Professional Documents
Culture Documents
English Conversation New Conversation Corpus Analysis of Storytelling
English Conversation New Conversation Corpus Analysis of Storytelling
English Conversation New Conversation Corpus Analysis of Storytelling
General editor
Merja Kytö (Uppsala University)
Editorial Board
Bas Aarts (University College London), John Algeo (University of Georgia), Susan Fitzmaurice
(University of Sheffield), Christian Mair (University of Freiburg), Charles F. Meyer (University of
Massachusetts)
The aim of this series is to provide a framework for original studies of English, both present-day and
past. All books are based securely on empirical research, and represent theoretical and descriptive
contributions to our knowledge of national and international varieties of English, both written and spoken.
The series covers a broad range of topics and approaches, including syntax, phonology, grammar,
vocabulary, discourse, pragmatics, and sociolinguistics, and is aimed at an international readership.
Geoffrey Leech, Marianne Hundt, Christian Mair and Nicholas Smith: Change in Contemporary
English: A Grammatical Study
Jonathan Culpeper and Merja Kytö: Early Modern English Dialogues: Spoken Interaction as Writing
Daniel Schreier, Peter Trudgill, Edgar Schneider and Jeffrey Williams: The Lesser-known Varieties of
English: An Introduction
Hilde Hasselgård: Adjunct Adverbials in English
Raymond Hickey: Eighteenth-Century English: Ideology and Change
Charles Boberg: The English Language in Canada: Status, History and Comparative Analysis
Thomas Hoffmann: Preposition Placement in English: A Usage-based Approach
Claudia Claridge: Hyperbole in English: A Corpus-based Study of Exaggeration
Päivi Pahta and Andreas H. Jucker (eds.): Communicating Early English Manuscripts
Irma Taavitsainen and Päivi Pahta (eds.): Medical Writing in Early Modern English
Colette Moore: Quoting Speech in Early English
David Denison, Ricardo Bermúdez-Otero, Chris McCully and Emma Moore (eds.): Analysing Older
English
Jim Feist: Premodifiers in English: Their Structure and Significance
Steven Jones, M. Lynne Murphy, Carita Paradis and Caroline Willners: Antonyms in English:
Construals, Constructions and Canonicity
Christiane Meierkord: Interactions across Englishes: Linguistic Choices in Local and International
Contact Situations
Haruko Momma: From Philology to English Studies: Language and Culture in the Nineteenth Century
Raymond Hickey (ed.): Standards of English: Codified Varieties Around the World
Benedikt Szmrecsanyi: Grammatical Variation in British English Dialects: A Study in Corpus-based
Dialectometry
Daniel Schreier and Marianne Hundt (eds.): English as a Contact Language
Bas Aarts, Joanne Close, Geoffrey Leech and Sean Wallis (eds.): The Verb Phrase in English:
Investigating Recent Language Change with Corpora
Martin Hilpert: Constructional Change in English: Developments in Allomorphy, Word Formation, and Syntax
Jakob R. E. Leimgruber: Singapore English: Structure, Variation and Usage
Christoph Rühlemann: Narrative in English Conversation
CHRISTOPH RÜHLEMANN
Philipps University Marburg
University Printing House, Cambridge CB2 8BS, United Kingdom
Published in the United States of America by Cambridge University Press, New York
Cambridge University Press is part of the University of Cambridge.
It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9780521196987
© Christoph Rühlemann 2013
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2013
Printed in Spain by Grafos SA, Arte sobre papel
A catalogue record for this publication is available from the British Library
ISBN 978-0-521-19698-7 Hardback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
To my mother and my father
Contents
Introduction 1
1 Towards a working definition of conversational
narrative 4
1.1 Introduction 4
1.2 Genre 4
1.3 Participation 6
1.4 Temporal sequence 14
1.5 Agent orientation 22
1.6 Narrative structure 24
1.7 Recipient design 30
1.8 Sense-making 34
1.9 Summary 37
2 Data, methods, and tools 40
2.1 Introduction 40
2.2 The Narrative Corpus 40
2.2.1 Corpus construction 40
2.2.2 Corpus annotation 44
2.3 Methods and tools 63
2.3.1 Methods 63
2.3.2 Tools 65
3 How do narrators and recipients co-construct
turntaking? 76
3.1 Introduction 76
3.2 Co-construction of turn order 78
3.2.1 Introduction 78
3.2.2 Data and methods 82
vii
viii Contents
3.2.3 Results 85
3.2.4 Discussion 88
3.2.5 Summary 90
3.3 Co-construction of turn size 92
3.3.1 Introduction 92
3.3.2 Data and methods 93
3.3.3 Results 98
3.3.4 Discussion 104
3.3.5 Summary 108
4 Recipient design I: How do narrators mark quotation? 110
4.1 Introduction 110
4.2 Interjections as quotation markers 123
4.2.1 Introduction 123
4.2.2 Data and methods 126
4.2.3 Results 128
4.2.4 Discussion 133
4.2.5 Summary 141
4.3 Pauses as quotation markers 142
4.3.1 Introduction 142
4.3.2 Data and methods 144
4.3.3 Results 146
4.3.4 Discussion 151
4.3.5 Summary 153
5 Recipient design II: How do narrators use discourse
presentation for dramatization? 155
5.1 Introduction 155
5.2 Climactic structure at micro level: Sequential
ordering of report units in utterances 156
5.2.1 Introduction 156
5.2.2 Data and methods 157
5.2.3 Results 160
5.2.4 Discussion 167
5.2.5 Summary 171
6 How do recipients co-author stories? 173
6.1 Introduction 173
6.2 How do recipients influence narrator verbosity? 176
6.2.1 Introduction 176
6.2.2 Data and methods 178
6.2.3 Results 180
6.2.4 Discussion 184
6.2.5 Summary 187
Contents ix
x
List of figures xi
xiv
Acknowledgements xv
Further, I’ve learnt a lot from Stefan Th. Gries’s books on statistics in
corpus linguistics and his R workshop held in Denton/Texas in 2011. Stefan
was the first to demonstrate to me in fuller detail the breathtaking possibilities
R offers. I am also greatly indebted to him for invaluable comments on an
earlier version of the case study on turn order in Section 3.2 and for useful
tips given in response to queries posted to the Google Group “Corp Ling
with R” (http://groups.google.com/group/corpling-with-r) which he cre-
ated and maintains. I am likewise grateful for inspirational thoughts on earlier
versions of individual case studies contributed by Michael Hoey (Section 4.2)
and Rebecca Clift (Section 6.4). Special warm thanks are due to Merja Kytö,
general editor of the Studies in English Language series, and Helen Barton,
commissioning editor at Cambridge, whose endorsement all the way through
from proposal to publication was unfailing. The Cambridge University Press
production team too deserve a big thank you. Under the guidance of assistant
editor Fleur Jones, production editor Beata Mako and copy editor Cheryl
Prophett have greatly helped out the finishing touches to the manuscript.
Despite all these significant contributions, the usual disclaimer applies: all
remaining errors and weaknesses of this book are entirely my fault.
Finally, I thank my family, my wife Andrea and my sons Ricardo and
Lionel, not only for bearing with me during the years I have diverted much
of my attention to this book (rather than to them) but also for providing
the lively and loving environment without which such an endeavor would
hardly be possible. Also, I owe much to my father-in-law Fred and, even
more, to my parents, Günther and Jolanthe Rühlemann. It was undoubt-
edly their stories of experiences of love, hunger, despair, struggle, and hope
that taught me the centrality of storytelling to who we believe we are and
who we wish to be.
List of tags
Textual component
CPR = Pre-narrative conversation
CNN = Narrative
CNI = Narrative-initial utterance
CNF = Narrative-final utterance
CNI-CNF= Utterance which is both narrative-initial and narrative-final
CPO = Post-narrative conversation
Embed level
ES = Stand-alone narrative
EC = Narrative chain (with sublevels EC1, EC2, EC3)
EN = Nested narrative
Reporting mode
MDD = Direct
MDF = Free Direct
MII = Indirect
MIF = Free Indirect
MSS = Representation of Speech Act
MVV = Representation of Voice
MVT = Representation of Voice with Topic
MUU = Representation of Use
MRR = Anaphoric Reference to Discourse Presentation
MRQ = Request for Discourse Presentation
MXX = Unclear Mode
Participation role
PNU = Unsupported Narrator
PNS = Supported Narrator
xvi
List of tags xvii
Quotatives
QSB = Base form “say”
QSZ = Third-person singular present tense form “says”
QSD = Past tense form “said”
QSG = Progressive form “saying”
QSN = Past participle form “said”
QTB = Base form “think,” QTZ, QTD, QTG, QTN
QGB = Base form “go,” QGZ, QGD, QGG, QG
QBB = Base form “be like,” QBM = “ ’m/am like”, QBZ = “ ’s/is like”,
QBR = “ ’re/are like,” QBD = “was/were like,” QBG = “being like,”
QBN = “been like”
QAB = Base form “ask,” QAZ, QAD, QAG, QAN
QLB = Base form “tell,” QLZ, QLD, QLG, QLN
QLK = Quotative “like” without preceding form of “be”
QOO = Other Quotatives
Narrative type
T10 = First-person experience narrative
T1G = First-person generalized recurrent experience story
T1D = First-person dreamtelling
T1F = First-person fantasy
T1M = First-person mediated story
T30 = Third-person experience narrative
T3G =Third-person generalized recurrent experience story
T3D = Third-person dream
T3F = Third-person fantasy
T3M = Third-person mediated story
T3J = Joke
Introduction
Whether in a store, along the road, at work, play, home, or other community
settings, when people are together, they are inclined to talk about events –
those they have heard or read about, those they have experienced directly,
and those they imagine.
Ochs & Capps (2001: 1)
Narratives, it is widely claimed, abound in conversation. Ryave (1978: 113), for
example, refers to them as a “commonplace conversational activity,” Schiffrin
(1996: 167) views them as “a pervasive form of text,” Labov (1997: 396)
maintains that narratives “play a role in almost every conversation” (Labov
1997: 396), and Ochs & Capps (2001: 54) consider them “a ubiquitous feature
of ordinary conversation”. Moreover, it is claimed, stories serve critical
functions. As Ochs & Capps (2001: 17) note: everyday conversational narra-
tive is “a site for working through who we are and how we should be acting,
thinking, and feeling as we live our lives” (see also Schiffrin 1996: 167).
Bamberg (2004a: 332) sees narrative as configuring self and identity. Pang
(2010: 1322) considers a person’s self “a macronarrative subsuming all her
life-narratives.” Blum-Kulka (1993: 361) goes as far as to maintain that “the
essential nature of human beings is captured by the metaphor of man as homo
narrans.”
Given their (assumed) extraordinary frequency and social significance in
conversation, it is hardly surprising that a plethora of research has been
dedicated to the topic. Indeed, in discourse analysis, oral narrative is “one
of the most developed areas” (Schiffrin 1984: 314). Most analyses, though,
have been limited in terms of numbers of narratives considered (e.g.,
Schiffrin 1996), or have foregrounded non-conversational genres such as
professional storytelling (e.g., Leith 1995) or stories elicited in sociolinguistic
interviews (e.g., Labov 1972, Gwyn 2000). By contrast, stories from everyday
multi-party talk in conversation used to attract much less interest. Only
recent research has moved conversational narrative center stage. Both
Ochs & Capps’s (2001) pathbreaking volume as well as research into ‘small
stories’ (e.g., Bamberg 2004a, Georgakopoulou 2006a) have advanced the
theory of conversational narrative considerably. Another major step ahead,
1
2 Introduction
1.1 Introduction
In this chapter, I attempt to delineate conversational narrative. This is a
daunting task given that narrative “bows to no simple generic blueprint that
sets it apart once and for all from other forms of discourse” (Ochs & Capps
2001: 18). The intention here is not to elaborate an all encompassing defi-
nition, but rather to sketch the outlines of the object under investigation by
defining key properties of conversational narrative and discussing concepts
considered fundamental in oral and specifically conversational narrative
theory, thus setting the scene for the empirical chapters to come. The
full-fledged case studies in the analytical Chapters 3 to 6, it is hoped, will
facilitate a much richer and more adequate picture of conversational narrative
than is possible in this preliminary chapter.
Conversational narrative is approached from several angles. Section 1.2
locates conversational narrative in a genre framework. In Section 1.3 I outline
a participation framework for conversational narrative. Section 1.4 is
concerned with temporal sequence as the semantic backbone of narrative.
Section 1.5 emphasizes the importance of agent orientation. In Section 1.6,
I introduce the notion of narrative structure and discuss the structural
complexity of conversational narratives. Section 1.7 portrays recipient design
as a macrostrategy underlying narrative-discourse production. Section 1.8 is
concerned with the central function of storytelling to make sense, construct
identity, and propagate moral stance. No definition of conversational narra-
tive would be complete without discussion of co-construction. Since the
whole book is dedicated to the topic of co-construction it was felt unnecessary
to add a section on co-construction to this introductory chapter. I intend to
give sufficient evidence of co-construction in Chapters 3 to 6.
1.2 Genre
On a simple technical definition, conversational narrative refers to the stories
occurring in conversation. Conversation, it is widely agreed, is a core genre.
Its special status is owed to three main reasons. To begin with, conversation
4
1.2 Genre 5
1.3 Participation
A truism holds that “any tale involves a teller, and that, therefore, narrative
study must analyze two basic components: the tale and the teller” (Toolan
2001: 1). It will become very clear in this study that narrative in conversation
goes far beyond the supposedly essential dyad of tale and teller. Blum-Kulka
(1993) proposes a more adequate conceptual framework, including not
only tale and teller, but also telling: “the act of narrating in real time, the
actual performance of a story before an audience” (Blum-Kulka 1993: 363).
Although this definition of telling is helpful, the wording ‘before an audience’
is problematic in that it suggests a neat division between teller and recipients.
1.3 Participation 7
Conversation
→ Service encounters
Language-in-action
Telephone calls
General conversation
Conversational narrative
→ Big story
Small story
→ First-person experience
Third-person experience
Generalized recurrent experience
Fantasy
Dream
Joke
Retold story
Second story
Prayer
Untold stories
...
were in serious danger of being killed, where you said to yourself – “This
is it”’ ” (Labov 1972: 354; emphasis in original). This initial yes/no question as
well as the succeeding wh-question by the interviewer are arrowed in (1.1).
Note that in Labov (1972), the two interviewer questions are given in
parentheses, as in (1.1):
(1.1)
→ (You ever been in a situation where you thought you were gonna get
killed?)
Oh, Yeh, lotta time, man.
→ (Like, what happened?)
Well, like we used to jump off the trestle
And the trestle is about six-seven stories high.
You know, we used to go swimmin’ there . . .
We used to jump offa there, you know.
An uh-like, wow! Ya get up there
An’ ya feel like
You’re gonna die and shit, y’know.
Couple a times I almost . . . I thought I was gonna drown, you know.
(Labov 1972: 361)
The story in (1.1) is “minimally contextualized” (Edwards 1997: 140): the only
context the reader is offered is provided by the two elicitation questions,
which serve as standard stimuli in the interviews. Apart from the elicitation
questions, Labovian stories “report nothing (no talk or other conduct)”
(Schegloff 1997: 100) by the recipient(s) either in the course of the telling
or on the completion of the story, nor do they systematically record forms of
hesitations (but see the silent pauses in (1.1)) or major restarts. In short, “there
is nothing interactional in the data at all other than the eliciting question”
(Schegloff 1997: 101).2 That is, once successfully put on the track of telling a
story, the interlocutor turns mute and the teller is left to their own devices.
Labovian stories are thus essentially monologic and monophonic: they are
stories merely initiated by an interlocutor but not received and responded to
by a ‘participant’ re-actively taking part in them.
Given this minimal contextualization, narrative analyses based on this
type of data have attracted criticism. Schegloff, for example, argued that
“storytelling abstracted from its interactional setting, occasioning, and uptake
is an academically hybridized form” (Schegloff 1997: 104) and bemoaned
the “artificial environment of the academic elicitation” (Schegloff 1997: 105).
More recent narrative analyses have attempted to avoid the danger of decon-
textualizing stories by (i) turning away from interview-elicited stories to
naturally occurring stories in conversation and, driven by the view of narra-
tive as “an organic part of its interactional environment” (Schegloff 1997: 101),
by (ii) investigating stories with a focus on both their conversational embedding
1.3 Participation 9
N PNS : PRR R
Table 1.1 Characteristics of the recipient roles Responsive Recipient (PRR) and
Co-constructive Recipient (PRC)
sense that they contribute to the content of the unfolding story, for example,
by requesting or adding orientational information (when, where, to whom,
etc. the events happened), adding details to the unfolding account, comment-
ing on events in characteristic ways or contributing discourse presentation
(see Section 6.3). Thus, they are not only interactionally relevant, as tokens
of listenership, but also topically, as tokens of co-tellership interfering with
the content of the story. The role of PRC collapses the two basic roles of
narrator and recipient: PRCs can at the same time be considered co-tellers. In
a nutshell, while PRR responses primarily register understanding and thus
provide interactionally relevant feedback, PRC responses provide interac-
tionally as well as topically relevant feedback. The defining features of the two
recipient subroles are summarized in Table 1.1.
In shared narrations, three subroles of narrator are possible: Supported
Narrator (PNS), Primary Narrator (PNP), and Ratified Co-narrator (PNC).
Narrators assume the subrole of PNS if the only type of recipient response
they receive throughout the narration is responses by PRR (that is, tokens of
listenership focussing on the telling). If, by contrast, a recipient contributes
substantially to the unfolding narrative, thus assuming the role of PRC, the
current narrator is no longer seen as the only teller. Their role is reduced to
Primary Narrator (PNP). This role, as the name suggests, still does the bulk
of the telling but concedes to recipients co-telling space and co-
authorship. Tellers are also seen as Primary if their interlocutor is a PNC,
that is, a participant who has privileged access to the events underlying the
story (see Goodwin 1986b). Privileged access can be gained in three ways: (i)
the participant was involved in the events underlying the story or a witness to
them, (ii) they are familiar with the tale from previous tellings, or (iii) they co-
construct, on an almost equal footing with a PNP, a fantasy, where future or
hypothetical events are made up ‘on the fly.’ It goes without saying that
PNCs too may receive recipient support, either in the form of listenership
forms or substantial additions to or elicitations of story content.
12 Towards a working definition of conversational narrative
In the NC, narrator and recipient roles were assigned on differential bases.
Assignment of narrator roles was based on the global level of the story as a
whole, asking ‘Who does (the main bulk of) the telling?’, whereas recipient role
assignment was based on the local level of the individual utterance, asking
‘What does the response do with regard to the telling and/or the tale?’ Thus,
except for the very small number of stories where the participant who started
out as narrator ostensibly passes on the teller role to another participant,
narrators normally keep their role throughout the narrative while recipients
may freely shuttle within the same storytelling between recipient subroles
(PRR and PRC).
Consider an illustrative example. Speakers S1 (aged 65) and S2 (aged 67),
husband and wife, are talking about Carolyn, who works as a nurse in a
hospital, and the danger of having intruders in the hospital. Note that the
narrative in (1.2), as all narratives in this chapter, is given with all annotations;
these are explained in detail in Chapter 2. Additionally, the likely highpoint of
the story is marked by an arrow (which is itself not part of the annotation):
In (1.2), speaker S2’s role as PNP remains largely stable: S2 does most of the
telling in terms of initiating the topic, remembering and relating the details,
and steering the story to its highpoint (encoded in the arrowed utterance):
Carolyn’s (implicitly ascribed) recognition of her helplessness vis-à-vis
the intruder (There was no way that they could have summoned help).5 Even
in utterance 17 (Yes. Yeah), S2’s role can be seen as primary in that the
preceding polar question by S1 assigns to her privileged-access rights as the one
‘in the know’ of the story background information, and her answer, though
minimal in terms of number of words, grants the sought confirmation. Only in
utterance 15, in requesting a repetition of S1’s previous utterance, does S2 take on
a non-narrative role (labelled PX0; see Section 2.2.2).
Speaker S1, by contrast, shifts back and forth between two distinct roles,
that of PRC and that of PRR. Utterances 2, 4, 14, and 16 help advance the
narrative in terms of content. Utterance 2 is a request for completion of the
narrator’s still incomplete utterance, in which S2 initiates the narrative (And
do you remember the <pause/>). The request for completion of S2’s utterance
is thus, effectively, a request for completion of the story launch. In utterance
4, S1 contributes details to the story background (Well they’re often having
intruders) and encourages the main teller to continue (go on). In utterance 14
and, in repeated form, utterance 16, he seeks from the teller confirmation of
information that helps to contextualize the story events: the proposition
underlying the question Did n’t she ha well, to get from A to B did n’t she
have to leave the building somewhere? serves to emphasize the danger of getting
molested by intruders that Carolyn is exposed to in her work, thus helping to
add substance to the story highpoint. Conversely, in responses 6, 8, 10, and
12, S1 merely acknowledges receipt of and, particularly in the non-minimal
response in utterance 10, agreement with the narrator’s account. The excerpt
thus illustrates two things: it shows that a story can indeed be seen as the
outcome of collaboration and that participant roles in conversational narrative
are dynamic (see Ochs & Capps 2001: 3). The principle of dynamicity favors
participation shifts mostly in recipients, who may easily and repeatedly
switch between PRR and PRC. It may, however, also apply to narrators: in
some cases, speakers who start out as PNP yield their primary role to other
participants and take on a PNC role or even a more receptive role in the
background.
There is, however, more to participation than the actual participation of
the participants physically co-present in the telling situation. The pervasive
use in conversational storytelling of a single stylistic device opens up yet
another, more subtle, dimension to participation, viz. what I will call virtual
participation. This single device is direct speech, also referred to as quotation
and constructed dialog (Tannen 1986). As will be specified in much more
14 Towards a working definition of conversational narrative
detail in Section 2.2.2 and, specifically, Section 4.1, constructed dialog pre-
tends to give the exact locution that (could have) occurred in a non-present
situation, while, in actual fact, it depicts selected aspects of the reported
locution only. To uphold the pretense of faithfulness, constructed dialog
is designed as a likeness of real dialog. Design features include the use of
non-clausal material (such as interjections, vocatives, exclamations, etc.), the
imitation of delivery aspects (such as intonation, voice quality, false starts,
etc.), and, perhaps most importantly, the reporting speaker’s switch into the
reported speaker’s referential system, including the use of deictics which are
appropriate to the reported speaker and the reported situation but inappro-
priate to the reporting speaker and the reporting situation (see Levinson
2004: 111, Hanks 2011: 318). The construction of dialog along these lines
effectively creates the illusion that the reported speaker speaks for him/
herself in the reporting situation. This illusion has serious implications as
regards participation. First, the reported speaker is virtually translocated
from the non-present told situation to the present telling situation, where
he/she becomes a virtual participant. Second, given the virtual co-presence
of the reported speaker, the recipient of the constructed dialog becomes a
virtual addressee of the reported speaker’s utterance. The foundations are
laid for virtual turntaking between addressor and addressee. Indeed, as will
be shown in Section 6.3, recipients occasionally do respond to the reported
speaker’s utterance thereby entering into a virtual turntaking relationship
with them. Precisely why narrators seek to construct the recipient as a virtual
addressee will be discussed in more detail in Section 4.1. As regards partici-
pation, the observation and acknowledgement of virtual participation as a
subtle dimension complementing the physical dimension of actual partici-
pation, suggests that the complexity of participation in conversational story-
telling can hardly be overstated.
a-then-b-then-c
If the order were changed, say, to a-then-c-then-b, as in
a Yeah I forget one of his books yesterday
c and he clobbered me
b and I went home
an altogether different story would result.
The a-then-b relation is not the only one at work in narrative. One also “finds
implied relations between clauses such as a-and at the same time-b or a-and
now that I come to think of it-b” (Labov & Waletzky 1967/1997: 25). An
example of a narrative in which the events described did demonstrably not
occur in the same order as they were told in is (1.5):
(1.5) “Mark cried his eyes out” (Type: T1D / Embed Level: EC1)
CNN
CNI-CNF
S1 PNU I woke up, last night and Mark had had a really
bad dream he was crying his eyes out.
Three narrative clauses can be isolated:
a I woke up, last night
b and Mark had had a really bad dream
c he was crying his eyes out
Clearly, the mother woke up when and because she heard her son crying; that is,
the event described in a occurred after the event described in b and, partly, at the
same time as the event in c was occurring. To re-construct the actual temporal
order, we could re-arrange the narrative clauses thus (allowing for two minor
changes: deletion of and as well as change of past perfect to past tense in b)
b last night Mark had a really bad dream
c he was crying his eyes out
a I woke up
Why does the narrator, instead of choosing this temporal order, which
reflects the objective sequence of events, use the ‘flashback technique,’
which, according to Labov, is generally dispreferred in oral narratives
(Labov 1997: 411), thus reversing the temporal order of events? The temporal
reversal is arguably related to Labov’s claim that “[t]he viewpoint in oral
narratives of personal experience is that of the narrator at the time of the
events referred to” (Labov 1997: 411). The event described in a is, seen from
the mother’s viewpoint, indeed temporally prior to the event described in b,
that Mark had had a bad dream, which she most likely learnt of only at a later
point of time (see also van Dijk & Kintsch 1983: 57). The important point
illustrated by (1.5) is, then, that the a-then-b relation need not necessarily be
1.4 Temporal sequence 17
Filtering out clauses and, specifically, narrative clauses strictly relating tem-
porally junctured events is by no means straightforward and some decisions
taken in the analysis may well be disputable. For example, I have counted
clause 11 (So I was standing there) as a narrative clause although it is clear from
the context that the narrator had been standing at the place from where he
was watching the scene already some time before clause 11. However, a central
theme in the narrative is precisely the narrator’s determination to stay where
he was so as to be able to prevent the potential car theft and it may be argued
that it was after his suspicion was aroused that he decided to keep standing at
his location. If the clause So I was standing there is taken as expressing this
1.4 Temporal sequence 19
Although merely skeletal in that much detail is omitted, this basic represen-
tation of the story conveys a good sense of what happened. Narrative clauses,
then, seem to form the backbone of stories. Other types of clausal material
primarily serve to flesh that backbone out.7
Two further points are worth making, one related to sequence and lin-
earity and one related to sequence and co-construction. It is important to note
that sequentiality does not necessarily equate linearity: conversational stories
“do not uniformly thread events into a unilinear time line and cause-effect
progression” (Ochs & Capps 2001: 41). Chronologies in conversational nar-
ratives may be non-linear in the sense that tellers occasionally tell subsequent
events before preceding events, a partially reversed progression which may
be mainly due to memory gaps or memory failure (see Norrick 2005b).
As regards sequence and co-construction, it seems that for most narrative
subgenres including the most frequent one, viz. first-person experience
stories, recipients can exert a powerful influence on the production of narra-
tive clauses. As will be shown in Section 6.3, among the most common types of
contribution by Co-constructive Recipients (PRCs) are questions related to
the story in progress. The two structural subtypes of questions – yes/no and
wh-questions – interact in different ways with the exposition of narrative
events and thus the formulation of narrative clauses: event-related yes/no
questions seem to qualify as hypothetical narrative clauses in the sense that
a possible event is added to the existing sequence of events. Consider (1.7):
(1.7) “Garlic at school” (Type: T10/Embed Level: EC1)
CPR
(. . .)
CNN
(. . .)
S3 PRC So what happened when you going to have the
le, the lesson were you
S1 PNP Well, I was already in and he came in and just
erm I was here and he just suddenly came
round that way and he sat down next to Louise
and and he saidQSD [MDDhow are you? ] I
saidQSD [MDDI’m okay. ] [MDFYou look good]
or something. And then that was it really.
→ S3 PRC [MVV He chatted to Louise did he? ]
S1 PNP Yeah. Why break the habit of the lifetime eh?
(KBY-N1)
Excerpt (1.7) highlights an instance of a narrative event offered, in the form
of a question, by a recipient. A female student, S1, and her mother, S3, are
talking about a boy at college. Apparently, the student is disappointed at the
boy’s lack of interest in her. When she hints at her disappointment (And then
that was it really.) the mother correctly guesses who instead he showed an
1.4 Temporal sequence 21
interest in (He chatted to Louise did he ?), a suggestion her daughter answers in
the positive (Yeah. . . .).8 The narrative sequence can be reconstructed as
follows, with narrative clause f genuinely co-authored by recipient and
narrator:
a I was already in and he just suddenly came round that way
b and he sat down next to Louise
c and he said, “How are you?”
d I said, “I’m okay.”
e “You look good.”
f He chatted to Louise
Wh-questions by recipients, conversely, do not add hypothetical narrative
events. Their co-constructive role is in eliciting narrative clauses encoding
narrative events: Consider (1.8), where a school girl is relating to her mother
how a boy threw a scratch book at her:
(1.8) “Dangerous boy” (Type: T10 / Embed Level: EC2)
(. . .)
CNN
CNI
S2 PNP she’s got a big scratch book mum and she glued it on
and came that close to me it’s about there and he
threw it at me and [unclear] there. Can you see it?
→ S1 PRC Mm. What did you say yourself?
CNF
S2 PNP Told Mrs [name] , and she goesQGZ [MDDoh yes,
I know ] and she just walked off!
(KBJ-N1)
The main narrator starts out by relating three distinct events encapsulated in
three narrative clauses:
a she glued it on and
b came close to me and
c he threw it at me
The recipient’s question What did you say yourself? prompts the main teller to
add three more narrative events (her report of the incident to the teacher, the
teacher’s verbal reaction and how the teacher walked off). The question
whether or not the narrator would have added these clauses without the
recipient’s enquiry cannot be answered. What can be said is that the recipient’s
question renders their addition highly likely in that the question constitutes the
first-pair part of a question-answer adjacency pair, which “requires the pro-
duction of a reciprocal action (or ‘second pair part’)” (Goodwin & Heritage
1990: 287; added emphasis). That is, the question sets up “specific expectations
which have to be attended to” (Levinson 1983: 306).
22 Towards a working definition of conversational narrative
anybody can prepare the dish following the recipe. Narratives, conversely,
essentially revolve around agents, or actors (van Dijk & Kintsch 1983: 55),
and the events that occurred are depicted as events that occurred to them:
events are hardly deemed worth reporting unless they affect human beings.
Recipes are, then, pseudonarratives because the events they depict are not
spatiotemporally grounded and not related to any specific agent(s); they are
minus in respect to agent orientation (Longacre 1983: 3). Personal narratives,
conversely, are plus: they do not only relate a series of time-sequenced and
causally-related events but also embed the events in terms of when, where
and, crucially, to whom they happened (see Semino & Short 2004: 20, van
Dijk & Kintsch 1983: 55). Even in the exceedingly rare case of stories where
no human protagonists are mentioned, the narrativization of events is justi-
fied only by virtue of their implicated relevance to humans. Consider for
illustration Ervin-Tripp & Küntay’s (1997: 137) “Earthquake story” occa-
sioned by the 1989 Loma Prieta earthquake in California. The story is
presented in (1.10) in simplified form. Al and Ned are brothers, co-telling
to Olga and Cynthia about the disastrous earthquake:
(1.10) “Earthquake story” (simplified from Ervin-Tripp & Küntay 1997: 137)
Al: you know that that nice glass china display in our
dining room?
Ned: in the dining room
Cynthia: ooh
Al: trashed
Cynthia: forget it
Ned: absolutely trashed
Al: whole thing absolutely yeah every single bit of glass
and pottery in th
Olga: and crystal?
Ned: all the crystal trashed
Al: crystal
Ned: everything trashed
Cynthia: ooh my god
Al: oh a er antiques genuine antiques
→ Ned: and the amount of money we have lost is going to be
astronomical
As Ervin-Tripp & Küntay (1997: 137) observe, there is “no human protago-
nist”;9 the implied protagonist is, arguably, the earthquake that caused the
damage related in the story. As regards the lack of explicit reference to human
agents, the story is not dissimilar to the recipe: leaving aside Al’s reference you
know (in the initial utterance), which is addressed to the recipients in the telling
situation, the only reference to agents (or rather, in this case, patients) occurs in
the final utterance, where Ned mentions the amount of money we have lost, with
we probably referring not only to the two brothers but their whole family. On
24 Towards a working definition of conversational narrative
the face of it, the agent orientation in this story, then, may seem peripheral. In
actual fact, however, it is central, indeed the essence of the story. This is
because the damage the earthquake caused is, by implication, damage inflicted
on Ned and Al and their family. And even if this implication did not surface in
the storytelling (that is, if Ned had mentioned only ‘the amount of money
lost’), this orientation would be crucial, for whether china, crystal, and antiques
get trashed is self-evidently relevant to humans valuing such objects: it requires
no explicit mentioning. Thus, despite the relative lack of surface presence, the
agent orientation even in this text is fundamental.
Thus, to extend the above-used metaphor, while the temporal sequence is
the semantic backbone, the orientation to human agents can be seen as the
heart of storytelling.
part while the grant of the floor for an extended period constitutes the
second-pair part. Note that the third characteristic of story prefaces, to
provide information the recipient can use to know when the story is over,
is, in the Labovian model, attributed to the abstract; preface and abstract thus
conceptually overlap.
While the notion of narrative structure has a wide currency in research on
oral narrative, it is not undisputed in research on narrative in conversation. As
noted, the model grew out of work on storytelling in sociolinguistic interviews,
a genre undoubtedly governed by constraints different from those governing
storytelling in casual conversation. It is as yet an open question to what extent
conversationalists actually exploit this superstructure in constructing conver-
sational narratives (see Ervin-Tripp & Küntay 1997: 133, Edwards 1997: 139,
Georgakopoulou 1997). An example which illustrates narrative structure and
some of the issues associated with it is excerpt (1.11); speaker S1 is a 36-year-old
woman, S2 and S3 are two children aged eight and ten:
(1.11) “Spilt paint” (Type: T10/Embed Level: EC2)
CPR
S3 Mum!
S2 Mm.
----------------------------------------------------------------------------------
S3 Do you Story preface
S1 What?
S3 know what?
CNN
CNI
----------------------------------------------------------------------------------
S3 PNP Andrew [name] spilled paint all over! Abstract
S1 PRC What?
S3 PNP Just all over the table.
S1 PRR All over the table.
S3 PNP Mm.
S2 PRC Well I can [unclear] erm [unclear] over
[unclear].
S3 PNP Over everything was it?
----------------------------------------------------------------------------------
S1 PRC How did they manage that? Orientation
S3 PNP Cos they we we were, we were doing it
[unclear] and and wha they were all kinds of
[unclear]
----------------------------------------------------------------------------------
and it’s got [unclear] and it’s got [unclear] Complication
and I like, rushed it over to the sink and,
----------------------------------------------------------------------------------
1.6 Narrative structure 27
and then just [unclear] and then [unclear] Evaluation ?/
with the table it fell Resolution
S1 PRR My goodness!
S3 PNP No th
CNF
----------------------------------------------------------------------------------
S2 PRC I couldn’t have spilt it by goingQGG [MDD Coda
whoop! ]
In (1.11), the identification of some sections will be largely consensual while
the identification of other sections will be less so. For example, in the initial
utterance Do you know what? by the speaker who is to become the narrator,
we do find a structure that matches some of the properties of a story preface –
the property ascribed to story prefaces which is missing here is ‘information
about what it will take for the story to be over’ (see above). Also instantiated is
a clear Abstract (Andrew spilled paint all over, later complemented by Over
everything was it?) and an Orientation section filling in background informa-
tion (note the use of the progressive form in we were doing it). Identifying
the Complication section is already less straightforward. This section will
probably be seen as containing the event depicted in I like, rushed it over to the
sink, which leads up to what might be either the Evaluation section presenting
the most critical event or the Resolution section showing the event’s result
(with the table it fell). Also, only a few items potentially qualify as evaluating
devices; they include the repeated collocation all over, which has ‘pervasive
meaning’ (Quirk et al. 1985: 684), the verb forms rushed, implicating haste and
fell implicating that a mishap occurred, and just, perhaps used as a comparator
adverb. Finally, it is unclear whether consensus can be achieved over the
status of the Co-constructive Recipient’s utterance I couldn’t have spilt it by
going, “Whoop!”, which might serve as a sort of Coda in that it re-orients the
discourse to the present situation by relating the most reportable event (the
spilling) to one of the participants or as a comment made by the listener after
story completion, in which latter case there would be no Coda.
While, then, in the excerpt, some narrative sections seem rather firmly in
place, other sections cannot be identified with confidence. (However, note
the many instances of unintelligible speech scattered across the excerpt,
which make a definite interpretation of the excerpt and its structure
problematic.)
It would be easy to present narratives with even less overt structure or
where this structure seems to be missing altogether (see Georgakopoulou
1997: 10). In some narratives, even more complex structural patterns can be
observed (see van Dijk 1975: 291). In KB7-N2, for example, where the
main teller is narrating a violent fight of the couple living next door, the
telling is cyclical in the sense that basically the same story is told not only once
but re-told in three recycles, with the recycles separated from one another by
28 Towards a working definition of conversational narrative
intervening nested stories (see Section 2.2.2). Given the text’s unusual
length, only the original version, from which the recycled versions are
made, is printed in (1.12); the likely climaxes in each of the three cycles are
arrowed, and of the recycles and the nested stories separating them I give a
structural skeleton and note the story details they add (for the full text, see
Appendix 1; for another instance of a cyclically structured narrative, see
Appendix 4).
(1.12) “The couple next door” (Type: T10 / Embed Level: ES)
CPR
(. . .)
CNN
CNI
S1 PNP Oh that young couple next door Abstract
[unclear] ah!
S2 PRC Yeah? [MXXThey been at it, have
they been at it again? ]
S1 PNP Oh I wish you’d have been at
home, I’d have gone round there.
S2 PRC What’ve they been up to now
then?
----------------------------------------------------------------------------------
S1 PNP Oh god! [MVVShe’s been Complication
screaming. ] He’s been reckon she
must have locked herself up in the
dark. And the only place with a
lock on is the bathroom and
----------------------------------------------------------------------------------
I reckonQOO [MDDthey’ll Evaluation
blooming knock the blooming
door down ] by the sound of it.
S2 PRR Gawd.
→ S3 PNP [MVVShe was screaming, ] [MSShe
was swearing. ]
Running up and down the stairs.
----------------------------------------------------------------------------------
S2 PRC What time was this? Orientation
S1 PNP Quarter to ten till about half past
ten. Or half an hour till half past
ten [unclear]
S2 PRC Well that wouldn’t please them
next door would it?
S1 PNP No.
1.6 Narrative structure 29
S2 PRR Coo. [unclear]
.....................................................................
Nested story 1: People usually don’t intervene in domestic fights
.....................................................................
Recycle 1 Added details:
-it was morning when she first heard the Orientation
couple fight
-it made her feel cold Complication
→ -he was calling her names
-the dog was barking
.....................................................................
Nested story 2: Why two acquaintances, Ann and Rita, didn’t intervene in the
fight
.....................................................................
Recycle 2 Added details:
-she could hear the fighting although the Orientation
telly was on full blast
-she turned the sound down Complication
→ -the fighting caused her distress
.....................................................................
Nested story 3: Why the main teller once stayed in a similar violent relationship
.....................................................................
Recycle 3 Added details:
-the fighting ended about half past ten Resolution
-the man may have left the apartment
CPO
(. . .)
(KB7-N2)
In (1.12), the narrator relates how she overheard the violent argument of a
young couple next door, in the course of which, apparently, the woman
locked herself up in the bathroom while the man tried to knock the door
down. As can be seen from (1.12), each recycle adds some new details,
some related to background information (Orientation) (when it was that
the narrator first heard the fight and that she was watching television at the
time), others to the main events (Complication) (that the man was calling
the woman names, how the fight caused the narrator distress, etc.).
Crucially, while the original story contains Abstract, Orientation (albeit
delayed after Complication), Complication, and Evaluation, it clearly does
not contain a Resolution section outlining how the events sorted themselves
out. That section is postponed until recycle 3 – that is, as far as after two
recycles and three intervening nested stories. The overall structure is
schematically depicted in Figure 1.3.
30 Towards a working definition of conversational narrative
Stories: Original R1 R2 R3
N1 N2 N3
Sections: ACOC OC OC R
As shown in Figure 1.3, given that the Resolution which is missing from
the original version is provided in recycle 3, the original story and recycle 3
can be seen as a frame, binding the original and the three recycles
together. The structural complexity inherent in this frame structure is
already considerable. Yet, two more factors add to the structural complex-
ity. First, the section order is partly ‘scrambled’ in that ‘earlier’ sections
occur later; cf., in the original story, the Orientation section, which follows
the Evaluation section. Further, the nested stories each have narrative
sections too (see the full text in Appendix 1). Finally, as a consequence of
the cyclical structure, each cycle seems to have climactic tension and
hence in each cycle a (candidate) climax can be identified (arrowed in
the excerpt), which makes the story, not a single-highpoint, but a
multiple-highpoint story (see Section 6.3). The complexity of the overall
structure of the text can thus hardly be overstated. Beside cyclically
organized stories, another type of narrative whose structure diverges
from the orthodox Labovian-type of structure are (mostly long) stories
with chapterlike divisions. See the discussion of a relevant example in
Section 6.3.
