify structure
of the interaction” (1986: 75) is further developed in Condon (2001). In this paper, she
investigates the (quantitative) correlation between discourse steps in task-oriented
dialogues and the occurrence of okay. The main question regarding such an analysis is
only for whom okay reflects the discourse structure, i.e., whether this function is attended
to by the participants in the interaction.
What we may ask now is which of the functions identified in the previous studies and the
analysis above can also be found in human-computer appointment scheduling dialogues.2
The question is how the functional spectrum of okay changes in this kind of interaction,
because only on the basis of an understanding of the mechanisms by means of which
discourse particles get their interpretations in very different situations can we identify the
determining factors and develop a model of their functional polysemy.
First of all it has to be mentioned that in the seven dialogues of between 20–30 minutes,
only three speakers used okay as a discourse particle in the main phase at all. In the other
four dialogues it was used by three speakers just once and by the fourth speaker thrice, but
exclusively in the cooperative phase at the beginning of the dialogues or directed to the
conductor of the experiments.
The following example shows okay turn-medially in a context in which the computer has
uttered nonsense. The speaker accordingly asks a clarification question first, before she
utters okay. Okay can therefore not signal acknowledgement here:3
(9) s4011107: I have noted the appointment for you.

e4011107: <%><;whispers>
s4012101: blablurb was soll date?
e4012101: sorry? <Cough> okay. what about Monday, the fourth of
January? from eight till fourteen-hundred.
Similarly, in the next example, okay occurs in reaction to a display of a misunderstanding

of her proposal, however, not immediately, but after pausing, breathing, sighing events
and an instance of the hesitation marker uh. Thus, whereas in human-to-human
communication speakers signal acceptance of the previous turn, this is a highly unlikely
438 K. Fischer
interpretation here. Furthermore, the information presented after okay constitutes no

relevant contribution to the previous topic:
(10) e4014302: <Cough> Wednesday, the thirteenth of January ninety-nine.

s4014303: the weekend is already occupied.
e4014303: <;groans> <;sighs> uh okay. Tuesday, nineteenth of
January ninety-nine?
In the following example from a different speaker, the context in which the first instance
of okay occurs is the same as in the previous example. Likewise, okay cannot function as
a signal of acknowledgement here because the rejection the computer has produced does
not refer to the speaker’s previous proposal. Thus, it is not even clear which weekend
could be occupied. The second okay then is framed by two metalinguistic statements, the
first one being interrupted. It consequently serves as a repair marker:
(11) e4072302: <Swallow> I have time on Thursday the twenty first of
January at two pm.
s4072303: the weekend is already occupied.
e4072303: <Smack> okay. let's try okay, I have a
another suggestion. how ’bout Monday, the eighteenth of January
 at twelve pm?
In contrast to those examples discussed in the human-to-human dialogues, these uses of

okay do not function as acknowledgement of the content because the computer’s reaction
does not make sense, and the occurrences cannot plausibly be argued to signal successful
perception and understanding, either, because they occur delayed and after irrelevant or
even not understandable utterances. Instead, they occur after communication problems
and before new attempts initiated by the human speakers to make themselves understood.
My suggestion is therefore that these occurrences of okay serve the speakers’ speech
management purposes, concluding a previous attempt to make oneself understood and
starting a new one. This interpretation is supported by the speaker’s use of metalanguage
in example (11) that makes the switch of strategy explicit. They therefore fulfil structuring
functions for the dialogue (see Condon, 2001).
To sum up, the functional spectrum of English okay seems to center around the two
functions of signaling acknowledgement and framing, but a number of further functions
has been identified as well. It can, for instance, occur as a repair marker in human-
computer interaction, and it may signal the completion of a requested action. The model
to be developed will have to account for this functional spectrum, including the variability
of its interpretation with respect to different situations.
4. Model
The questions to be addressed regarding okay include an account of how the functions of
okay identified above belong together. Thus, we have to ask how okay can signal
acknowledgement in many instances but not in others, how it can open, reopen, continue,
and terminate topics, and how its functional spectrum can change such that the partner-
oriented acknowledgement functions are lost when it occurs in human-computer
interaction. Another question concerns the surface cues that make the interpretation of
okay possible. Furthermore, it can be asked whether okay is really “virtually contentless”,
or whether it has invariant meaning aspects that license the different functions it can fulfil.
4.1. Discussion of the empirical results
The implications of the results from the empirical analyses for the model will be used here
to illustrate and motivate the model presented in the next section.
First of all, we can notice some dependency between the structural context in which okay
may occur and its interpretation. Condon (1986), for instance, proposes that the
acknowledgement reading of okay is tied to its position as, or at the beginning of, a
second pair part. This criterion, however, is not sufficient, since the structural contexts of
examples (2) and (3) are very similar. The only difference is that the information
following the utterance of okay constitutes a clarification question in example (2) and a
repetition of, or elaboration on, the same information as proposed by the communication
partner in example (3). We therefore have to distinguish between two structural contexts,
one in which okay signals compliance with the proposal made or request issued, and
another one in which it signals the successful acceptance of the communication partner’s
turn. Both structural contexts are quite different from the structural contexts in which okay
functions to mark, for instance, the beginning of a new, jointly established topic. Thus,
although the structural contexts may differ only slightly in some cases, they provide a
useful means to distinguish different readings in general.
Regarding the relationship between the different readings, Merritt’s proposal is that okay
can fulfil the bridging function between different interaction phases on the basis of its
acknowledging function. The acknowledgement is in such cases reduced to the
acknowledging that it is one’s turn to do something, therefore releasing the other one from
further obligations. Taking up this proposal for an invariant contribution of okay, we still
need to show how the whole spectrum of interpretations of okay can be accounted for.
This is an additional but necessary step, since just postulating a core meaning would leave
the actual uses of okay unspecified. The aim is therefore to find some common core and to
define the extensions further. In particular we need to ask whether, if we take the idea that
okay signals acknowledgement as a starting point, we can account for the individual
interpretations. We may, for instance, ask what exactly okay acknowledges when it
releases the communication partner from further obligations. Merritt’s (1980: 166)
example is repeated here:
(12) C: Do you have two dimes and a nickel for a quarter?

S: (rings cash register, opens drawer) We don’t have any dimes left.
C: OK. Thank you.
The proposal made here is that the acknowledgement signaled by okay in this example
refers to the successful perception and understanding of the communication partner’s
utterance. The relation of accordance expressed by okay thus refers to what is being said
on the one side and what is being heard and understood on the other. The same applies to
the uses of okay in dispreferred seconds, where okay does not, for instance, signal
acceptance of the proposal made. Thus, the relationship of accordance expressed may
refer to the content of the utterance in some uses or only to the successful acceptance of
440 K. Fischer
the communication partner’s presentation in other uses. This is the interpretation I would
like to propose for example (3) above.
Similarly, the topic function of okay can be explained in the same way: okay can either
open, continue, resume, or terminate a topic. Thus, it does not seem to signal something
like “I’m opening/continuing/resuming/terminating the topic”, because this would be
redundant information, obvious from the same cues that would allow the disambiguation
of the signal. The proposal made here is that okay only signals that the topic to be talked
about has been agreed upon, that what is going to follow is in accordance with the
supposed jointly acknowledged topic structure. This corresponds to Condon’s (2001)
finding that okay marks unmarked, default, discourse structures, and to Bangerter et al.’s
(2003: 20) proposal that okay is a device “for seeking or giving consent to a proposed
joint undertaking.” In the case of opening a new topic, we have seen that it is used when
the topic has been previously agreed upon or is situationally determined (as in the task-
oriented dialogues investigated here and in Condon, 1986, 2001). In the case of
continuing a topic, the acknowledged topic is the current topic, and in the case of
terminating a topic, what is jointly achieved is the fact that it can be terminated (Schegloff
and Sacks, 1973).
How about the function of okay as a repair marker and without partner orientation then, as
it has been found in human-computer communication? Regarding speech management
functions, the use of okay signals that after what has been said by the communication
partner, it is taken to be obvious that the previous strategy needs to be abandoned and a
new one has to be initiated. That is, the relation of accordance applies to the agreement
between the communication partners that one should not continue like this. The domain of
reference is the domain of speech management, including speech planning and linguistic
strategy selection (see Fischer, 2000a). In the action reading exemplified by Grosz (1982),
it would signal accordance between the action requested and the action carried out, thus
yielding the reading of completion of the requested action. So even the situation-specific
readings of okay can be explained on the basis of the underlying meaning proposed.
We can therefore conclude that to postulate a relationship of accordance as the core

meaning of okay may account for all of its different uses. This core meaning can not only
be assumed for the discourse particle okay, but also for its uses as an adverb. Thus, the
question “fkcf_5_05: would one thirty be okay with you?” can be understood as the
question for agreement on the content level, i.e., as a question whether two states of
affairs, here free time slots in the participants’ calendars, are in accordance with each
other.
This proposal is in contrast to Condon (2001), who argues that the acknowledgement
reading of okay can be explained on the basis of its function to mark unmarked sequences.
That is, in the same way as Schegloff (1982) has shown continuers to display
understanding by their sequential placement at transition relevance places and by passing
the opportunity to initiate repair, okay is proposed to mark a sequence as in accordance
with one’s expectations. However, such a proposal cannot explain the relationship with
okay as an adjective or adverb, and it reverses the order of grammaticalization proposed
by Diewald (this volume). It will therefore be assumed here that the acknowledgement
reading of okay is the basic one.
If the relationship between the different readings a discourse particle can fulfil is
explained by reference of the same core meaning to different communicative domains,
what is needed then is a list of the different domains to which discourse particles may
refer. What is proposed here is that these domains correspond to those aspects of the
communicative situation to which speakers attend as relevant. This can on the one hand be
shown by conversation analyses of the speaker categories involved, preferably on the
basis of analyses of a broader range of linguistic devices besides discourse particles. On
the other hand, these domains become apparent as being oriented to whether the speakers
can be shown to attend to them differently in different situations, for instance in human-
to-human and human-to-computer communication. For example, we can see that speakers
regularly signal to each other that they perceive and understand each other, especially in
potentially face-threatening situations. That speakers in natural conversation really attend
to their relationship to the communication partner, rather than, say, just marking the
dispreferred utterance, using the discourse particle to take the turn or to play for time,
these functions thus being secondary, is supported by the fact that speakers do not do so
when talking to an artificial communication partner. Consequently, okay serves functions
with respect to the regulation of the speaker-hearer relationship in human-to-human
communication, whereas in human-computer interaction, which is characterized by
communicative problems, speakers attend more to speech management by means of okay.
Thus, the use of comparable corpora allows the identification of communicative domains
as attended to by the participants.
4.2. The ingredients of the model
The model proposed consists of three parts, which interact. Each of the constituents of the
model accounts for a particular aspect of the observations made.
4.2.1. The invariant meaning aspects

The first part of the model is constituted by the invariant meaning of each discourse
particle morpheme. The morphemic meaning accounts for the relationship between the
different readings of the phonological/orthographic form of a particle, even across word
class boundaries. To postulate one core meaning is of course not the only way to account
for this relationship. Several different but related meanings may fulfil the same purpose.
However, it is taken to be an issue of “methodological minimalism” (Foolen, 1993) to
assume as few different meanings as possible.
The proposal for okay is to formulate its morphemic meaning as “after all I know (from X
or about Y) I think we think the same (about Z)”, thus including both the conclusive
meaning aspect and the relationship of accordance. This invariant meaning aspect
accounts for the relationship between the different readings.
Reference to a particular communicative domain accounts for the relationship between the
different readings and the particle morpheme. That is, the morphemic meanings may be
underspecified in a number of respects. Their formulations in NSM may contain place
holders such as something or someone, or they may, like okay, describe a relation of
accordance for which it needs to be specified to what it applies. Thus, the same invariant
meaning may evoke a considerable number of different readings, depending on how the
slots are filled or the reference of the relationship expressed is specified. The different
442 K. Fischer
contextual meanings of okay are created by reference to, for instance, those aspects the
speaker knows about and about which she thinks the same as her communication partner.
In the case of okay, the communicative domains to which it can refer are perception,
understanding, topic structure, the interpersonal relationship, action, speech management,
and content. The different contextual readings resulting are the following:
• after all I know from what I have heard, I think what you said and what I heard is the
same as a signal of successful perception;
• after all I know from what have heard, I think that what you want to say and what I
think you want to say is the same as a signal of successful understanding;
• after all I know I think we think the same about what to talk about as a signal of
accordance regarding the topic agreed upon;
• after all you know, do you think the same? as a search for agreement;
• after all I know I think we think the same as a signal of interpersonal accordance;
• after all I know I think that what you wanted me to do and what I have done is the
same as a signal of completed action;
• after all I know from what I have said and what you have said I think that we think the
same about talking about something new next as a speech management marker;
• after all I know I say that we think the same about this as a signal of agreement on
content.
Formulations in NSM are admittedly a bit clumsy (see Wierzbicka, 1995: 104). However,
describing the contextual meanings of discourse particles in this way allows showing the
interaction between invariant meanings and aspects of the context by specifying the
underspecified meanings of discourse particle lexemes with context-specific, yet general
aspects.
4.2.2. The communicative background frame

Now, the model would be quite useless if everything was possible, if the model was
completely unrestricted and the underspecified meaning aspects could be specified with
anything. A model that explains the polysemy of lexical items by reference to particular
communicative domains (e.g., Sweetser, 1990) needs to be so constrained as to allow only
those readings that can be identified as speaker meanings in interaction.4 Consequently,
we need to find a way to identify the domains of reference in a methodologically sound
way. Here we can return to our empirical analyses of our comparable corpora: showing
that speakers in one situation attend to a communicative task but not in another supports
the existence of independent, interactively attended-to communicative tasks which serve
as referential domains for the meanings of discourse particles. Therefore, the
communicative background frame, which specifies the possible range of discourse particle
readings, only consists of those communicative domains to which speakers attend as tasks
they want to fulfil in a particular communicative situation.5
The communicative domains (see also Schiffrin, 1987a; Aijmer et al., this volume; Frank-
Job, this volume) postulated here are furthermore by no means peculiar to this particular
model. For instance, it was proposed in the analysis of okay in the previous section that it
may fulfil functions regarding perceiving and understanding the communication partner.
The same communicative tasks have been found to be attended to by speakers in
telephone directory inquiries (Clark and Schaefer, 1989). Thus, the domains of perception
and understanding (see also Grosz, 1982: 153) can consequently not only be shown to be
relevant in the description of okay, but they have also been used as explanatory concepts
in analyses of other linguistic material and are thus validated on independent grounds.
The communicative background frame thus accounts for the relationship between the
different readings within the word class discourse particle. That is, the communicative
background frame, by combining the functional domains related to aspects of the
communication process itself (such as perception, understanding, speech management)
and to aspects of the direct communicative situation (such as the accompanying actions)
naturally accounts for the functional spectrum of the word class. Other particles display a
different functional range because they refer to other aspects than the communicative
background frame, for instance, the pragmatic pretext in the case of modal particles
(Diewald and Fischer, 1998) or aspects of host utterances (Nyan, this volume) that are
related to aspects of discourse memory (Roulet, this volume) as in the case of connectives.
The communicative background frames are therefore representations of the tasks that
constitute relevant aspects of a particular situation for the participants. Some aspects are
usually relevant in most situations, for instance, successful perception and understanding,
while speech management is particularly attended to in situations in which
communicative problems are likely. The situational descriptions resulting can be best
conceptualized as frames (e.g., Fillmore and Atkins, 1992; see Fischer, 2000a).
4.2.3. Constructions: Form-Meaning Pairs

The third part of the model is constituted by linguistic constructions (Goldberg, 1995;
Fillmore and Kay, 1999), which combine general form and meaning aspects of discourse
particle occurrences. They are general descriptions of the structural contexts in which
discourse particles may occur, including, on the form side, the position with respect to the
turn and the utterance and the intonation contour. On the functional side, it is assumed
that the reference to the respective communicative domains is connected to the structural
position. Thus, the meaning side of constructions determines the communicative function
the respective discourse particle may fulfil.
The constructions are word class-specific but not lexeme-specific. That is, it is assumed
that irrespective of which discourse particle functions as a repair marker, it will have
particular structural and functional properties. Similarly, it is possible to signal perception
and understanding in many different ways (see Clark and Schaefer, 1989). For instance,
corresponding to example (3) of okay, the function to signal successful perception,
understanding, and topic continuity, can also be fulfilled by other discourse particles, such
as hmm, oh, or well. However, all four discourse particles may do so in different ways,
because of their different invariant meaning aspects, for example:
(13) flmb_6_07: well, I have a meeting from, ten am until eleven pm, other than
that I'm free. so, when are you free?
mkps_6_08: hmm upon looking at my calendar, it looks like uh, that day may
not work out so well after all. um when did you say you were free
on Thursday?
(14) fjlv_5_02: either Tuesday afternoon or Wednesday afternoon. what do you

think.
mjcc_5_03: oh Wednesday afternoon sounds good.
444 K. Fischer
(15) mrac_3_07: the only day that's good for me next week would be Wednesday the
third. ah sometime after twelve. between twelve and five.
fcaw_3_08: well I do have some time late Wednesday afternoon.
The construction take-up specifies that turn-initial occurrences with falling or integrated
intonation contours will refer to the communicative domains perception, understanding
and thematic organization, signaling successful perception and understanding and topic
continuity. Okay, indicating agreement on the basis of broad evidence, fulfils these
functions by stating a relationship of accordance between what is said, what is heard and
understood, and what is being talked about, while, say, oh, fulfils the same functions by
indicating the receipt of new information (see Heritage, 1984).6
In this way, the constructions account for the relationship between functional
interpretation and structural position by specifying the structural aspects of the particular
reading and the different communicative domains to which a discourse particle may refer.
The constructions can be specified more or less formally in order to allow an integration
of discourse particles and modal particles in general grammatical descriptions.
A linguistic model should either be in accordance with and support established linguistic
distinctions, or it should shed new light on discussions regarding less well established
categories. In Fischer (2000b) I have argued that the model proposed can be used to draw
a line between semantics and pragmatics and thus contributes to a discussion on the
semantic/pragmatic interface. Furthermore, in that study a new way for understanding the
role of discourse particles with respect to turn-allocation (Sacks et al., 1974) was
proposed.
Here, I would like to outline two further contributions of the model to aspects of general
linguistic interest. First, I would like to comment on the notion of communicative
situation, since I believe that discourse particles can tell us very much about the aspects of
a communicative situation to which speakers attend, thus supporting a CA analysis of
context (e.g., Schegloff, 1997). Second, I would like to address the question of how
discourse particles contribute to matters of politeness.
5.1. Discourse particles as contextualization cues
The situations in which communication takes place have a strong influence on the
linguistic properties of the utterances that occur. Sociolinguistics, register theory, and text
linguistics are, for instance, concerned with describing such relations. However, both
inter- and intrapersonal variation in stable situations show that situations cannot be
defined by external criteria (Fischer, 2000c). What speakers consider the situation to
consist in needs to be indicated to their communication partners in order to provide the
information necessary for the interpretation of their utterances (cf. Gumperz, 1984).
One such contextualization cue can be the use of discourse particles. As we have seen in
the previous section, discourse particles get their functional interpretation in reference to
particular communicative domains. The functional spectrum of a set of discourse particles
used in a given situation thus depends on, and at the same time indicates, the
communicative tasks the speakers take to be relevant. In human-computer interaction, for
example, the speaker-hearer relationship is not much attended to, whereas more weight is
put on speech management. The communicative domains attended to thus determine the
situation definition and the functional spectrum a discourse particle may fulfil. Situation
definition, the communicative background frame, and the functional spectrum of
discourse particles are thus intimately related. This is the reason why a frame-based
representation seems more useful than, for instance, a once-and-for-all definition of
discourse planes (see Schiffrin, 1987a; but also Fischer, 2000a).
In line with Argumentation Theory, discourse particles are thus believed to contribute to
the construction of context (see Nyan, this volume). In particular they do so by claiming
to display (Fischer, 1999) the speaker’s mental processes. In the case of the items
discussed, it is both the speaker’s (locuteur) and the producer’s (producteur empirique)
voices that constitute the context.7 By bringing into play the speaker, the hearer, and the
relationship between both, discourse particles anchor the speech produced in the situation.
Thus, discourse particles make the human interlocutors (with their hopes, fears, desires,
and imperfections) part of the situation. It is in this way that they contribute to the
authenticity of speech (see Weydt, this volume). The elements of the speech situation
indexed are, however, taken as given information. Discourse particles, like modal
particles, do not present new information for the communication partner to consider
(Clark and Schaefer, 1989). On the contrary, the general pragmatic function of discourse
particles can be held to mark a contribution by the speaker as noninitial, as grounded in
the utterance situation. That is, by relating the current utterance to some aspect of the
communicative context, they minimize the speaker’s role in the contribution by presenting
the utterance as a natural consequence of the already given situation.
The idea that discourse particles are anchors in the utterance situation can also be found in
other analyses (see Aijmer et al., this volume), yet often less explicitly so. For instance,
Östman’s (1983) analysis of you know, but also Merritt’s (1980) analysis of okay and
Weydt’s (this volume) analysis of German denn, imply such a function. For instance,
Weydt paraphrases a sentence containing the modal particle denn: “I’m asking you [for]
your name because something within the situation motivates me to ask this question”.
Thus, in this paraphrase the current utterance is presented as a natural consequence of the
utterance situation.
Discourse particles may thus indeed guide the interpretation of utterances, yet not by
providing processing instructions but by constituting the situation and thus providing the
interpretative frame of the utterance. They do so by contextualizing the speaker, her
mental processes such as perception, understanding, or attitude, by indicating the role of
the communication partner and the relationship between the participants, and by
displaying the focus on particular tasks, such as extralinguistic action or speech
management. Thus, discourse particles are signs that contribute to the construction and
negotiation of context. Understanding the signals displayed as voices of the speaker
allows us furthermore to relate the pragmatic mechanism proposed for the interpretation
of discourse particle occurrences to approaches describing connecting devices or text
relation markers. Because of the voices displayed in the host utterances, the argumentative
nature of connectives is more obvious. However, as oppositely oriented discourse particle
uses, for example, in dispreferred seconds, show, discourse particles also fulfil
446 K. Fischer
argumentative functions. It is thus furthermore in this sense that discourse particles

contribute to context construction.
5.2. The contribution of discourse particles to politeness
Weydt (this volume) discusses different possible reasons why particles may be understood
as “friendly”. He argues that the reason is not that particles would “down-tone” an
utterance, as the German label Abtönungspartikel suggests, or that there is something in
the particles themselves that would cause this impression. In his view, particles “show
that the actual speaker takes into account his partner’s perspective on the subject, that he
cooperates”. In his words, “they create a network of relationships between the actual
hearer and the actual speaker.” On the basis of the model introduced in the previous
section, we can specify this function now further.
Discourse particles, by displaying mental processes, may serve politeness functions, for
instance, in dispreferred seconds, by showing that the speaker does not reject anything
presented by the communication partner thoughtlessly (Fischer, 1999). Thus, the sheer
playing for time by means of an um (see Levinson, 1983; Smith and Clark, 1993) can
already signal to the communication partner that the speaker is taking care of the other
person’s face needs. The polite effect is even increased if the speaker signals involvement
by means of, for example, the interjection oh. It is thus in this way that discourse particles
“mark” dispreferred utterances (Fischer, 2000b).
Furthermore, discourse particles, by displaying relations of accordance and by referring to

the domains perception, understanding and topic construction, for example, show the
speaker’s attending to the hearer and what she has to say. They allow a glimpse into what
the speaker understands the situation to consist in, how she conceptualizes the flow of
topics, what she perceives and understands. Thus, discourse particles contribute in making
the speaker transparent to the hearer and thus allow a joint construction of the interaction.
Finally, by marking the current utterance as noninitial by relating it directly in the

discourse situation, the speaker minimizes her own contribution to what is being said.
Thus, she can be held less responsible for what is being said; the situation is then taken to
provide the account (Heritage, 1988) of the utterance.
5.3. Conclusion
The study of discourse particles is, contrary to its treatment up to 40 years ago, not at all
peripheral to the concerns of linguistics. The investigation of their functional polysemy
and the factors conditioning their interpretation can be seen as a micro world study of
pragmatic interpretation in general. So while I agree with Pons Bordería (this volume) that
a model of the polyfunctionality of discourse particles should be embedded in a theory of
discourse, I think that the other direction of research, using the investigation of discourse
particles as a starting point, is also helpful. We can now begin to analyze both how the
three analytical concepts developed, frames, constructions, and invariant meanings, are
related to other parts of the vocabulary and what the three notions contribute to linguistic
theory development.
Furthermore, as has been discussed in this paper, the relationship between language and
discourse particles may contribute to the discussion of context, common ground, and
situation, as well as to the discussion of another key notion of pragmatic research:
politeness. The study of discourse particles thus allows a look at much more general
phenomena that constitute core questions in linguistic research.
Notes
1
Transcription conventions: = pause; ? = rising intonation; , = level intonation; . =
terminal intonation.
2
Condon (2001) looks at computer-mediated dialogues. However, the dialogues she
compares differ not only with respect to mediatedness but also with respect to mode:
spoken face-to-face communication versus typed computer-mediated dialogues. The
differences can therefore not be attributed to any particular variable.
3
Transcription conventions: e401 is the speaker ID, s401 the ID of the system, the digits
following number the turn. : pause, : breathing event; <%> = unintelligible
speech; ? = rising intonation; , = level intonation; . = terminal intonation.
4
The flexibility of the model is also intended to account for the dynamics of the discourse
particle meanings through time, as demanded by Hansen (this volume), see also
Diewald (this volume).
5
This may still allow more possible than actual meanings since some discourse particles
do not refer to all communicative domains, and there are clusters of readings that
regularly co-occur. The model therefore has to be restricted further by the third part of
the model, the constructions.
6
For a discussion of well, see Aijmer et al. (this volume).
7
Speech management functions are taken to be interactionally relevant (Jefferson, 1974),
and therefore also the producer’s voice contributes to the context construction.
Chapter 23
Discourse markers in Italian: towards a

