Clause Combining in Grammar and Discourse

CLAUSE COMBINING IN GRAMMAR AND DISCOURSE
TYPOLOGICAL STUDIES IN L A N G U A G E (TSL)
A companion series to the journal "STUDIES IN L A N G U A G E "
Honorary Editor: Joseph H. Greenberg

General Editor: T. Givón
Editorial Board:
Alton Becker (Michigan) Paul Hopper (Binghamton)
Wallace Chafe (Santa Barbara) Margaret Langdon (San Diego)
Bernard Comrie (Los Angeles) Charles Li (Santa Barbara)
Scott DeLancey (Oregon) Johanna Nichols (Berkeley)
Gerard Diffloth (Chicago) Andrew Pawley (Auckland)
R.M.W.Dixon (Canberra) Frans Plank (Konstanz)
John Haiman (Winnipeg) Gillian Sankoff (Philadelphia)
Kenneth Hale (Cambridge, Mass.) Dan Slobin (Berkeley)
Bernd Heine (Köln) Sandra Thompson (Santa Barbara)
Volumes in this series will be functionally and typologically oriented, cove-

ring specific topics in language by collecting together data from a wide variety
of languages and language typologies. The orientation of the volumes will be
substantive rather than formal, with the aim of investigating universals f
human language via as broadly defined a data base as possible, leaning
toward cross-linguistic, diachronic, developmental and live-discourse data
The series is, in spirit as well as in fact, a continuation of the tradition initiatec
by C. Li (Word Order and Word Order Change, Subject and Topic
Mechanisms for Syntactic Change) and continued by T. Givón (Discourse and
Syntax) and P. Hopper (Tense-Aspect: Between Semantics and Pragmatics).
Volume 18
John Haiman and Sandra A. Thompson (eds)
CLAUSE COMBINING IN GRAMMAR AND DISCOURSE

CLAUSE COMBINING IN
GRAMMAR AND DISCOURSE
edited by
JOHN HAIMAN
University of Manitoba
and
SANDRA A. THOMPSON
University of California
JOHN BENJAMINS PUBLISHING COMPANY

Amsterdam/Philadelphia
1988
Library of Congress Cataloging-in-Publication Data
Clause combining in grammar and discourse.
(Typological studies in language, ISSN 0167-7373; v. 18)
Includes bibliographies and index.
1. Grammar, Comparative and general - Clauses. 2. Discourse analysis. I. Haiman, John.
II. Thompson, Sandra A. III. Series.
P297.C54 1989 415 88-35006
ISBN 90-272-2893-0 (hb.)/90-272-2894-9 (pb.) (European; alk. paper)
ISBN 1-55619-022-0 (hb.)/l-55619-023-9 (pb.) (U.S.; alk. paper)
© Copyright 1988 - John Benjamins B.V.
No part of this book may be reproduced in any form, by print, photoprint, microfilm, or
any other means, without written permission from the publisher.
Contents
List of abbreviations vii

Introduction ix
John Haiman & Sandra . Thompson
Linking intonation units in spoken English 1
Wallace Chafe
A syntactic correlate of topicality in Newari narrative 29
Carol Genetti
Inconsequential clauses in Hua and the typology of clauses 49
John Haiman
Concessive clauses in English and Romance 71
Martin Harris
Clause integration in German and Dutch conditionals, concessive
conditionals, and concessives 101
Ekkehard König & Johan van der Auwera
Presentational cleft constructions in spoken French 135
Knud Lambrecht
Towards a typology of clause linkage 181
Christian Lehmann
Subordination in Tauya 227
Lorna Macdonald
Sequential chaining and discourse structure in Godie 247
Lynell Marchese
The structure of discourse and 'subordination' 275
Christian Matthiessen & Sandra A. Thompson
The grammaticization of coordination 331
Marianne Mithun
vi CONTENTS
The discourse function of clause-chaining 361

John Myhill & Junko Hibiya
Nominalization and assertion in scientific Russian prose 399
Johanna Nichols
Index
Cheng Luo 429
LIST OF ABBREVIATIONS
ABL ablative DAT dative

ABS absolutive DEF definite
ACC accusative DEM demonstrative
ACT active DEST destinative
ACTOR actor DET determinative
ADESS adessive DF directional focus
ADV adverb DIR directional
AF actor focus DO direct object
AGR agreement DS different subject
ANTI antithesis DU dual
AOR aorist DUR durative
AP antipassive
APPR apprehensional EMPH emphatic pronoun
ASSOC associative ERG ergative
ASP aspect EXCL exclusive
AUX auxiliary EXCLAM exclamative
BEN benefactive FEM feminine

FIN finite
 controller FOC focus
CAUS causative FUT future
CL clitic
CLAM clamative GEN genitive
COMP comparative GF goal focus
COMPL complementizer
CON conative IF instrumental focus
COND conditional IMP imperative
CONJ conjunction IMPERF imperfect
CONTR contrastive IMPFV imperfective
COORD coordinative INCL inclusive
COP copula INCONS inconsequential
viii LIST OF ABBREVIATIONS
INDEF indefinitive PP past participle

INF infinitive PR possessor
INSTR instrumental PRED predicate
INTR intransitive PREP preposition
IO indirect object PRES present
IRR irrealis PROG progressive
PROH prohibitive
LIG ligature PTCL particle
LINK linking particle PTCPL participle
LOC locative PURP purposive
MASC masculine Q interrogative

MED medial QUOT quotative
NEG negative RECIP recipient

NOM nominative RED reduplicative
NOML nominalizer REFL reflexive
NONFUT nonfuture REL relative
NONPAST nonpast REP repetitive
NP noun phrase
SG singular
NR nominalizer
SP species
SS same subject
OBJ object
STAT stative
OBL oblique
SUB subordinate
OF object focus
SUBJ subject
PART partitive TNS tense
PASS passive TOP topic
PAST past TRANS transitive
PERF perfect
PFV perfective UNM unmarked
PL plural
POSS possessive VOC vocative
Introduction
John Haiman
Sandra A. Thompson
Traditional and modern grammarians alike have restricted what they

call "syntax" to the study of what goes on within the boundaries of the pro
sodic sentence. This limitation is justified by the widespread belief that at
the paragraph level and beyond, anything goes, the ars obligatoria of gram
mar yielding to the anarchy of personal style. The vast majority of syntacti-
cians today continue to adhere to this view (as indeed do some of the con
tributors to the present volume).
On the other hand, the nature of clause combining within a prosodic
sentence has always been a central concern of traditional syntax. One has on
ly to glance through the monumental compendious grammars of languages
like Latin, Greek, German, French, and English to be struck by the subtle,
painstaking, and exhaustive treatment scholars like Schwyzer, Behaghel, and
Jespersen have lavished on the definition, description, and exemplification
of various kinds of coordination, parataxis, subordination, hypotaxis, and
embedding. The importance of at least the coordination/subordination
distinction is taken for granted in generative grammar, where it underlies im
portant research on deletion and anaphora.
There has always been something a little dubious, however, about a
research strategy which submits clause combinations to the most searching
analysis when there is no pause or intonation break between them, and of
ficially ignores virtually synonymous combinations which are separated by
such a pause or break. Unquestionably, the pause is a universal icon of
disassociation (for some examples of its function, see Mithun's article here).
But the implicit claim that utterances separated by such a pause are mutually
independent is an untenable one. It is, for example, a cliché that the distribu
tion of a great many words — among them now, well, why, and a host of
others — cannot be stated without reference to the content of sentences
 JOHN HAIMAN AND SANDRA A. THOMPSON
other than the one in which they appear. Three of the papers in this volume,
those by Genetti, Marchese, and Myhill and Hibiya, are attempts to define
the distribution and the meanings of such morphemes. Genetti argues that
in Newari the choice of nominative or ergative case marking on the first sub
ject of a chain of both transitive and intransitive clauses with the same sub
ject is determined by the clause in the chain which contains the most topical
referent. Marchese shows that the yi "sequential auxiliary" in Godie occurs
in clauses whose participants are highly topical (so that the focus is on the
events described rather than on the participants), but also that a sequence
of "yi clauses" signals the ending of a discourse unit and heralds either an
imminent climax, a change of narrative pace, or an entirely new discourse
unit to follow. Myhill and Hibiya's article deals with two unrelated clause-
chaining languages, Soddo (a Semitic language of southern Ethiopia) and
Japanese, and the distribution of non-final verb forms in multiclause con
structions in these languages. Among their findings they report that there is
some covariation between the non-final clause (in -m in Soddo, in -te in
Japanese) and the degree of foregrounding of the following clause.
The very existence of clause chaining in a large number of (mostly verb-
final) languages exposes the arbitrariness of the traditional limit on syntactic
investigations of clause combining, for in these languages we encounter pro
sodic units whose syntax is as subject to grammatical constraints as the com
plex sentence in English, but whose semantic content is comparable to entire
paragraphs. It may be claimed that languages like Soddo, Japanese, Newari,
and Godie belong to the class of languages where the sentence/paragraph
distinction is a pointless one, but that English, Latin, and German belong
to the (much larger?) class of languages where the distinction is absolutely
crucial.
But there is evidence, considered in many of the papers here, that such
a typological distinction is an artificial one: or, if it is valid, it not only
separates English from Soddo, but spoken English from written English,
and Vulgar Latin from French. Either obliquely or directly, a substantial
fraction of the papers in this volume deal with grammaticization and suggest
that grammatical coordination and subordination arise as universal
discourse structures become conventionalized, primarily in written registers.
Chafe's paper reminds us of the vast abyss between spoken and written
language and shows that in the former, clauses are connected by pauses far
more often than by explicit connectives. Moreover, the explicit connectives
include such unlikely ''conjunctions" as I mean, like, well, anyway, and of
INTRODUCTION xi
course. Mithun shows that the coordinating conjunction, both between

phrases and between clauses, is absent in many languages, and, where it ex
ists, is often derived from a hesitation marker, a perfective aspect marker,
or a word meaning originally 'with' or 'also'. Noting the frequency with
which coordination is expressed simply by asyndeton, she speculates that an
explicit word like and may become necessary only when a language becomes
written or comes into contact with languages that are written.
In a careful study of concessive relations in Romance languages, Harris
shows that almost all the evidence indicates that Vulgar Latin, unlike
Classical Latin, had no grammaticalized concessive construction. Rather,
the modern Romance languages (like Classical Latin) grammaticalized
various adverbs. Very much in the same spirit as Chafe and Mithun, he
speculates that "the archetypal embedded structures are perhaps the reflex
of a society in which literary registers develop and assume importance". A
similar point is made in passing in König and van der Auwera's study of the
integration of concessive clauses in German and Dutch. Since these are
languages with the familiar Germanic Verb-second constraint, one would ex
pect that complex sentences with an initial subordinate clause would force
subject-verb inversion in the following matrix clause. Under certain cir
cumstances initial concessive clauses (unlike initial conditionals) do not force
this inversion, that is, they are not integrated into the matrix clause and do
not "count" as constituents of the matrix sentence. While it is the descrip
tion of these circumstances which they are mainly concerned with, König
and van der Auwera also indicate that there was less integration of subor
dinate clauses in earlier stages of German and Dutch than there is presently,
and that integration or incorporation, like many of the other indices of
subordination discussed by Lehmann in his contribution to this volume, is
a function of evolution like synthesis or conventionalization in general.
Matthiessen and Thompson's paper makes the parallel claim that
"clause combining is a grammaticalization of the rhetorical organization of
discourse" but they support this claim not by reference to spoken language
or earlier stages of the same language, but by appeal to the parallelism bet
ween clause relations within prosodic sentences and sentence relationships
within larger texts.
A very different motivation for clause combining and incorporation is
suggested in the papers by Lambrecht and Nichols. Lambrechťs fine-grai
ned study of colloquial French focusses on apparent complex sentences of
the form yα NP qui VP, which, he argues, are in effect expansions of syntac-
xii JOHN HAIMAN AND SANDRA A. THOMPSON
tically well-formed but pragmatically unacceptable monoclausal W P VP

which are prohibited by a constraint which forbids lexical subjects (new in
formation) in initial position. Another constraint which forbids the inversi
on of subjects in all but a small and highly marked class of clauses rules out
the structure *VP NP which is used by languages like Italian. Not only does
this result in biclausal structures, but the syntactically "main" clause is
almost entirely empty semantically, while the 'subordinate" clause contains
far more information. Nichols' article deals with expository scientific prose
in modern Russian, which is characterized by nominalizations (heavy subor
dination, in terms of the indices discussed by Lehmann) of clauses which
convey new, frequently focussed information. Such nominalizations corres
pond to full clauses in any acceptable idiomatic English translation. While
this gives scientific Russian prose a gnomic flavour, it is motivated, in the
final analysis, by a low-level morphological fact about Russian: the absence
of an obligatory article on nouns and nominalizations. Nichols argues that
it is this absence which allows nominalizations in Russian, unlike
nominalizations in English, to be interpreted as possibly indefinite, and
hence, as a vehicle for new information.
The papers by Haiman and MacDonald deal with clause-chaining
languages of New Guinea, but focus on a possibly widespread but as yet
unreported construction encountered in these languages: the inconsequential
clause. In both Tauya and Hua, the inconsequential clause may function as
an independent utterance: however, when combined with a following clause,
it exhibits many of the properties (not to mention the morphology) of a
canonical subordinate clause. This behaviour is paradoxical only in the con
text of a theory which assumes that a coherent definition of "subordina
tion" exists which equates subordination with dependency.
No such equation is justified, of course, as many of the contributions
in this volume make clear. What does emerge from these papers is that the
motivations for clause creation and combination are semantically and
pragmatically heterogeneous, that correlations between formal indices of
subordination (as presented by Lehmann in particular) and pragmatic func
tion are inconsistent, and that by and large (but note Lambrecht's paper on
colloquial French) complex embedding constructions are a characteristic of
literary registers. In other words, clause combining is an area of grammar
where grammaticization is not as pervasive as it is elsewhere.
This is not to say, however, that it is of relatively little interest to the
grammarian. Rather, clause combining structures, like pidgin languages,
INTRODUCTION xiii
provide a field of inquiry where functional motivation for grammatical

structures are particularly transparent and thus provide new perspectives on
universal grammar.
Linking intonation units in spoken English
Wallace Chafe
University of California, Santa Barbara
There are several ways in which the discussion that follows is unusual,
even within the framework of this volume. Most importantly, it deals with
spoken English, not written English, and certainly not the mythical English
one often finds in discussions of syntax. Second, it does not exactly deal with
clause combining as such, but rather with the linkages that exist between suc
cessive 'intonation units,' as spoken language is produced in real time. The
majority of intonation units are clauses, but many are not. Third, since few
linguists are accustomed to examining language in terms of linkages between
intonation units, the discussion is of necessity more exploratory than defini
tive — only a tentative first step in the direction of an eventual clearer under
standing of what has to be an important aspect of spoken language, and by
extension all language.
Intonation units and clauses
When one listens carefully to spoken language, or even when one

examines a careful transcription of it, one can hardly help noticing that it
occurs in a series of relatively brief spurts of vocalization. In their prototypical
form, these spurts exhibit a single coherent intonation contour characterized
by one or more intonation peaks and a cadence that is recognizable as either
clause-final or sentence-final. Usually they are separated by pauses that last
anywhere from a fraction of a second to several seconds. It seems appropriate
to call them 'intonation units.' 1 By way of illustration, the following is a series
of intonation units produced by a speaker who is discussing a two-year-old of
his acquaintance who has a mother named Cindy:2
2 WALLACE CHAFE
(1) a. ... like almost anyone .. Cindy ... meets when we're like at the
beach or in the park,
b. ... will ... uh— .. ask how old he is.
 ... not s .. surprisingly,
d. ... and when she says two and a half,
e. ... they
f. ... my god.
g. he's so big and .. athletic.
h. ... I mean .. uh— ... a couple times we've run into ... other
kids his age.
i. ... he just ... you know ... makes them look like little babies
next to him.
j . ... and he loves to play.
k. .. he's always running around,
1. .. and playing sports,
m anyway.
We will be looking at two major kinds of linkages between such intona
tion units. First, there are the linkages signaled by intonation, particularly the
cadences distinguished here by commas and periods. Thus, there is a differ
ence between the continuation that is signaled by the non-falling pitch at the
end of (a) and the closure that is signaled by the falling pitch at the end of (b).
Second, there are the linkages signaled by various 'connectives,' words whose
chief function is to signal linkages: for example, the like in (a), the and when
in (d), the I mean in (h), the and in (d), (j) and (1), and the any way in (m). In
a sense, the discussion of this 'connective' category is an extension into spo
ken American English of the 'Conjunction' chapter of Halliday and Hasan
(1976: 226-273).
We can note in passing that the anyway in (m) signals a certain relation
between the entire preceding episode and the entire episode that is about to
follow. In the long run we should be concerned not only with linkages
between small segments such as intonation units, but also with those between
larger, episodic segments. Here, however, most of the discussion will be
restricted to linkages between smaller units.
It is productive to suppose that each intonation unit is a linguistic expres
sion of the particular information on which a speaker is focusing his or her
consciousness at a particular moment (Chafe, 1980b; 1987a). If there is validity
in that interpretation, then intonation units provide us with valuable windows
to the flow of thought by showing how much and what kinds of information
LINKING INTONATION UNITS 3
are being focussed on at different times, but also — of special interest to us

here — the kinds of movement that take place from one focussed piece of
information to the next while thought is being verbalized.
Suppose, just to have a label, we call each piece of focussed information
an 'idea.' There is no idea, in this sense, that exists in isolation; each is always
part of some larger sequence, situated within some larger context. Crucial to
a better understanding of human thought is an understanding of idea sequenc
ing, of how one idea may lead to another. By studying the possible linkages
between intonation units, we can gain some important insights into the nature
of this sequencing. What follows is intended as a start in that direction. It is
based on an examination of the linkages observed within about a thousand
pairs of adjacent intonation units, produced by eleven different speakers who
were participants in informal dinnertable conversations. These conversations
were recorded as part of a larger study of differences between spoken and
written language.3
Before going any further we need to confront the question of the relation
between intonation units and clauses. It may be noted that in example (1)
intonation units (d), (g), (h), (i), (j), and (k) are complete single clauses. The
sequence of (a) and (b) together forms one complex clause, with both (a) and
(b) containing their own embedded clauses. Intonation unit (1) is a clause with
a missing subject. The remaining intonation units — (c), (e), (f), and (m) —
are not clauses at all. It is typical for about 70% of the intonation units in con
versations among academic people to be complete single clauses. There is,
nevertheless, considerable individual variation, ranging from less than 60%
for some individuals to more than 80% for others. Clearly one aspect of con
versational style is the extent to which a person produces intonation units that
are also complete single clauses.
But what exactly is a clause, and why should clauses and intonation units
coincide much of the time, but not all of the time? One way to look at clauses
is in terms of their constituency: a clause is a segment of language that consists
of a subject and a predicate. In terms of the idea that underlies a clause, such
an idea is a composite, one of whose constituents is some referent that is taken
as a 'starting point.' The idea then builds on that starting point with some
piece of 'added information.' The starting point is typically the concept of a
person, less often of a thing, still less often of an event, state, or abstraction.
The added information is typically the concept of some event or state. The
starting point, verbalized as the subject of the clause, usually serves to link the
idea to an already established context. The added information, verbalized as
4 WALLACE CHAFE
the predicate, is the contribution made by the idea to the ongoing flow of
thought within the speaker's mind and, by the speaker's uttering the clause,
within the hearer's mind as well. Thus, in the following pair of intonation
units the same concept verbalized in (a) as my room and in (b) as it functions
both times as the starting point. (The speaker had already been talking about
her apartment, so that the concept of the room had been prepared for.) In (a)
the added information was the state verbalized as was small, in (b) it was the
state verbalized as was like nine by twelve or something:
(2) a. . .my room was small.
b. ..it was like ..nine by twelve or something,
Sometimes there is no starting point, but only added information, as when the
idea expressed by the clause does not apply to a particular referent, but rather
'ambiently' (Chafe, 1970: 102-103; Sasse, in press). In English the resulting
clause usually contains the dummy subject it:
(3) a. .. .it was really hot,
b. it was in the summer and,
It is intriguing to speculate that the intonation units found in spoken lan
guage manifest the flow of ideas, while clauses manifest the flow of language
— the way ideas are verbalized (cf. Chafe, 1979). Ideas, in themselves, are
successive activations of small amounts of information. Their verbalization
then typically requires that these successive activations be expressed in terms
of a starting point and added information, a subject and a predicate, a clause.
But ideas are often expressed as parts of clauses, or as fragments whose possi
ble expression in the clause format never materializes. Sometimes, on the
other hand, an idea may be verbalized as a combination of clauses.
To illustrate further, suppose that a speaker, having successfully uttered
a clause, goes on to add further information not expressed in clausal form:
(4) a. ..and then she .. .went faster.
b. ...in that class,
 ...you know than in the ...beginning class.
From an edited, written point of view it might seem that this speaker had said
'And then she went faster in that class than in the beginning class.' But from
the point of view of the successive activation of ideas, she first focussed on the
idea verbalized as (a), 'and then she went faster,' a complete sentence closed
off with a falling pitch. Having done that, she then focused on the idea that
the going faster took place 'in that class,' verbalized as an afterthought to (a),
but showing with a non-falling pitch that she already intended to add the com
parison verbalized as a second afterthought in (c), 'you know than in the
beginning class.' The following example is similar:
(5) a. ...so I was sure there was something wrong with my spine or,
b. ..you know brain tumor at least,
 ..uh ... spinal meningitis,
d. ..or something like that.
This speaker's successive focusing on different possibilities again extends the
clause beyond its original formulation through a series of separately focussed-
on alternatives.
There are also cases of the opposite sort, showing several clauses within
a single intonation unit. For example, a main clause may be followed by
another, 'adverbial' clause within the same intonation unit:
(6) but ...there were a few incidents that happened with me just
because I was a foreigner.
Or a relative clause may modify a noun in another clause without forming a
separate intonation unit:
(7) ... one of the things we had to do was type.
Or one clause may be embedded as the complement of a verb in a 'higher'
clause:
(8) I thought it was interesting.
These various devices with which a speaker may verbalize a single idea as
more than a single clause lie beyond the scope of the present discussion. Here
we will look only at various ways in which separate intonation units, verbaliz
ing separate ideas, may be linked one with another. In effect, the discussion
will focus on the kinds of linkages that occur between intonation units that are
either clauses or less than clauses.
Our starting point, then, is a view of the flow of thought and the flow of
language in which one idea is replaced by another through time. Each succes
sive idea is triggered by, and in fact contains information that overlaps with
the idea that preceded it, but at the same time each new idea exhibits some
change from the preceding one. We are interested in how language expresses
the links that may exist between one idea and the next as the flow of ideas is
verbalized. The discussion will be limited to links between intonation units
that are produced by a single speaker, and will thus ignore connections across
6 WALLACE CHAFE
the turns of different speakers. We will be examining the two major kinds of
linkages mentioned above: those that involve intonation, and those that
involve 'connectives' — conjunctions and similar words that overtly verbalize
the link itself. It is easiest to begin by looking at the purely intonational links.
Linking with intonation alone
There is a major distinction to be made between intonation units ending

with the kind of falling pitch that signals closure on an idea or idea sequence,
and intonation units ending with any of a number of other pitch contours that
signal continuation. I will refer to the falling contour as 'period intonation',
since it is most often represented in written English (as well as in the trans
cribed examples in this paper) with a period. It contrasts with a variety of
other contours to which I will refer collectively as 'comma intonations.'
In a sense the least marked linkage between intonation units is that in
which one unit ends with the period intonation and there is no connective
word linking it to the next unit. There are several functionally quite different
situations that lead to this result. The following are among the most conspicu
ous situations of this kind.
Afterthoughts. Quite often in spoken language a period intonation will

signal the end of a sentence, but the speaker will then tack on an intonation
unit that conveys a piece of supplementary information relevant to that same
sentence. One understands that the speaker chose at first to end the sentence
at the period, but then thought of something else that would also be useful for
the hearer to know, within the same cluster of information. More often than
not the additional information is expressed in a prepositional phrase, as in (9)
and (10), though it may have some other adverbial status, as in (11):
(9) a. ... well .. it was just obvious I couldn't ... I couldn't work.
b. ... uh .. with her in the office.
(1Q) a. ... but ... uh ... my father came .. into possession of some
papers that his mother had.
b. ... uh .. in German.
(11) a. ... it was quite .. striking when we were . the year we were in
Japan.
b. ... three years ago.
Wheel-spinning. A second, not at all rare situation in which one intona

tion unit ends with a period intonation and fails to be linked with an overt
connective to the next unit involves a temporary stoppage in the forward flow
of information, where the speaker momentarily engages in wheel-spinning. In
the following example the speaker had just been asked a question, to which he
responded:
(12) a. ... yeah.
b. .. that's right.
 .. yeah .. that's the idea.
The prosodies of these three intonation units are in important respects identi
cal: in each case the last word contains the intonation peak, which is followed
by the period intonation. It can be seen that the speaker says essentially the
same thing in (b) that he says in (a), and that he continues to say essentially
the same thing in (c). This is not to imply that he was wasting his or his inter
locutor's time; he may have judged his answer important enough to dwell on
it in this way. But the successive intonation units add nothing else that is new.
Here is a similar, if slightly more complex example:
(13) a. ... it's ... it's a pleasure.
b. ... I mean it.
 . . I enjoy it.
Again there are no overt connectives and there is a prosodic echoing. Intona
tion units (a) and (c) convey essentially the same idea; (b) expresses some
thing different, but its function is only to emphasize the speaker's sincerity in
verbalizing that idea.
Wheel-spinning is reminiscent of the 'couplets' that are a characteristic
feature of some oral literatures (McLendon manuscript; Mithun manuscript):
a pair of intonation units that repeat the same idea in different words. The fol
lowing is a translation of a sequence of couplets from a particular Maya ritual
(Gossen, 1984: 204):
(14) a. your children are gathered together,
b. your offspring are gathered together,
 for you to see,
d. for you to witness.
e. great San Juan,
f. great patron.
Couplets appear to be a stylized, aesthetically effective use of the same wheel-
8 WALLACE CHAFE
spinning that occurs often and unremarkably in colloquial speech in examples

like (12) and (13).
Crossing episode boundaries. In the next example, intonation units (b)

and (c) illustrate an extended form of wheel-spinning, each adding, after a
period intonation, a further comment on the smallness of the room. The same
period intonation appears again at the end of (c), still with no overt connec
tive linking (c) with (d). Nevertheless, the relation between (c) and (d) is
quite different from that between the earlier intonation units:
(15) a. . .my room was small.
b. ...it was like .. nine by twelve or something.
 it seemed spacious at the time.
d. ... I came home,
Here there is a major change in subject matter: from the repeated focuses on
the smallness of the room to a quite different focus on the speaker's arrival at
home on a particular occasion, an arrival that anticipates some newsworthy
events to follow. This leap forward in the flow of information is reinforced by
certain other prosodic features: a significantly longer pause at the beginning of
(d), together with a higher pitch and greater volume. The pause suggests a
major cognitive shift at this point, while the increased pitch and volume are a
signal that the speaker is beginning something new.
In the following example the speaker is talking about a course taken by
another student in her college dormitory:
(16) a. ... a—nd uh— ... her course .. required that they write ... short
papers.
b. .. and if you wrote five,
 you got an A,
d. .. if you wrote four,
e. you got a B,
f. if you wrote three,
g. you got a C,
h. it didn't matter what the papers were like.
i. ... I was taking a philosophy course with another professor,
j . ... and we also had to write short papers,
k. ... but ... we also got graded on our papers.
Of interest here is the transition from (h) to (i). Again there is a period into
nation at the end of (h), and no connective linking it to (i). Again there is a
longer pause, with a higher pitch and volume at the beginning of (i). A shift
of topic is again evident: from a description of grading practices in the other
student's course, the speaker shifts attention to a course she herself was tak
ing. In written language a paragraph boundary would have been appropriate.
We have seen three quite different cases in which intonation units suc
ceed one another with period intonation signaling closure at the end of the
first unit, and with no overt connective linking the first unit to the second.
Sometimes such cases show the tacking on of additional information that
appears as an afterthought. Sometimes they show wheel-spinning, where the
speaker reinforces an idea by continuing to focus on it, verbalizing it in a dif
ferent way, but failing momentarily to move on to the next idea. But some
times we find a major forward movement: a shift not only to a new idea, but
to a new paragraph-like episode.
Comma intonation: more to come. The consistent function of the falling

pitch prosody in the above cases was to signal closure: the fact that the
speaker had, at least provisionally, come to the end of an idea or of some
coherent sequence of ideas. Perhaps the simplest way to signal a linkage
between one intonation unit and the next is to end the first unit with a pitch
contour that does not fall to the lowest level. English has a collection of such
contours, all contrasting with the period intonation. There may be a rise, a
fall-rise, a level continuation, or even a partial fall. These contours have in
common their indication that there is more to come: that the intonation unit
just uttered is not the end of anything, but is going to be followed by some
thing else. It is here, then, that we have our first real example of overt linking:
the use of a comma intonation to show that another intonation unit is coming
before any significant closure is reached.
The example below followed a discussion of how the speaker had fainted
while taking a shower, and how she had never fainted before. Her inter
locutor then asked 'Where did you find yourself when you did faint?' The
reply was:
(17) a. ... I was standing up,
b. . . I was taking a shower,
 ... I had no warning,
d. ..no nothing.
At the end of each intonation unit there is a comma intonation, up to the clo
sure in (d). The rising pitches at the ends of (a), (b), and (c) function to signal
each time that the speaker has not yet finished what she is communicating.
10 WALLACE CHAFE
The comma intonation shows continuation rather than closure, but the
lack of any explicit connective provides no precise signal of the relation
between the one idea and the next. The above speaker is describing several
aspects of her activities and her state at the moment of fainting. These aspects
are simultaneously true; they each represent parts of her total state. There is
more here than wheelspinning. The speaker is progressing from one idea to
the next, but is at the same time circling around a central concept, not as
clearly moving forward as in the cases we will meet when overt connectives
are discussed below. It is as if she were viewing a single image from different
angles, before moving on to something new after the closure in (d).
Here is another example:
(18) a. ... I came home,
b. I was really exhausted,
 I was eating a popsicle,
d. ... I was sitting there in my chair,
e. ...just eating my popsicle,
Again the comma intonation signals a continuation from one idea to another,
but again there is a circling around to view different facets of a single larger
image. As in (17) this image involves the speaker's state and activities just
prior to a salient sequence of events. The comma intonation signals movement
to another idea, but the absence of a connective shows that this idea remains
within the same larger image and does not move significantly forward.
Linking with 'coordinating' connectives
All the examples so far have been characterized by the absence of any
word that explicitly marks the link between one intonation unit and another.
The speaker simply moves from one to the next, either closing off the first
with a period or showing continuation with a comma. About 44% of the cases
of unit linkages examined in this study are of this non-connective type. In
56% of the cases, then, some more specific linkage marker is present. Here is
a brief survey of the more conspicuous of these markers. Other recent works
have treated them in greater detail (for example Schiffrin, 1987 and Schourup,
1985).
and. Fully 50% of the cases with explicit connectives consist of or include
the maximally general connective and. This and is usually uttered at the
beginning of the second unit of a pair, but in a small number of cases it
appears at the end of the first unit. Both possibilities are illustrated in (19),
where and occurs at the beginning of (a) but at the end of (c):
(19) a. and then another day,
b. ...it was really hot,
 it was in the summer and,
d. . .my room was small.
The function of and is evidently to signal only that the idea expressed in the
upcoming intonation unit moves forward from the idea that preceded. If a
comma intonation signals that another idea is to follow before closure is
reached, the and signals that this next idea will be more than just a different
view of something already communicated, as in (17) and (18) above. The
more specific relation between this next idea and the preceding fails to be
specified; the speaker marks only that there is a relation of some kind. As
Schiffrin puts it (1987: 150), "All and displays is continuation, and/or
coordination: more precise identifications depend on discourse content and
structure."
Evidently, however, the most common relation is that of temporal order
ing: and is likely to be used when the event conveyed in the second intonation
unit took place after the event conveyed in the first:
(20) a. ..and we ...had gone past it maybe ...about a hundred yards,
b. and we ...decided we'd go back and investigate,
 ..and ..we did,
d. ..and ...it was ...a teenager,
e. ...quite dead.
f. ...in the ditch.
This sequence describes events that followed immediately after another
sequence; hence the and at the beginning of (a). The decision verbalized in
(b) followed the change of location described in (a). The implementation of
that decision, verbalized in (c), followed the decision itself. The discovery ver
balized in (d) was next. Wherever there is an and there was a forward move
ment through time. The absence of and at the beginning of (e) conforms to
the 'comma intonation without connective' pattern described earlier. Having
met the teenager in (d), we are told about his state with no connective; there
is no further forward movement through time, but only a circling back to
establish this property of the body. The period intonation at the end of (e) sig
nals closure, but then the speaker supplements what he has said by giving the
body's location in (f).
12 WALLACE CHAFE
If temporal sequencing is the most common relation associated with the

use of and, it is certainly not the only one:
(21) a. .. .the child is seated,
b. and . .the .. .research assistant starts explaining what's going on,
 and here's this divider and the ...this big wooden thing,
d. ...a—nd uh— ...and ...it's an art project,
e. ..and all they have to do is make a ...they have to .. follow a
medel and draw something.
The movement from (a) to (b) does involve temporal progression. Intonation
unit (c), however, involves a change in spatial focus simultaneous with (b); we
shift our attention from the seated child and the researcher to the apparatus
before them. With (d) the speaker shifts from this scene to a more abstract
view of the background against which these events were taking place. In (e)
he remains within that background perspective, but zooms in to a more
specific aspect of it. We see in (21) that intonation units connected by and may
communicate either a temporal or a spatial progression, a shift between con
crete and abstract, or a zooming in or out. Of course this is only a sample of
the relations that are possible in such cases.
but and or. The connectives but and or are alternatives to and, adding
additional flavors — in the one case of 'contrary to expectations,' in the other
of 'disjunction' — to the maximally nonspecific forward movement conveyed
by and. They are, however, considerably less frequent, with but occurring in
these data only one-fifth as often as and, and or less than one-sixteenth as
often. This is not the place to discuss in detail the specific functions of but or
or, but the following excerpts will supplement (20) and (21) as illustrations of
these two more specific kinds of forward movement:
(22) a. ... I mean I was successful,
b. but I .. I really worried,
(23) a. she's always had a thing for older men,
b. but this is getting ridiculous,
(24) a. ... then .. we also have .. two kinds of warnings,
b. ... or .. alertings.
(25) a. and she can't sleep,
b. ..or anything like that,
The vague extension of possibilities illustrated in (25b) may in fact be the most
common use of or in spoken English, as also in:

(26) a. I've never fainted.
b. ... or passed out,
 or gotten dizzy,
d. or anything.
The linkage between (a) and (b) in the last example suggests that and,
but, and or, may occur after a period intonation as well as after a comma.
Thus also:
(27) a. ... he just ... you know ... makes them look like little babies
next to him.
b. ... and he loves to play.
(28) a. so it was a lo—ng it was almost three thousand miles by the
time we were done.
b. but it was ... spectacular,
In the data examined here, and and or occurred much more frequently after a
comma than after a period, while but was equally common in both environ
ments.
All of the connectives to be discussed in the remainder of this section
were more common after period boundaries. Whereas and, or, and, more
ambiguously, but tend to relate ideas that are not self-contained but are ele
ments in a larger sequence, the following connectives tend to initiate some
thing new after a preceding sequence has been closed off with a period intona
tion. In the data examined, about 70% of these connectives occurred after
period intonations, only about 30% after commas.
These connectives are also distinguished by their ability to occur together
with and, a fact that confirms the notion that they add something more
specific to the general linkage signaled by the and. In other words, to take so
as the most common example, we find intonation units that are linked with
and, those that are linked with so, and those that are linked with and so.
Occasionally we find such longer sequences as 'and so then.'
so. By far the most common of these connectives is so, accounting for
about 40% of the occurrences of connectives of this type in these data. So
most commonly signals that the next idea, or often the decision to verbalize
the next idea, results in some fashion from what has already been said (com
pare the more extended discussion in Schiffrin 1987: 191-227):
14 WALLACE CHAFE
(29) a. ... I've had headaches maybe twice before in my life.

b. ... so I was sure there was something wrong with my spine or,
(30) a. ... they took a lot of tests.
b. ... so— by the ti—me I had the headache it .. they had the
results of the tests.
In (29) the idea in (b) does not follow literally from the idea in (a), but rather
from the the entire experience of which (a) is simply the last idea verbalized.
In (30) the idea conveyed in (b) was not itself a result of (a). Rather, the
speaker's decision to verbalize (b) followed from her verbalization of (a).
This so is not to be confused with the so (an alternative for so that) that
introduces a clause of result:
(31) a. ... and actually it .. hooks on,
b. so it doesn't fall off,
Examples like (31b) belong in the category of connectives to be explored in
the next section.
So, and especially and so, often follow an episode boundary:
(32) a. .. and all they have to do is make a ... they have to .. follow a
model and draw something.
b. ... and so the uh— .. research assistant explains I'll just pin the
model up here on the divider,
Intonation unit (a) closes off a general discussion of an assignment given to
some students, and (b) then introduces a new series of events beginning with
the research assistant's explanation. As at other episode boundaries, these
cases of so and and so tend to be accompanied by significant hesitating as well
as higher pitch and volume.
Our data show a motley collection of other connectives in this same gen
eral class. Each of them deserves an extended discussion, but I will only list
them here, giving a rough characterization and one or two examples of each.
The order is one of descending frequency. I mean, well, and like are signific
antly more frequent than the others.
I mean. While this phrase is not traditionally regarded as one of the Eng
lish connectives, in fact it does occur most often at the beginning of an intona
tion unit, where it signals an amplification or clarification of the idea that pre
ceded it (cf. Schiffrin, 1987: 295-309 and Schourup, 1985: 147-148):
(33) a. ... but uh ... it's going to be hard.
b. I mean I may not really be putting in full time even this quarter.
(34) a. ... I was thinking ... that she has kind of a tough life.
b. ... I mean she ... works,
 she has a husband,
d. she doesn't know what ... what's gonna happen,
e. ... she has a kid,
f. that she has to constantly take care of,
like. This word, frequently used in spoken English as a hedge, is also

used extensively as a connective, in which function it is similar to the for
example of expository writing. (Other functions of like are discussed in
Schourup, 1985: 37-63.) The question in (35a) related to the principal
speaker's experiences in a foreign country. There followed two intonation
units introduced by like:
(35) a. [Did you feel that people always looked at you?]
b. ... yeah.
 ... I felt .. like I felt a loss of identity when I .. when people
didn't.
d. .. like sometimes .. some days it would seem like people
weren't staring at me,
e. it felt really weird.
In (36) the speaker had been telling about a friend who had been offered a
new job. His interlocutor asked the question in (36a):
(36) a. [Is it something more satisfying to her?]
b. ... a li—tle,
 ... I mean .. her skills are mainly in education,
d. ...or—in secretarial.
e. ... like .. administrative assistant.
well. This connective signals a shift in the flow of information, a partially

new topic, but one that is at the same time closely relevant to what preceded.
As Svartvik (1980: 177) puts it, well "introduces a part of the discourse that has
something in common with what went before but also differs from it to some
degree." It often appears in conversation after such a topic shift has been
triggered by the interlocutor. After (37c) the speaker goes on to talk about
her own experiences:
(37) a. [You know Bill was really bothered by it.]
b. yeah.
16 WALLACE CHAFE
c. ... well I wasn't really much at all.

After (38g) the speaker continues to talk about her previous freedom from
fainting episodes (see (26) above):
(38) a. ... you heard that I fainted in the shower.
b. [Yeah.]
 [Bob said you fainted in the shower.]
d. [Right.]
e. ... okay.
f. ...well,
g. .. tha—t in itself was scary cause I never fainted before.
Other writers have characterized the function of well in other ways. Schiffrin,
(1987: 102-127), for example, finds that it marks the following material as a
'response,' while Shourup (1985: 64-93) finds it to be an indicator that the
speaker "'is engaged in private thinking."
then. Sometimes then as a connective straightforwardly signals temporal

progression. The event verbalized in (39) simply followed a series of other
events in time:
(39) ... then he decided he knew what she was trying to say,
Sometimes its function is more like in addition:
(40) a. ... so the ... situation is constructed to get us ... offers .. and
requests.
b. ... between the kids.
 ... then .. we also have .. two kinds of warnings,
In these data then occurs most often with a preceding and. The following
example shows two occurrences of and then, the first after a comma and the
second after a period. The greater strength of the period boundary between
(c) and (d) is evident; at this point there is a major shift in time:
(41) a. .. so I was sort of nervous about that,
b. .. for a day or two,
 .. and then I forgot about it.
d. .. and then the next week,
e. . . I woke up with this headache.
anyway. The function of anyway to signal a return to the main topic after
a digression is well known. After a long digression the speaker of (38) above
returns to the topic of fainting with:

(42) a. ... anyway,
b. ... and then I fainted again.
Asked about whether she received a lump from hitting her head, she replies:
(43) a. ... yeah.
b. .. yeah.
 ... it's just about going away now,
d. . . I had it about two weeks.
e. ... anyway,
f. .. so I was sort of nervous about that,
g. .. for a day or two,
h. .. and then I forgot about it.
(of) course. This connective, often simply course, has the paradoxical
function of introducing an idea that is contrary to expectations in an expected
way:
(44) a. I thought it was interesting.
b. .. course everybody else grumbled about it.
oh. As a connective oh functions similarly to to be sure: "what follows is

true, but it does not detract from the point I just made." The following
speaker had said she had never been involved in plagiarizing:
(45) a. ... oh I handed in somebody's ... paper,
b. but that was a little bit different.
We have followed, then, a progression of connectives from the maximally
general and with its relatives but and or, associated most often with the con
tinuation signaled by a comma intonation, to a set of more specialized connec
tives occurring most often after the closure signaled by period intonation.
Conspicuous among the latter are so, then, I mean, like, well, anyway, of
course, and oh. Common to these last connectives is the signaling of some
kind of new beginning in the information flow.
Linking with 'subordinating' connectives
In traditional terms the kind of linking discussed in the last section, so far
as it involves clauses, would be regarded as the linking of two clauses whose
18 WALLACE CHAFE
status is in some sense equal. Hence the term 'coordination.' There are other
examples of clause-linking in which one clause would traditionally be
regarded as a 'main' clause and the other as 'subordinate' to it. A recent arti
cle by Haiman and Thompson (1984) has called attention to the incoherent
nature of the various phenomena traditionally associated with subordination,
suggesting that the concept be abandoned as failing to embrace any unified set
of observations.
I want to amplify that point by suggesting that the clause-linking connec
tives that might be thought typical of 'subordinating conjunctions' raise seri
ous questions as to what subordination might mean, at least with respect to
spoken English. I am referring here to connectives like because, when, if, so
that, although, before, since, where, every time, and at which point, which hap
pen to be those present in the data before us. The following example shows
how a linkage of this kind is apt to occur in conversation:
(46) a. ... there're two kids,
b. ... who do it at the same time.
 ... so they have to share.
d. ... cause there aren't enough markers.
e. ... and that means they have to make requests for markers,
This excerpt is interesting because of the variety of linkages it illustrates. The
relative clause in (b) is a type to which we will return below. The so in (c) and
the and in (e) illustrate the 'coordinate' type already discussed.4 It is the rela
tion between (c) and (d) that illustrates the class of connectives before us now.
'(Be)cause' signals a relatively specific relation between the clause it
introduces and some other clause: whatever is verbalized in this clause is the
'cause' of (or at least gives evidence for) whatever is verbalized in the other
clause. But it is not clear why (d) should be considered 'subordinate' to any
greater degree than the other clauses in this sequence. The one clause here
that might possibly be regarded as 'independent' is (a), but the fact that the
presentative 'there' construction in (a) anticipates a following relative clause
casts doubt on its complete independence as well. What we have here is a
sequence of clauses, each linked in some way to one or more other clauses.
Perhaps the only property of (d) that is special is the fact that a clause
introduced by '(be)cause' may sometimes precede rather than follow the
clause to which it is linked:
(47) a. because there aren't enough markers,
b. they have to share.
This concocted example is the first we have seen with an anticipatory linkage
expressed at the beginning of the first of two clauses, rather than between the
linked clauses. Barring a change of mind on the part of the speaker, the first
clause can then only end in a comma intonation, not a period, since it has
already been established that another clause will follow. Apparently it is this
capacity for what might be called bidirectional linking that distinguishes these
'subordinate' clauses from those discussed earlier. Clauses like the following
imply only backwards linking:
(46c) so they have to share.
(46e) and that means they have to make requests for markers,
But those of the 'subordinate' type may be linked either to what precedes or
to what follows:
(46d) or (47a) (be)cause there aren't enough markers,
For speakers, however, this is not a balanced choice. In spoken English back
ward linking of the type illustrated by (46d) is the preferred type. Of all the
occurrences of 'subordination' in the data before us, 61% involve backward
linking, and 39% forward linking.
Although, as just mentioned, forward linking virtually requires a comma
intonation at the end of the first clause, backward linking has no such require
ment. Among such examples in these data, the proportions of comma and
period intonations at the end of the first clause are approximately equal. The
following three examples with because, when, and if all show a comma intona
tion at the end of the first clause, evidence that the linked clause was already
anticipated:
(48) a. ... you know so I can understand objections to ... people bring
ing in babies,
b. .. because it just doesn't work.
(49) a. so it doesn't fall off,
b. ... when we don't want it to.
(50) a. ... it's incomprehensible,
b. if you don't know it already.
The following examples are parallel, except that the first clause ends with a
period, suggesting that the linked clause was conceived subsequently and
more independently:
20 WALLACE CHAFE
(51) a. ... a friend of hi a— flute .. student of his called up and said she
can't come to her lesson.
b. ... becau—se ... she's sick.
(52) a. ... I went to the doctor after the first one.
b. ... when I fainted.
(53) a. ... so .. the purpose of the course is to— ... create something
like that.
b. ... if that's possible.
Among the postposed or backward-linking clauses, '(be)cause' is by far
the most common connective in these data, accounting for 50% of the total
cases of this type. Among the less common preposed or forward linking
clauses, 'if' is the most common connective, accounting for 44% of the cases.
The following examples with if, when, and because are arranged in order of
descending frequency:
(54) a. .. if I were to start over,
b. ... uh ... I would .. I would take a .. uh — . . a t least ... six
months maybe a year's leave of absence.
(55) a. ... when they ... see that this is ... happening,
b. ... they are engaged in conversation.
(56) a. ... uh because I'm an .. adviser,
b. I have to be on campus in the afternoons too.
In short, we have found no clear reason why linkages expressed by con
nectives like because, when, and if should be regarded as any more 'subordi
nate' than clauses introduced by and, except for the fact that they may antici
pate as well as follow the clause to which they are explicitly linked. We might
then think of them as bidirectional connectives, contrasting in at least that
respect with the connectives previously discussed.
But we cannot leave this discussion of 'subordination' without giving
some attention to relative clauses, as already exemplified in (46b) above. Our
concern here is only with relative clauses that constitute a separate intonation
unit. Some of them would be separated with commas in writing, and would be
classed in the traditional 'nonrestrictive' or 'appositive' function, expressing
some kind of 'aside.' But most of them carry forward the flow of information
on the main track of the discourse. All of them have in common the fact that
they say something additional about a referent already introduced and present
in the preceding intonation unit. This kind of linkage is relatively rare in these
data, constituting only 3.5% of all the linkages examined.

The most common of the relative pronouns in this sample is which. It
occurs with approximately equal frequency after a comma and after a period.
Some of the resulting clauses add information about a nominal referent:
(57) a. ... and they start the task,
b. ... uh— which involves coloring.
(58) b. ... and I was sort of ... en .. disentangling my hair .. from the
waste paper basket.
b. ... which is .. under there.
Others add information about an idea expressed as an entire predicate or
clause:
(59) a. ... the verb may not be a verb to the kid,
b. which of course is a ... a good point,
(60) a. ... and uh .. so everybody cracked up.
b. which was my intention.
Somewhat less common are the relative pronouns that and who. In these
data, that always refers to a nonhuman entity established by a previous nomi
nal referent:
(61) a. .... I'm going to make a proposal tomorrow.
b. ... uh— ... tha—t ... probably will go through
(62) a. and they send a ... nice .. typed .. letter,
b. ... that's ... you know.
 ... ve—ry ... well done,
Who also always refers to a nominal referent, but of course a human one:
(63) a. ... I noticed there was a gu—.
b. ... who walked by.
(64) a. .. they hope to get somebody else,
b. .. who's lined up for it,
It can be seen from these examples that the function of intonationally
separate relative clauses in spoken English is to provide a linkage in which a
referent established in one clause is repeated in the relative pronoun that
introduces the next. The relative pronoun is essentially a one-word substitute
for linkages of the type and he, and it, etc.
22 WALLACE CHAFE
Focusing on a connective
We began by looking at cases in which the linkage between two intona

tion units is not overtly signaled at all. Having seen how linkage can be most
simply expressed through comma intonation, we examined several varieties of
explicit connective words. It is appropriate now to end the discussion with
cases that are farthest from where we began: cases in which the linkage itself
is given full attention. Here the link is verbalized not with a connective that is
only one element in a larger intonation unit, but with a connective that
occupies a complete intonation unit by itself. A speaker, that is, may focus for
a moment entirely on the linkage.
To do this is not especially common in spoken English. Only about 1.5%
of the linkages examined separate off the connective itself as a separate into
nation unit. The connective most often treated in this way is well, which func
tions in part as a pause filler (see also example (38) above) :
(65) a. [But it was an apartment building?]
b. .. well,
 ... I lived in a totally farm neighborhood but,
d. ... on the grounds of this one farmhouse,
e. she'd built this little apartment house.
The next most common is anyway (see also example (42) above):
(66) a. ... it's just about going away now,'
b. . . I had it about two weeks.
 ... anyway,
d. .. so I was sort of nervous about that,
e. .. for a day or two,
Also occurring as complete intonation units are so, but, now, and actually.
Complete intonation units are also formed of the sequences well anyway and
well also I mean.
A comparison with written English
One of my purposes in constructing this tentative picture of linkages in

spoken English has been to show their relatively simple nature as compared
with those that are most often discussed by linguists, usually stemming from
memories (and occasionally observations) of what written, not spoken Eng
lish, is like. Several factors may be responsible for the differences between
spoken and written linkages. Speakers, as opposed to writers, have little time
to devote to making the linkages between intonation units explicit. Verbaliz
ing ideas on the run, they are too busy expressing them as conversationally
successful intonation units to add the time necessary to elaborate the connec
tions between them. Both the presence of a directly shared context and the
ability to supplement words with prosody and gestures help to make the con
nections between ideas more apparent. For writers, the absence of a directly
shared context and the lack of prosodic and gestural resources make it more
imperative to be explicit about connections between ideas.
In other words, both the advantages and the deficits of spoken language
work together to minimize the use of elaborate linking devices, while the com
plementary advantages and deficits of written language favor their elabora
tion. As a result, spoken language tends to favor the devices discussed earliest
in this chapter — linking with comma intonation and the 'easy' connectives
such as and — whereas written language tends to favor those discussed later
— the more specific and bidirectional connectives, relative pronouns, and the
treatment of connectives themselves as separate punctuation units. It is par
ticularly instructive to see how writers exploit this last device, used only rarely
by speakers.
As spoken language naturally divides itself into intonation units, written
language offers an analogous segmentation into 'punctuation units,' the
stretches of language that occur between punctuation marks. Although the
analogy is by no means perfect, there is a strong tendency in English writing
for punctuation units to mirror the intonation units of speech.5 That being the
case, it is of interest to examine the use of connectives as independent punctu
ation units in writing. The following remarks are based on comparably sized
samples of academic writing obtained from the same eleven individuals whose
spoken language was discussed above. I should stress that these remarks apply
to academic writing; the extent to which they are valid for other written
genres is an open question.
The striking finding is that such writing fosters not only the much more
frequent use of connectives as entire punctuation units, but also the use in this
way of a much larger variety of connectives. By far the most frequent in these
data are for example, however, and thus:
(67) a. the more target properties an item contained,
b. the more likely it would be to be recalled.
 For example,
24 WALLACE CHAFE
d. a subject might initially use each property separately to try to

recall and,
e. then,
(68) a. One contrary finding was that they found the subjects' time to
search the word lists did not correlate with the number of
items recalled.
b. However,
 the measures were total time and total recall,
(69) a. the corresponding expected numbers of positive decisions are
0.0,0.33, 1.0, and 3.0.
b. Thus,
 if the number of decisions improves memorability only when a
property of the word matches one of the target properties,
Whereas the different connectives treated in this way in the spoken sam
ple are limited to about 6, those found in the samples of academic writing
number at least 35. A sample from among them, in addition to those illus
trated above, includes again, also, apparently, as a result, as such, at least,
e.g., finally, for these reasons, for the most part, for one thing, furthermore,
i.e., if so, in fact, in particular, in addition, instead, later, more exactly, more
recently, moreover, nevertheless, of course, rather, second, similarly, so, speci
fically, that is, then, and therefore. Thus (academic) writing fosters both a
focusing on connectives themselves, and the invention and use of a great vari
ety of such connectives.
Conclusion
We have seen that the flow of ideas is manifested in spoken language in

the spurts of language I have been calling intonation units. It is important to
study the linkages between intonation units as a way of understanding the
associations that may exist between ideas. Since something like 70% of the
intonation units in conversational speech are complete single clauses (and
many of the others are clause fragments), there is good reason to think that
the clause is the typical way of verbalizing an idea. Clause linkage is the pre
dominant type of intonation unit linkage.
The limiting case of linkage is that in which the first unit ends with a
period intonation and the next begins with no connective word. There are at
least three distinguishable cases of this type: after-thoughts, wheel-spinning,
and episode boundaries. In contrast, a comma intonation at the end of the

first unit signals the speaker's intention at that point to continue with some
thing else.
Among the explicit connective words, by far the most common in spoken
English is and, signaling nothing more than that the idea expressed in the sec
ond unit moves forward in some way from the idea expressed in the first. The
less common, more specific cousins of and are but and or. These three are
commonly associated with a comma intonation. There are a variety of other
connectives like so, I mean, like, well, then, anyway, of course, and oh that are
more apt to occur after the closure expressed by a period intonation.
Somewhat different in nature are connectives like because, when, and if,
whose major distinguishing property appears to be their ability to signal a lin
kage with either the preceding or the following clause. It is difficult in spoken
language to find any clear content for the notion that they express 'subordina
tion.'
Relative pronouns provide another kind of linkage, combining in a single
word both the fact of the linkage and the repetition of a referent that
appeared in the preceding clause. Most relative clauses in spoken English lead
into new ideas in much the same way as linkages like and he or and it.
Finally, at the opposite extreme from the minimal linkages first discus
sed, there may be a focusing on the connective itself as an independent into
nation unit. This device is relatively rare in spoken English, but it comes into
its own in writing, where punctuation units consisting of nothing more than a
connective (like the finally at the beginning of this paragraph) are common
and diverse.
In concentrating on linking by means of intonation and connectives, I
have ignored various other ways in which ties between intonation units may
be expressed: for example, through anaphoric and lexical devices like those
mentioned in passing in footnote 4 (cf. Halliday and Hasan 1976). The entire
subject needs more thorough treatment. I hope, however, that I have at least
shown that focusing such research on real spoken English is likely to be fruit
ful, and to provide some surprises.
NOTES
1. Compare, among others, the 'information units' of Halliday (1967), the 'information blocks'
of Grimes (1975), the 'tone units' of Crystal (1975), the 'idea units' of Kroll (1977), the
'lines' of Hymes (1981), and in fact the 'clauses' of Pawley and Syder (1983).
26 WALLACE CHAFE
2. The sequences of two, three, and four dots indicate pauses. Two-dot pauses are less than
half a second in length, three-dot pauses between one-half and two seconds, and four-dot
pauses longer than two seconds. It may be noted that pauses also occur within intonation
units (Chafe, 1980a). A dash, as in 'uh— ,' indicates a lengthening of the preceding segment.
Commas and periods show intonational cadences in a way to be described.
3. The study was sponsored by Grant G-80-0125 from the National Institute of Education, with
additional assistance from the Sloan Foundation grant in support of cognitive science studies
at the University of California at Berkeley. I am grateful for the collaboration of Jane
Danielewicz, Pamela Downing, Tanya Renner, and Knud Lambrecht. Some related results
of this study have been reported in Chafe (1982 and 1985), and Chafe and Danielewicz
(1987).
4. Worth noting is the fact that the linkage between (d) and (e) is strengthened through the use
in (e) of the pronoun 'that,' referring to the idea expressed in (d), as well as through the verb
'means,' which goes further in specifying the relation that (e) bears to (d).
5. See now Chafe, 1987b.
REFERENCES
Chafe, Wallace. 1970. Meaning and the Structure of Language. Chicago: Uni
versity of Chicago Press.
1979. 'The flow of thought and the flow of language." In: Discourse and
Syntax, T. Givón (ed.), New York: Academic Press.
. 1980a. "Some reasons for hesitating." In: Temporal Variables in Speech,
H.W. Dechert and M. Raupach (eds.), The Hague: Mouton. Reprinted in
Perspectives on Silence, D.Tannen and M.Saville-Troike (eds.), Norwood,
NJ: Ablex.
. 1980b. "The deployment of consciousness in the production of a narra
tive." In: The Pear Stories: Cognitive, Cultural, and Linguistic Aspects of
Narrative Production, Wallace Chafe (ed.), Norwood, NJ: Ablex.
. 1982. "Integration and involvement in speaking, writing, and oral litera
ture". In: Spoken and Written Language: exploring orality and literacy,
Deborah Tannen (ed.), Norwood, NJ: Ablex.
1985. "Linguistic differences produced by differences between speaking
and writing." In: Literacy, Language, and Learning: the Nature and Conse
quences of Reading and Writing, David R.Olson, Nancy Torrance, and
Angela Hildyard (eds.), Cambridge: Cambridge University Press.
1987a. "Cognitive constraints on information flow." In: Coherence and
Grounding in Discourse, Russell Tomiin (ed.), Amsterdam: John Benja
mins.
. 1987b. "Punctuation and the prosody of written language." Technical

Report No. 11. Center for the Study of Writing, University of California,
Berkeley.
. and Jane Danielewicz. 1987. "Properties of spoken and written lan
guage." In: Comprehending Oral and Written Language, Rosalind
Horowitz and S.J.Samuels (eds.), New York: Academic Press.
Crystal, David. 1975. The English Tone of Voice. London: St.Martin.
Gossen, Gary H. 1984. Chamulas in the World of the Sun. Time and Space in
a Maya Oral Tradition. Prospect Heights, IL: Waveland Press.
Grimes, Joseph E. 1975. The Thread of Discourse. The Hague: Mouton.
Haiman, John, and Sandra A. Thompson. 1984. "'Subordination' in universal
grammar." In: Proceedings of the Tenth Annual Meeting of the Berkeley
Linguistics Society, Claudia Brugman and Monica Macaulay (eds.).
Halliday, Michael A. K., and Ruqaiya Hasan. 1976. Cohesion in English.
London: Longman.
Hymes, Dell. 1981. In Vain I Tried to Tell You. Philadelphia: University of
Pennsylvania Press.
Kroll, Barbara. 1977. "Combining ideas in written and spoken English: a look
at subordination and coordination". In: Discourse across Time and Space,
Elinor O.Keenan and Tina L.Bennett (eds.), Southern California Occa
sional Papers in Linguistics 5.
McLendon, Sally. "Eastern Pomo oratory." Paper presented at American
Anthropological Association Annual Meeting, 1985. Ms.
Mithun, Marianne. "Contrasting bilingual narrative strategies." Paper pre
sented at American Anthropological Association annual meeting, 1985.
Ms.
Pawley, Andrew, and Frances H. Syder. 1983. "Natural selection in syntax:
notes on adaptive variation and change in vernacular and literary gram
mar." Journal of Pragmatics 7: 551-579.
Sasse, Hans-Jürgen. "The thetic/categorical distinction revisited." Ms.
Schiffrin, Deborah. 1987. Discourse Markers. Cambridge: Cambridge Univer
sity Press.
Schourup, Lawrence C. 1985. Common Discourse Particles in English Con
versation. New York: Garland.
Svartvik, Jan. 1980. "Well in conversation." In: Studies in English Linguistics:
for Randolph Quirk, Sidney Greenbaum, Geoffrey Leech, and Jan
Svartvik (eds.), London: Longman.
A syntactic correlate of topicality
in Newari narrative
Carol Genetti
University of Oregon
This paper examines an alternation in case marking of subject NPs in

clause chains found in narrative texts in Newari, a Tibeto-Burman language
spoken in Nepal.1 Briefly, the alternation is found in clause chains with a
single co-referential subject mentioned once at the beginning of the chain. If
the chain has one or more clauses with a transitive verb, and one or more
clauses with an intransitive verb, then the case marking on the "subject" (i.e.,
the A or the S, to use Comrie's (1978) notation) can be either ergative or
absolutive. A full description of the construction will be given below.
I will show that the case marking alternation cannot be accounted for syn
tactically or semantically, but is controlled by discourse pragmatics. After sur
veying other pragmatic factors which might be expected to influence case
marking, I will demonstrate that the case marking correlates with the relative
topicality of the non-subject NPs in the two clauses, with the subject marked
ergative when the object of the transitive clause is more topical, and absolu
tive when an NP (typically a locative adverbial) in the intransitive clause is
more topical than the object of the transitive clause. We will see that the more
frequent reference to a participant is in the surrounding discourse, the more
likely it will be that a clause containing that participant will control case mark
ing. Following Givón (1983), I take frequency of reference to be a measure of
relative topicality, and since topicality of the arguments of a clause is one fac
tor which determines its importance, the results of the study show that this
alternation reflects the relative discourse importance of the various clauses in
a chain.
30 CAROL GENETTI
The problem
Newari is an ergative language with dominant SOV word order although

the order of subject and object NPs may be reversed due to pragmatic factors.
The ergative morpheme -n is often realized as a lengthened and nasalized
final vowel. The absolutive morpheme is unmarked. The following examples
illustrate the ergative/absolutive alternation:
(1) wƆ chē -ē won-Ɔ
he house-LOC go -PAST
'He went to the house'
(2) duku syat-Ɔ
he -ERG goat kill -PAST
'He killed the goat'
There is a split in the ergative pattern of Newari, such that the ergative
morpheme is obligatory in the perfective, but optional with the progressive
aspect. Thus example (3) is grammatical with or without the ergative mor
pheme:
(3) Ava(-ã) ja thuy-a cwon-Ɔ
Ava(-ERG) rice -PTCPL stay-PAST
'Ava is/was cooking rice'
Like other Tibeto-Burman languages, Newari is a clause chaining lan
guage, although much less so than the New Guinea languages which have
made this phenomenon famous (McKaughan, 1973; Franklin, 1983). Syntacti
cally, clause chains are marked by the verb of a non-final clause taking a par
ticiple form with a lengthened vowel (glossed in the examples as NF); the final
verb in the clause chain is finite. The clauses in such a chain often share the
same subject, which can be deleted in any or all of the clauses:
(4) chū -ū gohar yan-a -a ji-tƆ sya-e -mƆ -te dhal-Ɔ
mouse-ERG help do -PTCPL-NF I -DAT kill-INF-NEG-PROH say-PAST
'The mouse cried 'Help!' and said 'Don't kill me"
As in this example, this construction is often functionally equivalent to Eng
lish conjunction with implied sequential ordering.
For our purposes, adverbial subordinate clauses, marked by any of a
number of subordinators usually suffixed to finite verb forms, will be consid
ered, along with the non-final form, to constitute non-final clauses in a clause
chain. Complement clauses will be treated as dependent on the complement-
NEWARI TOPICALITY 31
taking verb and not as independent clauses in a chain. Auxiliary verbs will be
treated as part of the clause of the verb which they modify, and not as con
stituting separate clauses.2
The problem we are dealing with here has to do with the case marking on
shared subjects of clause chains.
(5) If the following conditions occur:
i. two or more clauses are chained;
ii. they share the same subject;
iii. that subject is mentioned only once, at
iv. the beginning of the first clause;
v. and if the verb of one clause requires an ergative marked NP,
while the verb of another clause requires absolutive subject
marking;
Then
the subject NP can be marked either as ergative or absolutive.
Consider the following example:
(6) a. wƆ duku syan-a -a won-Ɔ
he goat kill -PTCPL-NF go -PAST
'He killed the goat and left'
b. w- duku syan-a -a won-Ɔ
he-ERG goat kill -PTCPL-NF go -PAST
'He killed the goat and left'
This clause chain has a co-referential subject NP which in the first example is
case marked with respect to the intransitive verb won- 'go', and in the second
example with respect to the transitive non-final verb syan- 'kill'. There is no
intonational difference between the two sentences.
While ergative case marking in Newari is sensitive to semantic factors, in
particular vohtionality, this is not what governs the alternation in clause
chains. Both of these sentences could be used to describe the same event.
One could hypothesize a syntactic difference between the two sentences,
and claim that case marking identifies a main clause, as opposed to an adver
bial subordinate or a conjoined clause. Thus the structure of (6a) would have
wƆ won-Ɔ 'he left' as the main clause, and duku syan-a-a 'killed the goat' as an
inserted adverbial clause with a deleted co-referential subject. Then the
appropriate translation would be 'He, having killed the goat, left', which my
consultant does accept.
32 CAROL GENETTI
Sentence (6b) on the other hand, would identify w- duku syan-a-a 'he
killed the goat' as the main clause, leaving won-Ɔ 'left' as the functional equiv
alent of a conjoined clause with a deleted co-referential subject. The transla
tion in this case would be as above 'He killed the goat and left'.
The first problem with this analysis is that there is no other evidence for
this distinction. There is no morphological or syntactic behavior which distin
guishes the two constructions. Also, there are other grammatical devices for
specifically subordinating clauses as adverbial, and for conjunction.
However, the real problem with this analysis comes when we consider
clause chains of more than two clauses in length:
(7) w- gam -e woy -a -a duku syan-a -a won-Ɔ
he-ERG village-Loc come-PTCPL-NF goat kill -PTCL-NF go -PAST
'He came to the village, killed the goat, and left'
In this sentence, the subject is case-marked with respect to the middle of three
clauses. By the syntactic hypothesis, this would be the main clause, thas the
first clause would be considered subordinate, and the last clause conjoined.
There is no other syntactic distinction that would justify such an analysis.
There is also a problem with respect to sentences like (8):
(8) wo gam -e woy -a -a duku syan-a -a won-o
he village-Loc come-PTCPL-NF goat kill -PTCPL-NF go -PAST
'He came to the village, killed the goat, and left'
In this case, where there are two intransitive clauses either of which could
control the absolutive marking of the subject, the syntactic hypothesis must
claim that either the first or the last clause is the main clause and the other
conjoined or adverbial; however there is no way to determine which of the
two intransitive clauses should be considered the main clause. Thus we cannot
claim that case marking on the subject is determined by the syntactic main
clause. (In fact, in the absence of clear syntactic evidence for some other con
clusions, the obvious morphosyntactic criterion of finite verb inflection clearly
dictates that we consider the final clause to be the 'main' clause in all of these
examples).
The preceding hypothesis assumes that the subject NP belongs syntacti
cally to only one of the clauses, from which it inherits its case marking.
Perhaps a more appropriate way of conceptualizing the construction is as a
chain of clauses with one subject which precedes the chain. Then the case
marking alternation would not be controlled by any syntactic or semantic fac
tors, but would be pragmatically determined.
This is consistent with my informant's intuition about such sentences,

which he considers to be distinguished by presupposition. He says, for exam
ple, that (6a) is an answer to the question 'What did he do after he killed the
goat?'. Whereas (6b) answers the question 'What did he do before he left?'.
Thus the native speaker's intuition is that the clause whose verb controls case
marking is the new or asserted information, while the other clause is presup
posed. This paper examines not to what extent the various arguments in the
sentence are presupposed, which is a strictly anaphoric notion, but to what
extent they are topical. Thus a preliminary hypothesis is that the clause which
controls case marking is more salient or important in the discourse than a
clause which does not.
In order to test the validity of such a broad hypothesis, it is necessary to
look at a number of possibly interacting variables. I considered the possible
relevance of transitivity, clause position, sequencing, episode boundaries,
topic continuity, and topicality to the problem. The methodology will be dis
cussed in more detail below. Suffice it to say now that analysis of the results
led to a much narrower hypothesis:
(9) Hypothesis. A clause which contains an NP which is highly topical
for the surrounding discourse will be more likely to
control case marking than a clause with a less topical
NP.
Methodology
Data
The data used for this research came from four different sources. First, I
used thirteen spoken Newari texts, which I collected from one consultant dur
ing the summer and fall of 1984. These totalled fifty-four pages, glossed and
translated. All the texts are narrative. Most are fables and folk stories, while
three recount first hand experiences of the consultant. I also used the pub
lished texts of Hale and Hale (1970) to obtain further data on spoken Newari
discourse. Their series of thirty-six texts was taken from five consultants.
However, two of the consultants gave them the majority of the data; these
two both used the construction I am examining, while the other three did not.
Thus the spoken data that I used comes from three different consultants.
My data also include a considerable amount of written material: a book
of short children's stories and fables (Lal, 1962), and two long folk stories from
34 CAROL GENETTI
Baidhya (1982). All the texts are narrative, so no generalization from this to
other types of Newari discourse can be made.
Examples
Examples were chosen based on the criteria mentioned in the discussion

of the problem above and repeated here.
(10) i. two or more clauses are chained;
ii. they share the same subject;
iii. that subject is mentioned only once, at the beginning of the first
clause;
iv. and the verb of one clause requires an ergative marked NP,
while the verb of another clause requires absolutive subject
marking.
These initial examples then had to be culled for a number of reasons.
First, since Newari splits the ergative system based on aspect, a transitive
clause in the progressive could have either an ergative or an absolutive sub
ject, so that in a sentence like (10), there is no way to decide which of the two
clauses governs case marking on the subject NP:
(11) wƆ woy -a -a la khun-a cwon-Ɔ
he come-PTCPL-NF meat -PTCL stay -PAST
'He came and was cooking meat'
Thus I did not consider examples with progressive transitive verbs if the sub
ject was marked absolutive. If the subject was marked ergative, however, the
case marking could only reflect the transitivity of the progressive clause; thus
such examples were retained.
Second, this study utilizes a diagnostic for topicality of NPs which corre
lates with control of case marking. However, not all verbs which require erga
tive case marking have NPs which could be considered topical participants in
the narrative. There are three subgroups of clause types which fall into this
category. First, there are clauses with cognate object verbs, which consist of a
nominal or verbal morpheme, followed by a semantically bleached verb, usu
ally yan- 'do', such as sapha yan-a 'clean', khya yan-a 'joke', ujur yan-a 'com
plain', or dukh biy-a 'tease' (lit. 'give trouble'). Second, there are some syn
tactically intransitive verbs which, for semantic reasons irrelevant here,
require ergative subjects. An example of this is the verb hal- 'cry out'. Third,
there are many clause chains whose transitive verb is the quotative dha- 'say',
which takes as its object a direct quote. Since these three categories of clause
types do not have topical participants, I have excluded them from this study if
they were the only clause in the sentence which could take an ergative subject.
Further work needs to be done to evaluate what controls case marking in such
instances. After imposing these restrictions, I was left with fifty examples.
Table 1 shows the distribution of the examples by number of clauses.
Table 1. Number of clauses in examples

2 Clauses 29
3 „ 16
4 „ 3
5 „ 2
50
Counts
A number of hypotheses were tested for correlation with the case mark
ing alternation. Specifically, whether there is a preference for transitive versus
intransitive clauses, for in- versus out-of-sequence clauses, for clauses in a
particular position in a chain, or for adjacency to episode boundaries.
Case marking and position

In order to determine whether the syntactic structure of the chain plays a
role in the phenomenon, I counted the number of times which the final verb
controlled case marking and the number of times that the subject NP was
marked ergative versus absolutive. The results presented in Tables 2 and 3
show that the reasons for the alternation were more complex than a simple
syntactic rule.
Table 2. Case marking on subject NP

Ergative 29
Absolutive 21
Total 50
Table 2 shows that the distribution of ergative and absolutive morphemes

36 CAROL GENETTI
on subject NPs is fairly even, i.e. that there is no intrinsic preference for
marking with respect to transitive or intransitive clauses in a chain.
Table 3. Position of clause controlling case marking

Initial 16
Medial 3
Final 19
Total 38
Table 3 shows the distribution of data with respect to whether the clause
controlling case marking was in initial, medial or final position in the chain.
For twelve examples, all of which had more more than two clauses, it was not
clear which clause controlled case marking. For example, if a three clause
chain had two transitive verbs and one intransitive verb, it would not be obvi
ous which of the clauses with a transitive verb controlled ergative case mark
ing on the subject. Thus these examples are not represented in Table 3.
The results of the counts concerning clause position show too even a
spread to be able to account for the case marking alternation, i.e., there is no
preference for marking with respect to final versus non-final, initial versus
non-initial, and so forth. The low number of examples with medial clauses
clearly controlling case marking is expected considering the high number of
examples with only two clauses, and the indeterminate nature of many multi-
clause examples.
In-sequence / out-of-sequence
Recent work on the difference between foreground and background
information in discourse has shown that this distinction is based on a combina
tion of independent features (Givón, 1985). Myhill and Hibiya (this volume)
list four features which they consider to determine whether a clause in a
clause chain codes foreground or background information. These features are
+ / - human subject, same/different subject as preceding and following
clauses, perfective/imperfective, and in-sequence/out-of-sequence. Of these,
only the last one can be used to rank clauses within the clause chains chosen
for this study. The results of these counts are presented in Table 4.
NEWARI TOPICALITY
Table 4. In-sequence/ out-of-sequence

All clauses have equal sequencing value 35
In-Sequence clearly controls 5
Out-of-Sequence clearly controls 6
Either could control 4
Total 50
The results of these counts clearly show that this distinction does not cor
relate with the Newari phenomenon being examined. First, in thirty-five of
the examples all the clauses had the same sequencing values. In the eleven
examples which had one clause of each type, six in-sequence clauses con
trolled case marking, as opposed to five out-of-sequence clauses. In the
remaining four examples, all of which had more than two clauses, there was
no way to determine whether an in- or out-of-sequence clause controlled case
marking. The spread of results clearly indicates that the in-sequence/out-of-
sequence distinction does not correlate with the clauses controlling case mark
ing.
Episode boundaries
Episodes were defined as any continuous portion of the text which elabo
rates a particular theme of the story. Episode breaks, meaning any change in
time, place, or participant, were used to specifically delimit the endpoints of
episodes.
The counts of adjacency to episode boundaries similarly show that this is
not a pragmatic correlate of clauses controlling case marking. Of the fifty
examples, twenty-nine were directly adjacent to episode boundaries, whereas
twenty-one were not. The results are presented in Table 5.
Table 5. Distance from episode boundaries

Adjacent 29
Not Adjacent 21
Total 50
Since the counts for transitivity, clause position, sequencing, and episode
boundaries all showed no correlation with the case marking alternation in
38 CAROL GENETTI
question, I counted topicality and topic continuity in order to examine the

influence of these factors of clausal importance.
Topicality and topic continuity