The canonical Labovian-type structure is therefore far from being the
default structure in conversational stories. This reservation, however, does
not imply that narrative structure as such was irrelevant to conversational
narrative. As a culturally acquired schema, it cannot be expected to
perfectly dictate the way stories in conversation are structured. What it
does provide is a macrostructure to which speakers and recipients alike can
orient and which can aid them in producing and, respectively, processing
stories.
1.8 Sense-making
Why is a story told in the first place? What goals are participants pursuing
in co-constructing stories in conversation? These questions have been given
intriguing answers in recent research. For example, Johnstone portrays narra-
tives told by men and women as differential means to foster the ‘discourses’
of community (women) and contest (men) (Johnstone 1993), while in later work
(1997) she views narrative as a means of self-expression. Further, Holmes
notes, in passing, the ‘moral significance’ of stories (Holmes 1997a), a point
elaborated on in more detail in Walton & Brewer (2001). Most importantly
perhaps, Schiffrin (1996) and, more recently, Bamberg (e.g., 2004a, 2004b),
Georgakopoulou (e.g., 2006a, 2006b), and others have stressed the importance
of narrative in the construction of identity. Identity work is often achieved in
narratives via the sense-making function: “the essential function of personal
narrative – to air, probe, and otherwise attempt to reconstruct and make sense
of actual and possible life experiences” (Ochs & Capps 2001: 7).
1.8 Sense-making 35
Thus, seen in a wider perspective, the ‘point’ of a story is the sense that
participants can make of it: what the events reported signify in light of
personal and cultural scripts, values and expectations, and whether and
how script violation can be reconciled with the participants’ script systems
(see Daiute & Nelson 1997). Over and above the task of relating ‘inherently
tellable’ events, narratives fulfill functions infinitely more crucial: they are a
type of discourse through which we configure identity and self (Bamberg
2004a: 332). Pang (2010) even argues that the self itself is a composite
construct “constituted from a conglomeration of life-narratives” (2010:
13221–13222). The ways that stories configure self and identity (and, the
reverse, self and identity are stories themselves) are undoubtedly most
complex. One prime means by which stories play a role in identity formation
is as carriers of moral stance:
Everyday narratives of personal experience elaborately encode and perpetu-
ate moral worldviews. Personal narratives generally concern life incidents in
which a protagonist has violated social expectations. Recounting the violation
and taking a moral stance toward it provide a discursive forum for human
beings to clarify, reinforce, or revise what they believe and value. (Ochs &
Capps 2001:46)
Consider (1.14). S3, an 18-year-old male insurance clerk, is telling a delicate
story to two female students aged 13 and 16 about what appears to be a common
friend or acquaintance of theirs, Greg: a packet of condoms fell out of his
pocket at work. While the repeated instances of laughter both by the teller
as well as the recipients indicate a good deal of schadenfreude about Greg’s
embarrassment, the discourse immediately following the recount of the events
strikes a more serious tone. In the post-narrative component (CPO), it seems
that the participants are trying to find an explanation for Greg’s use of
condoms, which, we may assume, is seen as morally dubitable and which,
then, casts doubts over his moral credibility. The joint attempt to make sense
of Greg’s behavior is, thus, motivated by the participants’ desire to reconcile
their view of Greg, whom they value as a friend, and the suspicious behavior,
which is in violation of what they expect of him or, maybe more likely, what
they know others expect of him. Excerpt (1.14) is hence an illustration of the
observation that “[o]nce a person’s comportement is incorporated into narra-
tive, it is portrayed in relation to standards of right and wrong and is vulnerable
to public moral accountability” (Ochs & Capps 2001: 225).
(1.14) “Dropped your johnnies” (Type: T10 / Embed Level: EC2)
CNN
CNI
S3 PNP (singing) [unclear] Oh it was so funny at
work today, Greg fell off his chair.
S1 [. . .]
36 Towards a working definition of conversational narrative
S3 PNP Packet of condoms fell out of his pocket
[unclear]
S1 [laugh]
S3 PNP [laughing] And they were ripped .
S1 [laugh]
S3 PNP Ah no he was, he, he wouldn’t sit on his chair
[MSScos he’d just called me an arsehole ] and
I goesQGZ [MDDoh sit down [unclear] Greg! ]
I saidQSD [MDDsit down Gregory and shut
up.] So he went to sit down but his chair
weren’t there. All I saw were this pair of legs
sticking over the desk and him goingQGG
[MDDaaaaagh! ]
S2 PNC And his condoms [unclear]
S3 PNP And he got up and then one of the girls
saidQSD [MDDhi Greg dropped your
johnnies. ]
UN [laugh]
CNF
S3 PNP I’ve never seen anyone go so red in my life.
CPO
S2 How old’s he?
S3 Twenty three. He’s married.
S1 Well, so?
S3 He got married si what three months ago.
S1 Maybe he doesn’t want any children yet.
(KCE-N1)
1.9 Summary
In this initial chapter I have attempted to delineate key characteristics of
conversational narrative thus laying the ground for the empirical chapters to
come.
I argued that conversational narrative can be seen as perhaps the most
clearly discernible subgenre of conversation. Conversational narrative was
shown to represent a wide range of subgenres, including not only first-person
and third-person experience stories, as the most common forms, but also
an apparently open class of less prominent subtypes of stories such as
generalized experience, retold, fantasy, prayer, and so forth. I suggested
that, generally, conversational narratives are ‘small stories’ characteristically
concerned not with spectacular, extraordinary experiences but mundane and
often trivial-seeming happenings.
In outlining dimensions of narrativity, I followed Blum-Kulka’s (1993)
threefold model comprising the dimensions teller, tale, and telling. I defined
storytelling as the actual performance of a tale to and with recipients, thus
according recipients a central part in the emerging tale. Focussing on the
interaction of teller and recipients, a framework of participation in conversa-
tional narrative was elaborated. The framework proposed is based on a
distinction between actual participation and virtual participation. As regards
actual participation, the framework recognizes teller and recipient as hyper-
nyms of a broad range of subroles. Two recipient roles were distinguished:
Responsive Recipient producing tokens of listenership targeted at the telling,
and Co-constructive Recipient contributing content-oriented responses
with the potential of directly co-authoring the tale. Narrator subroles were
distinguished in close relation to these recipient types: Unsupported Narrator
receiving no feedback whatsoever from recipients, Supported Narrator receiv-
ing Responsive Recipient responses only, Primary Narrator receiving not
only Responsive Recipient utterances but also at least one Co-constructive
Recipient response, and, finally, Ratified Co-narrator, making contributions
to a Primary Narrator’s telling that exhibit privileged access to story
information. Finally, virtual participation was depicted in relation to the
pervasive use of constructed dialog, whose defining feature is the creation of
the illusion that the utterance reported is being produced by the reported
speaker themselves in the present telling situation. This illusion entails the
emergence of two virtual roles: while the reported speaker becomes a virtual
addressor, the recipient is made a virtual addressee.
Further, I discussed the role of the temporal sequence of events, which is
considered the most central semantic characteristic of narrative. I also
introduced Labov & Waletzky’s (1967/1997) a-then-b relation, a formal
term used to describe the succession of narrative clauses relating temporally
junctured events. The a-then-b relation was portrayed as the backbone of
narrative. It was noted that even recipients have recourse to the sequence of
38 Towards a working definition of conversational narrative
2.1 Introduction
This chapter is divided into two major subsections. Section 2.2 describes the
construction and annotation of the NC, the corpus of conversational narra-
tives, on which the analyses in this book are based. The account given here is
brief; see Rühlemann & O’Donnell (2012) for a more detailed description. In
Section 2.3, I outline major methods and tools used in this study. Specifically,
I introduce XPath and XQuery, two related query languages which operate
on the hierarchical tree structure of XML texts and which were consistently
used in generating data for the analytical chapters.
40
2.2 The Narrative Corpus 41
was achieved for the large majority of texts. The social composition of the NC
is, indeed, a fair reflection of the composition of the BNC-C; for more detail,
see Rühlemann & O’Donnell (2012).
As shown in Table 2.1, there are 500 speakers involved in the narrative
components of the texts in the NC, of whom 212 are female and 173 male.
The discrepancy is an echo of the small built-in speaker sex bias character-
istic of the BNC-C.1 The bias also comes to the fore in the number of
words spoken by men and women: female participants (44,476/56%) talk
almost twice as much as male participants do (24,268/31%). The greater
verbosity of women in the NC is also reflected in the greater mean number
of words: women produce, on average, 210 words, as compared to men’s
140 words (for a more detailed discussion and preliminary hypotheses to
explain the female overrepresentation in narrative, see Rühlemann &
O’Donnell 2012).
As can be seen from Table 2.2, the number of speakers not coded for class
(UU) is substantial: 272, representing 54% of all speakers. To judge from the
speakers with class information, the distribution of the social classes in the NC
too is similar to that of the BNC-C. The numbers of words per participant are
reasonably similar across all four classes. As regards absolute numbers of
participants per class, speakers coded for social class DE – that is, semiskilled
Table 2.1 Distribution of male and female participants in the narrative (CNN)
components of the NC
and unskilled manual workers, e.g., bus drivers, laborers, barmen – seem to be
underrepresented: not only are they far less numerous (39/8%) but they also
contribute far fewer words (7,625/10%) than speakers coded for the classes AB
(11,991/15%), C1 (15,239/19%), and C2 (11,064/14%). This might suggest that,
as regards class DE, the NC lost the BNC-C’s social balance out of sight. In
actual fact, however, DE speakers are similarly underrepresented in the
BNC-C (see Hoffmann et al. 2008).2 In other words, the social-class data in
the NC are a good match and reflection of the data in the BNC.
A similar distribution can be observed for speakers across the age param-
eter. Only one age group, viz. children in age group 0, are underrepresented
in the NC – just as they are in the BNC-C. All other age groups are
reasonably evenly represented.
On the whole, it seems fair to say that the social make-up of the NC is a
good fit of the social make-up of the BNC-C and hence that the social balance
characteristic of the BNC-C was also achieved for the NC.
Data extraction
Since the BNC contains annotation of parts of speech (POS) only, an automatic
retrieval of narratives was not possible. Instead, two main types of extraction
techniques were used: lexical extraction and extensive file reading.
Lexical extraction techniques were used by way of conducting searches in
the BNC-C for lexical strings that either the literature or inspection of
concordance lines suggested were frequent in storytelling. The strings used
included, for example, ‘(there’s/was) this guy/bloke/girl/man/woman,’ ‘it
was so (funny, weird, etc.),’ ‘did I tell you,’ ‘reminds me,’ ‘anyway,’ ‘sud-
denly,’ the interjections ‘bloody hell’ and ‘oh my god,’ the lemma
REMEMBER and so on.
Also, narratives were extracted by extensive horizontal reading of files in
the BNC-C, particularly those files for which the lexical ‘hooks’ did not help
retrieve any candidate texts.
2.2 The Narrative Corpus 43
Selection criteria
Three criteria were critical for conversational passages to be included in the
NC as narratives of which all three had to be fulfilled. The first such criterion
was ‘exosituational orientation.’ That is to say that linguistic evidence was
found that a speaker was referring to a situation different than the present,
storytelling, situation. In personal experience stories, exo-indicators include,
for example, past tense verbs, past-time deictics (this morning, yesterday, etc.),
remote-place deictics and reference to referents not present in the telling
situation. In fantasies, by contrast, which project future events, exosituational
orientation is typically realized by future time reference and/or expressions
of hypotheticality. Exosituational orientation in personal experience stories is
exemplified in (2.1); exo-indicators are in square brackets:
(2.1) “Rugby injury” (Type: T10/Embed Level: ES)
CPR
(. . .)
CNN
CNI
S1 PNP I [had to take [Tony] [to the casualty] on [Saturday]]?
S3 PRC Why?
In (2.1), the incipient narrator, S1, steers the discourse ostensibly away from
the present speech situation by referring to (i) a past event (had to take Tony
to the casualty on Saturday), (ii) a third-person character (Tony), who, upon
further reading, turns out to be absent from the present situation, (iii) a
remote location (the casualty), and (iv) a past weekday (Saturday). Thus, a
situational frame is established transferring the ‘scene’ from the here-and-
now to a narrative there-and-then.
The second criterion was Labov & Waletzky’s (1967/1997) ‘a-then-b’
relation (see Section 1.4), that is, the use of at least two temporally junctured
narrative clauses.3
Finally, inter-rater agreement had to be achieved between two independ-
ent researchers. Texts were excluded when the raters did not agree.
The texts of a few BNC-C files did not meet the criteria and hence no
narratives could be extracted from them. As shown in Table 2.4, a total of 279
texts were finally extracted from 143 BNC-C files. The NC contains 531
narratives and roughly 150,000 words.
Sampling
Since narratives are large linguistic units that unfold discursively, sampling
techniques that provide for same-length text samples were seen as inappro-
priate. Instead, narratives were included as wholes (see Section 2.2.2).
Moreover, given that a characteristic of conversational narratives is the use
of techniques to ensure coherence between turn-by-turn talk and stories both
at the beginning of stories and at story exit points (see Jefferson 1978) both
pre-narrative stretches and post-narrative stretches of turn-by-turn talk were
also included in the texts. The basic structure of the texts in the NC is thus
threefold:
Pre-narrative → Narrative → Post-narrative
As a rule, both the pre-narrative and post-narrative stretches included in the
texts were 15 utterances long. In many cases, however, this length could not
be achieved, for example, when stories succeed each other there are typically
very few non-narrative lines, if any, in between the stories. Also, the non-
narrative passages were not calculated over <div> (textual division)
boundaries.
XML editor
<oXygen/>XML Editor4 is a complete cross platform XML editor provid-
ing the tools, inter alia, for XML authoring. It also supports browsing,
managing, and querying XML databases. A special feature of this editor is
that all files are associated with a schema in which the annotations are laid
down and which automatically checks XML validity and well-formedness of
annotations implemented in the files. <oXygen/> was used for implement-
ing the annotations in the texts and for performing XPath queries, an
extremely useful query language for well-formed XML texts such as the
NC described in more detail in Section 2.3.2.
Tagsets
The tagsets used in the annotation of the NC were designed in accordance
with Leech’s (1997) ‘standards’ for corpus annotation. That is, the labels used
were intended to be concise, consisting mostly of no more than three
characters, perspicuous (that is, easy to interpret), and analyzable (that is,
decomposable into their logical parts). Consider, for example, the tag QSZ
(for quotative says). The tag is read from left to right: the left-hand character
designates the attribute (Q stands for quotative); the middle character
2.2 The Narrative Corpus 47
identifies the attribute (S stands for the verb SAY), whereas the right-hand
character identifies the value the attribute is given (Z stands for the third
person present tense form). This hierarchical left-to-right ordering is useful
in searches using wildcards. For example, a search for QS* finds all forms of
the lemma SAY used in quotative function, a search for Q* returns all verbs/
forms marked up as quotatives.
The annotation scheme provides for markup on five different levels: text,
textual components, utterance, speaker, and part of speech (POS). Readers
will be sufficiently familiar with speaker meta-information and POS-tagging.
These types of annotation are therefore not addressed in the following. The
remaining three levels are briefly explained in the order in which they are
located in the XML hierarchy. I begin with annotation related to text type.
Text
Annotation of the NC at text level is intended to grasp characteristics of
narratives as textual wholes. In XML, this information is recorded at the
level of the <div> element. In the NC, the <div> element records three
attributes. The first such attribute is the narrative’s ‘title,’ given by the
researchers. In devising the titles it was attempted to use labels which were
both concise (consisting of no more than one to four words) and precise
(capturing what the story is about). Further, <div> level annotation includes
information on the type of embedding of narratives included in the text.
Three main types of embedding are distinguished: narratives can stand alone
(that is, no response stories were found in the 15-utterance spans following
the story), they can be part of a sequence of narratives (what is referred to
here as a ‘narrative chain’), or they can be contained within a larger narrative
(‘embedded’).
The tagset for embedding includes these values:
Tagset Embedding
E ES Stand-alone narrative
EN Nested narrative
EC Narrative chain
EC1, 2, . . . First, second, . . . narrative in narrative chain
(2.3) shows the title (“Fathers and daughters”) given to text KBE-N2 and its
identification as a narrative chain (EC):
(2.3) <div title="Fathers and daughters" embedLevel="EC">
(KBE-N2)
Thirdly, <div> level annotation includes information on narrative sub-
genre. The annotation of text type in the NC includes not only identifica-
tion of first-person-experience stories but also a number of ‘non-canonical’
types of narrative. These include: third-person experience stories, which
48 Data, methods, and tools
are “told not from events remembered first-hand but about someone else”
(Norrick 2000: 149); ‘generalized recurrent experience stories,’ that is
“narrative-like representation of a recurrent shared past experience in
generalized form without reference to any specific instance” (Norrick
2000: 151); dreamtellings (Shanon & Eiferman 1984, Norrick 2000); ‘fanta-
sies,’ projecting future events or constructing a fictional world (Norrick
2000, Georgakopoulou 2006a, b); ‘jokes’ (Sacks 1978, Norrick 1994, 2000);
and, finally, what I will call ‘mediated tellings,’ including film retellings and
stories about texting and MSN interactions (see Blum-Kulka 1993,
Georgakopoulou 2008). The tagset for narrative subgenres includes these
values:
Table 2.5 Distribution of the 531 narratives in the NC across narrative types
No. of
Experiencer 1st person No. of stories % 3rd person stories %
narratives (see Section 2.2.1). That is, the boundary between pre-narrative
conversation and narrative was defined as that utterance in which a speaker
observably re-adjusts the frame of reference from the present telling situation
to a non-present told situation; for illustration, consider example (2.4). The
speakers are a group of teenagers of African descent, who have been discus-
sing the role of race in Hollywood films. In the exchange leading up to the
utterance tagged as narrative-initial utterance (CNI) the discourse is clearly
focussed on present states of affairs, as, for example, evidenced by the
frequently occurring present tense forms. In CNI, however, although the
present tense is retained at first (Kerry’s saying) and only later exchanged for a
past tense form (Truno was going), the focus shifts to events in a specific
situation in the past, viz. Kerry’s and Truno’s exchange of opinions on
Malcolm X:
(2.4) “Malcolm X” (Type: T30/Embed Level: EC1)
CPR
(. . .)
S6 Eddie Murphy films are good though [unclear]
S1 And Whoopie Goldberg and they’re brilliant actor
S3 She, she’s done some brilliant anti-racist films
S1 yeah they.
S2 But the thing is even people that, even though people
are saying, well you’ve gotta get more black people into
it, but if [unclear]
S4 You know, there’s, there’s a law in America.
S1 But that’s, but that’s what I mean, they’re making,
they’re saying let’s make, when the white people make
films black and white people watch them, they’re not
specifically made for white people, not specifically made
for black people, it just happens to be white actors. But
when they make a black film, they make a black film for
black people to sit and watch.
CNN
CNI
S1 PNP I’ve got it all on this tape [MDFMalcolm X], Kerry’s
saying QSG [MDDMalcolm X is brilliant, he was a black
man, he was a nigger ] and Truno was going QGG
[MDDshut up ] Truno’s black too and she was going QGG
[MDDshut up! Malcolm X, stupid black nigger, ] do you
know what I mean. But because they’re black
it don’t make no difference, do you know what I mean.
S3 PRC [MUUCos if we say it ]
(. . .)
(KPG-N1)
2.2 The Narrative Corpus 51
In story rounds, the boundary was set in that utterance which effected a
re-orientation from the sequence of events occurring in one told situation
to the event sequence occurring in another. The principle on which
decisions underlying text segmentation rested was thus clear-cut. Once
story boundaries were determined, determination of micro-components
followed automatically, as first (narrative-initial) utterance and last
(narrative-final) utterance; micro-componential annotation can thus, in
itself, be considered uncontroversial.6 The componential structure is
shown in Figure 2.1.
The rationale for annotating the major textual components CPR, CNN, and
CPO was as follows. Annotation of pre-narrative conversation (CPR) and post-
narrative conversation (CPO) was included because the shifts in the generic
framework from general conversation to narrative to general conversation
along with their repercussions on lexis, turntaking, topic, etc. (see Jefferson
1978) can thus be examined corpus-linguistically.
The main rationale for annotating the micro-components CNI and CNF
stems from recent research into textual positioning, revealing structural
associations between lexis and position in text, referred to as textual
Participant status
Utterance elements received attribute-value pairings for participant status.
Building on the participation framework outlined earlier, the tagset used for
participant status includes the following values:
Tagset Participant Roles:
For a detailed description of the concepts underlying the codes the reader is
referred to Section 1.3. The code PXX is used for utterances that are
narrative-related but whose status can not be determined with certainty;
the code PX0 designates utterances which are related not to the narrative
but some event occurring outside of the narrative, e.g., greeting a new
participant, admonishing a child, etc. The statistics for participant values
are shown in Table 2.6.
As noted, the theme of this study is the co-construction of narrative, that
is, the assumption that stories are interactionally achieved. The breakdown of
participant roles shown in Table 2.6 lends support to this assumption.
Unsupported Narrators (PNU), the label used in the NC which corresponds
most closely to Labov & Waletzky’s concept of the single teller, have very low
shares both in terms of number of words and utterances. PNU utterances
contain just 2,316 words, accounting for 3% of all words in the narrative
components (CNN). Only 45 utterances (accounting for 1% of all utterances
in CNN) are recorded for Unsupported Narrators (PNU). By default, one
54 Data, methods, and tools
PNU 2,316 3 45 1
PNS 10,009 13 288 5
PNP 52,807 67 2,488 42
PNC 3,783 5 450 7
PRR 1,677 2 1,264 21
PRC 6,270 8 919 15
PX0 1,632 2 313 5
PXX 329 0 293 5
Total 78,823 6,060
50
Recipient types
40
34.75
30
Mean
21.23 21.07
20
10
8.41
6.82
3.64
1.33
0
length continues to PNC, who, like PNP, can interact with any other narrator
type: their utterances are, on average, only 8.41 words long. The differences
are (mostly very highly) significant.8 Recipient responses to storytelling are
again much shorter than utterances by any narrator type: minimal, telling/
teller-related responses (PRR) are, on average, 1.33 words long, while
substantial, tale-related responses (PRC) are 6.82 words long – again a very
highly significant difference (see Section 2.3.2). (I will return to the differ-
ences in utterance lengths in Section 3.3.)
Discourse presentation
In annotating discourse presentation, two aspects were foregrounded:
(i) quotatives, that is, the verbs or other forms used to ‘announce’ that the
discourse forthcoming is not to be attributed to the speaker in the present
situation but to a speaker (who may be the same as the present speaker) in a
remote situation, and (ii) reporting mode, that is, the stylistic choice made by
the speaker in presenting the discourse. Note that quotatives are relevant for
two types of reporting mode only, Direct mode and Indirect mode. All other
forms of reporting are quotative-free.
There is broad agreement that, in conversation, a small pool of forms
dominate the quotative system despite the current “large-scale reorganization
56 Data, methods, and tools
Quotative Tag
SAY QS*
THINK QT*
GO QG*
BE like QB*
Like QLK
ASK QA*
TELL QL*
Other QOO
2.2 The Narrative Corpus 57
Quotative lemmas
SAY GO THINK TELL ASK BE like like Other
(2.9) MII: you know this erm, you know this young girl that was killed
along Benji Avenue with her mother ? Well I thought QTD
[MII it was Wendy ’s daughter] (KCP-N1)
(2.10) MUU: PS549> Oh aye, aye. Aye, but . . . you know . . . like you se –
[MUUshe mentioned one in particular], like
PS54D> What?
PS549> [MUU the word skeilth ]
PS54D> [MUU Skeilth? ]
PS54E> Mm. (KPD-N1)
(2.11) MSS: Milk up [MSS Nicola did n’t believe me when I told her.]
She thought QTD [MII I was playing a joke .] (KPN-N2)
(2.12) MVT: Well actually on the television I do n’t know whether you’ve
noticed when [MVT they start talking about burglaries
and things like that ] the picture that comes on is Newlands
Park you know (KDY-N1)
(2.13) MVV: [MVV She has talked, we’ve talked to her yesterday, ] was it
yesterday ? (KDS-N1)
(2.14) MRR: and he said QSD [MDDif you ain’t gone by then ] he said QSD
[MDDI’ll smash your fucking face in, ] [MRR that’s exactly
what he said to her.] (KC1-N1)
(2.15) MRQ: [laughing] [MRQ Well what did Mrs [name] say ? ] (KBW-
N2)
A total of 2,712 reporting mode values have been assigned altogether. Direct
discourse presentation, tagged MDD, is by far the most frequent reporting
mode in the NC. Figure 2.4 lists the observed frequencies and the respective
percentage values of all ten tags used for reporting modes.
The two direct modes, Direct (MDD) (58%) and Free Direct (MDF)
(8%), combined account for two-thirds (66%) of all discourse presentation
units in the NC, whereas the shares for the two indirect modes, Indirect
(MII) (11%) and Free Indirect (MIF) (1%), and the third-ranked
Representation of Speech Act (MSS) (9%) are far smaller. The ‘norm’ in
conversational narrative, then, is for discourse presentation to be direct (see
Leech & Short 1981). Further, Figure 2.4 shows that the four most frequent
types (MDD, MII, MSS, and MDF) taken together account for 85% of all
discourse presentation. As will be argued in Section 4.1, these four categories
can be said to support varied degrees of retrievability of the utterance
reported (the fifth category which facilitates retrievability, viz. MIF, is
ranked eighth). This suggests that in conversational narrative, a basic concern
in presenting discourse is in reporting with sufficient immediacy, offering the
2.2 The Narrative Corpus 61
Reporting 1% 0%
N Freq 1% 1%
mode 2% MDD
3%
MII
1 MDD 1,573
MSS
2 MII 292 6%
MDF
3 MSS 240
8% MVV
4 MDF 211
MVT
5 MVV 164
MUU
6 MVT 70 9%
58% MIF
7 MUU 61
MRR
8 MIF 35
11% MRQ
9 MRR 34
10 MRQ 14
11 MXX 32
Total 2,726
recipient (varied degrees of) access to the utterance reported. The notion of
immediacy in reporting is discussed in Section 4.1.
How do these proportions compare to proportions reported in related
research? One point of reference is Mayes (1990), who investigated direct and
indirect speech reports occurring in more than 22 hours of taped and tran-
scribed conversation. If, in the NC, the instances of Direct and Free Direct
and, on the other hand, Indirect and Free Indirect are combined and
compared, the relative proportions thus obtained exactly reflect the propor-
tions obtained in Mayes (1990): 85% for (Free) Direct and 15% for (Free)
Indirect. Looking further beyond the direct and indirect modes, the only
comparable corpus (I am aware of) which is consistently annotated for speech
(and also thought) presentation is the Lancaster Speech, Writing and Thought
Presentation Spoken Corpus (see McIntyre et al. 2004), a small, balanced
corpus of contemporary spoken English drawn from two sources: the con-
versational subcorpus of the BNC and oral history archives from Lancaster
University containing elicited interviews. While, as noted above, discourse
presentation in the NC is annotated for ten categories, McIntyre et al. (2004)
use six categories: Direct, Free Direct, Indirect, Free Indirect,
Representation of Speech Act, and Representation of Voice. The proportions
found in the LSWTPSC and the NC are juxtaposed in Figure 2.5.
As is shown in Figure 2.5, Direct mode (MDD) is the most common
discourse presentation type in both corpora; however, in the NC it accounts
62 Data, methods, and tools
NC vs. LSWTPSC
100
80
MDD
60
Proportions
MII
MSS
MDF
MVV
MIF
40
Other
20
0
NC LSWTPSC
for 58% and in the LSWTPSC it accounts for 38%. The NC also scores
higher on Free Direct mode (MDF): here, the percentage value is 8% in the
NC as opposed to 4% in the LSWTPSC. So, clearly, the (Free) Direct
modes are overrepresented in the NC as compared to the LSWTPSC. On
the other hand, the values obtained for Indirect mode (MII) are almost
identical in the two corpora with 12% in LSWTPSC and 11% in the NC.
Also, the proportions for Free Indirect mode (MIF) are very similar: 2% in
LSWTPSC and 1% in the NC. While (Free) Direct is overrepresented in the
NC and (Free) Indirect is comparably frequent in the corpora, MSS and
MVV are clearly underrepresented in the NC: here they account for 9%
(MSS) and 6% (MVV) respectively, while in the LSWTPSC they obtain
proportions three times as high (27% for MSS and 17% for MVV). The
differences between the two corpora are hence major, affecting mostly the
direct categories, which are more important in the NC, and both MSS and
MVV, which are more frequent in the LSWTPSC. It appears that the
differences can be explained with the distinct generic compositions of the
2.3 Methods and tools 63
two corpora. As noted above, the LSWTPSC contains data from BNC-C, as
does the NC, but the conversational data in the LSWTPSC are by no means
narrative data alone, as in the NC, but present a good cross-section of all
types of conversational subgenres. Second, the LSWTPSC also includes data
from oral history interviews, a genre which is decidedly not conversational
but much more formalized and restricted in terms of (fixed) role assignment
(informant and researcher) and interaction type (interview questions from
researcher to interviewee).
To conclude this section, the NC is a specialized corpus not only
because it targets narratives and their surrounding conversational con-
texts but also because it is richly annotated for various discourse features.
It thus adds to the small set of discourse-annotated corpora (see Garside
et al. 1997).
Statistical inference
Given that this study aims to provide insights on co-construction gained
from quantitative analyses of the narratives in the NC, particular emphasis is
placed on hypothesis and significance testing. The intention is to report
results that are meaningful not only for the NC, which is, after all, just a
minute sample of the population of all conversational narratives in English,
but for that very population as a whole. It is, however, essential to define that
population more exactly. The population of which the NC is a (hopefully
representative)13 sample is not the total of all stories told, being told, and to be
told in English, which is potentially infinite. That population is not known to
anybody; it might even be said not to exist as such (containing as it does
future stories). The population of which the NC is a likeness (in a nutshell) is
much smaller (but still frighteningly large): it is the total of stories told by
people living in a particular place on the globe at a particular point in time –
namely the British Isles in the last decade of the twentieth century (the time
the data for the BNC-C were collected). That is, the tests I am going to
conduct are intended to examine whether the findings made on the NC can
be generalized from the microcosm of the corpus to the macrocosm of telling
stories in English conversation in Britain during the 1990s.14 Almost two
decades have since gone by, the millennium has changed – has storytelling in
Britain changed as well (such that observations that could be generalized in
the 1990s may no longer be generally relevant)? To address this question a
representative corpus of storytelling in the early twenty-first century would
be required. It does not exist, so the question cannot be answered with
certainty. What is certain is that English has changed since the BNC-C was
created. A striking example at the level of the lexicon is the quotative BE like,
which had hardly entered British English in the first half of the 1990s and is
accordingly virtually absent from the BNC but widespread today (see end-
note 11). Grammatical change too has occurred; see, for example, the con-
tinued decline in modal usage (e.g., Leech et al. 2012). Storytelling in
conversation, however, is essentially neither a lexical nor a grammatical
phenomenon but a discourse phenomenon with crucial interactional under-
pinnings (such as recipient design, specific participation patterns, evaluation
in its various forms, etc.). These may have their roots in interaction principles
“that are at source independent of variations in language and culture”
(Levinson 2006: 40; see also Schegloff & Lerner 2009). They are therefore
unlikely to change in a 20-year span. What will be presented here as signifi-
cant findings is hence likely to have remained significant into the early
twenty-first century.
How is the generalization from sample to population achieved? The
answer is, by using inferential statistics. Its aim is “to infer from the proper-
ties of a part the likely properties of the whole” (Woods et al. 1986: 49). The
principle at the heart of inferential statistics is what Stefanowitsch (2005)
2.3 Methods and tools 65
Textual colligation
Finally, in a number of case studies, positional analyses will be carried out.
They build on recent work which has demonstrated the close association of
lexis with positions in texts (e.g., Hoey 2005, Scott & Tribble 2006, Hoey &
O’Donnell 2008, Mahlberg & O’Donnell 2008). In Hoey’s theory of ‘lexical
priming,’ the claim is made that “every word is primed to occur in, or avoid,
certain positions within the discourse” (Hoey 2005:13). This claim is referred
to as ‘textual colligation.’ Also, textual positioning is “likely to be genre and
even subgenre-specific” (Hoey & O’Donnell 2008: 300).
Research into text-lexis associations is an intriguing new area straddling
the fields of lexical studies and discourse analysis. Except for Römer &
O’Donnell (2010), who have examined (student) academic writing, work on
textual positioning has tended to pivot on (sub)genres of news reportage. To
the best of my knowledge, spoken data have not yet been examined in terms
of textual positioning at all.
2.3.2 Tools
In this section, I describe the three tools key in this study: the NC analysis
tools, the programming language R, and XPath and XQuery, two related
66 Data, methods, and tools
R
The tool used to perform significance tests of various sorts is R, a program-
ming language and environment for statistical computing and graphics.
While it has become standard among statisticians, its uptake in corpus
linguistics is still in its early stages. For a not too technical introduction to
R for corpus linguists, see Gries (2009a); for a more comprehensive and more
advanced introduction to R for linguists of all denominations see Gries
(2009b); an even more advanced introduction for linguists is Baayen
(2008). Readable introductions for non-linguists are Dalgaard (2008) and,
at a more advanced level, Crawley (2007). One of the major plusses of R is its
very powerful graphical capabilities, of which only a small proportion can be
exploited in this study (see for example, Mittal 2011).
One particularly useful graphical representation is the boxplot (see
Benjamini 1988, Frigge et al. 1989). It will be used on a number of occasions
in this book and is therefore introduced briefly in the following. The useful-
ness of the boxplot stems from the fact that it summarizes ‘at a glance’ several
pieces of crucial information related to the location and spread of data as well
as its potential skewness (see Crawley 2007: 155). A boxplot consists of five
graphical components: a box, a bold line across the box, notches within the
box, whiskers below and above the box, and, potentially, empty circles below
and above the whiskers. The box represents the interquartile range (IQR),
that is, the range from the first to the third quartile of the data (roughly, the
50% of the data around the median). The bold horizontal line depicts the
median, a robust measure of the central tendency of the data (the middle
value of a sorted distribution; see Woods et al. 1986: 30–32). The notches on
either side of the median line indicate the 95% confidence interval within
which the ‘true’ median is expected to fall (note that notches are an optional
feature; for example, the boxplots in Section 3.3.3 are drawn without
notches). Also, if the notches of two or more boxes do not overlap (that is,
2.3 Methods and tools 67
if they do not cover the same range on the y-axis), they “are likely to prove to
have significantly different medians” (Crawley 2007: 157). The vertical
dashed lines connecting the boxes and the whiskers “extend to maximally
1.5 times the interquartile range” (Baayen 2008: 30). The whiskers can be
understood as ‘fences’ separating “the outermost observations that are not
extreme enough to be flagged as outside by an exploratory rule of thumb”
(Frigge et al. 1989: 51) from outliers defined as “data points with values that
are surprisingly large or small given all data points considered jointly”
(Baayen 2008: 27). Finally, the empty circles represent individual data points
falling outside the fences, thus designating potential outliers. For illustration
consider boxplots showing the above-mentioned lengths of utterances by the
two recipient types, PRR and PRC.
In Figure 2.6, the boxplots are presented in two different panels. While the
left panel shows the distributions of the lengths of utterances by the recipient
types PRR and PRC entirely (with all outliers included), the right panel
presents the boxes in greater detail to permit a clearer view of the boxes and
the notches.
Lengths
70
60
15
50
Number of words
Number of words
40
10
30
20
5
10
0
In the left panel, it can be seen that the dispersion of the lengths of PRC
responses is much greater than the dispersion of the lengths of PRR
responses. This greater dispersion, and hence greater heterogeneity, is sug-
gested, not only by the larger range of outliers (indicated by the empty
circles), but also by the greater distance between the whiskers (the dashed
vertical lines) and the larger size of the box. This skewness shows up even
more clearly in the right panel. While, for PRC responses, the upper part of
the box (above the median line) and the lower part are reasonably sym-
metrical, there exists no lower part (below the median line) for PRR – the box
is decidedly asymmetrical. Also, the median (indicated in the bold horizontal
line in the box) for PRC is located much higher on the y-axis than for PRR,
suggesting that, on average, PRC utterances tend to be much longer than
PRR utterances. Further, the notches (indicating the 95% confidence inter-
vals) for PRR and PRC are far from overlapping. This is prima facie evidence
suggesting that the difference in median length is significant (see Gries
2009a: 205). (Indeed, according to a Wilcoxon rank sum test, the difference
is very highly significant: p-value < 2.2e-16.)
In the following, I introduce XPath and XQuery, two immensely powerful
tools for the analysis of XML texts.
Xpath and XQuery
The XML Path Language, XPath, is a metalanguage which allows the
navigation of XML documents (see Clark & DeRose 1999), such as the
texts assembled in the BNC and the NC. To understand how XPath works
an understanding of how XML texts are structured is helpful.
XML documents are characterized by the fact that any element in them is
part of a hierarchical tree structure. XML elements can be empty but, more
often, they have content, for example, a particular word form. For instance,
the narrative component (CNN) of text KP1-N2 starts with the following
words: That was the week when Peter came. The individual words are all
defined as <w>-elements, receiving both a start tag (<w>) and an end tag
(</w>):
(2.16) <w>That </w>
<w>was </w>
<w>the </w>
<w>week </w>
<w>when </w>
<w>Peter </w>
<w>came </w>
However, XML elements may be further defined by attributes, which, in
turn, have values. The same passage from KP1-N2 with all attributes and
values given in the NC reads as follows:
2.3 Methods and tools 69
It has three attributes and accordingly, three values: the title attribute, whose
value is Mining accident, and the attribute embedLevel, which states whether
the narrative contained in the text is part of a narrative chain or a stand-alone
story; here the value is ES, denoting that the text contains a single story. The
third attribute is narrativeType; its value is T10, the label used for first-person
narratives (see Section 2.2.2). At the next lower level (level 2) we find a <seg>
(from ‘segment’) element whose attribute Components has the value CNN,
that is, this <seg> identifies the beginning of a narrative component. One
level further down (level 3), there is a second <seg> element; it has the same
attribute Components but the value is CNI (for narrative-initial utterance).
Yet another dependency level further down (level 4) we find a <u> (utter-
ance) element. It has two attributes: who and Participation_roles. The who
attribute acts like a passport defining the speaker in terms of sex, age, and a
number of other social categories; the value is PS50U, which the text header
identifies as a 70-year-old housewife speaking the dialect of Northwest
Midlands. The Participation_roles attribute registers information about the
narrative role the speaker of the utterance is taking on in the course of the
incipient narration; the value here is PNP (Primary Narrator). Finally, at
the lowest dependency level (level 5), there are all the <w> elements. An
important term that comes into play at this stage is ‘node’: elements and
attributes (including their values) are referred to as nodes.15
XPath operates on that node tree in such a way that any given (series of)
nodes can be defined via what are called ‘axes.’ So, what we have so far referred
to as dependency levels 1-4 would be referred to in XPath parlance as the axes
1-4. XPath axes “tell the XPath processor which ‘direction’ to head in as it
navigates around the hierarchical tree of nodes” (Watt 2002: 82). To express
the various levels of superordination and subordination between the axes, the
XPath syntax uses a terminology which is borrowed from human family
relations. Thus, in (2.18) above, the first two axes stand in a parent-child
relationship: axis 1 is the parent of axis 2, or the reverse, axis 2 is the child of
axis 1. The first and all subsequent axes stand in an ancestor-descendant
relationship: axis 1 is the ancestor of axes 2-4, while axes 2-4 are descendants
of axis 1. Looking at axis 4 only, the Xpath syntax defines all seven <w>
elements, which are on the same ‘footing’ with one another, as siblings.
If we assume, for illustrative purposes, that (2.18) were a complete XML
document16 we could make use of the knowledge of the structure of XML
documents and of the XPath syntax to formulate ‘paths’ taking us to any
node in it.
For example, the simple XPath //div locates the text printed in (2.18).