“compositional” meaning
Carla Bazzanella
1. Introduction
1.1. State of the art
As an introductory remark, I would like to point out the significance of discourse markers
(henceforth, DMs; “a fuzzy concept” in Jucker and Ziv’s words 1998: 2-3) for the study
of language in context.1 In a pragmatic approach, DMs and their pervasive presence in
naturally occurring data (see, among others, Kwong Luke, 1990: 280) prove to be
precious cues, in a similar vein to what has been called “non-literal meaning” (Searle,
1979), for reaching a better understanding of how language works and how “intended”
meaning (Grice, 1989) is recognized.
Despite wide research interest in the area of so-called DMs from around 1985 (Schourup,
1985; Schiffrin, 1987a; Bazzanella, 1985; etc.) to the present (among others, Lenk, 1998;
Hansen, 1998a; Jucker and Ziv, 1998; Pons Bordería, 1998; Fischer, 2000a; Ferrer and
Pons Bordería, 2001b; and, in a diachronic perspective, Traugott, 1989, 1999b; Onodera,
1995; Brinton, 1996; Fleischman, 1999; Frank-Job this volume; Hansen, 2003; Waltereit,
this volume; Bazzanella, 2001a, b, c; 2003a), a generally accepted definition and unified
treatment of this phenomenon are still lacking. However, the following four points can be
considered as shared (starting from van Dijk, 1979; among others, Foolen, 1996; Hölker,
1991; Kwong Luke, 1990; Jucker and Ziv, 1998; and several contributions in this
volume):
1. DMs do not affect the truth conditions of an utterance, and do not add anything to the
propositional content of an utterance;
2. they are related to the speech situation;
3. they serve to indicate the mood of a sentence, and to express attitudes and emotions;
4. they are multifunctional, operating on several levels simultaneously.
450 C. Bazzanella
Given their characteristics (see section 2.2), the prototypical-model approach seems to be
the most appropriate to deal with DMs; several scholars (Jucker and Ziv, 1998; Pons
Bordería, 1998, this volume; Hansen, 1998a, this volume; Garcea and Bazzanella, 1999;
Cuenca, 2000a, 2001; Bazzanella, 2001b; Bazzanella, 2003c) have adopted it.
1.2. Methodology and data
My current approach has superseded the initial text-linguistics approach, which was
mainly concerned with cohesion problems (see Halliday and Hasan, 1967) and was
influenced by the seminal work of van Dijk (1979) (cf. the use of the label “connectives”
in Bazzanella, 1985). The interactional function, presented in Bazzanella (1990), has been
developed, together with the metatextual function, in a pragmatic perspective where
several parameters, both cotextual and contextual,2 play a role.
The methodology adopted here is mainly qualitative (see also Hansen, this volume), but
grounded on real data, i.e., a corpus of spoken Italian which contains various forms of
interaction.3 Recently, DMs in Old Italian have been analyzed by means of a
computerized corpus (ItalAnt):4 while the general issues have been confirmed, the
metatextual level (which was less relevant in the preceding corpus based on spoken
Italian) needed to be specified in more detail, and some other microfunctions were added
(see below).
A further analysis of Italian DMs is planned for the near future by using recent Italian
corpora, namely, AVI and IPAR.5
2. Definition
2.1. Problems
Terminology is in no way irrelevant to categorization; this is even more the case with the
class in question, which is not fixed by classical grammatical tradition and which should,
in my opinion, be defined functionally.
Many similar terms have been used to cover the same (or partly the same) class; in
English, for example, the following terms have been used: pragmatic markers, discourse
markers/particles, (phatic) connectives, pragmatic expressions, utterance particles.
As Jucker and Ziv (1998: 1) state, “the terminological diversity reflects both the wide
range of linguistic approaches that have been employed for their study, and the
multiplicity of functions which these elements are said to fulfill.”
Labels differ not only according to different national traditions,6 but also within the same
country, depending on the approaches, scholars, and period.7 They are not significant per
se, but only in their correspondence with the phenomena taken into account, which affect
inclusion and exclusion from the category. In other words, labels depend both on the
parameters of inclusion and on what is dealt with in the category: e.g., the number of DMs
in Cantonese Chinese may increase from 30 to 100 (cf. Kwong Luke, 1990).8
DMs in Italian 451
The term discourse marker, in my opinion, would highlight their functions better, be more
inclusive (also with regard to the discussion of similar phenomena in several languages),
be more adequate both for the different grammatical units which can be used in this
function (adverbs, conjunctions, verbal syntagms, etc.), and for the kind of linguistic
items which are not as “light” as a particle.
2.2. Characteristics
2.2.1. A functional class

DMs, unlike grammatical categories (such as verbs, nouns, adverbs, etc.), are identified
not on grammatical but on functional grounds (see Ariel, 1988; Pons Bordería, 2001b;
among others). In other words, various grammatical entities (such as adverbs, verbal
syntagms, interjections; e.g., well, you know, ah) can be used as DMs, thus creating a
highly heterogeneous class.
We could speak of DMs’ “transverseness” with regard to other grammatical categories;

the class of DMs is not grounded on fixed morphological and syntactical features but on
their contextual use in a given text (both written and spoken).
Since van Dijk (1979), a first attempt has been made to distinguish between different uses
of the same item, that is, its pragmatic and semantic uses. In proper semantic uses (e.g.,
ma ‘but’ in (1), the title of an old Italian movie), the item cannot be considered as a
discourse marker, but as a connective, with the same truth-value as the additive
conjunction e ‘and’, plus a conventional implicature (i.e., contrast/ opposition, Grice,
1989):
(1) Poveri ma belli

Poor but beautiful.
On the contrary, in (2), uttered by a teacher to noisy students, the opposition which is
conveyed by ma does not refer to two contrasting semantic values (or what are considered
to be, such as poveri/belli) but to the teacher’s request (i.e., stare zitti ‘be quiet’) with
regard to the actual utterance situation that s/he finds hard to bear (see point 3 of the
general features of DMs which the majority of scholars agree with, namely, they serve to
indicate the mood of a sentence, and to express attitudes and emotions).
(2) Ma state zitti!
The connection established by ma in (2) does not relate two linguistic items, as in (1), but
a linguistic act and its contextual components (see point 2 of the above-quoted features:
they are related to the speech situation). Ma in (2) seems to be a device used to reinforce,
on the one hand, the illocutive force of the utterance (see Bazzanella, Caffi, Sbisà, 1991),
by stressing the students’ failure to do their duty (to be quiet) and the teacher’s power to
enforce it (the use of the imperative: “be quiet”), and on the other hand, to mitigate the
commanding force of the pure imperative, thus fulfilling several, sometimes opposite,
functions (see point 5 of the above-quoted features, namely, they are multifunctional,
operating on several levels simultaneously).
452 C. Bazzanella
In the case of pragmatic uses, and of DMs in general, the semantic correspondence
provides only one clue to the interpretation (see point 1 above: DMs do not affect the truth
conditions of an utterance, and do not add anything to the propositional content of an
utterance), and the real or complex value has to be established on the basis of a number of
contextually and cotextually relevant parameters (that is, its compositionality, see section
4).
The compositionality of the pragmatic meaning of DMs becomes very clear when
translating them into other languages; this is an extremely delicate task, much more so
than with respect to other parts of discourse: “they pose important problems for
translation since they are language-specific in form and, mostly, in use: many languages
share identical or similar forms or word-formation processes but the conditions of use of
the interjections are not the same” (Cuenca, forthcoming). Their close discourse-
boundness and multifunctionality (which can be activated simultaneously by a given DM)
affect the choice of an equivalent in the target language: a translator must vary the
translated version according to the context in order to attempt to preserve the
multifunctionality of the DM in question. The difficulty of preserving the various
functions performed by a DM in its original context is evident, and at best the
predominant form is chosen (when the translation works; see, for example, Baker, 1992,
with regard to idioms; Stede, 2000); Aijmer and Vandenbergen (forthcoming); Cuenca
(forthcoming); Flores Acuña, 2004; Bazzanella and Morra, 2000; Aijmer, Foolen, and
Simon-Vandenbergen, this volume). When translating DMs, where “a substantial amount
of inferencing over and above their decoding” (Ariel, 1998:250) is required, there are two
main risks: over- and under-determination.
In translating (2), for example, we could simply omit ma, which is not crucial for the
semantic meaning (see 2.2.2), but this deletion would entail underdetermining the
translation by impoverishing it on a pragmatic level:
(2a) Be quiet!
The introduction of please, as in (2b), would be more polite than ma in Italian but would
not preserve the contextual reference which is conveyed by ma:
(2b) Please, be quiet!
Other translations would involve overdetermination and the choosing of one particular
function: on the one hand, as in (2c), which, given the social roles at issue, is too polite;
on the other hand, as in (2d) and (2e), which would express the teacher's exasperation, by
making the performative explicit, ((2d)—this is not very common), and by being very
direct and rather face-threatening (2e):
(2c) I would like you to be quiet
(2d) I order you to be quiet
(2e) Shut up!

DMs in Italian 453
In order to explain this complex issue better and to show how much a part it is of analysis,
it could be useful to look at another example of translating DMs from our Italian data. Let
us refer to the Italian translation of well in a literary text, Brothers and Sisters, by Ivy
Compton Burnett (Fratelli e Sorelle, published by Garzanti and Milan (1982), translated
by A. Micchettoni). Well has an Italian semantic equivalent, bene; the functions of these
expressions, when used as DMs, differ (see Bazzanella (1999) for a review of the several
studies on it). The main differences between English and Italian are that in Italian, bene
cannot be used as a turn-taking device, and it cannot carry out partial or complete
disagreement. In cases of disagreement, Italian resorts to be', which has lexicalized this
particular meaning (Bazzanella, 2001c: 93), whereas the English well contains both of
these opposite functions (see Svartvik, 1980; Owen, 1981; Carlson, 1984; Jucker, 1993;
Schiffrin, 1985; Watts, 1986, 1989; among others)).
In the quoted, accurate, Italian translation, we found clear cases of both under- and
overdetermination (Bazzanella and Morra, 2000). Thirty-nine percent of the occurrences
of well are deleted in the Italian translation, as in (3):
(3) Well, my children, I couldn’t have left you alone tonight.

Figli miei, non avrei potuto lasciarvi soli questa sera.
The Italian semantic equivalent of well, bene, is only used in 11.5% of the cases, as in (4):
(4) Well, you will have it as you will.

Bene, sarà quel che vorrete.
In 49.5% of the cases, there are 46 different functional correspondences, such as be’ in
(5), certo ‘sure’ in (6), allora ‘then’ in (7), and—even— bentornato ‘welcome’, which
makes the reproach for being late more mitigated in Italian, in (8):
(5) And Tilly says she will never like any one as much as him, even though she
marries somebody else. And Latimer says the same, don’t you Latimer?
Well, not quite the same, Father.
[...] Be’, non proprio lo stesso, papà.
(6) Don’t? why, what is the matter with all of you? Oh, you find it upsetting, do you?
Well, it is enough to knock you all out; it is indeed.
Non dirlo? Che cosa vi succede? Ah, lo trovate emozionante, vero? Certo,
basterebbe a mettervi tutti fuori combattimento; sicuro.
(7) Well, it is this, Robin.

Allora, è questo, Robin.
(8) Well, my dear, how late you are!

Bentornato caro. Come sei in ritardo!
Similar results are found by Cuenca (forthcoming) in the Spanish and Catalan dubbed
versions of the film Four Weddings and a Funeral; with regard to what she labels
interjections; the following types of translation are found:
454 C. Bazzanella
• literal translation (strategy a)

• non-literal translation
• interjection with dissimilar form but the same meaning (strategy b)
• non-interjective structure with similar meaning (strategy c)
• interjection with a different meaning (strategy d)
• omission (strategy e)
• addition of elements (strategy f)
Interestingly, “literal translation is not the most frequent strategy, which indicates that a
good translation should generally avoid it” (Cuenca, forthcoming).
In this paper, the literal English translation of an item used as a DM in Italian will be
provided for the sake of clarity and will be followed, after a slash, by a functional
correspondence.
A caveat: In the case of lack of contextual parameters, i.e., when DMs are dealt with
without referring to a specific corpus, the functional correspondence provided is
necessarily approximate.
2.2.2. Externality to propositional content and inter-replaceability

In contrast with several scholars, starting from van Dijk (1979) (see also Hoelker, 1991;
Kwong Luke, 1990; Yucker and Ziv, 1998; among others), who maintain that DMs
possess semantic cancellability, and/or the total absence of semantic content, I prefer to
speak about their externality to propositional content.9
In my opinion, the central semantic value is conserved in a given DM, even though by
using it in a particular co-text and context, it takes on other, mainly modal, values; it is the
core meaning (such as the contrast-feature in ma, see (1) and (2)) which allows for a
plurality of uses to come into play in relation to the linguistic and extralinguistic
context.10
DMs do not directly contribute to utterance truth-conditions, but affect the complex of the
utterance values by establishing its pragmatic meaning, which should be considered as
compositional (see section 4).
Their (partial) inter-replaceability is related to their externality to the propositional

content: DMs which differ with regard to their meaning can paradigmatically occur in a
given context if they fulfil the same function: e.g., vero? ‘true?/really?’,11 is replaceable
by no? ‘no?’ (or ‘tag question’) as a rhetorical request for confirmation; praticamente
‘practically/I mean’ is replaceable by voglio dire ‘I mean’ as a mitigating device, etc.
This functional inter-replaceability is responsible for the multifarious uses in translation

we referred to before with regard to the 49.5% of cases of translating well with 46
different functional correspondences.
2.2.3. Distributional and prosodic features

The fact that in several languages DMs take initial position has been pointed out many
times.12 However, there are some languages, such as Cantonese Chinese (Kwong Luke,
1990), where the unmarked position is the final one. In Italian, the variability is complete:
DMs in Italian 455
some DMs, such as diciamo ‘let’s say’ ecco ‘here it is’, can occur at the beginning, in the
middle, or at the end of an utterance.
By varying their position, their function may vary in covariance with other parameters
such as prosody; for example, diciamo ‘let’s say’ can be used both as a mitigating device
(such as in (9), where it works as a politeness device; Bazzanella, 2003b) and a boosting
device (such as in (10), where it is phonetically emphasized), depending also on the
intonation (Bazzanella, 1995: 250-1; Waltereit, this volume).
(9) Visto che sei, diciamo, in pensione, ma continui a dare una mano, non so, a
persone che te lo chiedono.
Since you are, let’s say, retired, but you go on helping, I don’t know/you know,
people who ask you.
(10) Siamo stati veramente duri, DICIAMOLO

We’ve been extremely severe, LET'S SAY/ LET’S BE HONEST
DMs can also occur together in two different functional forms. They can occur as
clusters,13 labeled cumuli by Bazzanella (1995): these are a sequence of DMs, each of
them fulfilling a different function, as in (11), taken from a broadcast program, where ma
works as a turn-taking device (see 1 in Table 2 below) and guardi as an attention-getting
device (see 3 in Table 2):
(11) Ma guardi, io eh quello che posso dire è questo [. . .]

But/Well listen, I eh what I can say is . . .
They can also occur as chains, labeled catene by Bazzanella (1995): these are a sequence
of DMs, each of them fulfilling the same function, usually as fillers (see Table 2), as in
(12), taken from an oral examination, and in (13), taken from a meeting between
university professors:
(12) A. Mi parli dei neogrammatici!

B. Sì (-) dunque allora i neogrammatici cioè [silenzio]
A. Tell me something about neogrammarians!
B. Yes (-) so then neogrammarians that’s to say [silence]/ Yes (-) well
hmm neogrammarians I mean
(13) Ecco, cioè, voglio dire, non sono del tutto d’accordo
Here it is/Well, that’s to say, I mean, I don’t completely agree
The distributional features of DMs (i.e., the variability of syntagmatic position, the
occurrence both of clusters and chains) seem to be related to their semantic externality.
A typical prosodic profile is that of parenthetical intonation or “intonational

independence”, but other contours are possible both in Italian and other languages (for
Spanish, see Pons Bordería, 2001b: 229), and seem to be accountable for, once again, in a
“family resemblance framework” (Pons Bordería, 2001b: 229).
456 C. Bazzanella
To take the variability of syntagmatic positions into account (see Cuenca, 2001: 215 for
Spanish), the prototypical approach should be referred to in order to provide a picture of
distributional features which include non-initial DMs.14
Discourse markers are items external to propositional content which are useful in
locating the utterance in an interpersonal and interactive dimension, in connecting and
structuring phrasal, inter-phrasal and extra-phrasal elements in discourse, and in
marking some on-going cognitive processes and attitudes.15
3. Interactional, metatextual and cognitive functions
As we have repeatedly stated, DMs are multifunctional, that is, they operate on several
levels simultaneously; two aspects of their multi- or polyfunctionality should be
considered:
• paradigmatic (in absentia):16 the same DM fulfills different, even opposite, functions
in different contexts (see well in 2.2.1, and diciamo in 2.2.3), according to position,
intonation, and other phenomena which co-occur in the pragmatic compositionality
(see below);
• syntagmatic (in presentia): several functions are performed by a DM in a given text,
be it written or spoken.17 See, for example, well in (7) where it works as a turn-taking
device, an inference marker, and a frame device:

Many scholars agree upon the multifunctionality of DMs, but there is no general
agreement on the specification of the various functions involved.18
For the sake of analysis, a general taxonomy which encompasses three macro-functions
will be proposed here: interactional (which includes conversational),19 metatextual, and
cognitive.20 These macro-functions have been subdivided into various microfunctions,
which are listed in Tables 1, 2, and 3. This list is not meant to be exhaustive but is merely
intended to outline the wide range of possibilities DMs can exploit in Italian (see
Bazzanella, 1985, 1990, 1994, 2001b, 2001c, 2003c for different analyses grounded on
real data) and is proposed for comparison with other languages.
Table 1. Cognitive functions of DMs
1. Procedural markers (related to cognitive processes, e.g., inference)

2. Epistemic markers (related to speaker’s subjectivity and commitment)
3. Modulation devices (related to propositional content and illocutionary force)
DMs in Italian 457
Table 2. Interactional functions of DMs21
Speaker Addressee
1. Turn-taking devices 1. Interrupting devices
2. Fillers 2. Back-channels
3. Attention-getting devices 3. Attention confirmed
4. Phatic devices 4. Phatic devices
22
5. Hedges and boosters 5. ––––––––
6. Checking comprehension 6. Comprehension confirmed; requests for
clarification
7. Requesting agreement, 7a. Agreement, confirmation, support
confirmation 7b. Partial or complete disagreement
8. Yielding the turn 8. ––––––––
Table 3. Metatextual functions of DMs
1. Textual markers
1.1 Structuring the parts
1.1.1 Introduction (as a frame device)
1.1.2 Transition
1.1.3 List
1.1.4 Digression
1.1.5 Ending
1.2 Quotation and indirect speech markers
2. Focusing devices
2.1. Local
2.2. Global
3. Reformulation markers
3.1. Paraphrase markers
3.2. Correction markers
3.3. Exemplification markers
458 C. Bazzanella
4. Meaning activation
The potential spectrum of the functions fulfilled by DMs is wide, as we have seen above.
How can the speaker choose, and the interlocutor/reader recognize, the intended meaning
of a particular DM in a given text?
The particular value of a DM is activated according to the co-occurrence of cotextually

(textual, paralinguistic, and gestural) and contextually (sociolinguistic, pragmatic,
emotive) relevant parameters.23 The indexical nature of DMs, that is, their relatedness to
the ongoing speech act (already stressed by van Dijk in 1977; see Frank-Job, this volume,
for specific aspects) is at issue here.
Let me briefly list the cotextual parameters, which include:
• Textual components: syntagmatic position, anaphoric and cataphoric relations,24, 25

coherence devices (e.g., Knott and Sanders, 1998; Hansen, 1998a; Lenk, 1998). In
other words, the context, in its sequential development, is crucial for activating the
specific meaning (see Roulet, this volume, with regard to text relations);
• Paralinguistic components, i.e., prosody (Schiffrin, 1987a; Gupta, this volume;
Hansen, this volume), variations in F0 (Fundamental Frequency), and other phonetic
correlates which are relevant to emotional states (for the general issue, see Scherer,
1986; Magno Caldognetto, 2002), and variations in volume (Bazzanella, 1995);
• Gestural components: different kinds of gestures can accompany the utterance of DMs,
thus (partially) affecting their value (e.g., in irony).
Contextual parameters include sociolinguistic components, i.e., related to space, time,26

social identity, style (see also register and channel variables (spoken/written discourse;
face-to-face conversations; broadcast programs, and their different settings and contextual
constraints; cf. Bazzanella, 2002).27 The most relevant ones seem to be kind and goal of
interaction, textual genre,28 social roles (Preisler, 1986), age (Thorne and Henley, 1975),
sex (see Bazzanella and Fornara, 1995 for both a review and some Italian data), variety of
language used, language background of both speaker and addressee(s) (Gupta, this
volume), ethnicity (Gupta, this volume), etc. To quote an example of a DM related to a
geographic or regional parameter (i.e., Piedmont, Lombardia, Canton Ticino; varietà
diatopica in Italian, see note 27): piuttosto che ‘rather’ is used as a disjunctive more than
a comparative marker, as in (14), taken from a telephone call. In cases like this, piuttosto
che works as an introduttore neutro di alternative divergenti (a ‘neutral marker of
divergent alternatives’; cf. Bazzanella and Cristofoli, 1998), and does not establish a
comparison between two contrasting concepts, as it usually does.
(14) A. Che cosa fai domani?

B. Non lo so. Vado in montagna piuttosto che stare a dormire piuttosto che
vedere qualcuno. Sono troppo stanca per decidere adesso.
A. What are you doing tomorrow?
B. I don’t know. Maybe I’ll go to the mountains rather than sleeping rather than
seeing somebody. I’m too tired to make up my mind now.
Though regionally marked, this form is becoming common in spoken Italian, mainly in
internet relay chat (IRC) and television and broadcast programs, which are more prone to
DMs in Italian 459
innovations and changes. In general, DMs are particularly inclined to change, not only in
young people`s jargon, but at any age, following some common trends (labelled facile
‘deperibilità’ by Bazzanella, 2001c). To exemplify, in just a few years, the common use
of niente’nothing’ as a filler has been replaced by non esiste ‘it doesn’t exist’.
The diachronic variable is particularly interesting in relation to the change of meaning and
functional roles in which actual DMs have been involved in a process of
grammaticalization and modal shift which has been studied in several languages, with
regard to several DMs: for example, French puis, alors (Hansen, 1998a); English indeed
(Traugott and Dasher, 2002), _a ‘then’ (cf. Brinton, 1996: 5), anon ‘at once, soon’
(Brinton, 2000); Greek lipòn ‘so, then, therefore, consequently’ (Georgakopolou and
Goutsos, 1998); Latin contra (Orlandini, 2003); Italian allora (Bosco and Bazzanella,
forthcoming), anzi (Bazzanella, 2003a), etc.
• Pragmatic components: speech acts, presuppositions, implicatures, (Grice, 1989;

Levinson, 1983; Sweetser, 1990), mutual/shared knowledge and “encyclopedia”,
conventionalization (Frank-Job, this volume; Hansen, this volume), etc. The functional
value of a DM may change according to the performed speech act, the various degrees
of conventionalization, and both the speaker’s intentions and the addressee’s
recognition of them (in fact, misunderstandings are likely to occur in everyday
interactions; see Bazzanella and Damiano, 1999); furthermore, shared knowledge may
be responsible for holophrastic uses of DMs, such as in (15), taken from Bazzanella
(1995):
(15) E’ morto mio padre, allora . . . capirai. . .