I could find no direct way of measuring clausal importance, but
hypothesized that it would tend to correlate with topicality of NPs within
clauses. If a clause contains a more topical participant for the immediately
surrounding discourse, then it will most likely code a more important event
than a clause containing a less topical participant. Conversely, if a narrative is
primarily concerned with one or two participants, much of the action of the
narrative will involve them, so reference to them will be made frequently.
Furthermore, the greater the number of references to a given participant, and
the smaller the distance between those references, then the tighter the textual
coherence will be.
On the basis of these assumptions, we can empirically assign a measure of
topicality to NPs by using the methodology developed by Givón (1983). Thus
I arrived at my diagnostics for topicality and topic continuity. A participant
will be said to be of 'higher topicality' if it occurs more frequently in the dis
course than another, hence less topical, participant. A participant will be said
to be a 'more continuous topic' if reference to it is made at smaller intervals
than to another participant. For the purposes of this research, this diagnostic
allows us to assign numerical degrees of topicality and topic continuity to par
ticipants, and thus rank the relative importance of clauses in discourse.
Having defined explicit means for counting clauses in Newari, it was then
necessary to set an arbitrary limit on the number of clauses to count in either
direction. For the continuity counts the reason behind this is obvious —
namely that beyond a certain number of clauses reference to a participant is
discontinuous. For the topicality counts two main reasons exist. One reason is
that a given topic may not be mentioned at all in a large portion of the text.
Another reason is the notion, yet to be empirically justified, that the speaker
and hearer at any point in the text are connecting and relating information
primarily to the immediately preceding — and for the speaker the
immediately following — discourse, as opposed to the whole of the narrative.
(For a fuller discussion see Givón (1983:12-15).) Therefore I chose to count
the ten clauses preceding and following each example. Thus the examples
were considered in a twenty clause environment. Direct and indirect quotes
are considered in this count. Upon examination it turned out that none of the
quotes in question contained material which diverged significantly from the
immediate action of the story, or the immediate participants. Thus participant

continuity and topicality will be maintained throughout.
To assess topicality ratings of a participant I simply counted the number
of references to that participant in the surrounding clauses, regardless of the
coding strategies or grammatical relations. This gives a raw number which I
assigned as the topicality value of the participant. To assess the topic con
tinuity ratings of participants, I counted the number of clauses to the next ref
erence to the participant and assigned a rating from 1.0, for one clause dis
tance, to 0.0 for no mention in the the ten clauses.3
Since the subject of both clauses was co-referential, counting the degree
of topicality and topic continuity of the subject would give no clue as to the
relative importance of clauses. Thus I instead counted the direct objects of
clauses with transitive verbs, and locative arguments of intransitive verbs.
These results will be presented in the left part of each of Tables 6 through 12
under each heading. The columns represent the number of times the object
and locative arguments had the highest topicality rating. These are correlated
with the rows, which represent the number of times the case-marking on the
subject was ergative or absolutive. This table does not include cases where, the
topicality ratings for both arguments were identical.
Following this tabulation the values of each example were compared to
see if one score was greater than another. If not, I counted the example as
'not-applicable (N/A)' since the method provides no basis for judging relative
importance. If there was a difference in topicality or topic continuity ratings,
I then checked to see if the clause whose participant had the greatest value
also controlled case marking. If it did, this was counted as a 'yes', if not, a
'no'. Thus the results presented in the right hand part of each table under each
heading were obtained.
Results and analysis
Topicality
The topicality ratings of participants were compared in three ways; one
was for the preceding ten clauses, one for the following ten clauses, and one
for the surrounding twenty clauses. Of these, the best correlation was found in
the twenty surrounding clauses. That is, the highest percentage of clauses that
both controlled case marking and contained the more topical participant was
found in this count. The results are displayed in Table 6.
40 CAROL GENETTI
Table 6. Frequency of reference in surrounding 20 clauses
TR OBJ INTR LOC # %
ERG 25 3 Yes 38 76.00

No 5 10.00
ABS 2 13 N/A 7 14.00
Total 50 100.00
t=9.284 (49 d.o.f.) ρ < .0005
Thus in a total of fifty examples, thirty-eight, or 76%, correlated as

expected. Five examples, or 10% did not follow this pattern, and seven
others, 14%, left no basis for judgement.
A surface inspection of these results seems to indicate a significant trend.
However, the low number of examples lends doubt to surface inspection.
Therefore these results were subjected to a statistical analysis called a t-test.
This compares the standard mean and deviation of the results with that of the
null hypothesis calculated on random probability. The test indicates a ninety-
nine percent certainty that the results obtained are different from the null
hypothesis. In other words, we can be ninety-nine percent certain that the cor
relation between topicality and case marking was not due to chance in the
selection of the examples.
The results are even more convincing if we omit from the statistics the
seven examples for which the diagnostic left no basis on which to judge corre
lation. For three of these seven examples, the topicality ratings for the object
and locative were identical. In this case, the diagnostic I am using fails to iden
tify one clause as more important than another. In three other examples, the
intransitive verb had no implied locative, so we could not compare topicality
ratings of arguments. And for the last example, the object NP was co-referen
tial with the locative NP. Again there was no basis for comparison. If these
are excluded from the total, then in thirty-eight out of forty-three cases, or
88%, the clause controlling case marking also contains the most topical partic
ipants.
As stated previously, the case marking alternation correlated most favor
ably with the topicality counts for the surrounding twenty clauses. However,
other counts did have interesting results. First, Table 7 shows that the exam
ples from spoken discourse had higher correlations than those from written
discourse:
Table 7. Frequency of reference in surrounding 20 clauses: spoken vs. written
SPOKEN
TR OBJ INTR LOC # %
ERG 18 2 Yes 30 81.08

No 4 10.81
ABS 2 12 N/A 3 8.11
Total 37 100.00
WRITTEN
TR OBJ INTR LOC # %
ERG 8 1 Yes 8 61.54

No 1 7.69
ABS 0 0 N/A 4 30.77
Total 13 100.00
Table 8 breaks down the examples of the spoken data by speaker. From
one of the speakers, NR, we have only two examples. But from the other two
speakers, there is enough data to indicate that the correlation is present for
both of them.
42 CAROL GENETTI
Table 8. Frequency of reference in surrounding 20 clauses

By speaker — spoken texts only
Speaker RS
TR OBJ INTR LOC # %
ERG 12 0 Yes 19 79.17

No 2 8.33
ABS 2 7 N/A 3 12.50
Total 24 100.00
Speaker NR
TR OBJ INTR LOC # %
ERG 2 0 Yes 2 .100.00

ABS 0 0
Speaker JNM
TR OBJ INTR LOC # %
ERG 4 0 Yes 9 81.81

No 2 18.19
ABS 2 5 N/A 0 0.00
Total 11 100.00
I also calculated the results for discourse preceding the examples, and
compared them with those for the discourse following. These results are pre
sented in Tables 9 and 10.
Table 9. Frequency of reference in preceding 10 clauses
TR OBJ INTR LOC # %
ERG 16 0 Yes 26 52.00

No 2 4.00
ABS 2 10 N/A 22 44.00
Total 50 100.00
Table 10. Frequency of reference in following 10 clauses
TR OBJ INTR LOC # %
ERG 24 2 Yes 31 62.00

No 5 10.00
ABS 3 7 N/A 14 28.00
Total 50 100.00
The results for the discourse following the example correlated with slightly
higher percentages than those of the preceding discourse. While this may indi
cate a tendency for cataphoric processing, the percentages are not strong
enough to substantiate this conclusion. I will mention that for one speaker,
RS, there seems to be a much stronger tendency for basing topicality on the
following discourse than for the others. Thus the difference between
cataphoric and anaphoric processing in determining clausal importance may
be speaker-dependent.
Topic Continuity
The results for the topic continuity counts did not show the high correla
tions of the topicality counts. This suggests that distance between references
to a given participant is not as important in determining clausal importance,
44 CAROL GENETTI
hence determining case marking, as is the frequency of reference. The topic

continuity counts also differed from the topicality counts in that preceding and
following distance correlated with case marking about the same percent of the
time. The results are presented in Tables 11 and 12.
Table 11. Distance of reference in preceding 10 clauses
TR OBJ INTR LOC # %
ERG 13 4 Yes 27 54.00

No 7 14.00
ABS 3 14 N/A 16 32.00
Total 50 100.00
Table 12. Distance of reference in following 10 clauses
TR OBJ INTR LOC # %
ERG 19 5 Yes 26 52.00

No 11 22.00
ABS 6 7 N/A 13 26.00
Total 50 100.00
Conclusions
The results obtained do suggest that the hypothesis is correct, namely

that the clause with the most topical participant controls case marking.
Although the number of examples is admitedly small, the second hypothesis
was substantiated empirically, and the results are statistically significant.
The first hypothesis, that the most important or salient clause will control
case marking, cannot be empirically justified, since there is no independent
measure of importance that can be quantified. This hypothesis can be

accepted if one accepts the intuitive notion that a clause which contains a
more topical argument is more important than one which does not. The impli
cation of this for the problem at hand, is that the speaker has some notion as
to which clause will be more important in the surrounding discourse, and case
marks the subject NP accordingly.
Furthermore, the higher the topicality of a participant, the more it contri
butes to the coherence of the surrounding text. Since the case marking alter
nation does correlate with high topicality, it in turn correlates with a tighter
cohesive structure.
The results of this study allow for some interesting observations concern
ing the Hopper and Thompson (1980) transitivity hypothesis. While their
work is concerned with the interaction of clause level phenomena with dis
course factors, the work here looks at the relations between clauses in a sen
tence. Therefore their hypothesis cannot be cleanly applied to the problem at
hand. Nevertheless, some comments can be made.
Hopper and Thompson propose that the notion of transitivity is multi-
factoral, with a number of independent elements, each of which contribute to
the degree of transitivity of the clause. They predict that the higher the degree
of transitivity of a clause, the more likely it is that the clause will be syntacti
cally or morphologically marked as transitive. One of the transitivity features
is individuation of the object, which in turn has a feature 'referential, definite'
(my 'topical'). Their prediction that a clause with a more topical object is
more likely to trigger transitive morphology is supported by the results of this
study.
The interesting implications of this study for their hypothesis, come with
examination of the intransitive clauses. The results here clearly suggest that an
intransitive clause with a highly topical locative argument is more likely to
result in intransitive (absolutive) morphology. However, it seems that a highly
topical locative argument increases the importance of an intransitive clause, in
the same way that a highly topical object increases the salience/transitivity of
a transitive clause. One would like to say that an intransitive clause with a
topical locative is somehow more 'transitive' in the multi-factoral sense of
Hopper and Thompson's theory, than an intransitive clause with no locative at
all. (This is supported, for example, by data from Nez Perce (Rude, 1986),
where highly topical locative arguments are promoted to direct objects and
trigger transitive morphology in a number of ways throughout the sentence.)
Yet in Newari, clauses with topical locatives do not result in more transitive
46 C A R O L GENETTI
morphology, but promote a stronger tendency for the subject to be marked by

the absolutive case. Since this is an interclausal phenomenon, we cannot
relate it directly to the transitivity hypothesis. However, the results of the
study do support their general position that transitive morphosyntax is dis
course dependent.
To summarize, this paper has presented an empirical study examining the
relationship between subject case marking and clausal importance in Newari
clause chains. The results indicate that topicality of non-subject arguments
plays ,an important role in determining which clause controls the ergative/
absolutive distinction. In addition, the correlation between topicality and case
marking is stronger in spoken discourse, as opposed to written discourse.
While one subject showed a strong tendency for cataphoric processing of topi
cality, the other speakers did not.
There is a clear intuitive relationship between clauses with highly topical
arguments and the general concept of clausal importance. Therefore the
results suggest that the speaker case marks the subject of a sentence in accor
dance with the most important clause. And, since subject case marking corre
lates with topicality of non-subject arguments, case marking thus contributes
to a more coherent textual structure.
In general, the field of interclausal relations is yet to be adequately map
ped. It seems that further work in this area will both rely on, and contribute
to, our currently limited knowledge of the relationship between grammar and
discourse.
Notes
1. This research was supported in part by the National Science Foundation, grant BNS-
8313502, and by a grant from the Joint Committee on South Asia, of the Social Science
Research Council and the American Council of Learned Societies, with funds provided by
the Ford Foundation and the National Endowment for the Humanities.
I wish to thank my consultant, Raj endra Shrestha, for his patience and cooperation. I
would also like to thank Steven Robeson for his advice on statistics. Scott DeLancey, Tom
Givón, Sandy Thompson, David Hargreaves and others have offered valuable advice. All
errors and inconsistencies are my own.
2. David Hargreaves (p.c.) has pointed out that there is variation between speakers in the mor
phological coding of clause chains, and the syntactic repercussions of various forms. Thus
some speakers appear to distinguish between two clause chaining constructions, although my
consultant clearly does not.
3. I am using slightly different terminology than that proposed by Givón (1983) and used in
other studies of this type. The usual term applied to counts of frequency of reference is 'per-
sistence', but since this often applies only to the text following the example in question, I
decided to use 'frequency' instead. Similarly 'topic continuity' is sometimes used to refer
only to anaphoric reference. I chose 'distance' to include cataphoric reference as well.
References
Baidhya, K. 1982. Nepa-yagu Pulã Bakh {Folk Tales of Nepal), 1.

Kathmandu: Man Das Sugat Das.
Comrie, Bernard. 1978. "Ergativity". In: Syntactic Typology, W. Lehmann
(ed.). Austin: University of Texas Press.
Franklin, K.J. 1983. "Some features of interclausal reference in Kewa". In:
Switch Reference and Universal Grammar, J. Haiman and P. Munro,
(eds.). Amsterdam and Philadelphia: John Benjamins.
Givón, T. (ed.) 1983. Topic Continuity in Discourse: Quantitative Cross-Lin
guistic Studies (= Typological Studies in Language, 3.) Amsterdam and
Philadelphia: John Benjamins.
1984. Syntax: A Functional-Typological Introduction, 1. Amsterdam and
Philadelphia: John Benjamins.
1985. "Foreground and background". In: Coherence and Grounding in
Discourse (= TSL, 11), R. Tomiin, (ed.). Amsterdam and Philadelphia:
John Benjamins.
Hale, A. and M. Hale. 1970. "Newari texts". In: Tone Systems of Tibeto-Bur-
man Languages of Nepal. Part IV: Texts, II (= Occasional Papers of the
Wolfenden Society on Tibeto-Burman Linguistics, 3), F.K. Lehman, (ed.).
Urbana: University of Illinois.
Hopper, P. 1979. "Aspect and foregrounding in discourse". In: Discourse and
Syntax (= Syntax and Semantics, 12), T. Givón, (ed.). 213-42. New York:
Academic Press.
Hopper, P. and S. Thompson. 1980. "Transitivity in grammar and discourse".
Lg 56:251-99.
Lal, J. 1962. Moca Bakh (Child Stories). Kathmandu: Ratna Pustak Bhan-
dar.
McKaughan, H., (ed.). 1973. The Languages of the Eastern Family of the East
New Guinea Highland Stock (= Anthropological Studies in the Eastern
Highlands of New Guinea, 1). Seattle: University of Washington Press.
Rude, N. 1986. "Topicality, transitivity and the direct object in Nez Perce".
UAL 51:2.
48 CAROL GENETTI
Tomlin, R. 1985. "Foreground-background information and the syntax of

subordination". Text 5(5):85-122.
1986. "The identification of foreground-background information in on
line oral descriptive discourse.". Papers in Linguistics 19(4): 465-494.
Inconsequential clauses in Hua
and the typology of clauses
John Haiman
Hua narrative, like that of many other Papuan and other languages, is
characterized by extensive clause chaining within the bounds of a prosodic
sentence. Medial (or sentence-internal) verb forms differ from final (or inde
pendent) verb forms, and may be of various types. A characteristic which
Hua shares with hundreds of other languages in New Guinea and elsewhere
is the marking of switch-reference on medial verbs, which indicates (roughly)
whether or not the subject of the verb in question is identical with that of
the following verb in a multi-clause chain. Both of the sentences in (1) below
could be translated as 'S/he shouted and s/he went outside', but in the first,
with a different-subject medial, the one who shouts and the one who goes
outside are different people; in the second, the one who shouts and the one
who goes outside are one.
(1) a. are dare higana hairga frufie.
b. are dare hurona hairga frufie.
A further characteristic shared with a smaller all still unknown number of
Papuan languages is a distinction between 'coordinate' and 'subordinate'
medial clauses. While coordinate medial verbs (like those of (1)) obligatorily
mark switch reference, subordinate medial verbs cannot and do not. Thus
sentence (2) below is ambiguous between the readings of (la) and (lb), while
differing from them in other ways as well:
(2) are dare bimana hairga frufie.
Finally — and this characteristic relates Hua to a rather small number of
closely related languages spoken around Goroka and Kainantu — and
medial verbs, whether coordinate or subordinate, whether different-subject
50 JOHN HAIMAN
coordinate or same-subject coordinate, are obligatorily followed by an an

ticipatory subject desinence which agrees in person and number with the sub
ject of the following verb. In all of the examples just given, this anticipatory
desinence -na signals that the subject of the following verb is third person
singular. In some medials (those of (la) and (2)), the anticipatory desinence
follows a medial desinence which agrees with the subject of the medial verb
(3sg. coordinate -ga-, 3sg. subordinate -ma-) while in coordinate same-
subject medials, like that of (lb) there is no medial desinence at all. Compare
the medial clauses when the final verb is frufane 'you went outside':
(3) a. are dare higa-ka hairga frufane. (Coordinate different-
subject)
'S/he shouted and you went out.'
b. are dare huro-ka hairga frufane. (Coordinate same-subject)
'You shouted and you went out.'
 are dare hima- hairga frufane. (Subordinate)
'S/he shouted and you went out.'
(Note that in (3c), it is clear that the subject of the medial clause and of the
final clause are distinct. But this is marked not by formal switch reference,
whereby identity of subjects is signalled by the absence of the medial desi
nence. Rather, it is marked incidentally, inasmuch as the person and number
of the medial subject are different from that of the following subject — and
thus, a fortiori, the two subjects must be distinct.)
An extremely powerful syntactic effect of the anticipatory desinence
(which is analogous to a 'pointer' in artificial languages like Pascal and
PL/1) is to morphologically reenforce the linear structure of Hua narrative.
Each medial clause literally points forward to the (subject of) the clause that
follows, thus establishing a morphosyntactic diagram something like
(4)
The medial clause chain, by far the most frequent clause chaining device in
Hua, is the standard means for the iconic expression of a variety of asymme
trical relations, among them the relations between cause and effect, protasis
and apodosis in ordinary (but not concessive!) conditionals, and, most gene
rally, anteriority in temporal succession.
The structure, however, is unsuitable for the expression of other rela
tions, in particular, for those where the event denoted by Sn does not (logi-
INCONSEQUENTIAL CLAUSES IN HUA 51
cally, or temporally) follow the event of Sn-1. Lacking conjunctions, Hua

uses the medial clause construction for 'and', 'and then', and 'and so'; but
it needs some other construction to handle the relationships typically expres
sed by English 'but'.
Hua has a variety of constructions for expressing these relationships. In
what follows, I will examine only one of these constructions: the inconse
quential. This is both a medial and a final construction with a wide range
of (to us) surprising and unexpected uses. A proper appreciation of its mea
nings will require us to consider some other constructions (including the
standard medial) whose distribution sometimes overlaps with that of the in
consequential. Finally, I will try to show how the inconsequential in Hua
poses insoluble problems for a number of categories which we take for gran
ted in our tradition, among them 'conditionals' and 'subordinate clauses'.
But first, a few words about the form itself.
Morphology
The inconsequential verb form consists of the verb-auxiliary complex
followed by the personal desinence -mana, which occurs in three forms:
i. lsg, 3sg., 2/3 pl. -mana (the unmarked form);
ii. dual -'mana(the dual form)
iii. 2sg., 1pl. -pana (the 'other' form).
Like the subordinate medial verb, it allows the future indicative auxiliary
-gu-, but not the future subjunctive auxiliary -su-: thus hi-ga- '2/3 pl.
will do (inconsequentially)', but not *hi-sa-mana '2/3 pl. may do (inconse
quentially)'. There are no other restrictions on the auxiliary combinations
that may occur with the inconsequential.
It differs minimally but crucially from the subordinate medial verb
form in two respects. First, the first syllable of the subordinate medial
desinential complex is invariably stressed, while the (homophonous) first
syllable of the inconsequential desinence never is. Thus, there is a phonetic
contrast between
(5) a. hímana (inconsequential; stressed on the verb)
'S/he said (inconsequentially)'
b. himána (subordinate medial; stressed on the medial
desinence)
'S/he said and s/he...'
52 JOHN HAIMAN
Second, while both desinential complexes are bimorphemic, the second

syllable of the subordinate medial desinential complex is an anticipatory
desinence which varies depending on the nature of the following subject. The
second syllable of the inconsequential, however, is the same irrespective of
the nature of the subject of the following clause, or even the existence of
such a clause. Thus (5b) belongs in a paradigm whose other members include
himaka '3sg. said and you...', himada '3sg. did and I...', himata'a '3sg. did
and the two of us...', and so on. No such variation is possible in the inconse
quential, which thus lacks the 'pointer' that is characteristic of standard
medials.
Semantically, the inconsequential differs from the subordinate medial
in two ways: unlike the subordinate medial, it may occur as a complete ut
terance without a following clause. However, if it is followed by another
clause within the prosodic sentence, then the relationship between the two
clauses cannot be translated by 'and', 'and then', or 'and so'. A relationship
of causal succession is almost totally denied. (We will return to the reasons
for this qualification below.)
The inconsequential is also opposed to the conditional (given or
hypothetical) protasis, which consists of the verb-auxiliary complex followed
by the desinence -mamo, as in
(6) a. hi - mamo (given conditional)
'Given that s/he said...'
b. hisi - mamo (hypothetical conditional)
'If s/he says...'
The morphological contrast lies in the second syllable only. The inconse
quential has -na, while the conditional has -mo. But there is some evidence
that the conditional desinence is itself bimorphemic, consisting of -ma,
followed by a nominalizing suffix -mo: this evidence comes from the
counterfactual conditional protasis, consisting of -hipa, followed by the
syllable -na. Both -mo and -na function elsewhere in the grammar of Hua
as nominalizers, and recognizing their identity allows us to relate
hypothetical with counterfactual conditionals, surely a desirable and
familiar identification. Moreover, once the functional similarity of -mo and
-na is admitted, the distinction between the inconsequential and conditional
clauses becomes morphologically trivial.
Even this rather minimal contrast is frequently suspended, a fact which
points to the deep and rather surprising kinship between these two clause
types. Semantically also, the contrast is minimal, and is confined to the

distributional distinction that the inconsequential can function as a complete
utterance, while the protasis cannot. Indeed, as we shall see, the conditional
'may replace the inconsequential in three of the four major functions I have
been able to identify.
These four functions of the inconsequential are
i. to mark actions taken in vain;
ii. to mark actions not as yet completed;
iii. to signal a change of speaker in a dialog;
iv. to mark verbs of perception preceding what we would call
their complements
In each of these functions, the inconsequential stands in a different
ecological relationship with competing forms which convey a similar mean
ing. It is from these different relationships that we can deduce the Gesamt-
bedeutung of both the inconsequential and the conditional, and contrast
them with the standard average medial and other forms of which we have
as yet said nothing.
The functions of the inconsequential
A. Actions taken in vain
It is only in this primal function that the inconsequential may occur as

a totally independent utterance. The action described in the inconsequential
clause is actually performed (that is, the inconsequential is not a conative
form), but the action did not lead to the desired or expected result, whatever
this is supposed to be. Frequently, the independent inconsequential is
followed by the optional clamative vocable -o, which also appears optionally
after imperatives, and proper names or kin terms in the vocative:
(7) a. hako - mana - ()
seek 1SG INCONS CLAM VOC
'I sought (but couldn't find)!'
'I looked (in vain)!'
b. Busa - ()
man's name CLAM VOC
'Busa!'
54 JOHN HAIMAN
. na bai - ()
thus stay (IMP SG) CLAM VOC
'Stay thus! Goodbye!' (said by person leaving an encounter)
As noted, a conditional could not replace the inconsequential in (7a),
but only because the conditional protasis is a dependent clause. As it hap
pens, the primal function of the inconsequential is carried out in sentence-
internal, or dependent, clauses as well. In the following examples, it is clear
from the context that the following clause after the inconsequential describes
an unexpected or unwished-for event:
(8) a. "Biga badeae!" hi - rgi' ve 'afie
up-there boy (voc) say 3SG INCONS really yeah? 3SG not = say
'He said "You up there!", but the other made no reply'
b. " A k u o ! " huna naroti' hi - nomo
elder-brother (voc) saying from-there say 3SG INCONS voice
'a'aiganahibo
3SG not make and 3SG then
'He called from there, saying "Elder brother!", but the
other made no sound and so he...'
 "Ma nagabo rgi' faota'a desu'e" ke hu -
this people really shooting let's eat word say 1SG INCONS
rgibo bura abademo gnumo kvamo huregitamo
really that girl person care she took and they
rgi' na ko'rimita
really that it dawned and they
kiko'itogamo vigamamone
to their homes indeed they will go
'I intended that we should shoot and eat these people but
that girl person was looking after them, and now it has
dawned and they will go to their homes.'
In the following example, the inconsequential desinence appears on on
ly the final verb in a string of verbs with the same subject. It is understood
(since the string is one of coordinate medial verbs) that the inconsequentiali-
ty is characteristic of all the verbs in the string. In this case, it is the action
of 'saying' which is inconsequential, since what is said is a lie which is not
believed (as the story makes clear):
(9) "Ai! Noviruo! zgahi' kaparó'zgaida rgi'

ai! daughter-in-law possum vine-on lurking really
kobo ribroe" huna vzahi -mana
I-saw-dawn saying (COORD SS) arrive 3SG INCONS
ninimo nago' navi' havaBo...apamo havaBo navi' ne
soot only-this in-there little mud little in-there is
' "Ai! Daughter-in-law! I lurked in wait for the possum by the
vine until dawn!", he said and arrived, but there was a little
bit of soot, a little bit of mud, in there.'
(In the folktale from which this passage is taken, a man goes up to a bush
campsite with his daughter-in-law. Telling her that he is going off to hunt
possum, he leaves her to fall asleep by the fire, and paints his face with soot
and mud to make it unrecognizable. He then returns to the campsite and
repeatedly has intercourse with her, leaving before dawn. After thoroughly
washing his face, he hopes that she will believe him when he claims to have
been absent hunting all night, but the trace of paint on his face gives him
away.)
Two final examples will provide our first evidence that the inconsequen
tial desinence - is indeed bimorphic. The second syllable - may be,
and regularly is, omitted in all cases where the inconsequential clause is
repeated:
(10) a. hakona hurmi -  hurmi - 
searching (3SG COORD MED) walk-about 3SG INCONS
'a'neganahibo
there-wasn't-any-and-so-he ...
'He walked about and walked about searching for it, but
there wasn't any and so he...'
b. hakona bai -  bai - 
searching (3SG COORD MED) stay 3SG INCONS
'a'eganahibo na kositamo ve
s/he-didn't-come-and-so-he that old man slept
'He stayed and stayed searching for her, but she didn't
come, and so that old man slept.'
The truncated inconsequential of (10) is morphologically identical with the
conditional, which also may lose its final syllable -mo under certain cir
cumstances.
56 JOHN HAIMAN
In all of the examples so far, I have translated the inconsequential as

'but', and it is important to note that this word has two uses in English:
(11) a. Inconsequential
I tried to lift the rock, but I couldn't.
They can ask me, but I will refuse.
He wrote her countless letters, but she never answered any
of them.
b. Antithetical
" I am a smooth man, but my brother Esau is a hairy man.
She answered some of his long letters, but didn't keep any
of them.
He is gone, but we remain.
In a number of languages, these two uses are translated by different
words or constructions. Hua is such a language: the inconsequential con
struction is limited to the representation of inconsequentiality illustrated in
the sentences of (11a). For antithesis, Hua employs a somewhat different
construction, in which the verb is followed by the desinential complex
-maborava, the exact analysis of which is problematic. In the purely an
tithetical usage of (1 lb) and (12), below, the inconsequential is (I believe) im
possible, or at least, not attested in over 200 pages of texts:
(12) Ai! nahita fumo ebgita deta nahita hisamihibai - maborava
ai thus pig killing eating thus they-are-about-to 2 / 3 P L ANTI
 dgaimo davugi' itagi' amo friramata'a
thus I dad-and mom-and they died-and-we-two
hava'amo baigu'mamone
empty-handed we-two-will-be-indeed
'Ai! They are about to kill and eat a pig, but I, with mom and
dad both dead, that is, the two of us, are left empty-handed.'
That is, the inconsequential cannot do duty for the antithetical function. On
the other hand, the antithetical in -maborava is able to function as an in
consequential, as in the following two examples. These examples are par
ticularly telling, as they come from alternate versions of the same myth from
which the inconsequentials of (9) are taken:
(13) a. Na hurmu - maborava  rgi' zgamo 'a'haoe

thus walk-about 1SG ANTI thus really possum I-didn'-t
shoot
'I walked about thus, but I didn't shoot any possums.'
b. ninimo aurga'avingamo aurga'amo frroprrufudi' ninimo
soot in-eye eye smeared soot
have' apamo hazona rgi' kvrguhuna rgi' sokohuna
white mud doing-right really scraping-off really well
ironahibo e - maborava bgota' aurga'a kaitapibo have'
getting rid come 3SG ANTI one eye in-corner white
hava'a ne.
nothing there-is
T h e soot smeared around his eyes and the white mud, he
carefully scraped it all off and got rid of it, and came back,
but in the corner of one eye there was a tiny bit of white
mud and soot left'
Thus, the antithetical adversative, in addition to marking antithesis, may
also serve to mark actions which have been undertaken in vain. However,
the antithetical in -maborava shares none of the other functions of the in
consequential. To these other functions we now proceed.
B. Actions not as yet completed
The limitations on the overlap between the antithetical and the inconse
quential may be seen most clearly when we compare the primal function of
the inconsequential with one which seems to differ from it so subtly as to
be indistinguishable. This is the function of denoting as yet fruitless or vain
activity. In this minimally contrasting function the adversative in -maborava
is never attested in any of my materials.
Prolonged or repeated activity is iconically indicated by a simple repeti
tion of the verb which denotes the activity: 'she talked and talked', and so
forth. A very common rhetorical device in Hua is to have a string of repeated
verbs conforming with the template:
(14) (V1INCONS)n (V1 MED)m V 2
where the first verb (V1) describes the repeated activity, and the second (V2),
the final successful outcome. It is striking that no alternative permutations
58 JOHN HAIMAN
of the inconsequential and the medial are attested other than those allowed
by the template of (14). For example, we never encounter a string of medials
followed by a string of inconsequentials, or a random alternation of medials
and inconsequentials. A semantic explanation of the existence of (14) is quite
straightforward, given the fundamental contrast between medial verbs and
inconsequentials. A prolonged activity may seem fruitless or inconsequential
for a long time. But, at long last, persistence pays off, there is 'light at the
end of the tunnel', and the desired outcome is clearly in sight. At this point,
the switch from inconsequential to medial verb forms for (V1) occurs.
As in its first function, the repeated inconsequential occurs in its trunc-
ted form here, that is, with the desinence -ma.
(15) kvina kemumo kvirina vi -  vi -  vi -
digging hole keep-digging go (3SG INCONS)
rgi' una una una
really going (3SG SAME SUBJ COORD MED)
rgi' bira Koigovu movibo rina
really there K. mountain-in taking
haDaupe
she-broke-through
'She dug a hole and kept digging and digging and digging and
digging until she finally tunneled to the surface and broke
through over there in Koigovu Mountain...'
The full form of the inconsequential (with the desinence - ) never seems
to occur in this function. However, the full form of the conditional protasis
(in -mamo) does:
(16) harorina zati zamu kegve' harorina rmi - mamo
flying river bank edge flying go-down (3SG COND)
rmi - mamo rmuna rmuna rmuna
going-down (3SG ss COORD MED)
rgibo hagogivamogi hagivamogi mnamo truhuremane
really tree SP and tree SP and bird it-perched!
T h e bird flew to the banks of the (Tua) river, going down,
down, down, down, until it alighted on the hagiva and the
hagogiva trees'
Strictly speaking, sentence (15) is 'ambiguous': the truncated verb form in

-ma is derivable from either the inconsequential or the conditional. I wish
to argue, however, that there is no ambiguity here, inasmuch as the inconse
quential and the conditional are virtually synonymous in this second func
tion of the inconsequential clause.
Having now witnessed two closely related functions in which the condi
tional (and no other form in Hua) may acceptably replace the inconsequen
tial, we turn to the only function in which it cannot.
C. Signalling a change of speaker
Hua represents not only speech but internal mental states by direct
quotes that function as the direct object 'complements' of the verb hu- 'do,
say'. Since Hua is a typical SOV language, these complements precede the
verb, as in
(17) "Dogue" hie
I-will-eat 3SG -says
'S/he says "I will e a t ' " (more typically, 'S/he wants to eat')
Reported dialogs have the structure
(18) Quote hu + inconsequential
Quote hu + inconsequential
Quote hu + inconsequential
The function of the inconsequential here is to signal a change of speaker in

a dialog. (If the same speaker continues, or if the last quote is followed by
some non-verbal action, the verb hu- occurs in the medial form.)
Structure (18) is frequent in mini-conversations which consist simply of
a question and answer:
(19) "Zahae?" hi - mana
what-do-2/3pL say 3SG INCONS
"nahuta huga'zamu hataimo ai'obaune."
doing-thus frogs bashing we-come-up
'What are you doing?'' he asked (and they answered) "We
have come up bashing frogs on the w a y " '
60 JOHN HAIMAN
The contrast between talk which leads to more talk (even of the sort which
advances the narrative), and talk which leads to action, is graphically il
lustrated in the interchange below:
(20) "Bgotako' ebgibroda ridai'obaue" huna hi - mana
only-one having killed I-am-bringing-up saying say 3SG INCONS
"Dmio!" huna higanahibo
give-me! saying he-said-and-she (COORD DS MED)
migana dobaie
she-gave-and-he (COORD DS MED) he-eats
'She said "I've only killed and brought up a single one", and
he said "Give it to me!" and she gave to him, and he ate it'
This function of the inconsequential seems unrelated to the first two: surely
a rejoinder which advances the narrative is not inconsequential. Yet it may
not be too fanciful to see embodied in the structure (18) a hardheaded world
view. Talk which leads to non-verbal 'action' is successful, while talk which
leads to more talk (even the answer to a question) is relatively fruitless and
hence, inconsequential. This sterniy pragmatic outlook is modified in two
ways. First, this is the only function of the inconsequential in which its
distribution overlaps, even partially, with that of the medial. In addition to
sentence types like (20) we also encounter (21):
(21) a.) "aitenemao!" higana
old woman! he said and she (COORD DS MED)
"Ve!" higana
yeah? she said and he (COORD DS MED)
"ora kipa dmio!"
that fire give-me!
'"Old woman!" he said. "Yeah?" she said. "Bring me that
fire!'"
b. "rmisumihuvao!" huna higana
I-want-to-go-down saying she-said-and-he (COORD DS MED)
"ega' a uka riroka eno!"
tomorrow going taking come!
'She said " I want to go down (now)" and he said "Go get
it tomorrow"'
Second, as noted, this is the only function of the inconsequential in which

it cannot ever be replaced by a conditional. We will return to the significance
of this restriction in our conclusion, where I will argue that the truly inconse
quential clause type in Hua (and perhaps in other languages also) is in fact
the conditional, nomenclature and traditional definitions to the contrary.
D. Marking verbs of perception
Given the syntactic representation of quotes, we might expect that Hua

treats verbs of perception in the same way; that is, what is perceived, like
what is spoken, is treated as the object complement of a verb, in this case,
a verb of saying, hearing, and so forth. This structure is possible in Hua,
but quite rare, and limited (it seems) to clauses which represent propositions
whose validity is presupposed. In these cases, the complement clause appears
with a head noun  "thing":
(23) eva' kutta'na ripa' - na- mo kgoe
money theft take (2SG REL) thing (NR) I-saw-you
Τ saw you stealing the money'
Sentences like (23) are acceptable only if the fact of your theft is already
common knowledge, and what I am reporting as news is only the fact that
I observed it. The sentence is impossible for the more usual Τ saw you steal
the money' or 'You stole it: I saw you', where my observation is what
validates the truth of what I describe.
In this more general case, a favourite construction is
(24) Verb of perception + inconsequential ... 'Complement'.
That is, the inconsequential signals that (something) of what follows is what
was perceived. There is no formal mark to signal the end of the 'comple
ment':
(25) a. ka - 'mana navibo rgi' 'a'baivao
look (2/3DU INCONS) in-there really they-aren't
'The two of them looked and saw that they really weren't
in there'
62 JOHN HAIMAN
b. havi - mana mina veta aimo bububabafita

listen (3SG INCONS) down-there sleeping shit farting
hikrufita fera'amo brave
snoring great they-put?
'He listened and heard them down below sleeping, farting
in their sleep and snoring mightily'
Frequently, the verb of perception is left out. What is left, and occurs with
the inconsequential desinence, is a verb of motion which denotes arrival at
a place from which perception is possible. In sentences of this sort, the in
consequential may be non-idiomatically (but I think happily) translated by
something like 'lo and behold' or a presentative sentence:
(26) vzahi - mana na kanabo na baimane
arrive 3SG INCONS that younger-brother that he-was!
'He arrived and lo and behold, his younger brother was there'
'He arrived and there was his younger brother'
Of all the regular functions of the inconsequential, this seems to be the most
exotic and inexplicable. In order to account for the appearance of the in
consequential here, it is again advisable to proceed, not from first principles,
but from a consideration of the other forms which can appear in these and
closely related constructions. From such a consideration, a (partially
satisfactory) account may follow.
First, it is notable that in most European languages, verbs of perception
are closely related, in their syntactic behaviour, with verbs of saying and
thinking and knowing. As we have seen, this is not the case in Hua. There
is, first, a gross difference in word order between complements of verbs of
saying and verbs of perceiving; and then, there is a difference in how these
verbs and their 'complements' are morphologically marked.
(27) a. 'QUOTE' verb of saying

b. Verb of perception +  'complement'
In both (27a) and (27b) the structure of the first component is independent
of the rest of the discourse, while the structure of the second depends entirely
on the nature and relationship of whatever follows. Are there any other con
structions which are relevant here?
We can immediately dispose of verbs of 'knowing'. Only one such verb
in Hua is in common use, and it is identical with the verb of perception k-
'see'. On the other hand, the common verb of thinking, keta havi- (literally,
'ear hear'), has a distinctive syntax:
(28) Verb of thinking + gasi' 'complement'
illustrated in a sentence such as
(29) d - geta havi - gasi' ogimamone
my ear hear GERUND 3SG-will-come-indeed
'I think s/he will certainly come'
This construction is possible for verbs of saying, as well. The semantic con
trast between (27a) and (28) in this case seems to be one of subjectivity:
(30) a. (Buro) "Orifamo baie" hie
Buro flying fox it is he-says
'Buro says it's a flying fox'
'S/he says it's a flying fox'
b. hu - gasi' orifamo baie
Buro say GERUND flying f o x it-is

'According to Buro, it's a flying fox'
In sentence (30a), constructed on the pattern of (27a), it does not matter who
makes the identification, since its correctness is accepted, or perhaps, irrele
vant, irrespective of the source. Consequently, no full NP (common noun,
proper noun, or contrastive pronoun) is required as the explicit source of the
quote. In (30b), on the other hand, the identification is a matter of opinion,
and it is important to specify whose opinion is being cited. The sentence is
therefore peculiar unless some NP is explicitly given as the source of the verb
of saying.
To return to the inconsequential of (27b): does it ever occur with verbs of
thinking or saying? Conversely, does it ever interchange with the 'subjective
gerund' in gasi' with verbs of perception? The answers to these questions de
pend on whether we use elicited sentences as data, or actual texts.
In elicited data, I found that whenever Verb + gasi' was possible, so too
was Verb + mana, and conversely. In real texts, this was not so. The pattern
of (27b), the inconsequential, was almost invariable with verbs of percep
tion; the pattern of (28), the subjective gerund, was equally regular with
verbs of opinion; while with verbs of saying, there was a semantically condi
tioned alternation between the direct quote pattern of (27a) and the inconse-
64 JOHN HAIMAN
quential of (27b), depending on whether the quote was presented as a subjec

tive assessment by a given speaker or a neutral object.
From this, it seems valid to infer the following:
i. opinions are treated as subjective; perceptions are not.
ii. Although both quotes and perceptions may be treated as non-
subjective (no -gasi' form) there is a difference between them.
A quote is treated as the object of a verb of saying (and, as an
object, precedes this verb), while a perception is not treated as
the object of a verb of perception (and, thus, follows this
verb).
Both (i) and (ii) reflect a plausible common-sense view. As regards (i):
whatever Plato and Berkeley may have said, there is a difference between
what we see and what we think for most of us. As regards (ii): a direct quote
is obviously the physical product of an act of speech, but a thing perceived
is not the created product of an act of perception.
The inconsequential form is then used where the complement is not sub-
jetive, and, as a non-object clause, follows the verb of perception. These
considerations will account for the fact that Hua does not use either the
direct quote or the subjective gerund constructions for describing the rela
tionship between the act of perception and the act perceived. But we now
need to account for the fact that Hua does not use the most common clause-
combining construction of all, the medial verb construction. Why not
(31) a. Verb of perception + medial 'complement'
b. 'complement' + medial Verb of perception
given that this construction has so many other uses? The answer, I believe,
lies in the linearity of the medial construction. The pattern (31b), in fact, is
quite common, and occurs wherever the event described is viewed as
preceding the act of perceiving it:
(32) αi'αnαdα kgoe
you-came-up-and-I (DIFF SUBJ COORD MED) I-saw-you
'You came up and I saw you.'
In (32), I saw you after and because you came up. Clause two is both post
and propter clause one, and the medial is entirely appropriate. But if we were
to reverse these two clauses, in accordance with the pattern of (31a), the
result would be unacceptable:
(33) * kgogaka ai'ane

I-saw-you-and-you (DIFF-SUBJ COORD MED) you-came-up
'I saw you and you came up.'
This sentence would only be appropriate if your coming up followed and
were a consequence of my seeing you. Given the linearity of the medial con
struction, it simply cannot be used to express the idea that I saw you come
up. And, as we have seen, none of the other possible contenders (the relative
complement clause on head noun  'thing', the direct quote, and the sub
jective gerund) is appropriate either.
This may, then, be the best explanation for the use of the inconsequential
construction for expression of the relationship between acts of perception
and the acts perceived. The construction is used because none of the other
available constructions is appropriate. They make unacceptable claims
about the semantic relationship between the two events in question: either
by claiming that the event perceived is already common knowledge (the
relative clause construction), or by claiming that the act perceived is a
physical product of the act of perception (the direct quote construction), or
by claiming that the event perceived is purely subjective (the -gasi' gerund
construction), or by claiming that the act of perception precedes and/or is
the cause of the event perceived (the medial construction).
The only other construction which is as noncommittal as the inconsequen
tial in this regard is — once again — the conditional, which occurs in my
texts as frequently as the inconsequential in this last function:
(34) rmuna ge - mamo rema' mna faipai' runa
going down see (3SG COND) coloured bird white bird
havireta ereta
going-up coming
hita na zumo kebgota' rita havita...
doing that work together doing going-up
'She went down and saw birds of all sorts, coloured birds and
white birds going up and down (the garden rows) doing all
that (garden) work together, going up and...'
While these remarks do not begin to explore the intricacies of all the rela
tionships among even the clause-combining devices I have mentioned here,
they do pretty well exhaust my present understanding of the ecology of the
inconsequential. It is time now to consider the implications of some of these
relationships.
66 JOHN HAIMAN
Conclusions
This paper has dealt mainly with two clause-combining categories, the in
consequential and the conditional, and their relative distribution. Concern
ing the first, there is relatively little more that can be said. There are, of
course, analogs of the words 'but' or 'in vain' in presumably every language.
But the semantic territory of the inconsequential in Hua differs markedly
from that of corresponding expressions in English, and even from that of
the corresponding construction in typologically closely related languages like
Tauya, discussed in Lorna MacDonald's contribution to this volume. While
the logic of the generalization of its functions makes a certain amount of
sense in Hua, this is very much a language-internal matter.
The conditional, on the other hand, is a highly grammaticalized and easily
identifiable category in almost every language, and traditional grammarians
have had little hesitation in assigning to it a universal meaning or
significance. One of the features of the meaning of the protasis, most people
seem to agree, is this: there is some causal connection between the protasis
(the 'if' clause) and the apodosis (the 'then' clause) (thus Ramsay 1931 and
Strawson 1953; among many).
Now, the point that seems to emerge from the data we have been discuss
ing here is that in Hua, at least, this contention is absolutely false. If we wish
to assert a causal connection between two clauses, then we should employ
the medial construction. If we wish explicitly to DENY that such a connec
tion exists, we should use the inconsequential or the conditional. Is this a
peculiar aberration of Hua, or does it reflect a general pattern? I would
argue that there is a pattern here, and that conditionals do not primarily exist
to mark purely causal relationships.
Two cross-linguistic pieces of evidence in favour of this contention are:
i. the identity or the near-identity of conditionals and concessive
conditionals in most languages (for example English 'if' and 'even
if). If an essential or defining property of conditionals was the
causal connection which they asserted between protasis and
apodosis, then we should expect the morphology and syntax of
concessive conditionals to be totally different in a reasonable
number of languages. It does not seem to be.
ii. the ability of paratactic clauses to function as causal, but not con
cessive, conditional protasis clauses. That is, the causal connec-
tion between two clauses is largely a function of the order in which

they appear, and has nothing to do with the presence of a diacritic
like 'if' or its congeners in other languages.
To this array of cross-linguistic evidence (some of which is presented in
Haiman 1978), the partial interchangeability of the conditional and inconse
quential· constructions in Hua is a small but compatible addition.
Finally, what is the Gesamtbedeutung of the inconsequential in Hua, and
where does it belong in an inventory of clause types? It seems to me that the
peculiarly close relationship between conditionals and inconsequentials
allows us a satisfactory answer to these questions.
First, like the conditional protasis, the inconsequential establishes some
proposition as a given in the discourse. Unlike coordinate medials, and
unlike gerund clauses in -gasi' , the mood, tense, and polarity of inconse
quentials is spelled out and specified in the inconsequential itself, and does
not depend on that of any following clause (cf. Haiman, 1978, where I
argued that this is what defines conditional protasis clauses in general). In
this respect, both conditionals and inconsequentials are aligned with in
dependent (or final) clauses, and with two dependent (or sentence-internal)
clause types: antithetical adversatives in maborava, and subordinate medial
clauses.
Second, like the conditional protasis, the inconsequential is at most non
committal with respect to any causal relationship between the proposition it
expresses and that expressed in the following clause. This property sharply
distinguishes conditionals, inconsequentials, and adversative antitheticals on
the one hand from subordinate (and, for that matter, also coordinate)
medial clauses on the other.
Third, there are two senses of the notion 'but': balance or antithesis on
the one hand, and inconsequentiality on the other. Like the conditional pro
tasis, the inconsequential is restricted to the expression of the latter sense
alone. This limitation distinguishes it from the antithetical adversative,
which encompasses both senses of 'but'.
Fourth, both the inconsequential and the conditional can be somewhat
weaker in force than English 'but'. Unlike the antithetical adversative, which
explicitly signals antithesis or inconsequentiality, both the conditional and
the inconsequential have one further use: they can signal the relationship
between a verb of perception and its complement.
Fifth — and this is a sense in which the inconsequential is less of a 'true'
inconsequential than the conditional — only the former can be used to signal
68 JOHN HAIMAN
a change of speaker in a dialog. This use of the inconsequential aligns it with

the coordinate medial here.
Finally, unlike the conditional, the inconsequential is capable of con
stituting a full non-elliptical utterance all by itself. This property aligns it
with final, or independent, clauses.
It is, however, very misleading to leave the impression that conditionals
and inconsequentials are interchangeable in all cases but the last two listed
above. Since our discussion has focussed on the functions and distributions
of the inconsequential, it may have been easy to overlook two functions that
this clause may never perform: it may never function as a causal (or 'if') con
ditional; and it may never function as a concessive (or 'even if') conditional!
(These limitations, it should be noted, may be entirely the consequence of
the fact noted at the beginning of this essay, namely, the fact that the in
consequential cannot occur with the subjunctive auxiliary -su-, and may
therefore never have hypothetical force. Following Jespersen and doubtless
many other scholars, we tend to assume that hypotheticality is an essential
property of conditionals (cf. Jespersen, 1940), a view which I have tried to
show is mistaken in Haiman, 1978). That is, the inconsequential seems to
share all the functions of the conditional except those that we, with our
Eurocentric perspective, may regard as the most basic.
Starting, as we must, from such a perspective, our goal must be to show
that the distribution and the functions of the inconsequential (or, indeed of
any grammatical category in any language) is characterized by a coherent
logic. I think it is possible to achieve this goal and say that the inconsequen
tial is characterized by givenness, independence, and causal non-
commitment. But we will be forced to recognize that these properties (to say
nothing of their clustering) are language-specific, and violate not only
universal notions of such categories as 'conditional', but our Eurocentric
'intuitions' of what constitutes principal or subordinate clauses. For exam
ple, the question whether inconsequentials are main or subordinate clauses
clearly makes no sense in terms of any traditional structural or semantic
definition of these notions. (Nor does it seem to be useful to enrich this
dichotomy, as Olson (1981), or Foley and Van Valin (1984) have done, and
recognize a third relationship of 'cosubordination' intermediate between
coordination and subordination.)
That the logic of grammatical categories in an alien language is
(somewhat) accessible to us is beyond question: this continues to be the best
evidence for language universals of some sort. But it is equally clear that the
categories we expect to find in all languages are not necessarily universals at

all, and when it comes to clause combining, we are really on virgin soil.
When we step out on this terra incognita, I suspect that our best research
strategy is to let morphology be our guide and adhere to the humble rule of
thumb that identical form is motivated by some significant similarity of
meaning. This paper is intended as another advertisement for this 'method'
of 'universal morphology' and the results it can bring.
REFERENCES
Foley, W., and R. Van Valin. 1984. Functional Syntax and Universal Gram
mar. Cambridge: Cambridge University Press.
Haiman, J. 1978 "Conditionals are topics". Language 54: 565-89.
Jespersen, O. 1940.  Modern English Grammar on Historical Principles.
V: Syntax. London: George Allen & Unwin.
Olson, M. 1981. "Barai clause junctures". Unpublished doctoral disserta
tion, Australian National University.
Ramsey, F. 1931. "General propositions and causality". Foundations of
Mathematics, 237-55. London: Kegan Paul.
Strawson, P. 1952. An Introduction to Logical Theory. London: Methuen.
Concessive clauses in English and Romance
Martin Harris
University of Essex
It is now generally recognized, as Ekkehard König (1986) has noted,

that many of the categories traditionally used for the classification
and characterization of adverbial clauses are not discrete ones. That this is
particularly true in the area of conditional and concessive clauses has been
clearly demonstrated by a number of scholars, exploring a field drawn to the
attention of contemporary linguistics by John Haiman as early as 1974, and
developed by him in a series of papers subsequently; I shall draw heavily in
what follows on his work and that of König. I have myself noted (Harris,
1986) that this semantic non-discreteness extends beyond the limits of con-
cessives and conditionals: specifically, we find "a semantic spectrum ranging
from causal clauses in which the causal link between subordinate and main
clause is asserted, via conditional sentences, in which the causal link is
hypothesized and proposed as a basis on which to continue the verbal in
teraction, to concessive clauses, in which the causal link between protasis
and apodosis is denied'' (Harris, 1986; for an earlier analysis on similar
lines, see Sandfeld, 1936:371 and Lehmann, 1973). Within this spectrum,
of course, there are clear foci, but the edges between the categories are blur
red, facilitating as we shall see shifts of meaning on the part of the variety
of connectives used with these values. The purpose of this is to look at
various definitions of 'concessive' and 'concessive conditional' clauses, us
ing primarily English data discussed in earlier contributions to the debate,
and then proceed to examine (changing) markers of these areas of meaning
in Romance, to see what light (if any) a diachronic study may throw on our
general understanding of this conceptually most complex area.
If we focus our attention now on the more limited semantic field outlin
ed at the start of this paper, namely that extending from conditionals via
concessive conditionals to concessives, we find that it is relatively easy to
72 MARTIN HARRIS
distinguish in logical terms the two end-points: in the words of König

(1985:365), "the former entail neither their antecendents nor their conse
quents, whereas the latter entail both of their component clauses". What is
at least as important in linguistic terms is that a concessive antecedent "con
tains a fact or notion in spite of which the main proposition stands" (Burn-
ham, 1911:4). Haiman (1974:357) makes much the same point when he says
that "in concessives, contrary to our justifiable expectation, the choice of
disjunct is totally irrelevant, at least to the result expressed in the main
clause," while Bruce Mitchell puts it thus (1985:706): "in a concessive
sentence, the truth of the main clause is asserted, despite the proposition
contained in the subordinate clause". This same fact is noted by König
(1985:366), when he points out that in addition to the entailment of both
antecendent and consequent alluded to above, a concessive connective im
plies that the antecedent and the consequent are by normal standards incom
patible: "This implication can roughly be described as follows: Normally (if
p, then not-q)". Notwithstanding the normal position, however, in a con
cessive sentence such as 'although it's raining, I'm still going swimming' the
assertion is that despite the rain, (which would normally preclude such an
activity) the swimming will take place. (Remember the spectrum ranging
from causals to concessives, and compare this sentence with 'Because the sun
is shining, I'm going swimming'). It is by virtue of the fact that the conse
quent is asserted that one frequently finds references to the 'actual' or 'fac
tual' nature of concessive sentences.
Concessive conditionals, as has often been pointed out (e.g. König,
1985:365-6), share features of both concessive and conditional concessives,
and this semantic overlap is reflected in their morphosyntax, as we shall see.
Here, we shall hazard an extremely simple analysis of the semantics of these
concessives, in the hope that it will prove adequate for the task in hand.
Essentially, we argue that a concessive conditional sets up a hypothesis, the
realization of which, if it occurs, would bring into being a situation in which
a concessive interpretation of the type described earlier then comes into play.
Compare again the simple concessive 'although it is raining'...(it is, and
despite this fact, the truth of the main clause is asserted) with 'even if it
rains...' (it may or may not rain, but if it does rain, then despite this fact,
the truth of the main clause is asserted). Note that 'even if'-concessives share
with 'if'-clauses all the various degrees of probability associated with simple
conditions (cf. Harris, 1986), while at the same time such concessive condi
tional clauses, by discounting (as a possibility, not as a fact as in the case
CONCESSIVE CLAUSES IN ENGLISH AND ROMANCE 73
of simple concessives) a specified circumstance which might be expected to

preclude the realisation of the consequent, entail the assertion of the proposi
tions expressed by the main clause. Note further that the three categories of
conditional, concessive conditional and concessive are in principle indepen
dent of the dimensions of time (although certain combinations may occur
more frequently in practice than others); compare:
Table 1
(C = Consequent)
(i) a. if he comes tomorrow, 

b. even if he comes tomorrow, 
c. although he's coming tomorrow, 
with
(ii) a. if he's here now, 

b. even if he's here now, 
c. although he's here now, 
and
(iii) a. if he had been here yesterday, 

b. even if he had been here yesterday 
c. although he was here yesterday, C1
In each case (with appropriate allowances for time) the (a) sentence entails
neither antecedent nor consequent2 but asserts the truth of  given the fulfil
ment of the hypothesis (and says nothing about  given its non-fulfilment);
the (b) sentence entails C3 regardless of the hypothetical realisation of a cir
cumstance normally seen as incompatible with  (the less expectable implies
the more expectable: cf. Bennett, 1982:412-3); and the (c) sentence entails
both antecendent and consequent, the former being as before normally seen
as incompatible with  but being in this case actually instantiated and
therefore no longer hypothetical. The way in which (b) sentences combine,
mutatis mutandis, certain characteristics of both (a) and (c) sentences is thus
clear.
If this analysis is accepted, it follows that concessive conditionals, in-
74 MARTIN HARRIS
sofar as they are concessives, will share the semantic features of the latter.
In particular, they share the property whereby the antecendent marks an ex
treme value (whether potential or actual, depending on the clause type)
within a set of possibilities, a value generally taken to be incompatible with
the consequent. There may be only two values postulated — in which case,
the parallel with yes-no questions is clear — or a whole set or scale of values,
with the antecedent lying clearly at one end of the relevant spectrum. Com
pare, for example, 'even though/even if he is right...' (concessive and con
cessive conditional respectively), where the implied choice is polar (' is he
right or is he wrong?'), with 'even though he spent/even if he had spent
every penny he had...', where the inevitability of the consequent if less
money had been spent is self-evident, and where the choice is scalar
('however much money he spent...'). 4 This concept of pragmatic scales is
clearly elaborated in Fauconnier (1975).
We can also now demonstrate convincingly that, insofar as concessive
conditionals are conditionals, then they share the semantic features of the
latter. In particular, the overlap between conditionals and interrogatives
discussed in Lehmann (1973) and Haiman (1978) can be seen to extend to
concessive conditionals where as we have seen 'even if he is right...' implies
a question 'is he right or is he not?' and further implies that, since this is
a concessive conditional, the answer does not affect the truth of the main
clause (as it would of course in a simple conditional). The overlap between
'if' and 'whether' in conditionals is well known: it is worth recalling further
in this context that 'whether he is right or not' is an alternative formulation
for a concessive conditional clause in English and that 'whether' derives
from OE hwaeðere 'which of the two' (cognate with Lat. uter) (cf. König,
1985:370n4). This overlap between conditions and questions does not of
course extend to concessives as such, where an earlier underlying question
has already been resolved ('even though he is right...'): hence once again the
factual nature of concessives.
In what follows, we shall seek to provide a unitary analysis of the
markers of concessive and concessive conditional clauses, in the hope of ad
ding to our understanding both of how the various clause-types we have been
discussing inter-relate and how and why the well-attested 'slippage' from one
category to another can so readily occur. We should recall that because the
semantic boundaries are not at all clear cut, particular conjunctions and in
deed entire morphosyntactic structures can pass readily — though as we shall
see overwhelmingly in one particular direction — from marking one value
to another. Instances of such changes appear frequently in what follows.

We are now in a position to turn to an examination of the range of con
junctions used to mark concessive clauses, limiting ourselves almost entirely
(for reasons of space) to structures of the 'concessive conjunction + finite
verb, main verb' type. We shall use Romance data for this part of the paper,
but will refer back to English material wherever it seems useful to do so.
It is possible to detect the following patterns underlying the various con
cessive conjunctions (or locutions which serve as or in due course become
conjunctions) in Romance:
Table 25
(i) the indefinite and therefore all-embracing pattern

(whatever you may do, nevertheless...)
(ii) the explicitly all-embracing pattern
(all that you may do, nevertheless...)
(iii) the volitive pattern
(let it be as you wish/let it be so, nevertheless...)
(iv) the explicit marking of an extreme scalar position
(Most X that it may be, nevertheless...)
((whatever) good/ill may come of it, nevertheless...)
(v) a temporal pattern (often reinforced)
(be it still/yet/already/now so, nevertheless...)
(while/when..., nevertheless...)
(vi) a conditional pattern (often reinforced)
((even) if/though he did it, nevertheless...)
From this table, it emerges that concessive markers are derived, on a

very widespread if not a universal basis, from markers originally having
other values: indeed König (1985:363, 380) goes so far as to speak of con-
cessivity itself as a derived notion. The explicit marking of this notion by
means of a specific subordinator is clearly an 'optional extra' within a
language, as the history of Romance demonstrates particularly clearly.
When such a specific subordinator does develop, a primary source will
necessarily lie in the semantics of concession, wherein as we have seen the
consequent of a concessive sentence is entailed whatever value is attributed
in respect of some specified parameter. Any circumstance, every cir
cumstance, whatever circumstance one may wish, all of these are clearly
76 MARTIN HARRIS
semantically appropriate sources for concessive conjunctions; so too is a

construct marking an extreme value on any scale, whether general ('however
most...') or more specific ('however good/ill that...'), given only the
necessary additional semantic component of prima facie incompatibility bet
ween antecedent and consequent. Certain temporal conjunctions, given once
again this element of incompatibility, also lend themselves readily to a con
cessive interpretation, as we shall see. Finally, conditional conjunctions
(generally but not invariably reinforced) readily develop concessive condi
tional value, from which (in appropriate circumstances) a more narrowly
concessive value may develop (as with English 'although'). All of these pro
cesses will be observed in Romance in the pages which follow.
Turning now to Classical Latin 6 , we find that there are essentially four
patterns favoured for the marking of concessives. Firstly, there is pattern (i),
the indefinite adverb of degree quamquam, literally 'as as', that is 'in what
ever way'. (Note the formal parallelism with quisquís 'whoever', quidquid
'whatever' etc.) Given that the choice of mood in Latin was still to a con
siderable extent semantically motivated (a point developed towards the end
of this paper), and given also that, as we have seen, a true concessive clause
relates something which, however unexpected or unwelcome, is nevertheless
a fact, the unmarked mood originally used in such clauses was (appropriate
ly) the indicative. Secondly, we find pattern (iii), including the explicitly
volitive items quamvis 'as you wish' and later Latin quamlibet 'as it pleases
you', together with licet, originally an impersonal verb with the value 'it is
alllowable' hence 'let it be so'. Clearly, if one's interlocutor can take as fac
tual any value he chooses to assign to the antecedent, but the speaker can
nevertheless assert the truth of the main clause, then we are right at the heart
of concessive territory and the frequently recurring morphological overlap
between verbs of choice or volition and concessive markers, noted in
Lehmann (1973) and Haiman (1974) can be said to be accounted for.7. In
terestingly, these conjunctions pattern with the subjunctive, presumably
reflecting a jussive origin (quamvis bonus sit 'let him be as good as you
please', i.e. 'although he is good'; Woodcock, 1959:201). Not for the last time
do we note a situation in which the use of one conjunction rather than
another overrides the underlying modal motivation of the clausal category
as a whole, a point developed more fully below.
The third method of marking concession in Latin is pattern (v), the use
of the temporal conjunction cum ('when' 'while'). 8 If one considers an
English sentence 'while he is certainly intelligent, he's also very lazy', the
semantic shift from 'while' to 'although' seems small indeed. (Cf. the
overlap between temporal and conditional conjunctions discussed in Harris,
1986.) We shall have more to say below about the relationship between tem
porals and concessives.
Finally, as instances of pattern (vi), we should note the range of com
pounds of si "if". Of course forms such as etsi9, etiamsi, tametsi 'even if'
were used primarily to mark concessive conditionals, but all of them, in par
ticular etsi (the normal concessive marker in Caesar, for example), could be
and were used to mark straight concessives. The use of reinforcers with con
ditional markers is often associated with a semantic shift via concessive con
ditionals into the 'factual' concessive domain (König, 1985:374), but is by
no means a pre-requisite, si alone for example also having this value at times.
When Ernout and Thomas claim (1953:351) in respect of this group of con
junctions that "la syntaxe est celle des conditionelles", this is acceptable
provided we recall the distinction between concessives (which are of course
not hypothetical) and concessive conditionals which do indeed behave syn
tactically like all other si-conditional clauses. Thus a number of instances
noted in the manuals of si, etsi and the like being used in 'concessive clauses'
clearly fail. Woodcock, for instance, cites (1959:200) 'si rex ipse esset (subj),
tamen ei non parerem (subj), ('(even) if he were the king himself, I still
should not obey him'), which is manifestly a concessive conditional and not
a concessive, as can be seen clearly if the distinctions outlined in Table 1 are
borne in mind. Contrast this with 'si mihi bona re publica fruì non licuerit
(indic), at carebo (indic) mala' (Gildersleeve and Lodge, 1895:377), where
the antecedent is taken as definite: 'if (= even though) I shall not be allowed
to enjoy good government, I shall at least be rid of bad'. 10 In principle, a
truly concessive si or etsi can have any tense of the indicative mood which
is semantically appropriate: an exactly comparable point applies, of course,
to the cognate forms si/se used non-conditionally in contemporary
Romance.
Before moving on (chronologically speaking, at least) to survey the ma
jor markers of concession in Romance, let us pause to re-emphasise the
distinction between concessive conditionals and concessives, which we have
just alluded to once again. In contemporary English (or at least in the
idiolect of this speaker of educated British English), unlike earlier periods
of the language (König, 1986:241; for OE, see Mitchell, 1985:705), (even)
though/although cannot be used with a concessive conditional value, but on
ly with a (factual) concessive value: these conjunctions therefore have only
78 MARTIN HARRIS
a narrowly limited semantic range. In saying 'although/even though he will

come tomorrow... ' one is taking his coming (albeit in the future) to be a fact,
an event moreover which, as we have seen, has or will have no effect on the
truth of the main clause. The reverse situation, however, when primarily
conditional conjunctions (e.g. '(even) if') are to be interpreted as true con-
cessives (as opposed to concessive conditionals), is found in English as in
Latin and indeed everywhere in Romance, and is part of a much broader
phenomenon whereby si and its cognates are used very widely indeed outside
their core semantic territory. 11 (I have developed this point fully in Harris,
1986 and shall not cover the same ground again here.) The possibility of
substituting '(even) though' for some other conjunction in modern English
is therefore a clear diagnostic as to whether one is in fact dealing with a
(non-conditional) concessive clause.12
Reverting now to Romance, we find that none of the Classical Latin
concessive conjunctions which we have been discussing survived to serve as
the etymon for any present-day form apart from si, which was of course
primarily a conditional marker. With one possible exception to be discussed
shortly, we may accept the conclusion of Herman (1963:233) that "pendant
une période assez longue, la langue parlée préromane n'a pas eu de conjunc
tions concessives systématiquement employées et largement répandues." 13
(Equally, there was no exclusively concessive conjunction in Old English.)
This absence of explicitly concessive conjunctions in Proto-Romance is part
of a very general phenomenon, whereby a very limited number of subordina-
tors, in particular derivatives of quid/quod, carried a very wide range of
meanings, with language users relying on parataxis, co-ordination and of
course context to disambiguate when necessary. Robert de Dardel (1983:86)
puts it thus: "le roman commun a débuté avec un nombre très réduit de
subordonnants morphologiquement distincts: et, ak13, k2, quando, quan-
tu, quomodo, si1-2, ubi, éventuellement makarie." It is this last item, seen
by Dardel as a 'possible' Proto-Romance conjunction, which is conceivably
an exception to Herman's general statement, noted above.
In Rum. (mǎcar), dialects of North and South Italy (e.g. magari, macari,
13th century Sicilian macara, Logudorese mancari), OSp. (maguar,
maguer (a)) and Port. (macar, with maguer probably a Hispanism via Galli-
cian) together with certain Rhetian and Occitan dialects, one finds
derivatives of a form *makarie, with various interrelated meanings, in par
ticular 'would that' 'in spite of' and 'although'. (All of these meanings can
still be found in various Italian dialects, with the first widely used in the stan-
dard language: cf. Corominas, 1954 (III): 190; and Rohlfs, 1969 (III): paras
783, 963.) The origin of this form, it is now generally agreed, is (Classical)
Greek makarie, a form of the adjective meaning 'blessed', used with the
value of 'would that' (a sense retained by makari na in modern Greek), ser
ving as "una especie de cortesía demostrada al interlocutor afectando desear
que suceda lo que él nos objeta" (Corominas: loc. cit.). I take this explanation
of 'courtesy' to complement rather than contradict the account of the use
of volitives in this sense outlined above; in a very early example in the 10th
or 11th century Glosas de Silos, one finds macare used as the gloss for quam-
vis, which of course had precisely this function, at least in origin.
The semantic shift from a volitive to a concessive will by now hardly
occasion surprise; what is of interest is the extent to which this form was
really part of (late) common Romance14 or was borrowed separately in a
number of Romance-speaking areas, and, even more striking, the fact that
what was ultimately such a 'grammatical' item as a concessive conjunction
was borrowed from another language, Greek, a development wholly atypical
of the normal pattern of borrowings from Greek into Latin (and indeed
more generally). Of course this latter process is less surprising if one recalls
that it was as an interjection (i.e. with the sense of 'would that') that it was
first borrowed and John Green (p.c.) draws to my attention parallels with
paratactic interpolations of deo volente in Christian Latin. Indeed, exactly
the same process can be seen recurring at the present time, in that Sp. ojalá,
borrowed (again quite atypically) from Arabic, also with the sense 'would
that' (lit. 'and may God will it...' cf. deo volente, above) is widely used as
a concessive conjunction in parts of Spanish America: see Bourciez (1967:
para 393(e)), who indicates a stress-shift to ojála and Corominas (loc. cit.),
who cites Argentinian examples such as "ojála cierto, no lo creería"
('even if it were true, I wouldn't believe it'). The parallelism between the
development of makarie and ojála, two instances of a quite unusual type of
borrowing and a highly specific semantic change, is quite fascinating.
Insofar as there was a Proto-Romance concessive conjunction, then, it
was the Greek borrowing makarie. Apart from Rumanian, however, where
mǎcarcǎis still a widely used concessive marker, none of the modern forms
of the standard languages make use of derivatives of this item. It is to other
possibilities within the range of potential concessive markers that the
Romance languages turn, in a series of developments which are in essence
independent of each other and language-specific, though we may of course
suppose some degree of mutual influence between the sister languages from
80 MARTIN HARRIS
time to time.
We mentioned above that in early Romance a very limited range of sub-
ordinators carried a very wide range of meanings. One should add that, not
infrequently, two propositions were simply juxtaposed, with the context and
the mood of the verb having to suffice to render the meaning clear
(Lehmann, 1973). Such paratactic structures occurred with concessive as
with other values, as indeed they had in Latin (Woodcock, 1959:87;
Gildersleeve and Lodge, 1895: paras 264, 608): thus for instance Klare
(1958:32) cites the following 12th century French example:15
Li reis Hugue li Forz nen at nul bacheler,

de tote sa maisnie, tant seit forz et membrez,
Ait vestut dous halbers et dous helmes fermez,
Si seit sor un destrier corani et sojornet...
Si ferrai sor les helmes,..
where the verbs in bold are in the subjunctive and the meaning of ait
vestut, literally 'may he have (subj) put on', is clearly 'even if he put on'.
Such structures survived into Middle French (Martin and Wilmet, 1980: para
384, 2), particularly as explicit disjuncts: cf. `viegne ou ne viegne 'may he
come or not come' (cf. 'Like it or not').
In their earliest stages, the Romance languages — for which we shall
take French as representative16 — maintained such structures, often reinfor
cing the 'concessive' subjunctive with an adverb, the whole sentence initially
remaining paratactic. In his exhaustive survey, Klare (1958) notes the use of
tout 'all', of (com) bien '(how) well' and of various temporal adverbs (Ja, en
core, or(es)). Thus, to give just one example, 'encore soit (subj) malades' in
OFr. (literally 'still be he ill') was an early way of expressing the concessive
relationship 'although he is ill' (von Wartburg, 1922-68: IV, 478). Since in
different ways, these adverbs are all related to conjunctions which gradually
emerge in French or elsewhere in Romance, it is as well to glance briefly at
the semantics of each, which correspond to possibilities (ii), (iv) and (v) in
Table 2.
The first of these categories, the use of words meaning 'all' needs little
explanation. On any scale, a situation which is depicted as being entirely at
one end is clearly ready made to be used concessively, provided that the end
specified is that least readily compatible with the main clause, which is never
theless represented as true. We find, for instance 'tut seit-il mort (lit. 'en-
tirely be he dead') in the sense of 'though he is dead' among many other ex
amples cited by Klare, (1958:40-44). English 'albeit' clearly has a similar
origin, and the same use is to be found in structures such as 'for all that he
is intelligent, he is (nevertheless) idle' and the equivalent nominal structure
'for all his intelligence'.17
Category (iv) is more problematic. 'Well' is widely used as a reinforcer
with 'concessive' subjunctives in Romance, as indeed it may be in English
('Well may you laugh but...' 'It's all very well for you to laugh (but)...';
Dickens, for instance, writes This is well enough...but nevertheless...'.) 18
The use of bien in French to reinforce concessives is found as early as 1164
(Chrètien de Troyes), and is often explained by French linguists as a quan
tifier, i.e. 'however much...'. One problem with this, however, is that the
designation of the opposite end of the same semantic spectrum is also found,
either simply (cf. Cat. mal que 'badly that' = 'although') or in compounds
(Fr. mal-gré (que) 'ill-fate (that)') and, so far as I am aware, the 'quantifier'
explanation is not available in this case. The same argument weakens the
temptation to see 'well' as an instance of a volitive, since once again it is dif
ficult to see how 'badly' could be subsumed within the same framework. It
seems preferable to see bien and mal and their cognates as being particular
instances of the marking of extreme (and unfavourable) positions within a
scale of possible situations, the consequence being a fortiori inevitable in less
extreme circumstances, while compounds like malgrè seem to add weight to
this explanation ('whatever ill (fate) may come of it...'; cf. 'good it may be,
nevertheless...'). It would be incautious to deny, however, that the use of
bien as a quantifier may have favoured the increased use of this particular
adverb: the use of com bien ('how well', later 'how(ever) much') in Old and
Middle French with the value 'although' (cf. note 18) does provide another
motive for bien (que) if one is required.
Finally as regards pattern (v), the explanation seems to be thus: in ap
propriate contexts, a temporal element may be used in conjunction with a
main clause the truth of which is unexpected at that time (for whatever
reason). Now the essence of a concessive sentence is that  is true despite
the content of the antecendent, which is (by definition) generally incompati
ble with that particular consequent. If one thinks back in this light to the
earlier mention of Latin concessive cum, intelligence is not usually consistent
with laziness (at least not given a particular world-view) and therefore 'while
he is certainly intelligent, he is also very lazy' can readily take on a concessive
value (cf. König, 1985:378). One of the definitions of still in the OED, for
82 MARTIN HARRIS
instance, (sv still adv 6a), refers to the continuance of a previous action or
condition 'after or at the same time with [sic] some event or condition im
plied to be adverse', in other words exactly the same pattern we have seen
in the other semantic areas examined. To revert finally to our earlier exam
ple, the OFr. sentence encore soit (subj) malades meant originally 'be he
(unexpectedly) ill until now, (nevertheless)...', that is, 'although he is ill'.19
It is therefore from a starting point with only derivatives of makarie
(and that not in most of France) and paratactic structures variously reinfor
ced in the way just discussed that the system of concessive conjunctions in
modern Romance developed. The first obvious thing to note is the gradual
development of explicit conjunctions from (some of) the adverbs we have
just been examining, by the absolutely straightforward method of adding the
unmarked subordinator que/che. Only a very small number of inherited sim
ple adverbs survive in Romance (derivatives of two of which, si and quando,
we shall shortly mention briefly): elsewhere, the structure is adverb/preposi
tion plus (normally) a subordinator. Thus, to give just one example, one
finds in Fr./Occ./Cat./Sp. and Port; bien que, ben que, benchè, bé que,
bien que and bem que, the composition in each case being transparent. It
would serve no purpose to seek to provide an exhaustive survey of all the
concessive conjunctions in Romance; rather, we shall show how all the pat
terns described earlier are exemplified, and then conclude with some general
observations.
The indefinite pattern (possibility (i) in Table 2) is probably best
represented in Romance by Fr. quoique, literally 'what that...'. Found in
early French in the form que que,20 this particular conjunction gains ground
steadily during the Middle French period (Martin and Wilmet, 1980:233), to
the point of being one of the main concessive conjunctions of the contem
porary language.
There are a number of representatives of pattern (ii), the explicitly all-
embracing pattern. In Rumanian, for example, one of the principal con
cessive conjunctions is cu toate cǎ, 'for all that', while tutto che is not
unknown in Italian (Rohlfs, 1969 (II): para 697; (III): para 784). (The
semantic links with structures in French such as 'tout grand qu'il soit' ('all
big that he is', i.e. 'however big...') are transparent.) One standard con
cessive conjunction in literary Provençal incorporating the 'all' element was
sitot, but this is not used in modern standard Occitan; tot i que is however
found in Cat.: 'tot i que hagues guanyat el plet' 'although she had won the
lawsuit'. Combining features of patterns (i) and (ii) (and therefore perhaps
supporting König's decision (1985:372) not to separate these two categories)