This is, of course, not surprising in that the document has only one <div>
element. If the document contained more narratives it would contain more
<div> elements and, accordingly, the XPath //div would find all these
elements. If, in that longer document, we were specifically interested in
just the text called Mining accident, we could exploit the text’s title attribute
2.3 Methods and tools 71
bncDoc “http://www.w3.org/2001/XMLSchema-instance”
teiHeader
stext “CONVRSN”
div “Cars”
div “Neighbourly advice”
seg “CPR”
seg “CNN”
seg “CPO”
div “I always lock the car”
seg “CNN”
seg “CNI”
u “PS01A”
div “Locked in the car”
seg “CNN”
seg “CNI-CNF”
u “PS01C”
s “153”
align “KB1LC019”
w “PNP” He
w “VHZ” ’s
w “VBN” been
w “AVP-PRP” out
w “AT0” the
w “NN1” car
w “CJC” and
w “VVD” locked
w “PNP” me
w “PRP” in
w “PNP” it
w “PUN” .
s “153”
u “PS01A”
seg “CNF”
seg “CPO”
div “Sure start”
To illustrate the intricate embedding, Figure 2.7 shows the tree structure of
text KB1-N1 entitled Cars. The text is enclosed in a <div> element, the
highest-order element embracing all other nodes in the document. This first
<div> element contains three narratives, each of which is assigned a <div>
2.3 Methods and tools 73
element of its own. The first such subordinate <div> element has the title
value Neighbourly advice. As most <div> elements, this first subordinate
<div> is superordinate to three major text components, CPR, CNN, and
CPO, each enclosed in a <seg> element. The second subordinate <div>,
whose title value is I always lock the car, does not contain a pre-narrative
component (CPR). It starts with the main narrative component CNN, the
first element of which denotes the CNI (narrative-initial utterance) com-
ponent. Immediately following the CNI <seg> and the <u> element it
encloses is yet another <div> element entitled Locked in the car. This
<div> interrupts, as it were, the chain of utterances making up the main
narrative in the superordinate <div> entitled I always lock the car. It is an
example of an embedded narrative. Like many such embedded narratives, it
consists of only one utterance, that is, the narrative-initial and the narrative-
final utterances are collapsed into a single utterance. This single utterance
is enclosed in a <seg> element whose Components value is CNI-CNF.
After that ‘interlude,’ the main narrative is resumed and brought to a
close in the narrative-final utterance in the CNF <seg>, which is subordi-
nate to the CNN <seg>. The CNF <seg> is followed by CNN’s sibling
<seg> CPO, the post-narrative component. As soon as the CPO <seg> is
completed, a new <div> element occurs containing the third story in the
story chain.
The levels of embedding of annotations in the NC, which are only
sketched out crudely (for more detail on this, see Rühlemann & O’Donnell
2012), are numerous and complex. Finding ‘paths’ that, where necessary,
circumvent or, where desirable, operate on these embeddings is not easy. It is
nonetheless feasible and, what is more, revealing. Because of the density of
annotation provided in the NC, XPath can retrieve almost any set of data
researchers might be interested in.
An extension of XPath is XQuery, which supports the same functions and
operators as XPath (for an introduction, see Walmsley 2007).17 The differ-
ence lies in the increased flexibility of XQuery: while Xpath is restricted in
terms of the output it can return (namely either frequencies or text but not
both at the same time), users of XQuery can program their own scripts to get
much more diverse and complex output. This is because XQuery supports
the use of what are called for loops. The advantage of using loops is that not
only can a specific type of node set be specified, which is also possible using
XPath, but diverse data can be retrieved for each individual instantiation of
that type. Take for example the node set Responsive Recipient (PRR) utter-
ances. Suppose we are interested in the length of PRR utterances in narra-
tives that contain exactly one such PRR utterance. If we use XPath only, we
can only determine the total number of words and the total number of such
PRR utterances. Dividing the former by the latter the average number of
words in the PRR units can be determined. As is well known, though, a
danger of averages is that they may harbor important variation. Grasping that
74 Data, methods, and tools
The for loop in line 1 instructs the processor to loop over those CNN
components in which the number of PRR units (which are ‘descendants’ of
CNN components) is 1. The let commands in lines 2, 3, and 4 define
parameters for which data are to be pulled: let $n_words_CNN counts the
number of words in each CNN component in the node set; let
$n_words_PRR counts the number of words in each PRR unit, and let
$text_PRR instructs the software to retrieve the words contained in the
PRR units. The return command in line 5, finally, defines the output to be
obtained: in this case, the three parameters targeted in the let clauses. The
first ten lines of the output look like (2.19) (where no text is printed, the PRR
response consists of non-verbal behavior, such as laughter):
(2.19) $n_words_CNN $n_words_PRR $text_PRR
0 1 yes
12 1 yes
191 1 yeah
82 1 blimey
78 0
45 1 mm
145 1 yeah
24 0
25 1 yeah
35 2 oh no
The detailed information generated by the XQuery script enables a wide
range of quantitative analyses and graphical representations: for example,
dispersion measures can be calculated (standard deviation, interquartile
range, etc.), the distribution of the data can be compared to similar
2.3 Methods and tools 75
3.1 Introduction
A number of scholars have portrayed spoken narrative as a type of ‘monolog.’
Swales, for example, sees narrative as a “monologic activity” (1990: 61).
Labov (1972: 366) and Chafe (1992: 43) argued that storytelling constitutes
“a single ‘turn’.” In a similar vein, in research in a conversation-analytical
(CA) tradition, storytelling is often conceptualized as a ‘multi-unit turn’:
the individual utterances by the teller are taken, not as individual turns, but
as turn constructional units (TCUs), together forming an extended turn. In
Goodwin & Heritage (1990: 299), for example, stories are “thought of as
built from many turn-constructional units” (see also Goodwin 1986a, Norrick
2012). While it is recognized that “[s]tories take more than an utterance to
produce” (Sacks 1992: 222) because recipients “are specifically invited”
(Sacks 1992: 227) to produce talk in response to the storytelling, their talk is
often not accorded the status of proper turns. For example, it is common
practice to view talk in the backchannel, such as mm and most uses of yeah,
whose main purpose is the display of understanding (see Gardner 1998),
as talk situated “between listening and speaking” (Gardner 1998: 204), and
therefore to deny it the status of a turn in its own right. Rather than as turns,
recipient utterances during storytelling are seen as ‘embedded sequences’
(Norrick 2012) which “do not disrupt the overall thrust of the multi-turn
unit” (Norrick 2012: 568). Their primary function is as “talk that deals with
the recognition that a story is being told” (Sacks 1992: 227). On this view,
recipient responses to the main speaker’s extended turn (the storytelling) are
treated as somewhat ‘ancillary’ (Edelsky 1993), that is, as subservient to the
‘global speech act’ (van Dijk & Kintsch 1983: 265) performed by the teller.
Conceptually, the view of storytelling as a multi-unit turn, and the denial of
turn status to recipient responses it entails, is not without problems. First,
labeling responses as ‘embedded sequences’ does not answer the question
exactly what sort of unit responses are and how they relate to stories as
multi-TCU turns. If indeed “the basic unit of talk is the TCU” (Selting
2000: 477, added emphasis), then responses should count as TCUs too, that
is, as units that can potentially construct turns. Second, TCUs, by definition,
76
3.1 Introduction 77
et al. 1974: 701). Turn order is not fixed because it is “locally controlled (i.e.
turn by turn)” (Sacks et al. 1974: 708) rather than pre-determined by some
higher-order control mechanism overriding local control. Since control is
exercised locally, that is, at every possible transition-relevance place, the rule
effectively ensures that “at each such place any party to the conversation can
speak next” (Sacks et al. 1974: 711). It should be obvious that the ‘every third
slot, from a first’ pattern is in blatant violation of the ‘no fixed turn order’
rule: it represents precisely such a fixed order.
In this section I aim to test the empirical validity of Sacks’s observation.
That is, the research question I am going to examine is: Do narrators in
conversational narrative control ‘every third slot, from a first’? If the obser-
vation could be validated statistically, this would be evidence, not only that
the ‘no fixed turn order’ rule postulated by Sacks and colleagues does
not hold for conversational narrative, but also that turntaking in narrative is
co-constructed in the sense of an agreement between participants to suspend
ordinary turntaking for the duration of the story and concede the narrator
privileged turntaking rights.
Before we approach the research question directly, some preparatory work
is required. To begin with, more convenient terminology will be helpful. To
get rid of the somewhat clumsy labeling of the postulated turn order pattern
as the ‘third slot, from a first’ pattern, I will henceforth refer to it as the
‘N-notN-N’ pattern, with upper case ‘N’ standing for Narrator and ‘notN’
for any participant that is not the narrator, that is, any recipient type,
Responsive Recipient (PRR) or Co-constructive Recipient (PRC), or a
potential Co-Narrator (PNC). The N-notN-N pattern will have to be kept
apart from what can be referred to as the ‘n-notn-n’ pattern, with lower case
‘n,’ which describes the control of the first and third turn slot, not by the
narrator, but by any same speaker irrespective of participant role. The
N-notN-N pattern thus represents a subtype of the n-notn-n pattern. Also,
since the pattern (in either form) concerns the succession of three turns each,
I propose to refer to the resulting turn sequences as ‘turn trigrams.’ In
establishing the possible trigrams available, an important constraint is related
to the fundamental rule that ‘speaker change occurs’ (see Sacks et al. 1974):
the same speaker cannot, by definition, occupy two successive turns. That is,
labeling speakers with letters, a trigram with the structure AAB is not
conceivable. If A self-selects, for example, after a pause because no other
participant has self-selected, the thus extended turn counts as one turn.
However, the same speaker can re-occur in the same trigram if they occur
in the trigram’s boundary slots, that is, the first and last trigram slot (ABA or
BAB), thus instantiating the n-notn-n pattern or, if that ‘same speaker’ is the
narrator, the N-notN-N variety.
Further, a note on number of parties is required. As Sacks et al. (1974: 712)
put it, “[n]umber of parties can vary” (1974: 701). That is, notwithstanding
the fact that “the system favours . . . smaller numbers of participants” (1974:
80 How do narrators and recipients co-construct turntaking?
Pattern Trigrams starting with N Trigrams starting with R1 Trigrams starting with R2
n=3 N-R1-R2 R1-R2-N R2-N-R1
N-R2-R1 R1-N-R2 R2-R1-N
n-notn-n/ N-R1-N R1-N-R1 R2-R1-R2
N-notN-N N-R2-N R1-R2-R1 R2-N-R2
conversation, can result. However, given the N-notN-N pattern (i.e., assum-
ing that the narrator controls ‘every third turn, from a first’), turn distribu-
tion will inevitably be biased to the narrator. This is because recipients are
allocated only ‘second’ turns. The restriction to second turns is unproble-
matic in two-party narrative where, given speaker change, turn distribution is
equally divided between narrator and listener (whereas turn size may differ
radically; see Section 3.3). In three-party talk, by contrast, with two recipi-
ents, second turns must be divided between the number of recipients: the
likelihood for each of the two recipients to get the second turn is effectively
halved. In four-party talk, with three recipients, the likelihood for each
recipient decreases to a third, and so on. The N-notN-N pattern therefore
most likely grants the narrator a greater turn share than their interlocutors.
That is, if the N-notN-N pattern is distinctive of turn order in conversational
narrative, this diversion from ordinary turn order in conversation will by
necessity trigger a second diversion – one from ordinary turn distribution.
Moreover, a closer look is warranted at what turn order patterns are
available beside the N-notN-N pattern. First, in a conversation with three
participants – the main narrator (N), recipient 1 (R1), and recipient 2 (R2) –
how many distinct trigrams can there be?
As shown in Table 3.1, in three-party talk, as many as 12 different trigrams
are possible. Six of them (or 50%) match the n-notn-n pattern (see the lower
two rows), where any one speaker regardless of role takes both the first and
the third turn. Two of these n-notn-n trigrams represent a special config-
uration, viz. the N-notN-N pattern (shaded grey in the table), where the
narrator occupies the first and third slot and which is the focus of this section.
The six trigrams which account for the remaining 50% form a pattern too.
They share the property that, unlike the n-notn-n/N-notN-N pattern
where only two participants are involved, all three participants take turns.
Technically, the pattern can be represented as n=3. Specifically, it can be
seen that the above-mentioned case in which two successive recipient
responses follow the narrator’s turn can be instantiated in the form of
two n=3 trigrams, viz. N-R1-R2 and N-R2-R1. A distinction might be made
between recipient responses which are temporally consecutive (that is, which
occur one after the other) and those occurring at the same time (that is, in
82 How do narrators and recipients co-construct turntaking?
overlap). For practical reasons, this distinction is not implemented here; all
double-responses are treated as consecutive turns. In sum, the 12 different
possible turn-trigrams fall into two distinct turn order patterns, n-notn-n
or n=3; two n=3 trigrams represent double-response trigrams while two
n-notn-n trigrams match the N-notN-N pattern.
Finally, a crucial question concerns the expected proportion of N-notN-N
trigrams out of all trigrams in three-party stories. This proportion depends
on the probability of occurrence of turns taken by the three participants. If we
follow Sacks et al.’s (1974) above-mentioned postulates for turn order and
turn distribution (that neither turn order nor turn distribution are predeter-
mined in advance), the probabilities are, in principle, equal for all three
speakers. That is, the chances for each of the three participants N, R1, and
R2 to occupy the first slot of the first turn trigram are 1/3 = 0.33. Once that
first slot is taken, given that the same speaker cannot occupy two successive
slots, the probability of occurrence increases to 1/2 = 0.5. To illustrate, since
recipients too can take the first turn in a story by initiating or eliciting its
telling, the chances that N gets the first turn are 0.33.2 In that case, the
chances that R1 gets the second turn are 0.5 given that only one competitor is
left, viz. R2. The same logic, and probability, applies to any further turn.
Based on these assumptions, a random simulation was performed in R for
three distinct speakers taking a total of 10,000 turns. The simulation cuts a
distinct ‘path’ into the web of all possible turn combinations, records the
resulting 9,998 trigrams and computes the proportion of N-notN-N trigrams
out of all trigrams in that simulation. The process was repeated 100 times and
the mean proportion of N-notN-N trigrams for all 100 simulations was
computed. It turned out that this mean proportion was 16.67694%. I will
round this figure to 17%.3
This proportion is decisive in that it constitutes a benchmark value against
which the proportions obtained in three-party narrative will be assessed in
the analyses below. Note that the same proportion, viz. 17%, is expected for
the two double-response trigrams taken together (N-R1-R2 and N-R2-R1).
Given these orientations, the hypotheses to be tested can be formulated
thus:
H0: In three-party narratives, the true proportion of N-notN-N trigrams is
smaller than, or equal to, 17%.
H1: In three-party narratives, the true proportion of N-notN-N trigrams is
greater than 17%.
3.2.3 Results
As noted above, turn order is intimately linked to turn distribution. If turn
order is significantly skewed towards the N-notN-N pattern this will inevitably
go hand in hand with a skew towards increased turn shares for the narrator and
decreased turn shares for the recipients.
The participants in the 19-story sample take together 232 turns. Their
distribution is skewed as expected: the narrators take together 107 turns
(46%), while those recipients who respond first in the narratives (R1) have
together 68 turns (29%) and those responding in second place (R2) have
57 turns (25%). Given that, based on Sacks et al.’s ‘no fixed turn distribution’
premise, turns should be distributed equally between the three participants,
according to a chi-squared test for given probabilities, these differences
are highly significant (X-squared = 17.85, df = 2, p-value = 0.00013).4
Thus, contrary to Sacks et al.’s (1974) postulate, in three-party narrative,
turn distribution is skewed towards the narrator. Due to the close linkage
between turn distribution and turn order, this finding is a strong indication
that turn order will be biased towards the N-notN-N pattern.
86 How do narrators and recipients co-construct turntaking?
Number of Proportion
Story title Total number N-notN-N N-notN-N
(short) of trigrams trigrams in %
Embarrassing 4 2 50
New bike 4 1 25
Old cricket 4 2 50
Old friend 4 2 50
Teenage 4 2 50
Emma 5 3 60
Tarmacked 5 3 60
Elly 6 2 33
Pushed 6 3 50
One hour 7 1 14
Roaring 9 5 56
Spilled 10 5 50
Stew 12 6 50
BNC 14 2 14
Sleep 14 6 43
Recording 15 5 33
Voucher 15 5 33
Bingo 21 3 14
Malham 35 8 23
Total 194 66 34
3.2 Co-construction of turn order 87
The data in Table 3.3 are presented in ascending order of the total number
of trigrams per story (see column 2). The number of trigrams that conform
to the N-notN-N pattern is given in column 3, column 4 gives the corre-
sponding percentage value. Note that the maximum percentage value for the
N-notn-N pattern is 60%. As can be seen, a good number of the percentage
values in Table 3.3 are within proximity of that maximal range. The total of
66 N-notN-N trigrams accounts for 34% of all trigrams in the sample – twice
as much as assumed under the H0, where the expected proportion was 17%.
Also, the data in the table reveal a curious tendency: as the number of
trigrams per story increases (see column 2), the share the N-notN-N trigrams
have in them (expressed as a percentage) seems to decrease. This suggests
that the two variables may be negatively correlated.
To test for this possibility, Kendall’s correlation test was chosen (see Gries
2009b: 144) to account for the non-normal distribution of the two variables.6
Generally, correlation tests calculate a correlation coefficient ranging from -1 to
+1, “where the extremes indicate perfect correlation and 0 means no correla-
tion” (Dalgaard 2008: 120). According to the test, Kendall’s correlation coef-
ficient τ is -0.41. That is, it is smaller than 0, suggesting that the two variables
are correlated; the strength of the correlation though is modest in that the
coefficient is slightly closer to 0, the value indicating that there is no correlation
at all, than 1, the value indicating perfect correlation. The p-value = 0.026
indicates that the correlation is significant. Moreover, the negative coefficient
indicates that the variables are negatively correlated: the number of N-notN-N
trigrams decreases as the total number of trigrams in a story increases. The
negativity of the correlation is brought out by a linear regression, plotted in
Figure 3.1, where the regression line exhibits a very clear downward slope.
The correlation is undoubtedly an interesting discovery, which I will dis-
cuss below, but it does not help address the central question of whether the
N-notN-N pattern is statistically significant in three-party narrative. What we
have so far are two critical figures. First, under the H0, N-notN-N trigrams are
expected to account for 17% of all trigrams. Second, in the sample, N-notN-N
trigrams account for 34% of all trigrams. The question then is whether this
difference in proportion is large enough to justify rejecting the H0.
One obvious method to answer this question is the 1-sample proportions
test with continuity correction (e.g., Dalgaard 2008: 146, Gries 2009b: 126). It
essentially computes a 95% confidence interval (CI) for the ‘true’ proportion.
CIs take into account sampling variability (see Woods et al. 1986: 96) (that is,
the variability due to the fact that, in the present case, a new sample of three-
party stories will produce a different proportion of N-notN-N trigrams than
exactly 34%) and estimate “a likely range within which the population value
may lie” (Woods et al. 1986: 96). Further, given that we have a prior value that
we want to test for, viz. 17%, the null probability was set to 0.17.7
The test’s results are straightforward: the difference in proportion is very
highly significant (p-value = 5.111e-10) and the CI spans between 0.28 and 0.41.
88 How do narrators and recipients co-construct turntaking?
60
50
proportion N-notN-N
30 2040
5 10 15 20 25 30 35
number_all_trigrams
That is, the proportion of N-notN-N trigrams which can, with 95% con-
fidence, be considered true in the population of three-party narratives lies
somewhere between 28% and 41%. The proportion suggested by the H0 was
17% – a value not included in the CI. To double-check, a bootstrap was
performed (for more details on this method, see Section 4.3).8 The bootstrap’s
result fully confirmed the result of the proportions test.9 Hence, N-notN-N
trigrams are significantly more frequent than assumed under the H0. We can
justifiably reject the H0 and accept the H1: the proportion of N-notN-N
trigrams is greater than 17%. That is, Sacks’s (1992) interpretation of narrative
as an attempt by the narrator to ‘control every third slot, from a first’ is valid.
3.2.4 Discussion
The previous section reported four findings, one related to turn distribution
and three to turn order in three-party narrative. They are discussed in the
following.
3.2 Co-construction of turn order 89
3.2.5 Summary
In this case study I have been concerned with patterns of turn order in
conversational narrative. The aim has been to examine the empirical and
statistical validity of Sacks’s (1992) observation that narrators attempt to control
‘every third slot, from a first,’ an observation which concerns multi-party
narrative but not two-party narrative (where this turn pattern is inevitable).
3.2 Co-construction of turn order 91
For practical reasons I limited the analysis to three-party narratives. Since the
turn pattern in question involves the succession of three turns each, the
analysis was geared towards what I referred to as turn trigrams. The turn
order pattern corresponding to Sacks’s observation was labeled the N-notN-N
pattern (with N standing for narrator), which can be instantiated, in three-
party narrative, in the form of two trigrams, N-R1-N and N-R2-N (with R1 and
R2 standing for the two recipients). Given that the N-notN-N pattern impli-
cates that turns are preferentially distributed to the narrator and that recipients
avoid double-responses, attention was also paid to turn distribution and to how
recipients cooperate in taking response turns.
Starting from a close analysis of the turn trigrams found in a single
narrative, I widened the database by extracting from the NC an exhaustive
set of personal experience three-party narratives, fulfilling a number of strict
conditions. A sample of 19 narratives was thus extracted and analyzed for turn
distribution and turn trigrams.
A number of closely intertwined findings were made. First, as expected,
turns were found to be distributed with a significant skew towards the
narrator, who gets the lion’s share in turns. Second, the N-notN-N trigrams
accounted for a much larger proportion, viz. 34%, than expected under the
H0, where the expected proportion was 17%. The significance of this differ-
ence in proportion was confirmed by two tests, a proportions test and a
bootstrap: in either case, the confidence interval excluded the stipulated 17%
but included the 34% found in the sample. The results thus fully confirm
Sacks’s observation: the N-notN-N pattern occurs with significantly greater
than random frequency; that is, narrators indeed succeed in gaining control
over ‘every third turn, from a first.’ The decay of the N-notN-N pattern in
stories with more extensive turntaking was seen in relation to the divergent
nature of the pattern, which can only supplant the default pattern tempora-
rily. Further, it was discovered that the N-notN-N pattern is negatively
correlated with the total number of trigrams per story: the proportions of
N-notN-N trigrams in the stories decrease as the numbers of trigrams in the
stories increase. Finally, it was found that, to ensure that the narrator gets
every third turn, from a first, the two recipients coordinate their turntaking
economically: instead of both recipients responding to any one narrator turn
either in overlap to one another or in two adjacent turns, recipients ‘take
turns’ at producing responses: double-responses are avoided, preference is
given to single-responses. It was hypothesized that, to achieve this ‘economy
of listening’, recipients must cooperate, closely monitoring each other’s
behavior. Precisely how the economy of listening is achieved and whether
it is also achieved in narrative with more than three participants, is a
rewarding objective for future research.
The significance of the N-notN-N pattern suggests an agreement between
all three participants to suspend ordinary turn order patterning and grant the
narrator privileged turn order rights. The local (turn-by-turn) control of
92 How do narrators and recipients co-construct turntaking?
may be expected to translate into greater turn size on the part of the narrator
and lesser turn size on the part of the recipient(s), at least in cases where their
responses are restricted to, as Sacks remarks, “ ‘Mm hm’s, ‘Uh huh’s, what-
ever else they put in” (Sacks 1992: 18).
I therefore assume that extended turn size by principal teller, coupled to
minimal turn size by recipient as well as constancy of fluctuation between them
is especially characteristic of personal experience narratives. I also assume that,
unlike the N-notN-N pattern, which was observed to be indicative only of
multi-party narrative, turn size fluctuation should be indicative of narrative
regardless of number of participants. However, for space considerations, the
analysis presented in the following will be limited to two-party narratives.
Table 3.4 Number of words per turn and TSF coefficients in the pre-narrative (CPR),
narrative (CNN), and post-narrative (CPO) components of text KB9-N2
Turn # CPR TSF coeff. CNN TSF coeff. CPO TSF coeff.
#1 4 33 11
#2 1 0.60 1 0.94 1 0.83
#3 14 0.87 21 0.91 14 0.87
#4 29 0.35 1 0.91 1 0.87
#5 1 0.93 44 0.96 10 0.82
#6 11 0.83 1 0.96 1 0.82
#7 1 0.83 47 0.96 14 0.87
#8 5 0.67 1 0.96 2 0.75
#9 2 0.43 27 0.93 10 0.67
#10 13 0.73 11 0.42 22 0.38
#11 14 0.04 2 0.69 1 0.91
#12 5 0.47 1 0.33 8 0.78
#13 7 0.17 10 0.11
#14 1 0.75 21 0.35
#15 4 0.60 1 0.91
3.3 Co-construction of turn size 95
(Ni – N[i+1])
(Ni + N[i+1])
Figure 3.2 Turn size fluctuation formula adapted from Leech and
Fallon (1992) (N: number of words; i: turn number)
To bring out the fluctuation pattern underlying the text components more
clearly, I adapted Leech and Fallon’s (1992) difference coefficient formula. It
compresses large differences into values ranging between 0 and 1, thus
making them more easily comparable.
What we obtain from applying the formula in Figure 3.2 are coefficients of
turn size fluctuation, henceforth TSF. As noted, the TSF coefficients for text
KB9-N2 are included in Table 3.4 above. Note that, for the present purpose
of measuring TSF, only the size of the coefficient is relevant, whereas the sign
is irrelevant. Negative coefficients were therefore converted into positive
ones. To illustrate the procedure, the TSF coefficient for turns #1 and #2
in CNN is calculated thus: (number of words in turn #1 minus number of
words in turn #2) divided by (number of words in turn #1 plus number of
words in turn #2):
(33 – 1) / (33 + 1) = 32 / 34 = 0.94
Importantly, TSF coefficients do not measure the length of individual turns.
Rather, each individual TSF coefficient reflects the length of one turn in
relation to the length of the immediately preceding turn. Coefficients are thus
a measure of turn bigrams. Consider this excerpt from the above used text
KB9-N2, where the lengths and coefficients but not the words are given:
(3.3) “Tax” (Type: T10 / Embed Level: ES)
CNN
CNI N Coeff.
#1 S1 PNP 33
→ #2 S2 PRR 1 0.94
#3 S1 PNP 21 0.91
→ #4 S2 PRR 1 0.91
Compare, for example, the coefficients for the two arrowed PRR responses
(#2 and #4) in (3.3): although the two responses have the same length
(1 word), the coefficient for the first response (0.94) is greater than the
coefficient for the second (0.91). This is because the narrator’s turn preceding
the first response is longer (33 words) than the narrator’s turn (21 words)
preceding the second response. As noted earlier (see Section 1.3) in the NC
annotation scheme, the two recipient roles and the role of Co-narrator were
assigned on a turn-by-turn basis; that is, the assignment of the role depended
on what the utterance ‘does’ in relation to the preceding utterance. This
96 How do narrators and recipients co-construct turntaking?
TSF in KB9-N2
CPR
CNN
1.0
CPO
0.8
TSF coefficients
0.4 0.6
0.2
0.0
2 4 6 8 10 12 14
Turn bigrams
preceding role is, by definition (since we are dealing with two-party narrative
and since the unclear and non-narrative roles were discarded), the role of
narrator, either in the incarnation of Primary Narrator (PNP) or Supported
Narrator (PNS). In other words, all TSF coefficients obtained for the roles of
(i) PRR, (ii) PRC, and (iii) PNC reflect the relation between the length of
turns by these roles and the length of the preceding (Primary or Supported)
narrator’s turn. This association of TSF with participant role will be key in
the analysis below.
Figure 3.3 depicts the TSF coefficients for the three components of text
KB9-N2 as line plots. It can be seen that the coefficients for the narrative
(CNN) component diverge from the coefficients for the pre-narrative (CPR)
and post-narrative (CPO) components on two counts: (i) the coefficient
values for CNN are greater (only bigrams #9-#10, #10-#11 and #11-#12
have small coefficients), and (ii) they exhibit less dispersion (or, positively,
greater homogeneity) among each other: the values remain almost level
throughout the first 9 turn pairs; again, only the last three bigrams see a
noticeable drop. Observation (i), that the coefficient values for CNN are, on
3.3 Co-construction of turn size 97
Obviously, ‘sizable’ and ‘homogeneous’ are relative terms that can take on a
wide range of values. I will operationalize TSF size in terms of the central
tendency of the data expressed in the median value and TSF homogeneity in
terms of the length of the Q1–Q3 interval around the median expressed as the
IQR. Further, I set the reference values, for median, at μ0 = 0.75 and, for
IQR, at 0.25.12
XQueries were instructed to retrieve turn lengths from texts that satisfied
four criteria:13 find among all narrative components (CNN) those (i) which
are tagged as personal experience stories (T10 or T30), (ii) which are more
than 7 turns long, (iii) which do not contain an embedded narrative (EN), and
(iv) which do not enclose any non-narrative or unclear participant role (PX).
Condition (i) was set to avoid the influence of narrative subgenres such as, for
example, fantasies, which should be considered atypical in that fantasies
“construct a fictional world” (Norrick 2000: 161) and where the construction
is in principle shared equally between participants, most typically between
a Primary Narrator (PNP) and a Co-narrator (PNC), and where hence
no regular and sizable fluctuation will be expected to occur (see below).
Condition (ii) (minimal length is 8 turns) was imposed because of the
difficulty to detect regular patterning of TSF in narratives consisting of
insufficient numbers of turns; given the minimum of 8 turns in the story,
the number of TSF coefficients for each narrative was at least 7 – a number
from which patterning (or lack thereof) of TSF can reasonably well be
observed. Conditions (iii) and (iv) served to exclude those types of interven-
ing turns that are (potentially) unrelated to the story and whose inclusion
might hide the otherwise regular TSF in the transcript of which they are a
part. To ensure that genuine two-party conversational narratives were tar-
geted one further criterion was that the number of distinctly identifiable
speakers had to be exactly two.
3.3.3 Results
The XQueries retrieved 43 first-person experience (T10) stories and 10
third-person experience (T30) stories that satisfied these criteria. Table 3.6
gives an overview of the data:
Table 3.6 Data analyzed for turn size fluctuation (TSF); CPR:
pre-narrative, CNN: narrative, CPO: post-narrative components
0.75
Coefficients
m0
CNN
0.0 0.2 0.4 0.6 0.8 1.0
m0
Coefficients
0.75
CPO
0.0 0.2 0.4 0.6 0.8 1.0
0.75
Coefficients
m0
1.0
1.0
0.8
0.8
0.75
0.6
0.6
TSF coefficients
IQRs
0.4
0.4
0.25
0.2
0.46
0.2
0.42
0.18
0.0
0.0
3.3.4 Discussion
The analyses have yielded two main findings: (i) TSF is significantly
greater (‘more sizable’) in conversational narrative than in non-narrative
conversation; (ii) TSF is significantly greater and more homogeneous
between narrators and Responsive Recipients (PRR) than between narra-
tors and Co-constructive Recipients (PRC) and, respectively, narrators and
Co-Narrators (PNC).
Finding (i), that TSF is greater in narrative than in non-narrative conver-
sation, suggests that turn size is subgenre-dependent. In non-narrative con-
versation, the average TSF was, for pre-narrative, 0.60 and, for post-narrative,
0.64, that is, close to 0.3, the ‘ideal’ value under the assumption that Sacks
et al.’s (1974) rule applies, which indicates that turn size is not predetermined
between the participants such that any participant can, at any point in the
conversation, take long or short turns.16 In conversational narrative, by con-
trast, the rule does not apply. Here, the fluctuation that holds between any two
adjacent turns, expressed as a TSF coefficient, is significantly greater, namely
0.8. Since in narrative with two parties, turn bigrams inevitably include a
narrator and a recipient/Co-narrator, the alternative ‘rule’ that could be
gleaned is that, in conversational narrative, turn size is predetermined: on
average, narrators take long turns while recipients/Co-narrators take short
turns. This finding not only ties in well with above cited observations of
extended turn length of main tellers but also, more importantly, provides
the hitherto missing empirical and statistical validation of these qualitative
observations. As regards the relationship between narrators and recipients,
who have no prior knowledge of the story events, the significant imbalance in
turn size lends itself to an interpretation in the light of the epistemic imbalance
that holds between them: while narrators in producing the story can work from
the cognitive representation of the events they have modeled via experience
(in the case of first-person experience stories) or hearsay (in the case of
third-person experience stories), recipients know little or nothing of the
events; they need to construct a mental representation of the events via the
story (see van Dijk & Kintsch 1983). That is, recipients lag behind the narrator
epistemically: while the narrator has knowledge of the events from the begin-
ning, the recipients’ knowledge only grows in line with the progression of
the storytelling production. The finding that TSF is significantly more sizable
in conversational narrative than in general conversation is the direct correlate
of this epistemic imbalance.
3.3 Co-construction of turn size 105
Finding (ii), that TSF is both greater and more homogeneous between
narrators and Responsive Recipients (PRR) than between narrators and
Co-constructive Recipients and, respectively, Co-narrators, invites a two-
fold reading. First, the finding suggests that size and homogeneity of
TSF depend on narrative roles and the interaction between the roles.
The fact that TSF between narrators and Co-narrators is less sizable and
less homogeneous is due to the specific epistemic status of Co-narrators,
which is less disadvantaged than that of recipients: Co-narrators share with
the Primary Narrator key knowledge of the story events. In the case of
generalized experience stories, Primary Narrators and Co-narrators can
have similar generalized representations and thus be on an equal footing
epistemically. Consider (3.4), where the details of how to order checks
are provided and discussed by the Primary Narrator and the Co-narrator
as partners with equal rights and where, accordingly, turn size is not
skewed to any of them and, consequently, TSF is, on average, small and
heterogeneous:
(3.4) “How to order cheques” (Type: T1G / Embed Level: EC2)
CNN
CNI N Coeff.
S3 PNP You write to the bank and sayQSB 7
S1 PNC [MDDI haven’t got any cheque books.] 7 0
S3 PNP Well. 1 0.75
UN [. . ..]
S3 [MSSDo you ring the printer then for
PNP 10 0.82
a cheque book.]
S1 PNC No, it’s suppose to come, you [unclear] 9 0.05
come automatically.
S3 PNP Well when you get another cheque 19 0.35
book, you write on the back please
can I have another cheque book.
(KCD-N2)
The fact that TSF between Primary Narrator (PNP) and Co-constructive
Recipient (PRC) is smaller and less homogeneous than between Primary
Narrator and Responsive Recipient (PRR) is related to the willingness of
PRC to interact with the tale, for example, by providing comments, extend-
ing story information, and asking questions (see Section 6.3), thus gaining
direct co-authorial rights. In (3.5), for instance, the Co-constructive Recipient
not only initiates the story by asking questions about David but also supports
and fuels the storytelling’s progress by posing two further questions. His
turns are only slightly shorter than the lengths of the Primary Narrator’s
turns, thus leading to a decrease in TSF:
106 How do narrators and recipients co-construct turntaking?
3.3.5 Summary
This case study concerned itself with the co-construction of turn size in two-
party conversational storytelling. The measure used to investigate this pos-
sibility was turn size fluctuation (TSF), expressed as a coefficient obtained
from a comparison of the lengths of the component turns in turn bigrams. It
was hypothesized that TSF would be more sizable and more homogeneous in
conversational narrative than in non-narrative conversation. While it was
demonstrated that TSF is significantly more sizable in two-party conversa-
tional narrative than in two-party non-narrative conversation, TSF was not
found to be more homogeneous in storytelling generally.TSF which satisfied
both criteria of sizability and homogeneity was discovered for recipients
interacting with narrators in the role of Responsive Recipient (PRR) only.
That is, for conversational narrative as a conversational subgenre the hypoth-
esis could not be adopted unreservedly. TSF in storytelling is distinguished
from TSF in general conversation only by its being more sizable: that is, the
fluctuation between the lengths of participants’ turns is significantly greater
than the fluctuation between turns of participants in general conversation. By
contrast, the hypothesis could be unreservedly adopted for one specific
interactional pattern, that of Responsive Recipients (PRR) interacting with
Primary Narrators (PNP): in this dyadic interaction, TSF was not only
significantly more sizable but also clearly more homogeneous than in story-
telling shared between Narrator and (i) Co-constructive Recipient (PRC) or
(ii) Co-narrator (PNC). Another key finding that emerged from the analysis
was that the type of recipient response varies significantly depending on the
length of the narrator turn preceding them. Long turns are correlated with
PRR responses, while shorter narrator turns are correlated with PRC and,
respectively, PNC turns. The possibility was noted that turn size is one of the
(presumably many) variables by which narrators can control the type of
response they elicit.
To explain the differences in TSF between the interactional pairings, two
possible sources were identified. First, the differences in TSF were seen in
relation to the specifications of the roles of Co-narrator (PNC) and the two
recipient roles Co-constructive Recipient (PRC) and Responsive Recipient
(PRR), viz. the privileged epistemic status of Co-narrators (PNC), which
allows them to interact with the narrator on an equal footing, (ii) the topical
3.3 Co-construction of turn size 109
4.1 Introduction
Speakers in conversation very frequently present discourse that occurred
in anterior situations. This phenomenon has been referred to under various
headings: as a change in footing (Goffman 1981), speech reporting (e.g.,
Carter & McCarthy 2006), speech, thought, and writing presentation (e.g.,
McIntyre et al. 2004) and discourse presentation (e.g., McIntyre & Walker
2011), the term used in this study. ‘Discourse,’ as understood in ‘discourse
presentation,’ casts the net wide to capture a broad variety of discourses
including internal thought and emotion, non-verbal gesture, and external
speech. For this latter category (external speech) further distinctions can
be made: external speech can be habitual speech (the words someone uses
or used to utter on similar occasions), hypothetical speech (the words
someone might or might have uttered in some situation), and actual speech
(the words someone purportedly uttered in a specific situation). This
variability alone makes discourse presentation a multifaceted phenomenon.
Yet two further aspects greatly add to the variability of the discourse
presentation system: quotatives and reporting modes. Quotatives are
typically verbs used to introduce forthcoming discourse presentation. As
shown in Chapter 2, the quotative forms used in the NC comprise a small
set of forms including SAY, GO, THINK and a few other forms.
Reporting modes, on the other hand, are different ways of presenting
discourse; in Chapter 2, I sketched the ten reporting modes annotated in
the NC. It was shown that direct quotation, also referred to as constructed
dialog (Tannen 1986, 1988), in its two incarnations as Free Direct and
Direct, is by far the most frequent report type in the NC. Given its
dominance in quantitative terms and its centrality in turning storytelling
into drama (Tannen 1986, 1988) this mode will be of particular interest in,
not only this present, but also the following chapter. The chapters will view
the use of discourse presentation as a means by which narrators, following
the principle of recipient design, attend to their interlocutors’ processibil-
ity needs (Chapter 4) and, respectively, intensify their interlocutors’ inter-
est in the telling (Chapter 5).
110
4.1 Introduction 111
Immediacy
The use of discourse presentation effects a discoursal intertwining of two
distinct discourse situations (see Short et al. 1996: 112): a (mostly) anterior,
told, situation (comprising its spatiotemporal setting, the speakers and their
communications) gets approximated toward a present, telling, situation
(comprising the present spatiotemporal setting, the present speakers along
with their present communications) in such a way that the two situations
come more or less to overlap discoursally.
The approximation is a gradient phenomenon: the degree to which the two
discourse situations can be overlaid varies depending on which discourse
presentation type is chosen. I will argue in the following that the degree of
overlay is in direct correlation with the degree of immediacy of the presenta-
tion type chosen.
Leech & Short (1981) assumed that discourse presentation types could be
aligned along a cline of narratorial interference, viz. in terms of the degree to
which the narrator apparently interferes with, or is in control over, the
discourse presented (Leech & Short 1981: 324).1 Their model of narratorial
interference has been highly influential in research on discourse presentation,
(e.g., Short et al. 1996, Semino & Short 2004, McIntyre et al. 2004, Short
2007, McIntyre & Walker 2011). Given that Leech & Short primarily exam-
ined discourse presentation in fiction it is not surprising that the model takes
the perspective of the narrator. Alternatively, discourse presentation, partic-
ularly in conversation, where interaction between participants looms large,
can be examined with a focus on what is achieved by it for the recipient. Seen
from the recipient’s viewpoint, another cline can be discovered: that of
immediacy.
Immediacy derives from Goffman’s (1981) analytical framework of ‘pro-
duction format’ and the choices among internal speaker roles it affords.
Goffman distinguishes, inter alia, ‘author’ and ‘animator,’ where author is
understood as “someone who has selected the sentiments that are being
expressed and the words in which they are encoded” (Goffman 1981: 144),
while animator is conceptualized as someone “mov[ing] his lips up and down
to the accompaniment of his facial (and often bodily) gesticulations, and
words can be heard issuing from the locus of his mouth” (Goffman 1981:
145).2 Clearly, under many (perhaps most) circumstances the two roles are
one: “the individual who animates is formulating his own text” (Goffman
1981: 145) and the speaker acts in the double role of author-animator. In
discourse presentation, however, the two roles do not simply overlay. While a
presenter, as any speaker, is, by definition, always in the role of animator, they
need not always and/or not fully perform the role of author, understood in
Goffman`s sense, of the discourse animated. The type of discourse presen-
tation where the two roles seemingly drift apart most is direct quotation, in its
incarnations as Direct and Free Direct: there, the presenter seems to animate,
not their own words, but the words of some story character. The author-
animator unity collapses: the presenter acts as ‘animator-only’ but not as
author-animator.