My father has passed away, then/well . . . you’ll understand/you know . . .
• Emotive components which affect the linguistic expression. Typically, interjections

are considered to convey emotion, as in the following example, which “shows a
special use of this secondary interjection [=Good Lord] since irony activates both
meanings, the literal and the idiomatic” (taken from Cuenca, forthcoming):
(16) Fiona: What do you do?

Gerald: I am training to be a priest.
Fiona: Good Lord. (FW, 15: 25)
Emotion, a complex phenomenon (Bazzanella and Kobau, 2002; Bazzanella, 2004), can
be displayed at all linguistic levels, and “permeates the entire linguistic system” (Ochs
and Schieffelin, 1989: 22). It motivates the use in a stressful situation, for example, of
chains of DMs (però, cioè, non so ‘but, that is, I don’t know’), of the repeated use of I
don’t know as filler, and of different epistemic markers (penso ‘I think’, m a h
‘goodness/heaven knows’), as in (17), taken from a television interview with an Italian
man who was kidnapped in Libya and has just been released:
(17) Più tardi, però, cioè, non so, il terzo giorno penso che qualche mulo, mah anche là
il mulo può andare bene non so per un’ora o due al giorno, è impossibile poter
andare, non so cinque o sei ore al giorno
460 C. Bazzanella
Later, but/well, that is/I mean, I don’t know, the third day I think that some mules,
goodness/heaven knows, even there mules are ok I don’t know one or two hours a
day, one can’t go, I don’t know, 5 or 6 hours a day
5. Broader perspectives
When considered in a pragmatic perspective, DMs lose their exceptional status,

contributing to the identification of the complex picture which has to be drawn in order to
obtain a complete linguistic analysis and/or architecture.
The analysis of DMs cannot be limited to the sentential level nor to the mere propositional
level. What is said does not usually coincide with what is intended in a Gricean sense, and
in order to understand the latter, one should look for what has been labelled pragmatic
compositionality (Bazzanella and Morra, 2000), which echoes Frege’s Principle of
Compositionality, though in quite a different perspective. Pragmatic compositionality
means to take into account the several parameters which are involved (i.e., the above-
mentioned contextual and cotextual parameters) in the use of a given word, utterance, or
text, in order to understand or recognize the meaning intended by its speaker, utterer, or
author.29
If we go back to the translation problems of DMs (see 2.2.1), which pinpoint some crucial
issues of this complex phenomenon difficult to handle with traditional linguistic
categories, the functional correspondence of a given DM should take the quoted
parameters into account in order not to violate the appropriate conditions of use. To quote
an example already referred to, well, at the beginning of a turn, can perform different
functions:
• as a turn-taking device on the speaker’s side (see 1 in Table 2), which is replaceable in
Italian by many DMs. To list some of them (the list is of course open, and the specific
choice depends on the overall configuration, as for the other following functions):
allora ‘then’, dunque ‘hence’, ecco ‘that is’, ma ‘but’, e ‘and’ (Bazzanella, 1995);
• as a filler device on the speaker’s side (see 2 in Table 2), which is deletable—as we
saw above—or replaceable in Italian also by ehm, diciamo ‘let’s say’, cioè ‘that is’,
niente ‘nothing’, non esiste ‘it doesn’t exist’;
• as a hedge and modulation device on the speaker’s side (see 5 in Table 2 and 3 in
Table 3), which is replaceable in Italian by, be’, in qualche modo ‘in a way’, diciamo,
praticamente ‘practically’;
• as an agreement device on the hearer’s side (see 7a and 7b in Table 2), which is
replaceable in Italian by the semantic correspondence bene, or by other DMs, such as
certo (see (6)), and esatto ‘exactly’ (which is very common nowadays, also uttered by
asymmetric partners, e.g., by a student during an oral examination to a teacher’s
statement);
• as a frame device (see 1.1.1 in Table 3 ), which is replaceable in Italian, for example,
by dunque, allora;
• as a correction marker (see 3.2 in Table 3 ), which is replaceable in Italian, for
example, by cioè, diciamo;
• as a procedural/inference marker (see 1 in Table 1 ), which is replaceable in Italian, for
example, by allora, dunque. As Jucker (1993: 438) states, “The discourse marker well
indicates that the addressee has to reconstruct the background against which he can
DMs in Italian 461
process the upcoming utterance.” In Italian, well in this function is translatable, for
example, by: allora, dunque, insomma ‘after all’;
• as an epistemic marker (see 2 in Table 3 ), which is replaceable in Italian, for example,
by voglio dire ‘I mean’, credo ‘I think’.
The translator's choice between all these virtual possibilities (and others which are not
listed) should respect all the relevant contextual and cotextual constraints, by preserving
as much as possible of the multifunctionality in the target language, as in (7), where
allora performs in Italian the functions fulfilled in English by well (i.e., a turn-taking
device, an inference marker, and a frame device):

A good translation, as is shown by the wide range of correspondences used in the

mentioned text, should be appropriate both to the specific linguistic cotext, that is, the
local functions, and to the global context (i.e., in this case, a literary text, written in a
given time, in the English society, with given characters, social roles, etc.).
In general, I would like to claim that the appropriateness conditions are the ones
spontaneously referred to in the real usage of native language, which is affected by all the
mentioned components, in a complex overall configuration. The same contextual and
cotextual parameters which have been identified as crucial for DMs are relevant for
correctly conveying and interpreting not only non-literal meaning in general (such as
irony, metaphor, etc.), but also literal meaning. In a pragmatic perspective, the literal/non-
literal distinction could be eliminated, and appropriateness conditions could be claimed to
apply to all uses of language, both metaphorical and non-metaphorical (Bazzanella and
Morra, forthcoming).
The “alleged priority” (Recanati, 1995) of literal meaning has been discussed and
superseded: in fact language use does not rest on mere semantic correspondence but
exploits all available means, both verbal and non-verbal (reflected by the increasing
research on multimodality). Recent pragmatic approaches, in moving from intention to
interaction and to the negotiation of meaning (see, for example, Weigand and Dascal,
2001),30 appear to provide the right framework: the “smoothness” and “vagueness” of
even “literal language”, which should be considered a myth more than an ideal,31 requires
contextualization both for language to be produced in a given time, by given people,
according to given goals, and for speakers’ meaning/intentions to be successfully
recognised (cf. Bazzanella and Morra, forthcoming).
DMs are a clear case in point.
Notes
I wish to thank Maria Josep Cuenca, Susan Eerdmans, Jacqueline Visconti, and the co-
authors of this book for discussing a previous version of this paper, and to congratulate
Kerstin Fischer for making it possible to compare and discuss several approaches to DMs
and DPs.
462 C. Bazzanella
1
Discourse obviously refers to both spoken and written texts.
2
The complex topic of context is crucial for the treatment of DMs; see below. For recent,
interdisciplinary approaches to context, see Akman et al. (2001), and Akman and
Bazzanella (2003).
3
In my opinion, any theory should be verified against empirical data. If this is a general
claim, it becomes an unavoidable must when one is speaking of discourse markers (see
Gupta, this volume; Fischer, this volume; Roulet, this volume; and Bazzanella, 1994,
2005; among others). All examples provided here, when not marked otherwise, are
taken from real interaction/corpora.
4
See Bazzanella (2001a) and (2003 b, forthcoming a); see also Renzi and Bisetto (2000)
for a general presentation of the project.
5
See Bosco and Bazzanella (2002) for a presentation of the corpora mentioned in the text.
6
The terminological choice is related to different national traditions, in the sense that
while, for example, in German grammar, discourse particles are prototypically
particles, in Italy we started studying these linguistic phenomena in a perspective
related to connectives.
7
I myself have changed: Bazzanella (1985, 1990) distinguishes phatic/pragmatic
connectives from semantic ones, Bazzanella (1994) analyzes only indicatori fàtici, and
Bazzanella (1995) labels them segnali discorsivi, which corresponds to discourse
markers.
8
The cross-linguistic dimension is a recent and promising field of research in DMs; it will
not be dealt with here (see, among others, Hansen, 1998a; Waltereit, 2001).
9
Some syntactic and textual tests have been proposed by Bazzanella (1995) (namely,
interrogabilità, sostituzione tramite pro-forme, (dominio della) negazione, parafrasi,
domande-eco) to maintain the externality of DMs to propositional content. See also the
degree of syntactic integration (Diewald, this volume), and the extra-syntactic nature
of discourse markers (Gupta, this volume).
10
“[. . .] the assignment of meanings in a particular context does presuppose the existence
of certain underlying properties” (Kwong Luke, 1990: 265-266). See also the invariant
component and the common core meaning in Fischer (2000a, this volume) and
Schwenter (2001).
11
See also for the following correspondences, the above-mentioned caveat.
12
“[. . .] discourse markers are related to initial placement from the perspective of both
binding and unfolding” (Georgakopolou and Goutsos, 1998: 898). “La presencia en
posición inicial es necesaria para el desarrollo de valores textuales o modales, ya que
estos actúan preferentemente sobre el conjunto de la oración, es decir, tienen ámbito
sobre esta” (Pons and Ruiz Gurillo, forthcoming).
13
The proposed clusters differ from traditional clusters, since the possibility of combining
them is variable, at least synchronically: “Like many other particles in the [Cantonese ]
language [. . . the particle] WO can be used in combination with other particles,
forming particle clusters (compound particles)” (Kwong Luke (1990: 199). In the
example given in text as (11), Ma guardi, io eh quello che posso dire è questo, ma
guardi is a cluster/cumulus common both in Old Italian (Bazzanella, 2001a,
forthcoming) and in spoken contemporary Italian (Bazzanella, 2001c).
DMs in Italian 463
14
Kwong Luke (1990: 6-7) himself states: “The problem is that most of these objects that
have been identified as sentence-final particles do not actually occur only at the end of
sentences [. . .] They also occur at the end of ‘smaller’ syntactic units such as clauses
and phrases.” See also Gupta (this volume) in relation to discourse particles of
Singlish.
15
In accordance with Pons Bordería (this volume), I would like to stress that DMs both
create and mark a relationship. While DMs create a relationship when they force us to
find a situation compatible with the meaning of the marker, they mark a relationship
when the idea conveyed by the DM could be expressed/inferred by other means; in this
case, DMs make it (more) explicit.
16
Kroon (1995: 43) distinguishes between this kind of polyfunctionality (category-
internal, i.e., “their apparent variety of modal uses”), and the external one, the so-
called cross-categorial polyfunctionality, i.e., “their property of having one or more
uses that are commoner for other grammatical categories (e.g., adverbs or
conjunctions)”, which I prefer to call categorial heterogeneity (see section 2.2.1).
17
In Fischer's (2000a: 99) words: “[. . .] a discourse particle often fulfils several pragmatic
functions at once.” See also Pons Bordería (2000: 232): “a given connective is able to
perform different functions at the same time.” From both these kinds of
multifunctionality, which means several possible functional correspondences in the
target text, the above-mentioned difficulties in translation (see section 2.2.1) derive.
18
The focalization on one function rather than another is often mirrored in the label that is
chosen to define DMs (see section 2.1).
19
“My results and findings suggest that they form a class of conversational objects whose
functions are primarily conversation organizational” (Kwong Luke, 1990: 286).
20
While interactional and metatextual functions are in accordance with previous works by
Bazzanella (1990, 1995, 2001a, b, c), where a description of these different uses can be
found, cognitive ones have been added recently (see Bazzanella and Baracco, 2004, for
a specific analysis of DMs in inferential processes). With regard to procedural
microfunctions, see, among others, Blakemore (1987); for epistemic ones, see
Sweetser (1990), Gupta (this volume), and Pons Bordería’s (this volume) expression
of the speaker’s stance.
21
The interactional functions are paired in Table 1 (Bazzanella, 1990): the speaker’s and
addressee’s DMs seem to match each other, even though the pairing is usually
displaced in the real stream of speech, so much so that we might speak of an in
absentia pairing.
22
This microfunction is related to the wide topic of modalization. Modalizers, such as
German Modalpartikeln, should be considered as hyponyms of DMs (see also Pons
Bordería, this volume). In my proposal, interactional functions of modalizers, such as
hedges and boosters, which work as social and politeness markers, should be
distinguished from cognitive functions of modalizers which concern propositional
content, speaker’s commitment, and illocutionary force (Bazzanella et al., 1991). Since
the latter refer to the speaker’s subjectivity, they should be registered under cognitive
functions (see below).
23
See Hansen’s (this volume) level of “sequential environment in which an utterance
hosting a discourse marker occurs”, and Bazzanella’s (1998) distinction between local
and global context.
464 C. Bazzanella
24
According to Levinson (1983: 88), “What [DMs] seem to do is indicate, often in very
complex ways, just how the utterance that contains them is a response to, or a
continuation of, some portion of the prior discourse”. See, for example, Ziccolella
(1998), who stresses both the anaphoric and cataphoric functions of well
(“simultaneously both backward and forward looking”) in her paper.
25
Diewald (this volume) underlines the indexicality of DMs.
26
Diachronic studies on DMs have increased in recent years (see section 1.1).
27
In Italian sociolinguistics, the five dimensions of variation which are mentioned in the
text are respectively labeled diatopica, diacronica, diastratica, diafasica, diamesica
(Berruto, 1995).
28
DMs in Aulus Gellius’ Noctes Atticae, analyzed within a prototype taxonomic model by
Garcea and Bazzanella (1999), differ according to the textual genre: “Discourse
particles seem to reflect the nature of the text: while in the narrative structure discourse
particles do not occur, in the incipit and quotations metatextual discourse particles
prevail; in commentaries and dialogues interactional discourse particles signal both
agreement and disagreement”.
29
In Bazzanella, Caffi, and Sbisà (1991) the notion of synergies was proposed to account
for the combination of effects produced by different linguistic devices on the resulting
illocutionary force of a speech act.
30
The interactional perspective has already been suggested, among others, by Brown and
Levinson (1987: 2), who underline: “a shift in emphasis from the current
preoccupation with speaker-identity, to a focus on dyadic patterns of verbal interaction
as the expression of social relationships; and from emphasis on the usage of linguistic
forms, to an emphasis on the relation between form and complex inference.”
31
“The notion of literal meaning, then, expresses an ideal of academic discourse rather
than the reality of everyday communication. Word meanings are context-dependent,
imprecise, and variable; there is no theoretical notion that can usefully serve as a
counterpart of our folk concept of literal meaning. [. . .] Literal meaning, in other
words, is a myth: as convenient as it may be, it is, in the final analysis, a fiction”
Leezenberg (2001: 304).
References
Abraham, W., 1991. Discourse particles in German: How does their illocutive force come about? In:
Abraham, W. (Ed.), Discourse Particles: Descriptive and Theoretical Investigations on the
Logical, Syntactic, and Pragmatic Properties of Discourse Particles in German [Pragmatics and
Beyond 12]. John Benjamins, Amsterdam, pp. 203-252.
Aijmer, K., 1997. I think––an English modal particle. In: Swan, T., Jansen Westvik, O. (Eds.),
Modality in Germanic Languages. Historical and Comparative Perspectives [Trends in
Linguistics, Studies and Monographs 99]. Mouton de Gruyter, Berlin.
Aijmer, K., 2002a. English Discourse Particles. Evidence from a Corpus. John Benjamins,
Amsterdam.
Aijmer, K., 2002b. Modal adverbs of certainty and uncertainty. In: Hasselgård, H., Johansson, S.,
Behrens B., Fabricius-Hansen, C. (Eds.), Information Structure in a Cross-linguistic Perspective.
Rodopi, Amsterdam, pp. 97-112.
Aijmer, K., Simon-Vandenbergen, A.-M., 2003. The discourse particle well and its equivalents in
Swedish and Dutch. Linguistics 41, 1123-1161.
Akatsuka, M., Noriko, C., Strauss, S., 2000. Counterfactual reasoning and desirability. In: Couper-
Kuhlen E., Kortmann, B. (Eds.), Cause, Condition, Concession and Contrast. Mouton de Gruyter,
Berlin, pp. 205-234.
Akman, V., Bazzanella, C., 2003. The complexity of context. Journal of Pragmatics, 35, 321-329.
Akman, V. et al., 2001. Modeling and Using Context. Springer-Verlag, Berlin.
Alcina Franch, J., Blecua, J.M., 1975. Gramática española. Ariel, Barcelona.
Alsagoff, L., Ho Chee Lick, 1998. The grammar of Singapore English. In: Foley, J.A. et al. (Eds.),
English in New Cultural Contexts: Reflections from Singapore. Singapore Institute of
Management/Oxford University Press, Singapore, pp. 127-151.
Altenberg, B., 1999. Adverbial connectors in English and Swedish: semantic and lexical
correspondences. In: Hasselgård, H., Oksefjell, S. (Eds.), Out of Corpora. Studies in Honour of
Stig Johansson. Rodopi, Amsterdam, pp. 249-268.
Altenberg, B., Granger, S., 2002. Lexis in Contrast. Benjamins, Amsterdam.
Andersen, G. 2000. Pragmatic Markers and Sociolinguistic Variation: A Relevance-theoretic
Approach to the Language of Adolescents. Benjamins, Amsterdam.
Andersen, G., Fretheim, T., 2000. Pragmatic Markers and Propositional Attitude. [Pragmatics and
Beyond. New Series 79]. Benjamins, Amsterdam.
Andersen, H., 1984. Language structure and semiotic processes. ArbejdsPapirer fra Institut for
Lingvistik ved Københavns Universitet 3. University of Kopenhagen, Kopenhagen.
Anscombre, J.-C., 1989. Théorie de l’argumentation, topoï et structuration discursive. Revue
Québécoise de Linguistique, 18, 3-56.
Anscombre, J.-C., 1995a. La théorie des topoï. Editions Kimé, Paris.
Anscombre, J.-C., 1995b. Topique or not topique: formes topiques intrinsèques et formes topiques
extrinsèques. Journal of Pragmatics, 24, 115-141
Anscombre, J.-C., Ducrot, O., 1976. L’argumentation dans la langue. Langages, 42, 5-27.
Anscombre, J.-C., Ducrot, O., 1977. Deux mais en Francais? Lingua, 43, 23-40.
Anscombre, J.-C., Ducrot, O., 1983. L’argumentation dans la langue. Mardaga, Bruxelles.
Anscombre, J.-C., Ducrot, O., 1989. Argumentativity and informativity. In: Meyer, M. (Ed.), From
Metaphysics to Rhetoric. Kluwer, Dordrecht, pp. 71-87.
466 References
Anscombre, J.-C., Ducrot, O., 1994. La argumentación en la lengua. Gredos, Madrid.