are conjunctions such as Port, conquanto (lit. 'with how(ever) much'), It.
per quanto ( 'per quanto sia difficile, è tuttavia risolvibile' ('however difficult
it may be, it is nevertheless solvable')) and OFr. quanque, where the quan-
titive element is explicit, while OFr. com bien que 'how well/much that' has
already been referred to as an instance where two distinct semantic strands
converge.
The third pattern noted in Table 2 is that involving volitives. Here, we
need to distinguish two sub-types. Firstly, there are those whose etyma incor
porate an explicitly volitive element. These include derivatives of makarie,
discussed earlier, in particular in Ital. dialects and in standard Rum. (mǎcar
cǎ/sǎ/de), while the OSp. form maguer(a) que (with both concessive and
concessive conditional values, and with modal usage to match, cf. Algeo,
1972 (3):533; see also Rivarola, 1976), was in regular use until the 15th cen
tury. A comparable semantic shift is found in Cat., in which one conjunc
tion with concessive value, baldament, developed from an adverb originally
meaning 'gladly' (still attested in béarnis). From this emerged the meaning
'would that' (Roussillon), whence 'in spite of' and, in (literary) Catalan,
'although' ('ve, baldamen t sigui (subj) tard' 'he's coming although it's
late'). (See Coromines, 1980-84: 584.) We should note also Port, embora
(<em boa hora 'in good hour'), where the volitive and temporal elements
which may carry concessive value intermingle ('Be it timely/propitious
that...', later simply 'be it so that...' and hence 'although'). 21 This particular
development provides the link with a larger group in which the volitive ele
ment is implicit, being found in the use of the 'jussive' subjunctive. The
simplest instance is perhaps It. avvegnaché (now very dated), the sense being
'may it (indeed) happen that... (nevertheless C)' (Tekavcic, 1972:656).
Slightly more complex, because of the presence also of a temporal element,
are structures like OFr. ja soit que (jaçoit que) OProv. jassiaisso, OCat. jat-
sia che 'be it (already) (thus) that' (ja < iam 'already'). (This particular
structure persisted in French until the 17th century.) It is interesting to
observe in passing that the nearest English equivalent, albeit, noted above,
is reinforced by all rather than by a temporal adverb, another clear instance
of different languages ringing the changes within a narrowly defined set of
possibilities.
The overtly volitive element in Sp. siquiera ('if you wish', i.e. 'even if')
is noted below.
Pattern (iv) relates to the explicit marking of an extreme scalar position.
84 MARTIN HARRIS
Here one finds conjunctions of the type Cat. per més que, Sp. por más que,
which parallel the pattern por + adj + que + noun/verb ('however adj
(that) noun/verb') 22 ; from their literal interpretation of 'however most(ly)
that' (that is, at one end of a given spectrum), the development of a con
cessive meaning ('although' 'even if') is well in line with what we have
already seen. Modern Occ. emai/amai ( < et magis 'and more') belongs here
also.23
Perhaps more striking, however, is the development of a phenomenon
we have already noted, namely the early use of adverbs meaning 'good' or
'ill' to reinforce the so-called 'concessive' subjunctive, that is, a structure in
which a subjunctive verb is to be interpreted concessively in context. In
French, the resultant conjunction bien que dates from the 14th century, and
the cognate conjunctions in all the sister languages except Rumanian, noted
earlier, are flourishing. Occitan tot ben que combines patterns (ii) and (iv).
It is interesting to note, in view of the discussion of the French structure
avoir beau, below, that in the Roman dialect, bello che 'fine that' is an alter
native to benchè, the use of this slightly different 'ameliorative' element 24
with the same value casting further doubt on the 'quantitative' explanation
of the concessive use of derivatives of bene. (The use of bene in conjunction
with si is discussed further below).
Once one inspects the relevant data closely, there are a surprisingly large
number of 'pejorative' elements which develop from reinforcers into the in
itial element of concessive conjunctions. We have already noted the simple
Cat. mal que, and also Fr. malgré que (cf. It. malgrado che, Cat., Occ.
malgrat que), the semantic development of this conjunction being particular
ly clear in the idiom ''malgré qu'il en ait' (sometimes written mal gré) literally
'ill-will that he may have for it', that is 'although he is unwilling', 'whether
he likes it or not'. The development of this structure is clearly via 'whatever
ill may come of...', as has already been noted, and malgré is the normal
preposition for 'in spite of' in contemporary French.25 Derivatives of male,
however, are by no means the only elements found in this role. One finds,
for instance, Cat. α pesar que, Sp. a pesar de que, (a pesar de = 'in spite
o f ) , where the original sense of the noun pesar is 'grief', whereas in a
despecho de the nominal element signifies 'despair'. These structures are of
course comparable to English 'despite', 'in spite of', where the source of the
lexical stem is despicere ('outrage', 'harm', 'scorn' etc.).26
Before leaving this discussion of words meaning 'good' and 'ill', we
should just note the intriguing French idiom avoir beau, as in 'on a beau pro-
tester, personne n'écoute' (lit. 'one has fine to protest, no-one listens', i.e.
'however much/although one protests, no-one listens'). This use of beau
(recall the remarks about Roman bello che earlier) developed as an epithet
of the infinitive ('one has a fine protest'), whence the sense 'fine for one to
protest (but)... ' (cf. 'well may you.. .but'). A particular clear example of this
transition in meaning can be found in Gougenheim (1951:206): "Tu as beau
la nommer cruelle Et bel estre son serviteur, Si n'en seras-tu pas vainqueur '
('It is fine for you to call her cruel and fine to be her slave, but you still won't
overcome her') 27
The fifth pattern noted in Table 2 consists of temporal conjunctions
used with the added nuance that an event etc. is unexpected at the time in
question. We have mentioned derivatives of iam 'already' above, and we
may observe that one (less frequent) value of a que in contemporary Sp.
is concessive. (Juanita, ya que no le amase, se deleitaba en su conversación:
'Juanita, although she did not love him, took pleasure in his conversation'
(Valera, cited by Harmer and Norton, 1957:223.) Ya que is also found with
this value in béarnais and gascon. (For a brief survey of derivatives of iam
used with a concessive sense, see Rivarola, 1976:149). Iam itself, however,
has largely given way to other temporal adverbs, in particular those with
values similar to that of encore, attested, it will be recalled, in the earliest
OFr. texts and still used, albeit in literary registers, in modern encore que
'although'. The etymology of nr seems to be inde ad horam28 'from then
up to the hour', i.e. 'until now' or 'still', and a similar phrase is the etymon
of the first element of It. ancorché and of Cat. and Occ. encara que. The
development of Port. ainda que (<ab inde 'from thence'), initially con
cessive conditional in value (Algeo, 1972 (3):536), is clearly semantically
comparable, while the first element of It. anche che is of uncertain but pro
bably temporal origin. (See, for example, Coromines, 1980-84, s.v. anc.) The
originally temporal structure which has prospered best is certainly the stan
dard Sp. concessive conjunction aunque (aun probably <adhuc 'until
now'), which since the 15th century has gradually ousted the earlier
maguer(a) (que), which was discussed at length above.
Many primarily temporal conjunctions can have a concessive value in
certain circumstances: cf. the earlier discussion of Latin cum and English
'while'. Reasons of space mean that we shall look here only at certain com
binations involving derivatives of Latin quando and directly analogous con
structions.
French, Spanish, and Italian all make use of derivatives of quando
86 MARTIN HARRIS
('when'), usually with appropriate and familiar reinforcers, in the concessive

conditional ('even if') sense. It will be recalled that when referring to events
still to come, there is a scale of likelihood ranging, in everyday language at
least, from 'when' through 'if + non-modal verb form' to 'if + maximally
modal verb form' ('when he comes'/'if he comes'/'if he came ~ were to
come'). A future-referring concessive conditional is both an unlikely
hypothesis about an event yet to occur, and a statement that if, and therefore
when, it is realized, the consequent is nevertheless still certain (or as certain
as any future events can be). In this sense, 'when' marked with an ap
propriate diacritic is not at all unsuitable to carry this particular value. (For
a discussion of words meaning 'same' used in this way, see below.) Certainly
one finds lors ê que, alors (ê) que and in particular quand (bien)
ê in the sense of 'even if', in structures such as quand bien ê tu
aurais (cond) raison, je n'irais pas ('when well even you would be right...'
i.e. 'even if you were right, I would not go'). 29 In Italian quand'anche is used
with similar values. In Spanish, the nearest equivalent structure aun cuando
'even when' may be equivalent to either 'even if' or 'even though', by virtue
of a modal opposition discussed below, while cuando alone is used on occa
sions with both concessive and concessive conditional value in the contem
porary languages (Alcina and Blecua, 1975:1110).
The final group of conjunctions which we shall look at comprises rein
forcements of derivatives of si 'if' (pattern (vi) in Table 2). We have seen
earlier that si can appear not only in concessive conditional ('even if') ter
ritory, but also with a straightforwardly factual concessive value, given an
appropriate context. This is even more true, as one would expect, of rein
forced forms, from etsi and the like in Latin (see above) through to Fr.
êsi, Rum. chiar dacǎ, It. anche se 'even if', Cat. si bé, Sp. si bien, It.
sebbene 'if well' and Sp. siquiera 'if you wish'. (Modal usage will be discuss
ed below.) The identity and value of the reinforcements will generally be
familiar from what has preceded, with the addition of words meaning
'same', the clear implication being that the out-turn is the same as was to
be expected despite the apparently incompatible antecedent. (For an analysis
of what she calls 'insinuating words' which have the effect of contradicting
expectations, see Manoliu-Manea, 1985 and the works cited therein.) Con
sider 'all the same' in the sense of 'nevertheless...' within the consequent,
(a usage found also, for example, with Cat. tanmateix 'so the same' i.e.
'nevertheless') alongside its use also within antecedents, in structures such
as 'it's all the same if he does come, nevertheless...' (A similar bi-clausal
distribution is found with Fr. quand même, discussed above.) There is also
one rather curious addition to the range of reinforcers, found in It. seppure
('even if). One contemporary meaning of pure is 'even', and it is in this
sense that it has been colligated with se, but it derives originally from an ad
jective/adverb meaning 'clean(ly)', apparently via meanings such as 'truly
you' (as opposed to others) (Rohlfs, 1969 (III): para 963), i.e. 'you alone'
'you yourself', thence via the familiar meanings of 'same' and temporal
'still' to 'even' (cf. the various values of même 'self' 'same' 'even' in
French). The semantic development of Rum. chiar, from Latin clarum
'clear', almost certainly parallels that of pure in Italian: for an interesting
attempt to account for this semantic shift within a pragmatic and
psychomechanical framework, see Manoliu-Manea (1985).
We are now in a position to look briefly at our final topic, namely
modal usage in concessive and concessive conditional clauses in Romance.
The pattern is less than clear because in this semantico-syntactic area more
than most, the different factors motivating the use of one mood rather than
another in diverse états de langue interact in a multiplicity of overlapping
ways. The position in Latin30 was, broadly speaking, that the use of one
mood rather than another was semantically governed, depending on the
speaker's wish, or otherwise, to add a modal overtone to the clause (main
or subordinate) in question. Already, however, a process had begun, very
slowly at first, whereby the use of the subjunctive in certain dependent
clauses came to be viewed as an appropriate marker of subordination as
such, that is, came to be syntactically governed. A further complication is
that particular lexical items — matrix verbs, conjunctions etc. — came at
times, not always for any obvious reason, to be colligated with a particular
mood, regardless of an earlier semantic distinction which called for the use
of different moods in different circumstances. If one adds to this already
complex picture the fact that a change in one area of a language — e.g. the
development of an adverb into a conjunction — may not, initially or even
ever, provoke an expected consequential change on another area, and also
the fact that the different Romance languages have evolved at different
speeds, then the difficulty of providing a simple and all-embracing analysis
of modal usage in concessive clauses should be clear.
Despite this, one can make a number of observations. Essentially, con
cessive clauses are factual in content, and hence the indicative mood would
appear to be appropriate, whereas in concessive conditionals the hypothesis
is logically prior to any subsequent concession, and modal usage should
88 MARTIN HARRIS
therefore correspond to that found in conditional clauses, that is, the sub
junctive mood in all cases except that of 'open' conditions, i.e. those most
likely to be realized. A look at Latin quamquam ('although'), almost always
found with the indicative at least until the time of Livy (the later subjunctives
being based on the analogy of quamvis), and at etsi and other compounds
of si ('even if), generally found with the subjunctive mood but with the in
dicative mood where this is semantically appropriate (in particular in factual
concessive clauses: cf. once again Ernout and Thomas, 1953:351), would sug
gest that the Latin position very closely corresponded to this ideal. Two per
turbations need to be noted, however: firstly, conjunctions such as quamvis
with a volitive origin generally31 pattern with the subjunctive (whether or not
the relevant clause content is presented as factual) either because of the
jussive origin of the dependent clause ('let him be as intelligent as you wish')
or because volitives as a semantic class pattern with subjunctives in
Romance, or (most likely) for a combination of these two reasons. (In
stances of the indicative must be taken to be on the analogy of quamquam.)
Secondly, concessive cum patterns with the subjunctive, even when actual
facts are being reported. This is a case where a process of grammaticalization
can actually be seen at work in Latin: in early Latin, factual concessives with
cum were expressed in the indicative mood, whereas later the syntactic class
of concessives was largely marked by subjunctives (but see also note 8) and
this therefore became the appropriate mood with cum — as it tended to do
even with quamquam — regardless of the semantics of the particular clause
in question.
The Romance concessive markers, as we have seen, are not the direct
descendants of their Latin predecessors; the concessive conditional markers,
however, broadly are, and we shall deal with these first. Si is one of the very
few conjunctions with specific meaning to have unbroken history in
Romance, and the history of Romance conditional clauses has been surveyed
in great detail. (For a recent summary, see Harris, 1986.) We can say quite
simply that modal usage in concessive conditional clauses, where the con
junction used is a derivative of si or a compound thereof, is exactly what
would be predicted in the language in question. Thus, where (for instance)
Sp. si bien introduces either an open condition or a (factual) concession, the
indicative mood is found; elsewhere, the subjunctive mood is used, as ex
pected. A similar pattern is found with It. seppure and anche se, while in
French, where the subjunctive has been ousted from conditional sentences,
êsi is used with the same paradigms of the indicative as si alone, in the
same circumstances. One apparent exception to the general picture concerns

It. sebbene, (se + bene), which patterns like benché (see below) rather than
like compounds of se. Schmitt Jensen (1970:420-1, 442, 506) sees this as a
consequence of the early fusion of this particular combination: to this, one
might add its concessive rather than concessive conditional value.32 Be that
as it may, however, the categorical statement in Battaglia and Pernicone's
standard Italian grammar (1963:559) that "il modo della proposizione con
cessiva è sempre il congiuntivo" ('always the subjunctive', italics added)
cannot be sustained: on the very same page of that very same grammar, we
read "sebbene mi annoierò (fut indic), pure ci vengo" (even though I'll (cer
tainly) be bored, I'm still coming'), showing that, in relation to non-past
time at least, a concessive:concessive conditional distinction can be made
even with this particular conjunction.
Concessive clauses introduced by compounds with the second element
que/che present a simpler picture, at least in literary registers of French and
Italian and generally in Portuguese: they require the subjunctive.33 Whether
this reflects the process of grammaticalization we saw beginning already in
Latin, or whether (more plausibly) it reflects the origin of these conjunctions
as adverbial reinforcers of paratactic 'concessive' subjunctives in the earliest
periods of Romance, the rules of normative grammar are clear. Equally,
however, the indicative has also been found with all the relevant conjunc
tions at all times, at least in French34 (Grevisse, 1980: para 2680,2), this
mood being used in the contemporary language, above all in popular
registers, to stress the factuality of a concessive antecedent (op. cit.: para
2677).,35 In Italian, benché, ancorchè and apparently sebbene (but recall the
example above) "sont en effet toujours suivi du subjonctif — dans n'im-
porte quel contexte" (Schmitt Jensen, 1970:421: in popular registers,
however, in the case of factual antecendents, the indicative is not unknown.
Finally, in the case of Sp. aunque we may note an explicit recognition of
what seems actually to be close to the truth in French and Italian: aunque
always takes the subjunctive when used non-factually, and may but need not
take the subjunctive when the antecedent is represented as factual. Put in
other words, and looking at the picture as a whole, we find that pressure for
consistent modal use with one (or a group) of conjunctions is in conflict with
pressure for the use of one or other moods to be semantically motivated, the
former pressure having (perhaps, decreasingly) the upper hand in contem
porary Romance. (For historical observations on a similar theme, cf. Algeo,
1972/3: 541.)
90 MARTIN HARRIS
The reference to the modal opposition in the case of Sp. aunque leads
naturally to a brief mention of aun cuando (lit. 'even when', used similarly
with the indicative in a concessive and the subjunctive in a concessive condi
tional clause (Alcina and Blecua, 1975:1110; for a brief historical survey, see
Rivarola, 1976:68, 115, 146).
There is clearly a great deal more that one could say about concessives
in Romance, both about the semantics of the clause-types themselves, and
of the fields of meaning which provide the adverbs which lead in the end to
the new conjunctions. One could talk about the interrelationship between
concessives and indefinite relatives, about the use of inverted word order as
a marker of concessives as it is of conditionals (Grevisse, 1980: para 2684),
and about non-finite and elliptical structures with concessive value, an im
portant method of clause linkage, as indicated by Lehmann (this volume).
One could mention other patterns, for example of the 'notwithstanding' type
(e.g. Sp. no obstante) (cf. type (iii) in König, 1985:373) or the 'but that' type
(Sp., Port. pero que, cf. Rivarola, 1976:53f, 87f) about which there seems to
be nothing to say that can be illuminated by Romance data. Perhaps most
interesting of all, one could examine in greater detail the link there must be
between words meaning 'also' 'already' 'even' 'still' 'yet' 'same' and the
like. This and much more remains to be done. Nevertheless, one can draw
some general conclusions on the basis of historical Romance data which
strengthen the view of concessives and their relationship to other clause types
which has been emerging from recent work.
The notion 'concession' is not always explicitly marked by a specific
subordinator or the equivalent in a particular language. A conditional
marker (particularly if reinforced) and/or an adversative co-ordinator will
often serve the purpose just as well. Where concession may be marked in
some way, the further distinction between concessive conditionals and con
cessives may not be formally made (König, 1985:367). When this last distinc
tion is made in a language, explicitly concessive conjunctions derive either
from (reinforced) conditional markers used as they may be in factual (non-
conditional) contexts (Lat. etsi, Eng. 'although' Ger. obwohl It. sebbene)36
or from a range of constructs reflecting the entailment of the consequent in
a concessive sentence whatever prima facie incompatible value is attached to
a particular antecendent. Various subdivisions of this second source are ap
parent, and have been highlighted in this presentation for expository
reasons, but this should not obscure the fact that there is here one seman-
tically unified category, albeit one containing a fairly wide range of nuances
within itself.37 Even the use of temporal conjunctions with a concessive value
('while A is true and this would normally preclude C, nevertheless C') can
readily be seen as a subdivision of this broad category. It is hoped therefore
that this paper, by concentrating on the history of one clause-type in one
language family, has not only illuminated the evolution of that clause-type
in that language family but also thrown further light on the semantics of
concession more generally and its relationship with other adjacent areas of
meaning.
More generally, the examination of concessive sentences over such a
prolonged period enables us not only to observe many of the patterns of
clause linkage described by Christian Lehmann (this volume), but also to
study changes in these patterns over time. In particular, the continuum
which he labels "hierarchical downgrading" is clearly illustrated in the
analysis presented earlier of the Romance data. In synchronic terms, one
may find at a given time structures ranging from the juxtaposition of two
independent clauses through to archetypal instances of what Matthiessen
and Thompson (this volume), following Halliday, call "enhancing
hypotaxis". Further, the material confirms that there is no clear-cut distinc
tion between 'co-ordination' and 'subordination' as these terms are tradi
tionally used, and that these grammatical categories are as fuzzy as the
cognitive categories which they may serve to mark.
The latter point, made explicitly by Matthiessen and Thompson (this
volume), is particularly apparent in the domain of conditionals and con-
cessives, as earlier work by König and myself clearly indicates. What is
more, the fascinating discussion of integrative and non-integrative word-
order presented by König and Van der Auwera (this volume) suggests strongly
that the precise grammatical form of an utterance along a given spectrum
may correspond to some extent at least with the precise semantic nuance
within what is increasingly recognised, as we have noted several times, as a
continuum of meaning rather than as a series of conceptually discrete
categories. The non-discreteness of the grammatical categories themselves,
however, initially discussed in Haiman and Thompson's illuminating (1984)
paper, has not previously emerged as clearly as it does from a consideration
of the phenomenon which Lehmann refers to (this volume) as "syntagmatic
interweaving" within concessive sentences. Broadly, one can see in the
evolution of these sentences in Romance (and indeed widely elsewhere — see
for example König and Van der Auwera, this volume) a structure consisting
of two "adjoined main clauses" (the 'adjunction' consisting of a semantical-
92 MARTIN HARRIS
ly appropriate adverb or reinforcing particle stressing the truth of the conse

quent) developing into an 'interlaced' structure by the passage of the 'ad
joining' element from the consequent to the antecedent, where it may either
form the basis of a new conjunction or be added to an already present con
junction. In terms of the "location of marking" parameter identified by
Matthiessen and Thompson (this volume), we may say that the exponent of
the precise semantic relationship in question passes from the nucleus to the
satellite, a development which seems to go hand in hand with a move along
the downgrading hierarchy referred to earlier. In these various de
velopments, however, there is no one decisive point where one can speak
definitively of the passage from, for instance, parataxis to adjunction to
hypotaxis: while the archetypes are (by definition) clear, the boundaries be
tween them are blurred and indistinct.
There are many instances of this process in the data presented here. The
development is particularly clear in the case of items with the value 'same'
(e.g. Fr. même), where the passage from 'if he comes, I'm going (all the)
same' to 'same if he comes, I'm going' (Fr. êsi) is semantically
transparent, although the precise steps whereby the item in question moves
from one clause to the other need further study. One possible route is the
ease with which an adverb/preposition can develop into a conjunction in
many languages: thus we can readily envisage a transition from 'he is com
ing; in spite (of that) I'll (still) go' to 'in spite that he's coming, I'll (still)
go'. (Consider French malgré (que) in this light.) The et of Lat. etsi, the bene
and pure of It. sebbene and seppure, the chiar of Rum. chiar daca all started
out in the (semantic) consequent, that is, in the clause which eventually
became the 'main clause' in traditional terminology. Further work is needed
on the stages between the beginning- and end-points of this shift.
One final point to note is the different preferences apparently displayed
at different periods for particular grammatical structures to express the
discourse-semantic category 'concessive'. To give just one example, a super
ficial examination of the evolution of concessive sentences from Latin to
contemporary Romance would suggest a swing from one extreme to another
and then back again in respect of at least one of the clause linkage continua
identified by Lehmann (this volume), namely from embedding to parataxis
and back to embedding. In fact, however, the position seems to be rather
that the paratactic structures have been available at all times and that the ar
chetypal embedded structures are, to some extent at least, the reflex of a
society in which literary registers develop and assume importance. Put in
other words, paratactic and simple co-ordinate structures have consistently

been grammatically appropriate methods of marking a discourse-semantic
category which has always needed to be expressed, but the 'enhancing
hypotaxis' so characteristic of (Classical) Latin was largely lost in early
spoken Romance (recall the failure to survive of any Latin concessive subor-
dinator) and was subsequently re-created in ways which this paper hopes to
have made clear. All in all, the study of concessive sentences seems to be a
particularly fruitful area to increase our understanding of the processes of
clause-combining in grammar and discourse.
ACKNOWLEDGEMENT
I am grateful to Neville Collinge, Giulio Lepschy, Nigel Vincent and Max

Wheeler for their most helpful comments on an earlier draft of this paper.
NOTES
1. Of course one can find the past simple in an //-clause but not with a hypothetical meaning
(Harris, 1986:427). Conversely, in a concessive clause, which is factual, a past simple is
quite appropriate, a counterfactual concessive being a contradiction in terms.
2. Nigel Vincent reminds me, in respect of sentence (iii a), that counterfactuals, unlike other
conditionals, do in fact entail the negation of their consequents. This does not however
affect the point at issue here.
3. It seems impossible to accept the conclusion relating to. 'even if' conditional sentences
reached by Bennett (1982:413), that 'even if' conditionals do not entail their consequents.
However, this does not seem to require the abandonment of the convincing unified ac
count of 'even' presented elsewhere in Bennett's paper.
4. It follows from this that we believe that, on semantic grounds, sentences of the type
'however much money he spends, he will not avoid the problem' are to be classified as
concessive. It seems that to reject them on formal grounds — as does, for example, Mit
chell (1985:727) — is to fail to grasp fully the non-discreteness of both semantic and mor-
phosyntactic categories in this area. Of course in formal terms these are 'indefinite'
relative or adverbial clauses of diverse sorts, as he points out (op. cit., 736ff), but in seman
tic terms, once we are dealing with a main clause whose truth is asserted regardless of any
possible antecedent, then we are obviously in the (semantic) domain of concessives (cf.
Quirk, 1954:91-101). Furthermore, the frequent use of volitional elements in structures
of this type (Lat. quivis, quilibet, Sp. quienquiera etc, where vis, libet and quiera are all,
as we shall see, cognate with verbs of wishing or pleasing) clearly tends to confirm their
links with concessives, for reasons elaborated below. Only by virtue of limitations of
space will such structures be largely ignored here.
94 MARTIN HARRIS
5. There is a considerable degree of overlap between the categories proposed here and those
suggested in König (1985:372-3), which reached me after this paper had been prepared.
In particular, our (i)-(iii) are subdivisions of his (i) ('universal quantification') while (v)
and (vi) are subdivisions of his (ii). His (iii) is largely passed over in this paper while in
respect of his and my (iv) rather different — though I suspect complementary — analyses
are proposed.
6. For a discussion of a number of the conjunctions found in Latin, cf. Haiman
(1974:348-9).
7. 'Will' in structures such as 'Try as you will' is presumably comparable in origin.
8. For discussion of modal usage with concessive cum, cf. Ernout and Thomas (1953:354)
The fact that cum is a temporal relative ('at such time that') and that generic relatives
came to require the subjunctive mood complicates any discussion of modal usage in con
cessive cum-clauses.
9. The first element of etsi is et 'and'. Another element with the same value in Latin is
enclitic -que, which is directly cognate with the final element of OE h 'though', the
-h serving to reinforce an adversative adverb in a way seen throughout this paper.
10. OE gif ( = 'if') could apparently be used concessively, albeit only rarely (Burnham,
1911:92; Mitchell, 1985:704).
11. It is interesting to note in this connection the observation by Algeo (1972 (3): 539), as a
result of his detailed analysis of concessive conjunctions in medieval Sp. and Port., that
"in every instance conjunctions that at first introduce only -R [i.e. concessive conditional]
events expand in function [i.e. to concessivesl". In general, changes from concessive con
ditional to concessive are markedly more frequent than the converse (Algeo, op. cit.,
541-2): König (1985:364) indeed speaks of the concessive development as being a "dead
end street", in effect the end point of a particular set of semantic changes. In this connec
tion, it is interesting to note that in all the examples which König (1985) gives (e.g.
pp.375-6) of 'even if' used with a concessive value in modern English, 'even though' is
at least equally good. The converse substitution would not however be possible.
12. Examples such as 'even though he may/might come tomorrow' do not appear to me to
contradict this claim. Such a clause, which is not synonymous with either 'even if he
comes' or 'even if he came', seems best interpreted as 'even though it is a fact that there
is a possibility of his coming...' We may perhaps see this as a 'subjunctive of non-asser
tion', used in this cáse not really to cast doubt on the truth of the antecedent but rather
to indicate that its relevance is unexpected or to be discounted.
13. It is perhaps worth recalling in passing the claims associated with one particular group
of sociolinguists in the nineteen sixties that one characteristic of so-called 'restricted code'
was the avoidance in 'uneducated' speech of certain subordinators such as 'although' in
favour of the equivalent co-ordinator, e.g. 'but'. (See for instance Dittmar 1976:60-1 for
a review of this aspect of claims made by Basil Bernstein.) For an illuminating discussion
of the semantic relationship between 'but' and 'although', cf. König (1985:368-9).
14. As we have seen, Dardel finds the case 'not proven': hence his use of 'éventuellement'.
The best guess seems to be a more or less simultaneous borrowing into the southern
Romance languages, perhaps as early as the sixth century, at least in north Italy. The
CONCESSIVE CLAUSES IN ENGLISH A N D R O M A N C E 95
word apparently never took root in northern Gaul.

15. Le voyage de Charlemagne à Jerusalem et à Constantinople, lines 454-9.
An approximate translation is
King Hugh the strong has no young man,
In all his household, however strong and wise he might be,
(even if) he were to put on two halberks and two helmets attached
And be seated on a speedy, calm steed...
(Whom) I shall (not) strike on the helmets...
Neville Collinge draws to my attention a paratactic concessive conditional in Latin
(Horace, Serm 1.3.15ff): "decies centena dedisses...nil erat in loculis" '(even if) you had
given him a million, there was nothing left in his cash-drawers.'
16. But see, for instance, Tekavcic (1972:656-7) for a clear exposition of the same develop
ment in the history of Italian: an earlier comprehensive summary of the position in that
language (with a brief postscript on Provengal) is to be found in Miltschinsky (1917).
Adverbial reinforcement of this kind was certainly found in Latin, and may be presumed
to have an unbroken history in Romance.
17. All is found in this sense sporadically in earlier periods of English: see OED s.v. all 
IO for examples such as 'We brought more than ye we able to answer, all were it no Scrip
tures, nor Councels, nor Doctours' (1560). The for all that structure is also found early:
cf. 'They have beaten us openly...for all that we are Romans' (1526) (OED s.v. all A 9).
All if and if all, also attested, have not survived, while the intensifying element prefixed
to (al)though has survived, in a conjunction which has become exclusively concessive in
value. (cf. It. sebbene, Ger. obwohl, below and note 32.)
18. In earlier periods of English, 'how well' was also found, in a usage stated categorically
by OED to be 'after French com bien que'. A 1471 example: 'I have not Intencion for
to obey his comandement how well that he is my fader' (OED s.v. well 8 c ) . Recall also
Ger. obwohl ('if well'), a normal marker of concession.
19. Of course in paratactic structures of this kind, a clear formal distinction between con-
cessives and concessive conditionals is not possible, at least not out of context.
20. Quoi is a normal tonic development of Latin quid/quod.
21. Sp. Amer. use of ojala mentioned above also belongs in this section. Note the general
absence of the explicit subordinator with this set of conjunctions, although in earlier Sp.
maguer(a) was found with que (by analogy with the newer concessives discussed later)
before it was lost from the language, while che is found in the cognate structure in various
Italian dialects. (For a discussion of this phenomenon, cf. Herman 1963:232).
22. For example 'por altos que sean los arboles' 'however tall that may be the trees'.
23. This use of 'most' may be compared with the structures discussed by Fauconnier
(1975:188-9), in particular the acceptability of
(a) The most elegant suit looks bad on Alec
and the unacceptability of
(b) *The most elegant suit looks good on Alec.
96 MARTIN HARRIS
If one reformulates (a) as 'However elegant a suit may be, it looks bad on Alec' and hence
as 'Although Alec wears a most elegant suit, it looks bad on him', then one can see clearly
the operation of the relevant scale. (If Alec looks bad in a most elegant suit, what hope
has he in any less 'extreme' form of attire?) This same observation explains the
strangeness of
(c) ?The bluest suit looks good/bad on Alec
since 'blueness' does not normally have any inherent connection with looking good or
bad: there is simply no relevant pragmatic scale.
The interpretation offered here is intended to complement and not to contradict that
offered by Fauconnier.
24. Recall too the earlier discussion of Cat. baldament. It is interesting to note also that one
current concessive conjunction in German is obschon ('if-already'), where schon derives
from schön 'beautiful' via the stages 'beautifully (prepared etc.)' to 'properly' to
'already'. In Danish, skjønt (without the conditional element but otherwise from the same
root) also has the value 'although'.
25 For a detailed discussion of the history of malgré (que) in French and its usage in the con
temporary language, cf. Grevisse (1980: para 2522).
26. König (1985:373, 377) notes of this latter group (and comparable forms in other
languages) that the notions concerned are applicable originally only to human agents and
experiencers, and that the subsequent loss of this restriction, which he calls 'bleaching',
is only a "special manifestation of the general type of semantic change leading from 'con
crete' to 'abstract'." Such factors may indeed have reinforced in these particular cases
a general predisposition on the part of structures having the value 'whatever good/ill may
come of it' to develop a concessive interpretation.
27. Giulio Lepschy draws to my attention comparable Italian examples of the type "hai un
bel dire, nessuno ascolta" lit. 'you have a fine saying' ie 'though you may say what you
will (however much you say), no-one listens'.
28. For a full discussion cf. FEW s.v. hora (Von Wartburg, 1922-68).
29. This example is taken from the Collins Robert Dictionary, s.v. quand (c), where the transla
tion of quand bien ê is given 'even though' or 'even if'. This is not in fact accurate,
for reasons discussed earlier: 'even though' is appropriate in contemporary English only
for antecendents taken, unlike this one, to be factual. 'Even though you were right' is of
course perfectly grammatical in English, but in a different sense and requiring a different
(concessive) translation in French, e.g. with bien que etc. The examples of these conjunc
tions noted by Grevisse (1980: para 2678) with the indicative mood refer in effect to
general truths, and are, as he himself notes, equivalent to 'even when' rather than to 'even
if' or 'even though'.
30. The changing use of the indicative and subjunctive moods in Latin and Romance is set
out in detail in Harris (1978:166ff).
31. Various conflicting analogies lead to exceptions to all of the general patterns being
described here.
32. Cf. Ger. obwohl lit. 'if + well' , now used in the sense of 'although'.
33. A similar pattern is found in OE, where the nearest equivalent conjunction eah 'though'
is used with the subjunctive mood quite independently of the reality of the concession
(Mitchell, 1985:720-3).
34. For a clear survey, cf. Haase (1969: para.83). It is interesting to note that Malherbe
specifically calls for the indicative in factual concessions, and that Vaugelas frequently
makes use of this mood.
35. Interestingly, in a language such as French where the subjunctive mood has lost virtually
all of its semantic value, it is the conditional (not the subjunctive) which is used to stress
non-factuality. This point is developed in more detail in Harris (1986).
36. Nigel Vincent points out that in certain languages, this may be the only way of marking
concession, citing Hewitt (1979:45): "Abkhaz has only the one means of expressing con
cession, and that is by saying 'even if' — the clitic gə 'even' is suffixed to the appropriate
conditional form."
37. David Holton had drawn to my attention the existence in Modern Greek of concessive
markers representing almost all the patterns described here. Structures involving ólo ('all'
n.sing.) preceded by a preposition are common in all registers, particularly frequent being
m'ólo  ('with all that (rel.)') and par'ólo  ('but for all that'), while the volitive pat
tern is represented by ki as ('and let [it be that...]') (ki < kai 'and'). Concessive markers
involving the element kalá ('well') are also attested: angalá, angalá kai ('if well (and)')
were common in late medieval texts from Crete and Cyprus (where there was significant
contact with Romance) and comparable structures apparently survive in dialects. Similar
ly, kalá kai ('well and'), with or without a marker of subordination, is very common in
older Cretan texts and also survives dialectally. The most frequent structure of all is an
kai ('if and'), a combination which we have seen frequently to develop concessive mean
ing elsewhere. {Ki an has rather the value 'even if', i.e. is a concessive conditional
marker.) It is interesting to note how many of the structures incorporate kai 'and', a pat
tern that was found already in the classical language, which used kaí ox kaíper with a par
ticiple as the principal mode of marking concession. This use of kai is reminiscent of the
use of an' in a conditional and/or concessive sense in earlier English: OED s.v. and 
cites for instance a 1658 example "Religious they will be an't be but for the benefit they
receive thereby" i.e. 'even though it will only be for...' Consider also the references to
-que and -h in note 9.
REFERENCES
Alcina Franch, J. and J.M.Blecua. 1975. Grammatica española, Barcelona:
Editorial Ariel.
Aigeo, J.E. 1972/3. "The concessive conjunction in Medieval Spanish and
Portuguese: its function and development". Romance Philology XXVI:
532-45.
Battaglia, S. and V.Pernicone. 1963. La grammatica italiana. Torino:
Loescher.
Bennett, J. 1982. "Even if". Linguistics and Philosophy 5:403-18.
98 MARTIN HARRIS
Bourciez, E. 19675; Éléments de linguistique romane. Paris: Klincksieck.

Burnham, J.M. 1911. Concessive Constructions in Old English Prose. Yale
Studies in English 39, New York.
Corominas, J. 1954. Diccionario crítico etimológico de la lengua castellana.
Bern: Francke.
Coromines, J. 1980-84. Diccionari Etimològic i Complementari de la
Llengua Catalana. Barcelona: Curial Edicions Catalanes.
de Dardel, R. 1983. Esquisse structurale des subordonnants conjonctionnels
en roman commun. Genève: Droz.
Dittmar, N. 1976. Sociolinguistics. London: Arnold.
Ernout, A. and F.Thomas. 19532. Syntax latine. Paris: Klincksieck.
Fauconnier, G. 1975. "Polarity and the Scale Principle". CLS 11: 188-99.
Gildersleeve, B.L. and G.Lodge. 1895. Gildersleeve's Latin Grammar. New
York and London: Macmillan.
Gougenheim, G. 1951. Grammaire de la langue française du seizième siècle.
Lyon: IAC.
Grevisse, M. 198011. Le Bon Usage. Paris-Gembloux: Duculot.
Haase, A. 19697. Syntaxe frangaise du XVII siècle. Paris: Delagrave.
Haiman, J. 1974. "Concessives, conditionals and verbs of volition". Foun
dations of Language (II):341-59.
—- 1978. "Conditionals are Topics". Language 54:564-89
-— and Thompson, S.A. 1984. "'Subordination' in Universal Grammar".
BLS 10:510-523.
Harmer, L.C. and F.J.Norton. 19572. A Manual of Modern Spanish. Lon
don: University Tutorial Press.
Harris, M.B. 1978. The Evolution of French Syntax: a Comparative Ap
proach. London: Longman.
-— 1986. "The evolution of conditional sentences in Romance". Romance
Philology, xxxix, 4: 405-436.
Herman, J. 1963. La formation du système roman des conjonctions de
Subordination. Berlin: Akademie Verlag.
Hewitt, B.G. 1979. Abkhaz (Lingua Descriptive Series). Amsterdam: North
Holland.
Klare, J. 1958. Entstehung und Entwicklung der Konzessiven Konjunk
tionen im Französischen. Berlin: Akademie Verlag.
König, E. 1985. "On the history of concessive connectives in English.
Diachronic and Synchronic evidence". Lingua 66:363-381.
König, E, 1986. "Conditionals, concessive conditionals and concessives:
Areas of contrast, overlap and neutralization". In On Conditionals, E.C.

Traugott et al. (eds.) Cambridge: Cambridge University Press.
Lehmann, C. 1973. Latein mit abstrakten Strukturen. München: Fink
(Structura).
Manoliu-Manea, M. 1985. Conventional Implicature and Cross Language
Analysis of "Insinuating Words". LAUT Series , Paper 118. University
of Trier.
Martin, R and Μ Wilmet. 1980. Manual du français du moyen âge, II: syn
taxe du moyen français. Bordeaux: SOBODI.
Miltschinsky, M. 1917. "Der Ausdruck des Konzessiven Gedankens in den
Altnorditalien Mundarten". Zeitschrift für Romanische Philologie,
Beiheft LXII. Halle: Max Niemeyer.
Mitchell, B. 1985. Old English Syntax, II: Subordination, Independent
Elements and Element Order. Oxford: Clarendon Press.
Quirk, R. 1954. The Concessive Relation in Old English Poetry. New
Haven: Yale University Press.
Rivarola, J.L. 1976. Las conjunciones concesivas en español medieval y
clásico. Tübingen: Max Niemeyer.
Rohlfs, G. 1966-9. Grammatica storica della lingua italiana e dei suoi dialet
ti, I-III. Torino: Einaudi.
Sandfeld, . 1936. Syntaxe du français contemporain, II: les propositions
subordonnées. Paris: Droz.
Schmitt Jensen, J. 1970. Subjonctif et hypotaxe en Italien. Odense: Odense
University Press.
Tekavčić, P. 1972. Grammatica storica dell'italiano, II: Morfosintassi.
Bologna: Il Mulino.
von Wartburg, W. 1922-68. Französisches Etymologisches Wörterbuch.
Basel: Zbinden.
Woodcock, E.C. 1959.  New Latin Syntax. London: Methuen.
Clause integration in German and Dutch
conditionals, concessive conditionals,
and concessives1
E k k e h a r d König Johan van der A u w e r a
University of Hannover Belgian National Science Fund
& University of Antwerp (UIA)
INTRODUCTION
The distinction between 'coordination' and 'subordination' as well as that

between 'main clause' and 'subordinate clause' belong to the stock-in-trade of
the ordinary working grammarian. They form a basis for many generalizations
in a great variety of languages, and are themselves typically based on general
syntactic and semantic properties like the following (see Andersson, 1975;
Davison, 1979; and esp. Haiman and Thompson, 1984):
i. Grammatically signaled incorporation of one clause into another;
ii. Interpretative dependence of one clause on another in the assign
ment of mood, tense, subject, illocutionary force, or referents for
its pronouns;
iii. The inclusion of a clause within the scope of scope-bearing ele
ments of another clause;
iv. Intonational linking between two clauses;
v. 'Main clause phenomena'.
Work done in the last ten years on various languages, however, has
demonstrated clearly that there are important exceptions to each of the above
criteria (Andersson, 1975; Green, 1976; Davison, 1979; Handke, 1984). The
criteria upon which the distinction between coordination (or parataxis) and
subordination (or hypotaxis) are normally based are often inconsistent, and
identify different classes of clauses as subordinate. Those who have drawn
attention to this dilemma have proposed various solutions. One consists in an
102 EKKEHARD KÖNIG AND JOHAN VAN DER AUWERA
appeal to prototype theory, and in making a distinction between prototypical

and marginal cases of subordination. Another approach is to replace the
dichotomy 'coordination' vs. 'subordination' by a more elaborate typology of
how clauses may be linked. Van Valin (1984) and Foley and Van Valin
(1984), for instance, treat 'embedding' and 'dependence' as totally indepen
dent parameters, and define four different types of clause linkage in terms of
these parameters (see also Lehmann, this volume). The most radical solution,
proposed inter alia by Haiman and Thompson (1984), is to completely give up
a unitary grammatical category of 'subordinate clause'. 'Subordination' is
regarded as a composite term, analyzable into a number of independent
parameters, 'where each of these parameters involves a different relationship
which two adjacent clauses in discourse can have with each other' (Haiman
and Thompson (1984:520) see also Matthiessen and Thompson, this volume).
Their list of parameters includes the ones given in (i) through (iv) above.
1 V-2 AS A SIGNAL OF INCORPORATION IN GERMANIC
One of the goals of this paper is to contribute to the theoretical discus

sions on subordination and, more generally, the nature of clause combining,
by examining one parameter in detail, in connection with some Germanic lan
guages and with a very restricted semantic domain. The notional domain is
that of conditionality and concessivity; the Germanic languages focussed on
are Dutch and German, but we will also mention Scandinavian.
Languages such as Dutch and German, which have 'Topic - Finite Verb -...',
otherwise known as 'Verb-Second' or, simply, "V-2" as a basic word
order in independent sentences, provide a clear signal for the grammatical
incorporation (or embedding) of one clause into another: If a clause is fol
lowed by the finite verb of another, the former can generally be taken to func
tion as a constituent within the other.
(1) G. Weil er krank ist, kann Fritz nicht mitkommen
D. Omdat hij ziek is, kan Fred niet meekomen
because he sick is can Fred not come along
'Because he is sick, Fred cannot come along'
As an indicator for incorporation and thus a parameter for subordination, V-2
here makes the same prediction as two other parameters traditionally taken to
indicate subordination in Germanic: V-late placement and the presence of a
CLAUSE INTEGRATION IN GERMAN AND DUTCH 103
conjunction (complementizer).2 The assumption that the first clause in exam

ples like (1) is subordinate to the second therefore seems to rest on a solid
foundation.
However, the word order in (1) is not the only possibility. The initial
clause may also be followed by a resumptive (correlative) element (dann or so
in German, and dan in Dutch) and, under certain conditions, the second
clause may exhibit the word order which it has as a separate sentence, despite
the fact that the other indicators characterize the first clause as 'dependent',
'embedded', or 'subordinate'.
(2) G. Wenn er krank ist, dann bleibt er zu Hause
D. Als hij ziek is, dan blijft hij thuis
if he sick is, then stays he at home
'If he is sick, then he stays at home'
(3) G. Selbst wenn sie nicht alle gekommen sind, wir
D. Zelfs als ze niet allemaal gekomen zijn, we
even if they not all come are we
G. können mit dem Besuch zufrieden sein
D. kunnen met het bezoek tevreden zijn
can with the attendance content be
'Even if they haven't all come, we can be pleased with the
attendance'
Even though the initial clauses in (3) have two properties normally associated
with German and Dutch subordinate clauses, they cannot be assumed to func
tion as the first constituents of the second clauses. The three word order pat
terns illustrated by (1) to (3) and symbolized in (4)
(4) a. Adverbial Clause -Finite Verb - ... (see (1))
b. Adverbial Clause - Resumptive - Finite Verb - ... (see (2))
c. Adverbial Clause - Topic - Finite Verb - ... (see (3))
exist as options in both Dutch and German in the notional area of conditional-
ity and concessivity. In these two languages, conditionals, concessives, and
concessive conditionals may under certain conditions exhibit any of these
three word orders, so that we may get minimal pairs like the following3:
(5) a. D. Als je me nodig hebt, werk ik op het kantoor
if you me necessary have work I in the office
Tf you need me, I will work in the office'
b. D. Als je me nodig hebt, ik werk op het kantoor

'If you need me, I am working in the office'
(6) a. G. Selbst wenn er ein wenig betrunken ist, singt Pavarotti
even if he a little drunk is sings Pavarotti
gut
well
'Even when/if he is a little drunk, Pavarotti sings well'
b. G. Selbst wenn er ein wenig betrunken ist, Pavarotti singt gut
'Even if he is a little drunk, Pavarotti is (actually) singing well'
Considered in the light of the elaborate system of distinctions made by
Halliday (1985) or Lehmann (this volume), all three structures listed in (4) are
cases of subordination, since they form an endocentric construction with a fol
lowing clause as head. All three constructions also exhibit 'hypotaxis' in the
sense of Halliday, Lehmann, and Matthiessen and Thompson (this volume).
Only (4a), however, can be considered as a clear case of embedding in the
sense of functioning as a constituent within another clause. (4b) is generally
taken to exhibit a kind of left dislocation.
The main purpose of this paper is to examine in detail the three types of
clause linking given in (4): the formal properties with which they correlate,
the semantic distinctions that may be expressed by them, the conditions rele
vant for choosing one rather than another, their historical development, and
the differences and similarities between German and Dutch and, to a lesser
extent, Scandinavian.
Some of the phenomena to be discussed have attracted the attention of
linguists before. They have been dealt with under headings such as "excep
tions to V-2", "double filling of the forefield", and "peculiarities of word
order in complex sentences" (Behaghel, 1929; Horacek, 1957; Fleischmann,
1973; Kaufmann, 1974; Faucher, 1977, 1984; Van de Veide, 1979a, b;
Baschewa, 1980, 1983; Schelfaut, 1982). These studies have given us both
ideas and data, but they do not actually give much attention to the issues and
questions raised in this paper. None of the studies is written from a cross-lin
guistic perspective or tries to relate the synchronic diversity to the diachrony.
Finally, many of the explanations given in these studies (e.g. "foreground
ing", "omission of a clause", "stylistic expressiveness") are unsatisfactory. As
to the data, however, we will not go beyond our predecessors. We felt there
was enough to say about the examples and corpora already discussed in the lit
erature. This is not to say there is no need for more data. All of our corpus
data, for example, come from written sources.

Structures such as (4b) and (c) are not the only patterns standing out as
exceptions among the regularities of word order in complex sentences in Ger
man and Dutch. Other structures commonly regarded as 'dependent', 'subor
dinate', or 'embedded', but exhibiting V-2 include:
i. Subject and object clauses without complementizer
(7) G. Ich glaube, er ist krank
' I believe he is sick'
ii. Clauses coordinated with a preceding complement or adverbial
clause; in these cases, traditionally called 'anacoluthon' in German grammar,
only the first clause has V-late (cf. Behaghel 1903; Kolb, 1963; Faucher,
1984:23-24).
(9) D. Als jullie onderweg zijn en er verrast jullie
if you on the way are and there surprises you
een onweer dan mogen jullie ons opbellen
a storm then may you us ring
'If you are on the way and you are surprised by a storm, then
you can call us'
iii. Clauses introduced by connectives like German als 'as if, kaum
'hardly', and zwar 'true enough' or Dutch als 'as if'/'that', nauwelijks 'hardly',
and al 'even if/though'.
(10) D. Al roept hij om hulp, ik weiger
even if/though calls he for help I refuse
'Even if/though he is calling for help, I refuse'
(11) G. Kaum hatte er mich gesehen, (da) rannte er weg
hardly had he me seen there ran he away
iv. Clauses introduced by German weil 'because' or obwohl 'although'
and Dutch hoewel 'although', which do not normally cooccur with V-2.
(12) a. G. Sie liebt ihn, obwohl verdient hat er es nicht
she loves him although earned has he it not
'She loves him, although, he doesn't deserve it'
b. G. Sie liebt ihn, obwohl er es nicht verdient
she loves him although he it not deserves
'She loves him, although he does not deserve it'
Problems such as these, all of which concern the word order within so-
called 'subordinate' clauses, have received more attention in the literature

than problems of ordering in connection with clause integration, and will not
be addressed here. Analogous problems in English have been discussed under
the headings 'root transformations' or 'main clause phenomena in subordinate
clauses' inter alia by Hooper & Thompson (1973), Green (1976), and Bolinger
(1977). Insofar as these studies are concerned with alleged conflicts between
various criteria assumed to be jointly relevant for embedding, subordination,
and hypotaxis, however, they provide an important background and source of
inspiration for our investigation.
2 CONDITIONALS, CONCESSIVES, AND CONCESSIVE CONDI-

TIONALS
An essential prerequisite for our investigation of the three types of clause

linking given in (4) is a brief discussion of the adverbial clause types that allow
these three possibilities. The distinction between conditionals, concessives,
and concessive conditionals can best be drawn on the basis of the semantic
relations existing between the propositions expressed by the relevant complex
sentence, and those expressed by its component clauses. Thus we can say that
conditionals like (13) entail neither their antecedent nor their consequent,
whereas concessives like (14) entail both of their component clauses.4
(13) If John gets some help, he will complete his work in time
(14) Even though John got a lot of help, he did not complete his work
in time
Concessive conditionals like (15) have properties in common with both
conditionals and concessives.
(15) a. Whether John gets any help or not, he will never complete his
work in time
b. No matter how much help he gets, he will never complete his
work in time
c. Even if John gets a lot of help, he will never complete his work
in time
Even though the sentence types exemplified in (15) are basically conditionals,
they differ from the latter in relating a set of antecedent conditions to a con
sequent. This set may be specified in at least three different ways, and accord
ingly we can distinguish at least three different types of concessive condition-
als. The set of conditions may be given by (i) a disjunction 'p or not p' as in
(15a), (ii) a free-choice expression as in (15b), or (iii) a focus particle or scalar
expression as in (15c). In (15c) a set of antecedents is specified by asserting a
conditional relationship for an extreme (unlikely) value on a scale of possible
values. By implication, this relationship can also be assumed to hold for other,
more likelv values on the same scale.
(15') c. if you drink (just) a little, your boss will
several glasses, fire you
a lot,
Sentences like (15b) can be assumed to have the following 'logical form':
(16) (Vx) ((he gets x amount of help) > (he will never complete his
work in time))
Since concessive conditionals assert a conditional relationship between a
consequent and a whole set of antecedent conditions, which exhaust the
whole spectrum of possibilities, the antecedent conditions are often irrelevant
for the consequent, and the latter is entailed just as in concessives. A second
property that concessives and concessive conditionals have in common is a
relationship of 'normal incompatibility' or dissonance between the two com
ponent propositions in the case of concessives, and between at least one of the
antecedent conditions and the consequent in the case of concessive condition
als.
3 HISTORICAL DEVELOPMENT
Even though the three types of word order given in (4), which we will call
'integrative' (4a), 'resumptive' (4b), and 'non- integrative' (4c), can all be
observed in older forms of Germanic (Behaghel, 1929), it is now generally
assumed that they are linked as stages in a historical development leading
from non-integrative via resumptive to integrative word order (i.e. (4c) >
(4b) > (4a)). This hypothesis has been proposed very explicitly by Ham
marström (1923:49-55) and Horacek (1957:429) for German - both also refer
to older work - and by Van der Horst (1981a:40- 41, 1982b: 181-182) (cf. also
Debrabandere, 1976; Bossuyt, 1985, forthcoming) for Dutch. It is further
more supported by the detailed investigations of Baschewa (1980, 1983), who
showed the sharp decline in the frequency of the resumptive word order in
German concessive conditionals and concessives from the the 18th to the 20th
century. For Dutch we get a similar picture if we compare the work of

Schmidt (1958), who studied concessive conditionals and concessives from the
13th to the 17th century, with that of Schelfaut (1982), who studied the corres
ponding present-day structures. Yet another type of support for the suc
cessive integration hypothesis is the present-day distribution of the three
clause types; we will discuss this in some detail in Sections 5 to 7.
That German and Dutch have undergone a process of successive clause
integration is probably related to their V-2 character. Moreover, if it is true
that German and Dutch have strengthened their V-2 patterns, i.e. moved
from a moderate V-2 language to a stricter one (see e.g. Gerritsen, 1978,
1984; Jansen, 1980; Bossuyt, 1980; but also Van der Horst, 1984:156-157,176-
177), then it is plausible to posit a relation between the increasing clause inte
gration and the increasing V-2 character. Both hypotheses are supported by
what we know about Scandinavian. The modern Scandinavian languages are
arguably stricter V-2 than German or Dutch - they are strongly V-2 even in
subordinate clauses; they can be argued to have increased their V-2 character;
and their conditionals, concessive conditionals, and concessives are clearly
better integrated than in either German or Dutch or in the earlier stages of
Scandinavian (see Bergqvist, 1886; Alving, 1916:93-98; and De Boor,
1929:108-122 on Swedish). Thus, the non-integrative word order has almost
completely disappeared. If it is acceptable at all, it requires a clear break,
indicated in the orthography by a dash or a colon, between the two compo
nent clauses. The normal word order in conditionals, concessive conditionals,
and concessives is the resumptive or integrative one. The examples below are
taken from Swedish.
i. Conditionals:
(17) Om det blir sol, (såldå) åker vi och badar
if it becomes sun then go we and bathe
?*vi åker och badar
Tf the sun shines, we'll go for a swim'
ii. Concessive conditionals:
(18) a. Åven om det är sent, (så) vill jag inte gå hem än
even if it is late then will I not go home still
?*/jag vill inte gå hem än
'Even if it is late, I still don't want to go home'
b. Antingen du går eller inte, (så) blir resultatet bra

whether you go or not then becomes result good
1*resultatet blir bra
'Whether you go or not, the result will be good'
c. Hur mycket du än vet, (så) behöver du veta mera
how much you still know then should you know more
?*du behöver veta mera
'However much you know, you ought to know more'
iii. Concessives:
(19) Fastän det inte är sent, (så) vill jag gå hem nu
even though it not is late then will I go home now
?*jag vill gå hem nu
'Even though it is not late, I want to go home'
4 CONDITIONS FOR WORD ORDER CHOICE — GENERAL

CONSIDERATIONS
Conditions that are relevant for the choice of the word order in the con
sequent come in three subtypes: they either force, allow, or favor a pattern.
Put differently, they make the pattern necessary, possible, or probable.
Among the patterns that get the label 'possible', we will distiguish between
marked and unmarked patterns. A pattern will be considered as 'marked' if it
only occurs when some special condition is met. A pattern will be 'unmarked'
if it occurs both when the special condition is met and when it is not. Note that
there is no direct relation between markedness and frequency. When the spe
cial condition allowing both the marked and the unmarked pattern obtains,
there is no prediction as to which will be more frequent. There is also no pre
diction as to how frequently the special condition obtains. Of course, there is
no principled objection against calling the frequent pattern 'unmarked' and
the infrequent one 'marked', but we will not do this. Yet another sensible way
of using the notions of markedness and frequency concerns the formal fea
tures of a construction. For example, given that nearly all German and Dutch
adverbial clauses start with a conjunction, and that only a few do not, the
former could be called 'unmarked' and the latter 'marked'. For this formal
distinction, however, we will use the terms 'canonical' and 'non-canonical'.
Again, there is no direct relation between canonicity, on the one hand,
and our use of the notions of markedness and frequency, on the other. A non-
canonical clause type allowed for the expression of some aspect of condition-
ality may well be unmarked as well as more frequent than a canonical clause
type expressing the same meaning.
Before we discuss each of the clause types described in Section 2, we have
to mention one factor that affects the word order patterns of conditionals,
concessive conditionals, and concessives alike, and, in fact, of all adverbial
clause types. In conditionals like (20), called "relevance conditionals" by
Johnson-Laird (1986), the antecedent does not make any contribution
to the truth conditions of the consequent, but describes the conditions under
which the latter might be relevant for the hearer.
(20) D. Als iemand mi] zoekt, ik ben in de bibliotheek
if someone me searches I am in the library
'In case someone is looking for me, I am in the library'
The word order exhibited by (20) is an instance of the general rule which
requires or, at least, strongly favors non-integrative word order in Dutch and
German, whenever an initial adverbial clause relates to the speech act per
formed in uttering the following clause, rather than to the proposition expres
sed by it. Thus, in (20), it is the assertion that the speaker is in the library that
is related to a state of affairs of someone looking for him/her and possibly ask
ing the hearer where the speaker is. The relevant adverbial clauses can there
fore be called 'speech act qualifiers' (cf. Andersson, 1976; Hermodsson,
1978:51; Davison, 1979:120; Haegeman, 1984; Longacre and Thompson,
1985). The following example illustrates the generality of this phenomenon:
(21) G. Um es deutlich zu sagen, wir sind bankrott
in order it clearly to say we are broke
'To say it clearly, we are broke'
As the adverbial clauses in (20) and (21) are about rather than part of the fol
lowing clauses, it is not surprising that languages choose to have them outside
the following clauses rather than incorporate them.
In Swedish, however, speech act qualifying antecedents do not require
non-integrative order.
(22) Om någon ringer, (då/så) är jag i köket
if someone rings then am I in kitchen
Tf someone calls, I am in the kitchen'
This shows again that Swedish favors clause integration more than either Ger-
man or Dutch. Interestingly, perhaps there is a difference between German

and Dutch, too. Given the acceptability of the German example in (23), dis
cussed in Hermodsson (1978:51), and the impossibility of its Dutch counter
part, German does not seem to be quite as strict about forbidding integrated
speech act qualifying antecedents as Dutch is.
(23) G. Wenn ich es offen sagen darf, halte ich das für
D. *Als ik het open zeggen mag, houd ik dat voor
if I it open say may hold I that for
G. einen Schwindel
D. een zwendel
a swindle
'If I may speak openly, I consider that a swindle'
In the rest of this paper, the speech act qualification uses of conditionals,
concessive conditionals, and concessives will not be considered anymore.
When we have an occasion to claim that e.g. all conditionals exhibit a prop
erty, we mean "all conditionals except for the speech act qualification uses".
A final preliminary point is that we will only discuss conditional, conces
sive conditional, and concessive assertions, and neglect the corresponding
questions and commands.
5 CONDITIONALS
Canonical conditionals in both German and Dutch have the V in final or

at least late position and are introduced by conjunctions, yet both languages
also have a non-canonical V-l pattern. It seems best to discuss them sepa
rately.
5.1 Canonical conditionals
5.1.1 Conditions allowing non-integration
For both German and Dutch canonical conditionals, resumption and

integration are unmarked patterns, while non-integration is marked. The lat
ter is possible only if the consequent is separately assertable, which simply
means it could occur as a separate assertion. There are several factors which
may lead to separate assertability in this sense:
i. the factual character of the conditional

ii.-iii. two subtypes of mismatch between the literally expressed con
sequent and the one that is meant
iv. the anaphoric repetition of the antecedent in the consequent
v. the (subjunctive) mood of the conditional
We will now discuss and illustrate these factors.
i. Non-integrative word order is permissible in what one could call 'fac
tual' conditionals, conditionals in which the states of affairs denoted by both
the antecedent and the consequent are not hypothetical, as is the case for
ordinary conditionals, but actual.
(24) D. Als ik als schilder siecht was, als metselaar
if I as painter bad was as mason
was ik echter een katastrofe I succes
was I however a catastrophe success
'If I was bad as a painter, as a mason however I was a
catastrophe/success'
Note that the clause describing the successful mason can easily get a conces
sive reading. This is the result of the factuality, inherent to factual condition
als, and the adversativity, due to the specific lexical content.
ii. The separate assertability of the consequent may also be due to the
fact that the antecedent does not directly relate to the consequent itself, but to
a proposition that is at least partially unexpressed, but reconstructable by the
hearer.
(25) G. Wenn ich die Bilanz der letzten Jahre zog,
if I the balance of the last years drew up
es blieb ein Plus
it remained a plus
'If I took stock of the last few years, the balance remained
positive'
In (25) it is of course the discovery of the fact that the balance was posi
tive rather than the fact itself that is the consequence of the speaker's calcula
tions. The claim that the balance was positive is clearly separately assertable.
The following is a interesting corpus example, discussed by Faucher
(1984:115).
(26) G. Wenn der Oberst nichts bekommen hat, die N

if the colonel nothing got has the N
Zeitung könnte es ja direkt aus Berlin haben
paper could it yes directly from Berlin have
Tf the colonel did not get anything, the N. paper could have
obtained it directly from Berlin'
The context makes clear that nichts 'nothing' refers to some news item, and
that the reconstructed consequent is something like 'that does not prove that
the news was false'. Of this reconstructed consequent the literal consequent is
not a part, as was the case in (25), but an explanation. But as in (25), the liter-
ally expressed consequent is separately assertable. Note that the explicans -
explicandum relationship in (26) is not a purely contextual inference; it is sig
naled by the particle ja.
iii. The following cases are almost the opposite of what we found in (25).
While the given consequent of (25) is only a part of the intended consequent,
the intended consequent in (27) is part of the one given.
(27) D. Als zij dat vroeger geweten hadden, ik geloof niet
if they it earlier known had I believe not
dat zij ook maar één woord ver anderd hadden
that they also only one word changed had
Tf they had known that before, I don't think that they would
have changed even one word'
It is not the speaker's conviction that is asserted to follow from somebody
else's prior knowledge. The sentence expresses the speaker's doubt that such
prior knowledge would have led to any changes in the text. Thus we may
assume that only zij hadden ook maar één woord veranderd is the intended
consequent, and that the whole conditional is the complement of the verb
geloven 'believe'. What (27) shares with (25) and (26), however, is that the
part that follows the antecedent is separately assertable.
iv. In examples like (28) and (29), the consequent contains an anaphoric
repetition of the antecedent. As a result, it is separately assertable, since
everything that is relevant for the specification of its truth conditions is given
in the consequent itself, even though it depends on the antecedent for the
interpretation of the anaphora.
(28) G. Wenn du mitkommen willst, ich habe nichts

if you come along want I have nothing
dagegen
there-against
'If you want to come along, I don't mind'
(29) D. Als hij het gedaan heeft, hij heeft het niet zo bedoeld
if he it done has he has it not so intended
'If he has done it, he hasn't meant it that way'
Note that the anaphoric reference can be indirect. In (30) the initial
expression of the second clause can also be assumed to repeat the antecedent.
(30) G. Wenn das mein Hund wäre, er bekäme keinen Zucker
if that my dog was he got no sugar
'If that was my dog, it would not get any sugar'
v. A conditional may have the non-integrative word order simply
because it is subjunctive.5
(31) D. Als ik in jouw plaats was, ik zou hem aanklagen
if I in your place were I would him sue
'If I were in your position, I would sue him'
A subjunctive consequent is separately assertable, not because there is no
truth conditional dependence on the antecedent, but because such depen
dence is conveyed by the subjunctive mood itself and does not, therefore,
have to be conveyed by the word order. Thus ik zou hem aanklagen is fine as
an independent assertion.
(32) D. I zou hem aanklagen
'I would sue him'
Unlike an indicative consequent detached from its antecedent, a subjunctive
consequent does not lose its hypothetical character. Compare (31)-(32) with
(33)-(34).
(33) D. Als ik hem terugzie, klaag ik hem aan
if I him see again sue I him
'If I see him again, I will sue him'
(34) D. Ik klaag hem aan
'I sue him' or 'I am sueing him'
ik klaag hem aan is hypothetical in (33), but factual in (34). ik zou hem aan
klagen, however, is hypothetical in both environments.
In each of the above non-integrative conditionals, integration or resump

tion would have been possible, too. Whether the separate assertability of the
consequent favors rather than merely allows non-integration we do not know.
We do not think that any large scale statistical study of either German or
Dutch has addressed this issue. It does seem clear, however, that non-integra
tion is infrequently called upon. Of the 365 German conditionals in a corpus
of Faucher (1984) only 5 (1.5%) had non-integration.
5.1.2 Conditions favoring resumption
Resumption is done with so or dann in German, and with dan in Dutch.

It is unmarked like integration, but not as frequent. Of the 471 German condi
tionals collected by Danielsson (1971:28,161), 118 (25%) exhibit resumption
and 353 integration. Recalculating the figures for the 4205 canonical Dutch
collected by Schelfaut (1982), one finds that only 9% are resumptive, while
91% are integrated.6
Interestingly, Schelfaut (1982) has discovered one factor that favors
resumption, and that is the case where an antecedent exhibits what was called
'anacoluthon' in Section 1 (example (9)). (35) is another example.
(35) D. Als ik in Parijs ben en ik zie de Seine, dan denk
if I in Paris am and I see the Seine then think
ik aan jou
I to you
'If I am in Paris and I see the Seine, then I think of you'
One would expect the second antecedent clause to have V-late order, just like
the first one, but it has V-2. The expected V- late order is possible and may
actually be more frequent.
(35)' D. Als ik in Parijs ben en ik de Seine zie, dan denk ik aan jou
Schelfaut has found 48 "als V-late en V-late" vs. 30 "als V-late en V-2" ante
cedents. Now, of the 30 "als V-late en V-2" cases, 21 (70%) exhibit resump
tion. That this preference for resumption does indeed correlate with V-2, and
not merely with the fact that the antecedent is biclausal or long is shown by
the fact that of the 48 "als V-late en V-late" cases only 4 (8%) exhibit resump
tion. As an explanation of this correlation we propose the following: V-2 is
the canonical word order for declarative main clauses, while V-late is the
canonical subordinate clause order. A clause that has V-2 and that is
nevertheless subordinate both lacks a canonical signal of subordination, and
furthermore exhibits a confusing signal. Hence there is a greater need for

another, clear sign of subordination. It is plausible to think that resumptive
dan is such a signal.
5.2 Non-canonical conditionals
Non-canonical, V-l conditionals allow the three patterns in German, but

only two in Dutch: non-integration and resumption.
(36) G. Wäre ich in Paris,
D. Was ik in Parijs,
was I in Paris
G. ich würde zum Louvre gehen
D. ik zou naar het Louvre gaan
I would to the Louvre go
G. (dann) würde ich zum Louvre gehen
D. dan zou ik naar het Louvre gaan
'If I were in Paris, I would go to the Louvre'
It is true that Dutch allows a structure that is superficially similar to the inte
gration of a V-l conditional, viz. the doubling of a V-l clause.
(37) D. Ben ik in Parijs, ga ik naar het Louvre
am I in Paris go I to the Louvre
'If I am in Paris, I go the Louvre'
What distinguishes a double V-l from an integrated V-l pattern is that the
former need not be conditional - (38) is purely temporal - and that it cannot be
counterfactual - cf. (36).
(38) D. Waren we net binnen, begon het daar te regenen
were we just inside began it there to rain
'We had just come in and it started to pour down'
This juxtaposition of two V-l clauses is still a marked structure; it is restricted
to colloquial, lively speech (Geerts et al., 1984:930). Both the linguistic
juxtaposition and its markedness are iconic: the double V-l pattern conveys
that the states of affairs referred to are similarly juxtaposed, and that this jux
taposition is somehow remarkable.
Of the two (Dutch) or three (German) patterns, the resumptive one is
clearly preferred. For Dutch this has been shown by Schelfaut (1982): of her
419 V-l conditionals 350 (81%) have resumption; for German, by Danielsson
(1971:161): of his 137 V-l conditionals 77 (56%) have resumption.

The distinction between canonical and non-canonical conditionals is thus
very clear: (i) integration is the favored construction in both German and
Dutch canonical conditionals, but in non-canonical conditionals integration is
either excluded (Dutch) or non-favored (German); (ii) non-integration is
marked for canonical conditionals, but unmarked for non-canonical ones; and
(iii) resumption is a non-favored order in canonical conditionals, yet it is
favored in non-canonical ones. Is this a purely arbitrary set of correspon
dences? Or do the facts 'make sense'? We think that the latter is the case, but
only if one accepts the diachronic hypothesis sketched in section 3. Given the
evolutionary schema that non-integration lead to integration via resumption,
the facts show that canonical conditionals have progressed further than non-
canonical ones. Canonical conditionals have fully realized the third stage, but
the non-canonical ones have not. In this view, we are also committed to the
claim that German has gone further than Dutch, a conclusion reached for the
word order possibilities of speech act qualifying antecedents, too (see section
4).
Table 1
canonical non-canonical
non-integration marked unmarked, not favored

resumption unmarked, not favored unmarked, favored
integration unmarked, favored impossible in Dutch
unmarked, not favored
in German
6 CONCESSIVE CONDITIONALS
There are three types of canonical concessive conditionals; they are all V-
late and they all start with a conjunction or WH word. As non-canonical we
consider V-l structures and an exclusively Dutch construction starting with al.
The exclusively German wenn/V1 + auch structure, which may also be
regarded as a non-canonical concessive conditional, will be dealt with in Sec
tion 7.2.
6.1 Canonical concessive conditionals
6.1.1 Conditions forcing, favoring, and allowing non- integration
Canonical concessive conditionals are either necessarily, preferably, or

possibly non-integrated, depending on the type (as described in Section 2) and
the language.
i. In both German and Dutch, non-integrative word order is required in
concessive conditionals whose antecedents are specified by a disjunction.
(39) G. (Ganz gleich) ob wir ihm helfen oder nicht,
no matter whether we him help or not
er schafft es nie
he makes it never
*(*so/*dann) schafft er es nie.
'Whether we help him or not, he will never make it'
In disjunctive concessive conditionals, consequents are necessarily separately
assertable. This is, no doubt, at least part of the explanation of the non-inte
gration requirement.
ii. When the antecedent is specified by a free-choice expression, non-
integration is required in Dutch, and preferred in German.
(40) G. Wieviel er auch verdient, er ist nie zufrieden
D. Hoeveel hij ook verdient, hi] is nooit tevreden
how much he also earns he is never content
'However much he earns, he is never content'
Just like in a disjunctive concessive conditional, the consequent of a free
choice concessive conditional is separately assertable. Why the preference for
non-integration is absolute for the disjunctive structure, but not for the free
choice one, may have something to do with the nature of the antecedent. In
disjunctive concessive conditionals even the antecedent is separately asserta
ble; this is not the case in sentences of type (40).
German, different from Dutch, allows so resumption and integration.
(40)' G. Wieviel er auch verdient, (sol*dann) ist er doch nie zufrieden
But so resumption is not as frequent as non-integration. The corpus of
Baschewa (1983:98) contains 81 free choice concessive conditionals starting
with a WH word, 68 (84%) of which are not integrated. We will say more
about the condition allowing resumption in Section 6.1.2. The only thing
known about the integrated pattern is its infrequency (Baschewa, 1980:218).