While this account is accepted widely (e.g., Holt & Clift 2007, Brendel
et al. 2011), it requires a conceptual extension. For the choice between
animator-only and animator-author by definition builds on ‘pretense’, a
notion discussed at large by Searle (1975) in the context of the illocutionary
acts performed in fictional writing. Anticipating the reservations (that there
may not be an original utterance and that quotations are selectively depictive
rather than faithfully descriptive), presenters, acting in the role of animator-
only, inevitably ‘co-author’ the quoted wording such that the exact locutions
in the anterior situation were different ones.3 In formulating the quote, the
presenter is, in actual fact, both animator and (co-)author of the quoted
utterances. But the utterance is presented as if no authorial interference
4.1 Introduction 113
+ Immediacy −
Animator Author
Pretended authorial interference
MDF MDD MIF MII MSS MVT MVV MUU MRR MRQ
with its wording had occurred. The effect the pretended dissociation from the
role of author has on the recipient is decisive: if the presenter seemingly
‘steps behind’ the characters merely lending their voice to them but letting
them speak for themselves, the illusion is created that “the incident is
presently occurring” (Mayes 1990: 346). Given this illusion, two additional,
virtual, participant roles emerge: the reported speaker becomes a virtual
addressor and the recipient a virtual addressee (see Section 1.3). As virtual
addressees, recipients ‘witness’ the reported discourse (see Holt 1996: 236),
and, as witnesses, they have immediate access to the reported utterance. In
terms of the above mentioned discoursal intertwining, the overlay of the
reporting situation and the reported situation is complete.
Crucially, in presenting discourse, the choice of animator-only and animator-
author is not an either-or choice: the pretended dissociation from the role of
author can be more or less. As a result, the recipient’s access to the discourse
reported can be more or less immediate. In Figure 4.2, I present a model of
discourse presentation where discourse presentation types are ordered on a cline
of immediacy understood as a function of pretended authorial interference.
To remind the reader of the differences between the categories, the
following examples may be helpful illustration; only the example for MDD
is authentic, all others are invented.
MDF Free Direct Can I have my videos?
MDD Direct Wayne said, can I have my
videos?
(KE5-N1)
MIF Free Indirect Could he have his videos?
MII Indirect Wayne said, can he have his
videos?
MSS Representation of Speech Act Wayne asked for his videos.
MVT Representation of Voice with He talked about his videos.
Topic
MVV Representation of Voice He sat there talking.
114 How do narrators mark quotation?
MUU Representation of Use He said “vids”.
MRR Reference to Discourse He really said that.
Presentation
MRQ Request for Discourse What did he say?
Presentation
Free Direct (MDF), on the extreme left end of the cline, represents the
category in which the reporting speaker’s pretense not to formulate his/her
own text but instead to merely animate that of a character is maximal. As
Leech & Short put it, in Free Direct, where the quote is not announced by a
reporting clause, a situation arises as if the narrator “has vacated the stage and
left it to his characters” (Leech & Short 1981: 334). The presented utterance is
maximally immediate to the recipient.4 In Direct (MDD), the category next
to MDF on the cline, the presence of the reporting clause begins to assert the
speaker’s rights as author: while the report itself is fully ascribed to the non-
present speaker whose discourse is animated, the introduction of the report
by the reporting clause is a reminder to the recipient of the presenter’s
potential to interfere with the report. Given this reminder (and its close
structural association with the quote5), the quoted utterance is slightly less
immediate to the recipient. In Free Indirect (MIF), the third category on the
cline, the utterance presented undergoes a decisive authorial revision: all
deictics anchoring the quoted utterance in the anterior speech situation are
deleted and replaced by deictics conforming to the presenter’s deictic system.
Thereby the presented discourse is transposed from the spatiotemporal
coordinates of the anterior context of utterance to those of the present context
of utterance. That is, in MIF, the animator-author balance is shifted even
more towards the presenter as author; consequently, the recipient’s access to
the original utterance is further restricted and immediacy is further
decreased. Indirect (MII), the subsequent category on the cline, strengthens
the authorial role of the presenter in that MII is distinguished from MIF by
the use of a reporting clause to announce that discourse presentation is
forthcoming. As with Direct (MDD), this announcement is an addtional
authorial intervention and has the effect of further constraining the recipi-
ent’s access to the original utterance.
The subsequent categories on the cline on the right of MII share an
important structural property. They are distinguished from the Direct and
Indirect categories by the fact that instead of a bi-clausal structure consisting
of reporting clause (such as he said and she goes) and reported clause (the
report itself), the presentation is expressed in one clause. The effect of this
structural reduction is that the discourse presented is no longer somewhat
detachable from the presenter’s own discourse: it has completely merged into
it. The mono-clausal structure greatly increases the presenter’s ‘authorship’
and, accordingly, diminishes the recipient’s access to the discourse pre-
sented. The first such mono-clausal category is Representation of Speech
4.1 Introduction 115
Demonstration
A second crucial distinction intimately related to considerations of immedi-
acy needs to be made between the two most immediate modes, Free Direct
and Direct, and the remaining modes. Three distinctive features separate the
two groups.
The first fundamental difference pertains to perspective, and, conse-
quently, reference. First, as noted, Free Direct and Direct quotes use the
reported speaker’s referential expressions including tensed verbs, person
deictics, and place deictics: reference is appropriate to the reported speaker,
not the reporting speaker (see, for example, Holt 1996: 222, Coulmas 1986: 2,
Mayes 1990: 346). The default ‘ego-centric’ organization of deictic expres-
sions is lifted: deictics are transposed, or ‘displaced’ (Hanks 2011: 318), to
some other origo, viz. “the person of the protagonist at the relevant time and
place in a narrative” (Levinson 2004: 111; cf. also Hanks 1992: 56). The non-
direct modes, conversely, use the reporting speaker’s referential system:
reference and origo are appropriate to the reporting speaker in the present
reporting situation.8 Second, given that Free Direct and Direct quotes are
main clauses (Mayes 1990: 338) (whereas, for example, indirect presentations
are realized in subordinate clauses), “direct quotes have fewer syntactic
restrictions” (Mayes 1990: 338). They thus allow for the enquoting of non-
clausal units, including most prominently vocatives, interjections, and excla-
matives, which may be highly emotively charged and are typically not
reportable in non-direct modes (see Banfield 1973: 7, Mayes 1990: 342–343).
Third, only Free Direct and Direct enable reporters to imitate ‘delivery
aspects’ (Clark & Gerrig 1990: 775) such as pausing, truncation, speech
defects, inarticulate sounds, within-speech laughter, intonation, voice effects
(pitch, quality, volume, drawling, etc.), and facial expressions and gestures
(see Buchstaller 2002, Romaine & Lange 1991:240, Clark & Gerrig 1990).
All three characteristics (imitation of referential system, non-clausal units,
and delivery aspects) concur such that in Free Direct and Direct quotation
118 How do narrators mark quotation?
an utterance model, this does not preclude the possibility that the grammar of
quotes diverges from the grammar of utterances. On the contrary, that
divergence is highly likely, for credibility is not the only constraint to
constructability. Another constraint is processibility: this principle demands
“that the text should be presented in a manner which makes it easy for the
hearer to decode in time” (Leech 1983: 64). As will be argued in more detail
below, processing quotation, particularly multiple quotation (that is, sequen-
ces of quotes), poses specific problems in that reference in such quote
sequences is in constant oscillation, thus exacerbating what I will call the
boundary issue (see below): the difficulty for the listener to tell the voice of
the quoting speaker from the voices of the quoted speakers. Therefore, the
utterance model can only co-determine the grammar of quotation: a perfect
utterance-quote fit is unlikely. In comparing quotes and utterances, a crucial
interest thus lies in the points of departure from the utterance model, that is,
the points where quotes significantly deviate from utterances. In Section 4.2,
for example, it will be shown that interjections occur significantly more often
at the onset of quotation than would be predicted on the basis of their
occurrence in natural utterances, thus alleviating the ‘boundary issue’ dis-
cussed in the following subsection.
It is only superficially “something of a paradox” (Holt 1996: 226) that
although quotation depicts rather than repeats an utterance, one of its key
functions is as a non-grammaticalized evidential (Clift 2006), that is, as
indexing the quoter’s epistemic authority as a reliable source of information,
a function which is significant in such diverse contexts as legal testimonies
(Galatolo 2007), political discourse (Kuo 2001), nursing shift handover
meetings (Bangerter et al. 2011), and even interactions between mediums
and their audience (Wooffitt 2007). Although selective and depictive (and
although participants in conversation are, probably, at least vaguely aware of
this), quotation inevitably has “an air of objectivity” (Holt 1996: 236). This
seeming objectivity is undoubtedly due to the three components of demon-
stration as a theatrical enactment outlined above: the switch into the reported
speaker’s referential system, the use of non-clausal units, and the imitation of
delivery aspects. The quoted utterance has all the ingredients it takes to
credibly uphold the pretense (see above) that the reported speaker issued the
reported utterance ‘in actual fact.’
latter, “[h]e relates a speech event as he would relate any other event: from his
own perspective” (ibid.). In other words: discourse presentation is not only a
discoursal intertwining but, in the case of demonstration/quotation, also an
intertwining of perspectives. I will argue in the following that the perspectival
intertwining entails the necessity for the narrator to appropriately mark the
boundaries of the perspectives adopted: I will refer to the issue as the boundary
issue. The following excerpt illustrates the issue.
The narrator is a 57-year-old housewife, relating to her interlocutor how a
friend’s husband reacted to learning that their daughter had apparently had
her first period:
(4.1) “Women problems” (Type: T30/Embed Level: EC1)
CNN
CNI
S1 PNS [MVT Did I tell you about her little one who had
stomach pains?] As she come back she said QSD
[MDDDad ] [MDFwhat? ] [MDFHow long’s our Mum
going to be before she comes in? ][MDFAnother hour.]
[MDFOh. ] [MDFWhy? ] [MDFOh well I’ve got a bit of a
stomach ache and I want to talk to her you know it’s
women problems. ] [MDD All right ] he saidQSD. Well
he knew QOO [MIIwhat it was. ] He said QSD [MDDyou
go up and lay in your bedroom ] he said QSD [MDDand
I can send her up when she comes in. ] All right. The
little ’un goes to bed. The little one’s heard Mummy
pull up on the drive and has come down the stairs
[MVVwell before anyone could say anything ] [MSShe
got it out. ] [MDFShe might have er period you’d
better go and sort her out. ] She said QSD [MDFwhat. ]
[MDFShe might have one of her periods you’d better
fucking go and sort her out. She’s your daughter ] She
said QSD [MDDthere’s no need to [unclear] up and say
it like that. ] She said QSD [MDDyou could have kept
this shut and I could tell her myself. ]
S1 PRR Oh!
CNF
S1 PNS [MDDWell ] he said QSD [MDDyour fucking daughter
you sort her out. ]
(KBE-N2)
The narrator uses 21 instances of discourse presentation. Only the first (Did
I tell you about her little one who had stomach pains?) refers to a discourse
situation in which the presenter was involved; all other instances are attrib-
uted to speakers not present in the telling situation. The narrator is animating
the speech of three non-present speakers: her friend’s husband’s, her
4.1 Introduction 121
friend’s, and their daughter’s. The narrator uses a restricted set of presenta-
tion types. As is shown by the tags, eight utterances are presented using
Direct (MDD), nine using Free Direct (MDF); one Speech Act
Representation (MSS) (He got it out) and one Indirect presentation (MII)
(Well he knew what it was) are used. Discourse presentation in the excerpt is
thus predominantly (Free) Direct.
In terms of perspective, the narrator is switching to and fro between no less
than four different perspectives: the three characters’ perspectives and her
own perspective (expressed for example in the preface Did I tell you . . .,
narrative talk, e.g., As she come back . . ., and all reporting clauses he said, she
said, etc.). The switches in perspective can be understood in terms of
Goffman’s (1981) production format as switches from ‘author,’ who is “for-
mulating his own text” (1981: 145), to ‘animator,’ that is, “the talking machine,
a body engaged in acoustic activity” (1981: 144) animating someone else’s text.
The perspectival and role switches performed in the main part of the
narrative are represented in (4.2); the three characters are referred to as
‘father,’ ‘mother,’ and ‘daughter’:
(4.2)
Perspective Role Realization
Presenter Author [MVT Did I tell you about her little one who had
stomach pains?] As she come back she said QSD
--------------------------------------------------------
Daughter Animator [MDDDad ]
--------------------------------------------------------
Father Animator [MDFwhat? ]
--------------------------------------------------------
Daughter Animator [MDFHow long’s our Mum going to be before she
comes in?]
--------------------------------------------------------
Father Animator [MDFAnother hour. ]
--------------------------------------------------------
Daughter Animator [MDFOh. ]
--------------------------------------------------------
Father Animator [MDFWhy? ]
--------------------------------------------------------
Daughter Animator [MDFOh well I’ve got a bit of a stomach ache and I
want to talk to her you know it’s women problems. ]
--------------------------------------------------------
Father Animator [MDD All right ]
--------------------------------------------------------
Father Author he saidQSD. Well he knew QOO [MIIwhat it was. ] He
said QSD
--------------------------------------------------------
122 How do narrators mark quotation?
Father Animator [MDDyou go up and lay in your bedroom ]
--------------------------------------------------------
Presenter Author he said QSD
--------------------------------------------------------
Father Animator [MDDand I can send her up when she comes in. ]
--------------------------------------------------------
Presenter Author All right. The little ’un goes to bed. The little one’s
heard Mummy pull up on the drive and has come
down the stairs [MVVwell before anyone could say
anything ] [MSShe got it out.]
--------------------------------------------------------
Father Animator [MDFShe might have er period you’d better go and
sort her out.]
--------------------------------------------------------
Presenter Author She said QSD
--------------------------------------------------------
Mother Animator [MDFwhat.]
--------------------------------------------------------
Father Animator [MDFShe might have one of her periods you’d better
fucking go and sort her out. She’s your daughter ]
--------------------------------------------------------
Presenter Animator She said QSD
--------------------------------------------------------
Mother Animator [MDDthere’s no need to [unclear] up and say it like
that. ]
--------------------------------------------------------
Presenter Author She said QSD
--------------------------------------------------------
Mother Animator [MDDyou could have kept this shut and I could tell
her myself. ]
In the excerpt presented in (4.2), the presenter constantly shuttles between
distinct roles (animator and author) and perspectives (her own, the father’s,
the mother’s, and the daughter’s). The narrator’s discourse is highly poly-
perspectival: altogether, 22 role and perspectival shifts occur. Crucially, the
shifts in role and perspective entail shifts in reference: the numerous occur-
rences of referring pronouns I/me/myself, you/your, and she/her oscillate in
reference across the whole narrative (only the male pronoun tokens referring
to the only male character involved do not undergo this shift in reference).
This oscillation in reference poses heavy demands on the listener: each time
the perspective shifts reference must be reassigned to changed referents. The
crux in reassigning reference is the question when to perform the reassign-
ment. The reference encoded in referring expressions is not decoded for each
expression individually but in the context of the discourse unit in which the
4.2 Interjections as quotation markers 123
roles they fulfill in normal conversation, a new function that is not found in
non-quotative discourse: a discourse-deictic function of marking quotation. I
will discuss the use of interjections as quotation markers in keeping with the
recipient-design principle: as a service to the recipients by which narrators
flag the imminence of quotation, the switch in reference system that is
effected by (Free) Direct quotation, and the need for the recipients to resolve
reference polyperspectivally, that is, as according to the perspectives of the
various reported speakers.
Before embarking on the case study, let us explore the terrain: what are
interjections? A useful starting point definition is as members of the class of
inserts, which Biber et al. (1999: 56) describe thus:
They do not form an integral part of syntactic structure, but are inserted
rather freely in the text. They are often marked off by intonation, pauses, or
by punctuation marks in writing. They characteristically carry emotional and
interactional meaning and are especially frequent in spoken texts.
That is, four features typify inserts and hence also interjections: syntactic
isolation, intonational separation, emotional and interactional meaning, and
virtual restriction to speech. Given that the NC is a speech-only corpus and
not annotated for intonation, only syntactic isolation and the expressive
and interactional significance of inserts are of special relevance in the present
study. Particularly distinctive is their ‘compositional’ agility: because inserts
are syntactically free elements which normally do not enter into constructions
with other words, they tend to occur either utterance-initially or constitute
utterances by themselves (see Ameka 1992: 105). This sets them apart from
other typically utterance-initial items such as the pronoun I and the coor-
dinator and, which, under normal circumstances, cannot constitute complete
utterances.10 Inserts come in a wide variety: besides falling into the classes
of greetings, discourse markers, attention signals, hesitators and other
subtypes (see Biber et al. 1999: 9394), inserts include first and foremost
interjections.
Ameka (1992) distinguishes two basic types of interjections: primary and
secondary. Primary interjections, such as oh and cor, are defined as “relatively
conventionalised vocal gestures (or more generally, linguistic gestures) which
express a speaker’s mental state, action or attitude or reaction to a situation”
(Ameka 1992: 106). Contrary to ‘secondary interjections’ such as Holy Christ,
shit, etc., which are forms that belong to other word classes and can thus be
used in non-interjectional functions, primary interjections are “not used
otherwise” (Ameka 1992: 105). Interjections pose serious problems for classi-
fication: they can serve as backchannels (mm), hesitation markers (erm),
attention signals (hey), expletives (god), ad hoc vocalizations (whoa), and
discourse markers (yeah) (see Quirk et al. 1985, Norrick forthcoming). A
functional classification of interjections is proposed in Ameka based on the
functions of language developed by Bühler (1990 [1934]) and Jakobson (1960).
4.2 Interjections as quotation markers 125
Rühlemann (2007: 139ff.) who reports on frequencies of oh, well, and look as
collocates of the quotatives SAY, THINK, and GO. Crucially, all these
studies converge in noting that interjections used as signals of quotation
occur quotation-initially.
What all of the above-mentioned studies fail to provide is evidence to
support the notion that the function of interjections as quotation markers
differs in any way from the usual functions of interjections in utterances. If
interjections occur utterance-initially and quote-initially, it does not auto-
matically follow that the occurrence in initial position in constructed dialog
signifies a (new) function as a ‘signal’ that speakers ‘are embarking on direct
speech quotation.’ It could be simply that the way interjections occur in
direct reports is a faithful mimesis of the way they occur in natural utterances.
For example, McCarthy (1997) notes that oh and well are often included in
quotation “in the way they might be expected to occur in real conversation”
(McCarthy 1997: 159). If that were indeed the case, no special signal, or
marking, function could be ascribed to them. In order to establish whether
or not interjections in quotation gain an additional discourse-deictic marking
function as auditory quotation marks one will have to compare the frequen-
cies of occurrence of interjections in (i) (Free) Direct reports and (ii) con-
versational utterances. Only if a significant overuse in quotes can be observed
will it be possible to claim such a function as a marker of quotation.
In the following case study on interjections, precisely this comparison
between interjections as quote- and utterance-first items will be performed.
The identification of interjections will be largely based on the POS tagging
available in the NC. One immediate limitation to this methodology needs to
be acknowledged: interjections are “not consistently marked in corpora”
(Norrick forthcoming). The BNC and, by consequence, the NC are no
exception to this: the Claws5 (c5) tagset recognizes primary interjections
only. There is thus no way of automatically identifying and analyzing secon-
dary interjections such as oh my god (components of) which can be used non-
interjectionally. While it seems that most secondary interjections are much
less frequent than primary interjections (see Norrick forthcoming), there is
one secondary interjection of outstandingly high frequency: the item well,
which is, in conversation, predominantly used as a discourse marker, partic-
ularly in first position. This item is consistently annotated as ADV (adverb)
in the BNC and the NC. Given its high frequency and prominent role in first
position, it will be included in the analysis.11
20
15
15
Density
Density
10
10
5
5
0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Positions Positions
4.2.3 Results
To initially approach the data, consider the histograms in Figure 4.3, repre-
senting the distribution of positions of interjections in the two subsets.
The density curves in the histograms in Figure 4.3 represent the proba-
bilities of occurrence of positions of interjections estimated on the basis of the
positional frequencies (see Dalgaard 2008: 59). The histogram in the left
panel provides some initial evidence to support the above cited observation
that interjections tend to occur early in utterances: the first bin, representing
the positional values between 0 and 0.1, is clearly the tallest, and the density
curve has a very clear maximum in this segment. However, the first bin in the
right panel, representing positions of interjections in quotes, is even taller than
4.2 Interjections as quotation markers 129
that for utterances. What is more, the height of the curve’s primary maximum
for interjections in quotes is many times the height of the curve for interjections
in utterances. That is, the relative probabilities of finding interjections in very
early positions in quotes are dramatically higher than to find them in early
positions in utterances. Note also that in both panels the peak is precisely at
positional value 0 for first position. This can be taken as preliminary evidence
that interjections both in utterances and quotes may be primed to occur in that
first position. What we do not know yet though is whether this is a general
tendency applying to all or most interjections or whether the steep peaks are
‘powered’ by only a few interjections and can thus not unreservedly be
generalized. It is therefore necessary to establish, not how the class of inter-
jections, but how individual interjections behave positionally.14
In the following, I identify the most frequent interjections in utterances
and quotes as well as their frequencies in first position. I begin with inter-
jections in conversational utterances.
Table 4.2 lists the top ten most frequent interjections in conversational
utterances; it also gives the frequencies of occurrence in position 0 (that is, as
the very first word).
As can be seen in Table 4.2, the three most frequent interjections in
conversational utterances in the NC are yeah, well, and oh – a good match
of the top three interjections in large general corpora such as the Longman
Spoken and Written English Corpus (LSWEC), where, in that order, yeah,
oh, and well (see Norrick 2008: 444) were the three most frequent interjec-
tions. As regards positioning, oh has both the highest absolute frequency for
first position (430 occurrences) and the highest proportion for that position
(74%). Generally, it can be seen that, for nine out of the ten most frequent
items, the proportion of occurrence in first position (where the positional
value is 0) is above 50% (only ha has a (much) lower percentage), while the
overall proportion for occurrence in first position for all ten items is 58%.
Note that the nine interjections with high proportions have a combined
130 How do narrators mark quotation?
frequency of occurrence of 2,727 (out of 3,007) and thus account for 91% of
all the interjections in conversational utterances. This is clear evidence to
suggest that the above-noted trend for interjections to overall occur in
utterance-first position is indeed a reflection of a general trend which is
powered not by a few isolated interjections but by almost all of the most
frequent interjections.
So far, we have focussed on frequency of occurrence in first position. How
are interjections in conversational utterances distributed across all positions?
Taking the median, indicated in the bold vertical lines in the boxes in
Figure 4.4, as a measure of central tendency, we see that only the positions of
ooh and ha have a median value greater than 0. The medians for the remain-
ing eight interjections are invariably 0. This is further evidence to suggest
that the trend for the most frequent interjections to occur in first position is
indeed a central trend. Also, the outstanding status of oh is confirmed: it is not
only, as noted, the item with the highest share in occurrence in first position,
but its IQR (the interquartile range comprising the 50% of the data around
the median) is by far the smallest. This indicates that its positional homoge-
neity is greatest. Given this homogeneity and its high frequency of occur-
rence, the notch in the box for oh (depicting the interval in which the true
median is estimated to lie) is minimal. This suggests we can be highly
confident that the true median is indeed 0. The notches for yeah, well, and
no are slightly wider suggesting we can be slightly less confident that the true
median will be 0 too. The notches for the remaining items, however, are
much wider, warranting much less confidence. As regards dispersion, the
second least dispersed positions (after those for oh) are for no, where the IQR
is 0.27, while the IQRs for yeah and well are both 0.4. Finally, the IQR for all
eight interjections with median position 0 is 0.33, that is, exactly within the
first third of the positional range between 0 and 1. In sum, we can say that
interjections in conversational utterances have a very clear tendency to occur
not randomly across positions but within the first third of the utterance and,
particularly, at the very onset of the utterance (that is, in first position). This
can be said with great confidence of the interjections oh and, with slightly less
but still sufficient confidence, of the items well, yeah, and no. In priming
terms: oh and, to a lesser degree, yeah, well, and no, are primed for use not
only initially but also, more specifically, for use in first position in conversa-
tional utterances.
How are positions of interjections distributed in quotes?
First, note that the list presented in Table 4.3 only partially overlaps with
the list of items identified in previous research (see above), which included
oh, well, ah, look, and okay. The forms look and okay mentioned in Biber et al.
(1999) are notably missing from Table 4.3. This is unsurprising since they are
not tagged as interjections. However, even lexical searches for positions of
look and okay in quotes returned very low figures: look occurred 7 times and
okay just once in quote-first position. I suspect that the items may be more
frequent in quote-first position in AmE, the variety characteristic of a large
part of Biber et al.’s corpus.
oh
well
no
yeah
ooh
yes
hello
ah
ha
blah
The following differences will be noted if we compare the list of the ten most
frequent interjections in quotes shown in Table 4.3 with those most frequent in
conversational utterances shown above in Figure 4.5. First, mm and aye are
missing from the list for quotes, which instead contains hello and blah. While
the frequency of aye in utterances is comparatively low, the frequency of mm in
utterances is not. Its absence from the list of interjections in quotes is therefore
curious and worthy of further investigation.15 Second, the proportions of
occurrence in first position seem to be higher in quotes than in utterances,
both overall (note the overall proportion of 77% compared to 58% for utter-
ances) and for specific items: these include most notably ah, where the
proportion of use in first position is 100% for quotes compared to 61% for
utterances, oh (90% vs. 74%), well (75% vs. 51%) , and no (69% vs. 56%). The
proportions of use in first position for yeah, by contrast, are the same (57%).
The boxplots in Figure 4.5 show that, not surprisingly, the notches
(indicating the 95% confidence interval) are extremely wide for the less
frequent items such as blah and ha, which we can therefore safely disregard.
4.2 Interjections as quotation markers 133
However, the notch for yeah is also wide so that, for yeah, it cannot be said
with confidence that, in quotes, its true positional median is 0. That this true
median is 0 can be maintained with great confidence for oh and well: for these
two items the attraction to, or priming for, first position is so dramatic that the
interquartile range (IQR) coincides with that median, which is why no boxes are
drawn around the median (ooh, hello, and ah, for which no boxes are drawn
either, are of much lesser frequency than oh and well and no priming claims will
therefore be made for them). Accordingly, the IQRs for oh and well are 0,
indicating that the homogeneity of their distribution could hardly be greater.
These two items, then, exhibit a very strong attraction to first position in
quotes. In priming terms: oh and well are primed for use as the immediately
first word in quotes. As regards the remaining eight interjections, their fre-
quencies of occurrence in quotes are below 100 and hence too low to warrant
strong priming claims. These items are therefore not investigated further.16
To wrap up the findings on oh and well so far, the analysis discovered
evidence that both oh and well are strongly attracted to first position both in
conversational utterances and in quotes. It was found that this attraction to
first position is greater in quotes than in utterances. To test this latter finding
for significance, chi-squared tests were carried out. The results are summar-
ized in Table 4.4.
We see that the associations for oh and well are very highly significant: the
p-values for the two items are by far smaller than 0.05. The effect sizes,
measured by the Cramer’s V values, can be considered modestly strong: the
values are roughly 0.2 for both items (Cramer’s V ranges between 0 and 1).
The H0, which stated that certain interjections occur as frequently in first
position in quotes as in conversational utterances, can hence be rejected: oh
and well are used significantly more often in quote-first than utterance-first
position. Oh and well, then, are precisely not included in quotes ‘in the way
they might be expected to occur in real conversation’ (see McCarthy 1997:
159), they are included in quote-first position much more often than might be
expected on the basis of their occurrence in utterance-first position.
4.2.4 Discussion
What to make of this significant result? At first, I briefly review previous
research on oh and well.
134 How do narrators mark quotation?
actual conversation and (ii) narrators use oh and well as auditory quotation
markers alleviating the boundary problem of quotation at the left periphery. I
will discuss the two interpretations in what follows.17
Upon closer scrutiny, interpretation (i) is less implausible than it may
appear at first sight. First, demonstration theory holds that speakers depict
utterances selectively (see Section 4.1). If selection is thus a design principle
in using quotation, it cannot be ruled out that speakers can also pre-select
certain utterance types for presentation. Given the two interjections’ core
functions as indices of surprise (oh) and dispreferreds (well), the pre-selection
of oh- and well-introduced utterances for quotation would amount to a pre-
selection of utterances expressing surprise and deviation from expectations.
Such utterances might be perceived as intrinsically interesting to the recip-
ient, loosely speaking, as ‘more dramatic.’ Their pre-selection could thus
serve to intensify the recipient’s interest in the storytelling (see Brown &
Levinson 1987). In other words, pre-selection of oh- and well-prefaced utter-
ances for quotation would be a strategy in the service of recipient design.
Second, to support this theory it may be argued that if indeed oh and well
acquired a discourse-deictic function as quotation markers this new function
would possibly have an effect upon the usual functions oh and well carry out
in (non-quoted) utterances. It might be the case that when used to flag
quotation the items will no longer fulfill the usual utterance functions such
as to express surprise, acknowledge receipt of new information, and preface
deviation from expectations and so on and instead only attend to the new task
of marking quotation. That deletion effect, however, does not seem to exist. It
is easy to find in the data examples of oh and well used at the onset of
quotation that still preserve their ‘original’ meanings. Consider (4.4) to
(4.7), which illustrate uses of oh and well in quote-first position.
oh:
(4.4) black welding glasses on, and he turned round and he made me jump.
LikeQLK [MDD oh, Colin, ] and
(KB1-N2)
(4.5) [MDDI locked all [unclear] off from under my windows ] I saysQSZ .
[MDDAnd I’m about six foot four and I’ll handle anyone that breaks in
here. ] [MDD Oh that’s alright then ] he saysQSZ
(KDY-N1)
well:
(4.6) she saidQSD [MDDit was all packed up in a box ] and I saidQSD [MDD
well you wouldn’t see it then would you? ]
(KE6-N2)
(4.7) Women have babies. I said QSD [MDD well it comes from a man. ] So
Luke said QSD [MDDwell what are they like? ] So I said QSD [MDDwell
they’re like little tadpoles. ] And he went QGD [MDD well I can’t feel
any. ]
(KC5-N1)
136 How do narrators mark quotation?
In (4.4), the presenter reports the speaker’s surprise (he made me jump) at
recognizing Colin (oh Colin) ‘hidden’ behind black welding glasses. In (4.5),
the quote Oh that’s alright then acknowledges the new information conveyed
by the interlocutor’s description of his tall build (I’m about six foot four) and
his reassurance that he’ll handle anyone that breaks in. In (4.6), the female
interlocutor is apparently looking for something although she had it all packed
up in a box, to which the quoted reply is a hint that she might not be able to
find it because she packed it in a box; the reply then carries a critical overtone
thus making it a dispreferred. (4.7), finally, contains quoted efforts by adults
to explain to a boy what sperms are. The explanations given are evasive (well
it comes from a man) and iconic (well they’re like little tadpoles), with well at the
onset signalling the adults’ uncertainty as to whether their descriptions will
be adequate for the child. The quoted responses by the boy also marked by
well seem indeed to contain a degree of dissatisfaction: in response to the
adults’ vague description it comes from a man the boy demands a more specific
description (well what are they like?) and the adults’ tadpole simile is overtly
rejected (well I can’t feel any). Note that these are randomly selected examples
and that more such examples could probably be found. Upon inspection of
the data it appears that oh and well keep their usual discourse-marking
functions even in quotation. The assumed additional quotation-marking
function does not cancel them out or noticeably compromise them. This,
then, the argument concludes, could be evidence that oh and well have no
additional quotation-marking function.
How to counter these arguments? First, one weakness of the selectivity
argument is that there is no evidence to support one important entail-
ment. This entailment is that if we assume that speakers select utterances
conveying surprise and dispreferreds for their greater intrinsic interest
for the recipient we would in a similar vein (have to) expect that speakers
generally select more frequently for presentation those utterances that
‘score’ higher in terms of dramatic potential, including, for example,
utterances conveying strong emotions such as disgust, anger, amazement,
or pain. I am not aware of any evidence (in any research) supporting
this notion. Nor does the NC provide such evidence. If interjections are
taken as possible indicators for these kinds of utterances, searches for
interjections commonly associated with these emotions in quote-first
and, respectively, utterance-first position do not uncover evidence to
support the notion. As can be seen in Table 4.5, candidate interjections
such as urgh and ugh for disgust, fuck for anger, wow for amazement, and
aargh or ow for pain, hardly ever occur either in quote- or utterance-first
position:
To be sure, utterances expressing strong emotions can be made in many
other ways. It can therefore not be ruled out that indeed the selection-by-
dramatic-potential is a principle that applies generally rather than only to
utterances conveying surprise and dispreferreds. However, its general
4.2 Interjections as quotation markers 137
oh well
Quotes in NUtts
Quotes in NUtts
CUtts
CUtts
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Positions Positions
In Figure 4.6, the dark grey boxes represent positions of oh and well in
quote-first position within their enclosing narrative utterances (NUtts), while
the light grey boxes represent their (already familiar) positions across con-
versational utterances (CUtts) (see Figure 4.4 above). First, note that the
whiskers for quote-first oh (but not for quote-first well) in NUtts extend to
the value 0. This indicates that indeed some instances of oh (but not of well)
occur in first position both in the quote and in the enclosing NUtt (i.e.,
utterance-initially). But they are a minority. The clear majority of instances of
quote-first oh, and also well, occur in positions far higher than 0. This can be
seen instantly from the fact that the (dark grey) boxes for quote-first oh and
well in NUtts are located much farther on the right than the (light grey) boxes
for oh and well in CUtts. This visual distance shows up mathematically as the
difference between the medians (indicated in the bold vertical lines in the
boxes): the median for oh and well in CUtts, which, as noted earlier, is 0 for
either item, contrasts starkly with the median for oh and well used in NUtts,
which is 0.4 (again for either item). The difference is very highly significant:
according to a Wilcoxon rank sum test, for oh, p-value < 2.2e-16, and for well,
p-value = 1.147e-07. This is clear evidence that the two primings for oh and
well, viz. for utterance-launch and quote-launch, can be kept apart on
positional grounds. Altogether, then, recipients can draw on three factors
4.2 Interjections as quotation markers 141
4.2.5 Summary
This case study on positioning of interjections in quotes as compared to
conversational utterances investigated the possibility that interjections serve
a quotation-marking function. It was shown that two interjections – oh and
well – occur significantly more often in very first position in quotes than they
do in conversational utterances. Two possible interpretations were discussed:
that utterances expressing surprise (oh) and deviation from conversational
expectation (well) are preferably pre-selected for quotation and, alternatively,
that oh and well are primed for, not only utterance-first (in general conversa-
tion), but, even more strongly, quote-first position (in narrative), thus serving
an additional discourse-deictic role of flagging the left-hand boundary of
presented discourse. The two interpretations are by no means mutually
exclusive: why should presenters not both pre-select certain utterance
types that promise to be of interest to the recipient and, at the same time,
value the discourse-structuring effect of certain interjections that typify these
utterance types? Also, the two interpretations align to a larger principle, the
principle of recipient design: no matter whether oh and well are interpreted in
the context of pre-selection or as left-boundary quotation marks, their use in
quotes will primarily serve the recipient: their interest (in drama) and need
(for processibility). Seen thus, as means in the service of recipient design, the
statistical overuse of the small items oh and well in quote-first position
contributes to the large enterprise binding narrator and recipient together:
that of co-constructing narrative in conversation.
142 How do narrators mark quotation?
While this section has investigated positioning in the use of verbal items,
viz. interjections, the next section targets the positioning of non-verbal items,
viz. unfilled pauses.
H1: Short silent pauses occur more frequently in initial and final positions
than in medial positions in quotes.
How are initial and final positions defined? I define as initial positions those
positional values that range from 0 to 0.2, mathematically: ϵ[0,20], and as final
positions those values ranging from 0.8 to 1, mathematically ϵ[80,100].
included within the scope of the quote although, in fact, they may have been
placed outside the <seg>-element delimiting Direct and Free Direct quotes.
Re-consider (4.9), printed here as (4.10), which gives a little more context as
well as the annotation for reporting mode:
(4.10) “Bloody ignorant pig” (Type: T10/embedLevel: EC1)
(. . .)
S1 PNP I thought <pause/> [MDD you shouldn’t speak to people
like that] but <pause/> [MRR I’d said it]
(KB0-N2)
It can be seen that the pause between thought and you is placed outside the
quote’s boundaries defined by the annotation. However, note also that no
verbal material (that is, in annotational terms, no <w>-element) intervenes
between the pause and the quote. Therefore, for the purposes of the present
research, pauses like the first in (4.10) were included, not within the quote (as
said earlier, pauses cannot be seen as objects of quotation), but within its
positional scope. That is to say that pauses occurring immediately before the
quote were assigned the positional value 0, that is, the same value that would
be obtained if the pause had been included inside the quote’s annotation. By
contrast, pauses occurring before the quote but separated from it by inter-
vening verbal material were excluded. By the same token, pauses occurring
immediately after the completion of a quote were included within the quote’s
positional scope, whereas after-quote pauses distanced from the quote by
inbetween word elements were excluded. For example, in (4.10), the pause
following the quote was discounted given the intervening conjunction but,
which signals the inception of a new clausal unit. Note that for silent pauses
occurring at the extreme right end of the positional scope (that is, as quote-
last element or as the very first element after the quote), the positional value is
1. This is because, as noted in Section 4.2, underlying the calculation of
positions is the division of the number of words preceding the item whose
position is to be determined by the total number of words in the unit (the
quote, in this case). Silent pauses, however, do not represent words but XML
elements of their own, which, in XPath terms, are treated as siblings to word
elements (<w>) (see Section 2.3.2). They are hence not counted as words.
Consider (4.11): the quote contains 18 words. The position of the pause
following the quote is 18/18 = 1, a value which adequately reflects the
item’s quote-boundary position.
(4.11) “Microwave” (Type: T10/embedLevel: EC2)
(. . .)
S1 PNP (. . .) And she QSDsaid [MDD no you have it my love. I’ve
just bought a new one and you can have it] <pause/> (. . .)
(KB6-N1)
146 How do narrators mark quotation?
Freq %
<pause/> 362 77
er 63 13
erm 47 10
Total 472 100
4.3.3 Results
At first, it seems appropriate to establish how frequently silent pauses occur
within the above-described positional scope of quotes. For comparison,
Table 4.6 also presents the observed frequencies of the filled pauses er and
erm, by far the most frequent forms (other potential forms such as ehm or uhm
are not attested in the NC).
Short silent pauses are by far the most frequent pause types in the data: the
362 instances of <pause/> account for far more than two-thirds of all pauses
(77%) , while the 63 ers and 47 erms combined take up the remaining 23%. So
pausing in the context of quotation is indeed mainly a non-verbal phenom-
enon and we are justified in restricting the focus of this research to silent
pauses. However, considering that the total number of quotes in the NC is
1,784 (see Section 2.2.2), the 362 silent pauses in the data are involved in less
than a fifth of quotes overall (‘less’, because, as will be shown below, many
quotes have two pauses, one on either end).
How do pauses distribute positionally? If the above-noted assumption is
correct (that silent pauses are more frequent in initial and final positions than
in medial positions) we will expect to find a U-shaped distribution across the
quotes’ positional scope as defined above; that is, a distribution will result that
has exactly two peaks: one for initial positions, defined as the positional
segment ϵ[0,20], and one for final positions, defined as ϵ[80,100].
Figure 4.7 shows the distribution of positions of silent pauses and of the
filled pauses er/erm as histograms. It will be immediately obvious from the
4.3 Pauses as quotation markers 147
4 <pause/> er/erm
4
3
3
Density
Density
2
2
1
1
0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Positions Positions
graph in the panel on the left that silent pauses behave almost exactly in the
way predicted, namely with the highest frequencies of occurrence in initial
and final positions, resulting in a density curve which indeed resembles a
U and peaks located exactly in the two critical positional segments ϵ[0,10] and
ϵ[90,100]. However, the peaks are of unequal size: the left-hand peak is
greater than the right-hand peak. This distribution suggests that the prob-
ability for silent pauses to occur quote-medially is smaller than to occur
quote-initially and also, to a lesser extent, smaller than to occur quote-finally.
By contrast, a U-shaped type of distribution is not found for er/erm:25 while
the filled pauses too have their primary peak in initial positions, their
secondary peak is not, as expected, in final but pre-final positions, more
specifically in the range ϵ[60,80]. Contrary to expectations, in the final
148 How do narrators mark quotation?
segment the curve even reaches, not its second high point, but its lowest
point. This distribution suggests that er/erm are less likely to occur quote-
finally than pre-finally and much less likely than initially.