Antos, G., 1982. Grundlagen einer Theorie des Formulierens. Max Niemeyer, Tübingen.
Anward, J., 2000. A dynamic model of part-of-speech differentiation, In: Vogel, P., Comrie, B.
(Eds), Approaches to the Typology of Word Classes. Mouton de Gruyter, Berlin, pp. 3-46.
Ariel, M., 1998. Discourse markers and form-function correlations. In: Jucker, A., Ziv, Y. (Eds.),
Discourse Markers. Descriptions and Theory. Benjamins, Amsterdam, pp. 223-259.
Asher, N. et al., 2001. Cooperativity in dialogue. In: Bras, M., Vieu, L. (Eds.), Semantic and
Pragmatic Issues in Discourse and Dialogue. Elsevier, Amsterdam, pp. 217-245.
Asher, N., 1993. Reference to Abstract Objects in Discourse. Kluwer, Dordrecht.
Asher, N., Lascarides, A., 2003. Logics of Conversation. Cambridge University Press, Cambridge.
Auchlin, A., 1993. Faire, montrer, dire. Pragmatique comparée de l’énonciation en français et en
chinois. Lang, Berne.
Auchlin, A., 1981. Mais hein, pis bon, ben alors voilà quoi! Marqueurs de la structuration de la
conversation. Cahiers de Linguistique Française, 2, 141-160.
Auer, P., 1996. The pre-front field in spoken German and its relevance as a grammaticalization
position. Pragmatics, 6, 295-322.
Auer, P., Günthner, S., 2003. Die Entstehung von Diskursmarkern im Deutschen–Ein Fall von
Grammatikalisierung?. InList 38, http://www.uni-potsdam.de/u/inlist/issues/38/.
Autenrieth, T., 2000. Heterosemie und Grammatikalisierung bei Modalpartikeln. Eine synchrone und
diachrone Studie anhand von eben, halt, e(cher)t, einfach, schlicht und glatt. Unpublished
dissertation, Universität Tübingen.
Baker, M., 1992. In Other Words. A Coursebook on Translation. Routledge, London.
Bakhtin, M.M., 1981. The Dialogic Imagination. Four Essays by M.M. Bakhtin. Holquist, M. (Ed.),
Emerson, C., Holquist, M. (Transl.). University of Texas Press, Austin, TX.
Bangerter, A., Clark, H.H., 2003. Navigating joint projects with dialogues. Cognitive Science, 27,
195-225.
Bangerter, A., Clark, H.H., Katz, A.R., 2003. Navigating joint projects in telephone conversations.
Discourse Processes, 37, 1-23.
Bao, Z., 2001. The origins of empty categories in Singapore English. Journal of Pidgin and Creole
Linguistics, 16, 275-319.
Barth, D., Couper-Kuhlen, E., 2002. On the development of final though: a case of
grammaticalization? In: Wischer, I., Diewald, G. (Eds.), New Reflections on Grammaticalization.
International Symposium, Potsdam, 17-19 June, 1999. [Typological Studies in Language 49].
Benjamins, Amsterdam, pp. 345-361.
Baskaran, L., 1988. Aspects of Malaysian English syntax. Doctoral dissertation, University College,
London.
Bateman, J., Rondhuis, K.J., 1997. Coherence relations: towards a general specification. Discourse
Processes, 24, 3-49.
Batliner, A., Fischer, K., Huber, R., Spilker, J., Nöth, E., 2003. How to find trouble in
communication. Speech Communication, 40, 117-143.
Bauhr, G.1994. Funciones discursivas de bueno en español moderno. Lingüística Española Actual,
16, 79-124.
Bazzanella, C., 1985. L’uso dei connettivi nel parlato: alcune proposte. In: Franchi De Bellis, A.
Savoia L.M., (Eds.), Sintassi e morfologia della lingua italiana d’uso. Teorie e applicazioni
descrittive. Bulzoni, Roma, pp. 83-94.
References 467
Bazzanella, C., 1990. Phatic connectives as interactional cues in contemporary spoken Italian.
Journal of Pragmatics, 14, 629-647.
Bazzanella, Carla (1994). Le facce del parlare. Un approccio pragmatico all’italiano parlato, second
ed. La Nuova Italia, Firenze/Roma.
Bazzanella, C., 1995. I segnali discorsivi. In: Renzi, L., Salvi, G., Cardinaletti, A. (Eds.), Grande
grammatica italiana di consultazione, vol. III. Il Mulino, Bologna, pp. 225-257.
Bazzanella, C., 1998). On context and dialogue. In: Cmejrkova, S., et al, (Eds.), Dialogue in the
Heart of Europe. Niemeyer, Tübingen, pp. 407-416.
Bazzanella, C., 1999. Corrispondenze funzionali di well in italiano: analisi di un testo letterario e
problemi generali. In: Skytte, G., Sabatini, F. (Eds.), Linguistica testuale comparativa. Memoriam
Maria-Elisabeth Conte. Museum Tusculanum Press, Copenhagen, pp. 99-110.
Bazzanella, C., 2001a. Persistenze e variazioni nell’uso dei segnali discorsivi: primi risultati di
un’analisi nell’italiano antico. In: Zsutsanna, F. and Salvi, G. (Eds.), Semantica e lessicologia
storiche. Bulzoni, Roma, pp. 183-206.
Bazzanella, C., 2001b. Segnali discorsivi e contesto. In: Heinrich, W., Heiss, C., Soffritti, M. (Eds.),
Modalità e substandard. CLUEB, Bologna, pp. 41-64.
Bazzanella, C., 2001c. I segnali discorsivi tra parlato e scritto. In: Dardano, M., Pelo, A., Stefinlongo,
A. (Eds.), Scritto e parlato. Metodi, testi e contesti. Casa Editrice Aracne di Roma, Roma, pp. 79-
97.
Bazzanella, C., 2002. Sul dialogo. Contesti e forme di interazione verbale. Guerini e Associati,
Milano.
Bazzanella, C., 2003a. Dal latino ante all’italiano anzi: la ‘deriva modale’. In: Garcea, A. (Ed.),
Colloquia absentium. Studi sulla Comunicazione epistolare in Cicerone. Rosenberg, Torino, pp.
123-140.
Bazzanella, C., 2003b. Discourse markers and politeness in Old Italian. In: Held, G. (Ed.), Partikeln
und Höflichkeit. Peter Lang, Wien, pp. 247-268.
Bazzanella, C., 2003c. Nuove forme di comunicazione a distanza, restrizioni contestuali e segnali
discorsivi. In Maraschio, N., Poggi Salani, T. (Eds.), Italia linguistica anno Mille-Italia
linguistica anno Duemila. Bulzoni, Roma, pp. 403- 415.
Bazzanella, C., 2004. Emotions, language, and context. In: Weigand, E. (Ed.), Emotions in Dialogic
Interaction. Benjamins, Amsterdam, pp. 59-76.
Bazzanella, C., 2005. Linguistica e pragmatica del linguaggio. Laterza, Roma-Bari.
Bazzanella, C., forthcoming. I segnali discorsivi. In: Renzi, L., Salvi, G., Cardinaletti, A. (Eds.),
Italant. Grammatica dell’italiano antico. Il Mulino, Bologna.
Bazzanella, C., Cristofoli, M., 1998. Piuttosto che e le alternative non preferenziali: un mutamento in
atto? Cuadernos de filologia italiana, 267-278.
Bazzanella, C., Damiano, R., 1999. The interactional handling of misunderstanding in everyday
conversations. Journal of Pragmatics, 31, 817-836.
Bazzanella, C., Fornara, O., 1995. Segnali discorsivi e linguaggio femminile: evidenze da un corpus.
In: Marcato, G. (Ed.), Donna e linguaggio. Cleup, Padova, pp. 73-86.
Bazzanella, C., Kobaum P., 2002. Passioni, emozioni, affetti. McGraw e Hill, Milano.
Bazzanella, C., Morra, L., 2000. Discourse markers and the determinacy of translation. In: Korzen I.,
Marello, C. (Eds.), Argomenti per una linguistica della traduzione, On Linguistic Aspects of
Translation, Notes pour une linguistique de la traduction. Edizioni dell’Orso, Alessandria, pp.
149-157.
468 References
Bazzanella, C., Morra, L., forthcoming. Metaphorical truth-conditions, context, and discourse. In:
Burkhardt, A., Nerlich, B. (Eds.), Tropical Truth. Mouton de Gruyter, Berlin.
Bazzanella, C., Caffi, C., Sbisà, M., 1991. Scalar dimensions of illocutionary force. In: Zagar, I.Z.
(Ed.), Speech Acts. Fiction or Reality? IPRA Distribution Center for Jugoslavia, Ljubljana pp.
63-76.
Beaulieu-Masson, A., 2002. Quels marqueurs pour parasiter le discours? Cahiers de Linguistique
Française, 24, 45-71.
Becker-Mrotzek, M., Fickermann, I., 1994. Versicherungsamt. In: Redder, A., Ehlich, K. (Eds.),
Gesprochene Sprache. Transkripte und Tondokumente [Phonai 41]. Niemeyer, Tübingen, pp.
110-135.
Beeching, K., 2002. Gender, Politeness and Pragmatic Particles in French. Benjamins, Amsterdam.
Beinhauer, W., 1978. El español coloquial. Gredos, Madrid.
Beinhauer, Werner (1968). El español coloquial, second ed. Editorial Gredos, Madrid.
Bell, A., 1991. The Language of News Media. [Language in Society 16]. Blackwell, Oxford.
Bell, R., Peng Quee Ser, L., 1983. ’Today la?’ ‘Tomorrow lah!’ the la particle in Singapore English.
RELC Journal, 14, 1-18.
Bellert, I., 1977. On semantic and distributional properties of sentential adverbs. Linguistic Inquiry,
8, 337-351.
Benveniste, E., 1966/1974. Émile Benveniste, Problèmes de linguistique générale, vols. 1, 2.
Gallimard, Paris.
Berger, T., 1998. Partikeln und Höflichkeit im Russischen. Slavistische Beiträge, 375, 29-53.
Bergmann, J.R., 1981. Ethnomethodologische Konversationsanalyse. In: Schröder, P., Steger, H.
(Eds.), Dialogforschung. Jahrbuch 1980 des Instituts für deutsche Sprache. Pädagogischer Verlag
Schwann, Düsseldorf, pp. 9-51.
Berrendonner, A., 1983. Connecteurs pragmatiques et anaphore. Connecteurs pragmatiques et
structure du discourse. Actes du 2ème Colloque de Pragmatique de Génève. Cahiers de
Linguistique Française, 5, 214-246.
Berrendonner, A., 1990. Pour une macro-syntaxe. Travaux de Linguistique, 21, 25-36.
Berruto, G., 1995. Fondamenti di sociolinguistica. Bari, Laterza.
Bestgen, Y., 1998. Segmentation markers as trace and signal of discourse structure. Journal of
Pragmatics, 29, 753-763.
Bestgen, Y., Vonk, W., 1995. The role of temporal segmentation markers in discourse processing.
Bestgen, Y., Vonk, W., 2000. Temporal adverbials as segmentation markers in discourse
comprehension. Journal of Memory and Language, 42, 74-87.
Bhat, D.N.S., 2000. Word classes and sentential functions, In: Vogel, P., Comrie, B. (Eds),
Approaches to the Typology of Word Classes. Mouton de Gruyter, Berlin, pp. 47-64.
Biber, D., Finegan, E.T., 1988. Adverbial stance types in English. Discourse Processes, 11, 1-34.
Biber, D., Johansson, S., Leech, G., Conrad, S., Finegan, E., 1999. Longman Grammar of Spoken
and Written English. Longman, London.
Björklund, M., 2000. Mikhail Bakhtin. In: Verschueren, J., Östman, J.-O., Blommaert, J., Bulcaen,
Ch., (Eds.), Handbook of Pragmatics, second ed. Benjamins, Amsterdam.
Blakemore, D., 1987. Semantic Constraints on Relevance. Blackwell, Oxford.
Blakemore, D., 1989. Denial and contrast: a relevance theoretic analysis of but. Linguistics and
Philosophy, 12, 15-37.
Blakemore, D., 1992. Understanding Utterances. Blackwell, Oxford.
References 469
Blakemore, D., 1996. Are apposition markers discourse markers? Journal of Linguistics, 32, 325-347.
Blakemore, D., 2002. Relevance and Linguistic Meaning: The Semantics and Pragmatics of
Discourse Markers. Cambridge University Press, Cambridge.
Blutner, R., Jaeger, G., 1999. Competition and interpretation: the German adverbs of repetition.
Unpublished manuscript.
Bolinger, D., 1987. Intonation and Its Uses. Stanford University Press, Stanford, CA.
Bolton, K., 2002. The sociolinguistics of Hong Kong and the space for Hong Kong English. In:
Bolton, K. (Ed.), Hong Kong English: Autonomy and Creativity (Asian Englishes Today. Hong
Kong University Press, Hong Kong, pp. 29-55.
Bosco, C., Bazzanella, C., 2002. Contextualization in spoken language corpora. In: Pusch, C.D.,
Raible, W. (Eds.), Romanistische Korpuslinguistik - Korpora und gesprochene Sprache/Romance
Corpus Linguistics - Corpora and Spoken Language. Narr, Tübingen, pp.19-30.
Bosco, C., Bazzanella, C., forthcoming. Corpus linguistics and the modal shift: pragmatic markers
and the case of allora. In: Pusch, C. (Ed.), Romance corpus linguistics. Corpora and historical
linguistics. Narr, Tübingen.
Bouchard D., 1995. The Semantics of Syntax. Chicago University Press, Chicago.
Bresnan. J., 2000. Optimal Syntax. In: Dekkers, J., van der Leeuw, F., van de Weijer, J. (Eds.),
Optimality Theory: Phonology, Syntax and Acquisition. Oxford University Press, Oxford, pp.
334-385.
Bresnan, J., Aissen, J., 2002. Optimality and functionality: objections and refutations. Natural
Language and Linguistic Theory, 20, 81-95.
Brinton, L., 1996. Pragmatic Markers in English. Grammaticalization and Discourse Function.
Mouton de Gruyter, Berlin.
Brinton, L., 2000. The importance of discourse types in grammaticalization: the case of anon. In:
Herring, S.C., Van Reenen, P., Schøsler, L. (Eds.), Textual Parameters in Older Languages.
Current Issues in Linguistic Theory. Benjamins, Amsterdam, pp. 139-162.
Brinton, L., 2003. I mean: the rise of a pragmatic marker. Paper presented at GURT 2003.
Brisand, F., van Rillaer, G., Sandra, D., 2001. Processing polysemous, homonymous and vague
adjectives. In: Cuyckens, H., Zawada, B. (Eds.), Polysemy in Cognitive Linguistics. Benjamins,
Amsterdam, pp. 261-284.
Briz, A., 1993a. Los conectores pragmáticos en español coloquial. I: Su papel argumentativo.
Contextos, 11, 145-188.
Briz, A., 1993b. Los conectores pragmáticos en la conversación coloquial. II: Su papel
metadiscursivo. Español Actual, 59, 39-56.
Briz Gómez, A., 1998. El español coloquial en la conversación. Esbozo de pragmagramática. Ariel,
Barcelona.
Briz Gómez, A., Val.Es.Co, 2002. Corpus de conversaciones coloquiales. Anejo I de Oralia. Arco,
Madrid.
Briz Gómez, A., et al., 1995. La conversación coloquial. Materiales para su estudio. Universidad,
València.
Briz Gómez, A., Val.Es.Co, 2000. Cómo se comenta un texto coloquial. Ariel, Barcelona.
Briz Gómez, A., Val.Es.Co, 2003. Un sistema de unidades para el estudio de la conversación
coloquial. Oralia, 6, 7-61.
Briz Gómez, A., Hidalgo Navarro, A., Pons, S., Portolés, J. et al., forthcoming. Diccionario de
partículas del español.
470 References
Broschart, J., 1997. Why Tongan does it differently? Categorial distinctions in a language without
nouns and verbs. Linguistic Typology, 1, 123-165.
Brown, A., 2000. Singapore English in a Nutshell: An Alphabetical Description of its Features.
Federal Publications, Singapore.
Brown, G., Yule, G., 1983. Discourse Analysis. Cambridge University Press, Cambridge.
Brown, P., Levinson, S.C., 1987. Politeness. Some Universals in Language Use. Cambridge
University Press, Cambridge.
Brown, R., Gilman, A., 1960. The pronouns of power and solidarity. In: Sebeok, T.A. (Ed.), Style In
Language. MIT Press, Cambridge, MA, pp. 253-276.
Bruxelles, S., Ducrot, O., Fradin, B., Nguyen, T.B., Recanati, F., 1982. Justement, l’inversion
argumentative. Lexique, 1, 151-164.
Bruxelles, S., Ducrot, O., Raccah, P.-Y., 1993. Argumentation et champs topiques lexicaux. Cahiers
de Praxématique, 21, 88-104.
Bühler, K., 1934. Sprachtheorie. Die Darstellungsfunktion der Sprache. Gustav Fischer, Jena.
Busquets, J., et al., 2001. La SDRT: une approche de la cohérence du discours dans la tradition de la
sémantique dynamique. Verbum, 23, 73-101.
Bustorf, W., 1974. Riflessioni sui cosiddetti «riempitivi» italiani. In: Medici, M., Sangregori, A.,
(Eds.), Fenomeni morfologici e sintattici nell’italiano contemporaneo. Atti del VO Congresso
SLI, Roma 1972, vol. 1. Bulzoni, Roma, pp. 21-25.
Bybee, J.L., Perkins, R.D., Pagliuca, W., 1994. The Evolution of Grammar. Tense, Aspect, and
Modality in the Languages of the World. University of Chicago Press, Chicago.
Cadiot, A., et al., 1979. Oui mais non mais ou: Il y a dialogue et dialogue. Langue Française, 42, 94-
102.
Cadiot, A., Ducrot, O., Fradin, B., Nguyen, T.-B., 1985. Enfin, marqueur métalinguistique. Journal of
Cadiot, A., Ducrot, O., Fradin, B., Nguyen, T.-B., Vicher, A., 1985. Sous un mot, une controverse:
les emplois pragmatiques de toujours. Modèles Linguistiques VII, 105-123.
Cadiot, P., 1993. Représentations d’objets et sémantique lexicale: qu’est ce qu'une boîte? Journal of
French Language Studies, 1-23.
Cann, R., 1993. Formal Semantics. An Introduction. Cambridge University Press, Cambridge.
Carel, M., 2002. ‘Occupe-toi d’Amélie’ emploi contrastif de mais et illustration. Cahiers de
Carel, M., Ducrot, O., 1999. Le problème du paradoxe dans une sémantique argumentative. Langue
Carlson, L., 1984. Well in Dialogue Games. Benjamins, Amsterdam.
Carston, R., 2001. Conjunction and pragmatic effects. In: Asher, N., (Ed.), The Encyclopedia of
Language and Linguistics. Pergamon Press, Oxford, pp. 692-698.
Carston, R., 2002. Thoughts and Utterances. The Pragmatics of Explicit Communication. Blackwell,
Oxford.
Carter, R., 2003. The grammar of talk: spoken English, grammar and the classroom. In:
Qualifications and Curriculum Authority, (Ed.), New Perspectives on Spoken English in the
Classroom: Discussion Papers. QCA, London, pp. 5-13.
Castell, S., 2000. Truth in Opinion-focused Discourse: the Language of Young Men. Masters thesis,
School of English, University of Leeds.
Castro Caycedo, G., 1999. Colombia X. Planeta, Bogota.
References 471
Chafe, W.L., 1980. The deployment of consciousness in the production of a narrative. In: Chafe, W.
(Ed.), The Pear Stories: Cognitive, Cultural, and Linguistic Aspects of Narrative Production.
Ablex, Norwood, NJ, pp. 9-50.
Chafe, W., 1986. Evidentiality in English conversation and academic writing. In: Chafe, W., Nichols,
J. (Eds.), Evidentiality: The Linguistic Coding of Epistemology. Ablex, Norwood, NJ, pp. 261-
272.
Chafe, W., 1987. Cognitive constraints on information flow. In: Tomlin, R. (Ed.), Coherence and
Grounding in Discourse. Benjamins, Amsterdam.
Chafe, W., 1993. Prosodic and functional units of language. In: Edwards, J. and Lampert, M. (Eds.),
Talking Data: Transcription and Coding in Discourse Research. Lawrence Erlbaum, Hillsdale,
NJ, pp. 33-43.
Chafe, W., 1994. Discourse, Consciousness and Time: The Flow and Displacement of Conscious
Experience in Speaking and Writing. University of Chicago Press, Chicago.
Chafe, W., 2001a. The analysis of discourse flow. In: Schiffrin D., Tannen D., Hamilton, H.E. (Eds.),
The Handbook of Discourse Analysis. Blackwell, Oxford, pp. 673-687.
Chafe, W., 2001b. Prosody and emotion in a sample of real speech. In: Fries P., Cummings M.,
Lockwood D., Sprueill, W. (Eds.), Relations and Functions Within and Around Language.
Continuum, London, pp. 277-315.
Chao, Y. R., 1968. A Grammar of Spoken Chinese. University of California Press, Berkeley, CA.
Chierchia, G., McConnell-Ginet, S., 1990. Meaning and Grammar: An Introduction to Semantics.
MIT Press, Cambridge, MA.
Chodorowska, M., 1997. On the polite function of ¿me entiendes? in Spanish. Journal of Pragmatics,
28, 355-371.
Chouieri, L., 2002. Des limites d’une représentation « reichenbachienne» du temps dans le discours.
Cahiers de Linguistique Française, 24, 73-107.
Cinque, G., 1999. Adverbs and Functional Heads. A Cross-linguistic Perspective. Oxford University
Press, Oxford.
Clancy, P.M., 1985. The acquisition of Japanese. In: Slobin, D.I. (Ed.), The Crosslinguistic Study of
Language Acquisition. Lawrence Erlbaum, NJ, pp. 373-524.
Clark, H.H., 1996. Using Language. Cambridge University Press, Cambridge.
Clark, H.H., forthcoming. Pragmatics of language performance. In: Horn, L. R., Ward, G. (Eds.),
Handbook of Pragmatics. Blackwell, Oxford.
Clark, H.H., Carlson, T.B., 1982. Hearers and speech acts. Language, 58, 332-373.
Clark, H.H., Gerrig, R.J., 1990. Quotations as demonstrations. Language, 66, 764-805.
Clark, H.H., Schaefer, E.F., 1989. Contributing to Discourse. Cognitive Science, 13, 259-294.
Concise Chinese-English Dictionary. Commercial Press, Beijing.
Condon, S., 1986. The Discourse Functions of OK. Semiotica, 60, 73-101.
Condon, S., 2001. Discourse ok revisited: default organization in verbal interaction. Journal of
Cortés Rodríguez, L., 1991. Sobre conectores, expletivos y muletillas en español. Editorial Librería
Ágora, Málaga.
Coseriu, E., 1981. Introducción a la lingüística. Gredos, Madrid.
Coseriu, E., 1982. Textlinguistik. Eine Einführung, herausgegeben und eingeleitet von Jörn Albrecht
[Tübinger Beiträge zur Linguistik 109]. Narr, Tübingen.
472 References
Cotter, C., 1996. Engaging the reader: The changing use of connectives in newspaper discourse. In:
Arnold, J., et al., (Eds.), Sociolinguistic Variation: Data, Theory and Analysis. CSLI, Stanford
University, Stanford, CA, pp. 263-278.
Couper-Kuhlen, E., Kortmann, B., 2000. Cause, Condition, Concession, Contrast: Cognitive and
Discourse Perspectives. Mouton De Gruyter, Berlin.
Croft, W.C., 1995. Intonation units and grammatical structure. Linguistics, 33, 839-882.
Croft, W.C., 1998. Linguistic evidence and mental representations. Cognitive Linguistics, 9, 151-
173.
Croft, W.C., 2000. Lexical and grammatical meaning. In: Booij, G., Lehmann, C., Mugdan, J., et al.
(Eds.), Morphologie. Ein internationales Handbuch zur Flexion und Wortbildung/Morphology.
An International Handbook on Inflection and Word-Formation, vol. 1.1 [Handbücher zur Sprach-
und Kommunikationswissenschaft]. Mouton de Gruyter, Berlin, pp. 257-263.
Cruse, D.A., 1986. Lexical Semantics. Cambridge University Press, Cambridge.
Cruse, D.A., 2000. Meaning in Language. Oxford University Press, Oxford.
Cuenca, M.J., 2000a. Definició i delimitació del concepte de connector. In: Macià, J., Solà, J. (Eds.),
La terminologia a l'ensenyament secundari Graó, Barcelona, pp. 77-90.
Cuenca, M.J., 2000b. Defining the indefinable? Interjections. Syntaxis, 3, 29-44.
Cuenca, M.J., 2001. Los conectores parentéticos como categoría gramatical. Lingüística Española
Actual, 23, 211-235.
Cuenca, M.J., 2002. Els connectors textuals i les interjeccions. In: Solà, J. (Ed.), Gramàtica del català
contemporani.
Cuenca, M.J., forthcoming. Translating interjections for dubbing. Proceedings of the II International
Contrastive Linguistics Conference, Santiago de Compostela, 25-27 d'octubre, 2001.
Damasio, A., 1994. Descartes’ Error: Emotion, Reason, and the Human Brain. G. P. Putnam’s Sons,
New York.
Damasio, A., 2000. Descartes’ Error. Quill, New York.
Dancieger, B., Sweetser, E., 2000. Constructions with if, since, and because: causality, epistemic
stance, and clause order. In: Couper-Kuhlen, E., Kortmann, B. (Eds.), Cause, Condition,
Concession and Contrast. Mouton de Gruyter, Berlin, pp. 111-142.
Dausendschön-Gay, U., Gülich, E., Krafft, U., 1991. Linguistische Interaktionsanalysen. Beiträge
zum 20. Romanistentag 1987. Niemeyer, Tübingen.
Davidsen-Nielsen, N., 1993. Discourse particles in Danish. In: Pre-publications of the English
Department of Odense University No. 69, August 1993.
de Groodt, S., 2001. Die Grammatikalisierung konzessiver Nebensatzkonjunktionen mit ob- im
Frühneuhochdeutschen. University Gent. Unpublished manuscript.
de Saz R.M., 2003. An analysis of English discourse markers of reformulation. Unpublished doctoral
dissertation, University of Valencia.
Deacon, T., 1997. The Symbolic Species. Norton, New York.
Deterding, D., 1994. The intonation of Singapore English. Journal of the International Phonetic
Association, 24, 61-72.
Detges, U., 2000a. Time and truth: the grammaticalization of resultatives and perfects within a theory
of subjectification. Studies in Language, 24, 345-377.
Detges, U., 2000b. Two types of restructuring in French creoles: a cognitive approach to the genesis
of tense markers. In: Neumann-Holzschuh, I., Schneider, E. (Eds.), Degrees of Restructuring in
Creole Languages [Creole Language Library 22] Benjamins, Amsterdam, pp. 135-162.
References 473
Detges, U., 2001. Grammatikalisierung: Eine kognitiv-pragmatische Theorie, dargestellt am Beispiel