Note that we can again claim that German has been more progressive than
Dutch.
iii. In both German and Dutch, concessive conditionals in which a series
of antecedents is specified by a focus particle - selbst (G) I sogar (G) I zelfs
(D) 'even'; auch (G) / ook (D) 'also'; or auch nur / ook maar 'so much as' -
have integration as their unmarked pattern. Such concessive conditionals
('even if conditionals for short) also allow non-integration as a marked
option. Just as for normal conditionals, the relevant question for the choice of
word order is whether the consequent is separately assertable, and again there
are several factors that may lead to such separate assertability, notably:
a. the nature of the relationship between antecedent and consequent
(irrelevance vs. high dependence)
b. the choice of the focus of the particle (whole antecedent or a part
of it)
a. How the word order signals the nature of the relationship between
antecedent and consequent can be illustrated by the following two examples:
(41) D. Zelfs als wij de snelheidsgrens met vijftig
even if we the speed limit with fifty
km/h verlagen, het woud is niet meer te redden
km/h reduce the forest is not more to save
'Even if we lower the speek limit by 50 km/h, the forest cannot
be saved anymore'
(42) G. Wenn du auch nur ein Wort verrätst, bin ich
if you also only one word betray am I
verloren
lost
'If you give away so much as one word, I will be in great trouble'
Just like concessive conditionals of the disjunctive or free choice type (cf. (39)
and (40)), 'even if conditionals may characterize the antecedent as irrelevant
for the consequent. In such cases, the latter is totally independent of the
former in its truth conditions and is thus separately assertable ((41)). In con
trast to (41), (42) expresses a high degree of dependence between two situa
tions, since it holds even for the lowest on a scale of values. Note also that
only (42) invites the inference 'if not p, not q', usually called 'conditional per
fection' (cf. Geis and Zwicky, 1971), i.e. 'if you don't give away so much as
one word, I will not be in trouble'. What the exact syntactic or semantic prop
erties are that determine the relationship between antecedent and consequent
is not completely clear. What is clear, however, is that certain focus particles
like sogar or auch nur always signal dependence and thus invariably cooccur
with integrative word order.
b. The other factor that is relevant for word order is the choice of the
focus of the particle. Focus particles combine with a structured proposition,
consisting of a focus and a propositional schema that is satisfied by the focus
(cf. Jacobs, 1983; König, forthcoming). One of the contributions that these
focus particles make to the meaning of a sentence is the presupposition that
there are certain alternatives to the value of the focus, i.e. other, less likely,
and therefore more surprising values of the same kind which satisfy the prop
ositional schema in question. Thus (43) presupposes that other people than
the president will attend the meeting.
(43) a. Even the president will attend the meeting
b. even (λx [x will attend the meeting] , the president)
In concessive conditionals, focus particles may focus on a part of the anteced
ent, as in (41) and (42), where the focus lies on vijftig and ein Wort, or on the
entire antecedent, as in (44).
(44) Even if he is a little slow, Fred is actually quite intelligent
What is under consideration in cases like (44) as an alternative value to the
focus 'p' is the negative counterpart of the antecedent, i.e. 'not p'. In other
words, 'even if conditionals of type (44) function just like concessive condi
tionals which relate a disjunction 'p or not p' to a consequent (as in (39)). It is
not surprising therefore that the word order in such conditionals should be
non-integrative.
(45) D. Ook als Fred vaak iets langzaam is,
also if Fred often something slow is
hij is in de grond zeer intelligent
he is in the ground very intelligent
'Even if Fred is often a little slow, he is basically very intelligent'
It follows from what was said before that examples (44) and (45) do not invite
conditional perfection as an inference. This inference would be incompatible
with the contribution made by the focus particle, i.e. with the presupposition
that the alternative under consideration ('not p') satisfies the conditional
schema 'if x, q'.
Finally we should mention that the presence of the free choice expression
noch so (G) / nog zo (D) 'still so' in an antecedent introduced by auch / ook
or selbst / zelfs puts the conditional into the same class as those listed in (40).
Non-integration is the only (Dutch) or preferred (German) option.
(46) G. Selbst wenn er noch so schnell läuft, er kommt zu spät
even if he still so fast runs he comes too late
'However fast he runs, he will be late'
6.1.2 Conditions allowing resumption
Resumption is never a preferred option, but it is a possible pattern under

certain conditions. Whenever resumption occurs in such conditions, the
resumptive marker is invariably accompanied by a focus particle or a conces
sive conjunct.
i. In both German and Dutch 'even if conditionals, the usual conditional
resumptives dann / dan can be used. But since these may only resume a condi
tional antecedent rather than a combination of focus particle and antecedent,
the relevant particle has to be repeated or paraphrased in the consequent. In
German selbst, auch, and sogar are used in this function, in Dutch we find
zelfs, ook, and nog.
(47) D. Zelfs als wij de snelheidsgrens zeer sterk
even if we the speed limit very strongly
verlagen, zelfs dan is het woud verloren
reduce even then is the forest lost
'Even if we lower the speed limit very severely, even then the
forest is lost'
There is one particularity in Dutch: in combination with nog, dan is optional.
Note that dan follows a resumptive, whereas zelfs and ook precede it.
(48) D. Zelfs als hi] hard fietst,
even if he hard bikes
(dan) nog komt hij te laat
then still comes he too late
(dan) komt hij nog te laat
'Even if he bikes very fast, he will be late
ii. In German, resumption is also possible with the marker so. We have
already mentioned this possibility for concessive conditionals specified by a
free choice expression (cf. (40)'). The condition for so resumption seems to be
the presence of the conjunct doch or dennoch - or even jedoch - all meaning
'still'. Of the 13 resumptive free choice concessive conditionals found by
Baschewa (1983:98), there is only one without a conjunct.
So resumption is also possible in 'even if conditionals. Again, a conces
sive conjunct in the consequent is nearly obligatory. Of the 42 selbst wenn /
auch wenn cases found by Baschewa (1980:158, 163), only 7 (17%) have
resumption. Six of those contain the conjunct doch.
(49) G. Selbst wenn die Pause nicht aus funktionellen Gründen
even if the pause not out of functional grounds
entstanden sein mag, so bildet sie doch auch ...
originated be may so constitutes it still also
'Even if the pause may not have come about for functional
reasons, it nevertheless makes ...'
When we discussed speech act qualifying adverbial clauses and non-
canonical conditionals and found that German allows integration and Dutch
does not, we interpreted this greater permissiveness as showing that German
had gone further than Dutch in the progression from non-integration to inte
gration. It is not clear that the greater permissiveness of German with respect
to resumptive concessive conditionals warrants the same conclusion. All the
relevant cases of resumption concern so, and not dann. We know from
Baschewa (1980, 1983) that the use of so has been decreasing for at least two
centuries. Dutch originally had a resumptive so, too (see Jansen, 1980;
Schmidt, 1958), but the frequency of Dutch so has decreased to the point of
total disappearance. From this point of view, therefore, it is Dutch that is
more progressive.
6.2 Non-canonical concessive conditionals
Both German and Dutch have V-l concessive conditionals. In terms of

the typology developed in connection with (15), such sentences have to be
analyzed as 'free choice' concessive conditionals: free choice for some
parameter of the antecedent is offered to the hearer by means of an impera
tive or optative construction.
(50) D. Vertel wat je wil, ik doe het op mijn manier
tell what you want I do it on my way
'Say what you like, I'll do it my way'
Non-integration is the norm for such constructions, in both German and

Dutch.
There is another non-canonical concessive conditional in Dutch that
should briefly be mentioned, viz. the one introduced by al, ook al, or zelfs al,
all of which meaning 'even if.7 This construction is non-canonical, because
clause-initial al functions like a subordinating conjunction, on the one hand,
but is directly followed by the finite verb, on the other. Historically, clause-
initial al is best analyzed as a strengthening particle that was prefixed to a V-l
conditional (cf. Van Es, 1949, 1951; but also Schieb, 1952).
Al, optionally preceded by ook/zelfs, requires non-integration, unless the
presence of zelfs, ook, or nog in the consequent leads to resumption by dan
(cf. (47) and (48)).
(51) a. D. Al fietst hij hard, ( hij komt te laat)
even if bikes he hard he comes too late >
*dan komt hij te laat
'Even if he bikes fast, he will be too late'
b. D. Al fietst hij hard, zelfslook *(dan) komt hij te laat
'Even if he bikes fast, even then he will be too late'
c. ľ). Al fietst hij hard, (dan) nog komt hij te laat
* (dan) komt hij nog te laat
( hij komt nog te laat )
'Even if he bikes fast, even then he will be too late'
None of the structures discussed in this section exhibit integration and
thus provide further evidence for our hypothesis that canonical structures
have progressed further on the way towards integration than non-canonical
ones.
7 CONCESSIVES
Canonical concessives in both German and Dutch are V-late and intro
duced by conjunctions, primarily obwohl, obgleich, obschon, and wenngleich
for German, and hoewel and ofschoon for Dutch. Non-canonical German
concessives have antecedents with auch that are either V-l or V-late with
wenn. A Dutch non-canonical concessive is the one that starts with al.
7.1 Canonical concessives
In both German and Dutch, canonical concessives have integration as

their unmarked and preferred pattern. In Dutch, integration is the only
unmarked pattern, while German also allows resumption as a minor option.
Of the 154 obwohl concessives in Baschewa's (1980:140) corpus, 8 (5%) have
resumption. Similarly, Metrich (1978:14) has 246 obwohl concessives, and
only 6 (2%) have resumption. Both in German and in Dutch, non-integration
does occur, but as a marked and infrequent option: of the 246 obwohl conces
sives collected by Metrich (1978:14) only 8 (3%) exhibit non-integration.
Given these facts, separate assertability cannot be a relevant factor for the
choice of word order in concessive sentences. After all, concessives entail
both of the component clauses, which are thus always separately assertable.
For these structures, the relevant factor seems to be 'assertive emphasis'.
This cover term is meant to include such cases as referred to as 'marked
focus' or 'contrastive (emphatic) stress' in the literature (cf. Taglicht 1984).
Here we may distinguish between two cases, on the basis of both intonation
and meaning. In the first case, we have one contrastive focus, one emphati
cally stressed element. As in all cases of marked focus selection, this focussed
element is characterized as a member of a presuppositional set. Moreover, it
is characterized as ranking higher than the alternatives under consideration. If
'a' is the denotation of the focussed element, and 'b' one of the alternatives
under consideration, the contribution of this type of focusing to the meaning
of the relevant sentence can be glossed as 'not only b but a'. The following
example is a case in point:
(52) G. Wenn dieser Satz unter die Literaten käme,
if this sentence among the literary came
steinigen würden sie mich
stone would they me
'If this statement became known to the literary people, they
would stone me'
The second type of example to which the terms 'marked focus' or 'con
trastive stress' may be applied is characterized by a bifocal intonation pattern:
a rising nucleus followed by a falling nuclear tone. Another property of this
construction is that semantic scope is not indicated by the left-to-right
sequence of elements, as it normally is (Jacobs, 1983:105). The denotation of
the first focus is again opposed to other members of the same presuppositional
set, but this time the alternatives under consideration are excluded as possible
values for the relevant predication, i.e. 'not-b'.
(53) D. Hoewel ik Fred niet verdragen kan, haten doe ik hem
although I Fred not stand can hate do I him
ook niet
also not
'Although I can't stand Fred, I don't actually hate him'
Examples like (53) show that the alternative entities the contrastively stressed
element of the main clause is opposed to may be given in the initial clause.
Another characteristic property of such concessives is that they do not imply
any factual conflict. The meaning of concessive conjunctions in such examples
is more or less identical to that of the adversative conjunction but.
The following example, documented in Kaufmann (1974:5), shows that
assertive emphasis may also result from the use of adverbs like einfach 'sim-
ply'.
(54) G. Obwohl die Dinge für jedermann offen zutage lagen -
although the things for everyone open to day lay
er wollte es einfach nicht wahrhaben
he would it simply not true have
'Although everything was clear to everybody, he simply
wouldn't see the truth'
7.2 Non-canonical concessives
German V-l + auch concessives are non-canonical because of the V-l

feature. That wenn + auch concessives are non-canonical is less easy to
defend: after all, wenn + auch clauses have the verb in final position. There
is one feature that distinguishes wenn + auch from other (also basically com
plex) concessive conjunctions (obwohl, obgleich, obschon, and wenngleich):
the two components are typically discontinuous:
(55) G. a. Wenn auch Fritz kein Geld hat, ...
if also Fred no money has
b. Wenn Fritz auch kein Geld hat, ...
'Even though Fred doesn't have any money, ...'
Just like obwohl, etc. wenn/W-1 + auch allows each of the three patterns,
but it is unclear which pattern is the marked or preferred one. So let us first
look at a few basic facts about wennFV-1 + auch:
i. Especially in writing, many wenn/V-1 + auch antecedents can get a

conditional reading. Thus auch in (55a) can also also function as a focus parti
cle focusing on the subject, rather than as a component of a conjunction.
Obwohl, on the other hand, is unambiguously concessive.
ii. For at least some speakers of German, wenn/V-1 + auch can have a
concessive conditional meaning ('even if) (cf. Metrich, 1978:153-195).
iii. wenn + auch is much more prone to non-integration than obwohl.
Metrich (1978:144) found 91 wenn + auch concessives (concessive conditio
nals), and 27 (30%) had non-integration - compare this with the 2-3 % found
for non-integrated obwohl.
iv. wenn + auch is much more prone to resumption than obwohl. Of
Baschewa's (1980:140) 154 obwohl concessives only 8 (5%) exhibit resump
tion, but of her 80 wenn + auch concessives and concessive conditionals 44
(55%) have it (Baschewa, 1980:171).
v. Most consequents of the integrated and resumptive wenn + auch con
cessives (and concessive conditionals) contain concessive conjuncts like doch
'still' (Metrich, 1978:144). In the consequents of the non-integrated wenn +
auch clauses (Metrich, 1978:144) as well as in obwohl consequents (Baschewa,
1980:140) such conjuncts are rare.
These facts suggest that both the presence of a concessive conjunct in the
consequent and non-integration mark concessivity: while a wenn/V-1 + auch
antecedent as such can get a purely conditional reading, the concessive con
junct or the non-integration make clear that the intended meaning is conces
sive or, at least, concessive conditional. The reason why non-integration lends
itself to this purpose is clear. It marks the consequent as separately assertable,
and this, we have seen, is a typical characteristic of concessive conditionals
and concessive sentences. As to the factors determining the choice between
integration, resumption, and non-integration, it seems clear that assertive
emphasis favors non-integration.
(56) G. Wenn ich auch nicht alles verstanden habe, einiges
if I also not all understood have something
ist mir klarer geworden
is me clearer become
'Even though I didn't understand everything, a couple of things
have become clearer to me'
But assertive emphasis, which arguably accounts for the 2-3% occurrence of
non-integrated obwohl, cannot explain the 30% for non-integrated wenn +

auch. We hypothesize that the high percentage for non-integrated wenn +
auch is yet again an illustration of the fact that non-canonical structures are
less progressive than canonical ones. The high percentage for resumptive
wenn + auch, as compared to the low percentage for resumptive obwohl,
points in the same direction.
At this point, we should briefly return to clause-initial al in Dutch. Like
concessive-conditional al, concessive al can be strengthened by ook and it
allows non-integration. Different from concessive-conditional al, however, it
cannot be strengthened by zelfs, and it does not allow resumption.
(57) D. Al fietst hij hard, hij komt te laat
even though bikes he hard he comes too late
*komt hij te laat
*dan komt hij te laat
'Even though he bikes fast, he will be too late'
There is no semantic explanation of the preference of concessive al for
non-integration: it is only a matter of the greater archaicness of non-canonical
constructions.
8 CONCLUSIONS
The preceding discussion has shown that there are three ways of combi
ning a German or Dutch, conditional, concessive conditional, or concessive
clause with a following main clause: (i) mere juxtaposition, (ii) juxtaposition
and linking through a resumptive element introducing the main clause, and
(iii) complete integration indicated by the initial position of the finite verb in
the main clause, which thus takes second position in the overall complex
sentence. These three strategies are easily interpreted as positions on a
continuum from parataxis to embedding (cf. Lehmann, this volume). This
continuum also has a diachronic significance. German and Dutch show
how subordinate clauses have tended to develop from non-integration to
integration via resumption.
At first sight, the choice between the three word order patterns in
present-day Dutch or German seems to be totally random, and the relevant
factors are frequently treated as totally idiosyncratic facts. It was the main
purpose of this paper to show that the choice between the patterns is, at least to
a large extent, determined by very general principles. Some of these principles

are semantic or pragmatic ones:
i. Speech act qualifying adverbial clauses, being about rather than a part
of the main clause, are generally non-integrated.
ii. The separate assertability of a consequent of a conditional or conces
sive conditional allows, prefers, or requires non-integration.
iii. Assertive emphasis on a consequent of a concessive allows non-inte
gration.
iv. Scope: (iv.a) If the consequent of an 'even if conditional contains a
focus particle that focuses on the antecedent, resumption is possible or
required, (iv.b) Non-integrative word order is found where some part of the
apparent main clause takes wide scope over the whole complex sentence.
v. Iconicity: (v.a) Non-integrative word order occurs in complex sen
tences which assert or imply the independence and mutual irrelevance of two
propositions or their cooccurrence against a background of general incompati
bility, (v.b) The markedness and juxtaposition of V-1 clauses in Dutch reflects
the remarkableness of the cooccurrence of two facts.
Another very general principle is that not all Germanic languages and not
all construction types have been equally affected by the development from
non-integration to integration:
i. The successive clause integration process has affected Scandinavian
more than German, and German more than Dutch.
ii. One feature of this process, however, the disappearance of resump
tion, is more thoroughly felt in Dutch than in German.
iii. Non-canonical structures have participated less strongly in the
development from non-integration to integration via resumption.
On one occasion, we have seen a psycholinguistic principle at work: the
resumption preferred after an anacoluthon in conditional antecedents may
well function as a comprehension expedient.
Finally, our investigation throws some light on Foley and Van Valin's
(1984:269) 'semantic relations hierarchy', and their assumption that the
stronger the semantic relation, the closer the syntactic bonding. Non-integra
tive word order can be observed in conditionals, which are placed at the lower
(weakest) end of the scale, but not in connection with any of the other seman
tic relations distinguished by Foley and Van Valin. Furthermore, our findings
CLAUSE INTEGRATION IN G E R M A N A N D D U T C H 129
suggest that concessive conditionals, which do not appear in their list, should
be placed below conditionals, i.e. further to the pole denoting the weakest
semantic relation. Interestingly, one would perhaps expect that concessives,
which have also been left out of account, belong even further to the pole of
the weakest semantic relation, yet this expectation proves to be wrong.
NOTES
1. Thanks are due to Emilia Baschewa, Alain Bossuyt, Georges De Schutter, Ad Foolen, Sieg
fried Thyssen, Joop van der Horst, Jos Rombouts, and Marc Van de Velde as well as the
participants of the Albany workshop for helpful suggestions. We would also like to thank
Leiv Egil Breivik, Östen Dahl, Hartmut Haberland, and Klaus von Bremen for helping us
with the Scandinavian data. We are grateful to the Deutsche Forschungsgemeinschaft for
supporting E. König's trip to Albany, and to the Alexander von Humboldt-Stiftung for sup
porting J. van der Auwera during a research stay at the University of Hannover and at the
Max Planck Institut für Psycholinguistik, Nijmegen.
2. In recent generative descriptions of word order in German and Dutch, the V-lateness is no
longer regarded as being necessarily linked to the subordinate status of a clause. It is instead
seen as a consequence of the lexical filling of a complementizer node (see Den Besten, 1983;
Lenerz, 1984; esp. Haider, 1984:115). In order to account for the complementarity in the
presence of a complementizer and V-2, the finite verb is moved into COMP position if that
node is not lexically filled. Of course, the presence or absence of a complementizer can still
be linked with the main clause - subordinate clause dichotomy. Thus Reis (1985:273) con
vincingly argues that all canonical German main clauses are not introduced by a compie
mentizer, whereas all canonical subordinate clauses are so introduced.
3. When Dutch and German follow the same rule, we will illustrate it with an example from
one language only.
4. We will illustrate the relevant distinctions with examples from English, which has the advan
tage that order problems do not complicate the presentation. See König (1986) for a more
detailed discussion of these distinctions.
5. Whether or not the subjunctive is a sufficient condition for non-integrative word order is not
completely .clear. The examples given in Faucher (1977:30) suggest that emphasis and con
trastive focusing may also be relevant. The distinctions in meaning and force that are typi
cally expressed by non-integrated vs. integrated subjunctive conditionals can be illustrated
by contrasting (a) with its integrating counterpart:
a. G. Wenn ich an deiner Stelle wäre, ich würde ihn verklagen
if I at your place were I would him sue
b. G. Wenn ich an deiner Stelle wäre, würde ich ihn verklagen

The first example would typically be used as a piece of advice in a situation where the
addressee is contemplating sueing a person. The second example" by contrast suggests that
the speaker is contemplating sueing a person. Here the consequent is of course much less
emphatic or 'assertive' than the one in (a), and it seems quite plausible, therefore, to estab
lish a link between integration and the argumentative orientation expessed by the sub
sequent utterance.
6. If these figures are representative, then German favors resumption more than Dutch, which
again confirms the hypothesis that German has progressed further from non-integration to
integration. Interestingly, in the 365 canonical conditionals collected by Faucher (1984:115),
resumption even outnumbers integration (202 vs. 158). Faucher's conditionals are taken
from somewhat archaic, literary prose.
7. Clause-initial al is genuinely vague in meaning and may also be interpreted as 'even though'.
The latter is discussed in Section 7.2.
REFERENCES
Alving, Hj. 1916. Det grammatiska subjektets plats i den narrrativa satsen i
svenskan. Uppsala: Berlings.
Andersson, L.-G. 1975. Form and Function of Subordinate Clauses.
Göteborg: University, Dept of Linguistics.
-----. 1976. "Talaktsadverbial". Nysvenska studier 55-56:25-46.
Baschewa, E. 1980. Der Konzessivsatz im Neuhochdeutschen - synchronische
und diachronische Untersuchungen zu seiner Syntax, Semantik und Stilistik.
Ph.D., University of Leipzig.
1983. "Untersuchungen zur Diachronie des Konzessivsatzes im
Neuhochdeutschen". Beiträge zur Erforschung der deutschen Sprache 3:77-
107.
Behaghel, O. 1903. "Die Herstellung der syntaktische Ruhelage im
Deutschen". Indogermanische Forschungen 14:438-459.
. 1929. "Der Nachsatz". Beiträge zur Geschichte der deutschen Sprache
und Literatur [PBB] 53:401-418.
Bergqvist, B J . 1884. Studier öfver den konditionala satsfogningen i
fornsvenskan. Lund: Berlings.
Bolinger, D. 1977. Pronouns and Repeated Nouns. Indiana: Indiana Univer
sity Linguistics Club.
Bossuyt, A. 1980. "Woordvolgorde in het Middelnederlands: een funktionele
aanpak". In: Linguistics in the Netherlands 1980, S. Daalder and M. Ger-
ritsen (eds.), 93-101. Amsterdam: North-Holland.
1985. "The typology of embedded predications and the SOV/SVO shift
in Western Germanic". In: Syntax and Pragmatics in Functional Grammar,
A.M. Bolkestein et al. (eds.), 15-29. Dordrecht/Cinnaminson: Foris.
. Forthcoming. Historical Functional Grammar. Dordrecht/Cinnaminson:

Foris.
Brugman, C. et al. (eds.). 1984. Proceedings of the Tenth Annual Meeting of
the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society.
Danielsson, E. 1971. Zum Ausdruck der Konditionalität im Deutschen. Lic.
Thesis, University of Uppsala.
Davison, A. 1979. "Some mysteries of subordination". Studies in the Linguis
tic Sciences 9:105-129.
De Boor, H. 1922. Studien zur altschwedischen Syntax in den ältesten
Gesetztexten und Urkunden. Breslau: Marcus.
Debrabandere, I. "De SVf-woordorde in zinnen met aanloop". Handelingen
Koninklijke Commissie voor Toponymie en Dialectologie 50.87-97.
Den Besten, H. 1983. "On the interaction of root transformations and lex
ical deletive rules". In: On the Formal Syntax of the Westgermania,
W. Abraham, (ed.) 46-131. Amsterdam/Philadelphia: Benjamins.
Faucher, E. 1977. "Exceptions à la règle V2?". Cahiers d'allemand 11:27-47.
1984. Vordre pour la clôture. Nancy: Presses Universitaires.
Fleischmann, K. 1973. Verbstellung und Relief theorie. Ein Versuch zur Ge
schichte des deutschen Nebensatzes. München: Wilhelm Fink.
Foley, W.A. and R.D. Van Valin. 1984. Functional Syntax and Universal
Grammar. Cambridge: Cambridge University Press.
Geerts, G. et al., (eds.). 1984. Algemene Nederlandse Spraakkunst.
Groningen: Wolters-Noordhoff & Leuven: Wolters.
Geis, M. and A. Zwicky. 1971. "On invited inferences". Linguistic Inquiry
2:561-565.
Gerritsen, M. 1978. "De opkomst van SOV patronen in het Nederlands in
verband met woordvolgordeveranderingen in de Germaanse talen". In:
Aspecten van de woordvolgorde in het Nederlands, J.G. Kooij, (ed.), 3-39.
Leiden: Vakgroep Nederlandse Taal- & Letterkunde.
1984. "Divergent word order developments in Germanic languages: A
description and a tentative explanation". In: Historical Syntax, J. Fisiak
(ed.), 107-135. Berlin/New York/Amsterdam: Mouton.
Green, G.M. 1976. "Main clause phenomena in subordinate clauses". Lan
guage 52:382-397.
Haegeman, L. 1984. "Pragmatic conditionals in English". Folia Linguistica
18:485-502.
Haider, H. 1984. "Topic, focus & V-second". Groninger Arbeiten zur Ger
manistischen Linguistik 25:72-120.
Haiman, J. and S.A. Thompson. 1984. "'Subordination' in Universal Gram-

mar". In: C. Brugman, C. et al. 1984:510-523.
Halliday, M.A.K. 1985. An Introduction to Functional Grammar. London:
Edward Arnold.
Hammarström, E. 1923. Zur Stellung des Verbums in der deutschen Sprache.
Lund: Håkon Ohlssons Buckdruckerei.
Handke, J. 1984. Descriptive and Psycholinguistic Aspects of Adverbial Subor-
dinate Clauses. Heidelberg: Groos.
Hermodsson, L. 1978. Semantische Strukturen des Satzgefüge im kausalen
und konditionalen Bereich. Uppsala: Almqvist & Wiksell.
Hooper, J. and S.A. Thompson. 1973. "On the applicability of root transfor-
mations". Linguistic Inquiry 4:465-498.
Horacek, B. 1957. "Zur Verbindung von Vorder-und Nachsatz im
Deutschen". Beiträge zur Geschichte der deutschen Sprache [PBB] Sonder-
band 57:415-439.
Jacobs, J. 1983. Fokus und Skalen. Tübingen: Niemeyer.
Jansen, F. 1980. "Developments in the Dutch left-dislocation structures and
the verb-second constraint". Papers from the 4th International Conference
on Historical Linguistics, E.C. Traugott (ed.), 137-149. Amsterdam: John
Benjamins.
Johnson-Laird, P.N. 1986. "Conditionals and mental models". In: Traugott
1986: 55-75.
Kaufmann, G. 1974. "Das konjunktivische Bedingungsgefüge". Zielsprache
Deutsch 1:1-21.
Kolb, H. 1963. "Rückfall in die Parataxe". Neue Deutsche Hefte 96:42-74.
König, E. Forthcoming "Gradpartikeln". To appear in Handbuch der
Semantik, A. von Stechow and D. Wunderlich (eds.). Königstein:
Athenäum.
. 1986. "Conditionals, concessive conditionals and concessives: areas
of contrast, overlap and neutralization". In: Traugott 1986: 228-246.
Lenerz, J. 1984. Syntaktische Wandel und Grammatiktheorie. Eine Unter
suchung an Beispielen aus der Sprachgeschichte des Deutschen. Tübingen:
Niemeyer.
Longacre, Robert E. and Sandra A. Thompson. 1985. "Adverbial Clauses".
In: Shopen 1985: 171-234.
Metrich, R. 1978. La concession en allemand. (Les groupes subjonctionnels).
Thèse de 3e cycle, Paris 4 - Sorbonne.
Reis, M. 1985. "Satzeinleitende Strukturen im Deutschen: Über COMP,

Haupt- und Nebensätze, w-Bewegung und die Doppelkopfanalyse". In:
Erklärende Syntax des Deutschen, W. Abraham (ed.), 271-311. Tübingen:
Gunter Narr.
Schelfaut, F. 1982. Eerste Deel: De Nederlandse Conditionele Bijzin, ingeleid
door ALS, WANNEER of ZONDER VOEGWOORD, met of zonder
DAN in de hoofdzin. Tweede Deel: De Nederlandse Concessieve Bijzin,
ingeleid door AL, ALHOEWEL, HOEWEL, OESCHOON, met of zonder
TOCH in de hoofdzin. Mémoire, Université de Liège.
Schieb, G. 1952. "al, alein(e), an(e) im Konzessivsatz. Ein Beitrag zur Vel-
dekekritik". Beiträge zur Geschichte der deutschen Sprache und Literatur
[PBB] 74.268-285.
Schmidt, C. 1958. De concessieve voegwoordelijke bijzin in het Nederlands van
de middeleeuwen en de zeventiende eeuw. Ph. D., University of Groningen.
Shopen, Timothy (ed.) 1985. Language Typology and Syntactic Description II.
Cambridge: Cambridge University Press.
Taglicht, J. 1984. Message and Emphasis: on Focus and Scope in English.
London: Longman.
Traugott, E.C. et al. (eds.). 1986. On Conditionals. Cambridge: Cambridge
University Press.
Van der Horst, J.M. 1981a. Kleine Middelnederlandse Syntaxis. Amsterdam:
Huis aan de Drie Grachten.
. 1981b. "Onderschikking en de plaats van de persoonsvorm in het mid-
delnederlands". Tijdschrift voor Nederlandse Taal- en Letterkunde 97:161-
184.
. 1984. "Over vorm en inhoud van bijzinnen". In: Vorm en Funktie in Tekst
en Taal, 154-179. Leiden: Vakgroep Nederlandse Taal- & Letterkunde.
Van de Velde, M. 1979a. "Zur mehrfachen Vorfeldbesetzung im Deutschen".
In: Wortstellung und Bedeutung. Akten des 12. Linguistischen Kol
loquiums, M.-E. Conte et al. (eds.), 131-141. Tübingen: Niemeyer.
. 1979b. Zum Vorfeld im heutigen Deutsch. Ph.D., Rijksuniversiteit Gent.
Van Es, G.A. 1949. "Oorsprong en functies van het voegwoord al". It beaken
12:105-117.
. 1951. "Syntaktische vormen van de concessieve modaliteit in het Neder
lands", Tijdschrift voor Nederlandse Taal-en Letterkunde 68:253-295.
Van Valin, R.D. 1984. "A typology of syntactic relations in clause linkage".
C. Brugman et al. 1984:542-558.
Presentational cleft constructions
in spoken French
Knud Lambrecht
University of Texas, Austin
1 Introduction1
In previous work on the grammar of spoken French (Lambrecht, 1984a

and 1987) I have tried to establish two fundamental facts concerning the re
lationship between syntax and discourse in that language. The first fact is
that the 'canonical' transitive clause of the SVO type, containing lexical NPs
in subject and in object position, cannot be considered the basic clause type
of spoken French at the level of discourse, largely because it hardly ever oc
curs in actual speech. The only clauses containing lexical subject NPs occur
ring with some regularity are SV clauses, i.e. clauses with no object NP. The
occurrence of these SV clauses is furthermore restricted to a relatively well-
defined discourse function having to do with low topicality of the subject NP
referent in the discourse. Instead of the canonical SV(O) clause type, there
appears with overwhelming freauency a clause type which, following Du-
Bois' (1984) work on Sacapultec, I call the preferred clause of spoken
French, and which is of the form V(X). In this preferred clause structure the
subject is an incorporated pronoun 2 which is also the unmarked topic of the
clause. Full lexical NPs follow the verb and are semantically non-agentive
and pragmatically non-topical. Furthermore, normally not more than one
lexical NP constituent per clause occurs as a primary grammatical relation.
It follows from this preferred clause typology that in prototypical transitive
clauses the agentive subject will be an (incorporated) pronoun. There is thus
systematic absence, in spoken French, of sentences such as Sapir's famous
The farmer kills the duckling (Sapir, 1921:ch.V). From the point of view of
discourse, the preferred V(X) clause is the basic clause type.
136 KNUD LAMBRECHT
The second important fact about the grammar of spoken French —

which correlates directly with the first — is that the language provides its
speakers with systematic means of preserving this preferred clause type, in
the form of ready-made grammatical constructions whose main function is
to allow lexical NPs to occur elsewhere than in initial subject position. Since
in French, as in many languages, clause-initial position is the preferred posi
tion for topic, agent, and subject, these constructions thus represent gram-
maticalized means of preserving the preferred mapping of topic, agent, sub
ject and pronoun. These preferred-clause-structure-preserving constructions
are of two major structural types: so-called 'dislocations' and so-called
'clefts'. I claim that these two major structural types, which are widely at
tested among the world's languages, are expressions of a universal cognitive
constraint on the processing of referential information in spoken language
(cf. Lambrecht, 1986 b., and Section 3.2 below). The two types have in com
mon that they allow some lexical constituent (typically a noun phrase) to ap
pear elsewhere than in the clause in which the referent of the constituent is
a semantic argument. Put differently, these structural types allow speakers
to separate the referring function of NPs from their semantic-relational
function as arguments in some propositional piece of information. The lexi
cal constituent, instead of being part of the relational network of the propo
sition or clause, appears either in a non-relational position altogether, as in
the case of dislocation, or in a special clause of extremely low transitivity,
as in the case of clefting. What distinguishes the two structural types is the
discourse status of the referent of the NP: in the case of dislocation, this NP
referent is already topical, whereas in the case of clefts it is non-topical or
not-yet-topical.
In this paper I would like to take a close look at one type of cleft
construction frequently used in spoken French, which for lack of a better
term I will call the presentational cleft or avoir-cleft construction. This
construction seems to have gone generally unnoticed in contemporary
French linguistics3, in spite of its strikingly frequent occurrence in spontane
ous speech and in spite of the interesting syntactic, semantic and pragmatic
problems it poses for the theory of grammar. Sentences (1) through (3) are
typical examples:4
(1) Y'a Jean qu'a téléphoné
there-has Jean that has called
'Jean called'
PRESENTATIONAL CLEFT CONSTRUCTIONS IN SPOKEN FRENCH 137
(2) Y'a le téléphone qui sonne!

there-has the phone that rings
'The phone's ringing!'
(3) J'ai les yeux qui m'font mai
I-have the eyes that I.SG.DAT-do bad
'My eyes hurt.'
The corresponding French utterances of the form SV(O) {Jean a téléphoné,
Le téléphone sonne, Les yeux m'font mal) are pragmatically unacceptable
in most discourse contexts (cf. Lambrecht, 1984a and 1987). Discourse unac-
ceptability of corresponding SV(O) sentences will be assumed for all
examples of avoir-constructions discussed in this paper. Examples (1)
through (3) are characterized by the fact that some discourse referent is co
ded first as a lexical NP in a short clause (hereafter the avoir-clause) contai
ning a form of the verb avoir and then as a relative pronoun in another
clause (hereafter the qui-clause). While the avoir-clause poses the presence
of the NP referent in the discourse, the qui-clause expresses the semantic role
of this referent as a participant involved in some action, state or process.
Both the avoir-clause (the presentational clause proper) and the following
qu(i)-clause are of the preferred form [pro-V NP]. 5
The sequence y'a in examples (1) and (2) consists of the locative adver
bial y 'there', and the 3p singular form of avoir. The phrase y'a NP cor
responds to standard French il y a NP 'there is NP', in which y is preceded
by the dummy subject pronoun il. The history of the expression shows that
the subjectless form y'a is not a popular innovation of modern spoken
French, as might be thought, but a form that has remained essentially un
changed since the beginnings of the French language, the subject pronoun
il (normally / i / before consonants and glides) appearing, disappearing and
reappearing over the centuries.6 Spoken French y'a is thus the mirror image
of Spanish hay (historically ha+y), both going back to a combination of
Vulgar Latin habet 'he/it has' with the distal deictic adverb ibi. Notice that,
unlike English there is, French (il) y'a can only be used 'existentially', the
corresponding deictic presentational construction being voilà NP, which I
will not discuss in this paper. The verb avoir tends to occur frozen in its pre
sent tense form, but tenses other than the present do occur (cf. Section 3.4.)
The label 'presentational cleft' is meant to capture two important
features of this construction. The first is that some piece of information
which could be expressed as a monoclausal proposition is 'clefted' and coded
138 KNUD LAMBRECHT
as a sequence of two clauses. The second feature is that this construction is

typically used when some referent which is not yet pragmatically available
as a topic in the current discourse register is to be introduced or 'presented'
as a participant in some action, event or state. However the avoir-cleft con
struction is not restricted to the presentational function proper, i.e. it is not
used only to introduce new discourse referents in order to make them
available for reference in subsequent discourse. It has another pragmatic
function, exemplified in (2) and (3), which I will refer to as 'event-reporting',
and which consists in presenting some non-topical referent as an element in
some unexpected or surprising piece of information. Such event-reporting
sentences are particularly interesting because they differ strikingly from
presentational constructions involving existential sentences in English and
other languages. They require most clearly some analysis in which the rela
tionship between the two clauses is shown to be different from the relation
ship traditionally assumed to hold between a main and a subordinate clause.
2 Typological considerations
The analysis presented in this paper is based on the assumption that

there is a relationship between the form of a sentence and its function in
discourse, and that grammatical form is in part determined by the pragmatic
circumstances under which the sentence is used as a unit of information. In
particular, I believe that syntactic constructions are often motivated by
various pragmatic requirements having to do with the processing of informa
tion in discourse.
Consider the following utterance and the associated pragmatic situa
tion. At a bus stop, the departure of a crammed bus is delayed because a
woman loaded down with shopping bags is boarding very slowly. The
woman utters the following sentence, with an apologetic smile:
(4) My car broke down. (K.L.)
Discourse-pragmatically, the utterance (4) is of the event-reporting type. The
constituent in boldface represents what I will call the focus of the informa
tion expressed by the utterance, coinciding here with the main participant in
the reported event. The focus carries the main sentence accent. It is
characterized by pitch prominence and, in this example, by low tone on all
following elements in the clause.7 The utterance in (4), an example of actual
ly observed speech, could have the following natural equivalents in Italian
and in French, given the same speech situation:

(5) Mi si è rotta la macchina.
to-me itself is broken the car
(6) J'ai ma voiture qui est en panne.
I have my car that is in breakdown
The communicative intention and situation being identical in the three
utterances, we can say that the three sentences have (approximately) the
same 'cognitive meaning' in that they all express the same state of affairs in
a given world. They presuppose the fact that the speaker has a car (a
pragmatic, culture-dependent sort of presupposition made possible by the
real world situation) and assert that this car is presently not in working con
dition. The semantic structure of the three sentences is similar: they contain
a proposition composed of a one-place predicate indicating mechanical
malfunction and of one argument designating the thing that is not function
ing, i.e. the car. Let us call the car semantically the 'theme'. The sentences
also have an utterance-specific conversational meaning. We can fully
understand the woman's utterance only if we understand the relevance of
this utterance with respect to the situation in the bus. I will ignore this
conversational-pragmatic aspect of the utterance for the purpose of my
argument.
How is this semantic structure related to syntactic structure and to in
formation structure in the three languages? Let us take for granted that the
basic word order in all three languages is Subject-Verb-Object. Let us fur
ther take for granted a basic information packaging principle in these
languages according to which the topic, whether pronominal or nominal, is
normally the first element, coinciding with the subject, and according to
which the focus is normally the last element in the clause. We then have to
account for the following differences among the three sentences.
In the English sentence My car broke down, the semantic role of theme
is mapped onto the syntactic relation of subject. The subject is the initial NP
in an intransitive sentence, resulting in a structure of the form NP-V. This
syntactic sequence is not directly motivated by the pragmatics of the ut
terance. Rather it is an independently motivated syntactic structure in the
language. That it must have some independent structural motivation follows
from the fact that the same sentence could be used, under different
pragmatic conditions and with a different intonation contour, to convey a
pragmatically different piece of information, in which the car is an already
140 KNUD LAMBRECHT
established topic (as when I ask What happened to your car? and you
answer, with perhaps somewhat unnatural explicitness, My car broke down).
Thus in (4), both the semantic role of theme and the pragmatic role of focus
are mapped onto the syntactic role of subject NP, and this subject NP oc
cupies normal preverbal position. Instead of being coded syntactically, the
information structure of the utterance is coded intonationally: the subject is
marked as non-topical (as a focus) via primary sentence stress and following
low tone. Symbolizing the focus with the letter Z and the non-focal part of
the sentence with the letter A, we can schematically represent the informa
tion structure of the English sentence as Z-A.
Let us now consider the Italian sentence Misi è rotta la macchina in (5).8
Ignoring the different coding of the possessive relation in the two languages,
we notice that in (5) the mapping of semantics and syntax is essentially the
same as in (4): the theme is expressed as the subject NP of an intransitive
sentence.9 However the mapping of information structure and syntax is quite
different. In Italian the normal topic-initial SV structure of the sentence is
adjusted to fit the pragmatic requirements of the utterance by inverting the
order of the two major constituents. We can see that some syntactic adjust
ment has taken place by comparing this utterance with a corresponding ut
terance in which the car is not the focus of a reported event but the topic
of a statement intended to convey information about this car (as in the
English sentence quoted in parentheses above). Such an utterance would
have to be of the form La mia macchina si è rotta, with the topical subject
NP occupying initial position. Thus, even though the Italian example
resembles the English one in that the theme, the subject and the focus are
all three combined under the same constituent, the syntactic structure is dif
ferent. It is not the syntactically unmarked sequence, but the pragmatically
unmarked sequence that is maintained, with the focus as the final element
in the clause. The information structure of the Italian sentence is thus A-Z.
A rather complex situation obtains in the French example (6). In the
sentence J'ai ma voiture qui est en panne, both the syntax and the semantics
are adjusted in order to fit the pragmatic function of the utterance. Due to
a powerful constraint against the comapping of the pragmatic relation focus
and the grammatical relation subject (cf. Lambrecht, 1984a), spoken French
uses the clefted construction in (6), thus avoiding the focus-initial SV
sentence Ma voiture est en panne, which would be pragmatically unaccep
table, though syntactically well-formed. In (6) the focus of the utterance,
voiture, does not appear as the subject NP of an intransitive clause, as in
English and Italian, but in a separate clause, as the syntactic object of the
verb avoir. The clefting process creates an additional postverbal slot in
which the focus NP may appear, preventing it from occurring in initial sub
ject position. (6) thus makes up for the absence, in French, of focus-final
VS sentences of the Italian type: verb-initial sentences like Est en panne ma
voiture, which would have the focus in the right place, are ungrammatical
in French. 10
In the thus created bi-clausal structure, the first clause (J'ai ma voiture),
even though it has the form of an independent proposition, does in fact not
function to assert that the speaker 'has her car'. Rather the function of the
avoir-clause is to pragmatically pose the NP referent in the discourse in such
a way that the lexical NP will not appear as a subject. The semantic and syn
tactic relation of the NP to the main predicate est en panne is then expressed
in the qui-clause, whose pronominal subject is anaphorically linked to the
object in the preceding clause. This qui-clause, even though it has the inter
nal structure of a relative clause, differs in important ways from a restrictive
relative. For example the fact that the 'head' NP in the presentational cleft
can be a proper name, i.e. have a unique referent, excludes the modifying
function associated with the restrictive relative clause. Moreover the infor
mation expressed in the qui-clause of the presentational cleft is not
(pragmatically) presupposed as in the restrictive relative. In fact it is the
predicate of the qui-clause, not of the avoir-clause, that codes the main
assertion expressed by the sentence.11
Thus in the French sentence both the semantic and the syntactic struc
ture of the utterance are accommodated to an independently motivated in
formation structure, at the price of complex semantic and syntactic ad
justments. Ignoring the problem of a possible secondary focus on the
sentence-final element panne, which does not arise in English and Italian, we
can represent the information structure of the French sentence as A-Z-A, i.e.
as a grammatical compromise between the English structure Z-A and the
Italian structure A-Z.
The structural facts emerging from our three examples are more or less
independent formal phenomena. They are manifestations of the various
grammatical forces which interact and compete with each other in determi
ning the actual shape of the sentence. As the examples demonstrate, the
competition among these forces leads to quite different results, even in such
closely related and typologically similar languages as English, French and
Italian. In the case of English, information structure 'loses out' on the syn-
142 KNUD LAMBRECHT
tactic level. However this loss is compensated for by the fact that in English
the sentence accent can move freely from right to left, allowing for focus
marking in any position in the sentence. Following Bolinger (1984), we may
consider intonation an autonomous structural component of grammar,
whose predominance in English would make any syntactic adjustment un
necessary in this language, or would at least make up for the rigid word
order constraints of English grammar. Typologically, English presents itself
as an example of extreme 'subject prominence' (Li and Thompson, 1976),
i.e. as a language in which a great variety of semantic and pragmatic func
tions can be mapped onto the invariant syntactic function of subject and in
which word order is to a large extent grammatically and not pragmatically
controlled.12
The competition of grammatical forces has different consequences in
Italian. Here, it is syntax that 'loses out' in the competition between formal
structure and information structure. The basic order of constituents is
altered to accomodate the requirements of discourse. We may say that the
formal structure (5) is directly motivated by the pragmatic function of the
utterance. Just as English is reluctant to tolerate a violation of its basic SV
order, Italian is reluctant to tolerate violation of the information structure
constraint that places the focus of an event-reporting utterance in post-
verbal position. 13 In Italian, word order is thus to a greater extent
pragmatically controlled than in English, even though the syntax of Italian
is far more rigid in this respect than the syntax of so-called free word order
languages, like Russian or Latin.
As for French, syntax and information structure both win and lose in
the competition. The constituent order in the French sentence being strongly
grammatically controlled, the language does not tolerate subject-verb inver
sion or other types of word order variation found in languages with
pragmatically controlled word order. Nevertheless the global structure of
sentence (6) directly reflects its pragmatic function. Even more than in
Italian, the structure of the French sentence is pragmatically motivated. The
presentational cleft has as its unique function the expression of a certain
pragmatic utterance type. As the discussion in section 3.4. will show, certain
formal properties of the construction can be made sense of only by making
reference to its pragmatic function. For example the use of the verb avoir
and its cooccurrence with a possessive NP, as e.g. in (6), is a feature so in
timately associated with the construction that it 'makes no sense' from the
point of view of compositional sentence-level semantics.
By using grammatical constructions of the clefting type, spoken French

achieves several things at once. It substitutes structures of the preferred
[V(X)] type for the pragmatically unacceptable SV(O) sequence; it preserves
its syntactically controlled basic word order without violating the informa
tion structure constraint on focus placement; and it avoids violation of its
basic sentence accent pattern with the main accent falling on the clause-final
constituent. The 'mixed strategy' of cleft formation allows the language to
have its cake and eat it too. It represents one of the specific solutions in
French to the conflict between syntax and pragmatics.
I would like to conclude these typological considerations with a general
remark concerning the role of discourse-pragmatics in grammatical analysis.
If we view the relationship between morphology, syntax, semantics and
pragmatics as one of 'competing motivations', as suggested in the above
discussion,14 i.e. if we view the structure of the sentence as the result of a
complex interplay of various grammatical forces, which compete with each
other for the limited coding possibilities afforded by the grammar of the
language, it would be misleading to say that discourse pragmatics can ex
plain the formal structure of the sentence. For example, discourse-pragmatic
analysis cannot explain why, in spite of the fact that clause-final position of
the focus is a basic feature of Italian, English, and French, only Italian can
use the VS inversion construction to express the propositional content of (4),
while the analogous English and French VS sentences *Broke down my car
and *Est en panne ma voiture are ungrammatical. Nor can it explain the for
mal mechanism of the clefting construction in French. But discourse
pragmatics can show the functional motivation for the various structures in
these languages. It can establish the need for formally unambiguous coding
of the topic and the focus relation as a general functional principle which
can be partly overridden by competing grammatical forces but which never
theless is an ideal towards which languages tend by different grammatical
means. Discourse pragmatics can then prepare the ground for a better
understanding of the various language-specific ways in which this functional
ideal is pursued, taking into account the various structural properties of the
different languages.
144 KNUD LAMBRECHT
3 Pragmatic properties of the avoir-cleft
3.1 Concepts and definitions
In order to describe the pragmatic appropriateness conditions for the

use of the avoir-cleft in discourse, I must introduce some discourse-pragma-
tic terminology. I will, in particular, need clear labels for the various
cognitive statuses which the referents of linguistic expressions may have in
the minds of the speech participants at any given point in a discourse. These
cognitive statuses are of two major types: (i) activation, which has to do with
the rapidly changing temporary statuses referents can be assumed to have in
the hearer's consciousness; and (ii) identifiability', which has to do with the
question of whether a particular referent is already stored in the addressee's
long-term memory or not. 15
With respect to activation, I essentially follow Chafe's (1987) defini
tions and terminology. According to Chafe, a referent can be in any one of
three activation states, which he calls '(discourse-)active', 'semi-active' (or
'accessible'), and 'inactive' respectively. An active referent is one 'that is
currently lit up, a concept in a person's focus of consciousness at a particular
moment'. An accessible (semi-active) referent is one 'that is in a person's
peripheral consciousness, a concept of which a person has a background
awareness, but one that is not being directly focused on'. An inactive
referent is one 'that is currently in a person's long-term memory, neither
focally nor peripherally active'. 16
The principal cognitive criterion for Chafe's distinction is whether a
referent has been, directly or indirectly, activated in the consciousness or
short-term memory of the addressee. The activation parameter therefore ap-
plies only to referents which in one way or another are capable of being
recalled from memory or otherwise retrieved from the universe of discourse.
Activation thus does not account for the difference between a
recallable/retrievable referent and one that is unknown to the hearer and
therefore pot recallable by him or her. To account for this difference we
need the concept of identifiability. I label as identifiable any referent that
can be retrieved from the context or recalled from memory in the above men-
tioned way, and as unidentifiable any referent for which a new referential
entry or model has to be created in the mind of the hearer before it can be
recalled in subsequent discourse. Following Chafe (1976), we may call a
referent identifiable if the speaker assumes that the hearer is able to pick out,
from the set of referents describable by some linguistic expression, the one
intended by the speaker in a particular utterance. An unidentifiable referent
is then one which the speaker assumes the hearer is not yet able to pick out
in this way. One important linguistic correlate of the cognitive status of iden-
tifiability is the grammatical category definiteness as manifested e.g. in the
use of the definite article in English and other languages. Note however that
there is no one-to-one correspondance between identifiability or non-identi-
fiability of a referent and the definiteness or indefiniteness of the noun
phrase used to code this referent. This is shown for example in the fact that
definiteness markers are not used homogeneously across languages. For ex
ample proper names (whose referents are always identifiable) carry the
definite article in Greek but not English; plural NPs with generic referents
are coded with the definite article in French but not in English or German,
etc.
Even though activation and identifiability are independent cognitive
parameters, they are related in certain predictable ways. A referent which is
assumed by the speaker to be unidentifiable by a hearer is necessarily also
assumed to be inactive in the hearer's consciousness. The reverse however
is not true: an already identifiable referent cannot necessarily be assumed to
be already active in the hearer's consciousness. Rather an identifiable
referent can be in any of the three activations states: 'active', accessible',
or 'inactive'. In the latter two cases, the identifiable referent is typically, but
not necessarily, coded as a definite lexical noun phrase rather than as a pro
noun. When a referent is assumed to be active in the hearer's mind, it is of
course also necessarily assumed to be identifiable. Active referents are
typically, but again not necessarily, coded as unstressed pronouns, whether
bound or free, phonologically overt or null.
A referent can be accessible due to roughly three factors. First a referent
may be accessible because it has been "deactivated" (Chafe) from some
earlier active state in the discourse, in which case I will call it textually ac
cessible; secondly it may be accessible because it is saliently present in the
extralinguistic context, in which case I will call it situationally accessible; and
finally it may be accessible because it can be inferred from some other active
or accessible element in the universe of the discourse, in which case I will call
it inferrable.
The connections between the two parameters of activation and iden-
tifiablity are represented in the following diagram: 17
146 KNUD LAMBRECHT
(7) unidentifiable
inactive
identifiable accessible
active
Among the six logically possible combinations of the two parameters, only
the four indicated by the lines in the diagram are linguistically relevant. The
four paths in the diagram connecting the identifiability parameter with the
activation parameter will be labelled (i) brand-new ( = unidentifiable and in
active), (ii) unused ( = identifiable and inactive), (iii), textually accessible or
situationally accessible or inferrable ( = identifiable and accessible), and (iv)
active ( = identifiable and active).18
3.2 The presentational construction: a cognitive hypothesis.
I would like to preface my pragmatic analysis of the French presenta

tional cleft construction with a summary of a hypothesis I develop elsewhere
(Lambrecht 1986 b) concerning the relationship between the cognitive sta
tuses of referents as just described and the pragmatic relation of topic, which
I define as the relation of aboutness holding between a referent and a pro
position in a particular context (cf. Reinhart, 1982). This hypothesis ac
counts for, among other phenomena, the existence and form of complex
presentational constructions. I assume that human language is subject to a
general cognitive constraint imposed on the simultaneous performance of
two independent tasks: (i) the task of determining the intended referent of
a topic expression and (ii) the task of processing propositional information
about this referent.19 This constraint can be expressed in the form of a simple
discourse-pragmatic maxim: 'Do not introduce a referent and talk about it
in the same clause.' On the basis of this constraint I argue (against Reinhart,
1982) that it is an inherent property of any topic expression that its referent
must be referentially accessible. As a corollary, I argue that the interpreta
tion of sentences with insufficiently accessible topic referents poses certain
difficulties of interpretation, which can lead to unacceptability and, in ex
treme cases, to ungrammaticality on the sentence level.
The difficulties arising in the interpretation of sentences with insuffi
ciently accessible topic referents can be accounted for by postulating a
general correlation between the activation/identifiability states of topic

referents and the pragmatic acceptability of sentences. This correlation can
be expressed in the form of a scale of topic acceptability. Allowing for a cer
tain amount of variability across languages (due to the competing motiva
tions factor discussed in Section 2), the degree of pragmatic well-formedness
of a sentence containing a topic expression can be measured by the position
of the topic referent on the following scale:
(8) The topic acceptability scale
most acceptable
active
semi-active (accessible/inferrable)
inactive-identifiable (unused)
unidentifiable (brand-new) least acceptable

This scale is to be interpreted in the following way. The most easily pro
cessed and, therefore, the most acceptable sentences cognitively speaking are
those whose topics are highest on the scale, i.e. whose topic referents are ac
tive in the discourse. Such referents are the preferred topics because the
cognitive effort necessary to understand a sentence containing such a topic
is not increased by the additional task of assessing the topic referent (e.g.
by retrieving it from long-term memory or by drawing inferences leading to
its assessment).20 Since activeness is typically coded by unstressed pronouns
(including phonologically null elements) this entails that the preferred topics
are unstressed anaphoric or deictic pronominals. Unstressed pronouns may
thus be called the unmarked topics. It is the preference for discourse-active
topic referents that explains e.g. the form of the preferred clause in spoken
French with its initial unstressed (incorporated) pronoun. Less easily inter
pretable but still acceptable, and indeed frequently occurrring, topics are
those with accessible (semi-active) referents. In the case of accessible topic
referents, the cognitive task of interpreting the proposition which adds infor
mation about the topic must be performed simultaneously with another,
'low cost', processing effort, which is the effort necessary to remember, in
fer, or otherwise determine the referent of the topic expression.
A borderline case of pragmatic acceptability arises when new informa
tion is expressed about an unused (i.e. identifiable yet inactive) topic
referent. The acceptability of sentences containing topic expressions with
148 KNUD LAMBRECHT
unused referents varies widely with the language, the type of discourse and
the speech situation. The cognitive effort required in the interpretation of
such sentences is relatively 'high cost' because, in addition to processing pro-
positional information about some topic, the interpreter must infer the
referent of the topic itself, which was not previously made available in the
discourse. Clearly unacceptable as topics are brand-new referents, i.e.
referents which are unidentifiable for the hearer at the time the new informa
tion is conveyed about them. This type of unacceptability has a straightfor
ward cognitive explanation: if a hearer cannot mentally identify the referent
of a topic, he or she cannot determine whether the predicate holds for ('is
true o f ) this referent or not. The hearer then cannot make sense of the piece
of propositional information he or she is presented with. Sentences contain
ing such topics, are in a sense incomplete pieces of information. 21
A well-known formal correlate of the kind of unacceptability arising in
sentences with brand-new topic referents is the constraint found in many
languages against indefinite noun phrases in initial subject position. Given
that brand-new topic referents are lowest on the acceptability scale, the need
to avoid sentences with such topics is greatest. This is no doubt the reason
why across languages the common type of presentational clause is one in
volving a brand-new discourse referent. In some languages, certain presenta
tional structures, involving a limited set of intransitive verbs ('be', 'appear',
'die' etc), are used exclusively or with strong preference for the introduction
of such referents (cf. Chinese 'inverted word order', English existential
there-clauses etc.). A prototypical presentational structure, with parallels in
many languages, is the well-known fairy tale starter Once upon a time there
was..., where the noun phrase following the copula must be indefinite. In
other languages, presentational clauses may contain NP whose referents are
unused or even accessible (cf. e.g. the Romance presentational inversion
construction, which tolerates definite NPs).
Often such presentational clauses are immediately followed by a se
cond, syntactically dependent, clause in which the 'presented' referent,
which has become activated by the very utterance of the NP, reappears in
pronominal form, i.e. as an unmarked topic about which information can
now be conveyed in the cognitively preferred form. A prototypical example
of this is again our fairy tale starter, this time followed by a relative clause:
Once upon a time there was an old king who lived in a beautiful castle. I
will refer to this two-clause structure whereby a referent is introduced via a
presentational clause and immediately followed by another proposition as
the presentational construction. From the point of view of discourse, the

presentational construction is essentially a 'topic-creating' construction. 22 It
is a pragmatically motivated grammatical device for the mapping of syntac
tic structure and information structure which has the function of 'pro
moting' referents on the topic acceptability scale from non-active to active
status in the discourse (hence from lexical to pronominal coding in the
sentence) and thus to allow speakers to adhere to the cognitively preferred
and grammatically unmarked topic type. It is important for our purposes to
observe that from the point of view of grammatical coding the referent in
troduced via a presentational construction is marked as a focus in the first,
i.e. the presentational, clause and as a topic only in the second clause. The
clause Once upon a time there was an old king is not about king but it in
troduces a king; it is only in the following clause, who lived in a beautiful
castle, that the referent becomes linguistically a topic, coded in the topic ex
pression who.
One common explanation for the presentational clause in the presenta
tional two-clause construction is that it asserts the existence of the referent.
But since existence always entails 'being in a place', as pointed out e.g. by
Van Oosten (1978), it seems preferrable, from the discourse-pragmatic point
of view, to interpret the function of the presentational clause as that of
presenting or introducing the referent into the 'place' or scene of the
discourse and, thereby, of raising it into the addressee's consciousness,
rather than asserting its mere existence. This pragmatic interpretation ac
counts well for the fact that in the French presentational construction the
'presented' NP can be a definite description and even a proper name (cf. exx.
(1) through (3) above), i.e. an expression whose referent is not only presup
posed to exist but also to be known to the addressee, so that assertion of its
existence would be tantamount to a redundancy. It is also well known that
'existential' clauses often begin with a place adverbial, such as English there,
German da, French y etc., making the presentational, i.e. 'place-oriented',
interpretation even more compelling. Nevertheless there is a logical link bet
ween the existential and the presentational function, and, following Van
Oosten (loc. cit.), I prefer to see the two as endpoints on a continuum rather
than as two discrete functions.23
3.3 The presentational function of the avoir-cleft.
As I mentioned earlier, the avoir-cleft construction may be used in two

150 KNUD LAMBRECHT
apparently quite heterogeneous discourse functions: (i) in the presentational

function proper, i.e. to introduce into the current discourse register new
referents that may then become topics in subsequent discourse; and (ii) in
an event-reporting function, in which the 'presented' NP referent is not a
topic in any clear sense. The two functions are not categorially distinct and
it is not always possible to define an utterance as belonging to one rather
than to the other type. In this section I will describe the construction in its
presentational function. In the following sections I will then account for the
event-reporting use as a case of pragmatic and semantic extension.
Below are two examples of what I referred to earlier as the prototypical
presentational construction, containing a brand-new discourse referent. Ex
ample (9) is taken from a conversation about traffic problems in Paris, and
(10) is part of a narrative about the food shortage in Marseille under the Nazi
occupation:
(9) à l'heure actuelle j'm'plains pas, y'a un camarade d'usine qui
m'ramène en voiture jusqu'aux Quatre Routes pour prendre
l'autobus (François, 1974:818)
'right now I'm not complaining, there's a friend of mine from
the factory who drives me back to Quatre Routes to take the
bus'
(10) moi j'avais une voisine qui était habituée à bien boire et bien
manger, et bien presque elle'en'devenait folle d'être privée de
la nourriture comme ça (Albert)
'(me) I had a neighbor who was used to drinking well and
eating well, well she almost went crazy being deprived of food
like that'
In both examples, a brand-new referent is 'presented' to the adressee in the
avoir-clause in the form of an indefinite noun phrase (un camarade d'usine,
une voisine) and thereby made pragmatically available for pronominal topic
coding in the following qui-clause. By the time the qui-clause is uttered, the
referent's pragmatic status has changed from brand-new to active, i.e. it has
become a topic of the preferred type, which is now available for further
reference in the discourse. The NP which codes the new referent is introduc
ed as an object of the verb avoir, i.e. it appears in the preferred post-verbal
position for NPs with non-active referents. The presentational information
about the NP referent is expressed in the qui-clause. As observed earlier, the
communicative function of the avoir-clause in a presentational cleft is not

so much to assert the existence of the NP referents but to pragmatically pose
the referents in the discourse so that some proposition may be expressed
about them. What counts in (9) is not the existence of some worker friend
but the fact that this friend gives the speaker a ride home; in (10), it is not
the fact that the speaker had a neighbor but that the person in question liked
eating and drinking (and therefore suffered from the occupation) that con
stitutes the relevant piece of information. Notice that the information ex
pressed by the two-clause sequences could be grammatically expressed in the
form of the monoclausal SV(O) sentences Un camarade d'usine m'ramene
en voiture and Une de mes voisines était habituée à bien boire et bien
manger. Both would by syntactically well-formed sequences. However they
would violate the pragmatic constraint on topic accessibility described in
Section 3.2 and would therefore be unacceptable on the discourse level.
One difference between the two examples is the choice of the sentence-
initial pronominal element. In both cases, this element serves as a reference
point with respect to which the new referent is 'located' in the discourse. But
in (9) this reference point is the place adverbial y, while in (10) it is the per
sonal pronoun je. I do not fully understand the semantic and pragmatic
criteria which cause speakers to use one or the other of these referent-
introducing strategies. A personal pronoun seems to be preferred when the
NP referent stands in a relationship of possession with the referent of the
pronoun, in particular when the NP itself is possessive. For example it is not
possible to use a personal pronoun if the introduced NP cannot have a
possessive determiner. While sentence (1) Y'a Jean qu'a téléphoné is well-
formed, the corresponding *J'ai Jean qu'a téléphoné is not. On the other
hand possessive NPs do occur with y. For example in (6) {J'ai ma voiture
qui est en panne) the sequence y'a could be substituted for j'ai, leading to
a subtle meaning difference which I find difficult to define. Other factors
influencing the choice of one or the other strategy include the type of
predicate (a stative or generic predicate usually requires a personal pronoun)
and the aspectual make-up of the sentence. In general, sentences with y seem
to have more of an 'all new' character, and in the clearest examples of event-
reporting it is y and not a personal pronoun that is used (cf. the discussion
in Section 3.5 below).
In order to understand how the adverbial y and the subject pronouns
je, tu, il etc. come to stand in a paradigmatic relationship in the avoir-cleft
construction, we must take a closer look at the verb avoir and its semantic
152 KNUD LAMBRECHT
properties. How can we explain the occurrence of the transitive verb avoir
in a clause whose pragmatic function is to introduce a referent into a
discourse and whose analog in English and other languages is the highly in
transitive verb 'be'? Following analyses by Clark (1970), Van Oosten (1978),
and Foley and Van Valin (1984), I interpret the verb avoir 'to have' as a syn
tactically transitive but semantically intransitive verb, whose subject has the
non-agentive semantic case role of locative. This interpretation is supported
by the fact that many languages code as the subject of a HAVE-verb what
is a direct object NP in English and French, and as a dative or locative what
is the subject in our languages. It is this peculiar semantic-syntactic status
of avoir as a verb with two argument positions but with a semantically non-
agentive subject that accounts for its use in the presentational construction
(cf. Lambrecht, 1987). The verb avoir allows a lexical NP to occur in post-
verbal position which by its semantic role would otherwise have to occur
clause-initially, as the subject of an intransitive clause. This analysis of the
verb avoir supports my interpretation of the avoir-cleft as a syntactic con
struction used to preserve the preferred clause structure [V(X)] and to avoid
the pragmatically constrained SV(O) clause type.
Thus on the semantic if not the syntactic level, the adverbial y in y'a and
the subject pronouns je, tu, il etc. are exactly parallel. I suggested earlier that
the adverbial y serves as a reference point with respect to which a new
referent is located in the discourse. We now see that the subject pronouns
have essentially the same semantic and pragmatic function of locative
arguments. This parallel between y and personal pronouns in the avoir-
clause explains no doubt the optionality of the 3p pronoun il in the expres
sion il y'a in a language which normally requires the presence of an overt
subject marker (cf. however Lambrecht 1981 and 1987 for exceptions to
this requirement). The parallel may also account for cases such as (11)
(11) t'y'as ces vagues qui viennent des fois s'jeter sur les rochers
(François, 2:16)
'you have (there) those waves that sometimes are thrown
against the rocks'
in which tu cooccurs with y before the verb avoir. Occurrences of t'y'as are
not uncommon in the corpuses I have consulted.
Use of the verb have in presentational constructions is by no means con
fined to French. I mentioned earlier the special case of Spanish hay as a form
of haber 'have'. Both French y'a and Spanish hay go back to Vulgar Latin
habet, whose use as an intransitive presentational verb is well-attested (cf.

Melander, 1921). Further examples include Chinese yoû 'have, exist' and
dialectal German es hat 'there is'. Spoken English too has a construction
very much like the one illustrated in (10) above. This construction is ex
emplified in the following observed utterances:
(12) Speaker A, seeing sheets on a living-room couch:
A: I thought maybe your grandmother was using the room.
B: No, we had a friend of mine from Norway was staying
here. (K.L.)
(13) I have one of my uncles was an engineer and he... (K.L.)
(14) Well, I had a friend of mine called me. (Prince, 1981:238)
Spoken English and spoken French thus have in common the discourse-prag
matic phenomenon whereby a referent which is judged too low on the topic
acceptability scale is placed in object position after the verb have instead of
being coded as a clause-initial subject.24 One difference between these
English examples and the French avoir-cleft is that, while (12) through (14)
are usually considered substandard or dialectal in American English, the
French construction is firmly established in general spoken usage. No French
speaker could do without it. Another difference is that in English the relative
pronoun can be omitted. But the most important difference within the con
text of the present study is the fact that the use of the English have-construc-
tion seems restricted to indefinite NPs, i.e. to the introduction of brand-new
referents.25 Spoken English seems to have a different threshold of tolerance
for subject NPs with non-active referents than spoken French, resulting in
a different cut-off point on the topic acceptability scale in (8). While English
seems to tolerate relatively freely semi-active and even unused subject refe
rents, French systematically avoids all non-active referents in subject posi
tion. Moreover, the matching requirement between subject, topic, and
clause-initial position is much stricter in French than in English (cf. the dif
ference between the focus-initial English example (1) and the corresponding
French avoir-construction in (3) above). I argue elsewhere (Lambrecht,
1984a) that the strong preference in spoken French for active subject refe
rents takes the form of a general pragmatic constraint against lexical subjects
in this language. Since lexical coding is the unmarked way of expressing
referents that are not yet discourse-active, this restriction on the use of lex
ical subjects is a predictable consequence of the information structure con-
154 KNUD LAMBRECHT
straints in the spoken French clause.

The strong constraint in spoken French against non-active referents in
initial subject position is no doubt the main factor motivating the occurrence
of definite NPs in presentational clefts. As I will demonstrate below, the
cognitive status of the referents of such definite NPs can range from unused
to accessible. However no active referents may appear in an avoir-cleft. The
occurrence of active referents would indeed run counter to the very function
of the construction, which is to introduce referents that are assumed to be
not yet active in the discourse. Thus parallel to (1), repeated here as (15a),
we do not find utterances such as (15b):
(15) a. Y'a Jean qu'a téléphoné
'Jean called'
b. # Y'a moi qui ai téléphoné
'I called'
The unacceptabihty of (15b) is due to the fact that moi T refers to the
speaker, who is necessarily present in the speech situation and does therefore
not have to be specially introduced via the presentational cleft.
Notice that the unacceptabihty of (15b) is pragmatic, not syntactic.
Autonomous pronouns such as moi (which behave syntactically like NPs)
may occur as the object of y'a in other pragmatic contexts. This is shown
e.g. in the attested utterance (16), which was produced in a discussion about
plans for attending a play for which special group discounts were available:
(16) y'a Beth qui veut y'aller, euh, y'a y'a Jean-Marc, y'a moi, bon
(Barnes)
'there's Beth who wants to go, uh, there's there's Jean-Marc,
there's me, ok'
In (16), the avoir-clause y'a moi is understood to be followed by the same
qui-clause as the one following the NP Beth (i.e. qui veut y'aller). The accep
tability of (16) makes this sentence an apparent counterexample to the con
straint illustrated in (15b). Closer examination of the discourse context
reveals, however, that (16) is not an instance of a presentational cleft, but
of a sentence type whose English equivalent has been referred to as the listing
there-construction (cf. Rando and Napoli, 1978). In this listing function, the
English there-construction, like the French construction above, tolerates in
dependent pronouns, as shown e.g. in the English gloss of (16) or in a
sentence like If you need help, there's always me. It is a defining feature of
the listing construction that the expressed or understood proposition of
which the NP is an argument must be pragmatically presupposed in the
discourse context, i.e. cannot in itself constitute new information. This
presuppositional nature of the proposition is reflected in the fact that, in
French as well as in English, the relative clause in the listing construction has
the low tone associated with pragmatically presupposed information. It does
not contain an information focus of its own in the form of an intonation
peak.
Thus in the presentational avoir-cleft, the cognitive status of the NP
referent as brand-new, unused, or accessible but not discourse-active is
criterial. For example if the statement 'Jean called' in (1) were made in a
context in which the referent 'Jean' was already a discourse topic (e.g. if in
preceding discourse someone had enquired about the whereabouts of Jean),
sentence (1) would be an eminently unacceptable utterance. In this modified
discourse situation the statement would either have to be of the form iVa
téléphoné (assuming that the referent of the NP Jean was still active), or it
would have to appear in one of the 'dislocated' forms Jean iVa téléphoné,
or IVa téléphoné, Jean (assuming the referent had been deactivated by in
tervening discourse; cf. Lambrecht, 1981 and 1987). Likewise the utterance
in (2) could not be used felicitously in the somewhat unlikely case that the
relevant telephone were indeed a current topic (e.g. if it had just been fixed
by the repairman because its bell was not working). Similarly, example (3)
could not be used if in previous discourse I had already been talking about
problems with my eyes.
The following is a representative example of an avoir-cleft construction
in which the introduced referent is coded as a definite NP. (The example also
contains a presentational construction involving an indefinite NP.) The text
in (17) is from a conversation about the rapid progress of modern
technology:
(17) A: étant quand quand j'ai fait mon apprentissage à la Lor
raine, on avait des voitures, ça'roulait tout de suite à
quarante, soixante à l'heure, quelle vitesse (...), y'avait un
prix de j'sais pas combien pour celui qui dépasserait I'cent
B: oui mais y'a eu un
C: la course du Mans, oui oui
A: à l'heure actuelle la course du, oui
156 KNUD LAMBRECHT
B: j'ai eu mon beau-frère moi qui a fait un euh Paris-Nice, le

la course de lenteur, iVa été pénalisé parce qu'il'allait trop
vite, mon beau-frère de Suisse, ah oui iVa eu ben il'avait
perdu parce qu'il'allait trop vite (François, 1974:817)
A: when when I did my apprenticeship at the Lorraine factory,

there were cars, they went forty sixty kilometers an hour
right away, what speed (...) there was a prize of I don't
know how much for the person who'd go faster than a hun
dred kilometers an hour
B: yes but there was
C: the Le Mans race, yes yes
A: nowadays the race what, yes
B: I had my brother-in-law (me) who did a uh Paris-Nice, the
the slowness race, he was penalized because he went too
fast, my brother-in-law from Switzerland, oh yes he had,
well he lost because he went too fast.'
In speaker A's first turn the new discourse referent 'cars' is introduced in
the form of the indefinite NP des voitures, in a special construction type in
which the second clause is not introduced by a relative pronoun but simply
juxtaposed to the avoir-clause.26 The relevant passage in (17) is the presenta
tional cleft at the beginning of speaker B's last turn, introducing the new
referent mon beau-frère 'my brother-in-law'. This referent, which is to
become the topic of the next few clauses, was not mentioned earlier in the
conversation. It is thus of the unused type. Because of this non-active
cognitive status, the relevant piece of information about the referent, namely
that the person in question participated in a particular kind of car race, is
expressed via the presentational cleft. As in (9) and (10), the corresponding
canonical SVO sentence {Mon beau-frère a fait Paris-Nice 'My brother-in-
law did Paris-Nice') would be unacceptable on the discourse level.
3.4 Semantic observations
Before I discuss further examples of avoir-clefts involving definite noun

phrases, I would like to point to some interesting semantic properties of the
construction type exemplified in (17). Consider first the expression of tense.
There seem to exist two options concerning the tense of the verb avoir. The
first option is for the verb to remain frozen in the present tense, the actual
event time being expressed only by the tense of the qui-clause. This option
is illustrated in the following example.27
(18) chaque fois qu'il'passait des aviateurs il'y'a la tante qu'elle

demandait des nouvelles (Deulofeu, 1984:116)
'
every time pilots came by the aunt would ask for news'
The time at which the (repeated) event took place is expressed in the iterative
past tense of the qui'-clause qu'elle demandait des nouvelles, while the tense
in the avoir-clause il y a la tante is frozen. Further examples can be found
in (1) (y'a Jean qu'a téléphone) and in (23) and (24) below.
In spite of the phenomenon of frozen tense in (18) it would be
misleading to characterize the sequence y'a NP as a syntactically fixed
phrasal 'idiom'. 28 There is a second option concerning the tense of the avoir-
clause, namely, that the verb avoir takes on the tense of the qui-clause. An
example of this option occurred in (17), the relevant portion of which is
repeated here as (17'):
(17') j'ai eu mon beau-frère moi qui a fait un eh Paris-Nice
'my brother-in-law did a uh Paris-Nice'
In (17'), literally glossed as 'I've had my brother-in-law who has done Paris-
Nice...', the avoir-clause and the qui-clause have the same tense. This tense
adjustment appears at first to be a simple case of tense concord. But it dif
fers in one remarkable way from normal tense concord: the direction of the
adjustment is reversed. Since the tense of the qui-clause expresses the time
of the event, it is the 'subordinate' clause that governs the tense of the 'main'
clause, and not the other way around. This instance of 'reverse tense con
cord', in which the avoir-clause anticipates the tense of the qui-clause, is by
no means an isolated case, as shown in the following observed utterances:
(19) mais Vy'a eu cent cinquante personnes qui ont été pendues en
Afrique cette année (Jeanjean, 1979)
'but there were one hundred and fifty people (who were) hang
ed in Africa this year'
(20) rentrer en prison à quatorze ans, vous'savez, c'est comme si
vous'aviez la montagne qui vous'tombait sur la tête (Blanche-
Benveniste, 1983)
158 KNUD LAMBRECHT
'going to jail at age fourteen, you know, that's if you had the
mountain (that was) falling on your head'
The concord phenomenon in (20) is particularly interesting. The conjunction
comme si 'as if requires the imperfect ('imparfait') or pluperfect indicative
of the verb, a form which expresses irrealis mood in all subordinate counter-
factual clauses. Expressed in the form of a canonical sentence, (20) would
be C'est comme si la montagne vous'tombait sur la tête ('It's as if the moun
tain was falling on your head'). Notice that the imperfect indicative form of
the verb is an automatic consequence of the presence of the conjunction
comme si. A counterfactual clause not introduced by comme si or a similar
conjunction must appear in the conditional mood. Thus the verb of a
(restrictive) relative clause embedded under a comme si-clause would be in
the conditional (provided that the relative clause is itself construed as
counterfactual). This is shown in the made-up clause sequence (20'), which
parallels the relevant sequence in (20):
(20') c'est comme si vous'conduisiez une voiture qui aurait (*avait)
pas de freins
it is as if you drove a car which would have (*had) no brakes
'it's as if you were driving a car that didn't have brakes'
Unlike in English, the verb in the relative clause in (20') could not be in the
past indicative. Yet the qui-clause in (20) does appear in the indicative form
tombait. It follows that this form must be directly governed by the conjunc
tion comme si (just like the unique verb in the canonical version above is
governed by it), in spite of the fact that syntactically the conjunction is part
of the avoir-clause. This confirms the repeatedly made observation that the
qui-clause in the presentational cleft is not 'subordinate' in any clear sense
but expresses the main predication of the two-clause sequence.
These cases of anticipatory tense or mood concord are all the more in
teresting since they are completely non-standard and as such could not be
based on any rule acquired through some normative learning process. I take
this non-standard character of the phenomenon as a guarantee of the deep
linguistic reality of the grammatical process in question, in agreement with
Fillmore (1985) who, in an analysis of a substandard construction of spoken
English, makes the following observation: "Whenever we find impressive
regularities in language that we know we didn't learn either at mother's knee
or in Miss Fidditch's classroom, we can be sure that we are in touch with
structures seated deep in the langauge, and not inventions externally impo
sed upon it" (1985:73).
The phenomenon illustrated here is an example of a more general pro
perty of the avoir-cleft construction, which we may refer to as its non-com-
positionality (cf. Kiparsky, 1976; Fillmore 1979). The construction is non-
compositional in that it cannot be construed with the ordinary compositional
principles of syntax and semantics. From the point of view of compositional
semantics, a clause such as j'ai eu mon beau-frère in (17') is doubly
anomalous. First the use of the past tense ('passé composé') form j'ai eu is
odd, given that the speaker still has her brother-in-law at the time of the ut
terance. Secondly, the combination of j'ai 'I have' with mon beau-frère 'my
brother-in-law' is oddly redundant in that the possessive phrase already ex
presses the information that the referent coded in the possessive pronoun my
'has' his or her brother-in-law.29 An English sentence like I had my brother-
in-law might make 'literal' sense if used e.g. as an answer to the question
Who was going to help you? It would then be semantically and pragmatically
analogous to the listing type there-construction There was my brother-in-law
(i.e. who was going to help me). But this listing interpretation is not available
in the French clause J'ai eu mon beau-frère in the context of (17). Besides
obvious pragmatic differences, the verb in the corresponding listing type
sentence would have to be in the simple past (i.e. J'avais mon beau-frère).
The same type of semantic anomaly is found in the following example:
(21) J'ai mon neveu là qui va...qui s'marie là (François, 1974:829)
'I have my nephew y'know who's going to...who's getting
married'
(21) is not meant to assert that the speaker 'has her nephew' but that the
nephew in question is getting married. Particularly problematic from the
compositional point of view is example (3) J'ai les yeux qui m'font mal. Here
the avoir-clause J'ai les yeux 'I have the eyes' (referring to 'my eyes') is not
in itself a semantically (or pragmatically) well-formed assertion. The clause
is meaningful only in combination with the following qui-clause. An extreme
example of semantic non-compositionality is the following:
(22) Moi j'ai encore un formulaire que j'ai pas (Blanche-
Benveniste, 1983)
me I have still a form that I don't have
'there's another form I don't have'
160 KNUD LAMBRECHT
The sentence in (22) expresses a logical contradiction if analyzed composi-