To what extent are the above observations significant? Since the focus is on
silent pauses, the question of significance will only be addressed for this type
of pause. That is, to what extent can significance be claimed for the prefer-
ences for initial and final positions observed for short silent pauses? The
‘natural’ answer to this question would be to perform an appropriate stat-
istical test. However, significance tests essentially involve juxtaposing the
distribution of a given variable against the distribution of another comparable
variable or against a specific distribution (e.g., normal distribution, exponen-
tial, etc.) or against an a priori stipulated value (Dalgaard 2008: 95). In the
present case, however, we have one sample only and no a priori benchmark
value against which to compare the positioning of silent pauses. One way to
nonetheless make reliable inferences is the non-parametric density estimation
and the use of the bootstrap to compute confidence intervals. For reasons of
space, this method cannot be discussed here in great detail (for a readable
introduction, see Mooney & Duval 1993); some basic characterization must
suffice.
The bootstrap is a non-parametric resampling method introduced in Efron
(1979). In linguistics, it is used occasionally in computational linguistics (e.g.,
Erk et al. 2010) and machine translation (e.g., Koehn 2004) and is beginning
to be used in corpus research (e.g., Gries 2006, Lijffijt et al. 2011). Underlying
it is the “approach to treat the sample as if it is the population” (Mooney &
Duval 1993: 9). From the sample of size n, a large number of resamples are
drawn randomly. Crucially, the resamples are of the same size n as the
original sample and are drawn with replacement. Therefore, each resample
may contain some sample data points more than once whereas other sample
data points are not included at all. As a result, “each of these resamples
will likely be slightly and randomly different from the original sample”
(Mooney & Duval 1993: 10). These differences have repercussions: the
mean and the standard deviation of each resample vary from those of one
another and from those of the original sample.26
Why generate resamples? Generating a large number of resamples with
slightly different kernel density estimations enables the calculation of point-
wise standard errors for the estimated density and, hence, the calculation of
confidence intervals (CIs). These are crucial to the method in that the CIs
facilitate statistical inference (see Section 3.2). Now, crucially in the present
context, CIs can be visualized by confidence bands. Confidence bands are
obtained from a six-step procedure (executed in R): (i) draw a large number
of resamples, (ii) for each resample estimate the density in 512 points (a
number predefined in R) dispersed across the minimum and the maximum
values in the data (that is, in the case of positional values for pauses, between 0
and 1), (iii) calculate the mean for each of the 512 points over the resamples,
4.3 Pauses as quotation markers 149
(iv) calculate from all resamples for each point one single standard error (a
measure of how widely the density estimations vary), (v) calculate on the basis
of the standard errors the 512 confidence intervals, and (vi) connect them
such that they form the confidence band: its bounds delimit all those density
estimates that cover in 95% of all cases the true density estimate of the
population in question.
The confidence band obtained from the procedure invites interpretation.
Crucial to its interpretation is its width. It determines the precision and hence
reliability of the density estimate for the original sample: the narrower the
band, the greater the precision and reliability; the broader the band, the
smaller the precision and reliability. Since the bounds of the band delimit all
those density estimates that cover the true density estimate in 95% of the
cases, any conceivable curve representing any conceivable estimate within
the band could, in principle, represent the true estimate. If the upper and the
lower bounds of the confidence band are so far apart that between them
curves are conceivable whose shapes are essentially different from the density
curve obtained for the sample, then the shape of the density curve for the
sample is an unstable and unreliable foundation for any inference as to the
shape of the curve to be obtained from the true population. Such an essen-
tially different shape would obtain, for example, if any two points within the
confidence band could be connected in such a way that a completely new
primary maximum would arise at a point along the x-axis where, in the
original sample’s density curve, there is no such maximum. Such a scenario
is the more likely the wider the confidence band. If, by contrast, the width is
narrow such that alternative contours would merely modify the existing
contour in negligible details, then the ground on which to infer from sample
to population is stable and reliable: we can be 95% confident that the micro-
cosmos of the finite sample at hand is a faithful mirror image of the macro-
cosmos of the population.
For a reliable bootstrap, a sample size of 30 values is often considered a
sufficient basis (see Mooney & Duval 1993). The sample size of 362 silent
pauses and their positional values is thus more than sufficient. On this
sample, a bootstrap was performed with 5,000 resamples drawn randomly
from the original sample; the significance level α was set to 0.05. The results
can be inspected in Figure 4.8. The figure shows the same histogram and the
same density curve for positions of pauses in quotes as in the left-hand panel
in Figure 4.7 above. The dashed lines below and above the density curve
delimit the confidence band calculated on the basis of the bootstrap.
As said above, the interpretability of the confidence band depends on its
width. If the band is very wide, admitting shapes of density estimates largely
different than the density estimate for the original sample, then inferences
made on the basis of the sample’s curve are hardly reliable. If, by contrast, the
band is narrow admitting at best variations of the existing curve, then
the curve is meaningful in terms of statistical inference. In the case of the
150 How do narrators mark quotation?
Bootstrap
4
3
Density
2
1
0
confidence band shown in Figure 4.8, the latter scenario is clearly the case:
the band is markedly narrow, particularly around the values 0.25 and 0.75. At
these points on the x-axis, any alternative curve must follow, on the left, the
original curve’s decline from its primary maximum between 0.0 and 0.1 as
well as replicate, on the right, the curve’s rise towards its secondary max-
imum between 0.9 and 1. Around the peaks, the band is a little wider. This
indicates that, here, alternative contours are conceivable which peak not, as in
the original sample, between, on the left, 0 and 0.1 and, on the right, between
0.9 and 1, but between 0.1 and 0.2 and, respectively, between 0.8 and 0.9.
However, peaks beyond 0.2 and, respectively, 0.8 are not conceivable.
Further, the highest point of the valley between the peaks is still much
lower than the lowest point of the primary maximum on the left and the
lowest point of the secondary maximum on the right. Even if the true density
estimate covered these extreme points, the overall shape of the curve would
form a U: a (slightly lower) primary maximum for initial positions and a
4.3 Pauses as quotation markers 151
(slightly lower) secondary maximum for final positions. The original sample’s
maxima would still be firmly in place and the overall contours of any
alternative density curves would still be the same as the contour of the density
curve obtained for the sample. That is, the confidence band calculated on the
basis of the bootstrap suggests that the true density estimate for short silent
pauses in English narrative will essentially resemble the density estimate
calculated for the sample drawn from the NC, exhibiting peaks for initial
positions and final positions and a valley for middle positions. This is
conclusive evidence, then, to suggest that the alternative hypothesis accord-
ing to which short silent pauses are more frequent in the positional segments
ϵ[0,20] and ϵ[80,100] than in any positional segments in between, can be
accepted.
4.3.4 Discussion
How to interpret these positional preferences for initial and final positions of
short silent pauses? Two possible interpretations come to mind. Rühlemann
et al. (2011) propose a speaker-centered interpretation that links the use of
pauses in quotation to processing constraints imposed on the presenter by the
necessity in quotation to perform a “shift into a non-present speaker’s voice
and reference system” (Rühlemann et al. 2011: 225). This interpretation is in
line with Allwood et al.’s (1990) view of pauses as choice-related speech
management expressions, serving “to enable the speaker to gain time for
processes having to do with the continuing choice of content and structured
expression” (Allwood et al. 1990: 10–11). Rühlemann et al. (2011) were con-
cerned only with initial positioning of pauses. They were not aware of the fact
that silent pauses also peak at final positions in quotes. This oversight has
reverberations for the interpretation of how pauses position. If it is assumed
that pauses peak only at the onset of quotation the speaker-centered inter-
pretation sounds highly plausible: the shift from the speaker’s own default
reference system into a ‘persona’s’ reference system is likely to incur high
costs in terms of processing. Rühlemann et al. also noted the costs of
assuming a character’s ‘voice’ system. If voice is understood in a narrow
sense as referring to mimicking intonation and voice quality, the change from
the speaker’s own voice to a character’s voice will incur further processing
costs. The double switch from one’s own reference and voice system to a
persona’s reference and voice system will require some planning time; hence
the use of silent pauses. However, the present research has discovered that
silent pauses also preferably occur upon the completion of quotes – that is, at
points in the narrator’s discourse where, under most circumstances, a return
can be observed to the narrator’s own words and thus his/her own reference
and voice system (only where the quote is directly followed by another quote
is there no such return to the narrator’s own words but instead a switch into
yet another reported speaker’s reference and voice system). It seems less
152 How do narrators mark quotation?
plausible to assume that the return to one’s own system, which can be
considered the default system, is as processing-intensive as the switch into
an alien system. Thus, the speaker-centered interpretation of pauses as
planning buffers is likely not sufficient alone.27
A complementary interpretation can be arrived at if the use of pauses in the
context of quotation is approached from the perspective of the recipient.
Given the non-verbal nature of silent pauses, their occurrence within dis-
course causes a temporary absence of discourse which “interrupts the artic-
ulatory flow of speech” (Klewitz & Couper-Kuhlen 1999: 476). Given their
preferred recurrence at both ends of quotation, pauses effectively divide
discourse into pre-pause and post-pause discourse thus ‘cutting’ it into
segments: the quote as a whole is exposed from the flow of discourse as a
discourse segment with discernible boundaries. The segmentation alerts the
recipient to the possibility that the narrator is preparing or concluding a
major transition of some kind: either a transition from one syntactic unit to
another (see Biber et al. 1999), or a shift from ‘one new idea’ to another new
idea (see Chafe 1992) or, as in the case of quotation, shifts in the narrator’s
perspective: (i) from author formulating their own discourse to animator
creating a likeness of someone else’s discourse, or (ii), the reverse, from
animator to author, or (iii) from animator of one character to animator of
another character, as is the case in quotes occurring in strings (see Section
5.2). Seen in this recipient-centered perspective, pauses inevitably serve as
quotation markers. Evidence to support this interpretation is the U-shape of
the distribution: silent pauses bracket quotes, most typically occurring right
before and right after the quote, thereby strikingly resembling quotation
marks used in writing. If pauses mainly or only served to buy the narrator
planning time we should expect that pauses tend to occur quote-initially only.
Their recurrence quote-finally is much less expected given that the narrator’s
return to their own words, reference, and voice system is likely to require
much less planning and hence much less pausing. That is, the tendency for
silent pauses to bracket quotation fulfills a strategic function in the service of
recipient design: the quote, as a whole, is separated from the neighboring
discourse, marking its onset and completion and alleviating the boundary
issue at either end. Pauses are thus designed as a quote-unquote signal – a
design supporting the recipient’s processibility needs vis-à-vis the narrator’s
multiple changes in footing.
The two views complement each other, in two respects. High-frequency
items generally and non-verbal items specifically are prone to perform multi-
ple functions. Therefore it seems not unlikely that narrators use pauses both
as planning buffers for themselves and as quotation markers for the recipient.
Even if narrators did not make use of pauses strategically, that is, as a means
of reaching the goal of aiding the recipient in processing quotation (see van
Dijk & Kintsch 1983: 65), the use of pauses, as a temporary absence of
discourse, would still have the convenient side effect for the recipient to
4.3 Pauses as quotation markers 153
divide the narrator’s discourse into discourse chunks. Given that the chunks’
boundaries are far from random but coincide with discourse boundaries – viz.
the onset and completion of quotation – the chunks as such are meaningful,
thus facilitating the identification of quotation and, as a consequence, aiding
its processing.
4.3.5 Summary
In this section I have investigated the role of short silent pauses in the context
of quotes. The data considered included both pauses occurring within the
annotational boundaries of quotes as well as pauses occurring immediately
before or after the quote. Starting from observations in the literature that
silent pauses preferably occur at major syntactic and discourse boundaries it
was hypothesized that the distribution of pauses forms a U, with high rates of
occurrence in initial and final positions. Initial positions were defined as
positional values in the segment ϵ[0,20], while final positions were defined
as positions in the range ϵ[80,100]. The positional analysis demonstrated that
indeed silent pauses preferably occur at the onset and, respectively, upon
completion of the quote. It was noted in passing that no such U-shaped
distribution could be found for the filled pauses er and erm, which exhibited a
peak in initial positions but no corresponding peak in final positions. To
establish the significance of the U-shaped distribution of silent pauses, a
bootstrap was performed calculating a confidence band for the distribution.
The confidence band thus obtained turned out to be extremely narrow
leaving no room for any essentially different density estimate than the density
estimate obtained from the original sample. The alternative hypothesis was
therefore fully accepted: silent pauses frame quotes, with peaks in initial and
final segments.
Two alternative interpretations were discussed. In a speaker-centered
perspective, pauses were seen as speech management expressions serving
to buy the narrator planning time, a function which is vital given the
considerable processing costs incurred by the fact that quotation requires a
shift from the narrator’s own reference and voice system to a character’s
reference and voice system or from one character’s to another character’s
systems. It was noted that the fact that pauses frame quotes, peaking not only
at the beginning but also at the end of quotation, is less well explained by the
view of pauses as indices of the narrator’s planning constraints in that the
return to the narrator’s own reference and voice system arguably requires less
such planning. In a recipient-centered perspective, it was argued that silent
pauses, as temporary absences of discourse, have the effect of separating pre-
pause discourse from post-pause discourse. Given their separating effect,
pauses can be seen as a convenience to the listener, aiding their under-
standing that, in the case of quote-initial pauses, the narrator is about to
embark on quotation and, in the case of quote-final pauses, the quote has been
154 How do narrators mark quotation?
5.1 Introduction
In the previous chapter I referred to the centrality of discourse presentation
to conversational narrative. In the present chapter my aim is to present
evidence for the claimed centrality of discourse presentation from a pri-
marily functional perspective. Specifically, I will examine how narrators
design discourse presentation with a view to dramatization. Dramatizing
storytelling by means of discourse presentation is an important substrategy
in the service of recipient design (see Section 1.7). Its importance derives
primarily from the recurrence of the mimetic modes, Free Direct and
Direct, which, as shown in Section 2.2.2, taken together account for two-
thirds of all discourse presentation in the NC. Mimetic discourse presen-
tation can thus be considered the ‘norm’ in storytelling (see Leech & Short
1981). The use of the mimetic modes has important interactional implica-
tions. Free Direct and Direct instigate a crucial shift in participation
structure: given their mimetic potential, which facilitates maximal imme-
diacy of the reported discourse for the recipients (see Section 4.1), the
two modes have the capacity of turning recipients into witnesses of the
discourse presented: “rather than simply hearing about it” (Tannen 1988:
105, added emphasis), they are enabled to hear it (and see it if gesturing is
involved) for themselves. As a result, recipients are no longer only parti-
cipants in the telling situation, they also become participants in the told
situation.
In the case study to follow, I will demonstrate that narrators structure
sequences of discourse presentation occurring within utterances in such a
way as to achieve precisely that dramatization, due to an intrasequential
trend towards what I will call immediatization. That is, it will be shown
that sequences of report units increase in immediacy over the sequence
and that they exhibit a trend to culminate in the mimetic modes
Free Direct and Direct, thereby inviting the recipients to participate as
witnesses in the told situation and creating at utterance level what has
so far only been observed at story level, viz. climacto-telic tension (see
Section 1.6).
155
156 How do narrators use discourse presentation for dramatization?
From the narrator’s utterance in (5.1), this sequence of report unit tags can be
extracted, consisting of 4 report units:
pos1 pos2 pos3 pos4
MII MDD MII MXX
(Indirect) (Direct) (Indirect) (Unclear)
The report unit positioned as the first unit (pos1) in the sequence is an
Indirect report, followed in position 2 (pos2) by a Direct report, followed
by another Indirect report (pos3), and concluded by an Unclear report unit
(pos4).
The question I am going to address in this section is, ‘Are there any patterns
in how speakers sequence report units at utterance level?’ While the idea of
sequentiality underlies the idea of collocation, as the co-occurrence of lexical
items, a phenomenon which has attracted a wealth of corpus studies, sequen-
tiality in the use of report units, as discourse units, has, to my knowledge, not
yet been examined elsewhere systematically, let alone quantitatively. The
examination presented here is owed to the consistent annotation of report
units in the NC and the use of XQuery.
If sequential patterns in the use of report units per utterance could be
discovered, what kind of patterning would we expect to find? As noted, there
is a serious dearth of previous research into this question and, to the best of
my knowledge, the only previous study offering some guidance is Holt
(1996), who observed that indirect forms of reporting tend to precede direct
forms. In Section 4.1, I proposed to order reporting types along a cline of
immediacy of the reported discourse to the recipient, with the direct types
Free Direct and Direct facilitating the highest degrees of immediacy and the
remaining types (including the indirect types Free Indirect and Indirect)
gradually declining in immediacy. Holt’s observation of a progression from
indirect to direct reporting, then, suggests a progression from less immediate
to more immediate reporting. Starting from these (crude) premises, the
hypotheses I am going to test are the following:
H0: For sequences of report units within utterances, there is no trend from
less immediate to more immediate reporting mode types.
H1: For sequences of report units within utterances, there is a trend from
less immediate to more immediate reporting mode types.
800
20
600
15
n report units
Frequency
400
10
200
5
0
0
1 3 5 7 9 >10
n report units
Figure 5.1 Boxplot (left panel) and barplot (right panel) showing the
distribution of number of report units across 1371 utterances
utterances with n report units. The distribution of the utterances with n report
units in the NC is shown in Figure 5.1.
As can be seen from the boxplot in Figure 5.1 (in the left panel), the median
shown in the bold horizontal line indicates that the central tendency is for
utterances to have 1 report unit each (the mean, by contrast, is 1.99). Indeed,
as shown in the histogram (in the right panel), the 809 utterances with exactly
1 report unit (n = 1) account for the great majority of all utterances in the
subset. The decrease in frequency of utterances with n > 1 is rapid: while
there are 275 utterances with n = 2, and 125 utterances with n = 3, the
frequencies of utterances with 3 > n < 9 decrease to numbers in the two-
digit range; finally, from n = 9, the frequencies drop to one-digit numbers.
To return to the boxplot, there are a large number of outliers, that is,
utterances with far more report units than the majority of utterances (the
most extreme outlier is an utterance with as many as 19 report units; in the
barplot, it is included in the bar on the far right for n > 10).
Given this distribution of the data, the question of which subset to work with
can be answered fairly easily. Since the interest is in sequences of report units
within utterances, n=1-utterances are irrelevant. Further, n=2-utterances will
not be included either because trends cannot be read off from sets of two items
only. Finally, given their low frequencies, utterances with n ≥ 9 will be
discounted as well. So, the following analysis of sequences of report units in
5.2 Climactic structure at micro level 159
Immediacy
Animator Author
Mode MDF MDD MIF MII MSS MVT MVV MUU MRR MRQ
Immediacy 9 8 7 6 5 4 3 2 1 0
Value
Figure 5.2 Immediacy cline with associated ordinal values; MDD: Direct,
MDF: Free Direct, MII: Indirect, MIF: Free Indirect, MSS:
Representation of Speech Act, MVT: Voice with Topic, MVV:
Representation of Voice, MUU: Representation of Use, MRR: Reference
to Discourse Presentation, MRQ: Request for Discourse Presentation
ordinal value, namely 9, while all other reporting modes receive gradually
lower values depending on their position on the cline. MRQ, at the end of the
cline, as a mere request for discourse presentation, does not afford any
immediacy for the listener (who in virtually all cases is the requester) and
hence receives the value 0.
It goes without saying that the assignment of ordinal values to categories
on a continuum is not to be taken as implicating that the categories will differ
exactly by one degree of immediacy each. Obviously, the difference in
immediacy for the two adjacent categories MDD (Direct) and MIF (Free
Indirect), which are divorced by important characteristics (for example,
changes in the speaker’s reference system) is larger than the difference
between MDF (Free Direct) and MDD (Direct), which share important
characteristics (viz. the mimetic potential). The same could probably be said
of most other categories and their mutual differences in terms of immediacy.
However, ordinal values by definition do not measure degrees; they simply
indicate an order, namely, in the present case, the order of reporting types in
terms of immediacy. Simply put, the ordinal values state that there is more
immediacy, for example, to MDF than to MDD but they do not state how
much more there is.
The methodological steps taken to operationalize immediacy in the use of
3 ≤ n ≤ 8 report units per utterance were as follows. In a first step, using
XQuery scripts, the reporting mode values used in the NC’s annotation
of reporting mode (e.g., MDF, MSS, etc.) were retrieved for utterances
containing sequences with 3 ≤ n ≤ 8 report units.1 Second, each reporting
mode value in the sequences was replaced by its corresponding ordinal value
(see Figure 5.2). For example, in (5.2), the first three sequences obtained for
n=4-sequences were transformed into the sequences of immediacy values
shown on the right hand:
(5.2)
pos1 pos2 pos3 pos4 pos1 pos2 pos3 pos4
-------------------------- -------------------------
MII MDD MII MXX → 6 8 6 0
MSS MSS MVV MSS 2 2 3 2
MDD MDD MDD MDF 8 8 8 9
5.2.3 Results
The data are first inspected visually. Figures 5.3 and 5.4 show ECDF
(empirical cumulative distribution function) plots (see Crawley 2007: 241).
The y-axis depicts the cumulative proportions of each immediacy value as a
flight of steps. Importantly, frequency is read off as a function of the ‘leap’
from one step to the next: the higher the leap, the higher the proportion and,
hence, the frequency of the corresponding immediacy value (on the x-axis).2
5.2 Climactic structure at micro level 161
n=3 n=4 n=5
1.0
1.0
1.0
0.8
0.8
0.8
0.6
0.6
0.6
Fn(x)
Fn(x)
Fn(x)
0.4
0.4
0.4
0.2
0.2
0.2
0.0
0.0
0.0
0 2 4 6 8 0 2 4 6 8 0 2 4 6 8
Immediacy Immediacy Immediacy
1.0
1.0
0.8
0.8
0.6
0.6
0.6
Fn(x)
Fn(x)
Fn(x)
0.4
0.4
0.4
0.2
0.2
0.2
0.0
0.0
0.0
0 2 4 6 8 0 2 4 6 8 0 2 4 6 8
Immediacy Immediacy Immediacy
Figure 5.3 ECDF plots of immediacy values for first position (pos1) and
last position for 3 ≤ n ≤ 8 report unit sequences
1.0
1.0
1.0
pos1 pos1 pos1
pos2 pos2 pos2
0.8
0.8
0.8 pos3 pos3 pos3
pos4 pos4
pos5
0.6
0.6
0.6
Fn(x)
Fn(x)
Fn(x)
0.4
0.4
0.4
0.2
0.2
0.2
0.0
0.0
0.0
0 2 4 6 8 0 2 4 6 8 0 2 4 6 8
Immediacy Immediacy Immediacy
1.0
1.0
1.0
0.8
0.8
0.8
0.6
0.6
0.6
pos7 pos7
Fn(x)
Fn(x)
Fn(x)
pos8
0.4
0.4
0.4
0.2
0.2
0.2
0.0
0.0
0.0
0 2 4 6 8 0 2 4 6 8 0 2 4 6 8
Immediacy Immediacy Immediacy
Figure 5.4 ECDF plots of immediacy values for all positions for 3 ≤ n ≤ 8
report unit sequences
separated from the direct categories (values 8 and 9). The evidence in
Figure 5.3, then, indicates that, in the majority of sequence types, final
positions score higher in terms of immediacy than first positions. I take this
as initial evidence that there may indeed be a trend towards increased imme-
diacy within report-unit sequences.
To further explore this possibility visually, consider the ECDFs in Figure 5.4.
This figure gives cumulative relative frequencies, not only of the first and final
positions, but of all positions in the sequences.
The ECDF plots in Figure 5.4 paint a nuanced picture of the distribu-
tions. For n=3-sequences, we find that, in pos1, the cumulative frequencies
for values ≤ 7 are greater than those for both pos2 and pos3, which are
almost the same. That is, cumulatively, immediacy will be lower in pos1
than in pos2 and pos3 each, where it is roughly the same. For n = 3, a trend
towards immediatization is hence likely. As regards n=4-sequences, there
5.2 Climactic structure at micro level 163
can be little doubt of such a trend: in no position is the leap greater from the
non-direct categories (values 0 to 7) to the direct categories (values 8 and 9)
than for the final position. Also, the cumulative frequencies for values ≤ 7
decrease gradually over the positions: they are highest for pos1, roughly the
same for pos2 and pos3 but at a lower level than pos1, and smallest for pos4.
So immediacy seems to increase relatively steadily over all positions. The
same cannot be said of the remaining sequences. For n=5-squences, the
leap from 7 to 8 is greater for all non-final positions than for final position
(pos5), suggesting that immediacy overall will be lower in that final position.
Similarly, for n=6-sequences, the leap to value 8 is greater for pos2, pos4
and pos5 than the same leap for pos6. By contrast, for n = 7 and n = 8, it
appears that the proportion of values ≤ 7 is greater for most positions
preceding final position than for that final position, suggesting that imme-
diacy will be lower in these positions and higher in final position. However,
whether the increase in immediacy for n = 3 and n = 4 is consistently
gradual over the positions cannot be gauged from the plots. In sum, the
visual evidence in favor of an immediatization trend is straightforward only
for shorter sequences (3 ≤ n ≤ 4), while it is less clear for longer sequences
(5 ≤ n ≤ 8).
To establish with certainty whether report sequences are immediatized,
statistical tests were carried out. The appropriate test is the test for trend in
proportions (see Dalgaard 2008: 150). In this test, the observed column totals
are compared to the maximally possible column total corresponding to 100%.3
The results of the tests for trends in proportion fully confirm the alter-
native hypothesis H1, according to which there exists a trend from less
immediate to more immediate report units in utterances. The results are
listed in Table 5.2.
The p-values in Table 5.3 are all far smaller than the threshold value of
0.05. The results are very highly significant (***) for all sequences inves-
tigated but for n = 5, where the result is significant (*), and for n = 8, where
the result is highly significant (**). We can then fully accept the H1: there is
an immediatization trend in the use of report sequences within utterances;
N p-value Sig
n=3 0.0002097 ***
n=4 8.257e-05 ***
n=5 0.0253 *
n=6 0.0001015 ***
n=7 0.0005902 ***
n=8 0.004152 **
164 How do narrators use discourse presentation for dramatization?
Mode MDF MDD MIF MII MSS MVT MVV MUU MRR MRQ
Mimetic value 1 1 0 0 0 0 0 0 0 0
that is, narrators exhibit a statistically very robust tendency to order report
units from less immediate to more immediate.
The immediatization trend can be explored in greater depth. Knowing that
the mimetic mode MDD, along with its variant MDF, is the default mode
(MDD alone accounts for 58%, while the two categories MDD and MDF
together account for more than two-thirds of all report units in the NC; see
Section 2.2.2) we should expect that the trend towards greater immediacy is
correlated with a trend towards increased use of the two mimetic report
types. Indeed, this was strongly suggested by Figures 5.3 and 5.4 above,
where, in all sequences, the leap from immediacy values ≤ 7 (representing the
non-direct and non-mimetic modes) to values 8 and 9 (representing the direct
and mimetic modes) was not only the largest one across all positions but also,
and more importantly, the leap was often larger in sequence-late positions
than in sequence-early positions. We will then expect to find the trend
‘powered’ by the report unit types located at that upper end of the cline
(values 8 and 9), the direct and mimetic modes, rather than those at the lower
end (values ≤ 7), the non-direct and non-mimetic modes.
How can the contribution of the mimetic types Free Direct and Direct to the
immediatization trend be investigated? One economical and elegant method is
by converting the reporting mode tags to binary variables, indicating whether
the tags denote non-mimetic types (0) or mimetic types (1).4 Then tests for
trends in proportions need to be carried out for each n separately. The report
types and their corresponding mimetic values are shown in Table 5.3.
Following this methodological path, first the data conversion was carried
out. By way of illustration, consider (5.3), presenting the first three lines from
the same data as in (5.2) above (n = 4):
(5.3)
pos1 pos2 pos3 pos4 pos1 pos2 pos3 pos4
-------------------------- -----------------------
MII MDD MII MXX → 0 1 0 0
MSS MSS MVV MSS 0 0 0 0
MDD MDD MDD MDF 1 1 1 1
As before, the distributions are explored visually using ECDF plots. Given
that mimesis is measured using binary values, the plots exhibit a single step,
that from 0 to 1:
5.2 Climactic structure at micro level 165
n=3 n=4 n=5
1.0
1.0
1.0
0.8
0.8
0.8
0.4 0.6
0.4 0.6
0.4 0.6
Fn(x)
Fn(x)
Fn(x)
0.2
0.2
0.2
0.0
0.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Mimesis Mimesis Mimesis
1.0
1.0
0.8
0.8
0.4 0.6
0.4 0.6
pos7 pos7
Fn(x)
Fn(x)
Fn(x)
pos8
0.2
0.2
0.2
0.0
0.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Mimesis Mimesis Mimesis
The ECDF plots in Figure 5.5 describe a mixed picture. The first two
panels, for n = 3, and n = 4, suggest clear trends toward what could be
called mimeticization. For n = 3, the mimesis proportions for pos2 and pos3
are almost equal but, given the larger leap from value 0 to value 1, they are
higher than for pos1. For n = 4, the rise in proportion of the mimetic modes
is consistent over all positions: the leap from 0 to 1 is by far the smallest for
pos1, it is clearly larger for pos2, yet a little larger for pos3, and again a little
higher for pos4. The trend towards increased mimesis could not be clearer.
For 5 ≤ n ≤ 8, the existence of such a trend is less obvious. We observe a
decisive departure from the ‘rule’ that the proportions for final position are
greater than the frequencies for first position. For n=5-sequences, for
example, pos2, pos3, and pos4 score higher in terms of mimesis than
pos5. The existence of a trend towards maximized mimesis seems highly
unlikely. For 6 ≤ n ≤ 8, the existence of trends is at least doubtful given that
166 How do narrators use discourse presentation for dramatization?
n p-value Sig.
n=3 0.01333 *
n=4 0.0008915 ***
n=5 0.6143 ns
n=6 0.0007627 ***
n=7 0.1106 ns
n=8 0.3545 ns
some positions are higher in mimesis while other positions are lower than
the respective final position.
To test the distributions for significance, again tests for trends in propor-
tions were carried out, again juxtaposing each of the observed column totals
against the maximally possible total.5 The results are summarized in Table 5.4.
As suggested visually, the results are mixed. For three sequence types
(n = 3, n = 4, and n = 6), the results turn out significant, for another three
sequence types (n = 5, n = 7, and n = 8), the results are insignificant. On the
face of it, this would seem to prevent the conclusion that the trend towards
immediatization is consistently carried by the two mimetic modes. However,
it should not be forgotten that the sequences n = 3 (125 occurrences), n = 4
(64 occurrences), and n = 6 (22 occurrences), for which the results are
significant, cumulatively account for 79% of all sequences investigated
(Table 5.5). That is, for almost four-fifths of all the data under scrutiny the
trend towards greater immediacy is consistently carried by the mimetic
modes Free Direct and Direct. Therefore, the results as a whole suggest
that the increased immediacy observed in the preceding subsection is
achieved to a large extent by means of increased use of the two types
facilitating mimesis, Direct and Free Direct. For the sequences n = 3, n =
4, and n = 6, we can say with confidence that narrators sequence report units
at utterance level climactically not only in the sense that the report units
progress in terms of immediacy but also in the sense that the progression is
towards mimesis, that is, re-enactment of the reported discourse.
Finally, a quick note on those utterances which contain exactly two report
units, and which, as noted above, represent in fact the largest group of report
sequences (there are 275 n=2-sequences in the NC). Given that the hypoth-
eses were concerned with the question of there being a trend, n=2-sequences
have so far not been investigated. However, it is worth mentioning that, while
not a trend in a statistical sense, the same type of tendency towards greater
immediacy discovered for 3 ≤ n ≤ 8 report units can also be observed for
n=2-sequences. As shown in Figure 5.6, both the cumulative proportions of
immediacy and the proportions of mimesis are smaller in pos1 than in pos2 for
5.2 Climactic structure at micro level 167
1.0
pos1 pos1
pos2 pos2
0.8
0.8
0.6
0.6
Fn(x)
Fn(x)
0.4
0.4
0.2
0.2
0.0
0.0
5.2.4 Discussion
Two closely correlated findings were reported on: (i) there is a statistically very
robust trend for 3 ≤ n ≤ 8 report units at utterance level to increase in
immediacy over the sequence (the immediatization trend) and (ii) for the
large majority of the data examined, there is a trend for increased use of the
mimetic modes Direct (MDD) and Free Direct (MDF) over the sequence,
168 How do narrators use discourse presentation for dramatization?
(5.5)
n = 4: sequence: MVV, MDD, MDD, MDF
MVV
S1 PNS [ Arf was saying for erm to Nat the other
night. ] He said QSD[MDDoh look I got a baby
inside me Nat. ] She said QSD [MDDno you can’t
have babies. Ladies can. Do you know how they
get there? ] [MDF Yeah but I ain’t telling you. ]
(KC5-N1)
In (5.4), the three report units reveal a sequential ordering from Voice with
Topic (MVT), which sees the narrator firmly in the role of author, to Direct
(MDD), where, given that the report itself is produced in the role of animator,
the preceding quotative is the sole ‘relict’ of the narrator’s previously held
authorial role, to Free Direct (MDF), where the narrator fills the role of
animator of a character’s words completely. The three reports are thus
correlated with a steady increase in immediacy: from less immediate (MVT)
to highly immediate (MDD) to very highly immediate (MDF), and with a
progression from non-mimetic (MVT) to mimetic (MDD and MDF).
Similarly, in (5.5), the progression is from less immediate/non-mimetic
(MVV) to highly immediate/mimetic (MDD) to very highly immediate/
mimetic (MDF).
Why do narrators immediatize report sequences? Immediacy, it was argued
earlier, is a listener-centered category: minimum immediacy in reporting keeps
the listener at a distance from the discourse reported, maximum immediacy
offers them full access to the discourse presented. It was shown also that the
two mimetic modes MDD and MDF, which afford the greatest immediacy, are
by far the most frequent types in the NC; mimetic reporting can hence be
considered the default type of reporting, the ‘norm’ (see Leech & Short 1981).
Why, then, are sequence-initial reports less immediate than sequence-final
reports? The answer, it appears, has to do with a functional division of labor:
sequence-initial reports have a scene-setting function, while sequence-final
reports have a scene-performing function.
These functions are best understood re-considering examples (5.4) and
(5.5). The first reports – MVT in (5.10) and MVV in (5.5) – provide the
recipient with background information specifying who spoke to whom under
what circumstances: in (5.4), the recipient learns that the police came to
interview the presenter and, in (5.5), that Arf recently talked with Nat. These
reports thus draw up the coordinates of the situation in which the utterances
occurred introducing the speakers and sketching the spatiotemporal setting.
The initial reports, then, serve a recipient-designed function of grounding, or
contextualizing, the following reports; metaphorically: they serve to set the
scene (see Holt 1996: 243).
This is a crucial function: without the reports, that is, without the scene
they set, the subsequent reports could hardly be ‘put on’ as mimetic reports,
170 How do narrators use discourse presentation for dramatization?
for the recipient would be unable to assign the quotes to the speakers whose
words are quoted. Given the scene-setting function, sequence-initial reports
serve, within their sequence, a similar function as Labov’s Orientation
section does within the story as a whole: to provide background information
the listener can use to contextualize what follows.
The function of the following sequence-medial MDD reports is inter-
mediate in the sense that it involves both scene-setting and scene-performing
elements: the narrator indicates which of the characters is going to speak by
use of a reporting clause (he said/she said); in so doing the narrator acts
as author, mediating between the recipient and the protagonists. Then,
animated by the narrator, the protagonists begin to perform their ‘text’ as
if they were present themselves.
The scene-performing function is carried out in its pure form by the
sequence-final MDF reports. The narrator, as author, has disappeared from
the scene leaving it entirely to the protagonists who seemingly speak under real
life conditions, taking the turn independently and thereby gaining complete
independence.
In other words, we see a three-stage progression from scene-setting, in
which the narrator acts as author, to an intermediate stage where authorial
intervention and animation intertwine, to scene-performing, where the char-
acters are animated purportedly without any authorial addition. Over the
three stages, immediacy is continuously increased: the more the author
pretends to withdraw the more the distance between the reported discourse
(which was produced in an anterior situation) and the recipients of the report
(which is produced in the present reporting situation) is minimized, or,
positively, the more the closeness of the discourses reported to the recipient
is maximized. Thus, report sequences reveal a climacto-telic structure:
starting from introductory scene-setting reports, the sequences reach their
highpoint, or climax, in maximally immediate reports in which the distance
between the recipients and the discourse reported is removed. The sequence
climax turns the recipients into witnesses: recipients gain direct access to the
discourse. In Yule’s (1996) words: the mimetic reports forming the highpoint
5.2 Climactic structure at micro level 171
5.2.5 Summary
This section has concerned itself with co-occurrence patterns of discourse
units, viz. sequences of discourse presentation units in utterances. It was
shown that the co-occurrence of report units within utterances is far from
random. Rather, it is systematic in the sense that a trend can be observed for
narrators to order reports in sequences such that the immediacy of the
reports is increased intrasequentially. This trend, referred to as immediati-
zation trend, was found to be statistically robust for all report sequences
investigated (3 ≤ n ≤ 8). The same significant tendency was observed for n=2-
sequences, by far the largest group of report sequences in the NC. Moreover,
it was demonstrated that the trend towards greater immediacy is driven to a
large extent by the two most immediate report types, Free Direct and Direct,
which are, at the same time, the only mimetic types, facilitating the dramatic
re-enactment of reported discourse. The trend towards increased use of
the two mimetic modes over the sequences was found significant for three
sequence types: n = 3, n = 4, and n = 6, which taken together account for the
overwhelming majority of all sequences investigated (viz. 79%). No trend
towards increased mimesis was observed for n=5-, n=7-, and n=8-sequences.
It was concluded that the consistent trend towards greater immediacy is to a
very large extent due to the intrasequential increase of occurrence of the
mimetic modes Free Direct and Direct. It was argued that, in parallel with
similar trends observed in research based on the story as the unit of obser-
vation, the trend towards immediatization, particularly where it culminates in
the use of the mimetic modes, reveals a tendency for narrators to climactically
structure report sequences at utterance level.
This finding suggests that Longacre’s (1983) ‘tension’, which characterizes
discourse types like narrative and which is underlying the climacto-telic
structure of storytelling (see Section 1.6) is also at work below the level of
discourse type, viz. at the level of the utterance. Tension appears to be the
design principle governing, not only storytelling as such, but also storytelling
utterances containing report sequences. The microcosm of the utterance
mirrors the macrocosm of the story: both are subject to tension. Going one
step further, we might hypothesize that the tension being built up within
utterances containing report sequences contributes to the build-up of the
tension which structures the story. The ‘waves’ of tension traveling from
utterance to utterance within a single story might not only reflect, on a small
scale, the large ‘wave’ of tension of the story, but actually co-generate it.
Utterance tension achieved by means of immediatization could be one factor,
among others, which leads to story tension. To substantiate this hypothesis,
much further research is needed. Future investigations might also examine
172 How do narrators use discourse presentation for dramatization?
6.1 Introduction
In this chapter, the recipients’ contribution to the co-construction of narra-
tive moves center stage. The analyses will benefit from the annotation
of participant roles. As noted repeatedly, underlying the annotation of
recipient utterances is the dichotomous criterion of telling/teller- vs. tale-
relation (see Section 1.3): recipient contributions were defined as telling/
teller-related, and assigned to the role of Responsive Recipient (PRR), when
they did not add or elicit new information on story background and events
but primarily served to signal, to the teller, active reception of the telling.
Given the focus on the teller’s telling, as feedback registering receipt and
comprehension of the teller’s discourse, the primary function of this type of
response is interactional, that is, vis-à-vis the teller. Conversely, responses
were defined as tale-related, and assigned to the role of Co-constructive
Recipient (PRC), when, in addition to registering reception, they added
or elicited new information on story background and events. Tale-related
contributions in this sense do two things. Like telling/teller-related feedback
(PRR), they testify to the recipient’s active listenership; as such, their
function is interactional. Unlike PRR, they interact with details of the
story; as such, their function is topical, vis-à-vis the tale. As reported in the
Introduction to this book, it is widely agreed in recent research that recipients
can actively reshape “the course of an emerging story” (Goodwin & Heritage
1990: 300). This claim amounts to ascribing recipients co-authorship: by
reshaping a story’s course, they become its co-authors. In the present
chapter, the overarching goal is to substantiate this claim by providing
quantitative evidence to support it.
Besides the clear difference in function vis-à-vis narrator discourse, the
two recipient roles are clearly distinguishable in lexical, structural, and
pragmatic terms (see also Section 2.2.2). Consider the randomly selected
examples of PRR and PRC utterances in Table 6.1.