romanischer und anderer Sprachen. Habilitation thesis, Tübingen.
Detges, U., Waltereit, R., forthcoming. Grammaticalization vs. reanalysis: a semantic-pragmatic
account of functional change in grammar.
DGD = Datenbank Gesprochenes Deutsch, Institut für deutsche Sprache, Mannheim, accessible at
http://www.ids-mannheim.de/.
Diewald, G.,1997. Grammatikalisierung. Eine Einführung in Sein und Werden grammatischer
Formen. Niemeyer, Tübingen.
Diewald, G.,1999a. Die Entwicklung der Modalpartikel aber: ein typischer
Grammatikalisierungsweg der Modalpartikeln. In: Spillmann, H.O., Warnke, I. (Eds.),
Internationale Tendenzen der Syntaktik, Semantik und Pragmatik. Akten des 32. Linguistischen
Kolloquiums in Kassel 1997 [Linguistik International 1]. Lang, Frankfurt/Main, pp. 83-91.
Diewald, G.,1999b. Die dialogische Bedeutungskomponente von Modalpartikeln. In: Naumann, B.
(Ed.), Dialogue Analysis and the Mass Media. Proceedings of the International Conference,
Erlangen, April 2-3, 1998 [Beiträge zur Dialogforschung 20]. Niemeyer, Tübingen, pp. 187-199.
Diewald, G.,1999c. Die Modalverben im Deutschen. Grammatikalisierung und Polyfunktionalität.
Niemeyer, Tübingen.
Diewald, G., Fischer, K., 1998. Zur diskursiven und modalen Funktion der Partikeln aber, auch, doch
und ja in Instruktionsdialogen. Linguistica, 38, 75-99.
Dik, S.,1968. Coordination: Its Implications for the Theory of General Linguistics. North-Holland,
Amsterdam.
Dobson, W.A.C.H., 1974. A Dictionary of the Chinese Particles. University of Toronto Press,
Toronto.
Doherty, M., 1985. Epistemische Bedeutung. Akademie-Verlag, Berlin.
Dostie, G., de Sève, S., 1999. Du savoir à la collaboration. Etude pragma-sémantique et traitement
lexicographique de t’sais. Revue de Sémantique et Pragmatique 5, 11-35.
Dowty, D., Wall, R., Peters, S., 1981. Introduction to Montague Semantics. Kluwer, Dordrecht.
Du Bois, J.W., Schuetze-Coburn, S., Paolino, D., Cumming, S., 1992. Discourse transcription. Santa
Barbara Papers in Linguistics 4.
Ducrot, O., 1972/1991. Dire et ne pas Dire. 3ème édition. Hermann, Paris.
Ducrot, O., 1980. Analyses pragmatiques. Communications, 32, 11–60.
Ducrot, O., 1983. Opérateurs argumentatifs et visée argumentative. Connecteurs pragmatiques et
structure du discours. Actes du 2ème Colloque de Pragmatique de Génève. Cahiers de
Ducrot, O., 1984. Esquisse d’une théorie polyphonique de l’énonciation. In: Ducrot, O. (Ed.), Le dire
et le dit. Les Éditions de Minuit, Paris, pp. 171-233.
Ducrot, O., 1987. L’interprétation en sémantique: un point de départ imaginaire. In: Ducrot, O. (Ed.),
Dire et ne pas Dire. 3ème édition. Hermann, Paris, pp. 307-323.
Ducrot, O., 1988. Topoï et formes topiques. Bulletin d’Étude de Linguistique Française, 22, 1-14.
Ducrot, O., 1995. Topoï et formes topiques. In: Anscombre, J-C. (Ed.), Théorie des Topoï. Editions
Kimé, Paris, pp. 85-99.
Ducrot, O., 1995. Les modificateurs déréalisants. Journal of Pragmatics, 24, 145-165.
Ducrot, O., 1996. Slovenian Lectures. ISH, Ljubljana.
Ducrot, O., Groupe lamda-l, 1975. Car, parce que, puisque. Revue Romane, 10, 248-280.
Ducrot, O., et al., 1976. Mais, occupe-toi d'amélie. Actes de la recherche en sciences sociales, 6, 47-
62.
474 References
Ducrot, O., Bourcier D., Bruxelles S., Diller A.-M., Fouquier E., Gouazé J., Maury L., Nguyen T.B.,
Nunes,G. Ragunat de Saint-Alban L., Rémis A., Sidar-Iskandar, C., 1980. Les mots du discours.
Editions de Minuit, Paris.
DWB, Neubearb., 1965. Deutsches Wörterbuch. Neubearbeitung. Akademie der Wissenschaften der
DDR, Akademie der Wissenschaften zu Göttingen (Eds.). Hirzel, Leipzig.
Dyvik, H., 1998. A translational basis for semantics. In: Johansson, S., Oksefjell S. (Eds.), Corpora
and Cross-linguistic Research: Theory, Method and Case Studies. Rodopi, Amsterdam, pp. 51-86.
Dyvik, H., 1999. On the complexity of translation. In: Hasselgård, H., Oksefjell, S. (Eds.), Out of
Corpora. Studies in Honour of Stig Johansson. Rodopi, Amsterdam, pp. 215-30.
Ebeling, J., 2000. Presentative constructions in English and Norwegian. A corpus-based contrastive
study. Department of British and American English. University of Oslo.
Edelman, G., 1989. The Remembered Present. Basic Books, New York.
Ehlich, K.,1986. Interjektionen. Niemeier, Tübingen.
ELICOP = Brosens, Veerle et al., 1997-2000. Etude linguistique de la communication parlée.
Katolieke Universiteit Leuven, accessible at http://bach.arts.kuleuven.ac.be/ elicop.
Ernst, T., 2002. The Syntax of Adjuncts. Cambridge University Press, Cambridge.
Fabricius-Hansen, C., 1999. Bei dieser Gelegenheit––on this occasion––ved denne anledningen.
German bei––A puzzle in a translational perspective. In: Hasselgård, H., Oksefjell, S. (Eds.),
Corpora and Cross-linguistic Research: Theory, Method and Case Studies. Rodopi, Amsterdam,
pp. 231-248.
Fasold, R., 1990. Sociolinguistics of Language. Blackwell, Oxford.
Fauconnier, G., 1984. Les espaces mentaux. Minuit, Paris.
Ferguson, C., 1959. Diglossia. Word, 15, 325-340.
Fernandez-Vest, J., 1994. Les particules énonciatives dans la construction du discours. PUF, Paris.
Ferrara, K., 1997. Form and function of the discourse marker anyway: Implications for discourse
analysis. Linguistics, 35, 343-378.
Ferrer M.H., Pons, S., 2001. La pragmática de los conectores y las partículas modales [Quaderns de
Filologia]. Universidad,Valencia.
Ferrer, H., Pons Bordería, S., 2001. La pragmática de los conectores y las partículas modales.
Facultat de Filologia, Universitat de València, Valencia.
Fetzer, A., Meierkord, C., 2002. Rethinking Sequentiality: Linguistics Meets Conversational
Interaction. Benjamins, Amsterdam.
Filliettaz, L., 2002. La parole en action. Nota Bene, Québec.
Filliettaz, L., Roulet, E., 2002. The Geneva model of discourse analysis, an interactionist and
modular approach to discourse organization. Discourse Studies, 4, 369-392.
Fillmore, C.J., 1982. Frame semantics. In: Linguistic Society of Korea (Ed.), Linguistics in the
Morning Calm. Hanshin, Seoul, pp. 111-137.
Fillmore, C.J., 1984. Remarks on contrastive pragmatics. In: Fisiak, J. (Ed.), Contrastive Linguistics:
Prospects and Problems. Mouton de Gruyter, Berlin, pp. 120-141.
Fillmore, C.J., Atkins, B.T., 1992. Towards a frame-based lexicon: the semantics of risk and its
neighbors. In: Lehrer, A., Kittay, E.F. (Eds.), Frames, Fields, and Contrasts. Erlbaum, Hillsdale,
NY, pp. 75-102.
Fillmore, C.J., Kay, P., O’Connor, M., 1988. Regularity and idiomaticity in grammatical
constructions: The case of let alone. Language, 64, 501-538.
Fischer, K., 1998. Validating Semantic Analyses of Discourse Particles. Journal of Pragmatics, 29,
111-127.
References 475
Fischer, K., 1999. Die Ikonizität der Pause. Zwischen kognitiver Last und kommunikativer Funktion.
In: Wachsmuth, I., Jung, B. (Eds.), KogWis99: Proceedings der 4. Fachtagung der Gesellschaft f
ür Kognitionswissenschaft, Bielefeld, 28. September - 1. Oktober 1999. nfix, Sankt Augustin, pp.
250-255.
Fischer, K., 2000a. From Cognitive Semantics to Lexical Pragmatics: The Functional Polysemy of
Discourse Particles. Mouton de Gruyter, Berlin.
Fischer, K., 2000b. Discourse Particles, Turn-taking, and the Semantics-Pragmatics Interface. Revue
de Sémantique et Pragmatique, 8, 111-137.
Fischer, K., 2000c. What Is a Situation? Gothenburg Papers in Computational Linguistics, 00, 85-92.
Fischer, K., Brandt-Pook, H., 1998. Automatically disambiguating discourse particles. Proceedings
of Coling/ACL ’98 Workshop on Discourse Relations and Discourse Markers, Montreal, Canada,
pp. 107-113.
Fischer, K., Drescher, M., 1996. Methods for the description of discourse particles: Contrastive
analysis. Language Sciences, 18, 853-861.
Fischer, K., Nemo, F., 2000. Aber, Mais, But: Integrating Semantics, Grammar and Pragmatics.
Working paper presented at SIC-CSP2000.
Fleischman, S., 1999. Like as discourse marker. Paper presented at Pragma99, Tel Aviv.
Flores Acuña, E., 2004. Analisi contrastiva spagnolo/italiano dei connettivi di riformulazione: studio
di en fin. In: D’Achille, P. (Ed.), Generi architetture e forme testuali. Cesati, Firenze, pp. 349-
362.
Foley, J., et al., 1998. English in New Cultural Contexts: Reflections from Singapore. Oxford
University Press, Oxford.
Foolen, A., 1989. Beschreibungsebenen für Modalpartikelbedeutungen. In: Weydt, H. (Ed.),
Sprechen mit Partikeln. Mouton de Gruyter, Berlin, pp. 305-317.
Foolen, A., 1993. De betekenis van partikels. Dissertation, University of Nijmegen.
Foolen, A., 1996. Pragmatic Particles. In: Verschueren, J. et al., (Eds.), Handbook of Pragmatics.
Benjamins, Amsterdam.
Foolen, A., 2001. Review of Hansen, Maj-Britt Mosegaard. The function of discourse particles. A
study with reference to spoken standard French. [Pragmatics and Beyond, New Series, Vol. 53,
Benjamins, Amsterdam. 1998. xi + 417 pages]. Studies in Language, 25, 349-356.
Foolen, A., Van der Wouden, T., 2002. Introduction. Special issue on ‘Particles’. Belgian Journal of
Fox Tree, J.E., Schrock, J.C., 1999. Discourse markers in spontaneous speech: Oh what a difference
an oh makes. Journal of Memory and Language, 40, 280-295.
Fox Tree, J.E., Schrock, J.C., 2002. Basic Meaning of you know and I mean. Journal of Pragmatics,
34, 727-747.
Franckel, J.-J., 1989. Etude de quelques marqueurs aspectuels du français. Droz Geneva.
Frantext Electronic Corpus, accessible at http://atilf.atilf.fr/frantext.htm.
Fraser, B., 1988. Types of English discourse markers. Acta Linguistica Hungarica, 38, 19-33.
Fraser, B., 1990. An approach to discourse markers. Journal of Pragmatics, 14, 383- 395.
Fraser, B., 1996. Pragmatic markers. Pragmatics, 6, 167-190.
Fraser, B., 1998. Contrastive Discourse Markers in English In: Jucker, A.H., Ziv, Y. (Eds.),
Discourse Markers: Descriptions and Theory [Pragmatics and Beyond 57]. Benjamins,
Amsterdam.
Fraser, B., 1999. What are discourse markers? Journal of Pragmatics, 31, 931-952.
Fraser, B., forthcoming a. Towards an understanding of but in English. Submitted for review.
476 References
Fraser, B., forthcoming b. Sequences of DMs in English. Unpublished manuscript.

Fraser, B., forthcoming c. Review of Blakemore, Relevance and Linguistic Meaning. Journal of
Pragmatics.
Fritz, G., 1994. 10 Grundlagen der Dialoganalyse. In: Fritz , G., Hundsnurscher, F. (Eds.), Handbuch
der Dialoganalyse [Band IV]. Niemeyer, Tübingen, pp. 177-201.
Fuchs, C., 1988. Encore, déjà, toujours: de l’aspect à la modalité. In: Tersis, N., Kim, A. (Eds.),
Temps et aspects. Peeters/SELAF, Paris, pp. 135-148.
Fuentes Rodríguez, C., 1993. Comportamiento discursivo de bueno, bien, pues bien. Estudios de
Lingüística, 9, 205-221.
Fuentes Rodríguez, C., 1993b. Conclusivos y reformulativos. Verba, 20, 171-198.
Fuentes Rodríguez, C., 1993c. Conectores ‘pragmáticos’. In: Alcaide, E., del Mar Ramos, M.,
Salguero, F. (Eds.), Estudios lingüísticos en torno a la palabra. Kronos, Sevilla, pp. 71-104.
Furberg, M.., 1963. Saying and Meaning. Blackwell, Oxford.
Garcea, A., Bazzanella, C., 1999. Vincoli testuali e funzioni dei segnali discorsivi in Gellio. Lingua
e Stile, 34, 89-115.
Gasiglia, N., Nemo, F., Cadiot, P., 2001. Meaning and the generation of reference. In: Bouillon, P.
(Ed.), Generative Approaches to the Lexicon. Université de Genève, Genève.
Gazdar, G., 1979. Pragmatics. Implicature, Presupposition and Logical Form. Academic Press, New
York.
Gelhaus, H., 1995. Die Wortarten. In: Duden: Grammatik der deutschen Gegenwartssprache, fifth ed.
Dudenverlag, Mannheim, pp. 85-398.
Georgakopolou, A., Goutsos, D., 1998. Conjunctions versus discourse markers in Greek: the
interaction of frequency, position, and functions in context. Linguistics, 36, 887-917.
Geurts, B., Van der Sandt, R., 2001. Too. In: van Rooy, R. (Ed.), Proceedings of the 13th Amsterdam
Colloquium. University of Amsterdam, Amsterdam, ILLC, pp. 180-186.
Ghosh, A., 2000. The Glass Palace. HarperCollins, London.
Gili Gaya, S., 1983. Curso superior de sintaxis española. Vox, Barcelona.
Givón, T., 1979. Understanding Grammar. Academic Press, New York.
Givón, T., 1993. English Grammar: A Function-based Introduction, vols. I and II. Benjamins,
Amsterdam.
Goddard, C., 1994. The meaning of lah: Understanding ‘emphasis’ in Malay (Bahasa Melayu).
Oceanic Linguistics, 33, 145-165.
Goddard, C., 2000. Polysemy: A Problem of Definition. In: Ravin, Y., Leacock, C. (Eds.), Polysemy:
Theoretical and Computational Approaches. Oxford University Press, Oxford, pp. 129-151.
Goddard, C., Wierzbicka, A., 1994. Semantic and Lexical Universals: Theory and Empirical
Findings. Benjamins, Amsterdam.
Goddard, C., Wierzbicka, A., 2002. Meaning and Universal Grammar: Theory and Empirical
Findings, vols. I and II. Benjamins, Amsterdam.
Goffman, E., 1971. Tie signs. In: Relations in Public. Basic Books, New York.
Goh, C., 2000. A discourse approach to the description of intonation in Singapore English. In:
(Brown, A., Deterding, D., Low, E.L. (Eds.), The English Language in Singapore: Research on
Pronunciation. Singapore Association for Applied Linguistics, Singapore, pp. 35-45.
Gohl, C., Günthner, S., 1999. Grammatikalisierung von weil als Diskursmarker in der gesprochenen
Sprache. Zeitschrift für Sprachwissenschaft, 18, 39-75.
Goldberg, A., 1995. Constructions. Chicago University Press, Chicago.
Gomez-Txurruka, I., to appear. The natural language conjunction and. Linguistics and Philosophy.
References 477
González, M., 2004. Pragmatic Markers in Oral Narrative: The Case of English and Catalan.
González, M., 2005 Pragmatic markers and discourse coherence relations in English and Catalan oral
narrative. Discourse Studies, 7, 53-86.
Gregori Signes, C., 1996. ‘Bueno, hasta luego’: El uso de bueno en conversaciones. Miscelánea: A
Journal of English and American Studies, 17, 157-170.
Grevisse, M., 1993. Le bon usage, treizième édition refondue par A. Goosse. Duculot, Paris.
Grice, P., 1975. Logic and conversation. In: Cole, P., Morgan, J.L. (Eds.), Syntax and Semantics, vol.
3. Academic Press, New York, pp. 41-58.
Grice, P., 1989. Studies in the Way of Words. Harvard University Press, Cambridge, MA.
Grobet, A., 2002. L’identification des topiques dans les dialogues. Duculot, Louvain-la Neuve.
Grosz, B.J., 1982. Discourse Analysis. In: Kittredge, R., Lehrberger, J. (Eds.), Sublanguage. Studies
on Language in Restricted Semantic Domains. Mouton de Gruyter, Berlin, pp. 138-174.
Grosz, B.J., Sidner, C.L., 1986. Attention, intentions, and the structure of discourse. Computational
Groupe lamda-l, 1975. Car, parce que, puisque. Revue romane, 10, 248-280.
Gülich, E., 1970. Makrosyntax der Gliederungssignale im gesprochenen Französisch. Wilhelm Fink,
Munich.
Gülich, E., 1991. Pour une ethnométhodologie linguistique: Description de séquences
conversationnelles explicatives. In: Dausendschön-Gay, U., Gülich, E., Krafft, U. (Eds.),
Linguistische Interaktionsanalysen. Beiträge zum 20. Romanistentag 1987. Niemeyer, Tübingen,
pp. 325-364.
Gülich, E., 1999. Les activitées de structuration dans l'interaction verbale. In: Barbéris, J.-M. (Ed.),
Le français parlé. Variétés et discours [Les Cahiers de praxématique]. Praxiling Université Paul
Valéry, Montpellier III, pp. 21-47.
Gülich, E., Kotschi, T., 1983. Les marqueurs de la réformulation paraphrastique. Connecteurs
pragmatiques et structure du discours.Actes du 2ème Colloque de Pragmatique de Genève.
Cahiers de linguistique française, 5, 305-351.
Gülich, E., Mondada, L., 2001. 48. Konversationsanalyse/Analyse conversationnelle. In: Holtus, G.,
Metzeltin, M., Schmitt, C. (Eds.), Lexikon der romanistischen Linguistik vol. I, no. 2,
Méthodologie. Niemeyer, Tübingen, pp. 196-250.
Gumperz, J., 1982. Discourse Strategies. Cambridge University Press, Cambridge.
Gumperz, J., 1984. Communicative Competence Revisited. In: Schiffrin, D. (Ed.), Meaning, From,
and Use in Context: Linguistic Applications. Georgetown University Press, Washington, DC, pp.
278-289.
Günthner, S., 1999. Entwickelt sich der Konzessivkonnektor obwohl zum Diskursmarker?
Grammatikalisierungstendenzen im gesprochenen Deutsch. Linguistische Berichte, 180, 409-446.
Günthner, S., 2000. From concessive connector to discourse marker: the use of obwohl in everyday
German interaction. In: Couper-Kuhlen, E., Kortmann, B. (Eds.), Cause, Condition, Concession,
Contrast. Mouton de Gruyter, Berlin, pp. 439-468.
Gupta, A.F., 1992. The pragmatic particles of Singapore Colloquial English. Journal of Pragmatics,
18, 31-57.
Gupta, A.F., 1989. Singapore Colloquial English and Standard English. Singapore Journal of
Education, 10, 33-39.
478 References
Gupta, A.F., 1991. Acquisition of diglossia in Singapore English. In: Kwan-Terry, A. (Ed.), Child
Language Development in Singapore and Malaysia. Singapore University Press, Singapore, pp.
119-160.
Gupta, A.F., 1992. The Pragmatic Particles of Singapore Colloquial English. Journal of Pragmatics,
17, 39-65.
Gupta, A.F., 1994. The Step-Tongue: Children’s English in Singapore. Multilingual Matters,
Clevedon.
Gupta, A.F., 1998. Singapore Colloquial English? Or deviant standard English?. In: Tent, J., Mugler,
F. (Eds.), SICOL, Proceedings of the Second International Conference on Oceanic Linguistics.
Vol. 1: Language Contact. Pacific Linguistics, Canberra, pp. 43-57.
Gussenhoven, C., 2002. Intonation and Interpretation: Phonetics and Phonology. Proceedings of the
Speech Prosody Workshop. Aix-en-Provence, France.
Hagège, C., 2001. Les processus de grammaticalisation. In: Haspelmath, M., König, E.,
Oesterreicher, W., Raible, W. (Eds.), Language Typology and Language Universals. An
International Handbook. [Handbücher zur Sprach- und Kommunikationswissenschaft, 20.1].
Mouton de Gruyter, Berlin, pp. 1608-1623.
Halliday, M.A.K., Hasan, R., 1976. Cohesion in English. Longman, London.
Hamblin, C.L., 1973. Questions in Montague Grammar. Foundations of Language, 10, 41-53.
Han Yang, S., 1987. A pragmatic study of some sentence-final and post-verbal particles in Mandarin
Chinese. Doctoral dissertation, University of York.
Han Yang, S., 1991. An Analysis of Particles in Mandarin: with Special Reference to Ba and Le.
Sympuusha, Tokyo.
Hanks, W., 1993. Referential Practice. Chicago University Press, Chicago.
Hansen, M.-B. Mosegaard, 1998a. The Functions of Discourse Particles. A Study with Special
Reference to Spoken French. Benjamins, Amsterdam.
Hansen, M.-B. Mosegaard, 1998b. La grammaticalisation de l’interaction, ou, Pour une approche
polysémique de l’adverbe bien. Revue de sémantique et pragmatique 4, 111-138.
Hansen, M.-B. Mosegaard, 1998c. The semantic status of discourse markers. Lingua, 104, 235-260.
Hansen, M.-B. Mosegaard, 2002a. From aspectuality to discourse marking: The case of French déjà
and encore. Belgian Journal of Linguistics, 16, 23-51.
Hansen, M.-B. Mosegaard, 2002b. Sémiotique peircéenne et analyse des interactions verbales. In:
Andersen, H.L., Nølke, H. (Eds.), Macro-syntaxe et Macro-sémantique. Peter Lang, Bern, pp.
361-381.
Hansen, M.-B. Mosegaard, 2005. From prepositional phrase to hesitation marker : the semantic and
pragmatic evolution of French ‘enfin’. Journal of Historical Pragmatics 6, 37-68.
Hansen, M.-B. Mosegaard, to appear. La polysémie de l’adverbe ‘toujours’. Travaux de linguistique.
Hansen, M.-B. Mosegaard, Rossari, C., 2005. The evolution of pragmatic markers. Special issue of
the Journal of Historical Pragmatics 6 (2).
Hare, R.M., 1971. Meaning and speech acts. In: Practical Inferences. London: Macmillan, pp. 74-99.
Haspelmath, M., 1999. Optimality and diachronic adaptation. Zeitschrift für Sprachwissenschaft, 18,
180-205.
Haspelmath, M., 1999. Why is grammaticalization irreversible? Linguistics, 37, 1043-1068.
Hasselgård, H., Oksefjell, S., 1999. Out of Corpora. Studies in Honour of Stig Johansson. Rodopi,
Amsterdam.
Heeman, P.A., Allen, J.F., 1999. Speech repairs, intonational phrases and discourse markers:
modeling speakers’ utterances in spoken dialog. Computational Linguistics, 25, 1-45.
References 479
Heim, I., 1983. On the projection problem for presuppositions. In: Barlow, M., Flickinger, D.,
Wescoat, M. (Eds.), WCCFL 2, The Proceedings of the Second West Coast Conference on
Formal Linguistics. CSLI, Stanford, CA, 114-126.
Heine, B., Claudi, U., Hünnemeyer, F., 1991. Grammaticalization: A Conceptual Framework.
University of Chicago Press, Chicago.
Helbig, G., 1977. Partikeln als Illokutive Indikatoren im Dialog. Deutsch als Fremdsprache, 14, 30-
44.
Helbig, G., 1988. Lexikon deutscher Partikeln. Enzyklopädie, Leipzig.
Helbig, G., 1994. Lexikon deutscher Partikeln. Langenscheidt/Verlag Enzyklopädie, Leipzig.
Helbig, G., Buscha, J., 1986. Deutsche Grammatik. Ein Handbuch für den Ausländerunterricht.
Enzyklopädie, Leipzig.
Helbig, G., Helbig, A., 1990. Lexikon deutscher Modalwörter. Enzyklopädie, Leipzig.
Held, G., 2003. Partikeln und Höflichkeit. Lang, Frankfurt/Main.
Hengeveld, K., 1992. Non-verbal predication. Theory, Typology, Diachrony. Mouton de Gruyter,
Berlin.
Henne, H., Rehbock, H., 1993. Einführung in die Gesprächsanalyse. Mouton de Gruyter, Berlin.
Hentschel, E., 1986. Funktion und Geschichte deutscher Partikeln. Ja, doch, halt und eben.
Hentschel, E., 1991. Aspect versus particle: Contrasting German and Serbo-Croatian. Multilingua,
101, 139-149.
Hentschel, E., 2003. Wenn Partikeln frech werden…. In: Held, G. (Ed.), Partikeln und Höflichkeit.
Lang, Frankfurt/Main, pp. 55-72.
Hentschel, E., Weydt, H., 1983. Der pragmatische Mechanismus: denn und eigentlich. In: Weydt, H.
(Ed.), Partikeln und Interaktion. Niemeyer, Tübingen, pp. 263-273.
Hentschel, E., Weydt, H., 1989. Wortartenprobleme bei Partikeln. In: Weydt, H. (Ed.), Sprechen mit
Partikeln. Mouton de Gruyter, Berlin, pp. 3-18.
Hentschel, E., Weydt, H., 1995. Die Wortarten des Deutschen. In: Agel, V., Brdar-Szabó, R. (Eds.),
Grammatik und deutsche Grammatiken. Budapester Grammatiktagung 1993. Niemeyer,
Tübingen, pp. 39-60.
Hentschel, E., Weydt, H., 2002. Die Wortart Partikel. In: Cruse, D.A., et al. (Eds.) Lexikologie.
Lexicology. Ein Internationales Handbuch zur Natur und Struktur von Wörtern und
Wortschätzen. An International Handbook on the Nature and Structure of Words and
Vocabularies. Mouton de Gruyter, Berlin, pp. 646-653.
Heritage, J., 1984. A change-of-state token and aspects of its sequential placement. In: Atkinson, J.
M., Heritage, J. (Eds.), Structure of Social Action: Studies in Conversation Analysis. Cambridge
University Press, Cambridge, pp. 299-345.
Heritage, J., 1988. Explanations as accounts: a conversation analytic perspective. In: Antaki, C. (Ed.),
Understanding Everyday Explanation: A Casebook of Methods. Sage, Beverly Hills, CA, pp.
127-144.
Heritage, J., 1998. Oh-prefaced responses to inquiry. Language in Society, 27, 291-334.
Heritage, J., Sorjonen, M.-L., 1994. Constituting and maintaining activities across sequences: And
prefacing. Language in Society, 23, 1- 29.
Hidalgo Navarro, A., Pons Bordería, S., 2001. Sobre las propiedades fónicas de los marcadores
discursivos y su grado de especialización funcional. Paper presented at the XXXI Congress of the
S.E.L. Almería, December, 14-17th, 2001.
480 References
Hirschberg, J., Litman, D., 1993. Empirical studies on disambiguation of cue phrases. Computational
Hirschberg, J., Ward, G., 1992. The influence of pitch range, duration, amplitude and spectral
features on the interpretation of the rise-fall-rise intonation contour in English. Journal of
Phonetics, 20, 241-251.
Hoelker, K., 1991. Französisch: Partikelforschung. In: Holtus, G., Metzeltin, M., Schmitt, C. (Eds.),
Lexicon der Romanistischen Linguistik. Niemeyer, Tübingen, pp. 77-88.
Hoffmann, L., 1994. Eine Verhandlung vor dem Amtsgericht. In: Redder, A., Ehlich, K. (Eds.),
Gesprochene Sprache. Transkripte und Tondokumente [Phonai 41]. Niemeyer, Tübingen, pp. 19-
90.
Hopper, P.J., 1991. On some principles of grammaticalization. In: Traugott, E.C., Heine, B. (Eds.),
Approaches to Grammaticalization, vol. 1. Benjamins, Amsterdam, pp. 17-35.
Hopper, P.J., Thompson, S.A., 1984. The discourse basis for lexical categories in universal grammar.
Language, 60, 703-752.
Hopper, P.J., Traugott, E.C., 1993. Grammaticalization. Cambridge University Press, Cambridge.
Horne, M., Hansson, P., Bruce, G., Frid, J., Filipsson, M., 2001. Cue words and the topic structure of
spoken discourse: The case of Swedish men ‘but’. Journal of Pragmatics, 33, 1061-1081.
Hossbach, S., 1998. Reformulative Indikatoren im Französischen und verwandte pragmatische
Indikatoren. Ein kommentiertes bibliographisches Wörterbuch. Freie Universität, Berlin.
Houtkoop, H., Mazeland, H., 1985. Turns and discourse units in everyday conversation. Journal of
Hoye, L., 1997. Adverbs and Modality in English. Longman, London.
Ickler, T., 1994. Zur Bedeutung der sogenannten ‘Modalpartikeln’. Sprachwissenschaft, 19, 374-404.
Ifantidou-Trouki, E., 1993. Sentential adverbs and relevance. Lingua, 90, 69-90.
Infantidou, E., 2001. Evidentials and Relevance. Benjamins, Amsterdam.
Inkova-Manzotti, O., 2002. Les connecteurs accommodants: le cas de autrement. Cahiers de
Institut national de la langue française (TLF), 1994. Trésor de la langue française, vol. 16. Gallimard,
Paris.
Jakobson, R., 1971/1957. Shifters, verbal categories, and the Russian verb. In: Selected Writings, vol.
II. Mouton, The Hague, pp. 130-147.
James, A. R., 1983. Compromisers in English: a cross-disciplinary approach to their interpersonal
significance. Journal of Pragmatics, 7, 191-206.
Jayez, J., 2002. Les impliquestions. In: Carel, M. (Ed.), Les facettes du dire. Hommage à Oswald
Ducrot. Kimé, Paris, pp. 141-154.
Jayez, J., Rossari, C., 1998. Discourse relations vs. discourse markers relations. Proceedings of the
ACL ’98 workshop on discourse relations and discourse markers, pp. 72-78.
Jayez, J., Rossari, C., 2000. The semantics of pragmatic connectives. The French donc example. In:
Abeillé, A., Rambow, O. (Eds, )Tree Adjoining Grammars: Formalisms, Linguistic Analysis and
Processing. CSLI, Stanford.
Jayez, J., Rossari, C., 2001. The discourse level sensitivity of consequence discourse markers in
French. Cognitive Linguistics, 12, 275-290.
Jefferson, G., 1974. Error Correction as an Interactional Resource. Language in Society, 3, 181-200.
Johanssson, S., Oksefjell, S., 1998. Corpora and Cross-linguistic Research: Theory, Method and Case
Studies. Rodopi, Amsterdam.
Johnson, S., 1773. Dictionary of the English Language, fourth ed. London.
References 481
Johnstone, B., 2002. Discourse Analysis. Blackwell, Oxford.