tionally as a main clause asserting that the speaker has a certain form, and
a subordinate clause presupposing that the speaker does not have this
form.30 But if the two-clause sequence is interpreted as the global construc
tion which in fact it constitutes in natural discourse, the sentence makes
good sense. The kind of anomaly observed in these examples constitutes
good evidence for my interpretation of the avoir-clause in the presentational
cleft as a pragmatically motivated, non-predicating clause, whose only pur
pose is to pose the NP referent in the discourse. It strongly suggests an inter
pretation of the presentational cleft construction as a holistic grammatical
gestalt, whose parts cannot be understood without reference to this global
semantic and pragmatic unit.
3.5 From presentation to event-reporting
Example (17) with its definite NP mon beau-frère illustrated the use of
the presentational cleft for the introduction of an unused referent into a
discourse. As further examples of avoir-constructions containing definite
noun phrases consider again our model sentences (1) Y'a Jean qu'a
téléphoné, (2) Y'a le téléphone qui sonne, and (3) J'ai les yeux qui m'font
mal. Pragmatic contexts for these made-up sentences are easy to provide.
The referent of the NP Jean in (1) could be characterized as unused, e.g. in
a situation in which (1) serves as a message from the person who received
the phone call to the person the caller tried to reach. In (2) the referent of
the NP le téléphone might be either unused or (situationally) accessible.
Finally in (3) the referent of les yeux, being part of the speaker's body, must
be (situationally) accessible.
There is an important semantic and pragmatic difference between ex
amples (1), (2), and (3) on the one hand, and examples (9), (10), and (17)
on the other hand. In the latter group the individual referents un camarade
d'usine, une voisine and mon beau-frère represent pragmatically salient
discourse participants (the factory friend, the neighbor and the brother-in-
law are to become topics about which further information will be conveyed
in subsequent discourse). By contrast, the referents of the NPs Jean, le
téléphone and les yeux do not have to be similarly salient in the context.
While it would be possible for the individual Jean in (1) to become the center
of subsequent talk, it is equally possible that he is mentioned only as a
necessary participant in the reported event, i.e. the phone call. The same is
true of les yeux in (3). For example one could utter (3) in reply to the ques
tion Why are you wearing sun glasses today? and then never mention one's
eyes again. If sentence (3) has a topic, it is more likely to be the speaker
himself than his eyes. Likewise in example (6) J'ai ma voiture qui est en
panne (as in the parallel examples (4) and (5) in English and Italian) the au
dience does not expect the speaker to go on talking about her car. It would
be quite surprising under the circumstances if she did. What counts in the
context of (6) is not the individual car referred to by the NP but rather the
fact that some event of mechanical malfunction took place which involved
the speaker's car and which caused the speaker to take the bus home from
the supermarket. The same observation applies to example (2) Y'a le
téléphone qui sonne! We do not expect the speaker who utters this sentence
to go on and convey more information about the individual telephone.
Rather (2) communicates some proposition like There is an event of phone-
ringing' to which the speaker hopes someone will react and in which the in
dividual telephone is irrelevant.
The event-reporting interpretation of the avoir-cleft is not confinėd to
occurrences of definite NPs. Consider example (23), which contains the in
definite NP une voiture:
(23) ben ouais ah ben dimanche quand on arrivait i'y'a une voiture
qui a passé sur les pattes du chien de Gomez (Jeanjean, 1979)
'well yeah oh well Sunday when we were coming home a car
ran over the legs of Gomez' dog'
As in the previous cases, what counts here is not so much the identity of the
NP referent involved in the event (in this case the agent une voiture) as the
reported event itself. If there is a pragmatically salient, topic-worthy,
referent here it is the dog, not the car, and subsequent discourse is more like
ly to be about it than about the automobile.
As I mentioned earlier, the function of these event-reporting utterances
deviates from the presentational function proper as defined e.g. by Hetzron
(cf. note 22). In some important sense, all these short two-clause utterances
express single, non-complex pieces of propositional information, com
parable to the short monoclausal English and Italian utterances (4) and (5).
They constitute self-contained messages of a kind which Chafe (1974:115)
calls 'conceptual unities'. In these unities, the NP referents have no
pragmatic saliency beyond the clauses in which they occur. They do not mat
ter as individuals, but only as necessary elements in the expressed proposi-
162 KNUD LAMBRECHT
tion. In sum, the self-contained pieces of information expressed in these

avoir-clefts are reports of events, not comments about some topic.
The distinction I am trying to establish here in rather crude terms be
tween event-reporting utterances and topic-comment utterances is related, I
believe, to the distinction drawn by Kuroda (1972) between the 'thetic' and
the 'categorical' sentence type. Even though Kuroda's paper is concerned
with a logico-semantic distinction between types of propositions rather than
with a pragmatic distinction between utterance types, the parallel between
his analysis and the one presented here is important enough to warrant a
brief discussion. The thetic/categorical distinction discussed by Kuroda was
originally established by the 19th century philosopher Franz Brentano as a
fundamental distinction between two types of human judgment. Brentano's
theory, which was further developed by Anton Marty, assumes that the
categorical judgment type, which is expressed in the traditional sentence type
containing a subject and a predicate, is logically complex and "consists of
two separate acts, one, the act of recognition of that which is to be made
the subject, and the other, the act of affirming or denying what is expressed
by the predicate about the subject" (Kuroda, 1972:154). In addition to this
categorical type, Brentano and Marty postulate another, logically non-
complex, judgment type involving only the recognition or rejection of some
judgment material. This judment type is called 'thetic'.
According to Kuroda, this distinction between two judgment types is
grammatically manifested in Japanese in the difference between the particles
ga and wa. For example the difference between the two sentences (i) Inu ga
hasitte iru and (ii) Inu wa hasitte iru, both translatable as 'the dog is run
ning', is described as follows. The former (the thetic) sentence, containing
ga, represents 'the fact that an event of running (...) is taking place, involv
ing necessarily one (...) participant in the event.' The speaker's intention is
directed in (i) toward the entity participating in the event, i.e. the dog, 'just
insofar as it is a constituent of an event'. In the latter (the categorical)
sentence, which contains wa, 'the speaker's interest is primarily directed
towards the entity (...), and the reason why he wants to give an expression
to the fact that he recognizes the happening of the event (...) is precisely that
he wants to relate the occurrence of the event to this entity' (1972:162ff).
The entity to which an event is related by the speaker in this way is refer
red to by Kuroda as the 'subject', which is grammatically manifested in
Japanese as a wa-marked NP. Thetic sentences such as (i), on the other
hand, in which the entity is only a necessary participant in an event, are call-
ed 'subjectless'. It is clear that Kuroda's notion of 'subject' corresponds

closely to the notion of 'topic' in my more pragmatically oriented analysis,
while Kuroda's 'subjectless' sentence corresponds to my event-reporting ut
terance type. 'Subjectless' is then also equivalent to 'topicless', since the NP
in the event-reporting sequence is a focus. Thus Kuroda's (i) would translate
as the event-reporting sequence Y'a le chien qui court, while his (ii) would
correspond to the topic-comment sequence Le chien iVcourt.
What makes Kuroda's account of thetic sentences in Japanese par
ticularly interesting in the context of this study is the fact, not mentioned by
Kuroda, that thetic ga-sentences are also (and perhaps primarily) used in the,
presentational function proper. Kuno (1972), in his functionally oriented
analysis of the wa/ga-contrast, observes that one discourse situation in
which the subject particle ga is used instead of the topic particle wa is the
situation he calls 'neutral description' (an unfortunate term, I think).
Neutral description obtains when a sentence conveys totally new informa
tion, in the sense that both the referent of the subject NP and the proposition
involving this referent are new to the discourse. Kuno also observes that
neutral description clauses tend to be highly intransitive, containing verbs in
dicating existence or coming into existence of some referent or appearance
of a referent at the scene of the discourse. These are of course precisely the
verbs found in presentational clauses across languages (cf. Section 3.2.
above). It is well known for example that the verbs which most frequently
trigger subject-verb inversion in Romance are of this type, the prototypical
Romance inversion construction being presentational in nature (cf. e.g.
Wandruszka, 1981).
How does this account of the difference between sentences in which an
NP referent is a topic and sentences in which it is merely a necessary partici
pant in an event fit the interpretation of the avoir-construction proposed in
this study? If the fundamental pragmatic function of the avoir-cleft is
presentational, i.e. if the construction is used primarily to promote referents
from non-topic to topic status, why is the same construction then used for
utterances in which the referent is not a topic but only such a necessary par
ticipant in an event?
I believe that the answer to this question has to do with the 'all new'
character of the information conveyed in both cases. Recall the pragmatic
constraint against active NP referents in the avoir-cleft (cf. example (15) and
discussion). Even though the NP referents in the event-reporting subtype do
not have the quality of potential discourse topics, event-reporting clefts re-
164 KNUD LAMBRECHT
main presentational, i.e. referent introducing, in that they can be used ap
propriately only if the clefted referent has a degree of newness in the
discourse, or, more precisely, if the referent is not yet an established topic
in its discourse context.
I claim that it is this constraint on the cognitive status of the NP referent
in the avoir-construction that accounts for the use of the avoir-construction
in the two apparently quite heterogeneous pragmatic functions. The
pragmatic 'all new' character of the message in the referent-introducing and
in the event-reporting functions is the unifying feature that sets them off
from the topic-comment function. While in the former two utterance types
both the NP referent and the proposition involving the referent are new
elements in the discourse, in the latter only the proposition is new, the topic
being a pragmatically recoverable element.31 Recall our presentational pro
totype, the narrative-initial Once upon a time there was... The discourse
world introduced by this narrative formula is by definition a 'new' world,
without a cohesive link to preceding discourse. The event-reporting subtype
is similarly detached from the discourse context. Its non-cohesive nature is
reflected in the fact that an event report may be preceded in discourse by cer
tain stereotypical utterances such as What's new?, What happened? or
Guess what! What distinguishes these utterances from information-
requesting utterances like e.g. How are you? or Who is that guy over there?
etc. is that they carry no presupposition as to the topic or the proposition
to be contained in the reply.
The use of the presentational construction for the event-reporting func
tion is thus a natural pragmatic extension of the 'all new' character of the
presentational utterance. This extension is motivated by the focus character
of the NP in the avoir-clause. Recall that in the first clause of a presenta
tional cleft the referent is not yet a topic and appears grammatically as a
focus NP. The difference between the two pragmatic subtypes may then be
characterized as follows. In the presentational function proper the referent
of the focus NP becomes a topic in the qui-clause, while in the event-repor
ting function this passage from focus to topic does not take place. We may
think of the passage from presentation to event-reporting as a case of
pragmatic reanalysis, or 'pragmatic re-bracketing'. While in the presenta
tional case the two-clause sequence is construed pragmatically as [[y'a —
referent][proposition]], in the event-reporting case it is construed as [[y'a]
[referent — proposition]]. In one case it is the referent that is presented, in
the other it is the event.32
3.6 A further extension
To make this account of the two utterance types and their common
coding complete it is necessary to deal with a set of apparent counterex
amples. These counterexamples involve referents which have all the
cognitive and pragmatic requirements for syntactic expression as topics but
which nevertheless occur as presentational foci in a y'a-cleft. Such examples
seem to contradict the claim that avoir-constructions are restricted to 'all
new' utterances. They occur typically with referents that are situated in that
shady area between discourse-active and inactive status referred to earlier as
the area of pragmatic accessibility.
One such example of an already accessible, topical discourse referent
being coded as the focus of a presentational cleft is illustrated in the follow
ing short narrative, which I have analyzed elsewhere (Lambrecht, 1987) as
an illustration of the different ways in which discourse referents can be syn
tactically coded in spoken French. The text in (24) is part of a discussion
about the problem of obesity. Speaker M. is a Frenchwoman married to an
American. Her interlocutors are French speakers living in the United States.
The passage in (24) narrates the first encounter between the speaker's
parents and the parents of her husband:
(24) M: ben alors, moi j'vais te'dire, quand mes parents sont
venus pour le marriage, alors euh...évidemment, mon
père a la même taille que moi, ma mère est plus
petite, euh, mon père fait euh cent dix pounds cent
dix pounds, c'est à dire que cinquante deux kilos
E: oui
M: cinquante deux kilos
C: c'est un moustique!
M: mais quand on 'l'a vu, alors la famille de
Bill...évidemment, son frère, i'fait deux cents et quel
ques pounds
E: deux fois ton père!
M: son père, qui, qui est vraiment trop gros...la mère
bon, ça'va, mais enfin, enfin, une famille typiquement
américaine de c'point d'vue-là. Quand is'ont vu papa,
tout petit, tout chétif, y'a mon beau-frère qui l'a
porté comme ça. Il'l'a soulevé comme ça.
Il'en'revenait pas de sa légèreté! (Barnes)
166 KNUD LAMBRECHT
'M: well then, (me) I'm going to tell you, when my

parents came for the wedding, then uh...of course, my
father is the same size as me, my mother is smaller,
uh, my father weighs, uh, 110 pounds, 110 pounds,
that is, only 52 kilos
E: yes
M: 52 kilos
C: he's a mosquito!
M: but when people saw him, well, Bill's family...of
course his brother he weighs two hundred and some
pounds
E: twice as much as your father!
M: his father, who, who is really too fat...the mother,
well she's ok, but anyway, anyway, a typically
American family from that point of view. When they
saw papa, so little, so puny, my brother-in-law carried
him like this. He picked him up like this. He couldn't
get over his lightness!'
In describing the physical appearance of the members of her husband's
family (son frère, son père, la mère), speaker M. resorts to the characteristic
topic coding strategy, i.e. Left Dislocation,33 This strategy is appropriate
here because the referents coded as topic NPs are pragmatically accessible
via the already evoked family frame. However in narrating the high point
of the story, at which the brother-in-law lifts the speaker's father up and car
ries him around, the speaker switches to the presentational cleft y'a mon
beau-frère qui l'a porté comme ça. Notice that the y Vclefted referent of the
NP mon beau-frère is identical to the referent of the previously named NP
son frère, which, together with the NPs designating the other members of
the family, was coded in a dislocation construction. In fact the referent is
included in the verb form is'ont vu 'they saw' at the beginning of the im
mediately preceding clause. The referent mon beau-frère must therefore be
cognitively of equal if not increased accessibility. In spite of this similarity
in pragmatic status, it would have been less felicitous at this point in the nar
rative to resort again to the dislocation strategy. Neither mon beau-frère i'l'a
soulevé comme ça nor I'l'a soulevé comme ça, mon beau-frère would sound
right in the context.
In example (24) the argument of the pragmatic 'all new' character can-
not be invoked to explain the occurrence of the y Vconstruction, since the

NP referent is an already topical referent in the context. We must therefore
look for some other explanation. I believe what motivates this occurrence is
not the non-active cognitive status of the NP referent but rather the peculiar
nature of the conveyed information itself. To see this, let us modify the
semantic content of the relevant portion in (24), leaving the remaining
discourse unchanged:
(24') quand is'ont vu papa, mon beau-frère il'en'croyait pas ses
yeux (...)
'when they saw papa, my brother-in-law he couldn't believe his
eyes'
In this semantically modified version, the left-dislocated topic-marking con
struction (mon beau-frère il'en'croyait pas ses yeux) is preferrable to a y'a-
construction. Since the difference between (24) and (24') is not one in
cognitive status of the referent but exclusively one in semantic content, the
motivation for the y''a-cleft in (24) must have to do with the type of informa
tion conveyed by the sentence. However it is not the propositional content
of the information itself that accounts for the construction (in some other
context, the content of the relevant passage in (24) could be packaged dif
ferently). Rather it is the expected reaction of the interlocutor to this pro-
positional content. The difference between (24) and (24') is thus essentially
discourse-pragmatic. What distinguishes the two versions is that the message
coded in the form of they'a-constructionis assumed to be more surprising
and unexpected, and thus of greater news value than the message in (24').
What is marked as 'new' in (24) is thus not the referent of the NP but rather
the entire piece of information in which this referent is involved as a salient
participant.
Here is another example. On a recent trip to Switzerland I visited
friends whom I had not seen in several years. During a conversation which
revolved about changes that had taken place during my absence, we came
to talk about some common friends, a couple with two sons. Once that fami
ly was introduced as a discourse topic into the conversation, all four
members of it had become cognitively accessible referents. After talking
about the wife and the husband, my interlocutor made the following in
troductory comment concerning one of the sons:
168 KNUD LAMBRECHT
(25) Y'a Sacha qui est grand! (K.L.)

'Sacha is tall'
Given the accessible status of the referent 'Sacha' in the context and its role
as a potential topic, the normal grammatical strategy would have been to use
a topic-shifting construction such as (25'):
(25') Sacha West grand maintenant
'Sacha he's tall now'
The choice of the y'a-construction in (25) is all the more striking as the
highly stative predicate of the sentence does not express the kind of informa
tion one would normally characterize as an event. Indeed statements of the
form 'X is tall', in which some property is attributed to some entity, are pro
totypical examples of the subject-predicate (topic-comment) sentence type
and normally exclude the existential (or presentational) interpretation.
However, given the surprise effect the statement of the child's tallness was
assumed to have on the listener in the discourse context of (25), the coding
of the message via the presentational cleft was more appropriate than its
coding via the topic construction in (25'). What is conveyed in (25) is not on
ly the message that the child in question has grown and is now tall but that
his tallness will cause the addressee to be surprised. It is not the propositional
content itself that accounts for the event character of the message but the
effect the information expressed in the proposition was assumed to have on
the addressee given the pragmatic situation.
There remains an analytical paradox to be accounted for. If French has
a special pragmatically motivated grammatical device for utterances involv
ing referents that are not yet available as topics in the discourse, why is this
device sometimes used in contexts where the 'presented' referent is in fact
already topical in the context? I suggest the following solution to this
paradox. Once the avoir-cleft is established as a grammatical device used not
only to introduce topical referents into a discourse but also to express a cer
tain type of 'all new' message, it becomes possible for the 'all new character'
of the device to get conventionally attached to the grammatical structure,
more or less independently of the cognitive status of the referent. The
cognitive constraint on the type of referent is then relaxed and it is the
'newness' of the piece of information that becomes criterial.
The relative ease with which this pragmatic shift can occur has to do
with the nature of the discourse-pragmatic notions of topic and focus.
Unlike the concepts of identifiability and activation, which have to do with

the assumed cognitive statuses of referents in the minds of the speech par
ticipants, the concepts of topic and focus have to do with the relation be
tween propositions and referents in given discourse situations. They are
communicative functions which to some extent can be freely decided upon
by speakers, given certain assumptions concerning the cognitive statuses of
the referents in the hearer's mind. In the case under analysis, the discourse
factor which allows the speaker to choose freely between topic or focus
coding has to do with the cognitive status of accessibility.
I would like to suggest a modification of Chafe's definition of ac
cessibility as summarized in Section 3.1. This modification will allow us, I
think, to confirm the above suggested explanation for our discourse-prag
matic paradox. Chafe's model assumes that accessibility (semi-activeness) of
a referent, in particular accessibility of the 'inferrable' type, entails that the
accessible referent is somehow, indirectly or peripherally, present in the
hearer's memory or consciousness. I suggest instead that we think of referen
tial accessibility as a potential for triggering inferential processes in the
hearer's mind. Given the possibility of drawing some inference leading to
determination of a referent, it is often the case that hearers will draw this
inference only if requested to do so on the basis of the form of the utterance,
whereas they will not draw this inference if such a request is not formally
expressed. The choice of a particular grammatical form can be understood
as a request by the speaker to the hearer to treat the referent of some expres
sion as if it had a certain pragmatic status, even though 'objectively' this
status is not granted. The deciding factor in characterizing the parameters
of activation and identifiability is often not so much the question whether
a referent is 'objectively' active or identifiable in the hearer's mind. Rather
what matters seems to be whether or not a speaker assumes that a hearer is
able and willing, in the presence of a particular grammatical from, to draw
the inferences needed to correctly interpret a message. In the case of the
event-reporting avoir-cleft involving already accessible referents, the choice
of the construction constitutes a request to the hearer to cancel a certain ex
pectation concerning the role of the accessible referent as a potential
sentence topic. And this request is understood because of the 'all new'
character which is conventionally attached to the grammatical construction
itself.34
170 KNUD LAMBRECHT
3.7 Crosslinguistic observations
I would like to end this analysis with a few crosslinguistic observations

which show that the phenomenon in question is not unique to French and
which suggest that my explanation of the French facts may be relevant for
other languages as well. The crosslinguistic facts I have in mind involve the
already mentioned wa/ga-alternation in Japanese and the phenomenon of
subject-verb inversion in Romance languages, an example of which was
discussed in Section 2. The facts concerning the Japanese wa/ga-alternation
are notoriously complex (cf. e.g. Clancy and Dowing, 1987), but this need
not affect the particular aspect of the alternation I have in mind.
It follows from the essentially presentational character of Japanese
'neutral description' ga and Romance subject-verb inversion that neither is
normally appropriate in discourse situations in which a referent is already
a predictable topic and in which the proposition expressed by the sentence
is to be interpreted as conveying information about this topic. However
there are some notable exceptions to this general rule, exceptions which
typically involve referents of the accessible type. Consider the following two
examples from the beginning of two Japanese newspaper articles:35
(26) Suzuki-shushoo wa hoobee o oeru ni atatte Washinton de
dookoo-kisha-dan to kondan,...
'
Prime Minister Suzuki, leaving the United States, had a press
conference with the attendant reporters in Washington...'
(27) Tanaka moto-shushoo ga nyuuin-shita. Taoreta no wa shi-
gatsu 21-nichi
T h e former Prime Minister was hospitalized. The day he col
lapsed was April 21st'
Since these two fragments appear at the beginning of the respective
newspaper articles, we can assume that the pragmatic status of the two
referents suzuki-shushoo and tanaka moto-shushoo is identical. Both are ob
jectively 'new' to the discourse in that they were not mentioned in previous
context; however both are also pragmatically accessible since they represent
potential topics of news reports. It is because of this accessible status that
the referent in (26) can be marked as topical via the particle wa. Yet the
referent in (27), which has the same accessible status, is marked as non-topi
cal via the particle ga.
As in the French examples discussed earlier, the difference lies in the

discourse-pragmatic nature of the information conveyed in the two
fragments. Having a press conference is an expected kind of activity for a
Prime Minister visiting the United States, but collapsing and being
hospitalized is a totally unexpected event. It is this unexpected character of
the information conveyed in the second example that accounts for the use
of ga. The parallel with the French facts is striking. In both languages, a
referent which has the necessary accessibility status to become a topic occurs
in a syntactic construction which is normally reserved for the introduction
of new discourse referents. The determining factor in both cases is the sur
prise character of the information expressed by the sentence.36
A strikingly similar case in Spanish involving an accessible referent
which is coded in a presentational sentence to express a higly unexpected
piece of information is illustrated in the following headline from the New-
York-based hispanic newspaper "El Diario", announcing the unexpected
resignation of Secretary of State Alexander Haig (June 26, 1982):
(28) Renuncia Haig!
'Haig resigns!'
The surprise character of the information is expressed syntactically via the
inversion construction. I was not able to find a parallel headline conveying
information about Haig or some other high official involved in some less
unexpected acticity, but I know that such headlines appear typically in the
unmarked form of SV(O) sequences. Example (28) shows that in Spanish,
as in French and Japanese, a syntactic construction which is normally re
served for brand-new or unused referents can come to be used for referents
which are accessible topics in the context, provided that the information ex
pressed by the sentence is assumed to have a high degree of news-value. In
all cases, the event-reporting function which is conventionally attached to
the construction in its prototypical usage is extended to discourse contexts
in which the NP referent does not have the cognitive and discourse-
pragmatic status which originally motivated the construction.
NOTES
1. I am grateful to the following people who, in one way or another, have helped me with
previous versions of this paper: Ruth Berman, Suzanne Fleischman, Chuck Fillmore,
172 KNUD L A M B R E C H T
Paul Kay, Johanna Nichols, and Sandy Thompson. Special thanks go to Cathy O'Connor
for last minute comments and encouragement with this version.
2. As a result of recent research by Zwicky (1977), Klavans (1985), and Macaulay (1984) on
the definition of clitics, I now prefer to call the French personal pronouns of the bound
series 'incorporated pronouns' (following a terminology used e.g. by Bresnan and
Mchombo (1987)) or 'pronominal affixes', instead of using the generally adopted label
'clitic' which I used in previous work and which is misleading because of a number of
systematic differences between French bound pronouns and clitics in other languages.
For a discussion of spoken French pronouns cf. Lambrecht (1981). Cf. also Stump (1980)
for a formal analysis of French pronouns as inflectional morphemes.
3. Not entirely unnoticed, of course: Blanche-Benveniste (1983) analyzes some of the formal
properties of the avoir-construction, and Wehr (1984) mentions its presentational use in
discourse; there is also an insightful discussion in Damourette and Pichon's monumental
grammar of French (1911-1934, vol.4:463ff).
4. Even though my research is based on actually observed data (transcripts of recorded con
versations), I use made-up examples whenever this seems appropriate. Observed data are
from various sources, indicated in parentheses after each example. The indication 'K.L.'
refers to my own data collection. 'Albert' refers to a short corpus established by Suzanne
Albert at the Université de Provence (Aix en Provence), 'Barnes' refers to Betsy Barnes'
(University fo Minnesota) extensive corpus of spoken French, which forms the data base
of Barnes (1985). I am indebted to Colette Jeanjean (University of Aix) for making the
Albert corpus available to me and especially to Betsy Barnes for allowing me to use her
corpus. In my transcription of spoken French data I take some minor liberties with the
traditional spelling conventions; in particular I always use an apostrophe to indicate the
morphological link between pronominal affixes and the verb, even when there is no eli
sion. Thus y'a instead of y a, il'en'a instead of il en a etc.
5. Since it does not contain a lexical subject NP, I count the sequence [qui V (X)] as a sub
type of the preferred clause structure [pro-V (X)], even though the relationship between
the relative pronoun and the verb is not one of incorporation. Notice that spoken French
frequently elides the final i in qui before vowel (cf. ex. 1), making for a tight mor-
phophonological connection.
6. "Le tour ya (...) existe depuis les origines même de la langue française" (Damourette and
Pichon, vol.4:524); cf. also Melander (1921).
7. Note that the notion of focus as used here differs from the more restricted focus notion
in e.g Chomsky (1970) or Akmajian (1973). In particular the focus in (4) and in the fol
lowing examples is not contrastive and does not have the stress intensity associated with
contrastiveness. Its presence does not entail that the information expressed by the pro
position."minus the focus" is (pragmatically) presupposed. To account for the focus in
tonation on the NP in these examples the notion of "new information" is of no help since
the reported event in its totality is new information. What is needed is some notion of
'default' focus, which in an 'all-new' utterance such as (4) will always fall on the lexical
NP rather than on the verb (cf. Lambrecht 1986b, Ch.5).
8. I am using Italian as a more or less random example of a language permitting expression
of the pragmatic information type in (4) via a VS structure. The 'inversion' phenomenon
in Italian is observable under the same discourse circumstances e.g. in Spanish (cf. Bol-
inger, 1954) and in Russian. Further cross-linguistic parallels will be drawn in Section 3.7.
9. One could argue that the post-verbal NP la macchina is in fact not a (prototypical) subject
but shares certain properties with objects (e.g. its position). However with respect to one
important criterion for subjecthood, verb agreement, the NP is clearly a subject; si è rotta
agrees in person, number and gender with macchina.
10. Non-clefted (presentational) sentences involving verb-initial structure are restricted in
modern spoken French to a semantic subclass of intransitive verbs and normally require
the dummy subject marker il (cf. Lambrecht, 1987). For a discussion of inversion
sentences in modern literary French cf. Bailard (1981) and references therein.
11. The semantic peculiarities of the avoir-cleft will be further discussed in Section 3.4. For
a syntactic analysis of the avoir-cleft construction and of the systematic differences be
tween it and the restrictive relative construction, cf. Lambrecht (1986a).
12. Cf. Mathesius (1928) and Thompson (1978). As Comrie (1981:Ch. 3.5.) has shown, this
subject-prominence of English correlates with the prominent existence, in English, of cer
tain grammatical constructions, such as the so-called rule of 'Tough-Movement', whose
function seems to be to allow a wide variety of semantic roles to be syntactically coded
as subjects, and as a corollary, to allow for the coding in initial position of topical
referents that otherwise would have to appear later in the sentence, resulting in a violation
of the principles of information structure. For a discussion of the semantic diversity of
the English subject (compared to subject in German) cf. also Hawkins (1981).
13. Italian, like Spanish, German and countless other languages, does permit placement of
a contrastive focus in clause-initial position. Thus an utterance like La macchina mi si è
rotta would be appropriate e.g. as a contradiction of a previously made statement that
some other object, e.g. the speaker's bicycle, broke down.
14. For the concept of competing motivations in grammar cf. in particular DuBois (1984).
15. For an in-depth discussion of these cognitive notions, as well as of the relational
pragmatic concepts 'topic' and 'focus', cf. Lambrecht (1986b, Chs 2-5).
16. The terms 'active' and 'inactive' replace the terms 'given' and 'new' in Chafe (1976). The
term 'accessible' corresponds roughly to the notion of 'recoverability' which I used in
previous work on spoken French. The notion of accessibility will be further discussed in
Section 3.6.
17. This diagram is adapted from Wehr (1984:9). Wehr's terminology and definitions how
ever do not coincide with mine in all respects.
18. The terms 'brand-new' and 'unused' are taken from Prince's (1981) taxonomy of infor
mation statuses.
19. The idea that the interpretation of a (certain type of) sentence involves these two indepen
dent tasks has a parallel in a basic logical assumption made by the philosophers Franz
Brentano and Anton Marty about the double judgment ('Doppelurteil') expressed in any
subject-predicate sentence. This parallel will be further discussed in Section 3.5.
20. Chafe (forthcoming) calls the cognitive effort necessary for interpreting a discourse-active
referent a 'low cost' effort.
174 KNUD LAMBRECHT
21. Notice that the acceptability scale I am postulating is meant to account only for NPs
whose referents are topics. I am not making claims concerning non-topical subject NPs.
Thus while (8) predicts e.g. that a sentence like A boy is tall is of low acceptability, it does
not make the same prediction for an event-reporting sentence such as A man just got run
over by a car! in which the subject is not a topic because the proposition expressed by
the sentence is not to be construed as being about 'a man'. Cf. the discussion about non-
topical referents in Section 3.5.
22. Cf. Hetzron's (1975:374) definition of the 'presentative function' as that of "calling
special attention to one element of the sentence for recall in the subsequent discourse or
situation. This recall may be needed because the element is going to be used, directly or
indirectly, in the ensuing discourse, because what is going to be said later has some con
nection with the element in question, — or because that element is relevant to what is go
ing to happen or be done in the reality".
23. Some languages have different lexical items to express the two functions. For example
standard German distinguishes existential es gibt from presentational da ist (sind).
24. In a brief discussion of (14) and other English examples, Prince (1981) suggests a
pragmatic explanation similar to the one I am presenting here for French.
25. These brand-new referents may be of the type called 'anchored' by Prince (1981). An an
chored brand-new referent is an unidentifiable referent which is made pragmatically more
accessible by being placed in a semantic relationship with an already active referent. For
instance the referents of the NPs a friend of mine and one of my uncles in (12) through
(14) are anchored and thus more easily identifiable than their unanchored counterparts
a friend, an uncle. This anchoring difference is shown by Prince to have syntactic conse
quences in English for the position of the NP in the sentence.
26. Substitution of a juxtaposed clause for the qui-clause is impossible with presentational
clefts of the event-reporting type.
27. The presence of the pronoun elle in the qui'-clause of this example is due to a non-standard
relative clause formation rule whereby the sequence qu(e)-pronoun appears in place of the
case-marked form qui.
28. For an analysis and formal classification of phrasal idioms cf. e.g. Kiparksy (1976). Cf.
also Lambrecht (1984b) for a discussion of Kiparsky's classification and for a proposal
which minimizes the idiomatic/generative dichotomy inherent in many modern
treatments of idiomaticity.
29. Compare the semantic peculiarity of the English sequences we had a friend of mine, I
have one of my uncles and I had a friend of mine in (12) through (14) above. Suzanne
Fleischman (p.c.) reminds me that a similar situation obtains in the perfectly 'normal'
English sentence I have my brother-in-law (who's) coming over this afternoon.
30. Chuck Fillmore observes (p.c.) that an analogous type of semantic anomaly, also involv
ing the verb 'have', can be observed in such English expressions as I have two missing
teeth.
31. This property also distinguishes the presentational cleft from what I call the 'identifica-
tional' or c'est-cleft construction (cf. Lambrecht, 1987). The identificational sequence
contitutes in some sense the reversal of the topic-comment sequence. In the c'est-cleft, it
is the proposition that is 'old' (i.e. pragmatically presupposed) and the referent involved
in the proposition that is 'new' (i.e. focal).
32. Further evidence for the existence of a conceptual link between presentation and event-re
porting may be found in the fact that in some languages the concept of 'arriving' (of a
person at some location) and that of 'happening' (of some event) are expressed by the sa
me lexical verb or verb root: cf. e.g. German ankommen 'arrive' and vorkommen 'hap
pen' or French arriver, which has both meanings (cf. Wandruszka, 1981).
33. I count the NPs la famille de Bill...and son père, qui, qui est vraiment trop gros.. as left-
dislocated even though they are not followed by a proposition. For an analysis of the
function of Left Dislocation in spoken French cf. Lambrecht (1981 and 1987), and in par
ticular Barnes (1985).
34. In a recent study dealing with the structure and function of event-reporting sentences
across languages, Sasse (1987) proposes an account of such sentences which is parallel in
many ways to the one I develop here but which differs from mine in one important
respect. Sasse notices certain problems inherent in previous treatments (e.g. Kuno, 1972;
Schmerling, 1976; Wehr, 1984) in which the difference between topic-comment (or
categorical) sentences and event-reporting (or thetic) sentences is assumed to be directly
derivable from facts of information structure (roughly, my identifiability and activation
parameters). He sets out to demonstrate (following essentially Kuroda's approach) that
"the thetic/categorical distinction must be regarded as a linguistic phenomenon in its own
right which cannot directly be derived from and explained in terms of another discourse
strategy. (...) (It reflects) two different points of view from which a state of affairs can
be seen. These are universally reflected in sentence structure in a way as basic to the syntax
of human language as, say, the distinction between declarative, interrogative and im
perative sentences." The main problem I see with Sasse's otherwise highly suggestive ap
proach is that it ignores the formal similarity between thetic (event-reporting) sentences
and presentational sentences, i.e. sentences in which the cognitive status of the NP
referent is a crucial factor. This similarity is particularly compelling in French, but it is
clearly attested in many other languages (cf. the Japanese and Spanish facts discussed
below). Sasse himself observes that all the languages he examines express the distinction
between existential sentences and topic-comment sentences "by means of those gram
matical mechanismes which we found to mark the distinction between thetic and
categorical statements". Since from the discourse-pragmatic point of view 'existential'
and 'presentational' are quasi-identical notions (cf. Section 3.2), a strong formal similari
ty between thetic and presentational cannot be denied. Finally, neither Sasse nor Kuroda
seem to pay much attention to the fact that the presence of formal markers indicating
theticity of a sentence is crucially dependent on the overt presence of a lexical NP, i.e.
on a syntactic expression type which in the unmarked case is reserved for the coding of
non-active discourse referents. I therefore believe that an analysis such as the one I pro
pose here which captures the formal similarity in question by interpreting the event-repor
ting function as an extension of the presentational function is preferrable to an analysis
in which the thetic-categorical contrast is seen as a primitive.
35. I am grateful to Shigeko Okamoto (UC Berkeley) for providing me with these examples.
36. The minimal pair in (26) and (27) is strikingly reminiscent of Schmerling's (1976:420
often cited English minimal pair involving an accented and a non-accented subject NP.
176 KNUD L A M B R E C H T
Schmerling contrasts her example (91) Trûman diéd with (92) Jóhnson diēd, the dif
ference being that the dying event was expected in the context of (91), but totally unex
pected in the context of (92). The event-reporting utterance in (92) is of course analogous
in form and function to our model utterance (4) My car broke down
REFERENCES
Anderson, S.R. and P. Kiparsky (eds). 1973. A Festschrift for Morris Halle.
New York: Holt, Rinehart and Winston.
Akmajian, Adrian. 1973. " T h e role of focus in the interpretation of
anaphoric expressions." In S.R. Anderson and P. Kiparsky (eds).
Bailard, Joëlle. 1981. "A functional approach to subject inversion." Studies
in Language 5,1: 1-22.
Barnes, Betsy. 1985. 'The Pragmatics of Left Detachment in Spoken Stan
dard French. ' ( = Pragmatics and Beyond, VI: 3). Amsterdam: John Ben
jamins.
Blanche-Benveniste, Claire. 1983. "Examen de la notion de subordination."
Recherches sur le Français Parlé 4: 71-115. Aix-en-Provence: Publications
Université de Provence
Bolinger, Dwight. 1954. "English prosodic stress and Spanish sentence
order." Hispania 37: 152-156.
—-. 1984. "Intonational signs of subordination." BLS 10: 401-413.
Bresnan, Joan and Sam A. Mchombo. "Topic, pronoun, and agreement in
Chicheŵa." Language, 63: 744-782.
Chafe, Wallace. 1974. "Language and consciousness." Language 50:
111-133.
— . 1976. "Givenness, contrastiveness, definiteness, subjects, topics and
point of view." In Li (ed), 1976.
-—. 1987. "Cognitive constraints on information flow." In R. Tomlin (ed).
Chomsky, Noam. 1970. "Deep structure, surface structure, and semantic in
terpretation." In R. Jakobson and Shigeo Kawamoto (eds). Reprinted in
N. Chomsky (1972), Studies on Semantics in Generative Grammar,
62-119. The Hague: Mouton.
Clancy, Patricia and Pamela Downing, 1987. "The use of wa as a cohesion
marker in Japanese oral narrative." In John Hinds, Shoichi Iwasaki &
Senko Maynard (eds), Perspectives on Topicalization: The case of
Japanese wa. Amsterdam: John Benjamins.

Clark, Eve. 1970. "Locationals: A study of the relations between 'existen
tial', 'locative' and 'possessive' constructions." Working Papers in
Language Universals, Stanford University.
Cole, Peter (ed). 1981. Radical Pragmatics. New York: Academic Press.
Comrie, Bernard. 1981. Language Universals and Linguistic Typology.
Chicago: The University of Chicago Press.
Damourette, J. and E. Pichon. 1911-1934. Des mots à la pensée. Essai de
grammaire de la langue française. Vol. 4. Paris: Bibliothèque du "français
moderne". Collection des linguistes contemporains.
Deulofeu, José. 1984. "L'étude des langues parlées et la typologie des
langues." Recherche sur le Français Parlé. 5: 103-124.
DuBois, John. 1984. "Competing motivatons." In J. Haiman (ed).
Fillmore, Charles, J. 1979. "Innocence: A second idealization for
linguistics." BLS 5: 63-76.
— . 1985. "Syntactic intrusions and the notion of grammatical construc
tion." BLS 11: 73-86.
Foley, William A. and Robert D. Van Valin Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
François, Denise. 1974. Français parlé. Analyse des unités phoniques et
significatives d'un corpus recueilli dans la région parisienne. 2 vols. Paris:
S.E.L.A.F.
Haiman, John (ed). Iconicity in Syntax. Amsterdam: John Benjamins.
Hawkins, John. 1981. "The semantic diversity of basic grammatical rela
tions in English and German." Linguistische Berichte 75: 1-25.
Hetzron, Robert. 1975. "The presentative movement, or why the ideal word
order is V.S.O.P." In Li (ed), 1975.
Jakobson, R. and S. Kawamoto (eds). 1970. Studies in General and Oriental
Linguistics Presented to Shiro Hattori on the Occassion of his Sixtieth Bir
thday. Tokyo: TEC Co. Ltd.
Jeanjean, Colette. 1979. "Soit y'avait le poisson soit y'avait ce rôti farci.
Etude de la construction il y a dans la syntaxe du français." Recherche sur
le Français Parlé. 2: 121-162.
Kiparsky, Paul. 1976. "Oral poetry: Some linguistic and typological con
siderations." In A. Stolz and R.S. Shannon (eds).
Klavans, Judith L. 1985. "The independence of syntax and phonology in
cliticization." Language 61: 95-120.
Kuno, Susumo. 1972. "Functional Sentence Perspective. A case study from
178 KNUD LAMBRECHT
Japanese and English". Linguistic Inquiry 3: 269-320.

Kuroda, S.-Y. 1972. "The categorical and the thetic judgment. Evidence
from Japanese syntax." Foundations of Language 9: 153-185.
Lambrecht, Knud. 1981. "Topic, Antitopic and Verb-Agreement in Non-
Standard French". ( =Pragmatics and Beyond, II:6) Amsterdam: John
Benjamins.
—-. 1984a. " A pragmatic constraint on lexical subjects in spoken French."
CLS 20: 239-256.
-—. 1984b. "Formulaicity, frame semantics and pragmatics in German
binominal expressions." Language 60: 753-796
-—. 1986a. ''Pragmatically motivated syntax: Presentational cleft construc
tions in spoken French." In Proceedings of the Twenty-second Regional
Meeting of the Chicago Linguistic Society. Papers from the Parasession
on Pragmatics and Grammatical Theory. Chicago, Illinois.
-—. 1986 b. Topic, focus, and the grammar of spoken French. PhD disserta
tion. UC Berkeley.
-—. 1987. "On the status of SVO sentences in French discourse." In R.
Tomlin (ed).
Li, Charles (ed). 1975. Word Order and Word Order Change. Austin:
University of Texas Press.
-—. (ed). 1976. Subject and Topic. New York: Academic Press.
—-. and Sandra Thompson. 1976. "Subject and topic: A new typology of
language." In Li (ed) 1976.
Macaulay, Monica. 1984. Special Field Examination. Ms. University of
California, Berkeley.
Mathesius, Vilem. (1928) 1964. "On the linguistic characterology of modern
English." Actes du Premier Congrès International de Linguistes à la
Haye. Reprinted in Vachek (ed).
Melander, J. 1921. "La locution il y a." Studier i Modem Sprakvetenskap
VIII, (1921): 59-70.
Prince, Ellen. 1981. "Toward a taxonomy of given-new information." In P.
Cole (ed).
Rando, Emily and Donna Jo Napoli. 1978. "Definîtes in There-sentences."
Language 54: 300-313.
Reinhart, Tanya. 1982. "Pragmatics and linguistics: An analysis of sentence
topics." Distributed by IULC.
Rohrer, Christian (ed). 1981. Studia linguistica in honorem Eugenio
Coseriu. Vol. 4, Grammatik. Berlin: De Gruyter.
Sapir, Edward. 1921. Language. New York: Harcourt, Brace and World,
Inc.
Sasse, Hans-Jürgen, 1987. "The thetic/categorical distinction revisited."
Linguistics 25, 511-580.
Schmerling, Susan F. 1976. Aspects of English Sentence Stress. Austin:
Stolz, A. and R.S. Shannon (eds). 1976. Oral Literature and the Formula.
Ann Arbor: University of Michigan.
Stump, Greg. 1980. "An inflectional approach to French clitics." Ohio
State University Working Papers in Linguistics. 26: 1-54.
Thompson, Sandra A. 1978. "Modern English from a typological point of
view: Some implications of the function of word order." Linguistische
Berichte 54: 19-35.
Tomlin, Russell (ed). 1987. Coherence and Grounding in Discourse. Amster
dam: John Benjamins.
Vachek, J. (ed). 1964. A Prague School Reader in Linguistics. Bloomington:
Indiana University Press.
Van Oosten, Jeanne, 1978. "Expletive-there sentences." Ms. UC Berkeley.
Wandruszka, Ulrich. 1981. "Typen romanischer Subjektinversion." In
Rohrer (ed), 1981.
Wehr, Barbara. 1984. Diskursstrategien im Romanischen. Tübingen: Gunter
Narr Verlag.
Zwicky, Arnold. 1977. "On clitics." Distributed by I.U.L.C.
Towards a typology of clause linkage
Christian L e h m a n n
Fakultät für Linguistik und Literaturwissenschaft
Universität Bielefeld
1 BASIC CONCEPTS
The aim of this contribution is to give a survey of the most important

aspects of complex sentence formation in the languages of the world. They
will emerge as generally applicable parameters of clause linkage. I will try to
ascertain to what degree they correlate and how clusterings among the pos
sibilities provided by them yield cross-linguistic types of clause linkage. Most
of these parameters have been found relevant to this issue by other authors
and thus make no claim to originality.
The parameters are based on traditional concepts. Some of these are con
troversial and therefore have to be defined. Let us start with the most general
concepts. A syntagm is relational iff it contains a grammatical slot which pre
determines a grammatical relation to be contracted by it. For instance, an
adverbial phrase is relational because it is by itself capable of modifying some
thing, especially a verb phrase; but a noun phrase is not relational since it does
not, by itself, contract any grammatical relation (see Lehmann, 1985 for
details).
A grammatical relation R connecting syntagms X and Y is a relation of
dependency iff X occupies a grammatical slot of Y or vice versa. In a depen
dency relation, Y depends on X iff X determines the grammatical category of
the complex and thus its external relations. Non-dependency relations among
syntagms are relations of sociation. Among them are coordination, apposition
and others to which we will return. For example, the relation of an attribute
to its head noun is a dependency relation, but the relation of an apposition to
its head noun is not a dependency relation.
An endocentric construction will be conceived semantosyntactically, as a
182 CHRISTIAN LEHMANN
binary syntagm Z with parts X and Y such that Y is the head, belonging to
the same general category as Z (cf. Lehmann 1985: 78f.). The syntactic
relation of X to Y may be one of dependency or of sociation.
In the application of the term clause linkage, we will assume a broad con
cept of the clause which comprises any syntagm containing one predication.
Syntactically, this means that — apart from nominal clauses — the uppermost
controller of dependency in the syntagm is a verbal form. Since a verbal form
may be finite or non-finite, this includes nominalized clauses. Clause linkage,
then, is a relation of dependency or sociation obtaining between clauses in this
sense. In what follows, I will confine myself to the consideration of binary
clause linkage. This should not be understood as excluding the possibility of
more than two clauses being linked at the same level.
Subordination1 may now be conceived as a form of clause linkage.2 If syn-
tagms (clauses) X and Y are in a relation of clause linkage, then X is subordi
nate to Y iff X and Y form an endocentric construction Z with Y as the head.
In the course of the paper, subordination will emerge as a prototypical con
cept.
The term proposition will be used (instead of 'state of affairs') for the
semantic correlate of a (possibly desententialized) clause.
Hypotaxis will be understood as the subordination of a clause in the nar
row sense (which probably includes its finiteness). The definition does not
impose any further syntactic or morphological requirements on the subordi
nate syntagm and thus corresponds fairly well to common usage.3
Embedding is the dependency of a subordinate syntagm.
With these definitions, we get hypotaxis and embedding as two particular
types of subordination, the former delimited with respect to the kind of the
subordinate syntagm, the latter delimited with respect to the kind of the rela
tion of subordination.
Current linguistic usage4 does not treat the term coordination as com
plementary to subordination, since coordination does not imply that the coor
dinated elements be of a sentential or verbal nature. Coordination is a rela
tion of sociation combining two syntagms of the same type and forming a syn
tagm which is again of the same type.
Parataxis is the coordination of clauses. No further restrictions are
imposed on the kind or structural means of coordination. In particular,
parataxis may be syndetic or asyndetic.
We are now ready to enter into the consideration of the various seman-
tosyntactic parameters which are relevant for clause linkage across languages.
TOWARDS A TYPOLOGY OF CLAUSE LINKAGE 183
The following will be considered:5

i. the hierarchical downgrading of the subordinate clause,
ii. the main clause syntactic level of the subordinate clause,
iii. the desententialization of the subordinate clause,
iv. the grammaticalization of the main verb,
v. the interlacing of the two clauses,
vi. the explicitness of the linking.
The reader will notice that the semantic nature of the relation between
the two clauses does not figure in this list. While this has always played a
prominent role in the classification of subordinate clauses, it does not appear
to be constitutive of cross-linguistically valid types of clause linkage. That is,
there is no cross-linguistic notion of, say, the concessive clause which would
possess any constant structural correlates.6 It rather appears that the gram
matical types that will emerge on the basis of the above six parameters cut
across the semantically different clause linkage relations.7
Although these parameters will be grouped in three pairs, they will be
presented separately, as if they were mutually independent. In Section 5.1. we
shall turn to possible correlations among them.
2 AUTONOMY VS. INTEGRATION
The requirement in our definition that a subordinate construction must

be part of a higher construction leaves much room for variation. For one
thing, the subordinate syntagm may bear a sociative or a dependent relation
to the main clause. For another, it may be subordinate either to the main
clause as a whole or to some constituent of the main clause. Neither of these
alternatives is clear-cut; the differences are gradual. The two aspects taken
together identify what I will call the integration of the subordinate construc
tion into the main clause.
2.1 Hierarchical downgrading
2.1.1 Consider the following examples (brackets set off subordinate syn-
tagms):
(1) I was trimming a boomerang, there you came up.
(2) Walbiri
ŋatjulu-lu lpa-na kali tjantu-nu,
I-ERG PAST-SUBJ I boomerang trim-PAST
kutja-0-npa ya-nu-nu njuntu.

[SR-AUX-SUBJ 2 walk-PAST-hither you]
'I was trimming a boomerang when you came up.' (Hale, 1976:79)
(3) Hittite
nu kwit LUGALu-s tezzi nu apat iyami.
[CONN REL:ACC SG INANking-NOM says] CONN D3:ACC SG INAN do:1SG
'And what the king says, that I do.' (KBo XVII 4 II 17f)
(4) Latin
[quei ager ex privato in publicum commutatus est], de eo agro
siremps lex esto.
'Any land that has been converted from private into public, to such
land the law shall apply in the same way.' (CIL I2, 585)
(5) B ambara
tílè tùn ka gòni, ù se-ra bàmakɔ tùma mìn na.
sun:DEF PAST COP hot [they arrive-coMPL Bamako time REL at]
'The sun was hot when they arrived in Bamako.' (Bird, 1968:45)
(6) Kobon
Nipe kaj al-öm (hainö) ribö yaŋ ar-öp.
[3SG pig shoot-3SGss] after river below go-PERF3SG
'He killed a pig and (later) went down to the river.' (Davies, 1981:36)
(7) Latin
L. Petrosidius aquilifer ... pro castris [fortissime pugnans] occiditur.
'L. Petrosidius, the colour-bearer, is killed in front of the camp,
fighting most bravely.' (Caes. B.G. 5, 37, 5)
(8) Latin
Telebois iubet [sententiam ut dicant suam].
'He orders the Teleboans to give their opinion.' (Pl.Am. 205)
These examples differ gradually on a parameter that I shall call hierarchical
downgrading. At the starting pole of the continuum, there is no hierarchical
relation between the two clauses forming the complex sentence. This is the sit
uation which we call parataxis. At the end pole, there is a clear hierarchical
relation between them, the subordinate clause being downgraded to a particu
lar, well-defined constituent within the main clause. This is the situation we
call embedding. Between the poles, there are various constructions in which
the subordinate clause is ever more downgraded.8 (1) shows parataxis. In (2)
from Walbiri (Australia) we find what has been called an adjoined clause
(Hale, 1976). One of the two clauses constituting the complex sentence con
tains a subordinative conjunction and may thus be identified as the subordi
nate clause. It has to either precede or follow the main clause. It cannot be
embedded or have a syntactic function within the main clause. This is the faint
beginning of hierarchical downgrading and of subordination.
(3) from Hittite and (4) show a correlative diptych. As Haudry (1973,
Sect. 3) puts it, this is halfway between parataxis and hypotaxis. The relative
clauses in these examples are subordinate, but not embedded. As they stand,
they could not be embedded. On the other hand, Latin and, to a much lesser
extent, Hittite have the alternative possibility of embedding similar clauses
within the main clause.
In (5) from Bambara (Mande, West Africa) the situation is but slightly
different. The subordinate clause has to occur in final position; however, it is
not outside the main clause, but is rather its last constituent. The word order
rules of Bambara determine sentence-final position for adverbials, simple or
complex. The subordinate clause has the structure of a relative clause, which
usually, but not exclusively, appears in a correlative diptych.
(6) from Kobon (Papua New Guinea; see Davies, 1981) shows clause
chaining. Here the clauses of a complex sentence come in chains which can
grow fairly long (see the Haiman, MacDonald and Myhill and Hibiya con
tributions to this volume). The last verb in the sentence is the main verb. All
the preceding clauses are subordinate to the final one. They and their verbs
are called medial. They lack tense, aspect and mood categories, which are
taken to be those of the final verb, and instead have a special set of person
agreement suffixes which signal whether the subject of the following clause is
the same or different from the subject of this clause. We will return to this
point in Section 4.1.2. Here it suffices to see that although the medial clauses
are subordinate, they cannot be said to be embedded in the final clause. Foley
and Van Valin (1984: ch.6) call this relation 'cosubordination'.
Again, there is a small difference between this construction and the Latin
conjunct participle of (7). This is clearly part of the main clause and insofar
embedded in it. However, its syntactic function is not crystal-clear (cf. Pinks
ter, 1984: ch.8). It is a blend, as it were, of an apposition and an adverbial,
and thus not subject to government.
(8) finally shows a typical embedded clause, namely an object clause gov-
erned by the main verb. We may, of course, find examples of more deeply
embedded clauses; but the hierarchical unequality of the two clauses cannot,
in principle, become clearer than this.
However, within the constructions in which a verb governs a subordinate
clause, differences which appear to relate to the degree of downgrading of the
latter are determined by the nature of the governing verb. In particular, so-
called implicative verbs such as 'force' appear to downgrade the subordinate
clause more strongly than non-implicative verbs such as 'believe'. This issue
has been much investigated recently under the guiding question: What are the
factors determining the form of a subordinate clause, and in particular, how is
it determined by the class of the governing verb? (For some recent research,
see Givón, 1980; Dixon, 1984; Bolkestein, 1985; and Carvalho, 1985.) The
issue is hard to divorce from the control properties of the governing verb,
which codetermine argument sharing among the two clauses. This topic will
be taken up in the latter connection in Sect. 4.1.2.
2.1.2 An important aspect of the integration of the subordinate syntagm into

the matrix clause which has to be accounted for in terms of its hierarchical
downgrading is the sequential position of the former relative to the latter.
Authors have tended to think of positional variability as an important prop
erty of a subordinate clause (cf. Matthiessen and Thompson, this vol., Sect.
4.3.4). It should be emphasized that the position of a subordinate clause as
against its main clause is generally more subject to grammatical constraints
than the position of one independent clause as against its neighbour. While
the two independent clauses in (1) may not seem invertible without damage to
the sense, grammar is certainly indifferent as to any order of them. Subordi
nate clauses, however, often find their position subject to grammatical con
straints (cf. (13) and (14) below and Lehmann, 1982[T]: ch.IV.3.3).
The subordinate syntagm may be either positionally included in the main
clause, or it may precede or follow the latter. Accordingly, we speak of central
vs. marginal position of the subordinate syntagm (cf. Longacre and Thompson
(1985)). Many languages possess the constraint that subordinate clauses of a
certain type have to have a marginal position with respect to the main clause.
This is not only true of the adjoined and correlative clauses that we saw in (2)
through (4), but also of subordinate clauses with a more downgraded status.
Thus in Hixkaryana (Carib, Brasil) and Abkhaz (North-West Caucasian),
subordinate clauses are lower on the hierarchy, but yet are generally margi
nal, in Abkhaz mainly preposed, to the main clause.
Quite generally, the marginal position of a subordinate clause is in accor

dance with the principles of functional sentence perspective. Just as
elsewhere, sentence-initial position usually identifies the topic (more pre
cisely, the exposition, in the terms of Lehmann, 1984: ch.V.5) of the sen
tence. This is well-known from left-dislocated NPs. It is perhaps not so well
known that a whole subordinate clause may also provide a topic for the fol
lowing main clause.9 Thompson (1984) shows that English purpose clauses ful
fill different discourse functions as they either precede or follow the main
clause. Consider her introductory example, a passage taken from a report on
a voyage in a primitive boat named Brendan:
(9) Brendan was rushing farther and farther out to sea. [To slow her
down] we streamed a heavy rope in a loop from the stern and let it
trail in the water behind us to act as a brake, and, hopefully, to
smooth the worst of the wave crests. From the stern also dangled a
metal bucket; only twenty-four hours earlier we had been using it
[to cook an excellent meal of Irish crabs]. Now it clanked mourn
fully every time a wave broke against it. (Thompson, 1985, Sect. 5.1)
The preposed purpose clause formulates a problem which has been expected
on the basis of the text preceding it. This problem requires a solution which is
provided by the following main clause. An initial purpose clause thus acts like
other topics in that it "provides a framework within which the main clause can
be interpreted, and that it does this by its role as a link in an expectation
chain." (Thompson, 1985, Sect. 5.1) A final purpose clause, such as the
second one in (9), does not do this, but merely appends a purpose locally
relevant with respect to some semantic unit in the preceding main clause.
Cf. also Chafe 1984.
We have seen one Latin example of this discourse function of left-margi
nal subordinate clauses in (4). (10) is another one.
(10) Latin
[Calcem partiario coquendam qui dant], ita datur:
"If lime is given for burning by shares, it is given thus:"
(Cat.agr. 16)
The relative clause ought to identify a referent to be taken up in the main
clause. Here it does not do this, but rather depicts a situation which functions
as a framework for the main clause. Thus, the topic function here suspends
the semantic function of the subordinate clause.
A subordinate construction whose topic-forming potential is not yet fully
explored is the absolute construction as exemplified in (11) and (12) (cf. Hoff,
1985 for the Latin ablativus absolutus).
(11) Latin
[Cognito Caesaris adventu] Ariovistus legatos ad eum mittit.
'When he learnt about Caesar's arrival, Ariovistus sent envoys to
him.' (Caes.B.G.1,42,1)
(12) Italian
[(Dopo) chiesto e pagato il conto], l'avventore pensa a andarsene.
Lit.: '(After) asked and paid the bill, the customer thinks to go
away.'
Here the introductory subordinate construction again formulates a state of
affairs coherent with an expectation chain formed on the basis of the preced
ing text; and this state of affairs provides the background for the action of the
main clause.
Another subordinate clause whose topical function has been investigated
is the conditional clause (cf. Haiman, 1978). We may conclude that quite gen
erally subordinate constructions at the left margin of the higher clause have
topical function. This apparently holds irrespective of the order freedom
enjoyed by the specific subordinate construction. It is true of the Hittite rela
tive clause, which is almost invariably preposed, and of adjoined and correla
tive clauses in general, which must have marginal position; but it is true as
well of English purpose clauses and the Latin ablativus absolutus, whose order
is freer.
Certain conjunctions may force a fixed position of the subordinate clause
introduced by them.
(13) German
Das verstehe ich nicht, zumal die Reise noch so billig war.
'That I don't understand, especially as the trip was so cheap.'
(14) German
Fliegen ist viel schöner als Autofahren, nur daß man noch mehr
aufpassen muß.
'Flying is much nicer than car driving, except that one has to pay
even more attention.'
German subordinate clauses introduced by zumal 'all the more so as' and nur
daß 'except that' have to follow their main clause, as they do in (13) und (14),
and cannot precede it.
Another kind of positional constraint can be seen in the medial clause of

Kobon (6). The essential point in all the examples is the following: the fact
that rules of grammar may determine the position of the subordinate syntagm
in or vis-à-vis the matrix is an aspect of their integration which follows from
their hierarchical downgrading.
2.1.3 We finally return to the gradience illustrated by (1) through to (8). The
idea of this continuum also underlies the 'typology of nexus' put forward in
Foley and Van Valin (1984: ch.6), although the details differ. The continuum
of hierarchical downgrading is represented in Figure 1:
parataxis < > embedding

independent adjoined correlative medial conj. governed
clauses clause diptych clause part, clause
Figure 1: Hierarchical downgrading
Hierarchical downgrading is not only an important parameter in the typology

of clause linkage, but is also a central criterion for the traditional notion of
subordination; a clause not affected by it is not called subordinate.
2.2 Syntactic level
2.2.1 A second aspect of the integration of the subordinate clause into the
main clause, closely related to its hierarchical downgrading, is the variation
according to the main clause syntactic level which the subordinate clause
belongs to. The first to use this parameter for a typology of complex sentences
was T. Milewski (1954). He distinguished three syntactic levels: the sentence
level (above the simple clause), the clause level (within the clause), and the
verb level. This hierarchy is mirrored in Foley and Van Valin's (1984: ch.6)
typology of clause linkage according to their three levels of juncture, viz.
periphery, core and nucleus. I do not accept such a threefold division but
rather assume a multiplicity of syntactic levels between the morpheme and the
paragraph, much as in constituent structure grammar.
The guiding idea is that the lower this level, the more tightly the subordi
nate clause is integrated into the main clause. We may first reconsider the
examples already given. In (1) there are only independent clauses; they are
thus on the text level. The subordinate clauses in (2) through (4) do not form
part of the main clause, but are on the same syntactic level as the latter,
namely the sentence level. The subordinate clause in (6) occupies an inter
mediate position between being outside and inside the main clause (this adds
up to the reasons why it is called medial). The adverbial clause in (5) is clearly
part of the main clause, but on its highest syntactic level (immediately domi
nated by S). The conjunct participle of (7) is on some level between clause
and VP. Finally, the subordinate clause in (8) is on the VP level.
Apart from the switch in the order of (5) and (6) this looks like a neat
correlation between hierarchical downgrading and syntactic level. However, a
moment of reflection will show that this need not be so. There can be subordi
nate predications on very low syntactic levels which are far from being gov
erned by a main clause constituent. Look at (15).
(15) Tarquinio vero quid impudentius, [qui bellum gereret cum iis [qui
eius non tulerant superbiam]]?
'Again, what is more impudent than Tarquinius, who waged war
on those who could not bear his arrogance?' (Cic.Tu.3,27)
Here the first relative clause is appositive, which gives it a place somewhere in
the first half of the continuum of hierarchical downgrading. The second rela
tive clause is restrictive, thus dependent on the head noun, but still not governed
by anything and therefore not at the rightmost pole of Figure 1. However, the
first relative clause is on a fairly low syntactic level, modifying an NP gov
erned by the predicate. The second relative clause is on an even lower level,
if we determine levels by counting nodes from the root of a constituent struc
ture diagram. If we don't do this, but just count the distance from the nearest
upper S, the second relative clause is still on a lower level than any subordi
nate clause in the examples up to (8).
Low syntactic level of a not maximally subordinate syntagm can also be
seen in (16).
(16) Kobon
Nipe wañib si ud ar-öp.
3SG [string bag illicitly take] go-PERF3SG
'He stole the string bag.' (Davies, 1981:203)
The phenomenon illustrated by this example is called verb serialization.
This in general involves the combination of verbals to complex verbals with
out the intervention of any connectives which might make explicit the relation
among them.10 Usually only the last verb in the series carries the finite inflec
tion, whereas the preceding verbs may be uninflected, as in the above exam-
ple. This device is operative both in a syntactically regular way, as in (16), and
in verb composition, with various intermediate degrees of lexicalization. What
is presently of interest to us is the hierarchical relation obtaining in verb series
and the syntactic level on which they abide. While detailed investigation is still
pending, this much at least appears to be certain: The preceding verbal is not
governed by the following one (or vice versa). It is not even clear that the
former depends on the latter; they might be coordinate (or cosubordinate, as
Foley and Van Valin, 1984:261f would have it). On the other hand, the main
clause syntactic level on which the preceding verb is adjoined is certainly not
above the VP.
From these examples we may conclude that although advanced hierarchi
cal downgrading of the subordinate clause implies a low syntactic level for it,
the converse does not hold. Thus the relation between the continua of hierar
chical downgrading and of syntactic level is one of unilateral implication.
Serial verb constructions are in a clear grammaticalization relationship
with clause chains. The concomitance of the two clause linkage types, here
represented by (6) and (16) for Kobon, therefore recurs in totally unrelated
languages. See, e.g., Todd (1975) for Choctaw.
2.2.2 Up to now the lowest syntactic level illustrated by our examples has
been that of the VP. (17) to (20) are meant to show that the various processes
of subordination are also operative at still lower levels.
(17) Latin
Cato [hoc dicere] solebat.
'Cato used to say this.'
(18) I will [go] to bed now.
(19) Italian
Ho fatto prendere a mio figlio un altra professione.
have:isG made [take:INF to my son an other profession]
T had my son choose another profession.'
(20) Quechua
Juzi-ka Juan-ta ruwana-ta awa-chi-rka.
Jose-Top Juan-Acc poncho-ACC [weave]-CAUS-PAST(3SG)
'Jose had Juan weave a poncho.' (Cole, 1982:135)
The verbs on which the subordinate constructions depend here are operators
forming complex predicates. Lat. solere as in (17) may be regarded as forming
a periphrastic habitual aspect by governing an infinitival. Engl, will as in (18)

is an auxiliary which takes an uninflected verb (an infinitive without to) as its
complement. So here we are down to the level of the verbal or even the
(analytic) verb form.
A causative construction may have the appearance of a complex sen
tence, with the verb of causation functioning as the main clause verb and the
caused event being expressed by a subordinate clause governed by the verb of
causation. The Italian causative construction illustrated in (19) comes close to
this characterization, except that the subordinate clause is reduced to an
infinitival. It also shows that the subordinate predicate has a tendency to
attach directly to the verb of causation. This process is further advanced in
(20) from Imbabura Quechua. The verb of causation here is reduced to a ver
bal suffix. The syntactic level we are dealing with is thus that of the verb.
2.2.3 We are now in a position to formulate the continuum of the syntactic

level:
sentence < > word

subordinate clause is complex predicate formation
outside at margin inside inside verb auxiliary verbal
main of main main VP serial- peri- deriva-
clause clause clause ization phrasis tion
Figure 2: Syntactic level
Although this parameter reveals important differences among types of clause

linkage, it does not seem to be crucial for the traditional notion of subordina
tion.
2.3 Before we conclude the Section on integration, we should at least men

tion intonation as an important factor (cf. Bolinger, 1984). A clause may be
downgraded by low pitch and may be integrated with another clause by the
absence of an intonation break between them.
3 EXPANSION VS. REDUCTION
3.1 Desententialization of subordinate clause
3.1.1 The third of the parameters that structure a typology of clause linkage
is the degree to which the subordinate clause is expanded or reduced (cf.
Lehmann, 1982[N] for the following). In the reduction process, it loses the
properties of a clause, it is desententialized to varying degrees. Components of
the clause which allow reference to a specific state of affairs are dropped; the
state of affairs is 'typified'.11 At the same time, the subordinate clause increas
ingly acquires nominal properties, both internally and in its distribution. At
the end of this process of nominalization, the clause becomes a nominal or
adverbial constituent of a matrix clause. In Sections 3.1.2 and 3.1.3, we will
deal with the internal and external aspects of desententialization, respectively.
3.1.2 . There is a variety of semantic components and categories with their

grammatical correlates which make up a full-fledged sentence. Let me enum
erate here illocutionary force, mood, tense, aspect, actants and circumstants
with their various syntactic functions. We observe, first of all, that a subordi
nate clause may not normally have its own illocutionary force. Certain parti
cles bound up with illocutionary force are not admitted in subordinate clauses.
We may mention the Quechua validator -mi and the Latin assertive particle
ne. This is a consequence of the fact that "illocutionary force is the outermost
peripheral operator", as Foley and Van Valin (1984:239) put it, which means
that one sentence, however complex, may have only one illocutionary force.
Look at the following examples:
(21) He believed [(that) I wrote a letter].
(22) Latin
Non temere est [quod corvus cantai mihi nunc ab laeva manu].
'It is not by chance that the raven now croaks at me from the left
side.' (Pl.Au.624)
The structure of the noun clauses in both of these sentences differs minimally
from that of a corresponding independent clause. (The same goes, inciden
tally, for ((2) — (4).) The introductory subordinating conjunction is even
optional in (21). However, even with such a low degree of desententialization,
one thing is missing, viz. the illocutionary force that the corresponding inde
pendent sentences would have. The same can be seen in all the examples so
far given except for (1).