The examples listed in Table 6.1 give a good impression of what character-
izes and distinguishes the two response types. It can immediately be seen
that, as already noted (see Sections 2.2.2 and 2.3.2), PRR responses are by far
173
174 How do recipients co-author stories?
PRR PRC
oh oh god i would ’ve crawled into a hole for the rest of my life
mm well maybe it ’s just been a deep a deep-rooted infection
aha but that ’s so pointless cos it ’s got all the addresses your addresses in it
yeah yeah it ’s mulled wine hot wine
mm what ’s one of them an e c
yeah was it a big coach
[laugh] where did he get them from
yes who did
what pauline that wo n’t come back
Gawd oh that ’s charming
shorter than PRC responses. None of the ten PRR utterances in the table is
longer than 1 word, which is a good reflection of the overall brevity of PRR
utterances: the mean length of PRR contributions is 1.33 words (for a very
similar observation, see Goodwin 2007: 26). By contrast, the shortest PRC
utterance in the table is 2 words in length, while all others have lengths
between 4 and 16 words. Again, this is a good reflection of the mean PRC
utterance length, which is 6.82 words. (As noted, the difference in length is
very highly significant; see Section 2.3.2.)
Another major difference relates to word class. The items in the left column
(for PRR) are mostly interjections (oh, mm, aha, yeah, yes); one response is
nonverbal ([laugh]), one consists of a question pronoun (what), and the item
Gawd can be seen either as a noun (a variant of ‘God’) or a (secondary)
interjection (see Section 4.2.1). The items in the right column (for PRC), by
contrast, display words from almost any word class. Again, the differences
observable for the random examples in the table reflect the differences in the
samples as a whole. As can be seen from Figure 6.1, the dominance in PRR of
items tagged as interjections (ITJ) is overwhelming, whereas the lexical
diversity in PRC utterances is much greater. Most notably, PRC utterances
score higher on all word classes commonly counted among the ‘content,’ or
lexical, words, including (going from left to right in Figure 6.1): (i) adjectives
(AJO), (ii) adverbs (AVO), (iii) nouns, in their incarnations as singular nouns
(NN1) and proper nouns (NPO), and (iv) lexical (past tense) verbs (VDD).
The distinguishing feature of content words is that they “carry most of the
lexical content, in the sense of being able to make reference outside language”
(Stubbs 2002: 40; see also Biber et al. 1999: 55ff.), whereas function words
(such as auxiliary verbs, pronouns, prepositions, conjunctions, etc.) serve as
mortar to keep lexical words together (see Stubbs 2002: 40) and inserts (such
as interjections) predominantly “carry emotional and interactional meanings”
(Biber et al. 1999: 56).1 The much greater proportions of content words
suggests that tale-related PRC responses are semantically much heavier than
6.1 Introduction 175
PRR
PRC
60
50
40
Proportions
30
20
10
0
AJ0 AV0 DT0 ITJ NN1 NP0 PNP VBZ VDD VVB
Pos tags
Figure 6.1 Proportions of top 10 most frequent word classes in PRR and
PRC utterances
telling/tale-related PRR responses, which are semantically light, but, given the
extreme dominance of interjections, ‘weighty’ in interactional terms.
The third clear difference between PRR and PRC utterances relates to
utterance structure. PRR utterances are, as noted, overwhelmingly interjec-
tions, the most important of the word classes grouped under the heading of
inserts (see Section 4.2.1). The distinguishing feature of inserts is that they
“do not form an integral part of syntactic structure” (Biber et al. 1999: 56);
they are non-clausal in structure. By contrast, PRC contributions are most
typically clausal; non-clausal PRC utterances are highly infrequent. Further,
while most PRC utterances consist of a single clause, they can occasionally
consist of two or even more clauses.
The differences in lexical and structural terms give rise to differences in
pragmatic terms, that is, in what the two recipient types can ‘do’ in issuing
the utterance. While recipients producing PRR utterances are confined to
registering understanding of incoming talk by the narrator and/or displaying
176 How do recipients co-author stories?
800
600
Number of words
400
200
0
Given that recipients fall into the two types Responsive Recipient (PRR)
and Co-constructive Recipient (PRC), a division of the data into three
independent subsets (Subsets 1–3) is warranted: stories in which narrators
receive feedback (i) from PRR only, (ii) from PRC only, and (iii) from PRR
and PRC. In Subsets 1 and 2, we can measure the potential impact on narrator
verbosity (the dependent variable labeled “w_narrator”) of the number n of
PRR responses alone (an independent variable labeled “n_prr”) and, respec-
tively, PRC responses alone (labeled “n_prc”). Subset 3 allows the study of
two potential effects: an effect due to the co-occurrence of the numbers n of
PRR and PRC responses (labeled “n_prr + n_prc”) and an effect resulting
from the interaction of PRR and PRC.
XQuery expressions were used to pull data for the three subsets.3 Subset 1
(stories with PRR responses only) contains data from 22 stories, Subset 2
(stories with PRC responses only) contains data from 8 stories, while Subset 3
(stories in which PRR and PRC responses co-occur) is the largest with data
from 51 stories.
6.2.3 Results
In each meansplot in Figure 6.3, the means of the variable w_narrator (which
contains different values in each subset) are plotted as a function of the
variable on the x-axis (n_prr, n_prc, and, respectively, n_prr+n_prc). In
all three plots, an upward trend is visible; in the case of PRC only (upper right
panel), the trend appears weaker than the trend for PRR only (upper left
panel). The trend for PRR and PRC co-occurring in stories (lower panel), by
contrast, is clearly upward (note that, due to the wider confidence intervals
shown in the vertical lines, the scale on the y-axis is much larger than for the
upper panels). The visual inspection then suggests that the number of
narrator words per story (w_narrator) and number of recipient responses
(n_prr, n_prc, and n_prr+n_prc, respectively) may be positively correlated
for all three subsets. The impression that the values of the two variables in
each subset increase in unison is confirmed by Kendall’s rank correlation
tests. The test statistics are given in Table 6.2.
According to the tests, there are positive correlations between all three
variable sets; their strengths, as expressed in the τ obtained, range from
modest (τ = 0.25 for PRC only) to medium (τ = 0.44 for PRR + PRC),
their significance levels range from high (for PRC only) to very high (for PRR
only and PRR + PRC).
While this may be reassuring information suggesting we are on the right
track, it is just initial information – certainly not sufficient to adopt the H1
stated above, according to which PRR activity (n_prr) leads to increased
narrator verbosity (w_narrator). This hypothesis states a cause-effect rela-
tionship, defining a dependent (or response) variable (w_narrator) which is
to be explained on the basis of an independent (or explanatory) variable
6.2 How do recipients influence narrator verbosity? 181
600
Number of words in narrator utterances
500
400
400
300
300
200
200
100
100
0
0
−100
−100
1 2 3 4 5 6 7 9 12 22 1 2 3 4 5 6 8
n_prr n_prc
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 23 25 26 27 28 32 36 38 51
n_prr+n_prc
Table 6.3 4 Negative Binomial models for the association of narrator verbosity and
recipient activity
Model 4
400
300
r)
E(w_narrato
200
12
10
8
6
100
4
n_prc
2
00
0 2 4 6 8 10 12
n_prr
and n_prc the plane hangs relatively loosely (suggesting high exponential
growth in w_narrator) it is being pulled more and more tightly with increas-
ing n_prr and n_prc indicating that the exponential growth in w_narrator is
losing in force (but not stopped) due to the interaction of n_prr and n_prc.5
6.2.4 Discussion
What do the findings reveal about recipient co-authorship?
First, it has been discovered that PRR and, respectively, PRC contribu-
tions have the same type of effect on narrator verbosity: increases in feedback
of either type of response over the course of a story lead to increases in
narrator verbosity. This observation holds for three types of stories: stories
with PRR only, PRC only, and texts in which both PRR and PRC occur. That
is to say that feedback as such (irrespective of the type of feedback) alters the
course storytelling is taking: narrator verbosity increases exponentially the
more often listener feedback is given. Recipients thus become co-authors of
stories: because of the recipients’ feedback (of whatever kind), stories are told
in more detail by the narrator.
6.2 How do recipients influence narrator verbosity? 185
However, it has also been observed that the effect of PRR responses on
narrator verbosity is greater than the effect of PRC responses. A plausible
explanation for this finding can be derived from consideration of the differ-
ential nature of the two response types. Responsive Recipient (PRR) and
Co-constructive Recipient (PRC) responses are not only distinguished func-
tionally (with PRR focussed on the telling/teller and PRC focussed on the
tale) but also lexically, structurally, and pragmatically (see Section 6.2.1), with
PRC being decisively more diverse and ‘heavier’ on all three counts. Now,
the heavier semantic and structural weight of PRC contributions has the
effect of alleviating the semantic and structural load narrators have to carry
in that topical aspects pertinent to the story are brought in by the
Co-constructive Recipient (PRC) instead of the narrator themselves. These
topical aspects are encoded in lexical material and structural templates the
narrator can incorporate into their telling. Re-using the recipient’s topical
input thus adds to the narrator’s economy of telling (e.g., van Dijk 1975,
Sacks 1992; see Section 1.7): the correlate of this economy is reduced utter-
ance length. For illustration, compare (6.1), an embedded generalized-
experience story in which the narrator describes her neighbors’ indifferent
behavior towards recurrent domestic quarrels and in which the recipient
remains in the role of PRR throughout, and (6.2), where a telephone con-
versation is reported with decisive intervention by PRC:
(6.1) “Irresponsible neighbours” (Type: T3G/Embed Level: EN)
CNN
CNI
S1 PNP And and really I mean you can’t I’m I’m s I’m saying
that [MSSyou, you can’t interfere ] but by god I’d long
for someone [MSSto come in and interfere] . I really
would have done. And er er I I could never
understand why people [MVVthat obviously could
hear what went on ] and [MSSpretended they never] . I
know you don’t wanna get involved
S2 PRR Mm.
S1 PNP and this that and the other [MVV but people must have
heard what went on ] and and you know the next
morning they’d see a black eye [MVVand just carry out
a normal conversation ] as though
S2 PXX [unclear]
S1 PNP and you know they hadn’t heard it all and
S2 PRR Yeah.
S1 PNP [MDFooh have you had a fall or something? ] Then you
know I mean they must have [unclear] what was going
on oh aye they just completely ignore it.
S2 PRR Mm.
186 How do recipients co-author stories?
CNF
S1 PNP I’d be walking round if I could get around.
S2 PRR Mm.
In (6.1), the recipient remains completely ‘in the back channel,’ displaying
comprehension and an understanding that extended talk by the main speaker
is underway by use of continuers such as mm and yeah but not interfering
with the storytelling topically. The lack of interference is correlated with a
lack in “semantic and syntactic complexity” (Goodwin 2007: 26) and, accord-
ingly, the ‘job’ of describing the neighbors’ non-involvement is completed by
the narrator alone. In terms of authorship: while the recipient gains only
indirect authorship rights by encouraging the narrator to tell more, the
narrator is the sole direct author of the story. In (6.2), by contrast, a kind of
‘semantic and structural job sharing’ and a sharing of direct authorship rights
can be observed:
(6.2) “Nice little voice” (Type: T10/Embed Level: ES)
CNN
CNI
1 S1 PNP [MXXShe was very nice on the er, phone, ] I
have to admit that’s what I er […] she’s
2 S2 PRC On the phone, or on the bed, on the floor!
[laugh]
3 S1 PX0 [laughing] On the floor exactly ! Yes! Exactly!
Erm
4 S2 PX0 On the table.
5 S1 PX0 So that’s the people.
6 S2 PRC Who answered the phone?
7 S1 PNP She did.
8 S2 PRC How did she answer it?
9 S1 PNP [MDFEr er er hello! ]
10 S2 PRR [laugh] Oh well [unclear].
11 S1 PNP [MUUShe said it in her nice little voice, ] and I
said QSD [MDDhello, is Donna there please? ]
And she saidQSD , [MDD (mimicking)
speaking ! ]
12 S2 PRC [MDF (mimicking) Ee ee ee ! ]
13 S1 PNP That’s right. [MDF (mimicking) Speaking ! ]
(KSV-N1)
In (6.2), the two questions by PRC in utterances 6 and 8 (who answered the
phone and how she answered the phone) effectively help develop the story.
Up until the point where the questions are inserted in the discourse the
would-be narrator has only provided, in utterance 1, a typical abstract
summarizing the plot (She was very nice on the phone). The question-answer
6.2 How do recipients influence narrator verbosity? 187
6.2.5 Summary
In sum, recipient activity impacts significantly on narrator verbosity on all
counts. It does so, however, in distinct ways. Both telling/teller-related (PRR)
responses and tale-related (PRC) contributions have the effect of fueling
narrator verbosity. This holds true for stories in which the teller receives
PRR and, respectively, PRC responses alone and it holds for the individual
influences of PRR and PRC co-occurring in stories. The alternative hypoth-
esis, according to which narrator activity was seen as depending on recipient
activity, can therefore be accepted: recipient feedback of either type leads to
188 How do recipients co-author stories?
Form
Question
Answer
Comment
Utterance completion
Extension
Discourse presentation
Comments
While answers are relatively infrequent, comments are a very common
response type. Comments may serve a number of functions; for example,
to ironicize content, display resolution of misunderstanding, and, maybe
most importantly, evaluate story details (see Goodwin’s (1986a) category of
‘assessments’). Excerpt (6.10) exemplifies an evaluative comment: the
recipient conveys his/her strong disapproval of the haste with which a
cancer patient was sent home after she had undergone an operation for
breast cancer:
(6.10) “Breast cancer” (Type: T30/Embed Level: EC2)
(. . .)
S3 PNP It’s like Peggy across the road
S2 PNP8 four days
S3 PNP she had, she’s had to have her breast off you know,
she had cancer
S2 PRR Has she?
S3 PNP and she had it er Tuesday
S2 PRR Mm
S3 PNP and I went across to see how she was and he says QSZ
[MDDI’m expecting her home today, ] that were
Thursday, so that was
S2 PRC It’s not long enough you know
S3 PNP that was two days
S2 PRC disgusting
(KB2-N1)
The excerpt in (6.11) illustrates two types of comment. The first comment is a
word play (crème de la choir) in response to the narrator’s mention of the small
size of his/her choir. Whether the word play is intended simply to entertain
the participants or to provide an ironic perspective on the narrator’s account
cannot be determined easily. The second comment is a remark intended to
portray the choir ‘master’ in a negative light (he hasn’t got any friends). Note
also how this evaluation is instantly taken up by the narrator, who replaces
friends by acquaintances:
(6.11) “Choir” (Type: T10/Embed Level: EC1)
CNN
CNI
S1 PNP Actually I tell you one good thing about being in this
choir yeah, I’m in this little choir tonight, [unclear]
choir yeah,
S2 PX0 So’s casual sex.
S1 PNP which is only about
S2 PRR [laugh]
6.3 How do Co-constructive Recipients co-tell stories? 193
S1 PNP which is only, which is only about like eight or nine
people yeah?
S2 PRC Oh so you’re the crème de la choir.
S1 PNP [unclear] and we’ve got to, we’ve got to sing for the
master
S2 PXX Crème de la [unclear]
S1 PNP and his, and his friends and that yeah and erm so
we’re all, we don’t have to go to supper and we’re
getting, we’re all getting
S2 PRC His friends?
S1 PNP pizza from Pizza Hut.
S2 PRC His friends, he hasn’t got any friends.
S1 PNP Well okay then, acquaintances. [unclear]
(KP0-N1)
In the following excerpt, the recipient’s comment evaluates not a non-present
character’s behavior but that of the present narrator, who has the habit of
looking around in other people’s houses:
(6.12) “Richard’s house” (Type: T10/Embed Level: EC1)
(. . .)
S4 PNP he is just a right prick, right? I said QSD to Russell
[MDDI’ll go upstairs and have a look at the bedrooms]
and I’ve got this thing that if it’s a nice house I’d go
and have a look at the bedrooms, he said QSD
[MDDwell why? ] I said QSD [unclear] So I said QSD to
this Richard, I said QSD
[MDDexcuse me ] I said QSD [MDDbut your phone’s
not working downstairs ] I said QSD [MDDhave you
got one upstairs? ]
S2 PRC You nosy cow!
S4 PNP Yeah!
(KCA-N1)
Utterance completions
Recipients often complete the narrator’s as yet incomplete utterance. This
phenomenon has been referred to under various labels, among them ‘sen-
tence-in-progress’ (Lerner 1991), ‘co-construction’ (Helasvuo 2004), and
‘utterance completion’ (Sacks 1992). In conversation, utterance completions
have been claimed to be “an extremely frequent and routinely doable thing”
(Sacks 1992: 651), while interactionally they are significant in that they are
direct evidence of ‘good listenership’ (Gardner 1998, see also Sacks 1992: 654)
testifying to the online processing of utterances in their course, and they
demonstrate the participants’ willingness to share turn and syntactic resour-
ces (see Ono & Thompson 1996, Goodwin 1979). Most of the utterance
194 How do recipients co-author stories?
completions in the NC are of what Ono & Thompson (1996) refer to as the
‘completion’ type; that is, the recipient completes a syntactic unit that the
narrator has left incomplete. Utterance completions range in function from
‘helping out’ the narrator with some wording that he/she has difficulty
coming up with to displaying intense involvement. In (6.13), the
co-occurrence of the hesitation markers er and erm may be evidence that
the narrator is temporarily unable to produce the appropriate word, which is
finally provided by the recipient:
(6.13) “Voucher for taping” (Type: T10/Embed Level: EC1)
(. . .)
S1 PNP he’ll give me twenty five pounds worth of er, erm
Marks and Spencer
S3 PRC Voucher
S2 PRR Marks
S1 PNP Yeah
(KBX-N2)
In (6.14), by contrast, there is less evidence of hesitation and processing
difficulties on the part of the narrator; only the choice of the vague expression
sort of may hint at a certain delay in delivery before the recipient offers the
verb baked spoken with a question intonation, somewhat uncertain as to
whether this is the right word in the context:
(6.14) “Conger” (Type: T10/Embed Level: EC3)
(. . .)
S1 PNP And then, the next day I went along and he actually
caught the five foot one and ki and killed it and he
gave us a load of steaks off it that I sort of
S2 PRC Baked?
S1 PNP Yeah. Yeah.
(KBD-N2)
In (6.15), by contrast, there are no traces of hesitation on the part of the
narrator. Rather, it is the dramatic events of a violent domestic which prompt
the recipient to co-construct the narrator’s utterance it was really distressing
me cos by adding You knew what was possibly going on. In the excerpt, then, the
utterance completion signals high involvement:
(6.15) “The couple next door” (Type: T10/Embed Level: ES)
(. . .)
S1 PNP But it was [MVV I mean I could hear it over the telly. ]
I know I don’t have the telly on full blast but [MVV I
could hear it]
S2 PRR Mm.
6.3 How do Co-constructive Recipients co-tell stories? 195
S1 PNP [MVVabove the telly. ] When I first heard the first
noise I thought QTD [MDDwhat the bloody hell’s
that] and I turned the sound down.
S2 PRR Yeah.
S1 PNP Thought QTD [MDDbloody hell, put it back up ] but
you could, it was really distressing me cos
S2 PRC You knewQOO [MIIwhat was possibly going on.]
S1 PNP Yeah.
(KB7-N2)
Utterance completion can at times even be mutual, initiated by the recipient
and continued by the narrator, thus presenting a case of intensified two-way
completion. Consider (6.16), an excerpt from the same text as (6.15):
(6.16) “The couple next door” (Type: T10/Embed Level: ES)
(. . .)
S1 PNP A woman and that don’t s I mean [MSSif you’re
having just an argument it’s raised voices ] [MVVbut a
woman doesn’t scream and cry like that ] unless
S2 PRC Unless there’s something really
S1 PNP something awful going on.
S2 PRR Mm.
(KB7-N2)
Extensions
Recipients not uncommonly co-author stories by extending them. Extensions
are made by adding story-relevant details. The details recipients can contribute
are limited: given their underprivileged status vis-à-vis the story events – that
is, the sequence of events forming the narrative backbone – recipients most
commonly contribute details based on a general epistemic basis: their world
knowledge. A straightforward extension is illustrated in (6.17), where the
recipient throws in the appropriate technical term for the kind of wine the
narrator somewhat clumsily described as a sort of wine that was warm:
(6.17) “Tobogganing I” (Type: T10/Embed Level: EC1)
(. . .)
CNN
CNI
S1 PNP I remember once erm when basically all erm there
was a gang of us had a sort of wine that was warm.
Have you ever had it, like hot wine?
S2 PRC Yeah, [MUU gluhwein they call that]
(KPA-N2)
Extensions can also be of a more speculative and tentative nature. In (6.18),
the narrator, who works as a shop assistant, is relating how he/she tried to
196 How do recipients co-author stories?
make a smoking customer aware of the non-smoking policy in the shop and
received a hateful look by the smoker. The recipient’s suggestion that the
smoker [p]robably just hates everybody! goes well beyond his/her epistemic
reach: neither has he/she seen the look nor does he/she have much evidence
on which to base the assessment. The role of co-author is thus interpreted
rather liberally: co-authorship is claimed despite seriously restricted access to
story information:
(6.18) “God-sent smoker” (Type: T10/Embed Level: EC2)
(. . .)
S2 PNP I can’t do it. You know like th if you look at someone
and they always say your eyes are the souls o yo , like
your soul aren’t they? You can always read people’s
eyes, especially with your kind of facial expressions,
cos the eyes always tell the truth. And he looked u
like he really hated me!
S1 PRC Probably did!
S2 PNP What, cos I told QTD him [MIIhe couldn’t smoke]?
S1 PRC No! Probably just hates everybody!
(KC6-N1)
Discourse presentation
The final type is discourse presentation. Discourse presentation by recipients
will be explored in greater depth below. It is therefore only cursorily dis-
cussed in this introductory section. Excerpt (6.19) exemplifies a request for
discourse presentation (MRQ) (which is of course also a question), maybe the
‘natural’ choice of presentation type for recipients while (6.20) presents an
instance of Use (MUU):
(6.19) “Dream about Mrs X” (Type: T1D/Embed Level: EC1)
(. . .)
CNN
CNI
S2 PNP It was a dream that Mrs [name] was there again.
S1 PRR [laughing] Mrs [name] !
S2 PNP Mm.
S1 PRC [laughing] [MRQ Well what did Mrs [name] say ? ]
(KBW-N2)
25
20
29
15
Frequencies
22
20
10
15 14
12
5
4
0
note that laughter occurs significantly more frequently in stories with recip-
ient dialog than in stories without recipient dialog.16
Given, then, that recipients using dialog slip into the narrator’s role
thereby indexing high involvement, it is intriguing to investigate, not only
the dialog’s local sequential context, as in the examples above, but also the
larger sequential context; that is, it is intriguing to ask where in the storytell-
ing’s structural progression this point of heightened ‘connection’ with the
storytelling may fit. Intuitively, it seems plausible to assume that the con-
nection will not typically emerge at an early stage in the telling, where the
scene is being set for the story events, or at a stage mid-way, where the
account of the events is gaining momentum, but rather later in the process,
where the telling arrives either at its highpoint (the climax), or at one of
several highpoints. (Recall from Section 1.6 that, for conversational narrative,
a distinction can be made between single-highpoint and multiple-highpoint
stories.) In sequential terms, we will expect to find recipient quotes in co-
occurrence with the highpoint.
How can the highpoint be identified? Given that the highpoint is not a
surface category but a notional category and that there is no annotation for it
in the NC (or any other corpus), the most obvious way is by inspecting stories
and trying to pinpoint the telling of the most reportable event. This method is
qualitative and intra-textual, involving interpretation of individual texts.
Another way is by investigating surface or annotated features that are known
to co-occur with the highpoint. To the extent that these co-occurrence
features can be retrieved automatically from the corpus, this method is
quantitative and inter-textual, involving examination of frequencies pulled
from all relevant texts together.
In the following analysis I will use a combination of both methods. While
the qualitative method has a wide currency, the implementation of the
quantitative method in the present connection requires some elaboration:
which co-occurrence feature can be used to locate the climax? A number of
observers have reported an association of story highpoint and quotation by
the narrator. Most prominently, Labov (1972: 372–373) described quotation
as an internal evaluation device used by narrators to highlight the highpoint
of a story. Subsequently, this initial observation has been confirmed in
numerous studies. Norrick, for example, notes that “[t]he climax of a con-
versational story is often realized in dialog” (Norrick forthcoming; for similar
remarks, see Longacre 1983: 31). Likewise, Li (1986) cites a number of case
studies in aboriginal languages in Latin America which reported that “direct
quote, but not indirect quote, typically occurs at the peak of narrative” (Li
1986: 40). Macaulay (1987) observes an association of quotation with the
highpoint of narratives in Scottish dialect. Mayes (1990: 350) rhetorically
asks, “[w]hat better way to show what the point of the story is than to
dramatise it and act it out using direct quotation?”. Finally, Clift & Holt
(2007: 2) view story climaxes as “one of the recurrent interactional sites for
6.3 How do Co-constructive Recipients co-tell stories? 203
Climax
Narrator Recipient
dialog dialog
reported speech.” So, the co-occurrence of climax and dialog (by narrators)
is well established. If we, then, assume that recipient dialog is associated
with the climax and that the climax is associated with narrator dialog, we will
expect to find a three-way co-occurrence pattern: recipient dialog co-occur-
ring with narrator dialog, co-occurring with the climax, co-occurring with
recipient dialog. I will refer to this pattern as the quote-climax-quote (QCQ)
triangle.
The QCQ triangle though is a sequential pattern realized within texts; it
can thus only be detected by means of qualitative intra-textual analysis. So of
what help is it for the quantitative analysis? Given that climaxes are not
annotated in the NC, the triangle as a whole cannot be scrutinized using
automatic retrieval and quantification. What can be scrutinized automatically
and quantitatively is the positional coincidence of two elements of the triangle,
viz. recipient dialog and narrator dialog (for which annotation exists). That is,
we can investigate whether the distribution across stories (i.e., the inter-
textual distribution) of quotes by narrators is similar to the distribution of
quotes by recipients. Indeed, the distributions will have to be the same if it is
true that both types of quotes attach to the same sequential element, viz. the
climax. More specifically, building on Mayes’s (1990) report that “many
stories contain the largest amount of direct quotation at the main point”
(Mayes 1990: 349; added emphasis) we will expect that recipient dialog too
has its ‘largest amount’ (i.e., its highest frequency of occurrence) at the main
point. That is, as regards positioning across stories, we can predict that the
peaks of both types of dialog will coincide in the same positional segment.
These premises suggest, for the quantitative analysis, the following stat-
istical hypotheses:
H0: The distributions of positions of quotes by Co-constructive Recipients
and those of narrators are the same.
H1: The distributions of positions of quotes by Co-constructive Recipients
and those of narrators are not the same.
Given that the quantitative analysis can take only two elements of the QCQ
triangle into focus (narrator dialog and recipient dialog) but not the third (the
climax) and not the interplay between all three, the results of the quantitative
204 How do recipients co-author stories?
analysis will by necessity only be partial and even preliminary. To get the full
picture, a qualitative analysis must complement the quantitative one. The
results obtained from the quantitative method are reported on in the follow-
ing section, while the results obtained from the qualitative method will be
reported on in the Discussion section.
6.3.3 Results
To test the hypotheses, positional analyses were carried out using XQuery
scripts17 for (i) quotes (that is, Free Direct and Direct report units) occurring
in utterances by Co-constructive Recipients (PRC) and (ii) quotes in utter-
ances by any narrator type (PN*) including Primary Narrator (PNP),
Unsupported Narrator (PNU), Supported Narrator (PNS), and Ratified
Co-narrator (PNC). Quotes occurring in utterances whose participant status
was uncertain (PXX) or unrelated to storytelling (PX0) were excluded. As
noted, there are 29 genuine PRC quotes; the number of PN* quotes is
dramatically higher, viz. 1,711 (the total number of quotes in the NC is
1,784; see also Section 2.2.2).
As noted earlier (see Chapter 4), in research into lexis-text associations,
position is typically defined as the proportion of words preceding the target
item divided by the total number of words in the text. The underlying
formula is this (with ‘TI’ meaning target item):
NWords preceding TI
NWords in text
Applying this procedure to quotes is not without problems. This is because
quotes are discourse units of unpredictable lexical realization and they are
normally much longer than a single word; in fact, the mean length of quotes
in the NC is 7.52 words and the standard deviation (a “measure of how closely
the data values cluster around the mean” (Woods et al. 1986: 41)) is 5.9. As
regards positions of quotes within stories, quote length and its variability will
be felt particularly strongly where narratives are short. To illustrate the issue,
consider this hypothetical example of a minimal story realized in just 16
words altogether, with 8 words outside and 8 words inside the quote (indi-
cated by square brackets):
w w w w w w w w [w w w w w w w w]
The position of the quote in this hypothetical narrative is final and we will
expect that the positional value it obtains should reflect that final positioning.
If we use the usual procedure of dividing the number of words preceding the
quote by the total number of words in the story, the resulting value is 8/16 =
0.5 – a value which suggests the quote is positioned in the middle of the text.
The value obviously distorts the quote’s actual position. One method to more
6.3 How do Co-constructive Recipients co-tell stories? 205
PN*
PRC
median PRC
median PN*
1.5
1.5
1.0
1.0
Densities
Densities
0.85 – 0.95 segment
0.5
0.5
0.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Positions of PRC and PN* quotes Positions of PRC quotes
Figure 6.7 Left panel: Density lines and medians for positions of quotes
by Co-constructive Recipients (PRC) and narrators (PN*); right panel:
Bootstrap confidence band for PRC quotes (B = 2000, α = 0.05)
6.3.4 Discussion
The quantitative analysis yielded strong (though statistically inconclusive)
evidence to suggest that recipient quotes distribute across narratives very
much like narrator quotes do and that the respective positional peaks neatly
coincide in the same, narrow, segment. This positional coincidence across
stories, however, is a merely provisional indication of the two-way sequential
co-occurrence of narrator quote and recipient quote within stories, and even
more provisional as regards the three-way sequential co-occurrence of nar-
rator quote, recipient quote, and story climax (the QCQ triangle). The
positional coincidence only suggests the possibility of there being such a
triangle in texts. To examine this possibility in sufficient detail, it is necessary
to move from corpus linguistic methods to discourse analytical methods, thus
combining the quantitative analysis with a qualitative examination of relevant
stories. This qualitative examination is presented in the following.
The analysis proceeds in two steps: first, I discuss recipient quotes whose
positional values fall into, or border on, the positional range between 0.85 and
0.95 in which the two distributions were found to peak; then, I investigate
recipient quotes with positional values smaller than 0.85. This two-step
methodology is warranted in that low positional values are more likely to
signify multiple-highpoint stories, whereas high positional values may tend
to occur in single-highpoint stories.
Close inspection of narratives in which recipient dialog was positioned
within, or around, the peak range confirms that these quotes indeed co-occur
not only with narrator quotation but also with what can be seen as the story
highpoint. Inspection also confirms that peak-range positions of recipient
dialog are typical of single-highpoint stories. Consider, for example, the
narratives in (6.23) to (6.25); the PRC quote is in bold and the likely highpoint
is arrowed. In all three examples, the PRC quote occurs in the 9th positional
208 How do recipients co-author stories?
re-starts the story, thus initiating a recycle and re-constructing the climax,
viz. her amused reaction to the girl’s complaining about her many corns, with
minor changes made: in the ‘original’ version her reported internal comment
is [MDDooh students don’t change.], in the recycled version it is [MDDwell at
your age what are you going to do ?]. If the latter comment is taken as the
‘original’s’ dialog-as-climax, the recipient’s dialog ([MDF [laughing] A corn
on every toe ]), which immediately follows, may be seen as a reaction to it.
However, functionally, the quote defies easy interpretation. First, although
positionally close, the quote does not immediately relate to the narrator’s
dialog-as-climax. This is because it echoes a quote presented by the narrator
in an earlier segment (double-arrowed in the excerpt). Seen thus, it is more
closely related to that segment than to the narrator’s dialog-as-climax.
Alternatively, however, it could be argued that the recipient takes as the
most reportable event, not the narrator’s report of what she (the narrator)
thought upon hearing the student complain, but her report of what the
student actually said (viz. that she had corns on every toe). In that case, the
link between climax and recipient quote would be direct. A difficulty, then, in
this story, as in some others in the data under scrutiny, is that the climax
cannot be established with certainty. Therefore, the token in (6.27) cannot be
cited unreservedly as evidence supporting the three-way co-occurrence
pattern of narrator dialog, recipient dialog, and climax.
(6.27) “A corn on every toe” (Type: T10/Embed Level: EC1)
CNN
CNI
--------------------------------------------------------
Original
S2 PNP Mm. So then I had [unclear] on the train. And these
two girls, I think everybody was going for interviews at
Man er Birmingham university yesterday.
S1 PRR Oh right.
S2 PNP And it was all crowded and so forth. And we met this gi
I don’t think these two were going to Birmingham
university but they were going somewhere.
S1 PRR Mm.
S2 PNP [unclear] So erm these two girls, I couldn’t I mean I
was reading but couldn’t tell but they could. And I
heard the girl saying QSG [MDDwell I don’t look for
smart shoes now. ]
S1 PRR [laugh]
S2 PNP [unclear] good gracious. I mean at least I do get a pair of
smart shoes and they were lovely those shoes all day
S1 PRR Mm.
6.3 How do Co-constructive Recipients co-tell stories? 213
S2 PNP that [unclear] bought [unclear] . Erm she said QSD
→→ [MDD got a corn on every toe. ]
S1 PRR [laugh]
S2 PNP [laughing] Well good gracious .
S1 PRC How old was she?
S2 PNP Oh oh only, she’d be oh twenty five, twenties like.
They [unclear]
S1 PRR Oh dear.
S2 PNP you know [unclear] . [MVVI listened and listened,] I
→ thought QTD [MDDooh students don’t change. ]
S1 PRR [laugh]
S2 PNP So erm
S1 PRC [MDF [laughing] A corn on every toe ].
S2 PNP Yes.
S1 PRR Dear.
--------------------------------------------------------
Recycle
S2 PNP So she said QSD [MDDI have to, ] so I was squin You
know I had a squint underneath to see [laughing]
whether she was wearing any .
S1 PRR [laugh]
S2 PNP You know those fancy boots that are very wide?
S1 PRR Mm.
S2 PNP Got er some of those on. She said QSD [MIIshe couldn’t
she couldn’t, she was crippled if she put a decent pair
→ of shoes on. ] I thought QTD [MDDwell at your age what
are you going to do ? ]
S1 PRR [laugh]
S2 PNP Eh? So she must have been wearing some funny shoes
to get corns on
S1 PRR She probably had, mm.
S2 PNP Yes. She learnt her lesson the hard way. So then it got
into
CNF
S1 PRC It’s too late though now.
(KBW-N1)
Another narrative that is both cyclically structured and in which the identi-
fication of the climax is somewhat problematic is (6.28). In the excerpt, only
the ‘original’ is given (see Appendix 4 for the full text) and question marks in
the left-hand column denote unidentified speakers; the position of the
recipient’s quote is 0.25. The narrator, apparently a female student at a
boarding school, is telling how she got denied access to ‘the boys’ area’ by
a teacher although she insists she was visiting during the official visiting
214 How do recipients co-author stories?
gone]
S1 PRR Yeah?
S2 PNP so I tap on the window, had just gone past and I was
just about to go in his room, you know, cos he’s
standing there [unclear] and I just tapped on the
window [unclear] come outside Nick [unclear] had
just walked by and Rick and [unclear] [MVVwere
talking ] [unclear] you know it’s not like the dead of
night and everything’s quiet
S1 PRR Yeah.
S2 PNP he said QSD to me [MDDit’s a bit late to go knocking on
windows isn’t it,] and I was like QBD [MDD <pause/>]
→ [MXXI didn’t say anything ] I just thought QTD [MDD well]
S1 PRC [MDFNo.]
S2 PNP Yeah.
S1 PRC [MSSBut you can’t really say that.]
UN [unclear]
S2 PNP and he saidQSD erm [MDDcome on this is the boys’
area ] or something like that erm and I just stood
→→ there, I was so amazed because he’s wrong, you’re
allowed to go [unclear]
S1 PRC I know that’s the visiting times you’re allowed to be
down there
? PNP Yeah.
S1 PRC till what, ten fifteen?
? PNP Yeah.
? PRC I wasn’t in his study.
? PRR Yeah.
(KPH-N1)
A few more such difficult, or even improbable, examples can be found in the
data. However, inspection of the 29 recipient quotes in context reveals that in
the majority of stories in which the climax can confidently be identified,
narrator quote, recipient quote, and climax co-occur forming the QCQ
triangle, provided chapter and cycle divisions are taken into account.
To sum up, the analysis of recipient quotes in the context of their enclosing
stories/chapters/cycles strongly suggests that for the use of recipient dialog,
the larger sequential context is crucial: recipient quotes preferably follow
216 How do recipients co-author stories?
6.3.5 Summary
In this section, I have presented a case study on the use of quotation, or
constructed dialog (that is, use of Free Direct and Direct reports), by Co-
constructive Recipient (PRC). The study demonstrates that, and how, by
using dialog, recipients gain direct co-authorship rights in co-constructing
the story climax.
The starting point for the inquiry was a comprehensive overview of types
of PRC contributions. The taxonomy presented included six distinct types:
questions, answers, comments, utterance completions, extensions, and,
finally, discourse presentation, the only PRC response type that, at present,
can be investigated exhaustively and quantitatively in the NC. It was shown
that quotation is the most frequent type of discourse presentation in PRC
responses. I hypothesized that recipient dialog would occur in the sequential
context of the story highpoint. The hypothesis was based on a number of
premises. First, it was argued that the use of dialog by the recipient correlates
with a switch in discourse role occasioned by the recipient’s affiliation and
identification with the narrator: by using quotation as an inferential evidential
the recipient flags his/her self-promotion to narrator. Further, it was argued
that this claim to the role of narrator can be seen as an index of the recipient’s
heightened involvement in the storytelling and that this involvement is most
likely to arise in the context of the climax. Moreover, in the light of obser-
vations reported in the literature of the association of narrator dialog with the
story highpoint, it was assumed that recipient dialog would co-occur not only
with the story highpoint but also with narrator dialog. This three-way
association was referred to as the quote-climax-quote (QCQ) triangle.
The analysis proceeded in two major steps, combining quantitative and
qualitative methods. First, positions within stories were compared for nar-
rator dialog and recipient dialog. Though statistically inconclusive (given the
problems inherent in demonstrating sameness) it was found that, in the NC,
the distributions are strikingly similar: positionally, narrator dialog and
recipient dialog distribute across stories in essentially the same way, with
6.3 How do Co-constructive Recipients co-tell stories? 217
7.1 Introduction
The fundamental aim in this study has been to provide quantitative evidence
of the co-construction of conversational narrative. In the following section,
Section 7.2, I summarize the findings presented in the analytical chapters,
Chapters 3 to 6, and discuss implications of the individual case studies. In
Section 7.3, finally, I consider the implications the study has as a whole.
distribution is very clearly tilted towards the narrator, who gets significantly
greater shares in turns than any one recipient. Third, the N-notN-N pattern
is remarkable in that it demonstrates that recipients fine-tune their response
behavior to each other. The analysis showed that double-responses (that
is, trigrams in which both the second and the third slot are taken by the
recipients) were significantly underrepresented. This suggests that recipients
strive to avoid double responses, and instead give preference to single
responses (that is, trigrams, in which only the second slot is taken by a
recipient). I referred to this striking coordination of response behavior as
an ‘economy of listening.’ Examining the precise mode of operation of this
economy was beyond the scope of the case study. Interesting questions could
not be addressed, for example: What types of signal do recipients rely on to
avoid both of them responding consecutively or in overlap? Are response
turns in the ‘notN’ slot of N-notN-N trigrams equally distributed between
recipients? Does the narrator’s behavior play a role in deciding which of the
two recipients takes the ‘notN’ turn (for example, by means of gaze)? These
questions certainly merit future investigation.