Jucker, A.H., 1993. The discourse marker well: A relevance theoretical account. Journal of
Jucker, A., 1997. The discourse marker well in the history of English. English Language and
Jucker, A., Smith, S., 1998. And people just you know like ‘wow’: discourse markers as negotiating
strategies. In: Jucker, A., Ziv, Y. (Eds.), Discourse Markers: Description and Theory. Benjamins,
Jucker, A., Ziv, Y., 1998. Discourse markers: Introduction. In: Jucker, A., Ziv, Y. (Eds.), Discourse
Markers: Description and Theory. Benjamins, Amsterdam, pp. 1-12.
Jucker, A., Ziv, Y., 1998. Discourse Markers: Descriptions and Theory. Benjamins, Amsterdam.
Karagjosova, E., 2001. Towards a comprehensive meaning of German doch. In: Proceedings of the
ESSLLI 20001 Student Session, ESSLLI, Helsinki.
Kärkkäinen, E., 1998. The marking and interactional functions of epistemic stance in American
English conversational discourse. Unpublished doctoral dissertation. University of California,
Santa Barbara.
Kärkkäinen, E., 2004. Epistemic Stance in English Conversation. Benjamins, Amsterdam.
Karttunen,L., 1974. Presupposition and linguistic context. Theoretical Linguistics, 1, 181-194.
Karttunen, L., Peters, S., 1979. Conventional Implicature. In: Oh, C.K., Dineen, D. (Eds.), Syntax
and Semantics 11: Presupposition. Academic Press, New York, pp. 1-56.
Kay, P., 1990. Even. Linguistics and Philosophy, 8, 59-111.
Kay, P., 1992. At least. In: Lehrer A., Kittay, E.F. (Eds.), Frames, Fields and Contrasts: New Essays
in Semantic and Lexical Organization. Lawrence Erlbaum, Hillsdale, NJ, pp. 309-331.
Kay, P., Fillmore, C., 1999. Grammatical Constructions and Linguistic Generalizations: The What’s
X Doing Y? Construction. Language, 75, 1-33.
Killingley, S.-Y., 1972. Clause and sentence types in Malayan English. Orbis, 21, 537-548.
Knott, A., 2000. An Algorithmic Framework for Specifying the Semantics of Discourse Relations.
Computational Intelligence, 16, 1-10.
Knott, A., Dale, R., 1994. Using linguistic phenomena to motivate a set of coherence relations.
Knott, A., Sanders, T., 1998. The classification of coherence relations and their linguistic markers:
An exploration of two languages. Journal of Pragmatics, 30, 135-175.
Koch, P., 1995. Une langue comme toutes les autres: latin vulgaire et traits universels de l'oral. In:
Callebat, L. (Ed.), Latin vulgaire — latin tardif IV. Actes du 4e colloque international sur le latin
vulgaire et tardif. Caen, 2-5 septembre 1984. Olms/ Weidmann, Zürich, pp. 125-144.
Koch, P., forthcoming. Romanische Sprachgeschichte und Varietätenlinguistik. In: Ernst, G.,
Gleßgen, M.-D., Schmitt C., Schweickard, W. (Eds.), Romanische Sprachgeschichte
[Handbücher zur Sprach- und Kommunikationswissenschaft]. Mouton de Gruyter, Berlin.
Koch, P., Oesterreicher, W., 1990. Gesprochene Sprache in der Romania: Französisch, Italienisch,
Spanisch. Tübingen.
König, E., Requardt, S., 1991. A relevance-theoretic approach to the analysis of modal particles in
German. Multilingua, 10, 63-77.
König, E., Siemund, P., 2000. Causal and concessive clauses: formal and semantic relations. In:
Couper-Kuhlen, E., Kortmann, B. (Eds.), Cause – Condition – Concession – Contrast. Mouton de
Gruyter, Berlin, pp. 341-360.
Kosslyn, S., Koenig O., 1995. Wet Mind. Free Press, New York.
482 References
Krifka, M., 1999. At least some determiners aren’t determiners. In: Turner, K. (Ed.,), The Semantics-
Pragmatics Interface from Different Points of View [CRiSPI 1]. Elsevier Science, Amsterdam,
pp. 257-291.
Krifka, M., 2000. Alternatives for aspectual particles: semantics of still and already. In: Proceedings
of Berkeley Linguistic Society Meeting.
Kripke, S., undated. Presupposition. Manuscript.
Kroon, C., 1995. Discourse Particles in Latin. A Study of nam, enim, autem, vero, and at. Giebven,
Amsterdam.
Kroon, C., 1998. A framework for the description of Latin discourse markers. Journal of Pragmatics,
30, 205-223.
Kuyumcuyan, A., 2002. Diction et mention. Lang, Bern.
Kwan-Terry, A., 1978. The meaning and source of the la and what particles in Singapore English.
RELC Journal 9, 22-36.
Kwok, H., 1984. Sentence particles in Cantonese. Centre of Asian Studies, Hong Kong.
Labov, W., 1972. The transformation of experience in narrative syntax. In: Language in the Inner
City. University of Pennsylvania Press, Philadelphia, pp. 354-396.
Labov, W., 1984. Field methods of the project on linguistic change and variation. In: Baugh, J.,
Sherzer, J. (Eds.), Language in Use. Prentice Hall, Englewood Cliffs, NJ, pp. 28-53.
Labov, W., Fanshel, D., 1987. Therapeutic Discourse. Academic Press, New York.
Laitinen, L., 2002. From logophoric pronoun to discourse particle: A case study of Finnish and
Saami. In: Wischer, I, Diewald, G. (Eds.), New Reflections on Grammaticalization. International
Symposium, Potsdam, 17-19 June, 1999. [Typological Studies in Language 49]. Benjamins,
Lakoff, G., 1987. Women, Fire and Dangerous Things. Chicago University Press, Chicago.
Langacker, R., 1990. Subjectification. Cognitive Linguistics, 5-38.
Lázaro Carreter, F., 1990. Diccionario de términos filológicos. Madrid, Gredos.
Leech, G., 1983. Principles of Pragmatics. Longman, London.
Leezenberg M., 2001. Contexts of Metaphor. Elsevier, Amsterdam.
Lehmann, C., 1985. Grammaticalization: synchronic variation and diachronic change. Lingua e Stile,
20, 303-318.
Lehmann, C., 1995/1982. Thoughts on Grammaticalization. Revised and expanded version. First
published edition [Lincom Studies in Theoretical Linguistics 1], Lincom Europa, München.
Lenk, U., 1995. Discourse markers and conversational coherence. In: Wårvik, B., Tanskanen, S.-T.,
Hiltunen, R. (Eds.), Organization in Discourse. Proceedings from the Turku Conference.
University of Turku, Turku, pp. 341-352.
Lenk, U., 1998. Discourse markers and global coherence in conversation. Journal of Pragmatics, 30,
245-257.
Lenk, U., 1998. Marking Discourse Coherence: Functions of Discourse Markers in Spoken English.
Gunter Narr Verlag, Tübingen.
Ler, S.L., 2001. The interpretation of the discourse particle meh in Singapore Colloquial English.
Asian Englishes 4, 4-22.
Ler, S.L., 2002. The discourse particle lah in Singapore English: A relevance-theoretic account. In:
Brend, R., Sullivan, W., Lommel, A. (Eds.), LACUS XXVIII. Linguistic Association of Canada
and the United States, USA, pp. 287-296.
Levelt, W.J.M., 1989. Speaking: From Intention to Articulation. MIT Press, Cambridge, MA.
Levinson, S.C., 1983. Pragmatics. Cambridge University Press, Cambridge.
References 483
Levinson, S.C., 1995. Three levels of meaning. In: Palmer, F.R. (Ed.), Grammar and Meaning:
Essays in Honour of Sir John Lyons. Cambridge University Press, Cambridge, pp. 90-115.
Levinson, S.C., 2000. Presumptive Meanings. The Theory of Generalized Conversational
Implicature. MIT Press, Cambridge, MA.
Lewis, D., 1979. Scorekeeping in a language game. Journal of Philosophical Logic, 8, 339-359.
Lewis, D.M., 1999. From modal adverbial to discourse connective: Some rhetorical effects in
present-day English. In: Verschueren, J. (Ed.), Pragmatics in 1998. Selected Papers from the 6th
International Pragmatics Conference, vol.2. IPrA, Antwerp, pp. 363-375.
Lewis, D.M., 2002. Rhetorical factors in lexical-semantic change: the case of at least. In: Díaz Vera,
J.E. (Ed.), A Changing World of Words: Studies in English Historical Lexicography, Lexicology
and Semantics. Rodopi, Amsterdam, pp. 525-538.
Lewis, D.M., 2003. Rhetorical motivations for the emergence of discourse particles, with special
reference to English of course. In: van der Wouden, T., Foolen, A., Van de Craen, P. (Eds.),
Particles. Belgian Journal of Linguistics, 16, 79-91.
Li, C.N., Thompson, S.A., 1981. Mandarin Chinese: A Functional Reference Grammar. University of
California Press, Berkeley, CA.
Liberman, M., Pierrehumbert, J., 1984. Intonational invariance under changes in the pitch range and
length. In: Aronoff, M., Oehrle, R.T. (Eds.), Language Sound Structure (. MIT Press, Cambridge,
MA, pp. 157-233.
Lichtenberk, F., 1991. Semantic change and heterosemy in grammaticalization. Language, 67, 475-
509.
Lim, C., 1986. English in Singapore: A Study of its Status and its Solidarity and the Attitudes to its
Use. Doctoral dissertation, National University of Singapore, Singapore.
Lima, J. Pinto de., 2002. Grammaticalization, subjectification and the origin of phatic markers. In:
Wischer, I. and Diewald, G. (Eds.), New Reflections on Grammaticalization. International
Symposium, Potsdam, 17-19 June, 1999. [Typological Studies in Language 49]. Benjamins,
Amsterdam pp. 363-378.
LIP = De Mauro, T., et al., 1993. Lessico di frequenza dell’italiano parlato. Ricerca a cura
dell’Osservatorio linguistico e culturale italiano (OLCI) dell’Università di Roma La Sapienza.
Etaslibri, Milan.
LIZ = Stoppelli, P., Picchi, E., 1995. Letteratura Italiana Zanichelli 2.0. CD-ROM dei testi della
letteratura italiana. Zanichelli, Bologna.
Local, J., 1996. Conversational phonetics: some aspects of news receipts in everyday talk. In:
Couper-Kuhlen, E., Selting, M. (Eds.), Prosody in Conversation. Benjamins, Amsterdam, pp.
177-229.
Loke, K.-K., Low, M.-Y., 1988. A proposed descriptive framework for the pragmatic meanings of
the particle ‘la’ in colloquial Singaporean English. In: McCarthy, B. (Ed.), Asian-Pacific papers:
regional papers presented at the 8th World Congress of Applied Linguistics, University of
Sydney, 16-21 August 1987. Applied Linguistics Association of Australia, Australia, pp. 150-61.
Longacre, R., 1976. Mystery particles and affixes. Papers from the 12th Meeting of the Chicago
Linguistic Society. CLS, University of Chicago, Chicago.
Longacre, R.E., 1983. The Grammar of Discourse. [based on An Anatomy of Speech Notions, 1976].
Plenum Press, New York.
Louwerse, M., 2001. An analytic and cognitive parameterization of coherence relations. Cognitive
484 References
Low, E.L., Deterding, D., 2003. A corpus-based description of particles in spoken Singapore English.
In: Deterding, D., Ling, L.E., Brown, A. (Eds.), English in Singapore: Research on Grammar.
McGraw Hill, Singapore, pp. 58-66.
Lucy, J., 1993. Reflexive Language. Reported Speech and Metapragmatics. Cambridge University
Press, Cambridge.
Lüdtke, H., 1988. Grammatischer Wandel. In: Ammon, U., Dittmar, N., Mattheier, K.J. (Eds.),
Soziolinguistik. Ein internationales Handbuch zur Wissenschaft von Sprache und Gesellschaft
[Handbücher zur Sprach- und Kommunikationswissenschaft]. Mouton de Gruyter, Berlin, pp.
1632-1642.
Luke, K.K., 1990. Utterance Particles in Cantonese Conversation. Benjamins, Amsterdam.
Luscher, J.-M., 1994. Connecteurs: des guides pour l’interprétation. In: Moeschler, J., et al. (Eds.),
Langage et pertinence. Presses Universitaires de Nancy, Nancy, France, pp. 175-227.
Luscher, J.-M., Moeschler, J., 1990. Approches dérivationnelles et procédurales des opérateurs et
connecteurs temporels: les exemples de et et de enfin. Cahiers de Linguistique Française, 11, 77-
104.
Lyons, J., 1977. Semantics, vols. 1 and 2. Cambridge University Press, Cambridge.
Lyons, J., 1980. Semántica. Barcelona, Teide.
Lyons, J., 1995. Linguistic Semantics: An Introduction. Cambridge University Press, Cambridge.
Magno Caldognetto, E., 2002. I correlati fonetici delle emozioni. In: Bazzanella, C., Kobau, P.
(Eds.), Passioni, emozioni, affetti. McGraw-Hill, Milano, pp. 197-213.
Mann, W.C., Thompson, S.A., 1987. Rhetorical Structure Theory: A Theory of Text Organization.
ISI Report RS-87-190. Information Sciences Institute, University of Southern California. Marina
del Rey, CA.
Mann, W.C., Thompson, S.A., 1988. Rhetorical structure theory: Toward a functional theory of text
organization. Text, 8, 243-281.
Manoliu, M.M., 2000. From deixis ad oculos to DM via deixis ad phantasma. In: Smith, J.C.,
Bentley, D. (Eds.), Historical Linguistics, 1995, I: General Issues and Non-Germanic Languages.
John Benjamins, Amsterdam, pp. 243-260.
Marcu, D., 2000. The Theory and Practice of Discourse Parsing and Summarization. MIT Press,
Cambridge, MA.
Martín Zorraquino, M.A., 1991. Elementos de cohesión en el habla de Zaragoza. In: Enguita Utrilla,
J.M. (Ed.), I curso de geografía lingüística de Aragón. Institución Fernando el Católico,
Zaragoza, pp. 253-286.
Martín Zorraquino, M.A., 1994. ‘Bueno’ como operador pragmático en español actual. Encuentro de
Lingüistas y Filólogos de España y México, 2, 403-412.
Martín Zorraquino, M.A., Portolés, J., 1999. Los marcadores del discurso. In: Bosque, I., Demonte,
V. (Eds.), Gramática descriptiva de la lengua española, vol. III. Espasa Calpe, Madrid, pp. 4051-
4213.
Maschler, Y., 1998. Rotse lishmoa keta? Wanna hear something weird/funny? Segmenting Israeli
Hebrew talk-in-interact. In: Jucker, A., Ziv, Y. (Eds.), Discourse Markers: Description and
Theory. Benjamins, Amsterdam, pp. 13-60.
Matras, Y., 1998. Utterance modifiers and universals of grammatical borrowing. Linguistics, 36, 281-
331.
Matsumoto, K., 1999. And-prefaced questions in institutional discourse. Linguistics, 37, 251-274.
Matthews, S., Yip, V., 1994. Cantonese: A Comprehensive Grammar. Routledge, London.
References 485
Mazeland, H., Huiskes, M., 2001. Dutch ‘but’ as a sequential conjunction; its use as a resumption
marker. In: Selting, M., Couper-Kuhlen, E. (Eds.), Studies in Interactional Linguistics.
Benjamins, Amsterdam, pp. 141-169.
Mei, L.-W. Song, 2001. The polemics of Singlish. English Today 17, 39-45.
Merritt, M., 1980. On the Use of OK in Service Encounters. In: Shuy, R.W., Shunkal, A., (Eds.),
Language Use and the Uses of Language. Georgetown University Press, Washington, DC.
Métrich, R., 1993. Lexicographie bilingue des particules illocutoires de l’allemand. Kümmerle,
Göppingen.
Mey, J., 2001. Pragmatics: An Introduction. Blackwell, Oxford.
Meyer-Hermann, R., 1978. Aspekte der Analyse metakommunikativer Interaktionen. In: Meyer-
Hermann, R. (Ed.), Sprechen – Handeln – Interaktion. Ergebnisse aus Bielefelder
Forschungsprojekten zu Texttheorie, Sprechakttheorie und Konversationsanalyse [Konzepte der
Sprach- und Literaturwissenschaft 26]. Tübingen, pp. 103-142.
Moeschler, J., 1989. Modélisation du dialogue. Hermès, Paris.
Moeschler, J., 1996. Théorie pragmatique et pragmatique conversationnelle. A. Colin, Paris.
Molnar, A., 2002. Die Grammatikalisierung deutscher Modalpartikeln: Fallstudien. Lang,
Frankfurt/Main.
Mondada, L., 2001. Pour une linguistique interactionnelle. Marges linguistiques, 1, 1-21.
Morris, C., 1938. Foundations of the theory of signs. In: Neurath, O., Carnap, R., Morris, C. (Eds.),
International Encyclopedia of Unified Science. University of Chicago Press, Chicago, pp. 77-138.
Motsch, W., 1989. Dialog-Texte als modular organisierte Strukturen. Sprache und Pragmatik, 11, 37-
66.
Muller, C., 1999. Encore et toujours les modifieurs aspectuels: de encore à toujours. In: Plénat, M.,
Aurnague, M., Condamines, A., Maurel, J.-P., Molinier, C., Muller, C. (Ed.), L’emprise du sens.
Structures linguistiques et interprétations. Rodopi, Amsterdam, pp. 217-237.
Narbona Jiménez, A., 1989a. Las subordinadas adverbiales impropias en español. Ágora, Málaga.
Narbona Jiménez, A., 1989b. Sintaxis española. nuevos y viejos enfoques. Ariel, Barcelona.
Narbona Jiménez, A., 1990. Las subordinadas adverbiales impropias en español. Ágora, Málaga.
Nemo, F., 1986. Contraintes de pertinence et argumentativité. Semantikos, 9, 21-33.
Nemo, F., 1988. Book review of Sperber, D., Wilson, D., Relevance: Communication and Cognition.
Journal of Pragmatics, 12, 791-795.
Nemo, F., 1992. Contraintes de pertinence et compétence énonciative. L'image du possible dans
l'interlocution. Doctoral thesis. Ecole des Hautes Etudes en Sciences Sociales, Paris.
Nemo, F., 1996. The modal frames of speech, In: Weingartner, P., Schurz, G., Dorn, G. (Eds.), The
Role of Pragmatics in Contemporary philosophy, vol. 2. Kirchberg am Wechsel, Austria, pp. 673-
680.
Nemo, F., 1998. Enfin, encore, toujours entre indexicalité et emplois. Actes du XXIIe Congrès
International de Linguistique et de Philologie Romanes, Bruxelles, tome VII. Sens et fonctions,
XIV. Niemeyer, Tübingen.
Nemo, F., 1999. The pragmatics of signs, the semantics of relevance, and the semantic/pragmatic
interface. In: Turner, K. (Ed.), The Semantics-Pragmatics Interface from Different Points of View
[CRiSPI Series, volume 1]. Elsevier Science, Amsterdam, pp. 343-417.
Nemo, F., 2000. ‘Enfin’, ‘encore’, ‘toujours’ entre indexicalité et emplois. In: Englebert, A., Pierrard,
M., Rosier, L., Van Raemdonck, D. (Eds.), Actes du XXII Congrès de linguistique et de
philologie romanes, vol. VII. Niemeyer, Tübingen , pp. 499-511.
486 References
Nemo, F., 2001a. Contributions, énoncés, constructions, morphèmes. Eléments pour une linguistique
de la signification et de l'interprétation. Thèse d’habilitation. Université Paris 8.
Nemo, F., 2001b. Pour une approche indexicale (et non procédurale des instructions sémantiques.
Revue de Sémantique et Pragmatique, 9-10, 195-218.
Nemo, F., 2002a. Morpheme semantics and the autonomy of morphology. The stable semantics of
(apparently) unstable constructions, In: Andronis, M., Ball, C., Elston, H., Neuvel, S. (Eds.),
Papers from the 37th Meeting of the Chicago Linguistic Society, vol. 2. Chicago Linguistic
Society, University of Chicago, Chicago.
Nemo, F., 2002b. Indexicalité ou catégorisation: le sens entre signification et dénomination. In:
Larrivée, P., Lagorgette, D. (Eds.), La représentation du sens en linguistique. Lincom Europa,
München.
Nemo, F., 2002c. But and mais as morphemes. Delta (Sao Paulo), n°18-2.
Nemo, F., 2003a. Indexicalité, unification contextuelle et constitution extrinsèque du référent.
Langages.
Nemo, F., 2003b. Sémantique des pertinences énonciatives. Peter Lang, Bern.
Nemo, F., Cadiot, P., 1997a. Un problème insoluble? (1). Revue de Sémantique et Pragmatique, 1,
15-22.
Nemo, F., Cadiot, P., 1997b. Un problème insoluble? (2). Revue de Sémantique et Pragmatique, 2, 9-
40.
Nguyen, T.-B., 1986. Toujours est-il. Revue Romane, 21, 192-207.
Nguyen, T.-B., 1988. Toujours en position finale: emploi pragmatique particulier. Revue Romane,
23, 36-46.
Ni, Y., Ler, S.L., 2000. Wordforms and their linguistic and cultural implications: a comparison
between two corpora in the International Corpus of English. In: Brown, A. (Ed.), English in
Southeast Asia 99: Proceedings of the Fourth English in Southeast Asia Conference. National
Institute of Education, National Technological University, Singapore.
Noël, D., 2003. Translations as evidence of semantics. An illustration. Linguistics, 41, 757-785.
Nølke, H., 1989. Polyfoni. En sprogteoretisk indføring [ARK 48, Sproginstitutternes Arbejdspapirer].
Handelshøjskolen i København.
Nølke, H., 1993. Le regard du locuteur. Kimé, Paris.
Nølke, H., 2000. Linguistique modulaire: Principes méthodologiques et applications. In: Nølke, H.,
Adam, J.-M. (Eds.), Approches modulaires: De la langue au discours. Delachaux and Niestlé,
Lausanne, pp. 17-73.
Noren, C., 1999. Reformulation et conversation. De la sémantique du topos aux fonctions
interactionnelles. Uppsala University Press, Uppsala.
Nyan, T., 1992. Ceci-dit. Revue Romane, 27, 181-206.
Nyan, T., 1993. Au moins. Paper presented at the Pragmatics, Discourse and ‘Énonciation’ workshop,
University of Salford, Greater Mancester, England.
Nyan, T., 1998. Metalinguistic Operators, with Reference to French. Peter Lang, Bern.
Nyan, T., 1999a. Simplement. Argumentation, 13, 275-295.
Nyan, T., 1999b. Vers un schéma de la différence: le cas de mais. French Language Studies, 9, 211-
238.
Nyan, T., 2000. Marginal cases. Paper presented at a conference on particles organized by the
Belgian and Dutch Foundation for Science. Brussels, Belgium.
Nyan, T., 2004. Meanings at the Text Level: A Co-evolutionary Approach. Peter Lang, Oxford.
References 487
Ocampo, F., 2002. De el asunto está bueno a bueno, el asunto está: El uso del adjetivo calificativo
bueno como partícula discursiva. Paper presented at the XIII Congreso Internacional de ALFAL,
Universidad de Costa Rica.
Ochs, E., 1979. Planned and unplanned discourse. In: Cole, P., Morgan, J. (Eds.), Syntax and
Semantics, vol 12. Academic Press, New York, pp. 51-80.
Ochs, E., 1996. Linguistic resources for socializing humanity. In: Gumperz, J.J., Levinson, S.C.
(Eds.), Rethinking Linguistic Relativity. Cambridge University Press, Cambridge, pp. 407-437.
Ochs, E., Schieffelin, B., 1989. Language has a heart. In: Ochs, E. (Ed.), The Pragmatics of Affect.
Special Issue of Text, 9, 7-25.
Ochs, E., Schegloff, E.A., Thompson, S.A., 1996. Interaction and Grammar. Cambridge University
Press, Cambridge.
Oesterreicher, W., 1997. Zur Fundierung von Diskurstraditionen. In: Frank, B., Haye, T., Tophinke,
D. (Eds.), Gattungen mittelalterlicher Schriftlichkeit [ScriptOralia 99]. Narr, Tübingen, pp. 19-41.
Ohala, J. 1984. An ethological perspective on common cross-language utilization of F0 of voice.
Phonetica, 41, 1-16.
Onodera, N.O., 1995. Diachronic analysis of Japanese discourse markers. In: Jucker, A. (Ed.),
Historical Pragmatics [Pragmatics & Beyond, New Series 35]. Benjamins, Amsterdam, pp. 393-
437.
Orlandini, A., 2003. Contra: le parcours controversé d’une grammaticalisation. Actes du XVII
Congrès International des Linguistes, Prague July, pp. 24-29. Comparative Linguistics and
Historical Linguistics section, CD-Rom.
Östman, J.-., 1983. You know: A Discourse Functional Approach [Pragmatics and Beyond II:7].
Owen, M., 1981. Conversational units and the use of ‘Well’. In: Werth, P. (Ed.), Conversation and
Discourse. Structure and Interpretation. Croom-Helm, London, pp. 155-178.
Oxford English Dictionary, 2000. Internet edition, Simpson, J. (Ed.). Oxford University Press,
Oxford. http://oed.com.
Pakir, A., 1992. Dictionary entries for discourse particles. In: Pakir, A. (Ed.), Words in a Cultural
Context. UniPress, Singapore, pp. 143-52.
Passonneau, R.J.. Litman, D.J., 1996. Empirical analysis of three dimensions of spoken discourse:
segmentation, coherence, and linguistic devices. In: Hovy, E.H., Scott, D.R. (Eds.),
Computational and Conversational Discourse: Burning issues – an Interdisciplinary Approach.
[NATO ASI Series F: Computer and System Sciences, vol. 151]. Springer, Berlin, pp. 161-194.
Paul/Henne 1992 = Paul, H., 1992. Deutsches Wörterbuch. 9. Vollständig neu bearbeitete Auflage
von Helmut Henne und Georg Objartel unter Mitarbeit von Heidrun Kämper-Jensen. Niemeyer,
Tübingen.
Peirce, C.S., 1932. Collected Papers, vol. 2. Hartshorne, C., Weiss, P. (Eds.). Harvard University
Press, Cambridge, MA.
Peterson, C., McCabe, A., 1991. Linking children’s connective use and narrative macrostructure. In:
McCabe, A., Peterson, C. (Eds.), Developing Narrative Structure. Lawrence Erlbaum , Hillsdale,
NJ.
Petric, T., 1995. Indexikalische Leistungen der Modalpartikeln und ihre natürlichkeitstheoretische
Bewertung. Linguistica, 35, 245-259.
Philippi, D., 1999. Après tout als pragmatischer Indikator im Französischen: Untersuchungen zu
Vorkommen und Funktion. Freie Universität Berlin, Berlin.
488 References
Pierrehumbert, J., Hirschberg, J., 1990. The meaning of intonational contours in the interpretation of
discourse. In: Cohen, P., Morgan J., Pollack, M. (Eds.), Intentions in Communication. MIT Press,
Cambridge, MA, pp. 271-311.
Placencia, M.E., 1997. Address forms in Ecuadorian Spanish. Hispanic Linguistics, 91, 165-202.
Platt, J., 1987. Communicative functions of particles in Singapore English. In: Stelle, R., Threadgold,
T. (Eds.), Language Topics: Essays in Honour of Michael Halliday, vol 1. Benjamins,
Platt, J., Weber, H., 1980. English in Singapore and Malaysia. Oxford University Press, Oxford.
Platt, J., et al., 1983. Singapore and Malaysia: Varieties of English around the World. Benjamins,
Amsterdam.
Polanyi, L., 1978. False starts can be true. Proceedings of the Annual Meeting of the Berkeley
Linguistics Society, 4, 628-639.
Polanyi, L., 1989. Telling the American Story: A Structural and Cultural Analysis of Conversational
Storytelling. MIT Press, Cambridge, MA.
Polanyi, L., 2001. The linguistic structure of discourse. In: Schiffrin, D., Tannen, D., Hamilton, H.
(Eds.), The Handbook of Discourse Analysis. Blackwell, Oxford, pp. 265-281.
Polanyi, L., Scha, R.J.H., 1983. The syntax of text and discourse. Text, 3, 261-270.
Pons Bordería, S., 1998. Conexión y conectores. Estudio de su relación en el registro informal de la
lengua [Cuadernos de filología XXVII]. Universitat de València, València.
Pons Bordería, S., 2000. Los conectores. In: Briz, E., Grupo Valesco (Eds.), ¿Cómo se comenta un
texto coloquial? Ariel, Barcelona, pp. 198-230.
Pons Bordería, S., 2001a. Las unidades de la conversación. Unpublished paper presented at Jornadas
de Pragmática, Valencia, 6-8 noviembre de 2001.
Pons Bordería, S., 2001b. Connectives discourse markers. An overview. In: Ferrer Mora, H., Pons
Bordería, S. (eds.), La pragmática de los conectores y las partículas modales. Universidad de
Valencia, Valencia, pp. 219-242.
Pons Bordería, S., 2003. From agreement to stressing and hedging: Spanish bueno and claro. In:
Held, G. (Ed.), Partikeln und Höflichkeit. Peter Lang, Bern.
Pons Bordería, S., Ruiz Gurillo L., 2001. De todas formas: fijación formal y pragmática. Revista de
Filología Española, 81, 317-351.
Pons Bordería, S., Ruiz Gurillo, L., forthcoming. De todas formas: fijación formal y pragmática.
Revista de Filología Española.
Portolés Lázaro, J., 1998. Marcadores del discurso. Ariel, Barcelona.
Posner, R., 1980. Semantics and pragmatics of sentence connectives in natural language. In: Kiefer,
F., Searle, J., Bierwisch, M. (Eds.), Pragmatics and Speech Act Theory. D. Reidel, Dordrecht,
The Netherlands, pp. 87-122.
Powell, M.-J., 1992. The systematic development of correlated interpersonal and metalinguistic uses
in stance adverbs. Cognitive Linguistics, 3, 75-110.
Preisler, B., 1986. Linguistic Sex Roles in Conversation. Social Variation in the Expression of
Tentativeness in English. Mouton de Gruyter, Berlin.
Qiu, S.J., 1985. Early language development in Chinese children. Masters honors thesis. University
of Sydney, Sydney, Australia.
Rampton, B., 1995. Crossing: Language and Ethnicity among Adolescents. Longman, London.
Razgouliaeva, A., 2002. La combinaison des connecteurs mais enfin. Cahiers de Linguistique
Française, 24, 143-168.
Récanati, F., 1995. The alleged priority of literal interpretation. Cognitive Science, 19, 207-232.
References 489
Redder, A., Ehlich, K., 1994. Gesprochene Sprache. Transkripte und Tondokumente [Phonai 41].
Redeker, G., 1986. Language use in informal narratives: Effects of social distance and listener
involvement. Doctoral dissertation, University of California, Berkeley.
Redeker, G., 1990. Ideational and pragmatic markers of discourse structure. Journal of Pragmatics,
14, 367-381.
Redeker, G., 1991a. Linguistic markers of discourse structure [review of Discourse Markers by D.
Schiffrin]. Linguistics, 29, 1139-1172.
Redeker, G., 1991b. Quotation in discourse. In: van Hout, R., Huls, E. (Eds.), Artikelen van de eerste
Sociolinguïstische Conferentie. Eburon, Delft, The Netherlands, pp. 341-355.
Redeker, G., 1992. ‘Kleine woordjes’ in spontaan taalgebruik – stoplapjes of signalen voor de lezer/
luisteraar? [‘Small words’ in spontaneous language use – fillers or signals for the reader/hearer?].
Toegepaste Taalwetenschap in Artikelen, 43, 55-65.
Redeker, G., 2000. Coherence and structure in text and discourse. In: Bunt, H., Black, W. (Eds.),
Abduction, Belief and Context in Dialogue. Studies in Computational Pragmatics. Benjamins,
Reichman, R., 1978. Conversational coherence. Cognitive Science, 2, 283-327.
Renzi, L., Bisetto, A., 2000. Linguistica e italiano antico, Lingua e Stile, 35, 537-743.
Richards, J.C., 1977. Variation in Singapore English. In: Crewe, W.J., (Ed.), The English Language
in Singapore. Eastern University Press, Singapore, pp. 68-81.
Richards, J.C., 1983. Singapore English: rhetorical and communicative styles. In: Kachru, B.B. (Ed.),
The Other Tongue: English Across Cultures. Pergamon , Oxford, pp. 154-167.
Richards, J.C., Tay, M., 1977. The ‘la’ particle in Singapore English. In: Crewe, W.J., (Ed.), The
English Language in Singapore. Eastern University Press, Singapore, pp. 14-56.
Rieschild, V.R., 1996. Lebanese-Arabic discourse: Adult interaction with young children (with
reference to Australian-English situations). Unpublished doctoral thesis, Australian National
University, Canberra, Australia.
Rizzi, L., 1997. The fine structure of the left periphery. In: Haegeman, L. (Ed.), Elements of
Grammar: Handbook in Generative Syntax. Kluwer, Dordrecht, pp. 281-337.
Roberts, C., 1996. Anaphora in intentional contexts. In: Lappin, S. (Ed.), The Handbook of