Often the assertive force of the subordinate clause depends on the sen
tence type and the polarity of the main clause. Observe (23).
(23) a. You stole the caviar [because you were hungry].
b. Did you steal the caviar [because you were hungry]?
c. You didn't steal the caviar [because you were hungry].
In (23a) the causal clause is understood to be asserted, while in b and c it is
not. One way of putting this would be to say that certain bits of meaning are
not expressed in the subordinate clause because they are determined for the
whole sentence by the main clause.
A few classes of subordinate clauses are, however, allowed to have their
own illocutionary force (cf. Green 1976; Lakoff 1984). Among them is the
non-restrictive relative clause:
(24) Latin
ad Italiani accedei, [in qua nos sedentes quid erimus]?
'He will approach Italy, where we will be sitting and doing what?'
(Cic. Att. 10,8,4)
(25) Latin
Perutiles Xenophontis libri sunt, [quos legite, quaeso, studiose]!
'Highly useful are Xenophon's books; please read them
thoroughly!' (Cic.Cat.M.59)
The semantosyntax of the clauses represented in (24) and (25) is similar to
that of parentheses. The relation linking them to a main clause constituent is
not one of dependency, but of sociation, which means they are close to the left
pole of the continuum of hierarchical downgrading. I predict that subordinate
clauses with an illocutionary force of their own will become rarer as we
approach the right pole of this continuum.
A further consequence of the desententialization of the subordinate
clause are constraints on its mood. It suffices to mention some relevant facts.
In Latin the choice of mood is constrained by syntactic rules in most kinds of
subordinate clauses. The subordinate clauses of (10) and (22) show an obliga
tory indicative, the one of (8) an obligatory subjunctive. In several languages,
among them Akkadian and Abkhaz, all finite subordinate clauses are in the
subjunctive. We understand that such restrictions are intimately connected
with the lack of illocutionary force in subordinate clauses. As soon as it comes
to nonfinite subordinate constructions, there is no longer any mood. Relevant
examples are (7) and ((17) — (20)).
Another phenomenon closely associated with the lack of illocutionary

force is the reduction of word order freedom inside the subordinate clause.
This mirrors a gradual loss of freedom in theme-rheme assignment and in the
applicability of syntactic processes in general. For example, word order in the
subordinate clauses of ((2) — (4)) is essentially determined by functional sen
tence perspective, while in those of (6) and (7), as well as (26) below, there is
virtually no functional sentence perspective, and consequently word order
essentially obeys rules of syntax.12
The next verbal categories that get reduced and finally lost in desententializa-
tion are tense and aspect. We saw that in Kobon medial clauses and nonfinal
serial verbs tense and aspect are missing and are understood to be those of the
final clause or verb, respectively.
(26) Turkish
Evvel-â çik-an güçlük-ler-in
[start-Loc emerge-ACT PTCPL trouble-PL-GEN
kendi-lig-in-den orta-dan kalk-tig-i gör-ül-ür.
self-Ø-poss 3-ABL middle-ABL disappear-NR-poss 3] see-PASS-HABIT
'It turns out that the problems occurring at first disappear
completely by themselves.' (Wendt, 1972:187)
In the Turkish noun clause illustrated in (26), the tense-aspect system is
reduced to a binary opposition between non-future (suffix -dig-) and future
(-eceg). This opposition is bound up with voice in that the -dig-form cannot
function as an active participle, as can the -eceg- form. Similarly, the Latin
conjunct participle shown in (7) allows only of a tense opposition between
simultaneous (present) and subsequent (future) action, apart from previous
(perfect) action, which is bound up with passive voice. The same goes for the
ablativus absolutus (11). In the Italian absolute construction of (12), there is
no choice of tense, aspect or voice whatsoever. Similar observations apply to
the dependent verbs in ((17) — (20)).
Finally, desententialization affects the relationality of the predicate verb
(cf. Dik, 1985, Sect. 2 and Mackenzie, 1985). The predication expressed in an
independent clause, and also in such relatively weakly desententialized subor
dinate clauses as those of ((2) — (6)) as well as several of the other (finite)
examples, gets lost, and simultaneously the subject slot of the subordinate
verb is either converted into an oblique slot or is entirely lost. The verb
becomes nonfinite. For verbs which conjugate for person when finite, this
involves loss of personal inflection. A relevant example is (26). When there is
an NP that would be the subject of the corresponding finite verb, this cannot
be in the nominative, but has to be in various oblique cases. Very often this is
the genitive. This can again be seen clearly in (26). The possessive suffix on
the subordinate verb here agrees with the syntactic possessor (corresponding
to the subject). Compare the situation in English:
(27) I heard [the man's/his singing].
(28) illustrates a verb that does not in principle inflect for person.
(28) Japanese
Ano hito-gal no hon-o kai-ta koto-ga yoku sirarete iru.
[D3 person-NOM/GEN book-ACC write-PAST SR-NOM] well known COP
Tt is well known that that person wrote a book.'
The subject of the subordinate clause can be either in the nominative or
in the genitive, this being the only reflex of the verb's finiteness and nonfinite-
ness, respectively. Converse to this is the case of the non-final serial verb in
Kobon (16), since this loses its personal inflection without, apparently, losing
its ability to take a subject.
In Latin, the genitive of the semantic subject is known as the genitivus
subjectivus accompanying verbal nouns. (29) illustrates this {animi vel cor
poris) as well as the genitivus objectivus {gravioris operis et muneris) to be
treated below.
(29) Latin
labor est [functio quaedam vel animi vel corporis gravions
operis et muneris].
'Labor is the performance of a relatively hard compulsory work by
the mind or the body.' (Cic.Tu.2,35)
What was the subject of the finite verb may also appear in the accusative when
the verb becomes nonfinite. This may happen to the causee in a causative con
struction, as in (20). It also happens in the accusativus cum infinitivo, as illus
trated in ((30) — (32)).
(30) Quechua
alku-ta kri-ni aycha-ta shuwa-ju-j-ta.
dog-Acc believe-ISG [meat-ACC steal-PROG-ACR PTCPL-ACC]
T believe the dog to be stealing the meat.' (Cole, 1982:35)
(31) I heard [the man/him singing].
(32) Latin
[Liberos suos ... beatos esse] cupiat.
'He may wish his children to be happy.' (Cic.inv.1,48)
Finally, the possibility of the semantic subject going into the dative may be
mentioned; this is illustrated in the causative construction of (19).
The subject is the first of the verbal actants to be affected by desenten-
tialization. On stronger nominalization, the whole verbal government is
destroyed, so that NPs corresponding to the actants of the finite verb appear
in the genitive or in adpositional phrases. We have seen the Latin genitivus
objectivus in (29). (33) shows clearly that the loss of object government goes
hand in hand with other symptoms of nominalization.
(33) a. She objected to [his constantly reading magazines].
b. She objected to [his constant reading of magazines].
The less strongly nominalized verb of (33a) takes a direct object and an
adverb; the more strongly nominalized one of (33b) does not take a direct
object and is modified by an adjective, just like any noun. (For further
crosslinguistic discussion of nominalizations in this connection, see Comrie
and Thompson, 1985.)
Complete loss of the subject slot of the subordinate verb can be seen in
(17) and (18). Similarly, while every finite verb in English needs to have a sub
ject, the nominalization illustrated in (33) can do without the possessor phrase
corresponding to the subject. Milewski (1954) observes that where the clause
is constructed according to the concentric strategy, the NPs which function as
appositive extensions of the pronominal verbal affixes are impossible in subor
dinate clauses. On the whole, the more a verb gets nominalized, the more it
starts behaving like an ordinary noun. It is in this sense that we may speak of
the increasing nominality (or 'nouniness') of subordinate clauses, when they
are reduced by desententialization.
In the present connection, we should note that such a strongly
nominalized construction as the ablativus absolutus is usually reduced to two
main constituents corresponding to the subject and the predicate of a full
clause. It is only in the literary style of Classical Latin that the absolute con
struction may be expanded by adjuncts and even subordinate clauses (s. Cole-
man, 1985). This is typologically highly marked. (Cf. Thompson, 1985, Sect. 1
on certain peculiarities of written language.)
At some stage of strong desententialization, the polarity of the subordi
nate clause is also affected. This usually means that it can no longer be inde
pendently negated. Thus, we cannot negate the participle depending on the
English auxiliaries, nor the infinitive depending on most of the modals such as
must and may. In Jacaltec, nonfinite subordinate predications cannot be
negated (cf. Craig, 1977:242f and Foley and Van Valin, 1984:287f).
(34) Jacaltec
a. ç-w-acoj yiŋ hin c'ul chubil ch-in to-yi.
in POSS ISG stomach [that PRES-ISG go-AUG]
PRES-ERG ISG-carry
T am thinking of going.'
b. ç-w-acoj yiŋ hin c'ul chubil mach ch-in to-yi.
PRES-ERG isG-carry in POSS ISG stomach [that NEGPREs-ISGgo-AUG]
T am thinking of not going.'
c. ç-w-acoj yiŋ hin c'ul hin to-yi.
PRES-ERGISG-carryin POSS ISG stomach [POSS ISG go-AUG]
T am thinking of going.' (Craig, 1977:242)
(34a) shows a finite subordinate clause, introduced by a subordinator, whose
negative counterpart is (34b). In (34c) we see a nonfinite version of (34a).
There is no way of negating this subordinate construction.
3.1.3 In the preceding, I have confined myself to those processes of desen-

tentialization which are operative inside the subordinate clause. We should
note that increasing nominality of a subordinate construction also endows it
with the distributional properties of a noun or NP. One such property that
deserves special mention is the combinability with adpositions and case
affixes. The rule is that the more a subordinate clause is nominalized, the
more easily it combines with adpositions and case affixes. There is but one
primary preposition which can govern the that-clause as shown in (21), namely
in. In addition, there are some prepositions such as before, after, since, till and
because, which may govern a finite subordinate clause lacking the introduc
tory subordinator. We then usually say that the preposition functions as a con
junction. When it comes to more strongly desententialized subordinate con
structions such as the gerund, many more prepositions such as by, upon,
towards etc. become possible.
The Japanese subject clause in (28) takes the appropriate nominative
case suffix. Similarly, the Quechua object clause in (30) takes the accusative
suffix.
No finite subordinate clause in Latin may be governed by a preposition,
let alone take a case suffix. The preposed relative clause in (4) cannot directly
be governed by the preposition de. Instead, it must be reinstantiated as an
anaphoric pronoun, and in this form it can combine with the preposition. It is
only nonfinite constructions such as the gerund in (35) which may be gov
erned by a preposition.
(35) Latin
palus ... Romanos [ad insequendum] tardabat.
'A swamp detained the Romans in the pursuit.' (Caes. B.G. 7, 26,
2)
In (8) the subordinate clause is the object of the main verb; but it is not
in the accusative, nor is there any way it could be. Even the nonfinite object
clause of (32), directly comparable to the Quechua construction of (30), is not
in the accusative. I will return in a moment to the accusatives in (32). Simi
larly, the relative clauses in (15), (24) and (25) are attributes to nouns in dif
ferent cases; but they do not agree in case with their head nouns, as adjective
attributes do (see Lehmann, 1984:187f for the general possibility of such
agreement).
Instead of combining a whole subordinate construction with a case suffix
(or a whole subordinate clause with a preposition), Latin chooses an appropri
ate constituent of the subordinate construction as a representative of the
whole and has this carry the case suffix (cf. Lehmann, 1979). In (11) it is not
the subordinate clause as a whole which shows the ablative (assuming that this
is the appropriate adverbial case), but rather its two main constituents, the NP
and the participle which correspond to the subject and predicate of a finite
construction. Similarly in (32), the nouns corresponding to the subject and the
predicate nominal are in the accusative, vicariously for the subordinate clause
as a whole.
In finite subordinate clauses, the solution is to show case on the subor
dinato. Many of the subordinative conjunctions, such as quod, cum, qua, ut
etc. are old case forms of the relative pronoun which functioned as a general
subordinator. All these are consequences of the fact that a language of the
type Latin belongs to must show syntactic relations on the word.
3.1.4 We are now ready to sum up our observations on the inner and outer
processes resulting from desententiahzation in a continuum which appears in
Figure 3 (cf. Lehmann, 1982[N]:76).
Strictly speaking, constructions such as the ablativus absolutus or the
Romance gerunds in -ndo are adverbial rather than nominal in nature. I will
not here consider the problem of whether all adverbials can be analyzed as
nomináis in a certain semantic case and merely note that the term 'nominality'
used above should be understood as including adverbiality.
Much of the typological importance of the parameter of desententiahza
tion lies in the following fact: to the degree that the subordinate construction
sententiality < > nominality

clause nonfinite construction verbal noun
no illocutionary force
constraints on illocutionary elements
constraints on/loss of modal elements and mood
constraints on/loss of tense and aspect
dispensability of complements
loss of personal conjugation
conversion of subject into oblique slot
no polarity
conversion of verbal into
nominal government
dispensability of
subject
constraints
on complements
combinable with adposition / agglutinative case affix / flexive case affix
Figure 3: Desententialization
is desententialized, the whole sentence ceases to be genuinely complex. Look

ing at things from the point of view of expansion, we may say that the mere
insertion of a verbal noun as a constituent of a sentence does not yet make this
sentence syntactically complex. Cf. Carvalho, 1985; Helander, 1985 and (36).
(36) Latin
a. Vicinus assum furatus est [quia pauper est].
'The neighbour stole the roast because he is poor.'
b. Vicinus assum furatus est [propter paupertatem].
'The neighbour stole the roast because of his poverty.'
(36a) is a complex sentence, while (36b) is not. This is a syntactic statement.
It is quite a different question, which we will not address here, whether the
latter sentence is easier to process than the former. Experience with languages
such as Turkish and Japanese, whose subordinate clauses are more desenten
tialized than standard Indo-European ones, would not appear to invite the
conclusion that sentences containing less desententialized subordinate clauses
are ceteris paribus psychologically less complex.
3.2 Grammaticalization of main verb
3.2.1 In the preceding section we saw that a subordinate clause may be

reduced to its core, the predicate. In the following we will see that there is a
reduction process in the main clause, too. However, the kind of desententiali-
zation which leads to nominalization does not occur in main clauses, since it
presupposes subordination. What we find, instead, is the grammaticalization
of the word governing the subordinate clause.
Grammaticalization is a diachronic process and a synchronic continuum
which lead from lexical to grammatical items (see Lehmann, 1982[T] for
details). Among other things, it turns full verbs into modals and auxiliaries.
Since what is grammaticalized are essentially words, the approach of the pre-
sent section presupposes that the subordinate clause have a grammatical rela-
tion to just one word in the matrix clause. There is a variety of subordinate
constructions we might look at under this premiss (cf., e.g., Rosen, 1985 for
cleft sentences). However, the grammaticalization of the superordinate word
leads to special construction types only when that word is a verb. The other
interesting case is when the subordinate clause is governed by an adposition.
Adpositions grammaticalize to case affixes. The combinability of subordinate
clauses with adpositions and case affixes has been dealt with briefly in Section
3.1.3.
3.2.2 We can start the illustration of the grammaticalization of the superordi-

nate predicate with causative constructions.
(37) Latin
sol efficit [ut omnia floreant]
'the sun makes everything blossom' (Cic.n.d.2,41)
(19) Italian
Ho fatto prendere a mio figlio un'altra professione.
have:1sG made [take:INF to my son an other profession]
T had my son choose another profession.'
(20) Quechua
Juzi-ka Juan-ta ruwana-ta awa-chi-rka.
Jose-Top Juan-Acc poncho-ACC [weave]-CAUS-PAST(3SG)
'Jose had Juan weave a poncho.' (Cole, 1982:135)
(19) and (20), which are repeated here for convenience, belong to a series
with (37). In this example, the verb of causation is a full lexical verb which
governs a finite subordinate clause. There is little that is special about this
verb, except that the class of verbs which govern an ut-clause is a closed one.
The causative verb of (19) is much more grammaticalized. It combines directly
with the subordinate verb to yield an analytic causative verb. The valency of
the latter derives in a regular way from the valency of the dependent verb plus
an additional causer-subject introduced by the verb of causation. In (20), the
verb of causation is grammaticalized to a suffix on the semantically subordi-
nate verb. The latter thus becomes a causative verb. The valency of the under-
lying simple verb is again increased by the causer-subject, the former subject
being downgraded to a causee-object.
Indo-European languages such as English, German and Latin lack the
grammatical means of forming causative verbs in such regular ways. Instead,
the formation of causative verbs tends to be a matter of the lexicon. Latin has
the type illustrated in (38).
(38) Latin
Miles picem fervefecit.
The soldier boiled the pitch.'
The underlying simple verb in (38) is fervere 'to glow'. The formation of
causative verbs by composition with facere 'do, make' never was very produc-
tive. It was exclusively based on intransitive verbs, but extended to such biva-
lent verbs as assuescere 'to get used to', whose ablative, dative or preposi-
tional complement remains totally unaffected by the derivation of assuefacere
'to accustom to'. On the other hand, it is certainly no accident that the histor-
ically identical Italian verb fare forms the totally productive analytic causa-
tives seen in (19). Similar remarks apply to the totally unproductive German
and English causatives of the type fallen/fall — fallen/fell.
However, both the grammatical causatives of (19) and (20) and the lexi-
cal causatives of (38) etc. prove the point which is essential here: to the
degree that the main clause predicate gets grammaticalized, the whole sen-
tence ceases to be syntactically complex (cf. Brettschneider, 1984, Sect. 5;
Drossard, 1984, Sect. 4). In (20) and (38), there is only one clause, the latter
being uncontroversially a simple one.
3.2.3 For the sake of variation, let us look at a second series of examples.13
The following are desiderative constructions.
(39) Latin
op to [ut in hoc iudicio nemo improbus ... reperiatur]
'I wish that in this trial nobody will be found bad.'
(Cic. Verr.I, 50)
(40) French
Je veux [aller au cinema].
T want to go to the movies.'
(41) Sanskrit
a. dēva-h soma-m piba-ti.
god-NOM SG MASCsoma-ACCSG MASC PREs:drink-3SG
'The god drinks soma.'
b. dēva-h soma-m pi-pā-sa-ti.
gOd-NOM SG MASC SOma-ACC SG MASC RED-drink-DESID-3SG
'The god wants to drink soma.'
In (39) we have a lexical verb of wishing, belonging to a class of verbs
which take an object clause in a certain syntactic form. The verb of wishing in
(40) is a modal verb. Its advanced stage of grammaticalization appears, among
other things, from the fact that it takes the subordinate verb in the pure
infinitival form, without an intervening preposition such as de. In (41) we see
desiderative derivation in Sanskrit. Here the morpheme signifying the wish is
fully grammaticalized to a verbal suffix appended to the semantically subordi
nate verb. The result is one derived verb.
Just as in the causative constructions, we start from a complex sentence
whose main verb has a certain subordinative potential, through which it gov
erns a complement clause. We end up with a simple clause whose main verb
is the former subordinate verb, carrying the former main verb as an affix.
Gradually the main verb gets interpreted as a functive operating on the subor
dinate clause, specifically its verb, turning this into a derived verb.
3.2.4 Although the degree of grammaticalization of the superordinate verb

constitutes important differences among types of clause linkage, it appears to
play a largely negative role in traditional accounts of subordination. Namely,
constructions with a highly grammaticalized superordinate verb generally
have been neglected altogether in this connection. The continuum is
schematized in Figure 4.
independent predicate grammatical operator

lexical evidential modal auxiliary derivational/
verb verb verb grammatical
affix
Figure 4: Grammaticalization of main verb
3.3 The conclusion is, of course, that there are two ways of reducing a com-
plex sentence to a simple one (and conversely, two ways of expanding a clause
to a complex sentence). First, we may desententialize the subordinate clause,
turning it into a simple constituent of the main clause. Second, we may gram-
maticalize the governing verb, turning it into an affix which modifies the
meaning of the semantically subordinate verb. In both cases, the subordinate
verb becomes a constituent of the main clause: in the first case, a dependent
one, in the second case, its main verb.
4 ISOLATION VS. LINKAGE
4.1 Interlacing
4.1.1 Another parameter structuring subordination is the degree to which

the linked propositions are interlaced.14 The semantic aspect of interlacing is
that the two propositions share some elements of their meanings. Its syntactic
correlate are the non-specification of the common elements in one of the
propositions and/or the syntagmatic interweaving of the two propositions. Cf.
Davison, 1979, Sect. 4.1. and Foley and Van Valin, 1984: ch.6.3.
In principle, the linked propositions may share any bit of meaning one
may please to imagine. I will not systematically cover the possibilities here,
but mention just three: the sharing of predicates, of tense and aspect, and of
actants.
The two propositions may share their predicate. When this is expressed
only once, we get a kind of brachylogy known as gapping. Here are two exam-
ples:
(42) Latin
[ut ager ... sine cultura fructuosus esse non potest], sic sine doctrina
animus.
'As a field cannot be fertile without cultivation, so the mind (can-
not be fruitful) without instruction.' (Cic.Tu.2,13)
(43) Latin
magis ea percipimus ..., quae nobis ipsis ... eveniunt, quam illa,
[quae ceteris].
'We perceive more strongly those things which happen to ourselves
than those which (happen) to others.' (Cic.off. 1,30)
In (42) the whole predicate of the main clause except for the privative adjunct
is gapped on identity with the preceding subordinate clause. Conversely in
(43), the predicate of the subordinate clause, except for the benefactive
adjunct, is gapped on identity with the preceding main clause. Gapping works
essentially the same way in parataxis and in hypotaxis. However, there are no
subordinate constructions specialized for the case that a certain predicate, or
a predicate of a certain kind, is shared among the two clauses. This is why
shared predicates are not central to this section.
The second kind of interlacing that I will mention only briefly concerns
tense and aspect. Their interlacing means that the tense and aspect of the sub-
ordinate clause are partly or wholly determined by those of the main clause
(cf. Van Valin 1984 on 'grammatical category dependence'). 15 Partial depen-
dence of the tense of the subordinate clause on that of the main clause already
occurs at a level of weak desententialization in the form of consecutio tem-
porum. The subordinate constructions of (7), (30) and (31), on the one hand,
and of (11) and (12), on the other, signal only simultaneousness with and tem-
poral priority to the main clause, respectively, the tense itself being deter-
mined by the main clause. Similarly, the subordinate verbs of (26) and (34a)
only show that their time is not posterior and not prior, respectively, to main
clause time. Complete dependence of subordinate tense and aspect can be
seen in the Kobon examples (6) and (16), as well as in the infinitives of (17)-
(19).
4.1.2 We now turn to the central issue of the present section, which is the
interlacing of actants of the main and subordinate clauses. I will skip here the
whole issue of relative clauses and just observe that the correlative diptych, as
in (3) and (4), is essentially held together by anaphora, i.e. by the fact that the
two correlative clauses share an actant (or another nominal or adverbial con-
cept). The fact that backwards anaphora is allowed in subordinate clauses but
heavily constrained in main clauses is another instance of the interlacing of
actants that I will just mention.16
Many languages make a distinction in the inflection of a non-final (me-
dial) verb depending on whether the verb of the following clause has the same
or a distinct subject. Kobon does this in finite medial verbs, which have two
oppositive paradigms of personal endings, one signalling 'same subject', the
other 'different subject'; cf. (6). Quechua signals sameness vs. difference of
subject at the end of most of its subordinate clauses, e.g. the temporal clauses
in (44).17
(44) Quechua
a. Kitu-man chaya-shpa-mi rijsi-ta riku-rka-ni.
[Quito-ALL arrive-ADVLR SS]-VAL acquaintance-ACC see-PAST-1SG
'When I arrived in Quito, I saw a friend.'
b. nuka Kitu-man chaya-jpi-mi
[I Quito-ALLarrive-ADVLRDS DS]-VAL
rijsi riku-wa-rka.
acquaintance see-OBJ 1- PAST(SUBJ 3)
'When I arrived in Quito, a friend saw me.' (Cole, 1982: 61)
The examples illustrate a fairly general phenomenon, namely switch-reference
(see Haiman and Munro (eds.), 1983). In some Indo-European languages
such as Latin, the distinction between 'same subject' and 'different subject' is
not grammaticalized in such a general way. Nevertheless, something close to
it is operative in several areas of clause linkage. The distribution of the con-
junct participle and the ablativus absolutus is mainly governed by the identity
or distinctness of the subjects of the matrix clause and the nonfinite construc-
tion: if they are identical, the conjunct participle is in order, while if they are
distinct, the ablativus absolutus is needed (details in Haiman, 1983 and Hoff,
1985). The conjunct participle is, more generally, possible as a predicative
adjunct to just any NP in the matrix clause (see Pinkster, 1984: ch.8.2).
Another place in Latin where the subordinate construction is found to be
sensitive to subject identity vs. distinctness is in object clauses. This issue has
to be put into the broader framework of complementation and control (cf.
Givon, 1980). The verbs which govern complement clauses fall into the fol-
lowing classes, as regards their valency (cf. Pinkster, 1984: ch.7.1; Foley and
Van Valin, 1984: ch.6.5):
i. monovalent verbs, taking a complement clause as actant;
ii. bivalent verbs, taking an oblique complement clause;
iii. trivalent verbs, taking a human object and an oblique complement
clause.
With monovalent verbs (such as Lat. accidere 'to happen', constare 'to be
a fact'), there is no question of subject identity. The subject of the comple-

ment clause of a bivalent verb (such as cupere 'to desire') may be identical to
the subject of that verb. The subject of the complement clause of a trivalent
verb (such as poscere 'to demand') may be identical to the subject or the
human object of that verb.
From the point of view of control, these verbs fall into three classes,
cross-classifying with the first three:
a. verbs which exclude subject identity, e.g. sinere 'to yield';
b. verbs which allow subject identity, e.g. sperare 'to hope', docere 'to
teach';
c. verbs which force subject identity, e.g. incipere 'to begin', coercere
'to force'.
Verbs of type B and C will be called optional and obligatory control verbs,
respectively. Subordinate clauses with subject control differ systematically in
their structure from subordinate clauses without subject control. Let us briefly
look at some examples of two subtypes of optional control verbs.
On the one hand, we have verbs such as cupere in (45), which signify
some kind of emotional relation of the subject towards the subordinate propo-
sition.
(45) Latin
a. [ut... nuntium accipias] ... cupio (Cic.Tu. 1,17)
T wish for you to receive the message'
b. Misere cupis .. [abire]. (Hor.Sat.1,9,14)
'You are struggling pitiably to get away.'
As in many other languages (cf. Thompson and Longacre, 1985, Sect. I.3.2.1.2
for same vs. different subject in purpose clauses), the object clause of such a
verb may (or even must) be finite if it has its own subject (45a), whereas it must
reduce to an infinitive if the subordinate action is one of the higher subject
(b).
On the other hand, an object clause occurs after verbs of cognition, as in
(46), and even occasionally as a headless relative clause after a plain bivalent
verb, as in (47).
(46) Latin
a. nescit [quo flectere puppem]
'He does not know where to steer the ship.' (Coripp.Joh. 1,273)
(47) Latin
a. non habent unde reddere tibi
NEG have:(PRES)3PL [from where refund:INF thou:DAT]
They have nothing to give you back.' (Itala Luc. 14, 14)
Such subordinate clauses are usually finite. Only if they have the same subject
as the main clause is the infinitive possible, as in these examples.
In both of the cases we have observed, the optional control construction
is just the same as the one appearing with obligatory control verbs. Cf. (45b),
(46) and (47) with (48).
(48) Latin
brevior iam in scribendo [incipio fieri]
T already start getting briefer in writing.' (Cic.Att. 5, 6, 2)
Incipere cannot but govern an infinitive complement. We thus see that the
language provides different kinds of complement clauses for obligatory con-
trol verbs and for non-control verbs and that it makes these available for var-
iation with optional control verbs. The important observation here is just as
above for tense and aspect: If main and subordinate clauses are interlaced by
sharing an element of their structure, this will be left unspecified in the subor-
dinate clause, the specification being supplied by the main clause.
4.1.3 The last example also displays syntagmatic interweaving. There is

another type of interlacing in which this structural aspect is much more pro-
nounced. I am referring to the constructions that used to be called prolepsis
and that are nowadays called raising (several Latin examples in Szantyr,
1972:471; see also Christol, 1985). Observe the following examples:
(49) Latin
non fuit luppiter metuendus, [ne iratus noceret]
Tt was not to be feared that Jove might be pernicious in his anger.'
(Cic.off. 3, 104)
(50) Latin
Viden me [ut rapior]?
'Don't you see how I am being kidnapped?' (Pl.Rud.869)
Here the subjects have been taken out of the noun clauses and have been
made directly dependent on the superordinate verb. This bisects the subordi-
nate clause and, in (49), renders it even discontinuous. This kind of inter-
weaving presupposes the complement status of the subordinate clause.18 It is
particularly common when the subordinate subject has some direct relation to
the superordinate verb, as it clearly has in the two examples. I recall the func-
tional explanation of prolepsis given by Lofstedt (1911:272): according to him
a particularly salient notion, dominating in the relation between the main verb
and the subordinate construction, is anticipated.
There are various other kinds of interweaving, many of them currently
described by raising processes. I will just give two more examples:
(51) Italian
Mi feci [radere la barba].
me made:1sG shave:INF the beard
T had my beard shaved.'
(52) German
Ich habe mich [rasiert].
I have me shaved
T shaved myself.'
In (51) the personal object has been extracted from the dependent infinitival
and made directly dependent on the main verb ('object-to-object raising').
In (52) the participle which depends on the auxiliary has been distanced
from it by the intercalation of another phrase depending on the verb. In all
these cases, the subordinate construction is tightly interwoven with the matrix
clause. Here as above, the subordinate syntagms are complements of the
superordinate verb. In the clearly syntactic constructions (i.e. above the level
of analytic morphology exemplified in (52)), they function either as subject
or as direct object.
4.1.4 I will not at this point try to set up a continuum of interlacing, but we
may accept as plausible the idea that two clauses may be more or less inter-
laced, variation being between a pole of complete disjunctness of the two
clauses and a pole of maximal identity. The tightness of the linking does not
so much depend on the semantic nature of the linking relation as rather on the
amount of material that the two propositions have in common.19 The principle
that we have found to be operative at the level of the (complex) sentence has
a close analog at the text level. Recent research in discourse analysis has led
to the following thesis regarding the mechanism of text cohesion at the para-
graph level (Thompson and Longacre, 1985:211): "Lexical overlap is the
primary mode of intersentential connection".
4.2 Explicitness of linking
4.2.1 We are now coming to a traditional issue, the problem of syndesis and
asyndesis.20 I already remarked in the introduction that there is much ter-
minological confusion in this area. Therefore I should like to stress again that
the presence or absence of a connective device between two clauses has
nothing to do with parataxis vs. hypotaxis, but is exclusively a question of syn-
desis. In particular, it is not the case that either the concept of hypotaxis or
the concept of subordination require the use of a conjunction, as has been
claimed variously. Let us therefore look first at par atactic, then at hypotactic
constructions with regard to syndesis (cf. Brettschneider 1984:14f as to indica-
tive vs. predicative representation of the interclausal relation).
(53) Latin
[Several chapters on a repelled assault of the enemy. First sentence
of new section:] His rebus gestis cum omnibus de causis Caesar
pacatam Galliam existimaret atque ita hieme in Illyricum profectus
esset, quod eas quoque nationes adire et regiones cognoscere vo-
lebat, subitum bellum in Gallia coortum est.
'These things being done, Caesar had every reason to assume that
Gaul was now pacified. Thus in the winter he set off for Illyria,
because he wanted to visit those peoples, too, and to learn about
the area. There a sudden war broke out in Gaul.' (Caes.B.G. 3, 7,
1)
(54) Portuguese
Todas essas agões do departamento evidentemente nao tinham outro
motivo do que nao dificultar ou ainda impossibilizar o meu
trabalho. Assim sendo, eu prefiro pôr o meu posto á disposicao.
'All those actions of the department obviously had no other motive
than to render my work difficult or even impossible. This being so,
I prefer to vacate my post.'
(55) Italian
Il lavoro in questo istituto mi soddisfa completamente, ed anche i
colleghi sono molto gentili. Ciononostante devo dichiarare che esiste
un problema insuperabile.
'Work in this institute satisfies me completely, and also the col-
leagues are very friendly. This notwithstanding, I have to admit
that there is an insurmountable problem.'
(56) German
Ihr Angebot befriedigt meine Anspruche vollkommen; deswegen
nehme ich es dankend an.
'Your offer serves my wants completely; therefore I accept it grate-
fully.'
(57) You are very kind, but I must contradict you.
(58) This is right and that is wrong.
(59) Latin
Veni, vidi, vici.
T came, I saw, I conquered.'
The logic inherent in this series of examples is self-explanatory. We start with
over-explicit syndesis, pass through a continuum of decreasing explicitness
and end up with asyndesis. Three things need to be noted here. First, the
explicit linking devices have a clear anaphoric internal structure. His rebus
gestis in (53) starts with the proximal deictic pronoun which conventionally
refers back to preceding text. The following examples feature assim 'thus'
(54), cid 'that' (55) and des 'of that' (56). Each of these anaphoric pronouns
takes up preceding propositions, thus representing them in the sentence so
introduced. The sentences are hooked to one another, as it were.21 The verbs
contained in the connective locutions of (53) and (54) are also anaphoric.
Recall what was said in Section 4.1.4 on the function of semantic overlap in
text and sentence cohesion.
Second, the connective phrases themselves in (53)-(55) are clearly
reduced adverbial clauses. It is intriguing to observe that in order to connect
two paratactic sentences explicitly, we use a subordinate clause (cf. Thompson
and Longacre 1985, Sect. II.4.5.). The linking element in (56) is still an adverb.
In the following examples we find connective particles (see Pinkster, 1984:
ch. 12.2.5 for adverbs, connectors and coordinators).
Third, in natural text the explicitness of the linking device is adjusted to
the size of the entities linked. This is not a question of grammar, but of unpre-
tentious style. (59) would not become ungrammatical if it contained the con-
nective phrase of (53); but it would be stylistically marked. The relationship
between small chunks of text immediately following each other is sufficiently
clear from the mere adjacency. Large passages need explicit linking in order
to form a cohesive text.
4.2.2 Here is the corresponding series of examples for subordination:

(60) Portuguese
O estudante comprou um monte de livros especializados, [a fim de
que o professor o tivesse por inteligente].
'The student bought a pile of specialized books in order that the
professor should consider him intelligent.'
(61) Italian
[Nonostante l'ópera fosse molto rumorosa], mi addormentai nel
secondo atto.
'Although the opera was very noisy, I fell asleep in the second act.'
(62) Latin
[Postquam aurum abstulimus], in navem conscendimus.
'After we had taken away the gold, we boarded a ship.' (Pl.Ba.277)
(63) Latin
[Haec cum Crassus dixisset], silentium est consecutum.
'When Crassus had said this, silence followed.' (Cic. de or. 1,160)
(64) Portuguese
A verdade e [que todos sairam].
'The truth is that they all left.'
(65) Latin
Si vis [amari], ama.
'If you want to be loved, love.' (Sen.ep.9,6)
The principle underlying this series is the same as above: The connective
and subordinating device is maximally explicit at the start and then is gradu-
ally reduced to zero. The linking phrases are again adverbial in nature. The
phrase a fim de que in (60), lit. 'to [the] end of that', is structured like a pre-
positional phrase introduced by a, with the governed noun fim in turn govern-
ing a dependent noun phrase, which is represented by the subordinate clause.
It is noteworthy that such conjunctional phrases quite commonly incorporate
a universal, unmarked subordinator, here que. The complex conjunction of
(61) contains a verbal participle taking the subordinate clause as a (subject!)
complement. Since the participle functions like a gerund, the whole complex
functions as an adverbial. The conjunction in (63) consists of a preposition
and a subordinator introducing the clause governed by the preposition. The
conjunction in (63) can be recognized, with small etymological effort, as a
case form of a relative pronoun (cf. above Section 3.1.3). Only in the last two
examples do the subordinating devices not have any adverbial character. In
(64) we have the unmarked subordinator. In (65), the interclausal relation is
not expressed at all and the subordination is hidden in the inflectional cate-
gory of the infinitive.
Thus, the first half of this series neatly confirms Delbruck's (1871:99f)
thesis that a conjunction is a subordinator plus a certain case; cf. also Biraud,
1985, Sect. 2f. However, the same cannot be said of the last stages of the con-
tinuum. It should come as no surprise that conjunctions of an adverbial nature
may be found in adverbial clauses, while mere subordinators are found in
complement clauses.
This time the correlation between the explicitness of the interclausal rela-
tion and the heaviness of the linked clauses is not so pronounced, one of the
reasons being that a desententialized clause has to be assigned a function in
the main clause in order not to dangle functionless. Another factor which con-
tributes to complicating the picture is the possibility of expressing the semantic
relation of a subordinate syntagm to the main clause not in the subordinate,
but in the main clause.
(66) Latin
Atque ibi [vehementissime perturbatus] Lentulus tamen et signum et
manum suam cognovit.
'And there Lentulus, being extremely embarrassed, yet recognized
his signature and hand.' (Cic.Cat.3, 12)
The subordinate construction in (66) is strongly desententialized, but its
semantic relation to the matrix clause is made explicit in the main clause con-
junction tamen 'nevertheless'.22
4.2.3 Thus, there does not appear to be a strong relationship between desen-
tentialization and explicitness of linking. I delay the discussion of the correla-
tions which do hold to the next section and conclude this section with the con-
tinuum of explicitness of linking.
svndesis asyndesis
anaphoric subordinate clause
gerundial verb
prepositional phrase
connective adverb
specific conjunction
universal subordinator
nonfinite verb
form
Figure 5: Explicitness of linking
5 TOWARDS A TYPOLOGY OF CLAUSE LINKAGE
5.1 Correlations among the parameters
The following six parameters have been found relevant to the under-
standing of clause linkage and subordination:
i. the hierarchical downgrading of the subordinate clause (Figure 1),
ii. the main clause syntactic level of the subordinate clause (Figure 2),
iii. the desententialization of the subordinate clause (Figure 3),
iv. the grammaticalization of the main verb (Figure 4),
v. the interlacing of the two clauses,
vi. the explicitness of the linking (Figure 5).
Each of these parameters is construed as a continuum extending between
two opposite poles. The question now arises as to the independence of the
parameters. Note that our six continua are directed in a parallel way. Always
looking from left to right, we find the first two continua extending from
parataxis to embedding and from sentence to word level, respectively. The
continuum of desententialization is between the poles of maximal sententiality
and nominality. The continuum of the grammaticalization of the main verb
starts from an independent predicate and ends with a grammatical operator.
Interlacing of the two clauses varies between their total disjunctness and their
maximal identity. Finally, the continuum of explicitness of the linking has
explicit syndesis at its left and asyndesis at its right pole. Rather than trying to
make the intuitive parallelism among the continua explicit, I will simply com-
ment on individual correlations between pairs of them.
As regards the relationship between the first two continua, we saw in Sec-
tion 2.2. that advanced hierarchical downgrading of the subordinate clause
implies a low syntactic level for it. We will thus be justified if in the following
we take advanced downgrading as a sufficient condition for high integration.
High integration of the subordinate into the main clause correlates posi-
tively with its desententialization.23 More precisely, nominalized subordinate
constructions can easily be downgraded, since they acquire the distributional
properties of a nominal expression (cf. Meillet, 1921). We may also say that
nominalization necessitates at least some downgrading, since a reduced con-
struction cannot remain totally independent. However, maximal nominaliza-
tion does not presuppose maximal downgrading, since a nominalized verbal
may be relatively independent (say, as an adjunct, or even left-dislocated) just
to the degree that any NP may be. The Latin participial constructions, e.g. in
(7), illustrate this.
Strong grammaticalization of the governing predicate presupposes either
advanced desententialization or strong interlacing. This is because the gram-
maticalization turns the governing predicate into a grammatical operator on
the subordinate construction, but at the same time cuts down its syntactic
scope (see Lehmann, 1982[T]: ch.IV.3.1). Thus the grammaticalized predi-
cate must be an operator on a construction of relatively low complexity (cf.
Foley and Van Valin, 1984: ch.6.4.5 on the correlative reduction of both the
choice of the governing verb and the syntactic level of the linkage). This is
possible either through desententialization of the subordinate construction or
by having the operator apply, through interlacing, only to one constituent of
the subordinate construction, normally the predicate. Desententialization of
the complement is evidenced by the grammatical causatives and desideratives
in (19) and (40), respectively, while interlacing can be seen in the derivational
counterparts, (38) and (41b). Both desententialization and interlacing occur in
(51) and (52). The latter point would also be illustrated by (48) except that
here the main verb is not strongly grammaticalized.
Interlacing of clauses as brought about by raising operations presupposes
downgrading, thus, integration of the subordinate clause. This is because rais-
ing is controlled by the main predicate, which means that the subordinate
clause is governed by it. There are, indeed, no instances of prolepsis except
out of subject and object clauses. As far as switch-reference is concerned,
Haiman (1983:120) shows "that DS clauses are less dependent than are SS
clauses".
Interlacing of clauses as brought about by dependent subject control
leads to desententialization of the subordinate clause. This follows necessa-
rily, insofar as dependent subject control means non-finiteness of the depen-
dent verb and this means strong desententialization. Cf. Givon, 1980, Sect. 3
and Cooreman, 1984 on the correlation between the 'binding scale' and
desententialization. As for switch-reference, SS clauses exhibit stronger
grammatical category dependence than DS clauses (Haiman, 1983:121).
Explicitness of linking correlates with syntactic level, because the seman-
tic relation linking clause A to clause B is rather constrained if the linkage per-
tains to a low syntactic level of B, whereas more diverse semantic relations
may obtain on higher levels of B; cf. Foley and Van Valin, 1984:196.
Explicitness of linking has some relationship to desententialization. We
saw that explicit conjunctions, either coordinative or subordinative, may have
the function of an adverbial in the linked clause, whereas pure subordinators

do not have such a dependent status within the subordinate clause, but are
just operators applied to it. From this it follows that those explicit conjunc-
tions find their place within the linked clause without necessarily changing its
status very much, whereas the pure subordinators do downgrade the clause
they are applied to (cf. Carvalho, 1985). However, this consideration only
applies to explicit linkers which do function as adverbials within the linked
clause; they do not apply, e.g., to (60).
Finally, asyndesis (lack of explicitness of linking) correlates with depen-
dent subject control and thus with interlacing. This is because the syntactic
relation of the subordinate construction to the matrix clause may be deter-
mined either by some property of the former or by some property of the lat-
ter. If the subordinate clause is linked to the main clause by some explicit con-
nective, this determines its syntactic relation to the main clause and makes the
subordinate clause seek its own place in the complex sentence. However, this
is unnecessary and even impossible when the subordinate clause is controlled
by the main verb.
5.2 A functional account of clause linkage
It should be clear that the correlations and implicative relationships for-

mulated above represent tendencies, not laws. If any of them were perfect
and exceptionless, it would be no use to establish distinct parameters.
Nevertheless, to the degree that they do obtain, it makes sense to combine
our six continua into the synopsis in Figure 6.
Figure 6 makes us see the common functional denominator underlying
the various continua: all of them extend from a pole of maximal elaboration to
a pole of maximal compression (or condensation) of lexical and grammatical
information.24 This means that two opposing forces are at work in clause lin-
kage. The first acts towards the elaboration of a phrase into a more fully
developed construction which contains its own predication with all the acces-
sories. Methodologically, this implies starting from the simple independent
clause and gradually elaborating it into a complex sentence by expanding its
constituents into clauses. To this corresponds the definition of the subordinate
clause as "a member of a sentence having a form similar to that of a sentence"
(Jespersen, 1937:166). This point of view has been predominant in the tradi-
tional analysis of the subordinate clause. It has also been useful in the analysis
of serial verb constructions as a technique of valence increasing (cf. Foley and
elaboration compression
Downgrading of subordinate clause
weak strong
parataxis embedding
Syntactic level
high low
sentence word
Desententialization
weak strong
clause
Grammaticalization of main predicate
weak strong
lexical verb grammatical affix
Interlacing
weak strong
clauses disjunct clauses overlapping
Explicitness of linking
maximal minimal
syndesis asyndesis
Figure 6: Parallelism of clause linkage continua
Van Valin, 1984: ch.5.2) and of cleft sentences as a technique in functional

sentence perspective (cf. Lambrecht, this vol.).
The opposing force acts towards the compression of a full-fledged clause
to a nominal or adverbial constituent of a matrix clause. Methodologically,
this implies a derivation of complex sentences which starts from a set of com-
plete clauses, reduces one of them through desententialization and combines
them into one complex sentence by embedding them into each other. This
point of view has been prevalent both in traditional diachronic approaches to
complex sentence formation and in the transformational analysis of the subor-
dinate clause.
Once we realize that the opposing forces are complementary, we can
combine the two methodological viewpoints.25 In a functional framework,
clause linkage may be viewed as either representing two states of affairs so
tightly interconnected that they form one complex state of affairs (compres-
sion), or on the contrary analyzing one state of affairs as composed of two

(elaboration; cf. Slotty, 1936). In either case the cognitive relatedness of the
two states of affairs is mirrored in the way they are linked in language.
5.3 Types of clause linkage
The complex continuum of Figure 6 also provides a framework for a

typology of clause linkage. Types may be found on this continuum to the
degree that the properties and processes assembled on the six parameters tend
to cluster. Figure 6 tells us that there are two extreme types of clause linkage,
at either of the two poles of the continuum. On the one hand, we have a com-
bination in which both the two clauses and the linkage are maximally elabo-
rated. They are both independent and syntactically equal, each of them is a
full clause. They are linked by an explicit connective device; and this contains
all that is semantically common to the two clauses. This type may be rep-
resented by (53).
On the other hand, we have one clause containing, apart from the main
predicate, a reduced predication. This may be the subordinate clause. It is
then embedded into the main clause on a low syntactic level, strongly
nominalized and interlaced with the main predication and lacks any specifica-
tion of the relation linking it to the latter. This type is represented by (45b).
Or again the matrix predication may be the reduced one. Then this is gram-
maticalized, but the other characteristics of the reduced type hold as well. (20)
embodies this type.
Between these extreme types, there is a great variety of intermediate and
concomitant types. The intermediate types differ from the extreme types
merely in that all the parameters are developed to a non-extreme, but parallel
degree. The concomitant types result from the fact that correlation among the
parameters is not perfect. For example, we quite freely get completely
asyndetic clauses which otherwise show no signs of reduction. (Note that there
was no claim that absence of downgrading correlated with explicitness of link-
ing.) Some of these intermediate and concomitant types are well represented
in the languages of the world. Among them are, from left to right in Figure 6:
- the correlative diptych illustrated by (3) and (4);
- the weakly desententialized noun clause introduced by a universal
subordinator, as in (21) and (64);
- the more strongly nominalized noun clause with a subordinating

verb suffix illustrated in (26);
- the strongly nominahzed adverbial clause represented by such
absolute constructions as in (11);
- the infinitival depending on a modal verb, as in (40) and (65);
verb serialization as exemplified in (16).
Comparative linguists have always felt that the prevalence of such a con-
struction type may characterize — together with other features — a given
language. For instance, the prevalence of the correlative diptych is characteristic
of Hittite; the strongly nominahzed adverbial clause is typical for Quechua
and Tamil. Future research will ascertain which features tend to cluster, what
are the principles intrinsic to the model of Figure 6 that account for such clus-
tering, and how the different construction types go together with other prop-
erties of each language, thus constituting more complex and richer linguistic
types.26
Grammatical category labels
ADVLR adverbalizer DESID desiderative

ALL allative HABIT habitual aspect
AUG augment INAN inanimate
CONN connective NR nominalizer
D3 demonstrative of SR Subordinator
3. person deixis VAL validator
For further abbreviations, see list on page vii.
NOTES
1. The term subordination is applied, in different schools of linguistics, to different kinds of

phenomena. In the broadest use, which may be found in certain trends of European struc-
turalism, the size and nature of the subordinate element is of no concern. Here subordina-
tion practically means the same as dependency. In the most narrow use, characteristic of
classical philology, only finite clauses can be said to be subordinate. Here subordination
practically means the same as hypotaxis; and consequently the two latter terms are mostly
used interchangeably.
There are also differences with respect to the kind of relation presupposed for subordina-
tion. Some (e.g. Touratier, 1985) require that the subordinate clause have a syntactic func-
tion in the matrix clause. Others (e.g. Br0ndal, 1937) admit of subordinate clauses without
matrix clause. In addition, there are a number of morphological, semantic and logical
criteria which have been invoked in order to distinguish subordination from coordination.
As has been shown repeatedly (already in Br0ndal, 1937), none of these is crucial, although
they may be used to characterize a type of subordination well represented in certain Indo-
European languages.
2. This implies that a syntagm will be said to be subordinate only if it contains a predication,
and represents, thus, a compromise solution as regards the nature of the subordinate ele-
ment; cf. note 1. It follows from the above definition that the presence of a subordinate syn-
tagm presupposes the presence of a main clause — which may be or contain Y — to which
it is somehow subordinate (against Br0ndal, 1937).
3. Matthiessen and Thompson (this vol.) regard the "nucleus-satellite relation" as constitutive
for hypotaxis. On the one hand, they conceive this as a basically textual relation; on the
other hand, the traditional notion of hypotaxis appears to them to be best captured as a
grammaticalization of this relation. It thus comes fairly close to the semantosyntactic notion
of endocentricity used above.
4. Except perhaps for the European schools of structuralism mentioned in note 1.
5. Several of these parameters are reviewed in Haiman and Thompson (1984). For at least
some of them, the authors show that they cannot serve as a basis for the concept of subordi-
nation. In what follows, I will use them to differentiate types of clause linkage.
6. Harris (this vol.) shows that this is even true within one language or a group of closely
related languages.
7. The Foley and Van Valin (1984: ch.6.3) semantic bondedness hierarchy embodies a claim to
the contrary.
8. Paul (1920:145) already speaks of "Herabdrückung eines Satzes zum Satzgliede".
9. Cf. Marchese (1977) for various kinds of subordinate clauses in Godie, and Lehmann, 1984:
ch.V.5.1 for relative clauses.
10. A full treatment would have to investigate the relationship between serial verb constructions
and motion purpose clauses (containing expressions such as 'go buy', 'come play' etc.),
which seem to be more widespread. Cf. Aissen, 1984 for a specimen analysis.
11. Cf. Mackenzie (1984). Dik (1985:11) says: "any form of argument reduction easily leads to
'deactualization' of the meaning of the output predicate-frame, where deactualization means
that the output predicate-frame tends to get a generic, habitual or potential reading rather
than a reading which directly designates some actual state of affairs."
12. Again, this is a scalar phenomenon. As Green (1976) shows, the speaker may choose to
couch his main point in a subordinate clause and then use, in this, functional sentence
perspective and word order as characteristic of main clauses. Konig and Van der Auwera
(this vol.) show that main-clause word order in German subordinate clauses signals non-inte-
gration of the latter in the main clause.
13. A third one might be provided by the grammaticalization of the governing main verb to a
conjunction introducing the subordinate clause; cf. Lord (1976).
14. Szantyr (1972:526) seems to be using the term synsemy for this phenomenon. The older term
'synsemanticity' has had too many applications to be recommendable.
15. The interlacing of tense can also be regarded as a special case of relativized deixis known
from indirect (as opposed to direct) speech.
16. Cf. also Tao (1985) for the role of zero anaphora in clause linkage.
17. Cf. the similar situation described for Malayalam in Jayaseelan (1984).
18. Prolepsis is to be clearly distinguished from second position of the conjunction within the
subordinate clause. This is, as Marouzeau (1946) shows quite conclusively, an instance of
Wackernagel's enclitic position and does not lend any special relief to the word preceding
the conjunction.
19. Cf. Davison, (1979:122). Foley and Van Valin, (1984: ch.6.3.2) claim the first factor to be
relevant, too, but do not present convincing evidence for it.
20. In traditional grammar, these terms have normative-stylistical connotations. In particular,
asyndesis is often understood as the absence of a linking device where one would be
expected.
21. For early insights in this matter, see Paul 1920:148f.
22. (66) thus shows that Matthiessen and Thompson's (this vol., Section 5.1) claim for English
that markers of subordination mark satellites is not valid for Latin.
23. Cf. Foley and Van Valin, 1984: ch.6.4.1 on the correlation between their "syntactic bonded-
ness" and nominalization, and also Givon (1980), Dixon (1984), Bolkestein (1985) and Car-
valho (1985) on the correlation between the type of the governing verb and the desententiali-
zation of the governed clause.
24. It should be noted that this is the common denominator worked out by the Cologne research
group UNITYP for their universal functional dimensions. The opposite principles are there
called predication and indication, respectively. Cf., in general, Seiler (1983), and on clause
linkage in particular, Brettschneider (1980 and 1984).
25. Looking back, it should be noted that the unilinear order in which each of the continua has
been presented was due to expository simplicity.
26. Comrie (1984) on the typological connections of control properties, and Mithun (1984) on
subordination in poly synthetic languages, are contributions to this goal.
REFERENCES
Aissen, Judith L. 1984. "Control and command in Tzotzil purpose clauses".

BLS 10:559-571.
Biraud, Michele. 1985. "Propositions pour une description du systeme des
subordonnants en latin". Calboli (org.) 1985.
Bird, Charles S. 1968. "Relative clauses in Bambara". JWAL 5:35-47.
Bolinger, Dwight. 1984. "Intonational signals of subordination". BLS 10:401-
413.
Bolkestein, Machtelt A. 1985. "Parameters in the expression of embedded

predications: Latin subordination and nominalization". Calboli (org.) 1985.
et al., (eds.). 1985. Predicates and Terms in Functional Grammar. Dor-
drecht: Foris.
Brettschneider, Gunter. 1980. "Zur Typologie komplexer Satze". In Wege zur
Universalienforschung, Brettschneider, Gunter and Lehmann, Christian
(eds.) 192-198. Sprachwissenschaftliche Beitrage zum 60. Geburtstag von
Hansjakob Seiler. Tubingen: G. Narr.
-----. 1984. "PARTIZIPATION verknüpft mit NEKTION". akup 59:2-22.
Br0ndal, Viggo. 1937. "Le probleme de I'hypotaxe: reflexions sur la theorie
des propositions". Melanges linguistiques et philologiques offerts a M.
Aleksander Belic 1937:241-249.
Calboli, Gualtiero. (org.) 1985. Proceedings of Third Colloquium on Latin
Linguistics, Bologna, April 1-4, 1985. Published 1989 as Subordination and
Other Topics in Latin. Amsterdam: John Benjamins.
Carvalho, Paulo de 1985. "Subordination, parties de langue et signification. A
propos de la 'construction' des 'verbes de sentiment' (verba affectuum) et
des 'verbes de perception, jugement et affirmation' (sentiendi, iudicandi,
dicendi)". Calboli (org.) 1985.
Chafe, Wallace 1984, "How people use adverbial clauses". BLS 10:437-449.
Christol, Alain. 1985, "Prolepse et syntaxe indo-europeenne". Calboli (org.)
1985.
Cole, Peter. 1982. Imbabura Quechua. (LDS, 5). Amsterdam: North-Hol-
land.
Coleman, Robert G. 1985. "Absolute constructions: A Latin case history".
Calboli (org.) 1985.
Comrie, Bernard. 1984. "Subject and object control: syntax, semantics, prag-
matics". BLS 10:450-464.
and Thompson, Sandra A. 1985. "Lexical Nominalizations". In: Lan-
guage Typology and Syntactic Description Shopen, Timothy (ed.), Cam-
bridge: Cambridge University Press; vol. III: 349-398.
Cooreman, Ann. 1984. "A semantic basis for the choice of complement clause
types in Chamorro". BLS 10:572-582.
Craig, Colette Grinevald. 1977. The Structure of Jacaltec. Austin & London:
Davies, John. 1981. Kobon (LDS, 3). Amsterdam: North-Holland.
Davison, Alice. 1979. "Some mysteries of subordination". Studies in the Lin-
guistic Sciences 9:105-128.
Delbrück, Berthold. 1871. Der Gebrauch des Konjunktivs und Optativs im

Sanskrit und Griechischen. (Syntaktische Forschungen, 1). Halle :
Waisenhaus.
Dik, Simon. 1985. "Formal and semantic adjustment of derived construc-
tions". In Bolkestein et al. (eds.), 1985:1-28.
Dixon, R.M.W. 1984. "The semantic basis of syntactic properties". BLS
10:583-595.
Drossard, Werner. 1984. "KAUSATIVIERUNG und TRANSITIVIERUNG
im Tagalog". akup 60:1-25.
Foley, William A. and Van Valin, Robert D., Jr. 1984. Functional Syntax and
Universal Grammar. {Cambridge Studies in Linguistics, 38). Cambridge:
Cambridge University Press.
Givon, T. 1980. "The binding hierarchy and the typology of complements".
Studies in Language 4:333-378.
Green, Georgia M. 1976. "Main clause phenomena in subordinate clauses".
Language 52:382-397.
Haiman, John. 1978. "Conditionals are topics". Language 54:564-589.
1983. "On some origins of switch reference marking". Haiman and
Munro (eds.), 1983:105-128.
. and Munro, Pamela (eds.). 1983. Switch-Reference and Universal Gram-
mar. Proceedings of a Symposium on Switch-Reference and Universal
Grammar. (TSL, 2). Amsterdam & Philadelphia: John Benjamins.
and Thompson, Sandra A. 1984. "'Subordination' in universal gram-
mar". BLS 10:510-523.
Hale, Kenneth L. 1976. "The adjoined relative clause in Australia". In:
Grammatical Categories in Australian Languages, Dixon, Robert M.W.
(ed.), (Linguistics Series, 22), 78-105. Canberra : Australian Institute of
Aboriginal Studies.
Haudry, Jean. 1973, "Parataxe, hypotaxe et correlation dans la phrase
latine". BSL 68(1): 147-186.
Helander, Hans. 1985, "Causal subordinate clauses and other expression of
cause". Calboli (org.) 1985.
Hoff, Francois. 1985. "Les ablatifs absolus irreguliers: un nouvel examen du
probleme". Calboli (org.) 1985.
Jayaseelan, K.A. 1984. "Control in some sentential adjuncts of Malayalam".
BLS 10:623-633.
Jespersen, Otto. 1937. Analytic Syntax. New York: Holt, Rinehart & Winston
(Transatlantic Series in Linguistics). Reimpression 1969.
Lakoff, George. 1984. "Performative subordinate clauses". BLS 10:472-480.

Lehmann, Christian. 1979. "Zur Typologie des Lateinischen". Glotta 57:237-
253.
1982. Thoughts on Grammaticalization. A programmatic sketch. (akup,
48). Köln: Institut fur Sprachwissenschaft der Universität.
1982. "Nominalisierung — Typisierung von Propositionen". In:
Apprehension. Das sprachliche Erfassen von Gegenständen. Teil I: Bereich
und Ordnung der Phanomene, Seiler, Hansjakob and Lehmann, Christian
(eds.) 64-82. Tubingen : G. Narr (LUS, 1,I).
1984. Der Relativsatz. Typologie seiner Strukturen, Theorie seiner
Funktionen, Kompendium seiner Grammatik. Tubingen: G. Narr (LUS,
3).
1985. "On grammatical relationality". FoL 19:67-109.
Lofstedt, Einar. 1911. Philologischer Kommentar zur Peregrinatio Aetheriae.
Untersuchungen zur Geschichte der lateinischen Sprache. Uppsala:
Almqvist & Wiksell; Leipzig: R. Haupt.
Lord, Carol. 1976. "Evidence for syntactic reanalysis: from verb to com-
plementizer in Kwa". In: Papers from the Parasession on Diachronic Syn-
tax, Steever, Sanford B. et al. (eds.), 179-191. Chicago: CLS.
Mackenzie, J. Lachlan. 1984. "Communicative functions of subordination".
In: English Language Research. The Dutch contribution I, Mackenzie, J.L.
and Wekker, H. (eds.), 67-84. Amsterdam: Free University Press.
1985. "Nominalization and valency reduction". In: Bolkestein et al.
(eds.), 1985:29-272.
Marchese, Lynell. 1977. "Subordinate clauses as topics in Godie". In: Papers
from the Eighth Conference on African Linguistics, Mould, Martin and
Hinnebusch, Thomas J. (eds.), 157-164, Los Angeles: UCLA {Studies in
African Linguistics, Suppl. 7).
Marouzeau, Jules. 1946. "La construction des subordonnants". REL 24:247
260.
Meillet, Antoine. 1921. "Remarques sur la theorie de la phrase". Journal de
Psychologie:609ff. Reimpr.: Meillet, Antoine 1936, Linguistique historique
et linguistique générale. Vol. II:1-8. Paris: Klincksieck.
Milewski, Tadeusz. 1954. "Les equivalences de phrases composees
indoeuropeennes dans les langues americaines". Biuletyn Polskiego
Towarzystwa Jezykoznawczego 13:143-146. Reimpr: 1967. Etudes
typologiques sur les langues indigenes de l;Amérique. Krakow: Polska
Akademia Nauk (Prace Komisji Orientalistycznej, Nr.7).
Mithun, Marianne. 1984. "How to avoid subordination". BLS 10:493-509.

Paul, Hermann (ed.). 1920. Prinzipien der Sprachgeschichte. 5. ed. Leipzig:
Niemeyer. Reimpr.: Tubingen: Niemeyer, 1975.
Pinkster, Harm. 1984. Latijnse syntaxis en semantiek. Amsterdam: B.R.
Grüner.
Rosen, Hannah. 1985. "General subordinators and sentence complements".
Calboli (org.) 1985.
Seiler, Hansjakob. 1983. POSSESSION as an Operational Dimension of Lan-
guage. Tubingen: Narr (LUS, 2).
Slotty, Friedrich. 1936. "Zur Theorie des Nebensatzes". TCLP 6:133-146.
Szantyr, Anton. 1965. Lateinische Syntax und Stilistik. Von J. B. Hofmann,
neubearbeitet von A. Szantyr. München: Beck (Handb. d. Altertumswiss.
2,2, Bd.2).
Tao, Liang. 1985. "Clause linkage and zero anaphora in Mandarin Chinese."
Davis Working Papers in Linguistics 1.
Thompson, Sandra A. 1985. "Grammar and written discourse: Initial vs. final
purpose clauses in English". Text 5.1.
--—. and Longacre, Robert E. 1985. "Adverbial clauses". In: Language
Typology and Syntactic Description. Shopen, Timothy (ed.), vol. II:171-
234. Cambridge: Cambridge University Press.
Todd, Terry Lynn. 1975. "Clause versus sentence in Choctaw". Linguistics
161:39-68.
Touratier, Christian. 1985. "La subordination, essai de definition". Calboli
(org.) 1985.
Van Valin, Robert D. 1984. "A typology of syntactic relations in clause lin-
kage". BLS 10:542-558.
Wendt, Heinz Friedrich. 1972. Langenscheidts Praktisches Lehrbuch Tur-
kisch. Berlin: Langenscheidt.
Subordination in Tauya
Lorna MacDonald
1 Introduction
In Tauya, a Papuan language spoken by about 400 people living in the

Upper Ramu Valley, Madang Province, Papua New Guinea, there are three
kinds of subordinate clauses marked by the suffix -na1. Two of these, com-
plement clauses and inconsequential clauses, are morphologically similar,
and in some cases are morphologically indistinguishable. However, there are
clear syntactic and pragmatic differences between them. Syntactic evidence
suggests that inconsequential clauses have the same structure as left-
dislocated complement clauses. Although inconsequential clauses could be
considered to be subordinate, they can occur alone as complete, grammatical
sentences.
In Section 2, the similarities and differences between complement and
inconsequential clauses are presented. In Section 3, left-dislocated com-
plements and inconsequential clauses are compared, and in Section 4 the use
of inconsequential clauses as independent sentences is demonstrated. Final-
ly, in Section 5, it is suggested that the ability of inconsequential clauses to
function as independent sentences is a result of their pragmatic and struc-
tural independence from the following main clause.
2 Subordination
2.1 Subordinate clauses
In Tauya, an SOV language, three kinds of subordinate clauses are

marked by the suffix -na. The first of these is relative clauses2:
228 LORNA MACDONALD
(1) 0 - tu - e - na fanu - ra yate fitau - a - ?a

3SG give 1/2 SUB man TOP go throw 3SG IND
T h e man I gave (it) to went away'
The other two kinds of subordinate clauses with -na are exemplified by (2)
and (3):
(2) bramani mene - e - na - ?aisami tauya - sa
stay 1/2 SUB ABL ADESS
?
fofe - e - a
come 1/2 IND
'I came to Tauya from Brahman' (lit., From staying in
Brahman I came to Tauya)
(3) fanu nipi tei - sa mene - a - na tauya - sa
man her ADESS stay 3SG SUB ADESS
mene - a - ? a
stay 3SG IND
'Her husband stayed in Teri and/but she stayed in Tauya'
The subordinate clause in (2) can be identified as a complement clause 3.
Complements are followed by case suffixes which specify the role of the
clause within the higher sentence; in (2), the case relation of the complement
is ablative, marked by the suffix -?aisami. In (3), on the other hand, the
subordinate clause is not marked for case, nor does it receive a case inter-
pretation. As is evident from the gloss, many such clauses are interpreted as
inconsequential clauses, indicating that nothing follows from them causally.
Complement clauses and inconsequential clauses are very similar mor-
phologically. In fact, they may be morphologically indistinguishable if the
case suffix on the complement clause is deleted (see Section 2.2). Further-
more, both are preposed to the main clause. However, as is demonstrated
in Sections 2.2 and 2.3, there are clear syntactic and pragmatic differences
between them.
All three types of clauses marked by the suffix -na are identified as
'subordinate' on the following grounds (cf. Haiman, 1980). First, unlike
coordinate clauses, -na-marked clauses in Tauya do not share the mood of
the following (final) clause, but are interpreted as indicative. For example,
consider the following sentence which includes a relative clause:
SUBORDINATION IN TAUYA 229
(4) mei fofe - a - na fena?a - ra 0 - yau - e - nae

here come 3SG SUB woman TOP 3SG see 1/2 Q
'Did you see the woman who came here?'
(*Did you see the woman who did she come here?)
Although the final verb in (4) is in the interrogative mood, marked by the
polar interrogative modal suffix -nae, the relative clause is interpreted as
assertive. In (5), on the other hand, the coordinate clause can be interpreted
as sharing the mood of the final clause.
(5) fenc?a mei fofe - a - te 0 - yau - e - nae
woman here come 3SG COORD 3SG see 1/2 Q
'Did the woman come here and did you see her?'
'The woman came here and did you see her?'
Second, unlike coordinate clauses, -na-marked clauses do not share the tense
of the following final clause. For example, in (6) the final verb is marked
for the aorist tense, while the complement clause is in the future tense:
(6) pate mei fofe - ?ai - na - pe ?ufiya
tomorrow here come 3PL FUT SUB BEN sweet potato
fei - ?afe - a - ?a
boil PROG 3SG IND
'They'll come here tomorrow so she's cooking sweet potatoes'
Coordinate clauses, on the other hand, are always interpreted as sharing the
tense of the following clause.
(7) mei fofe - a - te ya - ni 0 - tu - amu - ?a
here come 3SG COORD 1SG ERG 3SG give 1SG FUT IND
'He'll come here and I'll give (it) to him'
(*He came here and I'll give (it) to him)
Finally, and again unlike coordinate clauses, -na-marked clauses need not be
interpreted as tense-iconic, that is, the ordering of the clauses does not
necessarily reflect the order of events. Thus, the inconsequential clause in (8)
is not tense-iconic with the final clause.
(8) pate saniya te - amu - na apu ?ini - mo - pe
tomorrow work get 1SGFUT SUB now sleep NOML BEN
230 LORNA MACDONALD
?ai - e - ?a
do 1/2 IND
'Tomorrow I'll work and/but now I want to sleep'
Coordinate clauses are always interpreted as tense-iconic.
(9) ya yate - pa ni - e - ?#
1SG go COORD eat 1/2 IND
T went and ate'
(*I ate and went)
These three syntactic features clearly identify -na-marked clauses as subor-
dinate. In the following subsections, similarities and differences between two
of these, complement clauses and inconsequential clauses, are examined.
2.2 Complement clauses
Complement clauses function as NPs within the higher sentence. As

NPs, they may occur with any of the case suffixes which are possible for lex-
ical nouns.
(10) fanu nipi ?umu - a - na - ni wamasi
man her die 3SG SUB ERG/INSTR widow
mene - a - ?#
stay 3SG IND
'Her husband died [ERG/INSTR] SO she's a widow'
(11) fanu nipi ?umu - a - na -pe fanu teme - nani
man her die 3SG SUB BEN man other ADESS
yate - a - ?#
go 3SG IND
'Her husband died [BEN] SO she went to another man'
(12) fena^a ? e/ momune - ^afe - a - na - na nono
woman there sit PROG 3SG SUB GEN child
?umu - a - ?a
die 3SG IND
'The woman sitting over there's [GEN] child died'
(13) ni mo ? otu ?ini - ?afe - i - na - ?ai
3PL all sleep PROG 3PL SUB ADESS
yate - amu - ?a
gO lSGFUT IND
T i l go to where they're all sleeping [ADDESS]'
Note the use of the ergative/instrumental and benefactive case suffixes to

mark complements as cause clauses. As in Hua, the ergative/instrumental
case is used when the result is an involuntary response, as in (10), while the
benefactive case is used when the result is more-or-less voluntary, as in (11).
As subordinate clauses, complement clauses do not share the mood of
the following final clause. However, a constituent within a complement
clause may be questioned. In this case, the final verb includes the non-polar
interrogative modal suffix -ne~ -e\
(14) we ?ini - ?afe - a - na - ?ai yate - e - ne
who sleep PROG 3SG SUB ADESS go 1/2 Q
'You went to where who was sleeping?'
(15) wame wimo te - a - na - pe na - ?isafe - a - e
what steal 3SG SUB BEN 2 SG angry 3SG Q
'What did he steal such that you are angry?'
In Tauya, the topic suffix -ra is mutually exclusive with all non-null case
suffixes on lexical nouns 5 . Thus,
?
(16) i fanu - ni (* -ra)
DEM man ERG TOP
'that man' (ERG)
(17) mu nono - pe (* -ra)
DEM child BEN TOP
'for this child' (BEN)
Similarly, complement clauses with overt case suffixes may not be marked
as topics:
(18) fanu nipi ?umu - a - na - pe (* -ra) fanu
man her die 3SG SUB BEN TOP man
?a
teme - nani yate - a -
other ADESS go 3SG IND
'Her husband died (*TOP) SO she went to another man'
232 LORNA MACDONALD
(19) momune - ?afe - e- na - ?ai (* -ra) fofe - i - ?a

Sit PROG 1/2 SUB ADESS TOP Come 3PL IND
'They came to where I was sitting (*TOP)'
The case suffix on a complement clause may be deleted if its case relation
is recoverable, either from context or from the lexical content of the clauses.
Thus, (10) to (13), above, may occur as (20) to (23).
(20) fanu nipi ?umu - a - na wamasi mene - a - ?#
'her husband died so she's a widow'
(21) fanu nipi ?umu - a - na fanu teme - nani yate - a - ?#
'Her husband died so she went to another man'
(22) fena?a ?ei momune - ?afe - a - na nono ?umu - a - ?a
'The woman sitting over there's child died'
(23) ni mo? otu ?ini - ?afe - i - na yate - amu - ?a
'I'll go to where they're all sleeping'
However, after deletion of the case suffixes, these clauses behave exactly as
do complements with overt case suffixes in respect to the distribution of in-
terrogative pronouns and the topix suffix. That is, they can include an inter-
rogative pronoun, and they cannot be marked as topics.
(24) we ?umu - a - na wamasi mene - a - e
who die 3SG SUB widow stay 3SG Q
'Who died such that she's a widow?'
(25) fanu nipi ?umu - a - na (* -ra) wamasi mene - a - ?#
man her die 3SG SUB TOP widow stay 3SG IND
'Her husband died so she's a widow'
It is in relation to the distribution of interrogative pronouns and the topic
suffix that complement clauses differ crucially from inconsequential clauses.
2.3 Inconsequential clauses
Unlike complement clauses, inconsequential clauses do not occur with

case suffixes, nor do they receive case interpretations.
(26) mei fofe - a - na wate 0 - yau - e - ?a

here come 3SG SUB NEG 3SG see 1/2 IND
'He came here and/but I didn't see him'
(27) ya - ni nu - su?utu - e - na " ?we - ?a " o - i - ?#
1SG ERG 3PL ask 1/2 SUB no IND say 3PL IND
T asked them and/but they said " N o " '
(27) mei fofe - i - na wate ya - tu - i - ?#
here come 3PL SUB NEG ISG give 3PL IND
They came here and/but they didn't give it to me'
Inconsequential clauses are often used where coordinate clauses would
be inappropriate for syntactic or pragmatic reasons. For example, coor-
dinate clauses may not be used in clause sequences where lexical content
rules out a causal inference:
(29) ? ya - ni pai a?ate - e - te wate ?umu - a - ?a6
lSG ERG pig hit 1/2 COORD NEG die 3SG IND
'(I hit the pig and it didn't die)'
In such cases, the non-final clause must be inconsequential:
(30) ya - ni pai a?ate - e - na wate ?umu - a - ?a
SUB
T hit the pig and/but it didn't die'
Second, unlike subordinate clauses, coordinate clauses may be interpreted as
sharing the mood of the following clause, and are always interpreted as shar-
ing the tense of the following clause. Therefore, when a mood or tense
distinction between two clauses is required, the first must be an inconsequen-
tial.
(31) a. mei fofe - a - te 0 - yau - e - nae
here come 3SG COORD 3SG see 1/2 Q
'Did he come here and did you see him?'
'He came here and did you see him?'
b. mei fofe - a - na 0 - yau - e - nae
'He came here and did you see him?'
234 LORNA MACDONALD
?
(32) a. ya - ni 0 - yau - pa yate fitau - amu - a
1SG ERG 3SG see COORD go throw 1SGFUT IND
'I'll see him and go away'
b. ya - ni 0 - yau - e - na yate fitau - amu - ?a
1/2 SUB
'I saw him and I'll go away'
Although inconsequential clauses are morphologically similar to com-
plement clauses, they are clearly distinct syntactically. First, unlike comple-
ment clauses, constituents within inconsequential clauses may not be ques-
tioned.
(33) a. fanu nipi mei mene - a - na ni bramani yate - a - ?a
man her here stay 3SG SUB 3SG go 3SG IND
'Her husband stayed here and/but she went to Brahman'
b.*we mei mene - a - na ni bramani yate - a - e
who Q
('Who stayed here and/but she went to Brahman?')
Second, whereas complement clauses may not occur with the topic suffix,
inconsequential clauses may be marked as topics:
(34) fanu nipi mei mene - a - na - ra ni bramani yate - a - ?a
TOP
'Her husband stayed here (TOP) and/but she went to Brahman'
2.4 Complement clauses vs. inconsequential clauses
The syntactic distinctions between complement clauses and inconse-

quential clauses may be summarized as follows:
Interrogative Topic
pronouns suffix
Complement yes no
Inconsequential no yes
That is, a constituent within a complement clause may be questioned, and
these clauses may not be marked as topics. Conversely, constituents within
inconsequential clauses may not be questioned, and these clauses can be
marked as topics.
Since the case suffixes on complement clauses may be deleted, comple-
ment and inconsequential clauses may be morphologically indistinguishable.
In some cases, the lexical content of the clauses forces either a complement
or an inconsequential interpretation. For example, the -na-clause in (35) is
interpreted as a cause clause; it is therefore a complement clause with a
deleted (ergative) case suffix.
(35) amo foi - a - na ?utine - a - ?a
tree wet 3SG SUB fall 3SG IND
T h e tree was rotten so it fell
As a complement, a constituent within this clause may be questioned, and
it may not be marked as topic:
(36) a. wame foi - a - na ?utine - a - e
what Q
'What was rotten such that it fell?'
b. amo foi - a - na (* -ra) ?utine - a - ?a
TOP
T h e tree was rotten (*TOP) SO it fell'
In (37), on the other hand, the -na-clause is interpreted as an inconsequential
clause, since no case interpretation suggests itself. As an inconsequential
clause, none of its constituents may be questioned, and it can be marked as
topic.
(37) a. na tei - sa yate - e - na ya bramani yate - e - ?a
2SG ADESS go 1/2 SUB lSG gO 1/2 IND
'You went to Teri and/but I went to Brahman'
b.*na mafi yate - e - na ya bramani yate - e - ne
where Q
('You went where and/but I went to Brahman?')
c. na tei - sa yate - e - na - ra ya bramani yate - e - ?a
TOP
'You went to Teri (TOP) and/but I went to Brahman'
However, there are many cases in which morphological ambiguity
results in inconsistent interpretations. For example, consider (38) to (40),
below.
236 LORNA MACDONALD
(38) wame ni - e - na na - fo?u - a - e

what eat 1/2 SUB 2SG swollen 3SG Q
belly
'What did you eat such that your belly is swollen?'
(39) ? wame ni - e - na ?ini - e - ne
sleep 1/2 Q
('What did you eat such that you slept?')
(40) * wame ni - e - na bramani yate - e - ne
go 1/2 Q
('What did you eat and/but you went to Brahman?')
(38) is a grammatical sentence. Here there is a clear cause/effect relationship
between the two clauses. The first clause is therefore interpreted as a comple-
ment clause with a deleted ergative case suffix. As a complement, its consti-
tuents may be questioned. (39), on the other hand, does not elicit uniform
grammaticality judgements. A case interpretation for the -na-claust is possi-
ble, i.e., it can be interpreted as a cause clause, but this interpretation is not
necessarily made. Finally, (40) is an ungrammatical sentence. No case inter-
pretation for the -na-clause is possible; it must therefore be interpreted as
an inconsequential clause and, as such, none of its constituents may be ques-
tioned.
(41) is an another example of the varying interpretations which can
result from morphological ambiguity.
(41) sema fitau - e - na wate tepau - a - ?a
pot throw 1/2 SUB NEG break 3SG IND
The subordinate clause in (41) has two possible interpretations. First, it may
be a complement clause in the nominative case, with case suffix 0; with this
interpretation, the appropriate gloss is The pot that you threw didn't break.
As a complement, a constituent within this clause can be questioned:
(42) a. wame fitau - e - na wate tepau - a - e
what Q
'What that you threw didn't break?'
Second, the -na-clause can be interpreted as an inconsequential clause, i.e.,
You threw the pot and/but it didn't break. With this interpretation, no con-
stituent within the subordinate clause may be questioned:
(42) b.*wame fitau - e - na wate tepau - a - e

'(What did you throw and/but it didn't break?)'
Therefore, despite their morphological similarity, it is clear that com-
plement clauses and inconsequential clauses are distinct syntactically.
However, there is evidence that inconsequential clauses have the same struc-
ture as left-dislocated complement clauses; this evidence is presented in the
following section.
3 Left-dislocated complement clauses
Complement clauses in Tauya may occur as left-dislocations. They are

then not marked for case, but are followed by the resumptive pronoun ?i
which takes their case inflection7.
(43) amo foi - a - na ?i - ni ?utine - a - ?a
tree wet 3SG SUB PRO ERG/INSTR fall 3SG IND
T h e tree was rotten, that's why it fell'
(44) pai yapi wimo te - a - na ?i - pe ya - ?isafe - a - ?a
pig my steal 3SG SUB PRO BEN 1SG angry 3SG IND
'He stole my pig, that's why I'm angry'
(45) fena?a mei - ?afe - a - na ?i - na nono ?umu - a - ?a
woman cry PROG 3SG SUB PRO GEN child die 3 SG IND
T h e woman crying, her child died'
(46) mene - ?afe - a - na ?e - sami mei fofe - ene - ?a
stay PROG 3SG SUB PRO ABL here come 1/2PL IND
'(Where) he's staying, that's where we came here from'
If the dislocated complement clause is in the nominative case, with case suf-
fix 0, the resumptive pronoun is reduced to a clitic on the dislocated clause:
(47) sema fitau - e - na - ?i tepau - a - ?a
pot throw 1/2 SUB PRO break 3SG IND
T h e pot I threw, (it) broke'
Dislocated complement clauses, unlike their non-dislocated counter-
parts, may not include interrogative pronouns:
238 LORNA MACDONALD
(48) * wame wimo te - a - na ?i - pe na - ?isafe - a - e

what steal SG SUB PRO BEN 2SG angry 3SG Q
('What did he steal, that's why you're angry?')
Unlike non-dislocated complements, dislocated complement clauses can be
marked as topics:
(49) pai yapi wimo te - a - na - ra ?i - pe ya - ?isafe - a - ?a
pig my steal 3SG SUB TOP PRO BEN 1SG angry 3SG IND
'He stole my pig (TOP), that's why I'm angry'
That is, with respect to the distribution of the topic suffix and interrogative
pronouns, dislocated complement clauses behave like inconsequential
clauses:
Interrogative Topic
pronouns suffix
Complement yes no
Dislocated complement no yes
Inconsequential no yes
There are at least two more similarities between dislocated complement
clauses and inconsequential clauses. First, just as dislocated complements
are followed by the resumptive pronoun ?i, inconsequential clauses may also
be followed by this morpheme. It occurs as a clitic on the inconsequential
clause:
(50) ya bramani yate - e - na - ?i ?ei wate ya - yau - i - ?a
1SG go 1/2 SUB PRO there NEG 1SG see 3PL IND
'I went to Brahman and/but they didn't see me there'
If the topic suffix is present, it precedes the resumptive pronoun:
(51) mei fofe - a - na - ra - ?i wate Ø - yau - e - ?a
here come 3SG SUB TOP PRO NEG 3SG see 1/2 IND
'He came here (TOP) and/but I didn't see him'
Second, both dislocated complement clauses and inconsequential clauses are
typically followed by a marked pause which clearly separates them from the
rest of the sentence. With dislocated complements, the pause occurs between
the complement clause and the following resumptive pronoun:
(52) fena?a Ø - yau - e - na - ra PAUSE ?i - ni sini - pe

woman 3SG see 1/2 SUB TOP PRO ERG 1PL BEN
?ufiya fei - ?e - ?a
sweet potato boil 3SG FUT IND
'The woman I saw, she'll cook sweet potatoes for us'
However, if the dislocated complement is in the nominative case, the pause
follows the cliticized resumptive pronoun:
(53) fena?a ?ei ?ini - mene - a - na - ?i PAUSE pate
woman there sleep STAT 3SG SUB PRO tomorrow
Bundi - sa yate - ?e - ?a
ADESS gO 3SGFUT IND
T h e woman asleep there, she'll go to Bundi tomorrow'
Similarly, inconsequential clauses are typically followed by a pause, which
occurs after the resumptive pronoun, if present:
(54) tauya - sa mene - pe - i - na - ?i PA USE ?ite
ADESS stay HAB 3PL SUB PRO garden
topu - pa ni - pa saniya te - pa ...
plant COORD eat COORD work get COORD
'They lived in Tauya; they planted gardens and ate and work-
ed and ...'
If the inconsequential clause is not followed by a resumptive pronoun, it is
nonetheless typically followed by a pause:
(55) te - we - i - na PAUSE wate te - pa yau - i - ?a
get CON 3PL SUB NEG get COORD see 3PL IND
'They tried, but they couldn't find it'
All evidence suggests that inconsequential clauses have the same struc-
ture as dislocated complement clauses. They are similar morphologically,
and behave alike with respect to the distribution of interrogative pronouns
and the topic suffix. Furthermore, both are typically followed by pauses
which separate them from the rest of the sentence.
4 Inconsequential clauses as independent clauses
Subordinate clauses are generally defined as being dependent. As part of

240 LORNA MACDONALD
their dependent status, Mallinson and Blake suggest that subordinate

clauses, unlike coordinate clauses, are not separable, i.e., they cannot occur
alone as independent sentences (1981: 261). Thus, for example, coordinate
clauses are separable in English, while subordinate clauses are not.
(56) a. It rained yesterday and the ground was wet
b. It rained yesterday
(57) a. When it rained yesterday, the ground was wet
b.*When it rained yesterday
However, the situation in Tauya is quite different. Here, inconsequential
clauses can occur alone, and are then interpreted as complete, grammatical
sentences. When coordinate clauses occur alone, on the other hand, they are
interpreted as sentence fragments.
Example (59.a) includes a coordinate clause; if this clause occurs alone,
as in (59.b), it is marginal in grammaticality and is interpreted as a fragment:
(59) a. ya?e fomitiya mepi - a - te o?osi foi - a - ?a
water yesterday come 3SG COORD ground wet 3SG ?ND
down
Tt rained yesterday and the ground was wet'
(59) b.?ya?e fomitiya mepi - a - te - ?a
water yesterday come 3SG COORD IND
down
('Why is the ground wet?) 'Because it rained yesterday'
The inability of Tauya coordinate clauses to occur alone can perhaps be ex-
plained very simply. The suffixes which are used in clausal coordination (-te,
in the example above) are quite possibly derived historically from conjunc-
tions. In this case, the nearest English equivalent to (59.b) would be *It rain-
ed yesterday and.
Complement clauses rarely occur independently. When they do, they,
like coordinate clauses, are interpreted as sentence fragments.
(60) pai yapi wimo te - a - na - pe - ?a
pig my steal 3SG SUB BEN IND
(Why are you angry?) 'Because he stole my pig'
Left-dislocated complement clauses occur independently with slightly
greater frequency. Generally, some sort of verbal context is required:
(61) nono nipi ?umu - a - na ?i - pe - ?a

child her die 3SG SUB PRO BEN IND
(Why is she crying?) 'Her child died, that's why'
Dislocated complement clauses also occur independently in a construction
which requires no verbal context, but which is used as a threat. The resump-
tive pronoun is marked with the case suffix -pene; this seems to imply causa-
tion, and that there will be an immediate response. For example, (62) was
said by a man to a young child who was holding a lighted match to a grass
roof.
(62) wate fu- ?e - na ?i - pene !
house burn 3SGFUT SUB PRO CASE
T h e house will burn' (... then I'll get you!)
Inconsequential clauses, on the other hand, occur as independent
sentences very frequently. They are not interpreted as sentence fragments,
but as complete, grammatical sentences. Examples (63) to (65), below, are
all from texts.
(63) youse nutumo topi - te - pa ate
charred tapiok end cut get COORD old woman
?w atifo yate 0 - tu - a - te yau - a - na - ?a
friend go 3SG give 3SG COORD see 3SG SUB IND
'She cut off the charred tapiok end and gave it to her friend,
the old woman, and the latter looked at it.'
(64) ?afa?i - sami forou - te - pa mepi - a - na mepi
above ABL get set 2 get COORD come 3SG SUB come
down down
0 - tei - a - na - ?a
3SG catch 3SG SUB IND
'He got ready and came down from above, he came down and
caught her'
(65) ?e?e sa firo - tei - tei - tei ya- ?emamo
oh road roam ITERATIVE 1 SG stomach
mai - a - te ?atou - e - na - ?a
come up 3SG COORD arrive 1/2 SUB IND
'Oh!, I roamed and roamed and I got out of breath and I
arrived.'
242 LORNA MACDONALD
Independent inconsequential clauses are also accepted as grammatical in

elicitation, whether or not an appropriate context is provided. They are
usually translated with Neo-Melanesian tasol, 'just' or 'only'.
(66) ya bramani yate - e - na - ?a
lSG gO 1/2 SUB IND
'I only went to Brahman' (N.-M. Mi go long Brahman tasol)
These clauses are also used in conversation. For example, (67) was said by
a man to a woman who was getting impatient while the two were waiting for
a third woman to return from her garden:
(67) mai - a - na - ?a
come up 3SG SUB IND
'She's coming up'
The implication seems to have been, Relax, stop worrying, she's on her way.
Occasionally, independent inconsequential clauses include the resump-
tive pronoun ?i, which precedes the modal suffix. This provides evidence for
their dislocated status.
(68) " apuma - ra ate afe sinipi
right now TOP old woman mother our
Ø - sotou - fe - ene - na - ?i - ?a . pate
3SG bury PERF 1/2PLS UB PRO IND tomorrow
?unama yate te - a - nani - ?a " o - a - na - ?i
only go get 2SGFUT ASSER IND say 3SG SUB PRO
"We've just buried the old woman, our mother. Tomorrow
you'll go and get it", he said'
A similar construction is used in some exclamations, wherein the resumptive
pronoun is followed by the exclamatory suffix -e8
(69) pomu - a - na ?i - e
fall 3SG SUB PRO EXCLAM
'He fell!'
5 Interpretation
The examples presented in Section 4 reveal that inconsequential clauses

in Tauya are separable, i.e., they can occur alone as grammatical sentences,
despite their status as subordinate clauses and their status as dislocations.