The N-notN-N pattern is thus the result of co-construction both in terms
of turn distribution and turn order. Given that this case study was restricted
to three-party narrative, a worthwhile extension of this work would be to
investigate turn order in more-than-three-party narrative. If, in principle,
the same preferential treatment of the narrator could be observed, the
co-construction of turn order would have to be seen as an even greater
interactional achievement in that the turn space available for each individual
recipient decreases with each additional participant to the narrative.1
In Section 3.3, the co-construction of turn size was taken into focus. The
unit of analysis were turn bigrams, that is, pairs of immediately adjacent
turns. Given observations of extended narrator turn size and the concomitant
reduced recipient turn size, it was assumed that the fluctuation between long
narrator turns and short recipient or Co-narrator turns, referred to as turn
size fluctuation (TSF), would be sizable and homogeneous. It was shown that
TSF is indeed significantly more sizable in narrative than in non-narrative
conversation; however, TSF in narrative was not found to be more homo-
geneous than in non-narrative conversation. It was demonstrated that TSF
homogeneity is decisively influenced by participant role and the interactions
between the participant roles: TSF is high and homogeneous or low and
heterogeneous depending on whether Primary Narrators (PNP) interact with
Responsive Recipient (PRR), Co-constructive Recipient (PRC), or Ratified
Co-narrator (PNC): in bigrams involving narrator and Co-narrator (PNP &
PNC) and in bigrams involving narrator and Co-constructive Recipient (PNP &
PRC), TSF was found to be low and heterogeneous, whereas in bigrams
involving narrator and Responsive Recipient (PNP & PRR), TSF was found
to be high and homogeneous. The fact that TSF is more sizable (but not more
homogeneous) in narrative than in non-narrative conversation is therefore
220 Conclusions and implications
mainly owed to the influence of the latter interactional pattern, PNP & PRR.
The reason why participant role has such a decisive influence on TSF is the
fact that turn lengths by narrators and recipients/Co-narrators are finely
attuned to one another: not only are recipient/Co-narrator turns differently
sized, with sizes ranging from minimal for PRR to sizable for PNC, but also
the lengths of narrator turns are correlated to the type of response that
follows: while, on one end, narrator turns preceding PNC outsize their
successor turn only slightly, on the other end, narrator turns preceding
PRR are a multiple of the length of their successor turn.
This mutual attunement suggests a sophisticated co-construction of turn
size in conversational narrative. Turn size is far from being varied in Sacks
et al.’s (1974) sense. On the contrary, it is subtly predetermined in advance,
the key determining factor being participant role: on a general level, narrators
are conceded long turns, while the turns of all non-narrators are shorter.
Within this general ‘setting,’ though, turn sizes are even more finely pre-
determined inasmuch as the differential size of narrators’ turns was found to
have an impact on the type (and length) of the immediate response. A
‘chameleon effect’ (Danescu-Niculescu-Mizil & Lillian 2011) was observed:
depending on the length of the preceding narrator utterance, length and type
of recipient responses vary, with (very short) PRR responses matching up
with very long narrator turns, and (longer) Co-narrator (PNC) and, specif-
ically, Co-constructive Recipient (PRC) utterances correlating with narrator
utterances of intermediate or short length. It was hypothesized that turn
size is among the variables by which narrators can influence recipients to
contribute in distinct ways. This chameleon effect makes an interesting
analogue with other forms of cooperative behavior, for example, with the
verbal ‘duetting’ noted by Holmes & Stubbe (1997), the adjusting found for
the number of function words in adjacent utterances (Danescu-Niculescu-
Mizil & Lillian 2011), and the matching observed for the length of the
listener-controlled turn interval preceding the stand-alone backchannel
yeah and the length of the speaker-managed turn interval following the
listener’s yeah (Peters & Wong forthcoming) (for an up-to-date survey of
work on the ‘chameleon effect,’ see Danescu-Niculescu-Mizil & Lillian 2011).
Turn size and its fluctuation were examined in two-party narrative only.
However, there is no principled reason why TSF should not also be indica-
tive, to an extent, of narrative interaction between more participants. Cursory
inspection of more-than-two-party narratives suggests that sizable and
homogeneous TSF may not only be characteristic of two-party conversation
but also of three- and more-than-three-party conversation.2 Future research
is needed to investigate this possibility in more detail.
Further, both case studies have been based on analyses of turn n-grams,
trigrams in the case of Section 3.2 and bigrams in the case of Section 3.3.
Examining n-grams has grown into a key method in corpus lexical studies;
to my knowledge, n-grams have not yet been examined beyond lexis. The
7.2 Summaries of the analytical chapters and conclusion 221
present analyses have demonstrated that the notion can fruitfully be extended
to aspects of turntaking, suggesting that further extensions to non-lexical
patternings might be conceivable.
Finally, on a more general level, the case studies support the notion that
corpus linguistics and the conversation-analytical study of turntaking can not
only mutually benefit, but that “they can actually synergize” (O’Keeffe &
Walsh 2012: 178). If more widespread use is made of turntaking-related
annotation, as provided in the NC, this synergy may prove even more fruitful
in CA-inspired corpus research in the future.
Chapter 4
This chapter has attempted, in the main, to shed light on narrator co-
construction via one crucial aspect of recipient design: the marking of quotation.
The analytical sections reporting on quotation markers were preceded by
Section 4.1, a largely theoretical section, introducing concepts key in under-
standing discourse presentation. The concepts included the notions of
immediacy (the cline of degrees by which the reporting mode types make
the reported discourse more or less immediate to the listener), demonstration
(quotation as a selective depiction rather than verbatim reproduction), and
the boundary issue (the necessity for narrators to demarcate the boundaries
of demonstrated material, whose referential elements need to be processed ‘in
another light’). Two case studies were reported exploring the potential
functioning of items as ‘auditory quotation marks.’
Section 4.2 examined positioning in the use of interjections in quotes. The
primary focus was on the role of interjections in first position. Two items
were shown to be significantly more frequent in first position in quotes than
in conversational utterances: oh and well. Two alternative interpretations
were discussed. I noted the possibility that narrators preferentially pre-select
for quotation oh- and well-prefaced utterances. This pre-selection may
be based on the perception of oh- and well-prefaced utterances as of greater
‘interest’ (Brown & Levinson 1987) to the recipient given the essential
functions of oh (as a ‘change-of-state’ marker indexing the receipt of new,
unexpected or surprising information) and well (as a marker of dispreferreds
indexing that the utterance does not, or not fully, satisfy social expectations).
Their preselection could thus serve to increase the recipient’s involvement in
the storytelling. Alternatively, drawing on the theory of lexical priming and,
specifically, the notion of textual colligation (Hoey 2005), it was argued that
oh and well have two primings: one for utterance-launch and one for quote-
launch. It was demonstrated that the two primings are clearly distinguishable,
not only by means of lexical associations (for example, collocations with
quotative verbs) and genre distinctions (the priming for quote-launch is
associated first and foremost with conversational narrative rather than general
conversation), but also, perhaps more importantly, by means of position: seen
222 Conclusions and implications
against the backdrop of the narrative utterance within which the quote occurs,
quote-first oh and well occur significantly later than oh and well occur in
conversational utterances. In this perspective, the priming for quote-first use
alerts the recipient to the forthcoming quotation and signals the concomitant
need to resolve references in quotation as pertaining to the reported speaker(s)
rather than the reporting speaker. Both interpretations emphasize the role of
oh and well in the service of recipient design: seen as components of preferen-
tially pre-selected utterances, oh and well intensify the recipient’s interest, seen
as a priming for the inception of dialog, oh and well attend to the recipient’s
processing needs.
Section 4.3 was concerned with short silent pauses in the context of quotes.
A positional analysis suggested that silent pauses frame quotes: they exhibit a
significant attraction not only to initial but also final positions. Silent pauses
were thus interpreted not only as choice-related speech management expres-
sions in Allwood et al.’s (1990) sense serving the speaker to gain planning
time but also as recipient-related processing instruction demarcating the
onset and completion of quotation. Unlike interjections, which mark only
the left-hand boundary, silent pauses mark both the left- and right-hand
boundary of quotation, thus functioning as a quote-unquote signal.
The findings discussed in the case studies shed some light on how speakers
recipient-design quotation with a view to alleviating the considerable pro-
cessing constraints imposed by the switches in role, perspective, voice, and
reference that are characteristic of Direct and Free Direct discourse presen-
tation. Recipient design in the use of quotation can undoubtedly be further
enlightened. First, more marking devices are conceivable. For example,
mimicking voice quality is likely to “serve as an ongoing acoustic prosodic
marker [of quotes]” (Halliday & Matthiessen 2004: 447). Second, in the case
studies presented above, no distinction was made between Free Direct and
Direct quotation. Since Free Direct is, by definition, without preceding
reporting clause (a very obvious marker of quotation), the need for alternative
marking might be greater for Free Direct than for Direct quotes. Also, the
means used to mark quotation might differ between the two quotation types.
Third, in this chapter, interjections and pauses have been investigated
separately. However, there is the possibility that different types of quotation
marker interact in some way or another; for example, the use of one marker
might obviate the need for some other, or the reverse, markers might
typically work in unison only but lose in force as standalones. The analyses
presented in this chapter are hence merely starting points for further, more
fine-grained explorations into the recipient design of quotation in conversa-
tional narrative.
Finally, the two case studies have important methodological and theoret-
ical implications. Methodologically the positional analyses of interjections
and silent pauses in quotes have broken new ground in that Hoey’s (2005)
textual colligation claim, which has so far only been discussed with a view to
7.2 Summaries of the analytical chapters and conclusion 223
written data, has for the first time been tested against spoken data. Also, while
the statistics in much research on positioning tend to be basic, the positional
analyses in this chapter have been validated using rigorous statistical exami-
nation. The findings are of theoretical importance inasmuch as this study has
statistically substantiated the textual colligation claim for conversational
narrative. That is, the case studies have provided quantitative evidence to
suggest that speakers of English are primed (i) generally, to use interjections
utterance- and quote-initially, (ii) specifically, to use oh and well in quote-first
position, and (iii) to use silent pauses both in initial and final positions within
quotes. What is as yet missing is psycholinguistic evidence to complement
the statistical evidence. For priming is first and foremost a psychological
concept and whether language users are indeed primed to use, for example, oh
and well at the immediate onset of quotation needs yet to be confirmed by
psycholinguistic research in experimental settings.
Chapter 5
In this second of two chapters dealing with discourse presentation, the focus
has been on ways that narrators, following the recipient design principle,
utilize discourse presentation co-constructively in the sense of designing
utterances dramatically.
Section 5.2 was concerned with the notion of climactic structure at the level
of the individual utterance. The focus was on sequences of report units
within utterances. Building on the immediacy cline model developed in
Section 4.1, a robust trend was found across all report sequences investigated
(3 ≥ n ≤ 8) to increase in immediacy over the sequence. A similar significant
tendency towards immediatization was found for n=2-sequences, the largest
group of report sequences in the NC. Further, it was discovered that the
trend towards greater immediacy is to a very large extent due to the intra-
sequential increase in occurrence of the two mimetic modes Free Direct and
Direct.
It was argued that, given this immediatization trend, the same ‘tension’ can
be observed at the level of the utterance that Longacre (1983) characterized as
‘plus’ in narrative as a discourse type. Given that tension is the driving force
behind the climacto-telic structure of storytelling, the observation of tension
generated by immediatized report sequences suggests that the notion of
‘climax’ can be seen, not only in association with the large unit of the story,
but also in association with the smaller unit of the utterance: utterances
exhibit climacto-telic structure too. I noted that an open question relates to
the possibility that the thus generated utterance tension could, not only
reflect, but also co-generate the larger tension of storytelling, a possibility
which merits closer attention in future research.
Immediatization in the use of report sequences at utterance level and the
concomitant utterance tension are a clear expression of recipient design as the
224 Conclusions and implications
Chapter 6
This chapter moved recipients center stage, investigating the ways they
co-author and thus co-construct narrative.
Section 6.2 investigated the co-authorship gained by recipients via use of
feedback as such. Starting from research on general conversation which
suggests that increases in backchannel frequency correlate with increases in
turn length (Peters & Wong forthcoming) it was hypothesized that increases
in recipient responses lead to increases in the total number of words in
narrator utterances per story.
Three independent subsets of data were created: stories in which the
responses were only telling/teller-related (PRR), stories in which the teller
received only substantial tale-related responses (PRC), and stories with both
types of responses co-occurring. Based on negative binomial (NB) models it
was shown for all three subsets that recipient feedback, regardless of the type
of recipient response, impacts significantly on the number of words used by
the narrator: narrator verbosity increases exponentially with increasing fre-
quency of feedback by listeners.
Since the effect of stimulating narrator verbosity was found both for stories
in which only telling/teller-related (PRR) and, respectively, tale-related
(PRC) contributions occur, and for stories in which the response types
co-occur, the observed increase in narrator verbosity is thus independent
of response type. That is, whatever the type of response, narrator produc-
tivity is stimulated by recipient feedback as such. However, differences were
found with regard to the strength of the influence on narrator verbosity:
PRR feedback increases narrator verbosity more than PRC feedback. This
difference was explained with recourse to the differential nature of the two
response types. While telling/teller-related PRR responses, being semanti-
cally void and structurally non-clausal, leave the job of doing the telling
entirely to the narrator, a kind of semantic and structural job sharing was
observed for tale-related PRC responses: they interfere with the emergent
story topically thereby providing semantic and structural material that can
be economically re-used by the narrator. Finally, a negative interaction was
discovered for PRR and PRC which produces a moderating effect: increases
in co-occurring PRR and PRC responses slow down (but do not halt)
the exponential increase in narrator verbosity. The moderating effect was
226 Conclusions and implications
value of recipient dialog was < 0.85. It turned out that the majority of these
narratives were multiple-highpoint stories with chapter or cycle divisions.
Again, it was found in the majority of instances that recipient dialog
co-occurred with what could be identified as the chapter or cycle highpoint
realized in the narrator’s quote and again, like in single-highpoint stories, the
recipient quote served to echo, extend, or intensify the climax.
To conclude, recipient dialog is closely related to narrator dialog and story
highpoint: it follows the narrator’s quote-as-climax not only sequentially but
also functionally by echoing, extending, or intensifying it. Narrator quote,
climax, and recipient quote form a tightly knit three-way co-occurrence
pattern, the QCQ triangle. By inserting dialog at the story highpoint, recipients
powerfully lay claim to the narrator role thus becoming co-authors at, not just
any point in the storytelling process, but the most crucial. Given its central
role, the shared authorship of the climax can, arguably, be seen as the highpoint
of narrator-recipient co-construction of conversational narrative.
The case study invites further investigation in at least two respects. First, it
opens a window on the climax, a seriously underresearched key element of
stories. Since recipient quotes and narrator quotes sequentially co-occur
with the climax and since their positional maxima coincide in the same very
narrow positional range (between 0.85 and 0.95) not only in single-highpoint
stories but also in multiple-highpoint stories (provided the positions of
recipient quotes are re-calculated on the basis of chapter and, respectively,
cycle divisions), it seems safe to conclude that the peak range identified by the
positional analysis is the preferred positional context for the occurrence of
the story climax where it is realized in narrator and recipient dialog. This
positional identification has far-reaching implications for the automatic
identification of the climax in future corpora of conversational narrative:
the co-occurrence of narrator and recipient dialog within, or around, the
range between 0.85 and 0.95 is a highly probable indication of the occurrence
of the climax. Whether this range is indicative of the occurrence of the climax
even where the climax is not realized either in narrator and recipient dialog
or in narrator dialog alone is as yet an open but intriguing question. The
possibility that the identified positional peak range is the climax’s ‘habitat’
seems promising and a useful starting point for algorithms targeting the
automatic retrieval of the story highpoint.
Second, seen in a wider perspective, as briefly noted by Mathis & Yule
(1994: 71), recipient dialog merits further investigation in terms of speech
accommodation theory (e.g., Giles et al. 1991). By constructing dialog, recip-
ients perform an accommodative act: they converge with the narrator’s
speech style, with convergence understood as “a strategy whereby individu-
als adapt to each other’s communicative behaviours” (Giles et al. 1991: 7).
However, the convergence is not only asymmetrical, i.e., accomplished only
by the recipient (see, for example, White 1989, Schmid & Küchenhoff 2008),
but also partial (see, for example, Bilous & Kraus 1988). The partialness is due
228 Conclusions and implications
Conclusion
In sum, the quantitative evidence for the co-construction of conversational
narrative presented in the case studies can be considered substantial. The
evidence collected covers the most diverse research areas and related lingui-
stic fields; these include, in Chapter 3, the conversation-analytical study
of turntaking and corpus-linguistic research into n-grams; in Chapter 4,
research into positioning and textual colligation; in Chapter 5, investigation
of co-occurrence patterns at the level of discourse units; finally, in Chapter 6,
research into narrative structure, as well as speech accommodation theory.
The analyses thus provide quantitatively substantial and conceptually
diverse, empirical support for the view that conversational stories constitute
“interactive achievements involving all participants to varying degrees”
(Holmes & Stubbe 1997: 7).
. . . there is no limit to what we can hope for, for there is no doubt that, as
advances continue to be made in the field of corpus-processing software, our
quest will become less of a dream and more of a reality. But that is another
story. (Gilquin 2002: 207)
The overall implication of the present study, beyond the implications of
the individual case studies discussed in the previous section, pertains to the
study’s methodological base, the NC and its annotation. The manifold
insights reported in this book would not have been possible without access
to the intricate discourse and pragmatic annotation implemented in the NC.
7.3 Implications and outlook 229
does not tolerate uncertainty. It requires the annotators to take clear deci-
sions; even where annotators code a token as ambiguous, this coding is an
unambiguous decision. Whatever their decisions, however, they are unlikely
to be shared by everyone in the field. As a result, discourse-annotated corpora
will be less useful for those who work on contrary or diverging conceptual
bases than for those sharing or accepting the conceptual bases underlying
the markup. However, it should not be forgotten that even straightforward-
looking concepts such as what counts as a word or a word class upon closer
inspection turn out to be fraught with a number of uncertainties (see Sinclair
1991) and, still, despite these uncertainties, all POS-tagged corpora heavily
rely on them. So, full consensus is at best an ideal that the annotation can
orient to but which will never be achieved. The grain of doubt that will
inevitably surround any sophisticated annotation scheme will be compen-
sated by the unique perspective offered by using richly annotated corpora:
viz. the discovery of structures that are both innovative and quantifiable, and,
hence, “otherwise impossible to ascertain” (Biber & Conrad 2001: 332).
So, corpus linguistics is at a crossroads. It can either continue focussing its
resources on investigating the idiom principle and the structures that con-
tribute to it. To achieve this end, existing corpus resources can be exploited
more fully, such as the big reference corpora of recent years and the web as
corpus, as well as construct new up-to-date corpora which follow the clean-
text policy or which contain POS annotation or parsing. Alternatively, corpus
linguistics can continue on this path, which has proven successful, while, at
the same time, allocating some of its resources to investigating discourse
based on specific, richly annotated corpora targeting specific pragmatic and
discourse phenomena in specific genres. The outcome of this latter scenario
would be a corpus linguistics that is both surface- and annotation-driven
and that is capable of unraveling both lexical and grammatical structure as
well as discourse structure. If this is the path followed, to adapt the quote
from Scott & Tribble (2006), it may be that future observers will come to
note that ‘by the early 21st century discourse came to occupy the centre of
language study previously dominated by syntax, grammar, and lexis’ – but
that is, indeed, another story.
Appendix 1
S1 PNP It were just you know oh it’s a nice morning when I’m hanging
the washing out as though everything was wonderful you know
S2 PRR Yeah.
S1 PNP normal.
S2 PRR Yes.
S1 PNP Ooh I thought QTD [MIIhe was going to, ] ooh I felt made me go
cold and I I thought QTD [MIIif, if I hadn’t been here on my
234 Appendix 1
own I’d have been very tempted to go round and and say is there
anything I can do.] [MVTOr I’d have called the police ] I think. I
weren’t sure what I’d have done but
S2 PRC Bad as that?
S1 PNP Oh [unclear]. [MSSHe was a calling her effing this,] [MSSswearing ]
oh it was dreadful. [MSSShe was screaming ] and you could hear
her crying. Then you hear, she pounded up the bloody stairs and
he was after her and I reckon well the only room with a lock
S2 PRC As you say is the bathroom.
S1 PNP is the bathroom, so she must have been in there cos he’s hammer-
ing on the door and bashing it and kicking it. [MSS Bloody hell of a
row. ] Or whether she opened it or whether she was in the front
bedroom with something behind the door I dunno but
S2 PRR That’s his
S1 PNP and she run back down the stairs and the front door was opening
and banging and shutting, whether she run outside or what I
don’t know what happened, the dog was bloody barking.
S2 PRR Oh lord.
--------------------------------------------------------
S1 PNP But it was [MVV I mean I could hear it over the telly. ] I know I
don’t have the telly on full blast but [MVV I could hear it]
S2 PRR Mm.
S1 PNP [MVVabove the telly. ] When I first heard the first noise
I thought QTD [MDDwhat the bloody hell’s that ] and I turned
the sound down.
S2 PRR Yeah.
S1 PNP Thought QTD [MDDbloody hell, put it back up ] but you could,
it was really distressing me cos
S2 PRC You knewQOO [MIIwhat was possibly going on.]
S1 PNP Yeah.
S2 PRR Mm.
S1 PNP A woman and that don’t s I mean [MSSif you’re having just an
argument it’s raised voices ] [MVVbut a woman doesn’t scream
and cry like that ] unless
S2 PRC Unless there’s something really
S1 PNP something awful going on.
S2 PRR Mm.
--------------------------------------------------------
land would give him the house if it’s hers. [unclear] evicted and
then she could come back again.
S2 PRR Mm.
CNF
S1 PNP Well if she went to the police they’d chuck him out. He has no
legal rights to the property. It’s in her name, she’s, her father’s
helping her paid the deposit or something and she clears the
mortgage she has no right, he has no right to it all if she went to
the police. They’d just turf him out wouldn’t they?
S2 PRR Mm.
CPO
S1 Well why stay in a relationship like that? Don’t understand it.
S2 Certainly a bit rough, a bit hard.
--------------------------------------------------------
S1 PNP Went quiet about half past ten so whether he buggered off
out or what I don’t know.
S2 PRR Mm maybe.
CNF
S1 PNP Certainly the front door was opening and banging and
shutting.
CPO
S2 Say I’m surprised Ann and Rita haven’t sort of didn’t
S1 Well banging on the wall probably with all the other banging
and
S2 May maybe. Maybe. They [unclear] part of the banging.
S1 So much bloody banging going on.
S2 Probably part of the banging Ann and Rita.
S1 And the people on this side [unclear] you don’t know how
they were reacting as well. Say, they was a quite young
couple. We don’t hear them that much do we? Mornings is
the time I hear them most. Cos they go out to work at half
past eight and you hear them between half past seven and half
past eight
(KB7-N2)
Appendix 2
Introduction
1. Most stories in the SCoSE apparently originated in conversations between aca-
demics based in Northern Illinois (cf. Norrick 2000).
2. A somewhat similar corpus project is reported on in Carruthers (2008). This
project involves the collection, transcription, and encoding of ‘new story-telling’
in French, a type of narrative performed publicly to eclectic audiences.
246
Notes to pages 17–31 247
event is Carolyn’s discovering that she’d have to face the intruder all on her own,
the (likely) reason why the story is being told is the speakers’ wish to empathize
with Carolyn vis-à-vis the dangers inherent in her work.
6. Text labels in the NC, such as KC9-N2, refer to the source file in the BNC-C (the
file KC9 in this case) and indicate whether the narrative is the first (N1) or second
(N2) extracted from the BNC-C.
7. Norrick (2000: 3) maintains that an organizational principle competing with
sequentiality is “repetition and formulaicity.” His conversational data “exhibit
storytellers organizing their performances around repetition and formulaicity as
much as sequence” (2000: 3).
8. The mother’s ability to guess that the boy chatted to Louise may, of course, be
due to a number of contextual factors. One of these factors may be a successful
reading of her daughter’s use of ‘foreshadowing,’ a technique discussed in more
depth in Section 1.7, where narrators insert into early stages of the narrative,
details whose specific significance is revealed only later, when interpreted in the
light of upcoming information. In the excerpt, the foreshadowing detail is
the mention of Louise (and he sat down next to Louise), which, at the point in
the discourse where it is inserted, seems marginal and hardly relevant. Only at the
point where her daughter conveys her disappointment at the boy showing little
interest in her, does the mother grasp the implicated significance of the fore-
shadowing mention of Louise.
9. Also, as Ervin-Tripp & Küntay (1997) observe, there is no explicit temporal
sequence of events; the damage reported could occur all at the same time.
However, given that the damage is the consequence of the earthquake, there is
an “implicit temporal sequence” (Ervin-Tripp & Küntay 1997: 137).
10. A useful distinction of three types of material under the heading of ‘orientiation’
is proposed in Norrick (2000: 33): general frame (information about time and
place); background information (additional information related to the setting,
which may not be necessary for the point of the story), and narrow frame
(information directly leading into the main action of the story).
11. Note that, in this quote, Labov, somewhat confusingly, equates the most report-
able event with the story point (the reason why a story is told). More precisely, the
most reportable event constitutes the story highpoint, or climax.
12. Obviously, this narrative macrostructure does not originate ex nihilo; among the
cognitive processes involved may be the following. To begin with, consider
Sacks’s (1992: 237) notion of course-of-action organization, that is, the process
of imposing a temporal sequence on the events to be reported. Bearing in mind
that the succession of narrative clauses a and b is automatically interpreted as
implying the a-then-b relation according to which a must have occurred before b
occurred (cf. Labov & Waletzky 1967/1997: 24) the first, threefold, operation a
would-be-storyteller must perform is (i) to select from all events that occurred in
the told situation the story-relevant events by discarding events which have no
bearing on the story, (ii) to sequentialize the relevant events, some of which may
have occurred wholly or partly simultaneously, into a temporally coherent string
of events, and, finally, (iii) following the clause-event match postulated by
Labov & Waletzky’s (1967/1997) (cf. Section 1.4), to encode events one by one
in matching narrative clauses. That is, narrativization of experience entails a
series of cognitive procedures designed to match the expectations that recipients
248 Notes to pages 34–52
became noticeable; the resulting two parts were assigned to different components.
Consider:
“Bruise the ginger” (embedLevel="ES" narrativeType="T10")
1 <u who="PS08Y">
<s n="1672" sID="1672"/>
<w c5="PNI" hw="none" pos="PRON"> None</w>
<c c5="PUN"> ?</c>
</u>
</seg>
→ <!– utterance split –>
<seg Components="CNN">
<seg Components="CNI">
2 <u who="PS08Y" Participation_roles="PNP">
<s eID="1672"/>
<s n="1673" sID="1673"/>
<align with="KC0LC0WW"/>
<w c5="PNP" hw="i" pos="PRON"> I</w>
<w c5="VM0" hw="can" pos="VERB"> can</w>
<w c5="VVI" hw="remember" pos="VERB"> remember</w>
<w c5="AT0" hw="the" pos="ART"> the</w>
<w c5="ORD" hw="first" pos="ADJ"> first</w>
<w c5="NN1" hw="time" pos="SUBST"> time </w>
<w c5="PNP" hw="i" pos="PRON"> I </w>
<w c5="AV0" hw="ever" pos="ADV"> ever</w>
<align with="KC0LC0WX"/>
<s eID="1673"/>
</u>
</seg>
(KC0-N1)
For space considerations only the (originally single) utterance that was split in two
utterances, labeled 1 and 2, is given in the example. Inspection of the preceding context
shows that participant PS08Y’s None? in utterance 1 functions as a response to another
participant who has just declined PS08Y’s offer to have more ginger. In saying I can
remember the first time I ever participant PS08Y performs a decisive shift in orientation
leading the discourse from an offer-declination adjacency pair to the launch of a story,
thus justifying the utterance split implemented in the text’s annotation.
7. Interestingly, while the narrative component is larger than the surrounding non-
narrative talk by roughly 8,000 words, it has strikingly fewer utterances. Taken
together, CPR and CPO include 8,933 utterances, whereas CNN includes only
6,060 utterances. Utterances in narrative must hence be considerably longer than
in general conversation.
This is confirmed by a comparison of the mean numbers of words per utter-
ance: the mean lengths of utterances in CPR (7.75) and CPO (8.07) are rather
similar, yielding a combined mean of 7.91 words for the non-narrative compo-
nents. Utterances in the narrative micro-components, by contrast, are much
longer but also display much greater heterogeneity with regard to length. While
CNI utterances have a mean length of 29.18 words (between three and four times
250 Notes to page 53
40
250
30
Number of words per utterance
20
100
10
50
0
CPR CNI CNM CNF CPO CPR CNI CNM CNF CPO
In the figure, the boxplots showing the distributions of the lengths of utterances
in the different textual components are presented in two different panels. While
the left panel shows the distributions entirely (with the whiskers and all outliers
included), the right panel presents the boxes in greater detail, thus permitting a
clearer view of the boxes and the notches.
In the left panel, it can be seen that the dispersion in narrative-initial utterances
(CNI) and narrative-final utterances (CNF) is much greater than in the non-
Notes to page 53 251
The reasons for the peaks in length in narrative-initial (CNI) and narrative-final
(CNF) utterances cannot be explored in full detail here. A few, in part speculative,
remarks may suffice. Given that the narrative-first utterance marks the change in
discoursal orientation from the present situation to the anterior situation in which the
events occurred, the dramatic increase in length that characterizes CNI may be
associated with an increase in new vocabulary. Work, for example, by Youmans
(1991) shows that breaks in the discourse (for example, transitions from one topic/
section to another) may be correlated with vocabulary ‘bursts’, that is, sudden
increases in discourse-new words. By contrast, the verbosity of narrative-final utter-
ances is unlikely to indicate a similar influx of new vocabulary. Rather, it needs to be
borne in mind that CNF, like CNI, overwhelmingly represents narrator utterances,
whereas CNM comprises both narrator and recipient utterances. The length of CNF
is therefore likely to a large part simply representative of the typical narrator utter-
ance. This assumption is underscored by a comparison of the mean length of CNF,
which is, as noted above, 22.55, and the overall mean length of narrator utterances
(irrespective of narrator subrole), which is 21.07. According to a two-sided Wilcoxon
rank sum test, this difference is not significant (W = 742232.5, p-value = 0.1999).
Second, I would argue that the increase in length from CPR to CNI and the
decrease from CNF to CPO suggests that the segmentation of the texts in the NC
252 Notes to page 55
into the three major components pre-narrative talk (CPR), narrative (CNN), and
post-narrative talk (CPO) indeed corresponds to segments in the discourse. To
put it differently, I take the significant differences in utterance length in CPR and
CNI as well as CNF and CPO respectively as evidence that the componential
annotation in the NC reliably mirrors the shift in the generic framework from
general conversation to narrative and from narrative back to general conversation.
I also suspect that future automatic annotations of conversational narrative can use
the two major points of changes in utterance length (from CPR to CNI and from
CNF to CPO) as a means of identification of narrative boundaries.
Third, it is of interest to compare the utterance lengths obtained for the NC to
the lengths obtained for the BNC-C. According to Rayson et al. (1997), women
use, on average, 10.33 words per utterance, whereas men use 9.53 words per
utterance (1997: 136). Rayson et al. do not present utterance length independent
of sex but, given the figures for number of words and utterances per man and
woman in the study, the overall mean length (independent of speaker sex) is 9.99
words per utterance. The mean utterance length obtained for the NC is strikingly
similar: 9.97 words per utterance. As regards utterance length, the NC is thus a
near-perfect reflection of the BNC-C. The near coincidence of the mean lengths
can however be interpreted in another, wider, perspective. We can ask if the match
in number of words per utterance between the two corpora indicates a match in
the distribution of general conversation and conversational narrative in the two
corpora. That is, considering that the NC consists of two roughly equal halves, one
narrative (CNN) and one non-narrative (CPR and CPO combined), which are
both drawn from conversation, does the near coincidence of the mean utterance
lengths for the NC and the BNC-C indicate that, roughly, half of all conversation
is conversational narrative? Estimates regarding the proportions of narrative in
conversation abound. As noted in the Introduction, generally, very high frequen-
cies are ascribed to stories in conversation. To my knowledge, these estimates are
commonly not based on quantitative analysis and hence speculative. Admittedly, if
I were to claim that storytelling makes up roughly 50% of all conversational talk
such a claim would be not much less speculative given that the near coincidence of
means of utterance length is just one, preliminary, piece of evidence and that many
more factors may influence the generic make-up of conversation, a genre which is
fraught with “genre-mixing and embedding” (McCarthy 1998: 31). However, the
figures obtained for the NC do suggest that possibility and might serve as a
starting point for more quantitative studies aiming to establish the exact propor-
tion of storytelling in its host genre conversation.
8. To establish whether the differences in utterance length between participant types
are significant, pairwise comparisons were performed using Wilcoxon rank sum
tests. The results suggest that all possible pairings are significantly different:
9. Note, however, Barbieri (2007), who reports on the use of another ‘new’ quotative, be
all, which is apparently not (yet) attested in many varieties outside North-America.
10. The tagset for BE like includes seven values: QBB = base form “be like,” QBM =
“’m/am like,” QBZ = “’s/is like,” QBR = “’re/are like,” QBD = “was/were
like,” QBG = “being like,” QBN = “been like.”
11. Comparing this triplet to the quartet of SAY, GO, THINK, and BE like commonly
seen as the major quotatives across most English varieties (see above), we notice the
absence of BE like. It has a total frequency of a mere 9 tokens in the NC. Even if like
(without preceding BE), which occurs 9 times too, is counted in as a variant of BE
like, the resulting total frequency of 18 occurrences of (BE) like must be disappoint-
ing if compared to the increasingly high rates reported for this new quotative in
much research. For example, in Tagliamonte & D’Arcy (2004), who look at
quotative usage in Canadian youth, BE like was found to leave SAY, GO, and
THINK far behind: BE like accounted for 58% of all quotatives in Tagliamonte &
D’Arcy’s corpus. Similar top rates have been reported for other varieties of English
(for an overview of relevant research see Buchstaller 2008).
Why is (BE) like so greatly underrepresented in the NC? The reason is
straightforward: the recordings for the BNC-C, from which the NC is drawn,
were made in the early 1990s, that is, at a time when the attested global spread of
(BE) like from California, where it is commonly assumed to have originated (cf.
Fairon & Singler 2006), to other varieties of English had only just reached the
shores of Britain. In a number of studies based on conversational data from the
first half of the 1990s, BE like was observed to have entered the language but still
to trail far behind more traditional quotatives like SAY and also GO (cf. Miller &
Weinert 1998, Andersen 2001, Rühlemann 2007). In today’s usage in British
conversational narrative, the situation may have dramatically changed: extrapo-
lating, for example, from Tagliamonte & Hudson’s (1999) study, in which BE
like, THINK, and GO were equally represented (each accounting for 18%), BE
like is highly likely, at least in narratives told by (female) adolescents, to challenge
SAY’s leading role seriously. Note also Buchstaller (2011), in whose NECTE2
corpus, which comprises Tyneside English from 2007–2009, BE like has
increased to 21% overall (Buchstaller 2011: 73).
12. Given its metalinguistic focus, Representation of Use (MUU) may be somehow
related to what is often called ‘pure quotation’ (cf. Brendel et al. 2011: 4ff.), as in
“‘Boston’ is disyllabic” (Brendel et al. 2011: 4). The exact commonalities and
differences between the two categories need yet to be explored though.
13. But see the many uncertainties that surround the notion of representativity and
which, as noted in Leech (2007), make the quest for representativity resemble the
quest for the holy grail – an ideal that will be never reached.
14. I’m grateful to members of the “Corp Ling with R” Google Group at
http://groups.google.com/group/corpling-with-r for useful thoughts on the
issue of population in corpus studies.
15. Nodes include not only ‘elements’ and ‘attributes’ but also ‘root,’ ‘text,’ ‘name-
space,’ ‘processing-instruction,’ and ‘comment.’
16. For this document to be fully complete as an XML document, also <?xml
version=“1.0” encoding=“UTF-8”?>, the ‘XML declaration’ would have to be
included.
17. To run XQuery, the (free) eXist XML database was used; cf. http://exist-db.org.
254 Notes to pages 77–82
for(i in 1:9998){
if(path[i]=="N" & path[i+2]=="N") trigrams[i] <- 1
if(path[i]!="N" | path[i+2]!="N") trigrams[i] <- 0
}
tab <- table(trigrams)/sum(table(trigrams))
proportions[j] <- tab[2]
}
mean(proportions)
Note that the restriction that the first turn slot is equally open to any of the three
participants (and that hence the probability for them is 1/3 each) has no influence
on the resulting proportion of N-notN-N trigrams. That is, even if N is set as
defaulting to the first slot of the first trigram (a justifiable decision given that the
vast majority of stories are started by the narrator him/herself; see the previous
endnote), the proportion expected for N-notN-N trigrams is, according to the
simulation, 17%. (To verify simply replace the sixth line of code with the code
given after the “#”.)
4. What is more, looking at the stories individually, the narrator gets virtually always
more turns than each recipient. The barplots below show the differences in number
of turns taken by the narrator and turns taken by the recipient who responds
first (R1) (left panel) and, respectively, the recipient who responds second (R2)
(right panel).
N - R1 N - R2
Differences narrator (N) and first recipient
6
Differences narrator (N) and second
recipient (R2) per story
4
4
(R1) per story
2
2
0
The number of narrator turns is smaller than the number of turns by one of the
recipients (viz. R1) in one story, where the difference is hence negative (see the
single downward bar in the left panel). The number of turns taken by the narrator
256 Notes to pages 86–98
equals the number of turns taken by the recipient (R1 and, respectively, R2) in
two stories, where the difference is hence zero. In all other cases, the narrator’s
turn share is clearly greater than any one recipient’s.
5. The result was double-checked and fully confirmed by a bootstrap, the boot-
strap’s confidence interval (CI) spanning between 0.06 and 0.15.
6. According to Shapiro-Wilk tests for normality (cf. Gries 2009b: 150) the distribu-
tions of either variable violate normality (number_all_trigrams: p-value = 0.0005;
proportion_N_notN_N: p-value = 0.017). The p-values are below the 0.05 thresh-
old; therefore the H0 underlying the Shapiro-Wilk test, which assumes normal
distribution, can be rejected.
7. The null probability does not impact the confidence interval (CI). The CI is the
same for different null probabilities.
8. I’m grateful to Stefan Gries for suggesting the use of a permutation test for the
present analysis.
9. In performing the bootstrap, the number of resamples was set to 10,000. That is,
10,000 different estimations of the proportion of N-notN-N trigrams out of all
trigrams were generated. Based on these estimations and their differences a 95%
CI was calculated. The most widely used CI (cf. Crawley 2007: 322) in boot-
strapping is the BCa interval. It was found that the BCa CI obtained from 10,000
replicates ranges between 0.27 and 0.40. To compare, the CI obtained from the
proportions test was 0.28 to 0.41.
10. To illustrate the difference between mean and median consider this set of
five numbers 1, 2, 3, 4, 5. Both the mean and the median are 3. The numbers
are distributed exactly symmetrically around this central value. If, however,
the set of numbers is asymmetrical, as in the set 1, 2, 3, 4, 55, the mean changes
to 13 while the median remains the central value, namely 3 (cf. Woods et al.
1986: 31).
11. The results for the non-narrative components CPR and CPO deserve a closer
look. The medians and the IQRs for CPR and CPO reveal distinct tendencies:
both fluctuation size and fluctuation homogeneity are markedly greater in the
post-narrative component CPO than in the pre-narrative component CPR. They
are clearly more similar to the median and the IQR obtained for CNN: TSF size
and homogeneity behave in CPO more CNN-like than they do in CPR.To
establish with greater confidence whether this ‘spill effect’ for TSF from CNN
to CPO can be observed, not only for text KB9-N2, but generally for discourse
segments following narratives in conversation is an interesting question worthy of
further investigation in future research.
12. It may be felt that, compared to the medians and IQRs obtained for text KB9-N2,
the reference values are too ‘liberal.’ There are two reasons to prefer these values
to more rigid ones. First, the values obtained for CNN in the example text seem
extreme; for example, an IQR of 0.15 is unlikely to be representative of many
other texts. Second, one factor that is not controlled for in the present analysis
and which inevitably leads to smaller medians and higher IQR values is overlap
(simultaneous speech). Overlap is problematic for the present analysis in that
TSF coefficients are calculated on the basis of the assignment of words to distinct
<u>-elements (that is, utterances) which are invariably ordered successively and
assigned to distinct speakers. As a result, in the case of overlapping speech this
Notes to page 98 257
practice can lead to the intrusion of ‘fake’ turns in the transcript and hence
incorrect TSF coefficients. The issue is illustrated in the following extract; three
columns are added to the text: the column labeled N gives the number of words
per turn; next is the TSF coefficient (Coeff.), while the right hand column (Corr.)
gives the corrected coefficients that would result if the overlap were taken into
account
“Liverpool boys” (Type: T10 / Embed Level: EN)
CNN
CNI N Coeff. Corr.
1 S1 PNS That’s right. And all smoky. Oh I know. 30
Well fo well there was some lads from
Liverpool that was evacuated when well
were evacuated. Oh! God! They were
filthy!
2 S2 PRR Mm. 1 0.94 0.94
3 S1 PNS I mean not just filthy i er bodily but they 13 0.86 0.86
had filthy habits.