Contemporary Semantic Theory. Blackwell, Oxford, pp. 215-246.
Rosch, E., 1973. Natural categories. Cognitive Psychology, 4, 328-50.
Rosch, E., 1977. Human categorization. In: Warren, N. (Ed.), Studies in Cross-Cultural Psychology,
vol. 1. Academic Press, New York, pp. 1-49.
Rossari, C., 1994. Les opérations de reformulation. Peter Lang, Bern.
Rossari, C., 2000. Connecteurs et relations de discours: des liens entre cognition et signification.
Presses Universitaires de Nancy, Nancy, France.
Rossari, C., 2001. Les relations de discours: approches rhétoriques, approches pragmatiques et
approches sémantiques. Verbum, 23, 59-72.
Rossari, C., 2002. Mais que sont donc les mots du discours? In: Carel, M. (Ed.), Les facettes du dire.
Hommage à Oswald Ducrot. Kimé, Paris, pp. 283-296.
Rossari, C., Jayez, J., 1996. Donc et les consécutifs. Des systèmes de contraintes différentielles.
Linguisticae Investigationes 20, 117-143.
Rossari, C., Jayez, J., 1997. Connecteurs de conséquence et portée sémantique. Cahiers de
490 References
Rouchota, V., 1998. Procedural meaning and parenthetical discourse markers. In: Jucker, A.H., Ziv,
Y. (Eds.), Discourse Markers: Descriptions and Theory. Benjamins, Amsterdam, pp. 97-126.
Rouchota, W., 1996. Discourse connectives: What do they link? UCL Working Papers in Linguistics,
8.
Roulet, E., 1981. Échanges, interventions et actes de langage dans la structure de la conversation.
Études de linguistique appliquée, 44, 7-39.
Roulet, E., 1984. Speech acts, discourse structure, and pragmatic connectives. Journal of Pragmatics,
8, 31-47.
Roulet, E., 1987. Complétude interactive et connecteurs réformulatifs. Cahiers de Linguistique
Roulet, E., 1990. Et si, après tout, ce connecteur n’était pas un marqueur d’argument ou de prémisse
impliquée? Cahiers de Linguistique Française,11, 329-343.
Roulet, E., 1991. Vers une approche modulaire de l'analyse du discours. Cahiers de Linguistique
Roulet, E., 1996. Polyphony. In: Verschueren, J., Östman, J.-O., Blommaert J., Bulcaen, C. (Eds.),
Handbook of Pragmatics. Benjamins, Amsterdam.
Roulet, E., 1999. La description de l’organisation du discours. Hatier, Paris.
Roulet, E., 2000. Une approche modulaire de la complexité de l'organisation du discours. In: Nølke,
H., Adam, J.-M. (Eds.), Approches modulaires: de la langue au discours. Delachaux and Niestlé,
Lausanne, pp. 187-257.
Roulet, E., 2002a. Le problème de la définition des unités à la frontière entre le syntaxique et le
textuel. Verbum, 24, 161-178.
Roulet, E., 2002b. De la nécessité de distinguer des relations de discours sémantiques, textuelles et
praxéologiques. In: Andersen, H.L, Nølke, H. (Eds.), Macro-syntaxe et macro-sémantique. Actes
du colloque international d’Aarhus (, pp. 141-165. Bern, Lang.
Roulet, E., 2002c. Une comparaison entre deux approches de la description des relations de discours
dans un texte de presse. In: Carel, M. (Ed.), Les facettes du dire. Hommage à Oswald Ducrot.
Kimé, Paris, pp. 297-312.
Roulet, E., et al., 1991. L’articulation du discours en français contemporain, third ed., Peter Lang,
Bern.
Roulet, E., Filliettaz, L., Grobet, A., 2001. Un modèle et un instrument d’analyse de l’organisation du
discours. Peter Lang, Bern.
Rubattel, C., 1987. Actes de langage, semi-actes et typologie des connecteurs pragmatiques.
Linguisticae Investigationes, 11, 379-404.
Rubdy, R., 2001. Creative destruction: Singapore’s Speak Good English movement. World
Englishes, 20, 341-355.
Ruiz Gurillo, L., Pons Bordería, S., 1996. Escalas morfológicas o escalas argumentativas. Español
Actual, 64, 53-74.
Sacks, H., 1973. Lecture Notes. School of Social Science, University of California at Irvine.
Sacks, H., 1984. Notes on methodology. In: Atkinson, J., Heritage, J. (Eds.), Structure of Social
Action: Studies in Conversation Analysis. Cambridge University Press, Cambridge, pp. 21-27.
Sacks, H., Schegloff, E.A., Jefferson, G., 1974. A simplest systematics for the organization of turn-
taking for conversation. Language, 50, 696-735.
Sacks, H., 1992. Lectures on Conversation, vol. II. Jefferson, G. (Ed.). Blackwell, Oxford.
Saeed, J. J., 2003. Semantics, second ed. Blackwell, Oxford.
References 491
Sagerer, G., Eikmeyer, H.-J., Rickheit, G., 1994. Wir bauen jetzt ein Flugzeug. Konstruieren im
Dialog. Arbeitsmaterialien. Technical Report, SFB 360 Situierte künstliche Kommunikatoren,
University of Bielefeld, Germany.
Sanders, T.J.M., Spooren, W.P.M., Noordman, L.G.M., 1992a. Coherence relations in a cognitive
theory of discourse representation. Cognitive Linguistics, 4, 93-133.
Sanders, T.J.M., Spooren, W.P.M., Noordman, L.G.M., 1992b. Towards a taxonomy of coherence
relations. Discourse Processes, 15, 1-35.
Sanders, T.J.M., Spooren, W.P.M.,1999. Communicative intentions and coherence relations. In:
Bublitz, W., Lenk, U., Ventola, E. (Eds.), Coherence in Spoken and Written Discourse: How to
Create it and How to Describe it. Benjamins, Amsterdam, pp. 235-250.
Sanders, T.J.M., Noordman, L.G.M., 2000. The role of coherence relations and their linguistic
markers in text processing. Discourse Processes, 29, 37-60.
Sanders, T.J.M., Schilperoord, J., Spooren, W.P.M., 2001. Text Representation: Linguistic and
Psycholinguistic Aspects. Benjamins, Amsterdam.
Sasse, H.-J., 1993. Syntactic categories and subcategories. In: Jacobs, J., et al. (Eds.), Syntax. Ein
internationales Handbuch zeitgenössischer Forschung. Mouton de Gruyter, Berlin, pp. 646-686.
Saussure, F. de, 1972/1916. Cours de linguistique générale. Payot, Paris.
Schegloff, E.A., 1972. Sequencing in conversational openings. In: Gumperz, J.J., Hymes, D. (Eds.),
Directions in Sociolinguistics. Holt, Rinehart and Winston, New York, pp. 146-380.
Schegloff, E., 1982. Discourse as an interactional achievement: some uses of ‘uh huh’ and other
things that come between sentences. In: Tannen, D. (Ed.), Analysing Discourse: Text and Talk.
Georgetown University Press, Washington, DC.
Schegloff, E., 1997. Whose text, whose context? Discourse in Society, 8, 165-187.
Schegloff, E., Sacks, H., 1973. Opening up closings. Semiotica, 8, 2289-2327.
Scheler, G., Fischer, K., 1997. The many functions of discourse particles: a computational model of
pragmatic interpretation. Proceedings of CogSci.
Scherer, K.R., 1986. Vocal affect expression: A review and a model for future research,
Psychological Bulletin, 99, 143-165.
Scherer, K.R., 2000. Emotion effects on voice and speech: Paradigms and approaches to evaluation.
Proceedings of the ISCA Workshop on Speech and Emotion: A Conceptual Framework for
Research. Belfast, Northern Ireland.
Schieffelin, B.B., 1985. The acquisition of Kaluli. In: Slobin, D.I. (Ed.), The Crosslinguistic Study of
Language Acquisition . Lawrence Erlbaum, Mahwah, NJ, pp. 525-593.
Schiffrin, D., 1980. Metatalk: Organizational and evaluative brackets in discourse. Sociological
Inquiry, 50, 199-236.
Schiffrin, D., 1985. Conversational coherence: the role of well. Language, 61, 640-667.
Schiffrin, D., 1987a. Discourse Markers [Studies in Interactional Sociolinguistics 5]. Cambridge
University Press, Cambridge.
Schiffrin, D., 1987b. Discovering the context of an utterance. Linguistics, 25, 11- 32.
Schiffrin, D., 1990. Between text and context: Deixis, anaphora and the meaning of then. Text, 10,
245-70.
Schiffrin, D., 1992. Anaphoric then: Aspectual, textual and epistemic meaning. Linguistics, 30, 753-
792.
Schiffrin, D., 1994a. Approaches to Discourse. Blackwell, Oxford.
Schiffrin, D., 1994b. Making a list. Discourse Processes, 17, 377- 405.
492 References
Schiffrin, D., 2001. Discourse markers: language, meaning, and context. In: Schiffrin, D., Tannen,
D., Hamilton, H. (Eds.), The Handbook of Discourse Analysis. Blackwell, Oxford, pp. 54-75.
Schiffrin, D., 2002. Mother and friends in a Holocaust survivor oral history. Language in Society, 31,
309- 354.
Schiffrin, D., Tannen, D., Hamilton, H., 2001. The Handbook of Discourse Analysis. Blackwell,
Oxford.
Schilperoord, J., Verhagen, A., 1998. Conceptual dependency and the clausal structure of discourse.
In: Koenig, J.P. (Ed.), Discourse and Cognition. Bridging the Gap. CSLI, Stanford, CA, pp. 141-
163.
Schlieben-Lange, B., 1983. Traditionen des Sprechens. Elemente einer pragmatischen
Sprachgeschichtsschreibung. Kohlhammer, Stuttgart, Germany.
Schoessler, H., 2002. Aber, aber. Masters thesis, Department of Computational Linguistics,
University of Amsterdam.
Schourup, L.C., 1985. Common Discourse Particles in English Conversation. Garland, New York.
Schourup, L.C., 1999. Discourse markers. Lingua, 107, 227-265.
Schourup, L.C., 2001. Rethinking well. Journal of Pragmatics, 33, 1025-1060.
Schwenter, S., 1999. Pragmatics of Conditional Marking [Outstanding Dissertations in Linguistics].
Garland, New York.
Schwenter, S., 2001. Additive particles and the construction of context. In: Ferrer Mora, H., Pons
Bordería, S. (Eds.), La pragmática de los conectores y las partículas modales. Universidad,
Valencia, Spain, pp. 245-262.
Schwenter, S., 2001. Additive particles and the construction of context. Quaderns de Filologia,
Estudis Linguistics, 4, 245-261.
Schwenter, S., 2002. Discourse markers and the PA/SN distinction. Journal of Linguistics, 38, 43-69.
Schwenter, S., Traugott, E.C., 2000. Invoking scalarity: the development of in fact. Historical
Schwitalla, J., 1976. Dialogsteuerung. Vorschläge zur Untersuchung. In: Berens, F.J., Jäger, K.-H.,
Schank, G., Schwitalla, J. (Eds.,), Projekt Dialogstrukturen. Ein Arbeitsbericht [Linguistische
Grundlagen. Forschungen des Instituts für deutsche Sprache I/12]. Max Hueber, München, pp.
73-104.
Searle, J., 1969. Speech Acts. Cambridge University Press, Cambridge.
Searle, J., 1979. Expression and Meaning. Cambridge University Press, Cambridge.
Searle, J., 1979. Literal meaning. Erkenntis, 13, 207-224.
Searle, J., 1983. Intentionality. Cambridge University Press, Cambridge.
Searle, J., 2001. Rationality in Action. MIT Press, Cambridge, MA.
Segal, E., Duchan, J., Scott, P., 1991. The role of interclausal connectives in narrative structuring.
Discourse Processes, 14, 27- 54.
Serrano, M.J., 1999. Bueno como marcador discursivo de inicio de turno y contraposición: Estudio
sociolingüístico. International Journal of the Sociology of Language, 140, 115-133.
Simon, A.-C., 2003. La structuration prosodique du discours en français. Une approche
multidimensionnelle et expérientielle. Lang, Bern.
Simon, H.A., 1962. The architecture of complexity. Proceedings of the American Philosophical
Society, 196, 467-482.
Simon-Vandenbergen, A.-M., 2000. The functions of I think in political discourse. International
Journal of Applied Linguistics, 10, 41-63.
References 493
Simon-Vandenbergen, A.-M., Aijmer, K., forthcoming. The expectation marker of course in a cross-
linguistic perspective. Languages in Contrast. Elsevier, Amsterdam.
Slobin, D.I., 1985. Introduction: why study language acquisition crosslinguistically. In: Slobin, D.I.
(Ed.), The Crosslinguistic Study of Language Acquisition. Lawrence Erlbaum, Mahwah, NJ, pp.
3-24.
Smith, I., 1985. Multilingualism and diffusion: a case study from Singapore English. Indian Journal
of Applied Linguistics, 11, 105-128.
Smith, S., Jucker, A., 2000. Actually and other markers of an apparent discrepancy between
propositional attitudes of conversational partners. In: Andersen, G, Thorstein, F. (Eds.),
Pragmatic Markers and Propositional Attitude. Benjamins, Amsterdam, pp. 207-237.
Smith, V.L., Clark, H.H., 1993. On the course of answering questions. Journal of Memory and
Language, 32, 25-38.
Sornicola, R., 1981. Sul parlato. Il Mulino, Bologna.
Sperber, D., 1974. Le symbolisme en général. Hermann, Paris.
Sperber, D., Wilson, D., 1986. Relevance: Communication and Cognition. Blackwell, Oxford.
Sperber, D., Wilson, D., 1995. Relevance: Communication and Cognition, second ed. Blackwell,
Oxford.
Spitzer, L., 1922. Italienische Umgangssprache. Kurt Schroeder, Bonn.
Sprott, R., 1992. Children’s use of discourse markers in disputes. Discourse Processes, 15, 423-439.
Stalnaker, R.C., 1973. Presuppositions. Journal of Philosophical Logic, 2, 447-457.
Stalnaker, R.C., 1978. Assertion. In: Cole, P. (Ed.), Syntax and Semantics 9: Pragmatics. Academic
Press, New York, pp. 315 – 332.
Stede, M., 2000. Discourse particles and discourse functions. Machine Translation Journal, 15, 125-
147.
Stein, D., 1985. DM in Early Modern English. In: Eaton, R., Fischer, O., Koopman, W., van der
Leek, F. (Eds.), Papers from the Fourth International Conference on English Historical
Linguistics, Amsterdam, 10-13 April, 1985. Benjamins, Amsterdam, pp. 283-302.
Stenström, A.-B., 1994. An Introduction to Spoken Interaction. Longman, London.
Stewart, I., 1980. The Seizing of Singapore. Hamlyn, London.
Svartvik, J., 1980. Well in conversation. In: Greenbaum, S., Leech, G., Svartvik, J. (Eds.), Studies in
English Linguistics for Randolph Quirk. Longman, London, pp. 167-177.
Sweetser, E., 1988. Grammaticalization and semantic bleaching. Berkeley Linguistics Society, 14,
389-405.
Sweetser, E., 1990. From Etymology to Pragmatics. Metaphorical and Cultural Aspects of Semantic
Structure [Cambridge Studies in Linguistics 54]. Cambridge University Press, Cambridge.
Tabor, W., Traugott, E.C., 1998. Structural scope expansion and grammaticalization. In: The Limits
of Grammaticalization. Giacalone Ramat, A., Hopper, P.J. (Eds.). Benjamins, Amsterdam, pp.
229-272.
Tatian, 1872. Lateinisch und altdeutsch mit ausführlichem Glossar herausgegeben von Eduard
Sievers [Bibliothek der ältesten deutschen Literatur-Denkmäler 5], Schöningh, Paderborn.
Tay, M. Wan Joo, 1988. Code switching and code mixing as a communicative strategy in
multilingual discourse. In: McCarthy, B. (Ed.), Asian-Pacific Papers [Applied Linguistics
Association of Australia, Occasional Papers 2] (. ALAA, Wollengong, pp. 43-57. [Repr. in World
Englishes 8 (3, 1989), 407-417 and in Tay Wan Joo, M.,1993. The English Language in
Singapore: Issues and Development. UniPress, Singapore.]
494 References
Thomason, S.G., Kaufman T., 1988. Language Contact, Creolization, and Genetic Linguistics.
University of California Press, Berkeley.
Thompson, G., Zhou, J., 2000. Evaluation and organization in text: the structuring role of evaluative
disjuncts. In: Hunston, S., Thompson, G. (Eds.), Evaluation in Text. Authorial Stance and the
Construction of Discourse. Oxford University Press, Oxford, pp. 121-141.
Thompson, S.A., Mulac, A., 1991. A quantitative perspective on the grammaticization of epistemic
parentheticals in English. In: Traugott, E.C., Heine, B. (Eds.), Approaches to Grammaticalization,
vol. 2. Benjamins, Amsterdam, pp. 313-329.
Thorne, B,, Henley, N., 1975. Difference and dominance: an overview of language, gender and
society. In: Thorne, B,, Henley, N. (Eds.), Language and Sex: Difference and Dominance.
Newbury House, Rowley, MA, pp. 5-41.
Timberlake, A., 1977. Reanalysis and actualization in syntactic change. In: Li, C.N. (Ed.),
Mechanisms of Syntactic Change. University of Texas Press, Austin, TX, pp. 141-177.
Tomasello, M., 1995. Joint attention as social cognition. In: Moore, C., Dunham, P. (Eds.), Joint
Attention: Its Origins and Role in Development. Lawrence Erlbaum, Mahwah, NJ, pp. 103-130.
Tongue, R., 1974. The English of Singapore and Malaysia. Eastern Universities Press, Singapore.
Traugott, E.C., 1989. On the rise of epistemic meanings in English: An example of subjectification in
semantic change. Language, 65, 31-55.
Traugott, E.C., 1995a. The role of the development of discourse markers in a theory of
grammaticalization. Paper presented at the 12th International Conference on Historical
Linguistics, University of Manchester, Manchester, England.
Traugott, E.C., 1995b. Subjectification in grammaticalization. In: Stein, D., Wright, S. (Eds.),
Subjectivity and Subjectivisation: Linguistic Perspectives. Cambridge University Press,
Cambridge, pp. 31-54.
Traugott, E.C., 1997. The discourse connective after all: a historical pragmatic account. Paper
presented at the Sixteenth International Congress of Linguists, Paris, July 1997.
Traugott, E.C., 1999a. The role of pragmatics in a theory of semantic change. Pragmatics in 1998:
Selected Papers from the 6th International Pragmatics Conference. International Pragmatics
Association, Antwerp, Belgium.
Traugott, E.C., 1999b. The rhetoric of counter-expectation in semantic change: a study in
subjectification. In: Blank, A. and Koch, P. (Eds.), Historical Semantics and Cognition. Mouton
de Gruyter, Berlin, pp. 177-196.
Traugott, E.C., 2000. Constructions in grammaticalization. In: Joseph, B., Janda, R. (Eds.),
Handbook of Historical Linguistics. Oxford University Press, Oxford, pp. 1-22.
Traugott, E.C., 2003b. From subjectification to intersubjectification. In: Hickey, R. (Ed.), Motives for
Language Change. Cambridge University Press, Cambridge, pp. 124-139.
Traugott, E.C., König, E., 1991. The semantics/pragmatics of grammaticalization revisited. In:
Traugott, E.C., Heine, B. (Eds.), Approaches to Grammaticalization [vol. I, Typological Studies
in Language 19]. Benjamins, Amsterdam, pp. 189-218.
Traugott, E.C., Dasher, R., 2002. Regularity in Semantic Change. Cambridge University Press,
Cambridge.
Traugott, E.C., König, E., 1991. The semantics-pragmatics of grammaticalization revisited. In:
Traugott, E.C., Heine, B. (Eds.), Approaches to Grammaticalization [vol. I, Typological Studies
in Language 19]. Benjamins, Amsterdam, pp. 189-218.
Travis, C.E., 1998. Bueno: A Spanish interactive discourse marker. Berkeley Linguistics Society, 24,
268-279.
References 495
Travis, C.E., 2003. The semantics of the Spanish subjunctive: its use in the Natural Semantic
Metalanguage. Cognitive Linguistics, 14, 47-69.
Travis, C.E., forthcoming a. Discourse Markers in Colombian Spanish: A Study in Polysemy.
Mouton de Gruyter, Berlin.
Travis, C.E., forthcoming b. The semantics and pragmatics of discourse markers: an analysis of
Spanish pues. In: Andor, J., Pelyvás, P. (Eds.), Empirical, Cognitive-based Studies in the
Semantics-Pragmatics Interface. Elsevier, Amsterdam.
Tsuchihashi, M., 1983. The speech act continuum: an investigation of Japanese sentence final
particles. Journal of Pragmatics, 7, 361-387.
Turner, R., 1974. Words, utterances and activities. In: Turner, R. (Ed.), Ethnomethodology. Selected
Readings. Penguin, Harmondsworth. , pp. 197-215
Txurruka Gomez, I., 2001. The Semantics of and in discourse. Report No. ILCLI-00-LIC-9 ILCLI.
University of the Basque Country, Spain.
Umbach, C., 2001. Contrast and contrastive topic. In: Kruijff-Korbayova, I., Steedman, M. (Eds.),
Proceedings of the ESSLLI 2001 Workshop on Information Structure, Discourse Structure and
Discourse Semantics. ESSLLI 2001, Helsinki.
Van der Auwera, J., 1993. Already and still: Beyond duality. Linguistics and Philosophy, 16, 613-
653.
Van der Sandt, R., 1992. Presupposition projection as anaphora resolution. Journal of Semantics, 9,
333-377.
van Dijk, T.A., 1977. Connectives in Text Grammar and Text Logic. In: Petöfi, J., van Dijk, T.A.
(Eds.), Grammars and Descriptions. Walter de Gruyter, Berlin.
van Dijk, T.A., 1979. Pragmatic connectives. Journal of Pragmatics, 3, 447-456.
van Dijk, T.A., 1980. Macrostructures. An Interdisciplinary Study of Global Structures in Discourse,
Interaction and Cognition. Lawrence Erlbaum, Hillsdale NJ.
van Dijk, T.A., 1980. Texto y contexto. Cátedra, Madrid.
van Dijk, T.A., Kintsch, W., 1978. Toward a model of text comprehension and production.
Psychological Review, 85, 363-394.
van Dijk, T.A., Kintsch, W., 1983. Strategies of Discourse Comprehension. Academic Press, New
York.
Vasko, I., Fretheim, T., 1997. Some central pragmatic functions of the Norwegian particles altså and
nemlig. In: Swan, T., Jansen Westvik, O. (Eds.), Modality in Germanic Languages. Historical and
Comparative Perspectives. Mouton de Gruyter, Berlin, pp. 233-292.
Veltman, F., 1996. Defaults in update semantics. Journal of Philosophical Logic, 25, 221-261.
Verbmobil Database, 1995. Data Collection for a Speech-to-Speech Translation System – Scheduling
Domain. Institut für Phonetik, München/Carnegie Mellon University, Pittsburgh, PA.
Verschueren, J., Östman, J.-O., et al., 1995. Handbook of Pragmatics. Benjamins, Amsterdam.
Verschueren, J., 1980. A la recherche d’une pragmatique unifiée. Communications, 32, 274-284.
Vincent, D., 1993. Les ponctuants de la langue et autres mots du discours. Nuit Blanche, Quebec,
Canada.
Visconti, J., 2003. From temporal to conditional. Italian qualora vs English whenever. In: Jaszczolt,
K.M., Turner, K. (Eds.), Meaning through Language Contrast. Benjamins, Amsterdam, pp. 23-50.
Vogel, P, Comrie, B., 2000. Approaches to the Typology of Word Classes. Mouton de Gruyter,
Berlin.
Volosinov, V.N., 1973. Marxism and the Philosophy of Language. Seminar Press, New York.
496 References
Vonen, A.M. (2000). Polynesian multifunctionality and the ambitions of linguistic description. In:
Vogel, P., Comrie, B. (eds), Approaches to the Typology of Word Classes. Mouton de Gruyter,
Berlin, pp. 479-488.
Vonk, W., Hustinx, L., Simons, W., 1992. The use of referential expressions in structuring discourse.
Language and Cognitive Processes, 7, 301-333.
Waltereit, R., 1999. Reanalyse als metonymischer Prozess, In: Lang, J., Neumann-Holzschuh, I.
(Eds.), Reanalyse und Grammatikalisierung in den romanischen Sprachen [Linguistische
Arbeiten 410]. Niemeyer, Tübingen, pp. 19-29.
Waltereit, R., 2001. Modal particles and their functional equivalents: A speech-act-theoretic
approach. Journal of Pragmatics, 33, 1391-1417.
Waltereit, R., 2002. Imperatives, interruption in conversation, and the rise of discourse markers: a
study of Italian guarda. Linguistics, 40, 987-1010.
Ward, G., Birner, B., 2001. Discourse and information structure. In: Schiffrin, D., Tannen, D.,
Hamilton, H. (Eds.), The Handbook of Discourse Analysis. Blackwell, Oxford, pp. 119-137.
Ward, N., 2000. Prosodic features which cue back-channel feedback in English and Japanese. Journal
of Pragmatics, 32, 1177-1207.
Watts, R., 1986. Relevance in conversational moves: a reappraisal of well. Studia Anglica
Posnaniensia, 19, 37-59.
Watts, R., 1989. Taking the pitcher to the well: native speakers’ perception of their use of discourse
markers in conversation. Journal of Pragmatics, 13, 203-237.
Wee, L., 1998. The lexicon of Singapore English. In: Foley, J.A., et al. (Eds.), English in New
Cultural Contexts: Reflections from Singapore. Oxford University Press, Oxford, pp. 175-200.
Wegener, H., 2002. The evolution of the German modal particle denn. In: Wischer, I., Diewald, G.
(eds.), New Reflections on Grammaticalization. International Symposium, Potsdam, 17-19 June,
1999. [Typological Studies in Language 49]. Benjamins, Amsterdam, pp. 379-394.
Weigand, E., Dascal, M., 2001. Negotiation and Power in Dialogic Interaction. Benjamins,
Amsterdam.
Weinreich, U., 1953. Languages in Contact. Mouton, The Hague.
Wells, G., 1985. Language Development in the Pre-school Years. Cambridge University Press,
Cambridge.
Werth, P., 1999. Text Worlds: Representing Conceptual Space in Discourse. Longman, London.
Weydt, H., 1969. Abtönungspartikel. Die deutschen Modalwörter und ihre französischen
Entsprechungen. Gehlen, Bad Homburg vor der Höhe.
Weydt, H., 1983. Aber, mais und but. In: Weydt, H., (Ed.), Partikeln und Interaktion. Niemeyer,
Tübingen, pp. 148-159.
Weydt, H., 1984. Techniques of request: In quest of its universality. In: Manning, A., Martin, P.,
McCalla, K. (Eds.), The Tenth LACUS Forum 1983. Hornbeam Press, Columbia, SC, pp. 333-
341.
Weydt, H., 1993. Was ist ein gutes Gespräch? In: Löffler, H., (Ed.), Dialoganalyse, IV. Referate der
4. Arbeitstagung. Niemeyer, Tübingen, pp. 3-19.
Weydt, H., 2001. Partikelforschung/Particules et modalité. In: Holtus, G., et al. (Eds.), Lexikon der
Romanistischen Linguistik. Niemeyer, Tübingen, pp. 782-801.
Weydt, H., 2003. (Warum) spricht man mit Partikeln überhaupt höflich? In: Held, G. (Ed.), Partikeln
und Höflichkeit Peter Lang, Frankfurt/Main, pp. 13-39.
Weydt, H., 1983. Partikeln und Interaktion. Niemeyer, Tübingen.
Weydt, H., 1989. Sprechen mit Partikeln. Mouton de Gruyter, Berlin.
References 497
Weydt, H., Hentschel, E., 1981. Ein Experiment zur Entwicklung der verbalen Interaktionsfähigkeit
bei Kindern. Zeitsschrift für germanistische Linguistik, 9, 326-331.
Weydt, H., Hentschel, E., 1983. Kleines Abtönungswörterbuch. In: Weydt, H., (Ed.), Partikeln und
Interaktion. Niemeyer, Tübingen, pp. 3-24.
Weydt, H., et al., 1983. Kleine deutsche Partikellehre. Klett, Stuttgart.
White, P., 1999. A quick tour through appraisal theory. Background paper for Appraisal Workshop
University of Ghent, March 1999.
White, P., 2000. Dialogue and inter-subjectivity: reinterpreting the semantics of modality and
hedging. In: Coulthard, M.J., Rock, F. (Eds.), Working with Dialogue. Niemeyer, Tübingen, pp.
67-80.
Wierzbicka, A., 1976. Particles and linguistic relativity. International Review of Slavic Linguistics, 2,
251-312.
Wierzbicka, A., 1986. A semantic metalanguage for the description and comparison of illocutionary
meanings. Journal of Pragmatics, 10, 67-107.
Wierzbicka, A., 1988. The Semantics of Grammar. Benjamins, Amsterdam.
Wierzbicka, A., 1991. Cross-cultural Pragmatics. The Semantics of Human Interaction. Mouton de
Gruyter, Berlin.
Wierzbicka, A., 1994. ‘Cultural scripts’: A new approach to the study of cross-cultural
communication. In: Pütz, M. (Ed.), Language contact and language conflict. Benjamins,
Wierzbicka, A., 1995. Lexicon as a key to history, culture and society. In: (Dirven, R., Vanparys, J.
(Eds.), Current Approaches to the Lexicon. Peter Lang, Frankfurt/Main, pp. 103-155.
Wierzbicka, A., 1996. Semantics: Primes and Universals. Oxford University Press, Oxford.
Wierzbicka, A., 1986. Special issue on particles. Journal of Pragmatics, 10, 5.
Wilkins, D.P., 1992. Interjections as deictics. Journal of Pragmatics, 18, 119-158.
Wilkins, D.P., 1995. Expanding the traditional category of deictic elements: interjections as deictics.
In: Duchan, J., et al. (Eds.), Deixis in Narrative: A Cognitive Science Perspective. Lawrence
Erlbaum, Hillsdale, NJ, pp. 359-386.
Wilson, D., Sperber, D., 1988. Mood and the analysis of non-declarative sentences. In: Dancy, J.,
Moravcsik, J., Taylor, C. (Eds.), Human Agency: Language, Duty and Value. Stanford University
Press, Stanford, CA, pp. 77-101.
Wilson, D., Sperber, D., 1990. Forme linguistique et pertinence. Cahiers de Linguistique Francaise,
11, 13-35.
Wilson, D., Sperber, D., 1993. Linguistic form and relevance. Lingua, 90, 1-25.
Wischer, I., Diewald, G., 2002. New Reflections on Grammaticalization. International Symposium,
Potsdam, 17-19 June, 1999. [Typological Studies in Language 49]. Benjamins, Amsterdam.
Wolfram, W., 1986. Good data in a bad situation: eliciting vernacular structures. In: Fishman, J.A.,
Tabouret-Keller, A., Clyne, M., Krishmamurti, B., Abdulaziz, M. (Eds.), The Fergusonian
Impact: in honor of Charles A. Ferguson on the occasion of his 65th birthday, vol. 2. Mouton de
Gruyter, Berlin, pp. 3-22.
Wong, J.O., 1994. A Wierzbicka approach to Singlish particles. Masters thesis, National University
of Singapore, Singapore.
Wong, J.O., 2000. The ‘mE’ Particle of Singlish. Pagesetters Services Pte Ltd, Singapore.
Wong, J.O., 2004. The particles of Singapore English: A semantic and cultural interpretation. Journal
of Pragmatics, 36, 739-793.
498 References
Yang, L.-C., 1995. Intonational Structures of Mandarin Discourse. Doctoral dissertation, Georgetown
University, Washington, DC.
Yang, L.-C., 2001. Prosodic shape and expressive meaning: the expression and recognition of
emotions through prosody. Proceedings of the ISCA Prosody and Speech Recognition Workshop.
Red Bank, NJ.
Yang, L.-C., 2003. Visualizing spoken discourse: prosodic form and discourse functions of
interruptions. In: van Kupervelt, J. , Smith, R. (Eds.), Current and New Directions in Discourse
and Dialogue. Kluwer, Dordrecht, pp. 355-378.
Yang, L.-C., 2005. Extracting meaning from context: modelling the prosody of oh in Mandarin
conversation. Proceedings of the 31th Berkeley Linguistics Society Conference, Berkeley,
California.
Zeevat, H., 2001. The semantic role of modal particles. In: van Rooy, R. (Ed.), Proceedings of the
13th Amsterdam Colloquium. ILLC, Amsterdam.
Zeevat, H., 2002. Explaining presupposition triggers. In: van Deemter, K., Kibble, R. (Eds.),
Information Sharing. CSLI, Stanford, CA, pp. 61-87.
Ziccolella, S., 1998. Turn initial components in courtroom talk: ‘alright’ and ‘well’. In: Cmejrkova,
S., et al., (Eds.), Dialoganalyse VI. Niemeyer, Tübingen.
Zifonun, G., Hoffmann, L., Strecker, B., 1997. Grammatik der deutschen Sprache. Mouton de
Gruyter, Berlin.