Coordinate clauses, on the other hand, are not separable. The separability
of inconsequential clauses can be seen as a reflection of their syntactic and
pragmatic independence. That is, unlike coordinate clauses, inconsequential
clauses are independent of both the tense and the mood of the following
main clause. Further, unlike coordinate clauses, inconsequential clauses im-
ply the lack of any sort of causal connection with the following main clause,
and are not necessarily tense-iconic.
There is evidence of a continuum of dependency among the subordinate
clauses examined here. Non-dislocated complement clauses are dependent,
i.e., they cannot occur alone. This dependence is reflected in their structure.
Complement clauses occur within the main clause, not separated from it by
a pause or by the topic suffix, and they are marked with case suffixes which
specify their role within that clause, i.e.,
Complement:
s[[Clause + na-] + Case ...]s
Left-dislocated complements are intermediate in their dependency status.
Although they generally cannot occur alone, there is at least one construc-
tion in which they can occur without any verbal context, i.e., the threat con-
struction (see example (62)). As a reflection of their semi-dependent status,
they are separated from the main clause by a pause, and in some cases by
the topic suffix, yet they are represented within the main clause by an
anaphoric pronoun which takes their case inflection.
Left-dislocated complement:
[Clause + na] ( + TOP) ( P A U S E ) S [PRO + Case . . . ] s
Finally, inconsequential clauses approach complete independence, that is,

they can occur alone as complete, grammatical sentences. Where they occur
with a following main clause, they are separated from this clause by a pause,
and are not represented within this clause by a resumptive pronoun. If such
a pronoun is present, it occurs as a clitic on the inconsequential clause.
Inconsequential:
[Clause + na] ( + TOP) ( + PRO) (PAUSE) s [....] s
244 LORNA MACDONALD
Inconsequential clauses can perhaps be analyzed as reinterpretations of

left-dislocated complement clauses in the nominative case. The nominative
case suffix in Tauya is 0; when complements in the nominative case occur
as left-dislocations, the resumptive pronoun is reduced to a clitic on the
dislocated clause (cf. Section 3):
(70) fena?a Ø - tu - i - na - ?i bundi - sa
woman 3SG give 3PL SUB PRO ADESS
yate - a - ?a
go 3SG IND
'The woman they gave (it) to. (she) went to Bundi'
Thus, whereas dislocated complements in other case relations are
represented within the main clause by an anaphoric pronoun, dislocated
complements in the nominative case have no such representation. This struc-
ture may have been reinterpreted as signifying that the dislocated comple-
ment has no grammatical function defined in terms of the main clause, that
is, it bears no case relation and does not carry a causal inference. This inter-
pretation is like that of inconsequential clauses; note that (70) can receive an
inconsequential interpretation, i.e., They gave (it) to the woman and/but she
went to Bundi.
6 Summary
The evidence presented here suggests that complement and inconsequen-

tial clauses in Tauya are distinct both syntactically and pragmatically, even
though they are similar morphologically and share certain features
characteristic of subordinate clauses. Inconsequential clauses are more in-
dependent from the following main clause than are complements: they bear
no case relation to the main clause, and are not interpreted as cause clauses;
they are separated from the main clause by a pause, and are not represented
within it by an anaphoric pronoun; and they occur quite frequently as in-
dependent sentences.
Inconsequential clauses are, in some respects, similar to left-dislocated
complement clauses, and are perhaps derived from dislocated complements
in the nominative case. Unlike dislocated complements in other case rela-
tions, dislocated nominative complements have no anaphor within the main
clause. This absence of morphological representation, combined with the
SUBORDINATION IN T A U Y A 245
pause which typically follows all dislocated constituents, may have been
iconically reinterpreted as signalling the pragmatic independence of the
dislocated clause, i.e., it was interpreted as an inconsequential. This
pragmatic independence may have subsequently led to increased syntactic in-
dependence, such that these clauses may occur as independent sentences.
NOTES
1. Research was funded by a grant from the Social Sciences and Humanities Research Coun-
cil of Canada. Field work was carried out in Tauya Village, Madang Province, Papua
New Guinea, from July 1981 to October 1982, with the consent of both the National and
Provincial governments. My principle consultant was the councillor of Tauya Village, a
man between 40 and 50 years old who takes great delight in the complexities of his
language; I would like to take this opportunity to thank him for his generosity and pa-
tience. I would also like to thank the participants of this symposium for their comments,
particularly John Haiman and Ekkehard Konig.
2. The abbreviations used in this paper are as follows:
ERG/INSTR ergative/instrumental
PRO resumptive pronoun
For further abbreviations see the list on page vii.
3. Christian Lehmann has suggested (p.c.) that some of the clauses which I am calling "com-
plement" clauses might be better defined as internal-head relative clauses (see, for exam-
ple, (12)). However, there are also a number of cases in which the internal-head relative
clause analysis appears to be inappropriate (for example, (10) and (11)). Since these
clauses are morphologically and syntactically similar, and thus should be considered as
a single grammatical type, I have chosen to retain the term "complement".
4. -ne occurs as the modal suffix marking non-polar questions everywhere except after the
3sg aorist desinence -a-, where the modal suffix -e is used:
mafi yate - e - ne
where go 1/2 Q
'Where did you go?'
mafi yate - a - e
3SG Q
'Where did he/she go?'
5. As the topic suffix,-ra may occur on only one constituent per clause, and cannot occur
on either final verbs or interrogative pronouns. A few younger speakers do use the topic
suffix on NPs with overt case suffixes, for example,
yapi wate - ?ai - ra
my house ADESS TOP
'to/at my house' (TOP)
246 LORNA MACDONALD
6. One consultant suggests that (29) is a grammatical sentence if the speaker is known for
his/her physical weakness, i.e., if the coordinate clause can be interpreted as a cause
clause.
7. The resumptive pronoun is He/ in underlying form (see for example (46)), but ?i is its
most common surface realization.
8. Exclamations with the form of (69) appear to be used when both the speaker and the ad-
dressee(s) are witnesses to the event. For example, (69) was said by a man to several
others, all of whom witnessed the event. If the speaker is the only witness and wishes to
call attention to the event, the exclamation has the form of a final indicative clause with
the exclamatory suffix added, i.e.,
pomu - a - ?a - e !
fall 3SG IND EXCLAM
'He fell!'
REFERENCES
Haiman, J. 1980. Hua: A Papuan Language of the Eastern Highlands of

New Guinea. Amsterdam: John Benjamins.
Mallinson, G. and J. Blake. 1981. Language Typology. Amsterdam: North-
Holland Publishing Company.
Sequential chaining and discourse structure in
Godie
Lynell Marchese
UBS, Togo
1 Introduction
Within Godie discourse1, there is a peculiar syntactic construction

which I have labelled elsewhere sequential (Marchese, 1978b; 1979). These
clauses carry a special auxiliary marked yi which serves to link sequences of
events along the event line. In the following extract from a Godie folktale
about Spider, Gazelle and Hare, for example, three actions occur with the
sequential auxiliary:
(1) a. k + E 'ku-.
CR he die:PFV
b. 'n+ kwalIE yi-E yI 'kU'61!

and Spider SEQ-him now up take
c. 'n+ O yi 'duu m+
and he SEQ village :DEF-go
d. 'n+ wa yi-E l!
and they SEQ-him eat
'He (Gazelle) died and Spider SEQ-picked him up and SEQ-
went to the village and they SEQ-ate him'
From several points of view, the analysis of this construction is problematic.
Practically, in trying to edit stories in Godie for publication, it has never
been quite clear where, within the action chain, it is necessary to break for
punctuation. Analytically, the syntactic status of the construction is not easy
248 LYNELL MARCHESE
to define. The clause appears to be awkwardly between a 'subordinate' and

'main' clause. From a discourse perspective, the use of the clause seems at
first sporadic and unpredictable. The goal of this paper is to address these
issues: to examine the characteristics of this chaining device and to determine
what motivates speakers to encode sentences in yi sequential clauses.
2 The data base
The data used in this study come from oral recordings made by Carol
Brinneman and myself in southwestern Ivory Coast between 1972 and 1977.
The texts were transcribed and translated with the help of Zadi Sassi Michel,
from the village of Dakpadou. The data, which represent over 100 pages of
typed text, include various discourse types: folktales, songs, prayers, pro-
verbs, riddles, personal and fictional narratives, histories, and procedural.
These were provided by five speakers coming from two different Godie dia-
lects: jl-kO and kagbo (see Appendix 1).
3 Grammatical facts about clauses containing yi
While unmarked clauses in Godie are typically SVO as illustrated in (3a)

and (b) below, clauses containing yi are always OV, as seen in examples (b),
(c), and (d) above. This switch in word order is triggered by any auxiliary
in Godie including futures, negatives, conditionals and perfects and is a typi-
cal phenomenon in Kru languages (Marchese, 1979a; b). Like other auxilia-
ries in Godie, yi can be accompanied by a tense marker (either recent a or
remote wG), but it can never be inflected for aspect (such as the perfective-
imperfective distinction seen in (la).
Interestingly, the sequential auxiliary is homophonous with a potential
future auxiliary yi derived from the verb 'come' (Marchese, 1979b). Histori-
cally it is very likely that the sequential auxiliary is also derived from the verb
'come', having developed either directly from the main verb or indirectly
from the future auxiliary. Whether or not a clause containing yi has a future
or past sequential reading depends on the context. Thus a sentence contai-
ning yi initially will have a future reading as will all following clauses with
yi. If the preceding clause is in the perfective, however, yi will receive a past
tense reading:
SEQUENTIAL CHAINING IN GODIE 249
(2) a. O yi yi 'n+ O yi 1!
he FUT come and he FUT eat
'He will come and eat'
b. O -yi 'n+ O yi l!
he come:PFV and he SEQ eat
'He came and then he ate'
Thus yi looks back to the preceding clause for its time reference and from
this point of view, the two yi auxiliaries appear to be one and the same. They
represent actions which occur in sequence to each other. The auxiliaries can
be said to be distinct, however, inasmuch as they negate differently. The fu-
ture auxiliaryi/ has its own negative form 'naa, while the sequential yi (with
past reading) really has no negative counterpart. Future yi occurs sentence-
initially, while sequential yi typically does not. Furthermore, the future yi is
almost never found in narrative discourse (outside of dialogue), while the
past sequential yi — if not abundant — at least plays a major role in this
discourse type. Thus, the remainder of this paper will only make reference
to this past sequential yi which most often follows clauses containing perfec-
tive aspect.
Like perfectives, sequential appear to signal foregrounded or 'on line'
actions which move the story along. In a high percentage of cases (see (lb,
c, d) for example), clauses containing a sequential yi are preceded by the
conjunction n + 'and (then)' which is the most common linker of clauses oc-
curring in temporal sequence2. Clauses occur in the same order as they do
in the real world. They can never be switched around and maintain the same
meaning.
The presence of the yi sequential apparently has nothing to do with
switch reference. Example (1) shows that clauses with yi can include a change
of subject (examples (b) and (d)) or the same subject (example (c)). Later ex-
amples will serve to confirm this observation. However, it can be noted that
subjects in yi-constructions are always known, and may include personal
names (line (b)), pronouns (line (c)), definite names, or known entities.
Clauses in Godie may normally include topicalized or focussed NPs.
Topics appear sentence-initially and are replaced in the following clause by
a simple recapitulative pronoun. Focussed subject NPs, on the other hand,
are replaced by special focus pronouns:
250 LYNELL MARCHESE
(3) a. topic
ngw1OO O -6a ngw*dIO y*ku
womanrDEF she leave:PFv man:DEF side
'The woman, she left (her) husband'
b. focus
ngwlOO OmO -6a ngw*dlO y*ku
woman:DEF she:FOC leave:PFv man:DEF side
'It's the woman who left (her) husband'
Clauses containing yi may include topicalized material, occurring directly
before the subject pronoun (4) or extracted to the left of the conjunction (5):
(4) 'n+ wame wa yi m +
and them they SEQ go
TOP
'And them, they SEQ-went...'
(5) 'ngwlOO 'n+ O yi ngw*dIO y*ku-6a
woman:DEF and she SEQ man:DEF side leave
TOP
'And the woman, she SEQ-left (her) husband'
In contrast, in the entire data base, no focussed subjects have been found
in association with the yi sequential. This is a further illustration of the fact
that clauses containing ye contain only known subjects. They do not contain
new or highlighted information in subject position. Object complements in
yi-constructions are also typically introduced earlier in the discourse or are
so well known that they are already definite.
Most verbs occurring in such sequence chains are action- oriented. In
fact, on the average, 47% of all verbs occurring in sequential constructions
are verbs of movement ('go', 'come', 'descend', etc.), a fact which will
become significant in later discussions.
The sequential auxiliary is not unique to Godie; it appears in one form
or another in several Kru languages, where it often has a separate form from
the future auxiliary (Marchese 1979b: 239-240, 268-273). In a short paper on
Nyabwa, Bentinck (1979) attempts to determine the grammatical status of
clauses containing je, obviously a cognate auxiliary. Her main goal is to
determine whether the clause is subordinate or main, and she comes to the
conclusion that it is subordinate since it can never occur by itself. She cites
the following examples:
(6) Nyabwa
a. O na-o-o
he drink-AM-DEC
'He drank'
b. O li-o-o le O je na
he eat-AM-DEC then he SEQ drink
'He ate and then he sEQ-drank'
She notes, nevertheless, that a clause containing je can function as the main
clause of a sentence when preceded by a 'true' subordinate. Similar facts ob-
tain in Godie: clauses containing sequential yi (with a non-future reading)
never occur independently, though they may occur in what appears to be the
'main' clause, when preceded by sentence-initial temporal and/or relative
clauses:
(7) z + kpO pii, zl -zl nG,
morning early day dawn:PFv NF
'n+ -a yi gwall 'kU '61!
and we SEQ canoes up take.
'Early in the morning, when the day had dawned, we SEQ-took
the canoes...'
The question of whether Godie yi sequential are 'subordinate' or'main'
remains unanswered (and perhaps underlines the issue raised by Thompson
and Haiman (1984) as to the usefulness of such a distinction). It makes more
sense perhaps to ask how this clause compares to others in terms of certain
'subordinate' features (Marchese, 1978a). The sequential clause in Godie
does not function as the constituent of a higher clause, i.e. it is not the com-
plement of some higher verb; yet it curiously depends on the verb of the
preceding clause for its time reference. Speaking in Lehmann's terms (this
volume), there appears to be the 'faint beginning of hierarchical
downgrading and of subordination'. The sequential clause itself is still quite
sentential, though we begin to see restrictions on its syntax. As noted above,
it allows topicalization, but not subject focussing. Its word order is restricted
to OV. It would appear to have a finite verb (in this case, an auxiliary),
since it can be inflected for tense, but it cannot be negated. It can assume
the position and apparently the function of a 'main' clause as seen above,
252 LYNELL MARCHESE
and yet its position is restricted to non-initial environments. Finally, as we

will see later, sequential clauses encode main on-line events — not a very
typical place for a subordinate clause (Labov, 1982; Reinhart, to appear;
Thompson, 1987).
Going beyond the issue of 'subordinate' and 'main', another important
question seems to be: why is this construction used at all? Why are certain
actions linked together by the 'n + conjunction and encoded in clauses with
yi sequentials and others not?
4 Function of yi clauses in discourse
When we look at a wide data base including folktales, songs, prayers,

proverbs, procedurals, riddles, personal and fictional narratives, as well as
histories, it becomes obvious that yi sequentials do not occur in all discourse
types. They are extremely frequent in anything narrative-like, and thus the
greatest percentage occur in personal and fictional narratives, including
histories. They are infrequent in prayers and non-existent (in my data base)
in proverbs, songs, and procedurals 3 . Where riddles and folktales lend
themselves to recounting specific events taking place on a time line, yi se-
quentials are also found.
When attempting to analyze the function of the yi-constructions within
discourse, I first tried to analyze data preceding them, thinking that they
would signal same subject or different subject, parallelling consecutive con-
structions in other West African languages (Wiesemann, Nseme et Vallette,
1984). When this proved unfruitful, I gave my attention to what followed
in context. It is this environment that began to reveal the real function of
yi sequentials. From examining material immediately following the last
clause of a yi sequence, we can establish a list of contexts where yi sequen-
tials are most likely to occur. Normally, they may precede repetitive tem-
poral clauses which serve to introduce a new time frame (8), relative clauses
which introduce new participants (9), dialogue (whether or not it is overtly
introduced) (10), and finally conjunction-like words that function as
paragraph markers (11):
(8) chaining repetitive clauses
...'n+ wa yi 1c 'klc m+.
and they SEQ PTCL field go
> wa -m+ yI nc,

they go:PFv now NF
'And they SEQ-went to the field.
>Having gone....'
(9) relative clauses introducing new participants
... 'n + E yi m + .
and he SEQ go
>-0 -ka gbesuu nc, yuosE soli ki SO pl+ fIO
REL have gun:DEF NF almost week x 2 pass after
'And he (the panther) SEQ-left.
> H e who had the gun, almost two weeks passed,...'
(10) dialogue
a. overtly introduced:
...'n+ O yi -LaagO y*ku m + .
and he SEQ God side go
> O lc : " "
he said " "
'And he SEQ-went to (see) God.
> He said '....'
b. not overtly introduced
... 'n+ wa yi ngwu
and they SEQ roast
'n+ wa yi l!.
and they SEQ eat
>"'OO- -A yC 'nu.
Oh! we PF understand
'And then they SEQ-roasted and SEQ-ate (the bananas).
> ' O h ! We've understood...'
(11) particle used as a paragraph marker
...'n+ E yi zizio.
and he SEQ hide
> -BEE, -kOkwIO -ya 'O 'yua 'sO
Now Chicken and her children together
wa -kU -mOO -6utuu
they are there house:DEF
254 LYNELL MARCHESE
'And he (Eagle) SEQ-hid.

> Now chicken and her children, they were there in the
house'
These four contexts all have one thing in common. They serve to break the
discourse into significant units (marked in the text with the symbol > ) . Tem-
poral and relative clauses (as seen in (8) and (9)) as well as conditionals in
Godie often play the role of 'setting the scene' for an upcoming discourse
(Marchese, 1977, 1987). The 'breaking' effect of such subordinate clauses
has been attested in other languages as well (Givon, 1983). Dialogue, as seen
in (10), is also known to be discontinuous (Carlson, MS)4. Particles or con-
junctions such as the one in (11) are also common signals of discourse breaks
(Gratrix, 1978). In fact, looking at the above examples, we see many signs
of discontinuity immediately following the last yi clause. Example (9) shows
not only a new participant, but a shift in time. Example (11) includes a
discourse marker -bEE as well as a change in participants signalled by a full
noun phrase. Example (8) shows a subordinate clause which suggests
changes in both time and location.
In each case above, then, the end of a yi sequence coincides with the
beginning of a new discourse unit. What is the nature of this discourse unit?
It appears to be similar to the notion of paragraph, i.e. a cohesive unit higher
than the sentence which constitutes a 'building block' for a larger discourse5.
In narratives, paragraphs are typically signalled by changes in place, time,
and/or participants. Thus yi is one of the signals that a paragraph is winding
down and that a new one is beginning. What this means is that the speaker
has devices at his/her disposal to mark the end as well as the beginning of
a discourse chunk. Similar facts emerge from other languages. In related
Wobe, short, quick actions with no overt linkage show that a paragraph is
coming to a close (Link, 1978: 235). In Kaingang from Brazil, the lack of
morphological marking on the verb signals end of paragraph (Wiesemann,
et al., 1984: 107). In Biblical Hebrew, the formula of "rising and up and go-
ing to a different place...is one of the prevalent...conventions for marking
the end of a narrative segment". Apparently Godie makes use of a similar
device at least part of the time through the frequent use of movement verbs
within the yi-construction. As noted above, in this data base, almost half
(47%) of all yi clauses contain verbs of movement.
To show a typical paragraph structure, we can examine an extract from
a riddle. Three men (all previously introduced) are on the scene; one has a
pipe, one tobacco, and one matches. The paragraph begins with an introduc-
tory relative clause (functioning as topic (line a)), followed by two clauses
with normal VO order in the perfective aspect (lines (b) and (c)). After this
we have a series of eight actions — all marked by sequential yi (lines (d)-(k)),
until the close of the paragraph. The next paragraph begins with a temporal
clause referring back to the events in the preceding unit (line (1)):
(12) kagbo dialect

a. > -O -kC dikie na,
REL have:PFv pipe:DEF NF
b. kC O nyEE- d+ +d+kCnyOO
CR he give:PFV tobacco:have:person:DEF
c. EmEd++dlk+ dIkCnyOO 6l! yI IC 'kU
this tobacco:pers:DEF take:PFv now PTCL up
d. 'ni O yi-a yi lG 'm+ d++d+
and he SEQ-it now PTCL inside cut
e. 'nI O yi-a yI 'm + 'ngwU
and he SEQ-it now inside put
f. 'nI maclkCnyOO yiO yI 'O macIE 'nyE
and matches:PERS SEQ-him now his matches give
g. 'nI d+ +d+kCnyO yi 'mlC

and tobacco:pers SEQ drink
h. 'nl O yi-a dikikCnyO 'nyE

and he SEQ-it pipe:PERS:DEF give
i. 'nI O yi-a yI 'mlC
and he SEQ-it now drink
j . 'nI dikikCnyO yi-a yI -macIkCnyOO 'nyE
and pipe:PERS SEQ-it now match :PERS give
k. 'nI O yi-a yI mlC
and he SEQ-it now drink
1. > Wa 6IC yI lC mlClI nC...
they finish:PERF now PTCL drink.'NOMI NF...
256 LYNELL MARCHESE
> 'The one who had the pipe, he gave (it) to the tobacco
owner. It's that (the pipe) that the tobacco owner took and
he SEQ-cut it (tobacco), and he sEQ-put it inside, and the
match-owner SEQ-gave him his matches and the tobacco-
owner SEQ-smoked and he SEQ-gave it to the pipe-owner and
he SEQ-smoked it and the pipe-owner SEQ-gave it to the
match-owner and he SEQ-smoked it.
>When they had finished smoking...'
In some sense, the ye-sequential conveys a quickness of action, or the
impression that one event occurred immediately after another with no
significant breaks in time. Its use may be related to the discourse use of serial
constructions, which, while not occurring in Kru, are present in most of the
languages surrounding the Kru region. As Noss (MS: 16) points out for
Gbaya, a Camerounian language belonging to the Adamawa group, in
serials "...the action is stated with an economy of words that heightens
dramatic impact. The development of movement proceeds quickly and effi-
ciently without breaks in the action..." Yi-chains in Godie appear to have
a similar effect.
It is perhaps important to note that within the paragraph itself, sequen-
tials often follow a clause expressing a significant event, frequently marked
by the current relevance marker k + (or kC in the kagbo dialect). This
marker singles out actions which will have a significant effect on the rest of
the story and usually occurs at story-climax. Thus it seems to correspond to
the definition given to 'pivotal event' by Jones and Jones (1979). Within the
paragraph, k+ often occurs as the first main clause and serves as the begin-
ning of the action chain, as seen in (12b) above. A similar example comes
from a riddle about three men with miraculous powers:
(13) a. TanalOO, sC O 'kC 6(J nU?
third: one how he FUT Q do
b. k+ O 6l! ' O 'cIclIE 'kU
CR he take:PFv his cloth?6 up
c. 'n+ O yi-e'nyie 'klI lUlU
and he SEQ-it river :DEF face cover
d. 'n+ O yi-e naml +
and he SEQ-it walk
e. 'n + O yi 'tO.
and he SEQ cross
f. OO -wO 'sII- -wOtO.
he:NEG NEG also get wet
T h e third one, how would he do (it)? He took his, cloth
and he SEQ-covered the river and he SEQ-walked on it and he
SEQ-crossed. He also didn't get wet'
Here again it seems that a k + clause with SVO order and a perfective aspect
verb 'triggers' the sequential clause chain. Interestingly, in this example, the
speaker inserts an evalative statement at the end of the paragraph, a kind
of 'double signal' that the paragraph has concluded.
We have noted that yi sequences signal the end of the paragraph unit.
Yi may also function at a higher level, signalling the break of a unit larger
than a paragraph. In many instances yi is used to signal that the end of the
action in the story is at hand and thus serves to divide off macro-units such
as the body of a text from its conclusion. This 'higher level' use of yi is found
in several discourse genres. In a narrative-like folktale, a child renounces his
father. When he gets into trouble, only his father will help him. As the body
of the story comes to an end, a sequence of yi clauses occurs. Note also the
presence of the wC remote tense marker which characterizes the end of a
story (Marchese, 1978b). Immediately following these clauses, there is a ma-
jor break. Now the moral of the story is presented:
(14) a. ...'n + wa yi m +
and they SEQ go
b. 'n+ wa yi mlaa dII
and they SEQ meat:DEF cut
c. 'n+ wa ny'I 'sO yi wC yi
and they and-it two SEQ REM come
END OF ACTION — BODY
MORAL
> > (d)'I zE ny + kpO zIdI 'n+ O gC....
its reason man ?6 and he has:child
258 LYNELL MARCHESE
'... and they SEQ-went, and they SEQ-cut they meat and the
SEQ-came with it.
> > That's why, if a man has a child...'
Thus, the final yi (line (c)) marks the end of the story, while in line (d), the
speaker goes on to give the moral of the story, stating that one should never
renounce one's own parents.
Similarly, in a personal narrative about two schoolboys seeing an albino
for the first time, the story proper winds down with a sequential. Again, here
the remote tense marker also signals the end of the story. Another marker,
naa, provides a further clue that a new unit is about to start. Thus, the con-
tinuity of the story line is broken in line (a) and the speaker moves on to his
concluding remarks. As if often the case in Godie, the speaker 'signs off:
(15) a. 'n+ -a yi wC Zareko nyu mlC kC yi
and we SEQ REM Zareko water drink PURP come
END OF ACTION
CONCLUDING REMARKS
b. Naa' am(7 walI kl* nC, amC n nii- nC
Now that word piece DEM that I see:PFv NF
c. 7 zE n sC-a -lC 'mO
its reason I take-it PTCL there
d 'lO n '6ll!-
there I stop
'...and we SEQ-came to Zareko to drink water.
> > Now, this little story, I found it, that's why I'm tell-
ing it... There I stop'
One could argue, of course, that these macro-divisions are also
paragraph divisions, which indeed they are. What is crucial here, though, is
that yi clauses also occur as the story proper winds down. Many other ex-
amples of this phenomenon could be given. Roughly six texts out of twenty
in this data base use the sequential to mark the close of the story and/or the
beginning of the conclusion section. Thus again we see clause linking with
sequentials as part of the 'winding down' process of discourse, be it at the
level of paragraph or at a higher level, separating the body from the conclu-
sion. So the use of yi-chains is not limited to signalling quickness of action.

There is strong evidence that the sequential chain plays a very crucial role
in defining the structure of a given discourse.
Sometimes speakers choose to by-pass the typical division between the
body of the story and the conclusion. Or to put it another way, some
speakers may at times leave off the conclusion. When this is the case, yi will
be used to signal the end of the entire story and may often encode a kind
of story conclusion or resume. A good example comes from the story of
Elephant and Whale. This story is meant to teach why the elephant is on land
and why the whale is in the ocean (in fact, the Godie name for 'whale' is
"ocean-elephant"). The last actions are as follows, as the two fight a serious
tug-of-war:
(16) a. KUIlUE m+ k + E ngwU sII kpanyI fI
Elephant FOC CR-he put:PFv also much force
b. 'nI 6liyee yi dU
and rope:DEF SEQ cut
c. 'nI gilOE yi mOO gle 6l+
and Whale SEQ there ocean fall
d. 'nI klIlUE yi mOO klI too
and Elephant SEQ there bush stay
'It's Elephant who used much strength and the rope SEQ-cut
and Whale SEQ-fell in the ocean and Elephant SEQ-stayed in
the bush'
In this case, the final actions of the story turn out to be the most signifi-
cant. A similar style is found in story about Spider, Hare and Gazelle, where
the last line encodes the most important action, the final revenge. As noted
elsewhere (Marchese, 1986), the last sentence of some folktales resembles a
stage scene, where all the characters come out in full noun form to 'take a
bow'. This can be seen in example (17) below, where the remote tense marker
wC also surfaces:
(17) ..'n + kpOkpOnmU yi-O wC kwalIE 6lC-a
and Hare SEQ-him7 REM Spider kill-EMPH
'And Hare SEQ-killed Spider!!!!!'
Despite the fact that this clause is 'packed' with tense and full noun phrases,
260 LYNELL MARCHESE
it is still clear that only one thing is highlighted here — the killing.
Everything else is known from the beginning of the story, and the ap-
pearance of these elements story-finally is a typical Godie way to show that
the story has drawn to a close.
The occurrence of yi with full noun phrases and tense-marking is also
found story-finally in a folktale about Spider and two blind women. Spider
has tricked these poor women out of their food. In the final scene, God takes
his revenge:
(18) a. 'n+ wa yi p + pElIO
and they SEQ crawl
b. "n + wa yi 'mO '6a
and they SEQ there leave
c. 'n + LaagOtEpE yi z+p + a ninie
and God SEQ world.:DEF surprise
d. 'n+ 'cCkUU U yi wC KwalIE 'kU 6l!
and stone :DEF it SEQ REM Spider on fall
'And they (the women) SEQ-crawled and they SEQ-Ieft there
and God SEQ-surprised the world and the stone, it SEQ-killed
Spider'
Again the last lines (c) and (d) bring out the two main antagonists (God and
Spider) and the main prop (the stone) in full noun form, as well as the time
setting (in remote tense). Nevertheless, the main impact occurs in line (d) en-
coded in the verb 6l! 'fall.
5 Frequence and distribution of yi sequentials
We have said that yi-sequentials are used to wind down paragraphs and
the story body. When yi sequentials occur, how long are the sequence
chains? There is quite a lot of variation between speakers and even speakers
themselves vary from story to story. The longest chain in the present data
base is 9 clauses long (counting only yi clauses and not those which begin
the chain). The smallest is 1. On the average, the usual length of a chain is
1.68 (see Appendix 2). The average number of clauses per story which are
given to sequentials is around 10% (see Appendix 3). Note however that
some speakers are heavy users of yi. Speaker 2, for example, uses yi clauses
in an average of 23% of his texts, while Speaker 1 uses them much more
sparingly. On the average, only 4% of his total clauses contain a yi sequen-
tial.
It is possible that within the narrative genre, the specific type of nar-
rative (for example, folktale, history, personal narrative) may play a role in
the amount of sequentials found. Because of lack of data, I have been
unable to determine whether this is, in fact, the case. There is no hard
evidence showing that one narrative type has more sequentials than another.
Statistics in Appendix 3, however, give some weak evidence that speaker 3
uses yi more frequently in personal narratives than in folktales. This is a
matter for further study.
The facts discussed above show that in reality, yi clauses make up only
a very low percentage of the clauses in a given text. For example, the
sentences in (13) represent every case of yi found in the entire text (itself
made up of 83 clauses). How then are the yi sequentials distributed? We have
defined the environment where they often occur (8 - 11), but have neglected
to say they do not automatically occur every time one of these environments
is present8. Studying each text as a whole, it becomes obvious that yi sequen-
tials tend to occur more often in certain sections of the discourse than in
others. We have already seen that yi-sequentials are usually excluded from
conclusions and morals of stories. They are equally excluded from any in-
troductory material. So looking at the overall structure of a discourse we ex-
pect to find sequentials in the body. In the riddle seen in (12) about the three
men with miraculous powers, the structure is roughly the following:
Section 1 Setting Clauses 1-8
Section 2 Problem set forth Clauses 9-12
Section 3 1st action episode Clauses 13-25
Section 4 2nd action episode Clauses 26-34
Section 5 3rd action episode Clauses 35-40
Section 6 Conclusion: Clauses 41-42
problem presented
Yi sequentials will quite naturally only occur in sections 3-5, that is, in the
action episodes and nowhere else.
One common place where clauses tend to be linked by yi sequentials is
immediately preceding the story climax. In a folktale, Spider is confronted
with a problem. There is a genie (a spiritual being) on his back, and he can't
get him off. Finally an old woman gives him advice on how to get rid of him.
262 LYNELL MARCHESE
In the actions leading up to the genie's descent, we find clauses encoded with
the sequential:
(19) a. > Ma, clcEsC, kwallE -yi nC
But truly Spider come:PFv NF
b. yErEE EmE O-su yI
pepper:DEF that he crush:PFV now
c. 'n+ O yi mO booo ngwu

and he SEQ there bowl:DEF put
d. 'n+ O yi 'mlC
and he SEQ drink
e. > Ma, O kC kC-O gl+ 'O kO nC
But, he FUT CR-he look his back NF
f. 'Tuu-!! O yC-zO 6l+
BOOM he PF down fall
'Now really, Spider, having come, it's pepper he crushed
and he SEQ-put it in a bowl and he SEQ-drank it.
>And as he looked on his back, BOOM, he (the genie)
has fallen down!!'
It is interesting here to study the rhythm of the clause flow. Subordinate

clauses (which often contain secondary or repeated information) serve to
slow down the flow of discourse. Pauses following subordinates are quite
long and often the speaker draws out the non-final marker nC, sometimes
even waiting at this point for a verbal response from the hearer (Marchese
1977; 1987); Yi clauses, on the other hand, are pronounced quickly,
one after the other, and the change or rhythm (in relation to the rest of the
text) is quite noticeable. The pause between clauses containing yi is less than
between clauses without them. It sounds as if the speaker is waiting to finish
the chain before 'coming up for air'. Thus the above sequence has a definite
rhythm:
slow (a)
fast (b)
quicker (c)
quicker (d)
slow (e)
climax — end (f)
This corresponds to findings by Longacre (1985:85-86) who suggests that

phonological features such as an increase in the rate of utterance is a way
of indicating that the story peak (what I am calling here climax) is about to
occur. It also shows other devices in Godie which can be used to create a
'zone of turbulence' around the peak area. Note the short lull (line (e))
before the climax, which Longacre terms ''slowing the camera down". The
presence of an ideophone 'tuu- and a switch to the perfect auxiliary in line
(f) also makes this climactic clause stand out from its context.
Yi is found in a similar environment in a folktale about Viper and his
wife-stealing enemy, Eagle, told by the same speaker in (19). In the entire
story one yi clause occurs. This is when the action of the story is in full
swing, immediately preceding the climax of the story (line (c)):
(20) a. Gopl+ a -tlO 'mO 'a 'wluu nyl-6l+a 'ml +
Eagle he pierce:PFv there his head:DEF water-deep inside
b. 'n+ a yi yI 'mlC
and he SEQ now drink
c. ≥ -Ma sOOnadaa na, k+ E 6l!-a 'kU
But second:time NF CR he catch.'PFv -him up
T h e eagle pierced his head through the deep water and he
SEQ-drank.
> But the second time, he (Viper) grabbed him!'
In this extract, the other signals of climax are the slowing down provided by
the temporal clause ('but the second time'), and the appearance of the cur
rent relevance marker 9 . It seems significant that in the entire data base, the
k+ marker never occurs in the same clause with a sequential yi10, confirming
that each has a separate function in the discourse. Yi leads up to the climax,
while k+ may actually mark it.
In a riddle concerning the appearance of a genie which captures people
in a canoe, again only one sequential yi occurs. It appears paragraph-finally,
where it immediately precedes the most suspenseful and exciting moment:
(21) a. >Ma gOlUU wa -plC nC
Now canoe :DEF they enter:PFV NF
b. wa lu -kaa 'nyie n+ +d+n+ +d+
they paddle :PFV until river :DEF middle
264 LYNELL MARCHESE
c. 'n-h mOO 'nyie zl+a yi 'kU -6a

and there river :DEF genie :DEF SEQ up come out
d. >-Ma a -m + 'kU -6a -kC nC,
But he go:PFv up come out PURP NF
e. k+ a -61! gOlUU 'kU
CR he take:PFv canoe:DEF up
> 'Now having entered the canoe, they paddled until the
middle of the river and there the genie of the river SEQ-came
out.
>His having come out, he grabbed the canoe!'
As is expected, the yi clause occurs after a perfective aspect SVO
sentence with a known subject. It occurs paragraph-finally before a repeti-
tive chaining clause (introducing the new paragraph). And it occurs prior to
the climax (line (e)). However here we note another fact about yi. That is
that it not only leads up to the climax, but seems to introduce the second
most exciting line in the story as the tension and suspense are being built up.
Thus it can be concluded that yi does not occur indiscriminately every
time there are sequences of actions. It appears when actions are highlighted,
when they are in focus. It could be said yi-clauses tend to take the focus off
anything else in the clause (aspects, arguments, polarity) and concentrate on
the action involved. As noted earlier, subjects in yi-constructions are always
known (explaining why topicalization can occur, and not focussing).
Likewise objects in yi-clauses are almost always previously introduced
((1 b,d), (5), 12 b,c,e,f,h,i,j,k), (13 c,d), (14 b,c)) or seen as definite ((lc),
(8), (10), (19c)). Adjuncts (locations, purpose phrases, temporal phrases,
etc.) are uncommon in this environment; if they do occur, as in (15), they
have already been introduced or are taken for granted. We have already seen
that the sequential construction has no negative counterpart, and that its
aspect is 'frozen', being completely determined by the preceding context. In-
terestingly, these grammatical restrictions of yi clauses seem to be the result
of the function these clauses play in discourse, i.e. the highlighting of action.
If we were to make a scale of the foregrounding devices in Godie, then,
it appears that there are at least three distinctions being made:
pivotal events those events which are highly significant

(marked by k +) to the plot or theme of the story, including
the climax11
sequential events those events which are singled out to put
(marked by yi) emphasis entirely on the action, with the
special discourse function of building
suspense prior to climax and signalling end
of paragraph
perfective verbs those events which typically convey 'main
(marked by low tone) line' information, which move the story
along
All three of these markings may be used several times in the text. Thus,
there are often several events marked by k+ within a given discourse. In the
same way, yi-chaining may affect more than one area of a text. In the riddle
about the smokers (example (12)), there is an introduction, three episodes
and a conclusion. In the first two episodes, the characters do nothing except
discuss their dilemma, while in the third episode the real action takes off.
These actions are highlighted by a series of yi's. No other yi sequential oc-
cur until the riddle has almost come to an end, when again yi occurs
paragraph-finally — just preceding another crucual action in the story:
(22) a. ...-lOO O ngwUU- yI 'dikie 'E tUtUU
...there he put.'PFv now pipe:DEF its ashes:DEF
b. 'n+ wa yi m+.
and they SEQ go
c. >Wa-m+ -mOO-6a -kC nC,
they go:PFv there leave PURP NF
d. nglOgbanyO -kU -lOO -blO 'kU.
girl is there road on
There (on the road) he put the ashes from the pipe and
they SEQ went (away).
>When they had left, a girl was on the road.'
Line (d) is extremely significant. The ashes (which came as a result of the
actions in (12)) have magically turned into a woman of marriageable age. In
the first part of the riddle, the speaker helps the audience figure out how the
three men will cooperate so that each will be able to smoke. The question
266 LYNELL MARCHESE
to be solved in this part of the riddle (which is left for the audience to solve)
is which one of the three men will get to marry the woman who emerges from
the ashes. Thus, it is appropriate that line (b) contain a yi, as it leads up to
a significant point in the story. The repetitive clause in (c) serves, as we have
seen, to draw out the tension and suspense — to make the hearer wait —
to discover the big surprise: the ashes have turned into a woman.
6 Conclusions
What conclusions can be made about the encoding of events in sequen-

tial chains? Why do speakers 'clump together' certain events with this
device? They apparently combine events together when they want to focus
in on punctilear events, highlighting their quickness and/or the irrelevance
of time between them. They use sequential to wind down the paragraph,
giving a distinct signal that the unit has come to a close. But they do not use
the sequential to signal the end of every paragraph. Within the story they
use the clauses in the yi chain to set up expectations for the story climax.
They use them to point to significant conclusions at story-end, and they use
them to signal macro-level divisions, as when the actions of the story proper
draw to an end.
How should these chains be treated in editing texts for publication?
Should there be a division (i.e. a period) after each clause, as has occurred
in some story publications? Probably not. It would appear that sentences in
Godie can be made up of up to 9 clauses (as exhibited in this data base) and
there is no reason to think that such a unit should be artificially shortened
or divided because of Western biases.
Are these clauses subordinate? We have seen that in a very real sense
they are not fully independent. They occur in a fixed position, with a fixed
word order, depend on the preceding clause for their time reference and can-
not be negated. However, without a doubt, they do the job of foregrounding
(they move the story line along), and from the evidence given here, seem to
signal a higher degree of foregrounding than other clearly independent
clauses.
NOTES
1. Godie is an Eastern Kru language spoken in southwestern Ivory Coast. There are three
tones: high, marked by ', mid, which is unmarked and low, -. I have used special symbols
for the vowels. Capital letters represent the retracted set of vowels. /, + , *, and C repre-
sent central vowels, starting with the most closed and moving to the most open. 6
represents an implosive b.
Abbreviations used in this study include:
AM assertive marker
CR current relevance
NF non-final marker
PERS person
PF perfect auxiliary
REC recent past marker
REM remote past marker
SEQ sequential auxiliary
For further abbreviations see list on page vii.
I am grateful to Christian Lehmann, Phil Noss, and Sandra Thompson for their
helpful comments on an earlier version of this paper. I would also like to thank Sassi
Michel, Dapl+ Joseph, and K + ku Alphonse for sharing their stories with me and at the
same time express my gratitude to two great story tellers who have since passed away:
Natche Jean and LugbO Aye Grattoir.
2. 'N+ is the temporal marker per se, but is has other functions and may refer to logical
relationships rather than temporal ones.
3. In Godie procedural, the future auxiliary yi does occur, however, but only infrequently.
The most common modals are the conditional k+ (occurring in 79.6% of all subordinate
clauses and 25.6% of all clauses) and the volitive future 'kC (occurring in 45% of all main
clauses). In this context, it has a very special discourse function'. It apparently takes over
the role of the conditional, signalling hypothetical possibiliti ,, since the conditional has
assumed another role: that of indicating discourse chunks and important steps in a pro-
cedure (Marchese, 1987).
For example, in a text about how to determine who is guilty of a murder, it is stated
that a bamboo pole (animated by the spirit of the dead man) will point out the murderer.
A series of future yi clauses provides a hypothetical case:
Ny + kpOO yi 'klC zizio -kC m + .
man:DEF FUT bush hide PUR go
GbCgbEEE yi O tCC- -kCm+
stick :DEF it FUT him look for PUR go
'The man can go hide in the bush. The stick (animated by the spirit of the
dead man) will go to look for him'
Thus the future auxiliary in this context signals actions off the main line of the normal
actions in the procedure.
4. In Godie, apparently not all dialogues are equally discontinuous. This is an area for fur-
ther study.
268 LYNELL M A R C H E S E
5. The notion of paragraph is taken as a given in many works (Grimes, 1975; Longacre,
1983). The term itself is not crucial, but the notion is. There is definite 'chunking' of units
which are longer than the sentence. It is possible that yi-sequentials in Godie at times
signal the end of a unit smaller than a paragraph — a unit which might be called an
episode unit. However, it is not yet clear to me that there is a real difference in structure
between this smaller unit and the paragraph. It may be that paragraphs in Godie can be
very short (1-2 clauses) or very long (over 12 clauses). This is an area for further research.
6. A question mark ? indicates that the gloss is uncertain.
7. Here I am unsure of the transcription. I have never seen a repetitive (cataphoric) object
pronoun of this type. It may be a mistake, or it may be a speaker's hesitation, as he inserts
a pronoun object and then opts to put in the full noun phrase.
b. My analysis does not allow me to predict absolutely where yi sequences will occur (see
Hinds, 1979 and Marchese, 1987, for similar observations). The speaker obviously has
many stylistic devices at his/her disposal and yi is only one of them.
9. Note that there is not one set way in Godie to encode a story climax. I have noted here:
the presence of k+ on the clause, preceding clauses marked by yi, 'slowing down the
camera' (by a subordinate clause or temporal phrase) just before the climax, a change in
aspect (example (19)), and the appearance of ideophones. Another typical signal is a
reduction in the number of full noun phrases. Thus clauses marking climax tend to have
pronominal subjects and objects. These are the devices I have found to date. There may
be more. The existence of such a wide range of devices confirms Longacre's (1983:29)
observation that a language may have a 'bag of tricks' from which it picks and chooses
to designate high points in a narrative.
10. Though the k + current relevance marker does not cooccur with the yi sequential, it cooc-
curs with several aspectual/modal elements, including future and perfect auxiliaries, as
well as the perfective aspect (Marchese, 1978b).
11. The k + current relevance marker is found in discourse genres other than the narrative.
In procedural texts, for example, k+ may cooccur with a future auxiliary to designate
a crucial step or outcome in a process.
REFERENCES
Bentinck, J. 1979 "Les propositions sequentielles en niaboua - propositions

dependantes ou independantes?", Abidjan: CIRL. no. 6
Carlson, R. MS. 1984. "Narrative connectives in Supyire", paper given at
the meeting on African linguistics, UCLA, 1984.
Givon, T. 1983. "Topic continuity in discourse: the functional domain of
switch reference". In Switch-Reference and Universal Grammar,
Haiman, J. and P. Munro (eds). Amsterdam: John Benjamins.
Gratrix, C. 1976. "Godie narrative". In Papers on Discours, J.E. Grimes
(ed), Dallas: SIL.
Grimes, J.E. 1976. The Thread of Discourse. The Hague: Mouton.

Haiman J., and Thompson S., 1984. "'Subordination' in Universal Gram-
mar", BLS 10: 510-523.
Hinds, J. 1979. "Organizational Patterns in Discourse". In Discourse and
Syntax, Syntax and Semantics 12, T. Givon (ed). New York: Academic
Press.
Jones and Jones. 1979. Multiple Levels of Information in Discourse, Dallas:
SIL.
Labov, W. 1982. "Speech actions and reactions in personal narrative". In
Analyzing Discourse: Text and Talk Discourse and, D. Tannen (ed).
Link, C. 1978. "Units in Wobe Discourse". In Papers on Discourse,
Grimes, Dallas: SIL.
Longacre, R.W. 1983. The Grammar of Discourse, New York: Plenum
Press.
-—. 1985. 'Discourse peak as zone of turbulence". In Beyond the Sentence:
Discourse and Sentential Form. J. Wirth (ed), Ann Arbor: Karoma
Publishers.
Marchese, L. 1977. "Subordinate clauses as topics", Studies in African
Linguistics, Supplement 7: 157-164.
-—. 1978a. Subordination en Godie, Abidjan: JIA.
— . 1978b. "Time reference in Godie". In Papers on Discourse, J.E.
Grimes (ed). Dallas: SIL.
— . 1979a. Atlas linguistique kru: essai de typologie. Abidjan: ILA.
-—. 1979b. Tense/Aspect and the Development of Auxiliaries in the Kru
Language Family, UCLA, PhD dissertation.
-—. (1986) "Pronominalization and the appearance of full noun referents in
Godie discourse". In Proceedings of the 14th Conference of African
Linguistics (1983), Gerrit Dimmendaal (ed), Dordrecht: Foris.
-—. 1987. "On the role of conditionals in Godie procedural discourse". In
Coherence and Grounding in Discourse, R. Tomlin (ed), Amsterdam:
John Benjamins.
Noss, P. 1985. MS. "Translation and the African Oral Tale", paper given
at the 5th annual international conference on African literature and the
English language, University of Calabar, Nigeria.
Reinhart, T. (to appear) "Principles of Gestalt perception in the temporal
organization of narrative texts". In Poetics Today.
Thompson, S. 1987. "'Subordination' and Narrative Event Structure". In
Coherence and Grounding in Discourse, R. Tomlin (ed), Amsterdam:
270 LYNELL MARCHESE
John Benjamins.
Wiesemann, Nseme and Vallette. 1984. Manuel d'analyse du discours, PRO-
PELCA 26, University of Yaounde/SIL.
Appendix 1
Data on Story Tellers
Speaker Name Origin Age
(at time of recording)
1 LugbO Aye Grattoir Dakpadou 40
2 Dapl + Joseph Goguiboue 50
3 Zadi Sassi Michel Dakpadou 30
4 Natche Jean Dakpadou 70
5 K + kou Alphone Dakpadou 55
6 unknown (text mistakenly not marked)
The data base used in this study has two weaknesses. First, all the speakers
quoted here are male. The second weakness is that most of these speakers
are 'specalized' in a given discourse genre. Thus LugbO Aye has told the ma-
jority of the folktales, and K + kou Alphonse is the only one to have given
a history. It would have been preferable, of course, to have several different
speakers give histories, several give narratives, etc. This tendency towards
specialization reflects a true social situation, however, since different people
are known for being experts in a given discourse type.
Appendix 2
Length of yi sequences
Story Speaker # of clauses #of yi's length of sequences
Riddles
miracle 6 42 11 3,1,1,1,1,1,3
woman 6 31 9 8,1
Folktales
Spider & Elephant 2 62 17 1,1,1,1,1,2,1,
2,1,1,1,3,1
Animals' revenge 6 65 17 1,3,1,6,1,1,3,1
Why genies are in
the bush 1 59 5 1,2,2
Frog, panther,
gazelle 1 63 3 1,1,1
Two blind women 1 70 6 1,1,4
Viper and Eagle 1 43 1 1
Spider and God 2 54 10 1,1,1,1,6
Duck and Zuzu 1 47 3 2,1
Two snakes 1 62 4 4
Eagle and Chicken 1 46 3 1,1,1
Spider and Turtle 4 51 3 2,1
Elephant 3 76 4, 1,1,2
Calao 1 20 0
Chimpanzee 1 17 0
Man who left
his father 1 83 3 3
Narratives
{Personal)
Panther 3 154 3 3
Wife who left
her husband 3 72 11 2,9
Sassi's father 3 45 13 4,1,1,3,1,2,1
Boat 3 44 11 1,1,4,1,1,1,2
Albino 3 47 4 1,1,2
{Historical)
Asekpedou 5 58 3 1,1,1
Dakpadou 5 292 23 1,1,2,2,1,1,1

3,1,1,1,1,1,2,2,
1,1,1
272 LYNELL MARCHESE
Appendix 3
List of Data Base
Statistics on Yi
Genre Speaker # of clauses # of yi % of yi
Folktale
Viper and Eagle 1 43 1 2%
Why genies are in
the bush 59 5 8%
Two blind women 70 6 8.5%
Two snakes 62 4 6.5%
Man who left his
father 83 3 4%
Duck and Zuzu 47 3 6%
Calao 20 0 0%
Chimpanzee 17 0 0%
Frog, Panther,
& Gazelle 63 3 5%
Eagle & Chicken 46 3 6.5%
Woodpecker 30 0 0%
Spider & God 2 54 10 18.5%

Spider & Elephant 2 62 17 27%
Elephants 3 76 4 5%
Spider & Turtle 4 51 3 6%
Animals' Revenge 6 65 17 26%
Riddles
Miracle 6 42 11 26%
Genie 6 11 1 9%
Woman on the road 6 31 9 29%
Narrative
{Personal)
Panther 3 154 7 4.5%
Woman who left
her husband 3 72 11 15%
Sassi's father 3 45 13 29%
Boat 3 44 11 25%
Albino 3 47 4 8.5%
(Historical)
Asekpedou 5 58 3 5%
History of Dakpadou 5 292 23 7.9%
The structure of discourse and 'subordination'1
Christian Matthiessen
University of California, Los Angeles
& Information Sciences Institute
and
Sandra A . Thompson
1 INTRODUCTION
The nature of clause combining, including 'subordination' and its

relationship to 'co-ordination', in grammar has been the subject of much dis-
cussion.2 However, rather little of this discussion has addressed the general
problem of 'subordination' in terms of the structure of the discourse within
which the 'subordinate' clause appears. (Some exceptions include: Chafe
(1984), Dillon (1981:chap. 6), Golkova (1968), Grimes (1975), Halliday and
Hasan (1976), Longacre (1970), Longacre and Thompson (1985), Mithun
(1984), Thompson (1985a), (1985b), (1987) Tomlin (1985), and Winter (1982).)
In this paper we will try to achieve two objectives. First, we will suggest
that it is not possible to define or even characterize 'subordinate clause' in
strictly sentence-level terms. In other words, in order to characterize what it is
that distinguishes a 'subordinate' from a 'main' clause, one must appeal to the
discourse context in which the clause in question appears. To a discourse-
oriented linguist, this might seem self-evident. Yet it is striking that none of
the attempts to define 'subordinate clause' in the literature that we are aware
of has been made in recognition of the consequences of this perspective. Sec-
ond, we will propose and justify a discourse account of the notion of 'subordi-
nate clause'. Specifically, we will show that, in written English discourse, a
certain kind of what linguists have called 'subordinate clauses', namely
'hypotaxis', can usefully be viewed as a grammaticalization of a very general
property of the hierarchical structure of the discourse itself.
276 CHRISTIAN MATTHIESSEN AND SANDRA A. THOMPSON
This work can thus be seen as a contribution to the study of 'natural

grammar': we are proposing an answer to the question 'What discourse func-
tion motivates the grammar of hypotaxis?' But in order for this approach to
yield results, we have to identify the kind of clause combining we are
interested in. As we do this in the next section, it is important to keep in mind
that we are looking for grammatical interpretations of clause combining that
make functional sense, that is, that we can relate to function in discourse.
2 INTERPRETING THE CLAUSE COMBINATIONS TO BE

STUDIED
In discussions of clause combining, there is a good deal of variation both

in terminology and in the substance of different analyses. To avoid any confu-
sion, we want to be very clear about what kind of category we are looking at
and how it differs from other categories involving clauses. We will start by
exemplifying the kind of clause combinations we will be focusing on. We will
then characterize them informally as clause combinations without relying on
any traditional existing categories. The problem we face is that there is no
traditional, generally accepted interpretation that we can rely on. We will first
look at embedding and differentiate it from clause combining, and then look
at clause combining by coordination. Since the kind of clause combining we
wish to focus on cannot be interpreted as combining by coordination, we will
turn to Halliday's work for an additional kind of clause combining that has not
been clearly differentiated in much recent work. After discussing some of the
functions a clause may serve in grammatical structure, we will draw attention
to the need to keep considerations of functions distinct from the assignment of
a clause to a particular grammatical class.
While we will touch on some distinctions we think are crucial in the study
of clauses and clause combinations, we will not try to present a typology of
clause cbmbining or an exhaustive list of the parameters such a typology has
to take into account: That is the task undertaken by Lehmann in his contribu-
tion to this volume.
2.1 Characterization by exemplification
Let us first exemplify the kind of clause combinations we are interested

in. We interpret them as and call them clause combinations, but we don't want
THE STRUCTURE OF DISCOURSE AND 'SUBORDINATION' 277
to give them a specific traditional name, since any name is likely to presup-
pose a particular kind of interpretation. Further, the traditional names for the
clause combinations we want to study all imply interpretations we think are
both grammatically misleading and unhelpful when we try to account for their
discourse function.
I made an appointment with the best hand surgeon in the valley to see if my
working activities were the problem.
... the end result is no use of thumbs if I don't do something now.
While attending Occidental College ... he volunteered at the station as a clas-
sical music announcer.
As your floppy drive writes or reads, a Syncom diskette is working four ways
to keep loose particles and dust from causing soft errors, dropouts.
Before leaving Krishnapur to escort his wife to Calcutta, ... , the Collector
took a strange decision.
Figure 1: Examples of the clause combinations to be discussed
2.2 General characterization of clause combining
The clause combinations in the examples above present propositions that

are related circumstantially. The circumstantial relation is coded by a connec-
tive in one of the clauses in the combination; a number of these are listed in
Figure 2.
circumstantial relation connective
temporal when, while, as
before, after, etc.
conditional if, unless, provided
that, as long as, etc.
reason because, since,
as, etc.
concessive although, even though,
except that, etc.
purpose in order to, so that,
in order that, etc.
means by
manner as if, as, etc.
Figure 2: Connectives coding circumstantial relations
The clause combinations we illustrated in Figure 1 can all be diagrammed

as clauses combined by circumstantial relations. The last example in Figure 1
is represented in Figure 3. There are two circumstantial relations: motivate
glosses a causal relation (purpose) and precede a temporal one. The purpose
relation {motivate) holds between two simple clauses coding simple proposi-
tions. The temporal relation (precede) holds between this combination of
clauses and another, simple clause. The scoping over the combination of
clauses is indicated by the box in the diagram.
Figure 3: Visualization of clause combination

The diagram does not represent an analysis, but is just a visualization of
the combining relation. The main part of our paper will explore what kind of
discourse (text) organization clause combinations of the kind illustrated in the
figure reflect. Before we turn to the nature of discourse (text) organization,
we will explore the grammatical interpretation of the kind of clause combina-
tion visualized in the diagram. We will approach the grammatical analysis in
two steps. First we will distinguish clause combining from embedding, and
then we will ask how we can treat our examples as clause combining without
interpreting them as combinations by coordination.
2.3 Type of function: embedding vs. clause combining
We are concerned with clauses whose function can be stated in relation to

other clauses, in other words we are concerned with combinations of clauses.
2.3.1 Embedding
There is a different kind of environment for a clause, where it does not

combine with another clause:
Those were the days when every single poem had bristled with good qualities
like a hedgehog and had glutted itself with praise like a jackal, the happy
days before the Magistrate had been invited.
(Farrell)
Here the clause before the Magistrate had been invited is embedded within a
noun phrase:
[ the happy days before the Magistrate had been invited ]
This is not a case of clause combining: There is no other clause such that
before the Magistrate had been invited combines with it. Rather, it functions
within a noun phrase as a (post) modifier of the head noun days. It serves the
function a restrictive (defining) relative clause typically serves.
We also have an embedded because clause in the following example, this
time functioning as the complement in a clause:
All the same, the Padre sometimes had a worried look; this was because he
was afraid that the duties to which the Lord had called him might prove too
much for his strength.
(Farrell)
As in the case of embedding within a noun phrase, we do not have a
clause combination here, but simply a case of one clause functioning as a con-
stituent, a complement, within another clause.
There seems to be general agreement that cases of the kind we have just
illustrated are interpretable in terms of embedding.
The examples we listed in Figure 1 have generally been treated as cases of
embedding: One clause is said to be embedded within another clause as an
adverbial (or adjunct); they are often referred to as adverbial clauses. We find
this approach in many traditional grammars (cf. e.g. Jespersen, 1924). But
we also find it more recently in e.g. Quirk et al (1985)3 and in Foley & Van
Valin (1984). A few linguists have taken a different view: Longacre and Halli-
day, as well as other linguists working within the tagmemic and systemic tradi-
tions, have treated examples of the kind listed in Figure 1 not as embeddings
but as clause combinations.4
We are in agreement with this approach: As we have already indicated by
calling them clause combininations, we do not think that our examples can be
interpreted as clauses embedded within other clauses. In other words, these
clauses do not function as adverbials (or adjuncts). A detailed discussion is
beyond the scope of our paper, and we will just make two observations: if we
paraphrase our examples using a prepositional phrase functioning as an adver-
bial, the result is a grammatical metaphor; and we find combinations of one
clause with a combination of clauses. A brief discussion of each of these
observations follows.
i. Paraphrasing leads to grammatical metaphor
Sometimes a substitution test is used to show that a clause functions as an
adverbial in the same way a prepositional phrase does. But the test indicates
that so-called adverbial clauses do not in fact function as adverbials. When we
replace one of them with a prepositional phrase in context, trying to preserve
part of the meaning, we will typically find that the complement of the prepos-
ition is a nominalization, not an ordinary noun, and this is quite significant.
For instance, Before leaving Krishnapur, the Collector took a strange decision
becomes Before his departure from Krishnapur, the Collector took a strange
decision, which is different from e.g. Before noon, the Collector ..., with a
time noun rather than a nominalization. The nominalization departure is a
metaphor, which presents an event as an entity (see Hopper and Thompson,
1984; Halliday, 1985a: ch. 10). This is a marked way of presenting an event:
Rewording the "adverbial clause" with a prepositional phrase to show that it
is an adverbial does not show that at all; it shows that the result of represent-
ing the event of leaving as if it was an adverbial is a metaphor.
ii. Clauses may combine with clause combinations
When one clause combines with just one other clause, it may seem to
function as an adverbial, although it does not. But when one clause combines
with a combination of clauses, it is quite clear that there is no single clause it
could be an embedded constituent part of. Let's consider a fairly complex
example taken from a conversation between parent and child analyzed in Hal-
liday (1985a: 270). The part we want to focus on is in italics:
Child: How do you see what happened long ago before you were born?
Parent: You read about it in books?
Child: No, use a microscope to look back.
Parent: How do you do that?
Child: Well, if you're in a car or you're in an observation coach, you look
back and then you see what happened before but you need a microscope to
see what happened long ago because it's very far away.
The italicized part breaks down into a conditioning disjunction of clauses,
if you're in a car or you're in an observation coach, and a conditioned coordi-
nated sequence, you look back and then you see what happened. The point of
the example is that the condition does not relate to a simple clause but to a
clause combination: There is is no simple clause that the condition could be
analyzed as embedded in. Here are three additional examples, the first two
taken from Longacre (1970) and the third from Halliday (1985b). In all three,
there is a combination with a coordinative clause combination, in italics:
While Ed was coming downstairs, Mary slipped out the front door, went
around the house, and came in the back door.
Although Ed never slept more than five minutes overtime, his father got
cross with him and made things generally unpleasant.
When you have a small baby in the house do you call it it or do you call it she
or he?
We can diagram the first example in the same way we did in Figure 3 to
bring out the organization of the clause combinations; see Figure 4.
Ed was coming
downstairs
Figure 4: Clause combining with clause combination

Note that the diagram in Figure 4 is quite similar to the one in Figure 3
above. In the earlier example, the temporal relation combines a clause with a
clause combination. (The difference is that that clause combination in the ear-
lier example is not a coordinative one, but is of the kind we are studying in
this paper.) In fact, you need a microscope to see what happened long ago
because it's very far away is quite parallel to the example we diagrammed
above in Figure 3. If we treated the reason clause as an embedded part of you
need a microscope, we would be unable to bring out the fact that its domain is
the whole clause combination. This kind of situation where a clause combines
with a clause combination rather than a single clause is quite common. In Sec-
tion 3, we will see that it reflects a very basic organizational property of dis-
course in general.
2.3.2 Combining by coordination
We have shown that embedding has to be distinguished from clause com-

bining and that our examples illustrate clause combining rather than embed-
ding. There is a potential problem with this interpretation, since clause com-
bining where two or more clauses combine without being constituent parts of
one another traditionally implies coordination or apposition: In many tradi-
tional grammars, we only have two categories at our disposal, embedding and
clause combining by coordination. (Sometimes apposition is recognized as dis-
tinct from coordination.) The are usually simply called subordination and
coordination.5 Given that only 'subordination' and 'coordination' have been
available as analytic tools to many grammarians, it is perhaps not surprising
that they have tried to force examples such as those given in Figure 1 into the
subordination model by calling them adverbials.
What we need at this point is a framework that is richer than the tradi-
tional one and allows us to interpret our examples as instances of clause com-
bining without having to treat them as coordination (or apposition). We will
follow Halliday and other systemic linguists in assuming that there are two
degrees of clause combining, parataxis (e.g. coordination) and hypotaxis.
2.4 Two degrees of clause combining: hypotaxis and coordination
Halliday and grammarians working within the systemic tradition distin-

guish for English between 'embedding', which includes essentially clauses
embedded as restrictive relative clauses and subject and object complements,6
and clause combining or clause complexing in Halliday's terms.
Halliday recognizes a number of different types of clause combining. In

terms of the degree of interdependence, he distinguishes between paratactic
combining, 'parataxis', and hypotactic combining, 'hypotaxis'. Parataxis
includes coordination, apposition, and quoting.7 Hypotaxis includes essen-
tially clause combining involving non-restrictive relative clauses, clauses of
reported speech, and the clause combinations we exemplified in Figure 1.
These are clause combinations where we do not gain any grammatical or dis-
course insight by interpreting one clause as a constituent part of another
clause. Although the clauses are interdependent and stand in a kind of head -
dependent relation to one another at some level, there is no sense in which
one is a part of the other. 8 In our discussion of discourse organization, we will
show that the head-dependent organization is a general characteristic of dis-
course and will suggest that the grammatical facts of hypotactic clause combin-
ing reflect this kind of discourse organization.
2.5 Type of interdependence in clause combining
Having characterized our examples in Figure 1 as hypotactic clause com-

binations, we can follow Halliday one step further in differentiating them
from other types of clause combination, in particular from other types of
hypotaxis. Our examples are enhancing hypotactic clause combinations: In
addition to the distinction between parataxis and hypotaxis in terms of degree
of interdependence between clauses being combined, Halliday distinguishes a
number of different types of interdependence:9 projection vs. expansion, and
within expansion elaboration vs. extension vs. enhancement. Elaborating
hypotaxis involves combinations with non-restrictive relative clauses and
extending hypotaxis involves clauses of e.g. replacement {instead of spending
a lot of money in a restaurant we'll get ourselves invited to dinner somewhere)
and addition {besides visiting our relatives we went down to Texas).
We will not discuss this range of types, since our focus is only on enhanc-
ing hypotaxis. But the point is that it is not enough only to look at the degree
of interdependence, which is what often happens in discussions of clause com-
bining.10 The discourse properties of enhancing hypotaxis are related to those
of other kinds of hypotaxis but also crucially different and we need to recog-
nize enhancing hypotaxis as a distinct category if we are to fully understand it
in discourse terms.
As we have said, we will concentrate on what Halliday has called enhanc-
ing hypotaxis. Enhancing hypotaxis refers to hypotactic clause combining
involving some kind of circumstantial relation like condition, reason, purpose
and others kinds of cause, time, space, manner, and means: One clause
enhances another clause circumstantially.
The notion of enhancing hypotaxis takes us back to the list of clause com-
binations we started with in Figure 1 to exemplify the phenomenon we want to
look at in this paper. When we introduced the examples, we said that we did
not want to name them. There is simply no satisfactory term for them in trad-
itional accounts. We can now use Halliday's name, enhancing hypotaxis, for
this kind of clause combining. This name is now meaningful in that it sepa-
rates the examples in Figure 1 from embedding and it indicates how they are
distinct from other types of clause combination, both from parataxis (e.g.
clause combining by coordination) and from other types of hypotaxis.
2.6 The function of clauses in combination vs. the class of a clause
Up to now, we have concentrated on the grammatical function a clause

may serve. We have met with before clauses in several examples. Repeating
two of them, the first in changed form, we have:
Before he left Krishnapur, the Collector took a strange decision,
the happy days before the Magistrate had been invited
In terms of grammatical function, the two before clauses are quite differ-
ent. The first is part of an enhancing hypotactic clause combination, whereas
the second is embedded in a noun phrase. And yet we refer to them both as
before clauses and this captures a number of similarities; e.g. they are both
introduced by the marker before; neither could be interrogative. We can cap-
ture these similarities by assigning the clauses to the same class of clause. For
example, we may choose to set up a class of before clauses, of temporal
clauses, of circumstantial clauses, etc.. If we do, this will in no way entail that
they all have the same grammatical function. We can set up classes of
clauses to make generalizations about markers like before, about mood alter-
natives, about transitivity patterns, about thematic possibilities, about word
order, and so on. Similarly, we can set up classes of noun phrases to make
generalization about case, person, determination possibilities (e.g. proper vs.
common), and so on. In neither case does a class necessarily entail a particular
grammatical function.
Our point, then, is that grammatical function and grammatical class have
to be kept distinct when we study clause combining just as they have to be in
other areas of grammar. To say this may seem to be to state the obvious, but
there are four reasons why it is important to draw attention to the distinction.
First, the terminology in this area is often unclear. For example, the term
subordinate clause is sometimes used to refer to a particular function a clause
may have, the clause is 'subordinated' in relation to another grammatical unit;
and sometimes to a particular class of clause.11 Similarly, the term adverbial
clause is sometimes used to denote a function a clause may have, and some-
times it is used to name a particular class of clause.12 We have chosen to avoid
both terms in our paper.
Second, the terminological unclarity may in fact reflect a mixture of class
and functional criteria in the treatment of clauses. Br0ndal pointed to this
problem in 1937 and his observation that the two viewpoints are often mixed
is still applicable.
Third, various criteria for recognizing a particular class of clause are often
used in arguments for a particular analysis of what grammatical function or
functions a clause serves. For example, observations about word order in e.g.
German, Dutch, or Swedish are sometimes used as evidence for a treatment
of the whole class of clauses with a particular kind of word order as embed-
ded. We are not claiming that there are no correlations between class and
function, but we are claiming that the function a clause has is not a necessary
consequence of a particular word order pattern. To take an example from
English: In clauses with a wh- item, the sequence is typically wh- item + the
rest of the clause. This is true of direct and indirect interrogative clauses as
well as of relative clause, i.e. of 'wh- item clauses', but we would obviously
not be tempted to say that relative clauses and interrogative clauses have the
same grammatical function. In the same way, it does not follow that all clauses
in German with the finite verb in final position must have the same kind of
grammatical function.13
Fourth, when we are interested in the use of clauses in discourse, as we
are here, grammatical function is a more sensitive tool in distinguishing
clauses than is grammatical class. Clauses of the same class but with different
functional potentials are sometimes counted in the same category in discourse
studies. Clauses functioning in grammatically different ways usually have very
different discourse properties, although they may belong to the same class.
2.7 Why not 'subordinate' or 'adverbial' clause?
By way of summary, we can now state why we chose not to refer to the
clause combinations in Figure 1 as involving 'subordinate' or 'adverbial'
clauses. The problem with both terms is that they have been used to refer both
to grammatical function and grammatical class, as we have just indicated
above. However, even if we make it clear that we intend grammatical function
rather than class, the terms are still misleading. If subordinate clause is taken
to mean a clause that functions as subordinated to another grammatical unit,
this fails to make the distinction between embedding and clause combining. If
adverbial clause is taken to mean a clause that functions as an adverbial, this
treats it as embedded within another clause rather than as an instance of
clause combining, and we have rejected the embedding interpretation for our
set of examples.
2.8 Towards a discourse characterization
The approach to the study of the grammar of clause combining we have

outlined sketchily in this section is highly differentiating; in this respect, we
follow e.g. Halliday and Longacre. Clause combining by enhancing hypotaxis
is just one of many functionally distinct types of clause combining. The high
degree of differentiation is grammatically motivated, as is the particular
interpretation of enhancing hypotaxis we have adopted. But we have in fact
described the grammatical reflection of the rhetorical organization of dis-
course. We now have a good starting point for investigating the discourse
function of clause combining, since the grammatical interpretation does not
lump together distinctions such as embedding vs. hypotaxis that are crucial to
an understanding of discourse function. What we want to do now is leave the
grammar of clause combining for a while to explore the rhetorical organiza-
tion of discourse (text). This means taking a step up one level in abstraction,
from grammar to the level of discourse. Having given an independent account
of the rhetorical organization of discourse, we will then return to the grammar
of clause combining to show how it reflects the higher-level organization of
discourse.
3 RELATIONS IN DISCOURSE STRUCTURE
Since our claim is that the grammar of clause combining reflects discourse
organization, we will now indicate what we think some of the important
aspects of this organization are.
3.1 Text relations: an overview
It is uncontroversial that discourse is coherent, that is, that parts of a dis-

course 'go together' to form a whole. As background to the discussion of the
discourse basis for hypotaxis, we will in this section consider one type of dis-
course, small written expository texts in English, and describe one factor
involved in the creating and interpreting of texts as coherent. This factor is the
existence of perceived organizational, or rhetorical, relations between parts of
the text. We will consider one type of discourse, small written expository texts
in English; this restriction is a useful limitation for illustrative purposes.
Our study of expository English has revealed that a number of relations
continually recur; in line with a number of other discourse linguists (see
below), we have identified a set of twenty or so such relations.
We will show that these relations, often not directly signalled, are essen-
tial to the functioning of the text as a means for a writer to accomplish certain
goals. These relations involve every non-embedded clause in the text and they
form a pattern of relations which connects all the clauses together.
Let's begin the discussion by considering first a short text as an illustra-
tion. The text has been broken down into components, which we will call
(rhetorical) 'units', roughly coded as what most grammarians would call
'clauses', except that clausal subjects and complement14 and restrictive rela-
tive clauses represent units that are part of their matrix unit rather than sepa-
rate units.15
(from Language Sciences, April, 1969)
1. Sanga-Saby-Kursgard, Sweden, will be the site of the 1969 International
Conference on Computational Linguistics, September 1-4.
2. It is expected that some 250 linguists will attend from Asia, West Europe,
East Europe including Russia, and the United States.
3. The conference will be concerned with the application of mathematical
and computer techniques to the study of natural languages, the development
of computer programs as tools for linguistic research, and the application of
linguistics to the development of man-machine communication systems.
This text can be seen as a general claim (Unit 1), followed by two pieces
of detail elaborating this claim (Units 2-3).
Accordingly, we can postulate Unit 1 as being the most central of the
three units to the writer's purpose in creating this text. Without Unit 1, Units
2 and 3 are senseless, whereas without Units 2 and 3, Unit 1 would still convey
a message, albeit a difficult one to respond to. We thus postulate a text rela-
tion of Elaboration between Unit 1, on the one hand, and Units 2 and 3, on
the other, with Unit 1 as the nucleus and Units 2-3 as the satellite.
And we can postulate a List representation for the fact that Units 2 and 3
function together as co-equal realizations of the Elaboration satellite.
Schematically, we can represent this relation of Elaboration as shown in
Figure 5.
Figure 5: Relational structure of the 'Computational Linguistics' text

Before discussing the implications of these claims, let's consider one
more text:
(from UCLA Senate Report, June, 1984)
1. The Senate Office will be closed from July 30 through August 10 during
the Olympic Games.
2. The Executive Board decided to close the office during this period in
response to the fact that so many departments on campus will also be curtail-
ing their operations during the Olympics.
In this text, Unit 2 can be seen as providing a reason for the action
described in Unit 1.
Once again, Unit 1 represents the writer's main goal in creating this text.
We can therefore propose that a text relation of Reason exists between
these two parts of this text. Schematically, we can diagram the relational
structure of this text, then, as follows:
Figure 6: Relational structure of the 'Closed for Olympics' text