4 S2 PRR Mm. 1 0.86 0.86
→ 5 S1 PNS They were terrible! And in fact, I think 20 0.90 0.91
they only la they didn’t last a week. They
were sent
6 S2 PRR Mm. 1 0.90 0.83
→ 7 S1 PNS back. 1 0.00 -
8 S2 PRR Yeah. 1 0.00 -
CNF
9 S1 PNS I’d like to have seen that. 7 0.75 0.75
What the transcription presents as if they were two distinct utterances (the
arrowed utterances 5 and 7) are obviously just a single turn to which the
recipient’s Mm in turn 6 is a response occurring simultaneously with sent (in
turn 5) and back (in turn 7). The turn recorded as turn 7 thus constitutes a
‘fake.’ Its intrusion in the transcription has serious consequences for the
TSF coefficients. Compare the medians: the median is 0.65 with the fake
included as a turn in its own right but 0.82 if the overlap is corrected (that is,
if the ‘stranded’ back is counted as the 21st word in utterance 5 and the
recipient’s Mm and Yeah are drawn together as constituting one turn). The
difference in median is noticeable. Since overlapping speech is “very common
in natural, informal conversation” (Crowdy 1994: 26), its cumulative effects on
TSF coefficients are certainly not negligible. However, lacking evidence to
suggest otherwise, overlap will occur equally often across the three textual
components and the decreasing effect it has on TSF coefficients will thus be
the same for them.
258 Notes to pages 98–112
13. To retrieve turn lengths for two-party narratives, this XQuery was used:
14. The number of coefficients is always the number of turns in a story minus 1. This
is, as noted, because the coefficients are calculated on the basis of the lengths of
two adjacent turns (turn bigrams).
15. It may be noted that the TSF medians for PRC (0.52) and PNC (0.50) are lower
than the TSF medians for the non-narrative components CPR (0.60) and CPO
(0.64). This is because in the present analysis only coefficients for turn-bigrams in
which the narrator’s utterance comes first and the recipient’s response comes
second were included; coefficients for turn pairs in which the recipient’s response
is first and the narrator’s utterance is second were not taken into account.
16. The value of 0.3 as the median TSF – if turn size is not predetermined – is
suggested by a simulation in R. In the code below, random turn lengths varying
between 0 words (for non-verbal ‘turns’) and 300 words (the longest turn in the
NC) are generated for a two-party conversation with 1,000 turns taken. The
rounded median TSF of that conversation is 0.3:
categories, implicate the narrator’s pretense not to interfere with the discourse but to
“leave the characters to talk entirely on their own” (Leech & Short 1981: 324).
2. The third internal speaker role is ‘principal’: “someone whose position is being
established” (Goffman 1981: 144).
3. Indeed, Golato (2002) cites research suggesting that, like in other-quotation,
whose non-verbatimness has been sufficiently demonstrated empirically (e.g.,
Wade & Clark 1993), self-quotation too is subject to selective depiction and thus
far from faithful to the ‘original’ speech.
4. Note however that mimicry, which seems to occur in Direct and, particularly,
Free Direct discourse presentation only (Mathis & Yule 1994, Stenström et al.
2002), may not only bring the reported utterance closer to the recipient but also
signal ‘quotative distance’ (Stenström et al. 2002: 110) on the part of the mimicking
speaker toward the mimicked person. Culpeper (2011), for example, discusses
mimicry as an impoliteness strategy on the grounds that it can be seen as “a
caricatured re-presentation” (Culpeper 2011: 161) or ‘echo’ of anterior discourse,
“reflect[ing] the negative attitude of the echoer towards the echoed person”
(Culpeper 2011: 165). In these cases, then, the immediacy of the report is relativ-
ized. See also Holt (2000), who stresses the ‘duality’ of dialog, which can “convey
both the attitude of the reported speaker and, more implicitly, the attitude of the
current speaker” (Holt 2000: 438).
5. As noted earlier, the quote (i.e., the reported clause) can be seen as the direct object
of the reporting clause (Quirk et al. 1985: 1022–1023).
6. Because MSS presents (what the presenter has taken as) the utterance’s illocu-
tionary force, the recipient of the report is enabled to form an understanding of
what was communicated and, given this understanding, to imagine locutions that
convey the presented illocution. For example, in [MSS Nicola did n’t believe me
when I told her.], a recipient can easily construct potential locutions for Nicola’s
illocution: for example, I don’t believe you/I can’t believe you/ That’s unbelievable
and so forth. That is, based on the reported illocution, the recipient can retrieve,
not the actual, but possible locutions.
7. Brown & Levinson (1987) cite direct speech as a positive politeness technique
operative in the service of “Strategy 3: Intensify interest to H [hearer]” (1987: 106).
8. In Clark & Gerrig’s (1990) demonstration theory, the difference in use of reference
is related to the difference in ‘seriousness’: non-direct reporting is ‘serious’ in the
sense that it means what it encodes. For instance, to use Clark & Gerrig’s (1990:
764) example (2), the Indirect report “she tells him that uh she wants to buy an
ant” means the female referent informed a male referent that she wishes to
purchase an ant. By contrast, a Direct speech report like “she says ‘well I’d like
to buy an ant’,” Clark & Gerrig’s (1990: 764) example (1), is ‘nonserious’ in the
sense that the reporting speaker “isn’t ‘really or actually or literally’ making a
request” (Clark & Gerrig 1990: 770).
9. Clark & Gerrig (1990) report an experiment in which subjects were asked to
memorize a brief scene from Breakfast at Tiffany’s until they could recite it by
heart. One group were asked to reproduce the scene ‘as accurately as possible’; here,
the verbatim reproduction rate was 99%. Another group were asked to ‘simply tell
the story of the scene’ to a partner; there, the subjects produced verbatim quotations
only 62% of the time (Clark & Gerrig 1990: 797); for more experiments providing
support for demonstration theory, cf. Wade & Clark (1993).
260 Notes to pages 124–126
10. See Ameka’s (1992) discussion of ‘free’ uses of and: he argues that and, when used
as a stand-alone item, “usually carries a rising intonation signalling that it is an
incomplete utterance” (1992: 105).
11. It has been noted in the literature that interjections are typical of direct quotation
but untypical of (or even ungrammatical in) indirect or any other form of non-
direct reporting (e.g., Mayes 1990). This observation is fully borne out by the data
in the NC. The following are exhaustive lists of interjections occurring in (Free)
Direct quotes and all other report units:
(Free) Direct units All other report units
For a start, the two lists differ markedly in terms of length. The list for (Free)
Direct comprises 35 types and 492 tokens, the list of interjections in all other
modes contains merely 7 types and 34 tokens. Moreover, the two lists differ in
terms of the kind of interjections included. The list for (Free) Direct is headed by
oh with 199 occurrences and other conventionalized forms such as no and yeah.
However, more characteristically it seems, there are a great number of much less
conventionalized forms such as ee, aargh, gee, wahey, whoa, and so on. The
typological range of interjections in non-direct units is much more restricted:
the list mostly contains conventionalized forms, perhaps the only exception being
tt. In sum, the contrast between the two frequency lists could not be greater: Are
these differences significant? This question can be answered easily by use of the
NC keyness analysis tool. The results are unambiguous: key POS analyses for ITJ
in (Free) Direct compared (i) to (Free) Indirect and (ii) to all report types except
the direct categories show the tag ITJ consistently ranked first, that is, as the POS
tag most key; the p-values are far below 0.001 (note the loglikelihood values for
keyness which are far above the critical value of 10.83):
Key POS tag in (Free) Direct vs (Free) Indirect (p < 0.001; critical value: 10.83)
Rank Item Freq. (Free) Direct Freq. (Free) Indirect Keyness (LL)
1 ITJ 492 9 130.042
Key POS tag in (Free) Direct vs all remaining report mode types (p < 0.001;
critical value: 10.83)
Rank Item Freq. (Free) Direct Freq. all other report types Keyness (LL)
1 ITJ 492 34 216.241
12. The underlying formula to calculate positions is not without problems. For example,
according to the formula, the second word in a three-word utterance has position 1/3
= 0.33, whereas the second word in a six-word utterance has position 1/6 = 0.17.
That is, positions of words used within utterances (or any other unit such as quotes)
are relative to, and hence dependent on, the length of the utterances. However,
speakers hardly ever know in advance how many words long their utterances will be.
Therefore, as far as within-utterance positions are concerned, it will be hard to make
any large claims as to speakers being ‘primed’ to use words in particular positions.
Fortunately, this caveat need not worry us in the present connection where the focus
is on first position, which is invariably 0 whatever the length of the utterance.
13. To compute positions of interjections the following XQueries were used:
for conversational utterances longer than 1 word (n>1):
for $u in //seg[starts-with(@Components,’CP’)]//u[count(descendant::w)
>1 and descendant::w[@c5=‘ITJ’]]
let $n_w_utt := count($u//w)
for $itj in $u//w[@c5=‘ITJ’]
let $n_w_prec_itj := count($itj/preceding::w[ancestor::u is $u])
let $pos := $n_w_prec_itj div $n_w_utt
order by $pos
return
concat($pos,’ ‘,normalize-space(string-join($itj//w/text(),’ ‘)))
262 Notes to pages 129–133
14. An obvious question arising at this point is the question why interjections are so
strongly attracted to first position in utterances. For space limitations, pursuing
the question in sufficient detail is far beyond my aims. A few notes must suffice
here. First, it may be helpful to consider that interjections, being inserts (cf. Biber
et al. 1999), do not enter into syntactic constructions with other words. Given
their syntactic independence, interjections are an obvious choice for filling the
first slot in talk (cf. Tao 2003). Further, drawing on Tao’s (2003) observation that
the definite article the, which is among “the most frequent initial items in written
discourse” (Norrick 2009: 870), seems to avoid first position in utterances, Hoey
suspects that the high use of interjections in first position serves “to avoid the
appearance of (over-)definiteness that presumably fronting an utterance with
the might be interpreted as implying” (Hoey, personal email communication).
Finally, McCarthy (2003: 38) sees the utterance-first position “as the locus of
choice where speakers frequently select items that contribute to the non-
transactional stratum of talk.” In this view, the utterance-first slot, as the juncture
between two utterances, has an interpersonal focus, tying not only contributions
but also contributors to one another.
15. Further analysis shows that the absence of mm from inclusion in quotes is not only
restricted to occurrences in quote-first position but applies much more generally,
viz. to other positions and also to quotes consisting of one word only. Its absence
from quotation is thus virtually total.
16. Despite their low frequencies, the remaining eight items were examined statisti-
cally. It turned out that the differences in frequency of occurrence in first position
in quotes as compared to utterances are insignificant for yeah, no, yes, hello, ha,
and blah. The results for ah defy interpretation. While the p-value 0.0452
obtained from a Fisher’s exact test is still smaller than 0.05, suggesting a
significant association, the confidence interval poses a problem: not only does it
contain 1 (which it should not if the result were indeed significant) but it also
extends from 0.8 to infinity (a fact which is due to the 0 in the 2x2 table for ah).
This is obviously far too large an interval to be confident about ah’s status in first
position and to allow any definitive statement regarding significance. As regards
the item ooh, this interjection is significantly differently distributed in first
position in quotes and utterances: p-value = 0.003353, Cramer’s V = 0.3695652.
Consider:
Notes to pages 135–138 263
I know but he was so funny, for a dog to be that perceptive, do you see what I
mean? As soon as I came downstairs, [MDF Ooh God, I shouldn’t be in here ]
[laughing ] and out he jumped ! (KD5-N2)
Despite its apparent significance, no claims will be made regarding ooh on the
grounds that this item may not be seen as an independent interjection but rather
merely a phonologically lengthened variant of oh.
17. A third possible interpretation is that “quoters use oh and well more frequently or
in a wider range of ways because they believe that their use at the beginning of
quoted speech makes the quotation sound authentic” (Hoey, personal email
communication). This interpretation ties in well with the earlier proposed notion
of the conversational utterance as the model for constructed dialog: in construct-
ing the dialog, reporters design it in such a way that it maximally ‘sounds’ like real
speech. The weakness to this view, however, is that it does not explain the overuse
of oh and well in quotes. Only if the two items were used equally frequently in
utterances and in quotes, as is the case for the interjection yeah, would this view
have to be preferred over the two other interpretations discussed.
18. Yet another interpretation is worth considering: it could be argued that well and oh,
because of their unusual frequency, cannot be seen as part of the quote but rather as
discourse markers outside the quote, for example, as part of the reporting clause
which introduces the quote (cf. James 1983; for a brief discussion of the issue see
Jucker 1993). To support this view one could refer to the unusual frequency of the
two items: they are overrepresented in comparison to the utterances they are meant
to represent, hence they must constitute something else – something outside the
quote. There are two weaknesses to this view. First, it builds on the verbatim
assumption, which holds that quotation attempts to faithfully replicate locution, a
view which is, as noted repeatedly, untenable for cognitive (speakers’ memory
limitations) and functional (speakers’ selectiveness in demonstrating) reasons.
Second, if it is assumed that well and oh are overrepresented and therefore not
part of the quotes themselves it follows statistically that they are underrepresented
within quotes. To illustrate, if oh is taken as a component of the quote, as in the
analysis presented above, its frequency of occurrence in first position is 168, which
accounts for 90% of all occurrences of oh within quotes (the proportion of
occurrence of oh in utterance-first position was 74%, that is, significantly less).
Conversely, if oh in quote-first position is discounted as not belonging to the quote,
we are left with a total of 18 occurrences of oh within quotes. It turns out upon
inspection that 5 occurrences are in what has become the new first position (which
is, according to the ‘old’ count, exactly the second word in the quote). The
percentage for oh in that position is merely 5/18*100 = 27%. In comparison to
the proportion of 74% with which oh occurs in utterance-first position this is a
dramatic drop, which is very highly significant: X-squared = 16.339, df = 1,
p-value = 5.296e-05. An even more drastic discrepancy can be observed if the
same recalculation is performed for well: X-squared = 58.2926, df = 1, p-value =
2.259e-14. The discrepancies would call for an explanation: why should speakers so
blatantly underuse oh and well at the onset of quotes as compared to utterances? It
is hard to imagine a plausible answer to this question. By underusing oh and well,
which, in conversational utterances, are, as demonstrated, massively attracted to
first position, narrators would fail to provide recipients with important clues as to
the forthcoming quotation. Given the importance of recipient design governing
264 Notes to pages 139–146
20. XQuery used to retrieve positions of quote-first oh and well relative to their
enclosing narrative utterance:
for $u in //seg[@Components,’CNN’]//u[count(descendant::w)>1 and
count(descendant::seg[starts-with(@Reporting_modes,’MD’)])=1 and
count(descendant::seg[starts-with(@Reporting_modes,’MD’)]//w)>1 and
descendant::w[@c5=‘ITJ’]]
let $n_w_utt := count($u//w)
let $quote := $u//seg[starts-with(@Reporting_modes,’MD’)]
for $itj in $quote//w[1][@c5=‘ITJ’]
let $n_w_prec_itj := count($itj/preceding::w[ancestor::u is $u])
let $pos := $n_w_prec_itj div $n_w_utt
order by $pos
return
concat($pos,’ ‘,normalize-space(string-join($itj//w/text(),’ ‘)))
21. The fact that silent pauses are far more numerous than er and erm is not the only
reason why the analysis foregrounds silent pauses. There are two more reasons.
First, the distinction between the filled pauses er/erm and silent pauses is not
negligible in that er and erm sit somewhere at the boundary between words and
non-words, while silent pauses are decidedly non-verbal. Secondly, in terms of
annotation, er and erm are treated as <w> elements, while occurrences of
<pause/> are given an XML tag of their own, which is a ‘sibling’ to <w>
elements (cf. Section 2.3.2). This annotational difference has consequences for
the calculation of positions: for example, because silent pauses are not counted as
words, they can effectively obtain the value 1, while er and erm occurring within
the quote can not, even if they are in quote-ultimate position. So, the positional
values for er/erm and silent pauses would not be easily comparable.
22. In Rühlemann et al. (2011), long pauses were found to occur less frequently in
narrative components (CNN) than in non-narrative components (CPR and CPO).
They are thus likely to play a less central role in narrative than in non-narrative
conversation, where they seem to be closely associated with extra-linguistic action.
23. XQueries used to retrieve positions of filled pauses:
As can be seen in the table, the value 0.76 (Case [10,] in the original subsample) is
not included in Resamples 2 and 3, whereas the value 0.62 (Case [8,] in the
subsample) is found twice in Resamples 1 and 3. Given these different values in the
subsample and the resamples it is not surprising that both the means and the standard
deviations are different too – in fact, they are quite different: the mean of Resample 1,
for example, is 0.54, while the mean for Resample 4 is only 0.21. Similar differences
can be seen for the standard deviations. Obviously, the differences are so stark due to
the extremely small sample size of 10; they will decrease with increasing sample size.
27. On the other hand, the peak for pauses in final positions is less marked than the
peak for initial positions. This might be taken as evidence in defense of the
assumption that pauses primarily serve to buy the narrator planning time: it
might be seen as reflecting the lesser processing costs of re-instantiating one’s
own reference and voice system.
in the case of n=3-sequences, are 827 (in pos1), 901 (in pos2), and 900 (in pos3).
The column totals and maximal values (‘max.’ in Table 5.5) are the following:
Table 5.5
pos1 pos2 pos3 pos4 pos5 pos6 pos7 pos8 max
4. Another methodology would be to single out data sets for manual analysis. Take for
example the data set for n = 4. This data set can be processed in the following way.
First, frequency lists for all report unit types occurring in the set as a whole are
generated. Second, frequency lists for all report unit types per position (that is, per
column) are computed. Finally, the percentage values are computed for the report
unit types per position as proportions of the total frequencies. To illustrate, the tag
MII (Indirect) occurred 20 times in n=4-sequences as a whole while it occurred 8
times in pos1 (first position). Hence, its proportion of occurrence in that position is 8/
20*100 = 40%. The figure below represents these proportions for all tags with total
frequencies greater than 10.
60.000
50.000
40.000 pos1
pos2
%
30.000
20.000 pos3
10.000 pos4
.000
MDD MDF MII MSS MVV
This XQuery instructs the processor to loop over stories (CNN) which meet two
conditions. First, the number of PRR is greater than 0; this constraint excludes stories
told by PNU (which by definition does not receive any recipient feedback). Second,
the number of PRC contributions is set to 0; this excludes stories in which both
PRR and PRC contributions occur. The query outputs two types of data: number of
words in narrator utterances per narrative ($w_narrator) and number n of PRR
contributions per narrative ($n_prr).
4. In the present analysis, the NB model is preferred over the Poisson model in that
the latter requires equidispersion – that is, “in the Poisson model, the variance of
the random variable is constrained to equal the mean” (Greene 2008: 585) – a
requirement not met here: the variances are greater than the means.
5. The interaction effect is brought out more clearly if the model with the interaction
is compared to what the model would look like without the interaction:
1000 1000
E(w_narrator)
E(w_narrator)
500 500
0 0 10
10
0 0
c
pr
c
5
pr
5
n_
5 5
n_
n_p n_p
r r r r
10 10
0 0
Notes to pages 187–197 269
The figure makes immediately obvious that if there were no interaction effect
between n_prr and n_prc, the growth in w_narrator would be much more
dramatic, as shown in the steeply rising plane in the plot on the right, than is
the case when the interaction is taken into account. Its effect can be seen in the
much flatter plane in the plot on the left. Note, however, that, for the sake of
comparability, the scale of the plot for the model with the interaction has been
changed to match the scale for the plot without the interaction. Note also that to
more correctly illustrate the interaction effect it would be necessary to juxtapose
Model 4 and Model 3 (for PRR + PRC without interaction); in that comparison,
however, the interaction effect would not become salient enough.
6. In van Dijk & Kintsch’s terms, the ‘textbase,’ which “represents the meaning of a
text in all its detail” (van Dijk & Kintsch 1983: 52), is expanded by the recipient.
7. Fox & Thompson (2010: 136) refer to this type of wh-question as ‘telling question’
(which, on Fox & Thompson’s account, contrasts with ‘specifying question’).
8. It will be noticed that in the narrative in (6.10), the first three consecutive turns
were marked as Primary Narrator (PNP) turns. This annotation is correct due to
the fact that the narrative is the second (embedLevel EC2) in a story round and
that it overlaps with the first (EC1): what appears as the second PNP turn in EC2
is in fact the concluding turn in EC1.
9. This is because instances of discourse presentation were marked up using <seg>
elements not only within <u>-elements (utterances) that have attribute values
for narrator roles PNP, PNC, etc. but also within <u>-elements with values for
recipient roles. So queries for discourse presentation segments contained within
<u>-elements that have the attribute value PRC can be conducted and exhaus-
tive quantitative analysis is possible.
10. The total number of PRC report units in the NC in its present version is 120.
Upon close inspection, however, for four occurrences of reporting the annotation
was found erroneous.
11. An example of Free Indirect discourse presentation by Co-constructive Recipient
not included in the NC is given in this excerpt from the BNC-C. The main
narrator, speaker PS18L, is relating a session at a fortune teller’s. When she
describes her skeptical thoughts about the weird looking room, a co-participant,
speaker PS18K, produces the thought presentation What are [you] letting yourself
in for?. This presentation is indirect: note the participant’s switch into the primary
narrator’s deictic system evidenced by the reflexive pronoun yourself:
PS18L >: Exactly! We went into this room and there was this music playing like
on the erm . . . like John
PS18E >: Indian music.
PS18L >: would play on the
PS18E >: Oh no.
PS18L >: on erm
PS18K >: Like in fantasy
PS18L >: Yeah.
PS18E >: John’s sort of thing.
PS18L >: You know water [ . . . ] . . . a blue light bulb in the middle of the room
PS18E >: [laugh]
PS18L >: and I’m thinking, Oh!
270 Notes to pages 198–204
PS18E >: [laugh]
PS18L >: Come on!
PS18K >: What are [yoo] letting yourself in for?
PS18L >: You know. But it was just . . . I mean, the things that they said . . .
honestly . . . weird.
(KBU)
12. The p-values reported in this paragraph were obtained from chi-squared tests.
13. Note that epistemic modality (the level of reliability of the speaker’s knowledge) is
included in Clift’s (2006) notion of evidentiality, whereas it is excluded, for example,
in Aikhenvald’s (2004: 3) definition of evidentiality as “a linguistic category whose
primary meaning is source of information.” Although the more inclusive notion is
prone to “conceptual and terminological confusion” (Aikhenvald 2004: 5), it has a
wide currency and will therefore be adopted in the present analysis.
14. In this case, then, the sequential placement of the recipient quote after the
narrator’s implicit assessment would make an interesting contrast with the sequen-
tial order observed by Holt (2000), which is the converse in that, there, tellers first
produce a quote to which recipients then offer an assessment (Holt 2000: 450).
15. Extracts (6.23) and (6.24) are counterexamples in that here the recipient mimics
an utterance by a speaker who is not co-present in the telling situation. My sense
is that, in assuming the character’s role, the recipient does not affiliate with the
character but rather takes the dissociative attitude of mockery. The quote thus
serves, not identification, but, via ridicule, dissociation.
16. ‘Laughter’, here, refers to both (i) between-speech laughter, coded as [laugh] and
tagged, in the BNC and the NC, as a <vocal>element, and (ii) within-speech
laughter, coded as laughing voice quality and tagged as a <shift>element. The
two types have a combined frequency of 82 occurrences in 26 narratives contain-
ing recipient dialog, whereas, in the 505 narratives without recipient dialog, there
are 524 instances of laughter. According to Fisher’s exact test, this difference is
very highly significant: p-value = 4.8e-07 (odds ratio = 0.33; CI: 0.20 - 0.53).
17. To retrieve positional data for quotes by recipients (PR*) and, respectively,
narrators (PN*):
for $seg in //seg[@Components=‘CNN’ and descendant::seg[starts-
with(@Reporting_modes,‘MD’)] and
not(ancestor::div[@embedLevel=‘EN’])]
let $n_quotes := count($seg//seg[starts-with(@Reporting_modes,‘MD’)])
let $n_words_all := count($seg//w)
let $n_words_inside_quotes := count($seg//w[ancestor::seg[starts-
with(@Reporting_modes,‘MD’)]])
let $n_words_outside_quotes := $n_words_all - $n_words_inside_quotes
let $title := //div[@title and child::seg[.=$seg]]
for $targetquote in $seg//seg[starts-with(@Reporting_modes,‘MD’)]
[ancestor::u[starts-with(@Participation_roles,‘PR’)]]
let $n_words_preceding_targetquote :=
count($targetquote/preceding::w[ancestor::u[ancestor::seg[.=$seg]]])
let $n_quotes_preceding_targetquote := count($targetquote/preceding::seg
[starts-
with(@Reporting_modes,‘MD’) and ancestor::u[ancestor::seg[.=$seg]]])
Notes to page 205 271
Lengths of stories
1000
800
Total number of words
600
400
200
0
median: 0.85
IQR: 0.1
(KB2-N2)
274
References 275
2006b. “The other side of the story: towards a narrative analysis of narrative-in-
interaction,” Discourse Studies 8(2): 265–287
2008. “On MNS with buff boys. Self and other-identities in the context of small
stories,” Journal of Sociolinguistics 12: 597–626
Giles, H., J. Coupland, and N. Coupland (eds.) 1991. Contexts of accommodation.
Cambridge University Press
Gilquin, G. 2002. “Automatic retrieval of syntactic structures. The quest for the holy
grail,” International Journal of Corpus Linguistics 7(2): 183–214
Glenn, P. 2003. Laughter in interaction. Cambridge University Press
Goffman, Ervin. 1981. Forms of talk. Philadelphia University Press
Golato, A. 2000. “An innovative German quotative for reporting embodied actions:
Und ich so/und er so ‘and I’m like/and he’s like’,” Journal of Pragmatics 32:
29–54
2002. “Self-quotation in German: Reporting on past decisions,” in T. Güldemann
and M. von Roncador (eds.) Reported discourse: A meeting ground for different
linguistic domains. Amsterdam: John Benjamins, pp. 49–70
Goodwin, C. 1979. “The interactive construction of a sentence in natural conversa-
tion,” in G. Psathas (ed.) Everyday language – Studies in ethnomethodology. New
York: Irvington Publishers, pp. 97–121
1986a. “Between and within alternative sequential treatments of continuers and
assessments,” Human Studies 9, 205–217
1986b. “Audience diversity, participation and diversity,” Text 6(3): 283–316
2007. “Interactive footing,” in E. Holt and R. Clift (eds.) Reporting talk. Reported
speech in interaction. Cambridge University Press, pp. 16–46
Goodwin, C. and J. Heritage. 1990. “Conversation analysis,” Annual Review of
Anthropology 19: 283–307
Greene, W. 2008. “Functional forms for the negative binomial model for count data,”
Economics Letters 99: 585–590
Gries, S. Th. 2006. “Exploring variability within and between corpora: some meth-
odological considerations,” Corpora 1(2): 109–151
2009a. Quantitative corpus linguistics with R. A practical introduction. New York/
London: Routledge
2009b. Statistics for linguistics with R. A practical introduction. Berlin: Mouton de
Gruyter
Gwyn, R. 2000. “‘Really unreal’: Narrative evaluation and the objectification of
experience,” Narrative Inquiry 10(2): 313–340
Halliday, M. A. K. 2006. “Afterwords,” in G. Thompson and S. Hunston (eds.) System
and corpus: Exploring connections. London/Oakville: Equinox, pp. 293–299
Halliday, M. A. K. and R. Hasan. 1989. Language, Context, and Text: Aspects of Language
in a Social-semiotic Perspective. Oxford University Press
Halliday, M. A. K. and M. I. M. Matthiessen. 3rd edn. 2004. An Introduction to
Functional grammar. London: Edward Arnold
Hanks, W. F. 1992. “The indexical ground of deictic reference,” in A. Duranti and
C. Goodwin (eds.) Rethinking context: Language as an interactive phenomenon.
Cambridge University Press, pp. 43–76
2011. “Deixis and indexicality,” in W. Bublitz and N. R. Norrick (eds.) Handbook
of pragmatics. Vol. 1 Foundations of pragmatics. Berlin: Mouton de Gruyter,
pp. 315–346
References 279
2006. “On the human interactional engine,” in N. Enfield and S. C. Levinson (eds.)
Roots of human sociality: Culture, cognition and interaction. Oxford: Berg, pp. 39–69
Li, C. L. 1986. “Direct and indirect speech: A functional study,” in F. Coulmas (ed.)
Direct and indirect speech. Berlin: Mouton de Gruyter, pp. 29–45
Lijffijt, J., P. Papapetrou, K. Puolamäki and H. Mannila. 2011. “Analyzing word
frequencies in large text corpora using inter-arrival times and bootstrapping,” in
Proceedings of the European Conference of Machine Learning and Principles
and Practices of Knowledge Discovery in Databases ECML-PKDD 2011,
available at: http://users.ics.tkk.fi/panagpap/lijffijt_157.pdf
Longacre, R. E. 1983. The grammar of discourse. New York: Plenum Press
Lyons, J. 1977. Semantics. Volumes I and II. Cambridge University Press
Macaulay, R. 1987. “Polyphonic monologues: Quoted direct speech in oral narra-
tives,” IPRA Papers in Pragmatics 1: 1–34
2001. “You’re like ‘why not?’ The quotative expressions of Glasgow adolescents,”
Journal of Sociolinguistics 5(1): 3–21
Mahlberg, M. and M. B. O’Donnell. 2008. “A fresh view of the structure of hard
news stories,” in Proceedings of the 19th European Systemic Functional
Linguistics Conference and Workshop. http://scidok.sulb.uni-saarland.de/
volltexte/2008/1700
Mathis, T. and G. Yule. 1994. “Zero quotatives,” Discourse Processes 18(1): 63–76
Mayes, P. 1990. “Quotation in spoken English,” Studies in Language 14: 325–363
McCarthy, M. J. 1998. Spoken language and applied linguistics. Cambridge University Press
2003. “Talking back: ‘Small’ interactional response tokens in everyday conversa-
tion,” Research on Language and Social Interaction 36(1): 33–63
McEnery, T. and A. Wilson. 1996. Corpus linguistics. Edinburgh University Press
McIntyre, D. and B. Walker. 2011. “Discourse presentation in Early Modern English
writing. A preliminary corpus-based investigation,” International Journal of
Corpus Linguistics 16(1): 101–130
McIntyre, D., C. Bellard Thomson, J. Heywood, T. McEnery, E. Semino, and
M. Short. 2004. “Investigating the presentation of speech, writing and thought
in spoken British English: A corpus-based approach,” ICAME Journal 28: 49–76
Miller, J. and R. Weinert. 1998. Spontaneous spoken language: Syntax and discourse.
Oxford: Clarendon Press
Mindt, D. 2000. An empirical grammar of the English verb system. Berlin: Cornelsen
Minsky, M. 1975. “A framework for representing knowledge,” in P. H. Winston (ed.)
The psychology of computer vision. New York: McGraw-Hill, pp. 211–277
Mittal, H. V. 2011. R graphs cookbook. Birmingham/Mumbai: PACKT Publishing
Mooney, C. Z., and R. D. Duval. 1993. Bootstrapping. A nonparametric approach to
statistical inference. Newbury Park, CA: Sage Publications
Morley, J. and A. Partington. 2009. “A few Frequently Asked Questions about semantic –
or evaluative – prosody,” International Journal of Corpus Linguistics 14(2): 139–158
Nelson, G., S. Wallis, and B. Aarts. 2002. Exploring natural language: Working with
the British component of the International Corpus of English. Amsterdam/
Philadelphia: John Benjamins
Norrick, N. R. 1994. “Involvement and joking in conversation,” Journal of Pragmatics
22: 409–430
2000. Conversational narrative. Storytelling in everyday talk. Amsterdam/
Philadelphia: John Benjamins
282 References
1997. “‘Narrative Analysis’ thirty years later,” Journal of Narrative Inquiry and Life
History 7(1–4): 97–106
Schegloff, E. A. and G. H. Lerner. 2009. “Beginning to respond: Well-prefaced
responses to wh-questions,” Research on Language and Social Interaction 42(2):
91–115
Schiffrin, D. 1984. “How a story says what it means and does,” Text 4(4): 313–346
1985. “Conversational coherence: the role of well,” Language 61: 640–667
1987. Discourse markers. Cambridge University Press
1996. “Narrative as selfportrait: Sociolinguistic construction of identity,” Language
in Society 25(1): 167–203
Schmid, H.-J. and H. Küchenhoff. 2008. “Investigating style. Lexico-grammatical
corpus evidence for the socio-situational construction of discourse identities,”
paper read at the First Conference of the International Society of the Linguistics
of English (ISLE), Freiburg, Germany
Scott, M. 1997. “PC analysis of key words – and key key words,” System 25(2): 233–245
2010. WordSmith Tools version 5. Liverpool: Lexical Analysis Software
Scott, M. and C. Tribble. 2006. Textual patterns. Key words and corpus analysis in
language education. Amsterdam/New York: John Benjamins
Searle, J. R. 1975. “The logical status of fictional discourse,” New Literary History
6(2): 320–332
Selting, M. 2000. “The construction of units in conversational talk,” Language in
Society 29(4): 477–517
Semino, E. and M. Short. 2004. Corpus stylistics: Speech, writing and thought pre-
sentation in a corpus of English writing. London: Routledge
Shanon, B. and R. Eiferman. 1984. “Dream reporting discourse,” Text 4(4): 369–379
Short, M. 2007. “Thought presentation 25 years on,” Style 41(2): 227–257
Short, M., E. Semino, and J. Culpeper. 1996. “Using a corpus for stylistics research:
speech and thought presentation,” in J. Thomas and M. Short (eds.) Using
corpora for language research. Studies in honour of Geoffrey Leech. London/New
York: Longman, pp. 110–131
Sinclair, J. McH. 1991. Corpus, concordance, collocation. Oxford University Press
Sperber, D. and D. Wilson. 1986. Relevance: Communication and cognition. Oxford:
Blackwell.
Stefanowitsch, A. 2005. “New York, Dayton, Ohio, and the raw frequency fallacy,”
Corpus Linguistics and Linguistic Theory 1(2): 295–301
Stefanowitsch, A. and S. Th. Gries. 2003. “Collostructions. Investigating the inter-
action between words and constructions,” International Journal of Corpus
Linguistics 8(2): 209–243
Stenström, A.-B. 1990. “Pauses in monologue and dialogue,” in J. Svartvik (ed.) The
London-Lund Corpus of Spoken English. Lund University Press, pp. 211–252
Stenström, A.-B., G. Andersen and I. K. Hasund. 2002. Trends in teenage talk.
Amsterdam/Philadelphia: John Benjamins
Stenström, A. and J. Svartvik. 1994. “Imparsable speech: Repeats and other non-
fluencies in spoken English,” in N. Oostdijk, N. and P. de Hahn (eds.) Corpus-
based research into language. In honour of Jan Aarts. Amsterdam/Atlanta, GA:
Rodopi, 241–254
Stivers, T. and F. Rossano. 2010. “Mobilizing response,” Research on Language and
Social Interaction 43(1): 3–31
References 285
Stubbs, M. 2002. Words and phrases. Corpus studies of lexical semantics. Oxford:
Blackwell
Svartvik, J. 1980. “Well in conversation,” in S. Greenbaum, G. Leech, and J. Svartvik
(eds.) Studies in English linguistics. For Randolph Quirk. London/New York:
Longman, pp. 167–177
Swales, J. M. 1990. Genre analysis. English in academic and research settings. Cambridge
University Press
Tagliamonte, S. and A. D’Arcy. 2004 “He’s like, she’s like: The quotative system in
Canadian youth,” Journal of Sociolinguistics 8(4): 493–514
Tagliamonte, S. and R. Hudson. 1999 “Be like et al. beyond America: The quotative
system in British and Canadian youth,” Journal of Sociolinguistics 3(2): 147–172
Tannen, D. 1978. “The effect of expectations on conversation,” Discourse Processes 1:
203–209
1986. “Introducing constructed dialog in Greek and American conversational and
literary narrative,” in F. Coulmas (ed.) Direct and indirect speech. Berlin/New
York/Amsterdam: Mouton de Gruyter, pp. 311–332
1988. “Hearing voices in conversation, fiction and mixed genres,” in D. Tannen
(ed.) Linguistics in context: Connecting observation and understanding. Norwood,
NJ: Ablex, pp. 89–113
1989. Talking voices: Repetition, dialogue, and imagery in conversational discourse.
Cambridge University Press
1990. You just don’t understand. New York: William Morrow
(ed.) 1993. Gender and conversational interaction. New York/Oxford: Oxford
University Press
Tao, H. (2003). “Turn initiators in spoken English: A corpus-based approach to
interaction and grammar,” in P. Leistyna and C. F. Meyer (eds.) Corpus analysis:
Language structure and language use. Amsterdam: Rodopi, pp. 187–207
Thorndyke, P. W. 1977. “Cognitive structures in comprehension and memory of
narrative discourse,” Cognitive Psychology 9: 77–110
Tognini Bonelli, E. 2010. “The evolution of corpus linguistics,” in A. O’Keeffe and
M. McCarthy (eds.) The Routledge handbook of corpus linguistics, London/New
York: Routledge, pp. 14–27
Toolan, M. 2nd edn 2001. Narrative. A critical linguistic introduction. Oxford/New
York: Routledge
Upton, Th. A. and U. Connor. 2001. “Using computerized corpus analysis to
investigate the textlinguistic discourse moves of a genre,” English for Specific
Purposes 20: 313–329
Wade, E. and H. H. Clark. 1993. “Reproduction and demonstration in quotations,”
Journal of Memory and Language, 32(6), 805–819
Walmsley, P. 2007. XQuery. Sebastopol, CA: O’Reilly
Walton, M. D. and C. L. Brewer. 2001. “The role of personal narrative in bringing
children into the moral discourse of their culture,” Narrative Inquiry 11(2):
307–334
Watt, A. 2002. XPath essentials. New York: John Wiley
Weisser, M. 2010. Annotating dialog corpora semi-automatically: A corpus-linguistic
approach to pragmatics. Unpublished professorial thesis, University of Bayreuth
Forthcoming. “Speech act annotation,” in K. Aijmer and C. Rühlemann (eds.) The
Cambridge handbook of corpus pragmatics. Cambridge University Press
286 References
287
288 Index
mimesis 118 reference 117
mimicry 222, 259 response-turn order
multi-turn unit 78 double-responses 80
multi-unit turn 76 single responses 219
mutual orientation 90 self-quotation 259
narrative clause 15, 43, 208, 247 single teller 177
by recipient 17 small stories 1, 37
co-constructed 187 social balance 40
narrative sections 27 sociolinguistic interviews 1, 6, 26
narrative speech act 92 speech accommodation 227
narrative structure asymmetrical 227
chapterlike divisions 30, 210 partial 227
climacto-telic structure 25, 168, 170 spoken paragraphs 142
evaluation section 25 standard deviation 204
multiple-highpoint story 30 story boundaries 49
narrative sections 24 story climax See story highpoint
nested stories 28 story grammars 24
recycles 27 story highpoint 247
single-highpoint stories 207 story rounds 52
story highpoint 13, 25, 202, 207 structural analysis 10
story preface 25 tellability 25
structural centrality 32 tests
structural models 24 1-sample proportions test 86, 87
tension 25, 30, 214 bootstrap 88, 91, 148, 149, 206, 207
narrative subgenres 6 chi-squared test for given probabilities 85
narratorial interference 112 Fisher’s exact test 262
near-death experience 25 Kendall’s correlation test 87
N-notN-N decay 89 Kolmogorov-Smirnov test 205
origo 117 negative binomial (NB) model 182
oscillation of reference 119, 122 test for trend in proportions 163
participation framework 9 Wilcoxon rank sum test 100
perspective 117 theatricalization 118
population 64 transition-relevance point (TRP) 77
positional scope 145 turn distribution 80
prayer 37, 246 turn n-grams 63
pretense 14, 112, 259 turn bigrams 95, 96
processibility 119 turn trigrams 79, 80
processing constraints 151 turn order patterns 81
production format 112 turn size gap 109
animator 121 turn-constructional unit (TCU) 76
animator-only 112 turntaking 78
author 121 unquote 123
author-animator 112 utterance completions 194
principal 259 utterance model 119
proportions of narrative in conversation 252 utterance split 248
query languages utterance structure 175
XPath 68 utterance tension 171
XQuery 73 utterance vs. turn 77
quotation as evidential 119, 199 verbatim reproduction 118
quote-climax-quote (QCQ) triangle 203 virtual participation 13, 37, 38, 116
R 66, 148, 253, 254, 272 XML
R path 82 document structure 69
R simulation 82 element structure 68
recipe 22, 23 overlap 45
recipient model 31 Text Encoding Initiative
reciprocal listenership 226 (TEI) 45
recycles 211 XPath axes 70