Approaches To Discourse Particles

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Approaches To Discourse Particles

Uploaded by

Copyright:

Available Formats

APPROACHES TO DISCOURSE PARTICLES

The Studies in Pragmatics series is dedicated to publishing

Other titles in this series:

AIJMER & SIMON- Pragmatic Markers in Contrast

HABERLAND, CAFFI, Future Prospects for Pragmatics

FETZER & FISCHER Lexical Markers of Common Grounds

Amsterdam • Boston • Heidelberg • London • New York • Oxford

First edition 2006

Copyright © 2006 Elsevier Ltd. All rights reserved

No part of this publication may be reproduced, stored in a retrieval system

British Library Cataloguing in Publication Data

Library of Congress Cataloging-in-Publication Data

For information on all Elsevier publications

Printed and bound in The Netherlands

Maj-Britt Mosegaard Hansen

Kent Bach, San Francisco State University, USA

1 Towards an understanding of the spectrum of approaches to discourse

Part I: Polysemy-based approaches

2 A dynamic polysemy approach to the lexical semantics of discourse markers

3 Discourse markers in English: a discourse-pragmatic view 43

4 The rise of discourse markers in Italian: a specific type of language change 61

5 A functional approach to the study of discourse markers 77

6 Pragmatic markers in translation: a methodological proposal 101

7 The description of text relation markers in the Geneva model of discourse

8 A dynamic approach to discourse particles 133

Part II: Monosemy-based approaches

9 A relevance-theoretic approach to discourse particles in Singapore English 149

10 From procedural meaning to processing requirement 167

11 Towards a theory of discourse markers 189

12 What are particles good for? 205

13 The Natural Semantic Metalanguage approach to discourse markers 219

14 Epistemic modalities and the discourse particles of Singapore 243

15 Integrating prosodic and contextual cues in the interpretation of discourse

16 Formal properties of a subset of discourse markers: connectives 299

17 Discourse marker research and theory: revisiting and 315

18 Discourse markers as attentional cues at discourse transitions 339

19 A dynamic-interactional approach to discourse markers 359

20 Discourse particles as morphemes and as constructions 375

21 Discourse particles and modal particles as grammatical elements 403

22 Frames, constructions, and invariant meanings: the functional polysemy of

23 Discourse markers in Italian: towards a “compositional” meaning 449

Towards an understanding of the spectrum of

The first challenge consists in providing a definition of discourse particles by describing

• the distinction between discourse particles and other particles,

• the many different functional interpretations discourse particles may support,

• how to account for the relationship between the different readings,

2. Results from the discussion

Figure 1. Discourse particles versus discourse markers

3. Towards an understanding of the spectrum of approaches

Figure 2. Dimension of integratedness

items integrated into host items constituting

Figure 3. Dimensions of integratedness and function

items integrated into host items constituting independent

connecting functions regarding

The dimension of integratedness thus corresponds in part to a focus on particular

Figure 4. Dimensions of integratedness, function, and data

items integrated into host items constituting independent

connecting functions regarding

spoken and conversation

Thus, structuring the spectrum of approaches to discourse particles according to the

Figure 5. Dimensions of integratedness, function, data, and host units

items integrated into host items constituting

connecting functions regarding