Now let's see what apparatus we have constructed so far. We have seen that a
description of the "rhetorical (relational) structure" for a text can be given in
terms of the rhetorical relations between its component parts. This relational
structure shows the rhetorical relations between parts of the text, and it shows
how each component unit is linked to the rest of the text by a network of such
relations. The perception of texts in terms of hierarchically organized groups
of units is a linguistic reflex of a general cognitive tendency; Lerdahl and Jac-
kendoff's description of this process for the interpretation of tonal music
(1983:13), could equally well have been written to describe the interpretation
of texts:
"the process of grouping is common to many areas of human cognition. If
confronted with a series of elements or a sequence of events, a person spon-
taneously segments or 'chunks' the elements or events into groups of some
kind. The ease or difficulty with which he performs this operation depends
on how well the intrinsic organization of the input matches his internal,
unconscious principles for constructing groupings."
Their comment that 'grouping can be viewed as the most basic compo-
nent of musical understanding' (p. 13) holds equally well for text understand-
ing.
We have also seen that two types of relations can be distinguished: those
in which one member of the related pair is ancillary to the other (diagrammed
with an arc from the ancillary portion to the central portion), and one in which
neither member of the pair is ancillary to the other (diagrammed as descen-
dents from a List node). This distinction is crucial. The first type we might call
a 'Nucleus-Satellite' relation, the second a 'List' relation.16
We are suggesting, then, that all text can be described in terms of such
hierarchical relations among its various parts.17 It is important to note that
these relations are defined in terms of the functions of segments of text, that
is, in terms of the work they do in enabling the writer to achieve the goals for
which the text was written. The rhetorical structure of texts, then, is claimed
to be composed of function-specific elements.
The rhetorical structure of a text, then, can be composed of both 'Nu-
cleus-Satellite' relations and 'List' relations. Of the two types of relations, we
will focus on the Nucleus-Satellite relation.
The Nucleus-Satellite distinction is one which characterizes the organiza-
tion of all of the texts we have analyzed, and which furthermore seems to be
pervasive as a text-organizing device.18 We take it to reflect the fact that in
any multi-unit text, certain portions realize the central goals of the writer,
while others realize goals which are supplementary or ancillary to the central
goals.19 That is, the nuclear part is the one whose function most nearly repre-
sents the function of the text span 'covered' by that relation. Thus, for exam-
ple, in the 'Computational Linguistics' text considered above and diagrammed
in Figure 5, the analysis claims that Unit 1 is the nucleus of this text, with
Units 2 and 3 providing supplementary, elaborating, material. This nuclear-
satellite distinction reflects the fact that the central goal for the writer of the
text, as perceived by readers, is to convey the information that a particular
computational linguistics conference will be held.
Judgments about what is nuclear and what is supplementary, then, are
made by readers as part of the general cognitive tendency, referred to' above,
which has been investigated in depth by Gestalt psychologists (see, e.g., Ellis,
1938, Koffka 1935, and Kohler 1929), to impose structure reflecting 'central'
and 'less central' on certain types of perceptual input. For texts, these
judgments are based on our perceptions, as ordinary readers, of what the text
is designed to accomplish. Such judgments turn out, in general, to be easy to
make, though there may be problematic cases; the analysis of texts into
hierarchically organized nuclear and satellite parts reflects the fact that read-
ers consistently make such judgments as part of their comprehension of texts,
and writers construct texts expecting them to be able to do so.
It should be clear that nuclearity and hypotaxis are quite distinct from
each other: there are many Nucleus-Satellite relations which do not involve
hypotaxis, such as that illustrated in the "Computational Linguistics" text just
above: Units 2 and 3 are joint satellites to the nuclear Unit 1, but each of
these units is a sentence; there is no hypotaxis anywhere in this text. We wish
to emphasize that Nucleus-Satellite relations are pervasive in texts indepen-
dently of the grammar of clause combining.
Now, if the number of relations one needs to posit to describe the rela-
tional structure of any coherent text turns out to be relatively small, and if
precise definitions of these relations can be given, then we have the founda-
tions for a theory of the organizational structure of texts. In fact, this number
does seem to be small (about 20), definitions can be given, and such a theory
has actually been proposed. The description of text structure that we are
offering here is an adaptation of this theory, modified in the interests of acces-
sibility.
We will not attempt to present or justify this theory here, since a full
description of the theory and careful definitions of each of the relations would
distract us from the goal of relating issues of hypotaxis to discourse structure
(for more rigorous, though brief, treatments, see Mann 1984, and Mann and
Thompson 1985, 1986, and to appear). Instead, since our point can be
made independently of the precise definitions one gives for each relation, we
will proceed to simply identify and exemplify some of the other relations
which recur in the analysis of texts, and discuss the question of hypotaxis in
terms of these relations.20
The list of twenty or so relations which we have found to be useful were
subjectively arrived at through analysis of more than one hundred texts.21
While our list may not precisely match those of other researchers, our point
here is simply that some such list of relations, as has been recognized by such
researchers as those mentioned in note 20, is central to an understanding of
the organizational structure of texts.22 We will discuss only the most fre-
quently recurring of these relations in this paper.
3.2 Individual relations
3.2.1 Enablement
In texts which issue directives or make offers, it is common to find an

Enablement relation. Here is an example:
(from The Linguistic Reporter, Fall, 1971)
1. The University Press of Kentucky has announced the establishment of the
Kentucky Foreign Language Conference Award to be given annually for the
best manuscript dealing with some aspect of foreign language and/or litera-
tures.
2. The Award, $500 and acceptance of the manuscript for publication, is
offered in conjunction with the Kentucky Foreign Language Conference.
3. The deadline for submission of manuscripts for the 1972 Award is
December 1, 1971.
4. For further information write: KFLCA, The University Press of Ken-
tucky, 104 Lafferty Hall, Lexington, Kentucky 40506.
This offering text can be diagrammed as shown in Figure 7.
Figure 7: Relational structure of the 'Kentucky Award' text

An Enablement relation holds, then, between a nuclear portion of text

issuing the directive or offer and a portion providing the means which enable
the reader to comply with the directive or take advantage of the offer.
Figure 7 claims that the 'Kentucky Award' text is structured in the follow-
ing way: Unit 1, the announcement of the award itself, is nuclear to the Elab-
oration satellite, Unit 2. Then both Units 1-2 are nuclear to the Enablement
satellite, jointly realized by Units 3-4. Here the Enablement portion of the
text tells readers how to avail themselves of this offer by specifying the dead-
line and where to find further information.
This multi-unit text provides us with an opportunity to point out two
further features of our analysis. First, note that Figure 7 appropriately reflects
the scope relations which readers assign to the 'Kentucky Award' text. That is,
as mentioned just above, Units 3 and 4 are understood as being in an Enable-
ment relation with the combination of Units 1 and 2. The scope relations are
reflected in the levels in the diagram, i.e. in the hierarchic organization: The
Enablement relations and the Elaboration relation are not on the same level
(they are not 'sister' satellites). Rather, there is one satellite per level. Sec-
ond, note that Figure 7 reveals Unit 1 to be the nucleus, not only for its own
satellite, Unit 2, but also for the text as a whole. In fact, the analysis allows us
to claim that the unit which is found by tracing only vertical (nuclear) lines
from the top of the relational structure diagram will always be the nucleus of
the text as a whole, that unit which most succinctly represents the goal for
which the text was created.23
3.2.2 Motivation
Another relation commonly found in texts which issue directives or make

offers is the Motivation relation. Here is an example:
(from the ISI librarian, a message appearing on the electronic ISI bulletin
board, May 14, 1984)
1. Some extra copies of the Spring 1984 issue of AI Magazine are available
in the library.
2. This issue includes a 'Research in Progress' report on AI research at ISI.
By convention, we restrict the Motivation relation to instances of one part

of the text providing motivation to the reader, rather than to the subject of the
clause realizing the nucleus. Thus a Motivation relation is invoked only for
texts which convey offers, requests, or directives.
Figure 8 shows the relational structure of this text:
Figure 8: Relational structure of the 'Al Magazine' text

This example illustrates a Motivation relation between the nuclear offer
and the satellite providing motivation for the reader to take advantage of the
offer. In this case, the writer of the text intends to motivate ISI researchers to
consult the latest issue of AI Magazine by pointing out that they may find
themselves mentioned in it.
3.2.3 Background
One portion of a text can provide the background for another portion, as
in this text:
(from an ISI researcher, message appearing on the ISI electronic bulletin
board)
1. Someone left a coffee cup in my office over the weekend.
2. Would the owner please come and get it
3. as I think things are starting to grow?
In this text, the writer of the message is implying that the coffee cup is not
welcome in the writer's office because mold has started to grow in it. Accord-
ing to our definitions, the Background relation holds for a text span which
provides for the comprehensibility of an item mentioned in another text span.
We can thus analyze this small text as follows: Unit 1 provides background
information for the request expressed in Unit 2, for which Unit 3 provides
motivation. This grouping can be represented as in Figure 9.
3.2.4 Concession
A Concession relation holds when one portion of a text concedes a point

potentially damaging to the argument the writer wishes to make. Here is an
example taken from the beginning of a personal letter:
1. Your kind invitation to come and enjoy cooler climes is so tempting
2. but I have been waiting to learn the outcome of medical diagnosis
3. and the next 3 months will be spent having the main thumb joints replaced
with plastic ones.
Here, with respect to a previous invitation to come and visit, the letter-
writer concedes in the satellite Unit 1 that the idea is a tempting one. But her
nuclear argument (in Unit 3) is that thumb surgery will preclude her accepting
the invitation. With Unit 2 as background to Unit 3, here, then, is the analysis
of this text:
Figure 10: Relational structure of the 'Cooler Climes' text
3.2.5 Circumstance, Elaboration
Here is an excerpt from the description of a national public radio

announcer:
1. Peter Moskowitz has been with KUSC longer than any other staff
member.
2. While attending Occidental College,
3. where he majored in philosophy,
4. he volunteered to work at the station as a classical music announcer.
5. That was in 1970.
The Circumstance relation is exemplified twice in this passage: not only
do Units 2-3 together form a Circumstance satellite for Units 4-5, but Unit 5
forms a Circumstance satellite for Unit 4, as shown in Figure 11: (Note again
the scoping with one satellite at each level in the hierarchy.)
Figure 11: Relational structure for the 'Peter in 1970' text

Figure 11 shows that this excerpt also contains two instances of the Elab-
oration relation: Units 2-5 form an Elaboration satellite for Unit 1, and Unit
3 forms an Elaboration satellite for Unit 2.
3.2.6 Solutionhood, antithesis, purpose, condition
In this excerpt from a few lines later in the same personal letter consid-
ered in connection with Figure 10, we can see four new relations. Having
announced that she won't be able to come and visit because thumb surgery is
going to be necessary, the writer is giving the background story, which
involves hereditary arthritis:
1. Thumbs began to be troublesome about 4 months ago
2. and I made an appointment with the best hand surgeon in the Valley
3. to see if my working activities were the problem.
4. Using thumbs is not the problem
5. but heredity is
6. and the end result is no use of thumbs
7. if I don't do something now.
The Antithesis relation is one of those which is more involved with textual
concerns of presentation rather than with spatial or temporal relations
between events. This relation can be characterized as follows: the satellite
'thesis' expresses a proposition which the writer refuses to identify with, while
the nuclear 'antithesis' expresses a contrasting proposition which the writer
does indentify with. In Unit 4 of the 'Thumbs' text, the writer offers the thesis
that the use of thumbs at work might be the problem, and she signals her
refusal to identify with this proposition by the use of the negative. In Unit 5,
she offers the antithesis, which she does identify with, that heredity is causing
the problem.
This extract from this letter can be rhetorically represented as shown in
Figure 12.
Figure 12: Relational structure of the 'Thumb Heredity' text

As Figure 12 shows, this extract also illustrates three other relations.
First, Unit 3 can be seen to relate to Unit 2 in a Purpose relation, where the
satellite, Unit 3, provides a purpose for which the action in the nuclear Unit 2
is undertaken. Second, Units 6 and 7 are in a Condition relation: the satellite
provides the conditions under which the situation expressed in Unit 6 holds.
Note that both Purpose and Condition are examples of relations which involve
interaction among reported events more than presentational aspects of text
organization, which characterized the Antithesis relation.
The third new relation illustrated by this example is the Solutionhood
relation holding between Units 1 and Units 2-3. Here Unit 1 states a problem,
the troublesome thumbs; the solution, making an appointment with the best
specialist available, is expressed in Units 2-3.
Note that Figure 12 provides a particularly clear illustration of the way
groupings are hierarchically organized, as mentioned above: for instance, each
of the two occurrences of the Background relation in the relational structure
for this text is at a different 'level' of structure: Units 1-3 are background for
Units 4-5, and Units 1-5 are background for Units 6-7. Note that there is one
satellite at each level in the hierarchy.
3.2.7 Typology of rhetorical relations
Having illustrated a number of the relations which we have found to

characterize coherent text, we can now present a partial typology of rhetorical
relations.24
Recall that we have distinguished between 'List' relations and 'Nucleus-
Satellite' relations, as shown in Figure 13:
List Nucleus-Satellite
Figure 13: The distinction between 'List' and 'Nucleus-Satellite' relations

Within the category of Nucleus-Satellite relations, we can distinguish (i)
relations having to do with the success or felicity of a rhetorical act such as
requesting, and (ii) relations pertaining to the subject matter of the text.25
Within (ii), we also distinguish two subtypes. They are tabulated in Figure 14.
Nucleus-Satellite
[i] Rhetorical Act Motivation

Background
Solutionhood
Antithesis
[ii] Subject Matter
Elaborating Elaboration
Enhancing Purpose
Condition
Circumstance
Concession
Figure 14: Typology of rhetorical relations
All of the relations in [i] are used to ensure the success or felicity of a
rhetorical act. Thus, for example, Motivation is used to ensure the success of
a request or an offer by specifying information intended to make the reader/
listener feel inclined to comply or accept. Similarly, Background is used to
provide the reader/listener with information that will enable him/her to com-
prehend an item mentioned in the nucleus. The relations in [i] tend to be scale
insensitive, i.e. they may occur at any level in a text where there is a request,
offer, or claim, but they often scope over a whole text.
The subtypes of [ii] are taken from Halliday (1985a); they were mentioned
in Section 2.5 above. An elaborating relation is used when there is a relation
of 'being' between two or more units; this is the very general relation that
obtains between an attribute and a value, between a set and its members, or
between a generalization and its specific instances. Elaborating relations are
distinct from other rhetorical relations in that they do not necessarily hold
between propositions per se, but may relate terms in the propositions, e.g.
one term in the proposition may be related to another as type to subtype. Like
the 'rhetorical act' relations in type [i], elaborating relations also tend to be
scale-insensitive; they may occur at any level in a rhetorical structure.
Enhancing relations are used when units are circumstantially related to
one another. 'Circumstantial' has to be understood in a very general sense
here. In principle, the relation corresponds to a parameter of the physical
world: temporal, spatial, causal, instrumental, etc.. Events or situations are
related temporally, spatially, and so on.26
Unlike the other two types of relations in the typology, there is a ten-
dency for enhancing Nucleus-Satellite relations not be used at the very top of
a rhetorical structure, i.e., not to be scale-insensitive; in fact, they are typi-
cally found towards the bottom, a fact which is related to their special role in
clause combining.
3.3 Summary
We have been assuming that part of people's creating and interpreting of

texts as coherent is the imposition of organizational, or rhetorical, relations
between parts of the text. We have described and exemplified some of the
relations in terms of which the organizational structure of texts can be made
explicit, and we have shown how these relations are essential to the function-
ing of the text as a means for a writer to accomplish certain goals, and how
they form a pattern of relations which connects all the clauses together.
We have also emphasized that the Nucleus-Satellite distinction is one

which characterizes the organization of all of the texts we have analyzed. It
reflects the fact that in any multi-unit text, certain portions realize the central
goals of the writer, while others realize goals which are supplementary or
ancillary to the central goals. Among the properties of rhetorical units defined
by the Nucleus-Satellite relation, the following are especially important for
present purposes:
1. Use and markedness: whether or not a satellite is used in a text
depends on the goals of the speaker or writer; if it serves a purpose it
is used, otherwise it isn't. The nucleus derives its justification and
function directly from its place in the rhetorical structure, but the
satellite is related to the rest only via its nucleus. The satellite is thus
the marked member of the Nucleus-Satellite pair.
2. Distinctness: The nucleus and the satellite in a Nucleus-Satellite rela-
tion are rhetorically distinct.
3. Scope: Rhetorical units defined by enhancing rhetorical relations
always consist of a nucleus and one satellite. That is, a text is always
scoped in such a way that there is one satellite per nucleus.
4. Sequence: For all Nucleus-Satellite relations we find instantiations of
both sequences of nucleus-plus-satellite and satellite-plus-nucleus.
Finally, we have offered a typology of the relations we have discussed; we
expect this to be useful in the rest of this paper.
In the next section, we will relate these observations to the nature of
enhancing hypotaxis.
4 TEXT RELATIONS AND ENHANCING HYPOTAXIS
After presenting the text relations in the previous section, we are now in
a position to examine how their role in the organization of text is related to
enhancing hypotaxis, as we described it in Section 2. First we will draw an
analogy between the organization of clause combinations and of discourse
(text) in general. Then we will hypothesize that clause combining is a gram-
maticization of the rhetorical organization of discourse. The rest of the section
shows how the hypothesis is borne out, argues that some properties ascribed
to enhancing hypotaxis are not necessary properties that they have, and also
shows why it is important to distinguish embedding from hypotaxis when dis-

course function is studied.
4.1 Analogy between clause combining and text relations
There are several implications of the preceding discussion of rhetorical

relations for the issue of clause combining. As a first step in our exploration of
the relationship between rhetorical text relations and clause combining, we
can ask how clause combining is to be understood in the light of rhetorical
organization. We can ask the question in the following way: How is a clause
combination like a text?
The question is meaningful, because we made a distinction between the
rhetorical organization of discourse (text) and the grammatical organization
of clauses into clause combinations. In other words, having made the distinc-
tion, we can now ask whether there is an analogy between clause combining
and the rhetorical organization of discourse. We think there is a fundamental
analogy between the two.
The first point to notice is that the relationships among the units coded by
clauses in clause combinations in our texts are of exactly the same type as
those among the higher-level rhetorically defined text spans. This suggests
that the principles of clause combining should not be thought of as different
from those governing the way texts in general are organized.
The second point is that clauses also combine in the same two ways as the
rhetorical units of a text. Both display a kind of structure where the elements
are interdependent either as the members of a list or as a nucleus to a satellite.
In rhetorical organization, we recognized a distinction between Listing and
Nucleus-Satellite relations, which is exactly paralleled by the grammatical dis-
tinction between parataxis (e.g. coordination and apposition) and hypotaxis.
As a consequence of the structural similarity, both clause combining and
rhetorical organization show the same kind of scoping characteristics. For
example, a clause or rhetorical unit may combine with a combination of
clauses or a combination of rhetorical units. We diagrammed the scoping for
clause combinations in Figures 3 and 4 above, which reflect the same kind of
scoping characteristics as the rhetorical structures in Figures 7 and 9-12 above.
There is an implicit contrast in the analogy between the structuring of a
clause combination and the rhetorical organization of a text: A clause combi-
nation is organized like the rhetorical structure of a text, but it is not
organized like a clause. We can clarify the contrast by referring to Bloom-
field's (1933) distinction between exocentric constructions and endocentric

constructions. He pointed out that the actor-action relation is exocentric; the
generalization is that a clause is exocentric in its case-frame-like organization.
In contrast, we have shown that a clause combination is structured endocentri-
cally, and so is a text. In fact, Bloomfield recognized two types of endocentric
structuring, subordinative and coordinative. We can now reinterpret these in
the context of clause combining as hypotactic clause combining and paratactic
clause combining.27
4.2 Hypothesis
Two sum up our two observations about the analogy between clause com-
bining and discourse organization:
1. The same relations define clause combinations as texts in general.
2. Clause combinations and texts are structured and scoped in the same
way; we find both Listing and Nucleus-Satellite organization.
We suggest, then, that there is a fundamental analogy between a clause
combination and the rhetorical organization of a text. But we can take one
step further in order to explain why clause combining is organized the way it
is. We suggest that clause combinations represent (code, symbolize) units of
text rhetorically combined. Let us make this explicit by stating it as a
hypothesis:
Hypothesis: Clause combining in grammar has evolved as a grammaticaliza-
tion of the rhetorical units in discourse defined by rhetorical relations.
Our hypothesis predicts a number of characteristics of clause combining
based on what we observed about rhetorical organization in Section 4. We will
confine ourselves to Nucleus-Satellite relations of the enhancing type. The
more specific version of our hypothesis we will look at is:
Narrower hypothesis: Enhancing hypotactic clause combining has evolved as
a grammaticalization of rhetorical relations in text of the enhancing Nucleus-
Satellite kind.
4.3 Predictions about the grammar of clause combining based on rhetorical

relations
If we hypothesize that clause combining of the enhancing hypotactic kind

is a grammaticalization of enhancing rhetorical Nucleus-Satellite relations, we
can make the following predictions about the characteristics of enhancing

hypotaxis based on the characteristics of rhetorical relations listed in section
3.3 above. We will elaborate on each below.
1. Use of hypotactically enhancing clause: Whether or not a satellite is
used in a text depends on the goals of the speaker or writer; if it serves
a purpose it is used, otherwise it isn't. Consequently, we should
expect hypotactic enhancement to be used under the same conditions.
2. Scope: Rhetorical units defined by enhancing rhetorical relations
always consist of a nucleus and one satellite. Consequently, we should
expect enhancing hypotactic clause combinations to have the same
scoping property. (This can be contrasted with multi-nuclear units
which may continue indefinitely without any change in scoping.)
3. Grammatical marking: The nucleus and the satellite in a Nucleus-
Satellite relation are rhetorically distinct. Consequently, if there is
grammatical marking at the level of clause combining, we would
expect that the nucleus and the satellite should be grammatically dis-
tinct, that is, there should be marking such that one clause is identifi-
able as coding the nucleus and the other clause is identifiable as cod-
ing the satellite.
4. Sequence: For all Nucleus-Satellite relations we find instantiations of
both sequences of nucleus-plus-satellite and of satellite-plus-nucleus.
Consequently, we should find both orderings in enhancing hypotactic
clause combinations. Further, the two different sequences are typi-
cally systematically different. We should expect to find the same sys-
tematic differences in the alternative sequence of clauses.
5. Typical coding: If enhancing hypotactic clause combinations are
grammaticalizations of enhancing Nucleus-Satellite rhetorical units,
the prediction is, of course, that rhetorical units of this kind are
indeed coded as enhancing hypotactic clause combinations in text, if
they are coded as grammatically related clauses. In other words, we
predict that they are predominantly hypotactic rather than paratactic
or coordinative clause combinations.
4.3.1 Use of a hypotactically enhancing clause
Whether a clause is enhanced through hypotaxis or not depends entirely

on whether it is the nucleus in a Nucleus-Satellite relation or not. If there is a
satellite to express, the clause may be hypotactically enhanced. There is

nothing in the grammar of English that demands hypotactic enhancement. In
this respect, hypotaxis is just like coordination (or parataxis in general). How-
ever, it is quite different from embedding, where the use of the embedded
clause depends on what it is embedded as, for example whether it is embed-
ded as subject (in which case it is affected by the same constraints as subjects
in general) or as post-modifier in a noun phrase (in which case it is grammati-
cally optional, as post-modifiers generally are).
4.3.2 Scoping and number of clauses
We have already pointed out that clause combinations are structured and
scoped in the same kind of way as texts are. But since we have now also
claimed that a clause combination represents rhetorical organization, it fol-
lows that a major discourse function of a clause combination is to reflect the
scoping and structuring of a rhetorical unit in a text. Halliday (1985a: 201)
points to this kind of motivation for using a clause combination (complex):
"The clause complex is of particular interest in spoken language, because it
represents the dynamic potential of the system — the ability to 'choreograph'
very long and intricate patterns of semantic movement while maintaining a
continuous flow of discourse that is coherent without being constructional."
Rhetorical units defined by an enhancing Nucleus-Satellite relation have
only one satellite. This satellite may be realized by a list of rhetorical units,
but it is still a single satellite. Consequently, we predict that the same charac-
teristics will hold for enhancing hypotactic clause combinations. Consider the
following clause combination 'spoken by a girl aged nine' (Halliday, 1985b):
Our teacher says that
1. if your neighbour has a new baby and
2. you don't know whether it's a he or a she,
3. if you call it 'it'
4. well then the neighbour will be very offended.
Concentrating on the conditions, we can identify the following condi-
tional parts: if your neighbour has a new baby, (if) you don't know whether if's
a he or a she, and if you call it 'if. Rhetorically, these are not three sister satel-
lites in a flat structure organized around the same nucleus. Ultimately, they
are all related to the neighbour will be offended, but there is nesting (layer-
ing). Rhetorically, as shown in Figure 15, we analyze it as one conjoined con-
dition, Units 1 and 2, if your neighbour has a new baby and you don't know
whether if's a he or a she, which scopes over the combination of Units 3 and 4,
where Unit 3 is a condition on Unit 4, if you call the baby it 'if well then the
neighbour will be very offended. There are, then, two rhetorical units of con-
dition. The rhetorical nesting is reflected in the grammatical analysis given to
the example; see Halliday (1985a).
Figure 15: Relational structure of the 'Neighbour' text
4.3.3 Grammatical marking
As our prediction suggests, the grammatical marking of the hypotactic

clause combination is such that nucleus and satellite are differentiated. In fact,
there are essentially two types of marking, both on the satellite.
1. Connective'. The clause coding the satellite is typically introduced by a
connective; see the list in Figure 2 above. The connective is either a
conjunction or a preposition.
2. Finiteness: The clause coding the satellite may be marked with respect
to finiteness by being non-finite (infinitival or participial).
These two types of marking are inter-related in an interesting way. Finite
clauses that are hypotactically related are always marked by a connective,
whereas non-finite ones often are not. For example, purpose satellites are
often coded as infinitival clauses without a special connective like in order or
so as to.
As Halliday notes (1985a: 217-218), there is a reason for this pattern of
marking:
A finite clause is in principle independent; it becomes dependent only if
introduced by a binding (hypotactic) conjunction. If it is joined in a clause
complex, its natural status is paratactic. ... A non-finite clause, on the other
hand, is by its nature dependent, simply by virtue of being non-finite. It typ-
ically occurs, therefore, without any other explicit marker of its dependent
status.
In other words, unless a finite clause is introduced by a connective of the

kind listed above in Figure 2, it does not code the satellite-status explicitly. It
may still code a satellite, of course, but the grammar does not mark it as a
satellite.
4.3.4 Sequence of clauses
We have noted that rhetorical units defined by the Nucleus-Satellite rela-

tion occur in texts both in the sequence nucleus-plus-satellite and in the
sequence satellite-plus-nucleus. (The relative frequency of each sequence is
highly dependent on the kind of rhetorical relation that obtains.) Our predic-
tion is that the same ordering characteristics hold for rhetorical units coded as
enhancing hypotactic clause combinations.
Indeed we do find both Nucleus-Satellite sequences and satelllite-nucleus
sequences of the same enhancing type coded as clause combinations. In fact,
the property of being positionally variable has sometimes been taken to be
criterial for a definition of 'subordination'. However, a consideration of dis-
course structure reveals that it is by no means the case that such clause types
are simply 'preposable' with respect to their 'main clauses'; that is, there is no
free variation between initial and final position. Rather, as might be expected,
the sequence in which nucleus and satellite occur is highly discourse-deter-
mined.
Differences in sequence are manifestations of differences in thematic
status. In particular, when the clause coding the satellite precedes the clause
to which it is related, it is thematic (in the sense of Halliday 1967, and 1985a).
Fries (1983) shows how thematic status is manipulated in discourse to reflect
the method of development of a text. There are various consequences of his
findings.
For example, certain relations are often found at major breaks in the
relational structure of a text, reflecting the fact that the satellites in these rela-
tions, because of their semantic ability to serve as 'guideposts' (to use Chafe's
1984 term), are typically involved in transitions from one major portion of
text structure to another: see Longacre, 1970; Longacre and Thompson,
1985; and Winter, 1982. The grammatical consequence of a hypotactic
clause serving this text function is that such a hypotactic clause would appear
before its nucleus. One such relation is that which we have called Cir
cumstance, as exemplified by the italicized Unit 3 in this extract from the
beginning of an advertisement for Syncom diskettes for cleaning floppy drive
heads:
1. What if you're having to clean floppy drive heads too often?

2. Ask for Syncom diskettes, with burnished Ectype coating and dust-
absorbing jacket liners.
3. As your floppy drive writes or reads,
4. a Syncom diskette is working four ways
5. to keep loose particles and dust from causing soft errors, dropouts.
The relational analysis for this extract is shown in Figure 16:
Figure 17 shows the relational analysis of this extract:
Figure 17: Relational analysis of the 'Gardening Revolution text

Figure 17 shows that the purpose clause given as Unit 4 occurs at a major
structural break in the text. To orient the reader to the new structural ele-
ment, which contains the 'solution' to the problem posed by the purpose
clause, namely how to plant your own flower basket, the problem is articu-
lated in the form of a clause expressing purpose.
In these terms, then, the issue is not so much one of hypotactic 'subordi-
nate' clause types being 'preposable', but of certain relations, such as Cir
cumstance and Purpose, being particularly appropriate for orienting the
reader to a new text span.
Thus, the apparent variation in the possible positions for circumstantial
and purpose clauses, and by extension other clauses with this apparent 'pre-
posability' property, is not simply a grammatical option in terms of which to
attempt to categorize clause types. Rather, this variation, we suggest, finds its
explanation in the radically different discourse functions which certain rela-
tions can serve. (See Chafe, 1984; Ford and Thompson, 1986; Longacre and
Thompson, 1985; and Thompson, 1985b, 1987 for further discussion of this
point, and Schiffrin, 1985b for a discussion of factors involved in the order of
clauses in causal sequences in conversational English.)
4.3.5 Predominant coding type in text
We have suggested that hypotactic enhancing clause combinations are a

grammaticalization of enhancing Nucleus-Satellite rhetorical units. From this
claim it follows that the characteristic coding in text of rhetorical units of this
kind should reflect the grammaticalization.
To a satisfying extent, this prediction is borne out. Counts made of 18

short texts in our data base revealed the following correlations for Nucleus-
Satellite units and List units in general; see Figure 18:
hypotactic 'co-ordinate'
Nucleus-Satellite 45(92%) 4 (8%)
Nucleus-nucleus (List) 3 (11%) 24(89%)
Figure 18: Coding correlations for expanding relations
Figure 18 shows that the ratios for the mappings are roughly 9:1; thus the
hypothesis accounts for about 90% of the data. Now consider a comparable
count for enhancing relations only; see Figure 19. Here hypotaxis seems to be
even more highly favoured, since the correlations are even stronger.
Enhancing: hypotactic 'co-ordinate'

Nucleus-Satellite 36(95%) 2 (5%)
Figure 19: Coding correlations for enhancing relations
Note that the numbers in Figures 18 and 19 show clearly that there is no
circularity in our argument here, since there are some instances of hypotaxis
which do not reflect Nucleus-Satellite relations and vice-versa. As an example
from a narrative, we offer the following excerpt, in which the hypotactic
when-clause would be analyzed as the nucleus of the Circumstance relation in
which it occurs with the clause beginning with my temper...)
Towards the finish, however, we must have held rather too
independent a line, for we lost the hounds, and found
ourselves plodding aimlessly along miles away from anywhere.
It was fairly exasperating, and my temper was beginning
to let itself go by inches, when on pushing our way
through an accommodating hedge we were gladdened by the
sight of hounds in full cry in a hollow just beneath us.
(H.H. Munro)
Our point is that we have an argument in favor of our claim that
hypotaxis is revealingly viewed as a grammaticization of Nucleus-Satellite
relations in the fact that when such relations are grammatically coded, they
are often, but not always, coded as hypotaxis.
As an example of a Nucleus-Satellite relation that is coded as one clause
and not as a hypotactic clause combination, consider the sentences of the fol-
lowing excerpt with italicized phrases:
(From a capsule history of the Aztec calendar.)

Soon after the Conquest, the stone was buried on the main plaza area
together with other religious stonework and idols. When accidentally disco-
vered in 1560, it was reburied for fear that it might accidentally create unrest
among the Indians. Uncovered again in 1790, this time it was affixed to one
of the walls of the Cathedral, where it remained for almost a century before
being moved to the former downtown museum.
(L. Carlson)
The text is organized around time and the time indications are thematic
in initial position (cf. Fries, 1983). The temporal circumstance satellites are
coded as 'adverbial clauses', except for the first, which coded as a preposi-
tional phrase with a nominalized complement and is part of a clause. Further
down, there is a reason satellite, giving the reason for why the stone was
reburied: 'they feared that it might create unrest among the Indians.' It is
coded not as a clause but again as a nominalization and the prepositional
phrase for fear that ... is part of the clause that codes the nucleus 'it was
reburied'. Thus, in these two examples, the Nucleus-Satellite relation is coded
grammatically but not as a hypotactic clause combination.
4.3.6 Summary
So far, then, we have suggested that if our hypothesis that hypotactic

clause combining is a grammaticalization of the rhetorical structure of a text is
correct, then certain predictions can be made about characteristics of clause
combining. We have seen that these predictions are borne out.
4.4 The grammarian's intuition about hypotaxis
One important implication of our interpretation of hypotaxis is that it is

only in some such discourse-oriented terms that we can begin to make sense of
the characterizations of a hypotactic clause as 'subordinate' to, 'less impor-
tant' than, 'dependent' on, or 'not on a par' with its main clause. That is, only
when we see hypotaxis as a special case (namely the single-clause case) of the
satellite relation can we specify what it is that makes it seem 'subordinate': it
is 'subordinate' because of its subordinate role with respect to the nucleus, as
these function in the organization of the text to serve the writer's goals.
For example, Quirk et al. (1985: 918) try to characterize the difference
between 'subordination' and 'co-ordination' this way:
Both co-ordination and subordination involve the linking of units of the

same rank; but in co-ordination the units are constituents at the same level of
constituent structure, whereas in subordination they form a hierarchy, the
subordinate unit being a constituent of the superordinate unit.
Since the only examples they give are of phrases, it is not clear just how
one would determine whether, for example, an although or a since clause is
'on the same level' as the clause with which it is associated. The intuition that
is being expressed here is correct, that such clause types as those in Figure 2
are 'on a different level' from the unit to which they are 'linked'. But it is dif-
ficult, if not impossible, to make this intuition explicit as long as one tries to
do it in terms of isolated sentences. Only when one considers the clause types
in the discourse context where they actually occur can one identify text goals
and begin to associate clause types with central and supporting goals.
Similarly, Lyons (1968:178) suggests that:
"Complex sentences are divided into: (a) those in which the constituent
clauses are grammatically co-ordinate, no one being dependent on the others
... ; and (b) those in which one of the clauses is 'modified' by one or more
subordinate clauses grammatically dependent on it..."
Our interpretation of the intuition expressed in this quote is, once again,
that Lyons is attempting to state in grammatical terms the rhetorical fact that
there is a set of clause types in English which typically function as satellites in
discourse.
One aspect, then, of the linguist's intuition that hypotactic clauses are
'subordinate' or somehow 'not on a par with' their main clauses is that
hypotactic clauses function rhetorically in discourse in satellite roles.
One clear indication of the source of the grammarian's intuition comes
from situations of the kind illustrated by the Munro text above in Section
4.3.5. In that excerpt, a when-clause codes a nucleus rather than a satellite.
Linguists often characterize clause combinations like this one as very much
like coordination, thus reflecting sensitivity to the rhetorical facts.
We now return to the distinction between hypotaxis and embedding to
show that the hypothesis we have been arguing for could not have been made
without a clear distinction between these two types of grammatical construc-
tions.
4.5 Embedding
The characterization of hypotactic clauses as non-principal, appended,

etc. information isn't new, of course; cf. for example Wellander (1948: 473)
on Swedish. The important point we want to make in this paper is that the
rhetorical status of the clause is determined by the explicit rhetorical organiza-
tion of the type discussed above in Section 3. There have been arguments
against interpretations of the type exemplified in Wellander (1948). For exam-
ple, both Jespersen (1924: 105) and Andersson (1975: 9-10) argue against an
interpretation of main clause vs. subordinate clause as 'principal' vs. 'ap-
pended'. However, both writers use clauses embedded as participants in a
clause (subordination at the core level in Foley and Van Valin's terms (1984:
250 ff)) when they give examples intended to show that the distinction of
'principal' vs. 'appended' is invalid.
One of Jespersen's examples is
This was because he was ill
His observation is that "the principal idea is not always expressed in the 'prin-
cipal clause'" (the non-italicized part in his example). Jespersen's example is
a clause with another clause embedded as a constituent and is thus different
from the clause combinations under discussion in this paper (cf. Section 2.3.1.
above for similar example of embedding, quoted in context).28 However, this
difference is not explicit until we distinguish enhancing hypotaxis and embed-
ding: Jespersen's argument is correct, but only for embedding; there is no
reason to generalize to e.g. enhancing hypotaxis.
Andersson falls into the same kind of 'subordination' trap as Jespersen.
He refers to Wellander's observation that principal information is given in
main clauses and subordinate (or appended) information is given in subordi-
nate clauses. He writes:
According to such a generalization, a sentence like (1) should be more or
less without importance. But as far as I can see, it is not.
(1) That Sweden cooperates with Vietnam shows that
Sweden can hardly be regarded as a member of the free world
(Our italics, CM & ST)
The only word in (1) that is not a member of a subordinate clause is the
verb shows. Is that the only important word in the sentence? The answer is
obviously: No.
(Andersson, op cit.)
Andersson's example is one of embedding and not one of any kind of
hypotaxis. Consequently, although his remarks apply to embedded clauses of
this kind, the example is not a counterexample to our proposal. Rather, just
as Jespersen's example, it serves to indicate how extremely important it is not
to group embedding and hypotaxis together as 'subordination'. They are quite

different not only grammatically, but also from a discourse point of view.29
There is another problem with Andersson's remarks. It is worth pointing
out that 'important' is not the same as 'main', 'principal', or 'nuclear'. 'Subor-
dinate', 'appended', or 'satellite' information may also be important for the
success of a text. There is nothing to suggest that Wellander thought that 'sub-
ordinate' information was unimportant or 'more or less without importance'
in Andersson's terms.
What we hope to have shown in this section is that we can only arrive at
an understanding of the relation between hypotaxis and discourse structure
which we have been arguing for in this paper if we carefully distinguish
hypotaxis from embedding, for they can be shown to have radically different
discourse functions.
4.6 Examples of non-necessary alleged properties of hypotaxis
Some of the traditional criteria for hypotaxis include variations in word

order (as in Swedish, Dutch, and German), dependence, givenness, and pre-
posability, among others. Given that there are in the literature discussions of
counterexamples to most of these criteria (see, e.g., Andersson, 1975; Davi-
son, 1979; and Schuetze-Coburn, 1984), we will confine ourselves to just a few
examples here.
4.6.1 Information status: Are hypotactic clauses 'known'?
Our discussion of the typically satellite role of certain clause types may
help us to uncover what the discourse parameter is which underlies claims
about 'subordinate' clauses being 'known' or 'given' (see, e.g., Winter, 1982
and Givón, 1979a, 1982).
For example Winter (1982:45) insists on "the contrast between indepen-
dent clause and subordinate clause in respect of their information status as
'new' and known' or 'given'". But this claim is simply not supported by the
facts: none of the adverbial clauses in the texts we have considered in this
paper (which Winter would consider 'subordinate'), for example, are 'known'
in the sense that the reader is expected to know the information they express
in advance.30 As an example, let us consider the "Thumb Heredity" text once
more:
1. Thumbs began to be troublesome about 4 months ago

2. and I made an appointment with the best hand surgeon in the Valley
3. to see if my working activities were the problem.
4. Using thumbs is not the problem
5. but heredity is
6. and the end result is no use of thumbs
7. if I don't do something now.
Each of the two italicized clauses, Units 3 and 7, is one which most gram-
marians (including Winter) would be content to consider 'subordinate', yet
there is no sense in which the information in either of them can be said to be
'known' (see Dillon, 1981:129ff for further arguments against considering
'subordinate' clauses to convey 'known' information).
Givón (1980:372) proposes that "subordinate-adverbial clauses tend to be
primarily presupposed", where by "presupposed", he means "not part of the
same assertion" as the main clause. Again, a perusal of Units 3 and 7 in the
above extract about the thumb problem suggests that matters are not quite so
straightforward, since both Units 3 and 7 could defensibly be argued to be
part of the same assertion as the main clauses with which they are associated.
(For a discussion of problems with the presupposition analysis, see
Andersson, 1975).
The italicized hypotactic clauses in the "Thumb Heredity" text both occur
in final position; functionally, they are non-thematic (in the sense discussed in
Section 4.3.4) and they are not 'given' or 'known'. There are, however, cer-
tain conditions under which hypotactic clauses encode information which can
be said to be 'given' or 'known'. In particular, a thematic clause may serve to
link the current clause combination to preceding text, for example by repeat-
ing or summarizing information already presented in the preceding text: See
Longacre (1970), Longacre and Thompson (1985), and Thompson (1985b) on
hypotactic purpose clauses relating to expectations raised in the text. This use
of thematic hypotactic clauses points to the connection between thematic
status and the method of development in a text as described by Fries (1983).
When a hypotactic clause is used to repeat or summarize prior text, there are
typically cohesive chains (in the sense of Halliday and Hasan, 1976); we find
ellipsis, substitution, anaphoric reference, lexical repetition, and so on. Thus
we find the anaphoric do + this in the following example with a thematic pur-
pose clause (italicized):
We have done enough for this first lesson on using the

UNIX editor, and are ready to quit the session
with edit. To do this we type 'quit' (or 'q') and
press carriage return: ...
(Kernighan)
But even a thematic hypotactic clause is by no means necessarily 'known'
or 'given' in the sense of repeating or summarizing information presented in
prior text. It often functions to re-orient the reader or listener by introducing
'new' information. The clause While attending Occidental College in the radio
announcer text presented in Section 3.2.5 above serves this kind of function;
it introduces a new frame.
To sum up our argument, we suggest that when grammarians describe
hypotactic clauses as 'given' or 'known' they treat one specific discourse use of
these clauses as characteristic of hypotactic clauses in general.
4.6.2 Elocutionary force: Are hypotactic clauses 'unchallengeable'?
In a later work, Givón (1982) considers 'subordinate' clauses to be 'un

challengeable', which brings us closer to a characterization of a role they often
play in texts. It is an interpretation of these clauses in terms of 'illocutionary
force' rather than information status ('known', 'given', etc.). Illocutionary
force seems like a more promising candidate, since it is signalled grammati-
cally by the mood selection of a clause, and hypotactically related clauses are
either non-finite (and hence without a mood selection) or finite but with an
invariably declarative mood selection. In other words, there is no way in
which illocutionary forces can be differentiated by different mood selections
in a hypotactically related clause, and the default declarative mood is not in
opposition with other mood selections. By 'unchallengeable' Givón does not
mean that the message of the clause literally cannot be challenged,31 but
rather that the writer is implicitly asking the reader to 'accept for the sake of
argument' the message expressed by certain clauses in relation to other
clauses. Dillon (1981:132) refers to this as 'subordinate clauses establishing
common grounds with the reader'. For instance, both Units 3 and 7 in the
above text, and many of the satellite clauses in our data, can be considered
'unchallengeable' in this sense. But even 'unchallengeability' may not cover
all occurrences of hypotactic clauses.32
For example, consider the italicized as-clause which forms Unit 3 of the
'Coffee Cup' text diagrammed above in Figure 8, repeated here:
(from an ISI researcher, message appearing on the ISI electronic bulletin

board)
1. Someone left a coffee cup in my office over the weekend.
2. Would the owner please come and get it
3. as I think things are starting to grow?
There seem to be no circumstances under which it would be appropriate
to call the as-clause in Unit 3 'unchallengeable', let alone 'known'; yet it is
clearly a hypotactic clause. Only certain occurrences of satellite clauses, then,
are 'unchallengeable' in their discourse context (see Chafe (1984) for further
discussion of this point).
What makes such satellite clauses often interpretable as 'unchallenge-
able', we suggest, is that certain text relations typically convey information
which is intended as unchallengeable with respect to the nuclear information.
One clear example can be found in the italicized Unit 3 in this extract from the
beginning of the advertisement for Syncom diskettes for cleaning floppy drive
heads:
1. What if you're having to clean floppy drive heads too often?
2. Ask for Syncom diskettes, with burnished Ectype coating and dust-
absorbing jacket liners.
3. As your floppy drive writes or reads,
4. a Syncom diskette is working four ways
5. to keep loose particles and dust from causing soft errors, dropouts.
As illustrated by Unit 3 in this extract, the satellite of a Circumstance
relation is typically unchallengeable with respect to its nucleus, since it con-
veys material which describes the circumstances under which the proposi-
tion(s) expressed by the nucleus holds. Readers are expected to accept these
circumstances as a context within which to interpret the nucleus. For such
relations, then, the satellites, and hence the hypotactic clauses which may
realize them, are often taken as unchallengeable. Such an interpretation is
especially likely when these satellites precede their nucleus, as discussed in
Chafe (1984), Dillon (1981: 134ff), Longacre and Thompson (1985), and
Thompson (1985b).33 Since not all relations are of this type, however, it seems
a mistake to equate either hypotaxis or satellite status with 'unchallengeable'.
4.6.3 Another alleged grammatical property: pronominalization
Backward pronominalization in English has been suggested as a charac-

teristic of what we are calling hypotactic relations. However, it does not serve
to single out hypotactic clauses. On the one hand, it works with constituent
phrases (in thematic position) as well as clauses. For example:
By the door of her1 son's room the mother1 knelt
upon the floor and listened for some sound from within.
(S. Anderson)
On the other hand, we find the pattern in coordinative clause combina-
tions (the first two examples), just as in hypotactic clause combinations (the
third example):
I found him1 hilarious, but Antoine1 had a serious
and gifted side to him1 like most of our kind.
(T. Williams)
Other men on the farm about him1 worked too hard and were
too tired to think, but to think of the farm and to be
everlastingly making plans for its success was a relief for
Jessee .
(S. Anderson)
When he1 talked to George Willard, Wing Biddlebaum
closed his fists and beat with them upon a table or on
the walls of his house.
(S. Anderson)
In the coordinative examples, a rhetorical interpretation in terms of Antithesis
seems quite plausible, with the pronoun occurring in the thesis satellite. Thus
rhetorically, there is a satellite-nucleus relation, but grammatically, there is no
hypotaxis.
It is not clear what the patterns of 'backward pronominalization' with
respect to hypotaxis and parataxis are in naturally occurring discourse.34
Bolinger (1979) gives constructed examples that would seem to indicate that
the phenomenon is not confined to Antithesis relations nor to satellite-nucleus
relations in general. In both the natural coordinative examples above, the
referents had already been introduced at the point at which the backward pro-
nominalization occurred; the pronominalization was backward only in a local
sense, i.e., local to the grammatically combined clauses. Whether this would
be true of all such cases is a matter for empirical investigation. (It is also true
of the third hypotactic example: The three examples are quite parallel in this
respect.) What is clear, however, is that backward pronominalization is not a

criterion of hypotaxis.
5 CONCLUSION
In this paper we have tried to show that hypotactic clause combining is

best understood as a grammaticalization of the Nucleus-Satellite relations
which characterize the rhetorical organization of certain types of written dis-
course.
In other words, the independently motivated textual notion of nuclearity
can be seen to underlie the grammatical notion of hypotaxis. In attempting to
answer the question of what discourse function motivates this aspect of gram-
mar, then, we are able to offer a more satisfying account of the phenomenon
than have previous approaches, including traditional ones, which have
attempted to characterize hypotaxis in sentence-level semantic terms. We
hope to have shown that systematicity in grammatical structure flows out of
systematicity in functional demands placed on language.
We have also suggested that there is no advantage to postulating a gram-
matical category of 'subordinate' clause; rather the grammar of English at
least, and perhaps of other languages as well, suggests that a distinction
between what we have been calling 'hypotaxis' and 'embedding' is crucial.
There is an interesting consequence of these suggestions for attempts to
consider clause combining from a cross-linguistic perspective: if hypotaxis in
English is a grammaticalization of rhetorical relations, then it follows that the
grammar of clause combining may differ radically from one language to
another. Indeed, preliminary discourse-based investigations of such
phenomena in unrelated languages strongly suggests that this is the case. If
the basic approach to clause combining taken in this paper is correct, then the
interesting cross-linguistic issue is how and to what extent the grammar of
clause-combining in a given language reflects the rhetorical organization of
discourse in that language (for some intriguing suggestions on differences
among languages with respect to rhetorical organization, see Hinds, 1983).
NOTES
1. We are grateful to the Netherlands Institute for Advanced Study for fellowship support for
Sandra Thompson during part of the preparation of this paper. Christian Matthiessen's part
of the work was supported by the Air Force Office of Scientific Research contract no.
F49620-79-C-0181. The views and conclusions contained in this document are those of the
authors and should not be interpreted as necessarily representing the official policies or
endorsements, either expressed or implied, of the Air Force Office of Scientific Research of
the U.S. Government. We are also grateful to a number of people for discussion, advice,
and feedback on the ideas in this paper: Henning Andersen, Joan Bybee, Susanna Cum-
ming, Cecilia Ford, Barbara Fox, Erica Garcia, T. Givón, Mike Hannay, Ruqaiya Hasan,
Nikolaus Himmelmann, Teun Hoekstra, Christian Lehmann, Lachlan Mackenzie, Lynell
Marchese, Marianne Mithun, Tom Payne, Hans-Jürgen Sasse, Anne Stewart, and Sebastiaō
Votre. None of these people necessarily agrees with the use we may have made of their
input, but we hope all of them will recognize their positive influence on the shape it has
taken. The contribution of each of the authors is equal.
2. See for example, Andersson (1975), Beekman and Callow (1974), Davison (1979), (1981),
Grimes (1975), Haiman and Thompson (1984), Halliday (1985a), Halliday and Hasan (1976),
Handke (1984), G. Lakoff (1984), R. Lakoff (1984), Longacre (1970), (1976), (1983),
McCray (1981), (1982), Martin (1983), Olson (1981), Payne (1973), Posner (1972), Quirk et
al (1985), Rappaport (1979), Rutherford (1970), Schiffrin (1985b), Smaby (1974), Sopher
(1974), Talmy (1978a, b), Van Valin (1984).
3. They make finer distinctions and separate adjunct clauses from disjunct clauses, but they
consider both to be instances of embedding.
4. Longacre (1970) differs from Halliday (1985a) in that he takes a clause combination to con
stitute a sentence. A sentence may have constituents other than clauses, but the important
point in the present context is that there is a distinction between being a constituent of a
clause and being a constituent of a sentence.
5. This is not to suggest that all traditional grammarians who made a distinction between coor
dination and subordination believed that it is a clear-cut dichotomy. For example, Kruisinga
(1932: 501) remarks: "It is perhaps hardly necessary to observe that the distinction between
coordination and subordination is a relative one, allowing of intermediate cases." He discus
ses cases of apparent coordination and apparent subordination, drawing a distinction
between formal appearance (of coordination or subordination) and real function. Thus, a
clause may appear to be subordinate, but function as a main clause. Kruisinga does not,
however, revise the distinction between coordination and subordination or make his criteria
very explicit.
6. These are all instances of rankshifted clauses in Halliday's terms; cf. Halliday (1961).
Clauses of reported eech combined with clauses of 'saying' are not treated as embedded
clauses by Halliday; cf. Halliday (1980, 1985a). See also Haiman and Thompson (1984: 519-
520), Longacre (1970: 266-267), and Munro (1982); cf. also Foley and Van Valin (1984:
252), who recognize that such clauses are not objects, though they still think of them as
embedded.
7. Parataxis is thus more general than the traditional notion of coordination. But since our
focus is on hypotaxis, we will not discuss Halliday's notion of parataxis.
T H E STRUCTURE O F DISCOURSE A N D 'SUBORDINATION' 319
8. For discussions in the systemic literature of embedding, clause complexes, parataxis, and
hypotaxis, see e.g. Halliday (1963) on intonational differences, Halliday, Mcintosh, and
Strevens (1964: 25-28), Halliday (1965), Huddleston (1965), Scott, Bowley, Brockett, Brown
and Goddard (1968), Hudson (1968), Muir (1972: Section 2.4), Berry (1975: chapters 6 and
7), Young (1980: chapters 17 - 21), Halliday and Hasan (1976: 136), Halliday (1979), Halli-
day 1985b), and Halliday (1985a: chapter 7). In addition to hypotaxis and parataxis, Hudson
operates with a third type of clause combining, indeterminate between the two. He uses it
for appositional and asyndetic clause combinations. In a different framework, Foley and Van
Valin (1984) and Van Valin (1984) propose a typology of clause combining where they rec-
ognize 'cosubordination' in addition to the traditional notions of coordination and subordi-
nation. They suggest that the concept cosubordination was Originally developed' in Olson
(1981), but in fact it is similar to Halliday's earlier notion of hypotaxis; both entail depen-
dency but not embedding. Their three-way distinction is thus comparable to Halliday's
parataxis vs. hypotaxis vs. embedding, but they don't use the distinctions in the same way as
he does for English. For example, they treat 'adverbial clauses' as embedded in the
periphery of another clause, which is very much like the traditional analysis we have
rejected, but Halliday treats them as hypotactically related to (and not embedded within)
another clause. Their distinction between cosubordination and subordination is also reminis-
cent of Longacre's distinction between inner peripheries in sentence structure and con-
stituents of clause structure (Longacre, 1970).
9. Names referring to the type of relation in the literature have included 'logico-semantic rela-
tion', 'conjunctive relation', 'rhetorical relation', 'rhetorical predicate', 'interclausal rela-
tion', and 'clause relation'.
10. The two sets of distinctions, degree and type of interdependence, are not arbitrary, of
course; they are also the dimensions linguists have recognized in rhetorical organization: cf.
Section 3 and references cited there.
11. For a discussion of the uses of the term and a carefully defined use of it, see Lehmann's
paper in this volume.
12. In this respect, the potential for terminological confusion is very similar to the problem with
the term adverb. Adverb is sometimes used to refer to a class of word, sometimes to a func-
tion in the clause (= adverbial or adjunct).
13. As König and van der Auwera show in their paper in this volume, word order in German or
Dutch in any case varies according to discourse factors even in so-called 'subordinate'
clauses.
14. With the exception of 'reported speech' related to clauses of saying; cf. remarks in Note 6.
15. The size of the units is arbitrary; we choose the clause as the basic reference point for ease
of exposition.
16. Grimes (1975) makes a similar distinction between two types of what he calls 'rhetorical
predicates', which are very similar to our text relations. By 'rhetorical predicates' he means
predicates which take propositions, rather than noun phrases, as their arguments. See Mann
and Thompson (1985a) for discussion of the similarities and differences between Grimes'
theory and ours. Grimes (p. 209) distinguishes between 'paratactic' predicates, which 'domi-
nate all their arguments in coordinate fashion', and 'hypotactic' predicates, which 'relate
their arguments to a proposition that dominates them' (p. 212). Beekman, Callow, and
Kopesec (1981) also make a similar distinction between addition and head-support.
320 CHRISTIAN MATTHIESSEN A N D SANDRA A. THOMPSON
17. These relations differ among themselves to some extent since they may reflect more closely
either relations among 'real-world' events or more closely considerations of text organiza-
tion. We return to this issue below in Section 3.2.7 and in Figure 14.
18. This distinction is reminiscent of the 'nucleus-margin' distinction of tagmemic linguistics, as
exemplified in the analysis of texts in Pike and Pike (1983:chap. 1).
19. In this respect, the process of producing text is like human action (behaviour) in general
undertaken in the pursuit of goals; writing or speaking is symbolic action. Planning in gen-
eral seems to rely on the distinction between central goals and subsidiary or supplementary
goals.
20. This theory, which we call Rhetorical Structure Theory, is being developed by William C.
Mann and us, with valuable input from Barbara Fox, at the USC Information Sciences Insti-
tute (ISI) in Los Angeles in the context of work on text generation, designing computer pro-
grams that have some of the capabilities of authors. It has been influenced by the work of
Beekman and Callow (1974), Beekman, Callow, and Kopesec (1981), Crothers (1979),
Grimes (1975), Halliday and Hasan (1976), Hobbs (1979), (to appear), Longacre (1976),
(1983), Martin (1983), and McKeown (1982).
21. These texts ranged from one paragraph to several pages in length, of the following types:
administrative memos, personal letters and letters to the editor, advertisements, Scientific
American articles, newspaper articles, organizational newsletter articles, public notices in
magazines, travel brochures, and recipes. There seems to be no limit in principle on the
length of the text whose organizational structure can be analyzed in terms of this theory.
22. It is an open question whether analysis of texts from other languages will reveal the same set
of relations as has our analysis of English texts. The implication in the work of people
involved with translation, such as Beekman, Callow, and Longacre (cited above) is that the
proposed lists of relations are cross-linguistically valid, but our impression is that texts from
other cultures may be organized according to different conventions and hence may be call
for a slightly different set of relations from those we propose here. For example, Thomas
Payne (p.c.) has shown that relations of location are significant in the oral prose of some lan-
guages. The possibility of such differences has, of course, implications for differences in
grammatical coding.
23. In the case of a text with more than one central goal, of course, a multi-nuclear structure
would be the best representation.
24. This is a partial typology in that there are more categories which could have been included;
it suffices for our present concerns, however.
25. It is possible to identify cross-classifications in this taxonomy. For example, some of the rela-
tions of type (i) are enhancing (cf. below) just as are some of the relations of type (ii). How-
ever, for our purposes the typology presented here is sufficient. Although we will not pursue
this matter here, we may note that the rhetorical distinction between (i) and (ii) is conse-
quential for clause combining.
26. It is clearly possible to expand the typology of enhancing relations, e.g. to posit a causal-con-
ditional subtype of enhancing relations (what Grimes (1975) calls 'co-variation'), with its
own subtypes Cause, Reason, Purpose, and Condition, However, we have taken the typol-
ogy as far as is necessary for the purpose of this paper.
T H E STRUCTURE O F DISCOURSE A N D 'SUBORDINATION' 321
27. If we analyse text in genre-specific terms, we find a type of organization that differs from the
rhetorical organization we have identified and described in this paper: that is the rhetorical
organization found in the generic structure of a narrative, a nursery tale, a folk story, a ser-
vice encounter, an advertisement, and so on. These structures are specific to a particular
genre of discourse. They are exocentric rather than endocentric and as Halliday has pointed
out (Halliday, 1981, 1982) there is an analogy between this kind of discourse organization
and the organization of a clause. The tentative generalization is that there are two modes of
organization simultaneously in a text, an exocentric kind (the generic structure of a particu-
lar genre) and an endocentric kind (what we have called rhetorical structure). There is an
analogy between one of these, the exocentric one, and the clause, and between the other,
the endocentric one, and the clause combination.
28. Jespersen's example illustrates a not uncommon strategy: A consequence (of a cause) is
stated in one clause. It is then referred to anaphorically in the next clause, in which its cause
is ascribed to it.
29. Incidentally, it is not at all clear that Andersson's example is a counterexample to what Wel-
lander had in mind. In the section where Wellander makes his observation, he characterizes
the subordinate clause as follows: "(it) determines the main clause in one respect or another,
specifying more specific circumstances: time, place, manner, cause, consequence, purpose,
etc." (1948: 473). This characterization fits adverbial clauses much better than it fits embed-
ded complement clauses, but as long as both are known as subordinate clauses it is hard to
be sure which type claims like Wellander's apply to.
30. Kruisinga, as long ago as 1932, made the same observation (p. 410): he suggested that differ-
ent kinds of clauses of cause differ in information status: "A clause of cause can bring for-
ward a cause that is an explanation of an action or occurrence in order to inform the reader
of this explanation; but it may also take the reader's knowledge for granted, and serve only to
remind him of the reason for the action of the main. The most important conjunctions in
clauses expressing a reason that is assumed to be known or acknowledged as correct, are 'as'
and 'since'." (Our italics, CM and ST.) There is, in fact, no particular reason to think that
the differentiation of different degrees of clause combining should serve to signal differences
in information status. As e.g. Halliday (1967) has shown, information status is signalled by
tonal prominence in English.
31. As Michael Halliday, p.c., in attributing the property 'arguable' to a given clause, has
pointed out, it is often possible, in fact, to explicitly deny or question the proposition in a
hypotactically related clause. Huddleston (1965: 583) points out that non-defining relative
clauses (elaborating hypotactic clauses) may contain interrogative tags: They're going to
Alford, which is near Skegness isn't it The tag serves to invite the listener's comment on
whether he/she agrees with the information offered by the speaker (cf. Halliday (1985a)),
thus allowing for a challenge.
32. As an indication, we note that it is possible to at least construct examples of although or
because clauses with tags (cf. footnote 31), as in: You should register for classes tomorrow,
although you're taking only the seminar, aren't you?, or We'd better take an umbrella, because
it might rain, don't you think?
33. This was also noticed by Kruisinga (1932: 465): "Clauses of reason will mostly precede, just
as those of cause will generally follow; this explains why as-clauses of this type precede,
those with because follow. Of course, this is not a law, ...". His distinction between reason
and cause was mentioned above.
34. Carden (1982) offers an enlightening discussion of the phenomenon with examples from
actual texts, but all of them involve hypotactic clauses.
REFERENCES
Allén, Sture (ed.). 1982. Text Processing. (Proceedings of Nobel Symposium

51.) Stockholm: Almqvist & Wiksell.
Allerton, D.J., Carney, E., and Holdcroft, D. (eds.). 1979. Function and
Context in Linguistic Analysis. Essays Offered to William Haas. Cam-
bridge: Cambridge University Press.
Andersson, Lars-Gunnar. 1975. Form and Function of Subordinate Clauses.
Gothenburg Monographs in Linguistics 1. Göteborg: University of
Göteborg Department of Linguistics.
Bailey, Charles-James and Roger Shuy. (eds.). 1973. New Ways of Analyzing
Variation in English. Georgetown: Georgetown University Press.
Beaman, Karen. 1982. "Coordination and subordination revisted. Syntactic
complexity in spoken and written narrative discourse'. In: Tannen 1982:45-
80.
Beekman, John and John Callow. 1974. Translating the Word of God. Grand
Rapids, MI: Zondervan Publishing House.
J. Callow, and M. Kopesec. 1981. The Semantic Structure of Written
Communication. Dallas: Summer Institute of Linguistics.
Benson, James D. and William S. Greaves (eds.). 1985. Systemic Perspec
tives on Discourse, Vol. 1. (Advances in Discourse Processes, Vol. XV.)
Norwood, N.J.: Ablex.
.(eds.),(in press). Systemic Functional Approaches to Discourse: Selected
Papers from the 12th International Systemic Workshop. Norwood, N.J.:
Ablex.
Berry, Margaret. 1975. Introduction to Systemic Linguistics 1. Systems and
structures. London: Batsford.
Bloomfield, Leonard. 1933. Language. London: Allen & Unwin.
Bolinger, Dwight. 1977. "Pronouns and repeated nouns". Bloomington:
Indiana University Linguistics Club.
— . 1979. "Pronouns in discourse". In Givón 1979b: 289-309.
Brøndal, Viggo. 1937. "The problem of hypotaxis". Reprinted in Househol
der 1972.
Carden, Guy. 1982. "Backwards anaphora in discourse context". Journal of
Linguistics 18:361-387.
Chafe, Wallace. 1980a. "The deployment of consciousness in the production

of a narrative". In: Chafe 1980b: 9-50.9-50.
(ed.). 1980b. The Pear Stories. Norwood, N.J.: Ablex.
. 1984. "How people use adverbial clauses". Proceedings of the Tenth
Annual Meeting of the Berkeley Linguistics Society, 437-449. Berkeley:
Berkeley Linguistics Society.
Cole, Peter (ed.). 1981. Radical Pragmatics. New York: Academic Press.
Copeland, James E. and P. Davis (eds.). 1980. The 7th LACUS Forum. Col
umbia, S.C.: Hornbeam Press.
Crothers, Edward. 1979. Paragraph Structure Inference. Norwood, N.J.:
Ablex.
d'Arcais, Giovanni B. Flores. 1978. "The perception of complex sentences".
In: Levelt and d'Arcais 1978.
Davison, Alice. 1979. "Some mysteries of subordination". Studies in the Lin
guistic Sciences 9(1):105-128.
. 1981. "Syntactic and semantic indeterminacy resolved. A mostly prag
matic analysis for the Hindi conjunctive participle". In Cole 1981:101-128.
Van Dijk, Teun Α. (ed.). 1985. Handbook of Discourse Analysis. Vol. 2.
Dimensions of Discourse. New York: Academic Press.
Dillon, George L. 1981. Constructing Texts: Elements of a Theory of Com
position and Style. Bloomington: Indiana University Press.
Ellis, Willis D. (ed.). 1938. A Source Book of Gestalt Psychology. London:
Routledge & Kegan Paul.
Farr, Cynthia, Ivan Lowe, and Carl Whitehead. 1985. "Explanation in English
and Korafe". In: Benson and Greaves 1985: 135-152.
Foley, William and Robert Van Valin. 1984. Functional Syntax and Universal
Grammar. Cambridge: Cambridge University Press.
Ford, Cecilia and Sandra A. Thompson. 1986. "Conditionals in discourse: a
text-based study from English". In: Traugott et al. 1986.
Fries, Peter H. 1983. "On the status of theme in English: arguments from dis
course". In: Petöfi and Sözer 1983: 116-152.
Givón, T. 1979a. On Understanding Grammar. New York: Academic Press.
. 1979b. (ed.) Discourse and Syntax. (Syntax and Semantics Vol. 12.)
New York: Academic Press.
. 1980. "The binding hierarchy and the typology of complements".
Studies in Language 4(3):333-377.
. 1982. "Logic vs. pragmatics, with human language as the referee: toward
an empirically viable epistemology". Journal of Pragmatics 6(2):81-133.
Golkova, Eva. 1968. "On the English infinitive of purpose in functional sen-
tence perspective". Brno Studies in English 7:119-128.
Greenberg, Joseph, Charles Ferguson, and Edith Moravcsik (eds.). 1978.
Univers als of Human Language. Stanford: Stanford University Press.
Grimes, Joseph E. 1975. The Thread of Discourse. The Hague: Mouton.
Haiman, John and Sandra A. Thompson. 1984. "'Subordination' in universal
grammar". Proceedings of the Tenth Annual Meeting of the Berkeley Lin
guistics Society, 510-523. Berkeley: Berkeley Linguistics Society.
Halliday, Michael A.K. 1961. "Categories of the theory of grammar". Word.
17(3):241-92.
1963. "Intonation and English grammar". Transactions of the Philologi
cal Society, 143-69.
—-. 1965. "Types of structure". In: Halliday and Martin 1981:29-42.
. 1967. "Notes on transitivity and theme", Part II. Journal of Linguistics
3:199-245.
. 1979. "Modes of meaning and modes of expression. Type of grammatical
structure, and their determination by different semantic functions". In:
Allerton, Carney, and Holdcroft, 1979: 57-80.
1981. "Text semantics and clause grammar: some patterns of realiza-
tion". In: Copeland and Davis 1980:31-60.
1982. "How is a text like a clause?" In: Allén 1982:209-49.
1985a. An Introduction to Functional Grammar. London: Edward
Arnold.
1985b. "Dimensions of discourse analysis. Grammar". In Van Dijk
1985:29-57.
, and Ruqaiya Hasan. 1976. Cohesion in English. London: Longman.
, and James R. Martin (eds.). 1981. Readings in Systemic Linguistics. Lon-
don: Batsford.
, Angus Mcintosh and Peter Strevens. 1964. The Linguistic Sciences and
Language Teaching. London: Longmans.
Handke, Jürgen. 1984. Descriptive and Psycholinguistic Aspects of Adverbial
Subordinate Clauses. Heidelberg: Groos.
Hinds, John. 1983. "Contrastive rhetoric: Japanese and English". Text
3(2):183-195.
Hobbs, Jerry. 1979. "Coherence and coreference". Cognitive Science 3:67-90.
To appear. "On the coherence and structure of discourse". In: Polanyi,
to appear.
Hopper, Paul J. and Sandra A. Thompson (eds.). 1982. Studies in Transitiv
ity. (Syntax and Semantics Vol. 15.) New York: Academic Press.
1984. "The discourse basis for lexical categories in universal grammar".

Language 60(4):703-752.
Householder, Fred (ed.). 1972. Syntactic Theory 1. New York: Penguin
Books.
Huddleston, Rodney. 1965. "Rank and Depth". Language 41: 574-86.
, Richard Hudson, A. Henrici, and Eugene Winter. 1968. "Sentence and
clause in scientific English". University College, London: Communication
Research Centre.
Hudson, Richard A. 1968. "The Clause-Complex". In: Huddleston, Hudson,
Henrici, and Winter.
Jespersen, Otto. 1924. The Philosophy of Grammar. London: Allen &
Unwin.
Koffka, Kurt. 1935. Principles of Gestalt Psychology. New York: Harcourt,
Brace & World.
Köhler, Wolfgang. 1929. Gestalt Psychology. New York: Liveright.
Kruisinga, E. 1932. A Handbook of Present-Day English. Part II. (English
Accidence and Syntax. Fifth edition. Groningen: Noordhoff.
Kuno, Susumo. 1973. The Syntactic Structure of Japanese. Cambridge: MIT
Press.
Labov, William. 1973. "The boundaries of words and their meanings". In:
Bailey and Shuy 1973, 340-373.
Lakoff, George. 1984. "Performative subordinate clauses". Proceedings of the
Tenth Annual Meeting of the Berkeley Linguistics Society, 437-449. Ber-
keley: Berkeley Linguistics Society.
Lakoff, Robin. 1984. "The Pragmatics of Subordination". Proceedings of the
Tenth Annual Meeting of the Berkeley Linguistics Society, 481-492. Ber-
Lanin, Dene. 1977. "You can take the sentence out of the discourse but you
can't take the discourse out of the mind of the speaker". Papers from the
Thirteenth Regional Meeting of the Chicago Linguistic Society, 288-301.
Chicago: Chicago Linguistic Society.
Lerdahl, Fred and Ray Jackendoff. 1983. A Generative Theory of Tonal
Music. Cambridge: MIT Press.
Lenerz, J. 1984. Syntaktische Wandel in Grammatiktheorie: Eine Unter
suchung an Beispielen aus der Sprachgeschichte des Deutschen. Tübingen:
Niemeyer.
Levelt, W.J.M. and G.B. d'Arcais (eds.). 1978. Studies in the Perception of
Language. New York: Wiley.
Longacre, Robert. 1970. "Sentence structure as a statement calculus. Lan

guage 46:783-815.
. 1976. An Anatomy of Speech Notions. Lisse: The Peter de Ridder Press.
1983. The Grammar of Discourse. Notional and Surface Structures. New
York: Plenum Press.
. and Sandra A. Thompson. 1985. "Adverbial clauses". In Shopen
1985:171-235.
Lyons, John. 1968. Introduction to Theoretical Linguistics. Cambridge: Cam-
bridge University Press.
McCray, Alexa. 1981. "Clause initial elements in German". Proceedings of
the Seventeenth Regional Meeting of the Chicago Linguistic Society, 205-
220. Chicago: Chicago Linguistic Society.
. 1982. "The left dislocation structure in German". Proceedings of the
Eighteenth Regional Meeting of the Chicago Linguistic Society, 336-347.
Chicago: Chicago Linguistic Society.
McKeown, Kathleen R. 1982. "Generating natural language text in response
to questions about database structure". University of Pennsylvania Ph.D.
dissertation.
Mann, William C. 1984. "Discourse structures for text generation". ISI
Research Report ISI/RR-84-127; available from Information Sciences
Institute, 4676 Admiralty Way, Marina del Rey, Ca. 90292-6695.
. and Sandra A. Thompson. 1985. "Assertions from discourse structure".
Proceedings from the Eleventh Annual Meeting of the Berkeley Linguistics
Society. Berkeley: Berkeley Linguistics Society.
. 1986. "Relational propositions in discourse". Discourse Processes.
9(l):57-90. Also available as ISI Research Report ISI/RR-83-115; from
Information Sciences Institute, 4676 Admiralty Way, Marina del Rey, Ca.
90292-6695.
. To appear. "Rhetorical structure theory: a theory of text organization
and its implications for clause combining". In: Polanyi (to appear).
Martin, James R. 1983. "Conjunction: the logic of English text". In: Petöfi
and Sözer 1983:1-72.
Mithun, Marianne. 1984. "How to avoid subordination". Proceedings from
the Tenth Annual Meeting of the Berkeley Linguistics Society, 493-509. Ber-
Muir, James. 1972. A Modern Approach to English Grammar: An Introduc
tion to Systemic Grammar. London:Batsford.
Munro, Pamela. 1982. "On the transitivity of 'Say' verbs. In: Hopper and
Thompson 1982:301-19.
Olson, Michael. 1981. "Clause junctures in Barai". Australian National

University Ph.D. dissertation.
Payne, John R. 1973. "Sentence conjunction and subordination in Russian".
You Take the High Node and Tll Take the Low Node. Papers from the
Comparative Syntax Festival, 305-325. Chicago: Chicago Linguistic Soci-
ety.
Petöfi, János S. and Emel Sozer, (eds.). 1983. Micro and Macro Connexity of
Texts. Hamburg: Helmut Buske Verlag.
Pike, Kenneth and Evelyn Pike. 1983. Text and Tagmeme. Norwood, N.J.:
Ablex.
Polanyi, Livia, ed. To appear. Discourse Structure. Norwood, N.J.: Ablex.
Posner, Roland. 1972. Theorie des Kommentierens. Frankfurt: Athenaum.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik.
1985. A Comprehensive Grammar of the English Language. London and
New York: Longman.
Rappaport, Gilbert. 1979. "Detachment and adverbial participle clauses in
Russian". UCLA Ph.D. dissertation.
Reinhart, Tanya. To appear. "Principles of gestalt perception in the temporal
organization of narrative texts". Linguistics.
Robins, R.H. 1966. "The development of the word class system of the Euro-
pean grammatical tradition". Foundations of Language, 2.
Rutherford, William. 1970. "Some observations concerning subordinate
clauses in English". Language 46:97-115.
Schiffrin, Deborah, (ed.). 1985a. Meaning, Form, and Use in Context: Lin
guistic Applications. Proceedings of the 1984 Georgetown University
Roundtable on Linguistics. Georgetown: Georgetown University Press.
. 1985b. "Multiple constraints on discourse options: quantitative analysis
of causal sequences". Discourse Processes 8(3):281-303.
Schuetze-Coburn, Stephan. 1984. "On the borders of subordination: signaling
relative clauses in spoken German". Proceedings from the Tenth Annual
Meeting of the Berkeley Linguistics Society, 650-659, Berkeley: Berkeley
Linguistics Society.
Scott, F.S., C.C. Bowley, C.S. Brocket, J.G. Brown, and P.R. Goddard.
1968. English Grammar: A Linguistic Study of its Classes and Structures.
London: Heineman Educational Books.
Shopen, Timothy, (ed.). 1985. Language Typology and Syntactic Description.
Cambridge: Cambridge University Press.
Smaby, Richard. 1974. "Subordinate clauses and asymmetry in English".
Journal of Linguistics 10:235-269.
Sopher, H. 1974. "Coordinate and subordinate clauses — a formal analysis of

the traditional classification". English Language Teaching Journal
28(3):202-213.
Talmy, Leonard. 1978a. "Figure and ground in complex sentences". In:
Greenberg et al 1978:625-649.
Talmy, Leonard. 1978b. "Relation between subordination and coordination".
In: Greenberg et al 1978:487-513.
Tannen, Deborah (ed.). 1982. Coherence in Spoken and Written Discourse.
Norwood, N.J.: Ablex.
Thompson, .Sandra A. 1985a. "'Subordination' in formal and informal dis-
course". In: Schiffrin 1985a:85-94.
1985b. "Grammar and written discourse: initial vs. final purpose clauses
in English". Text 5(l/2):55-84.
1987. "'Subordination' and narrative event structure". In: Tomiin 1987:
435-454.
Tomlin, Russell. 1985. "Foreground-background information and the syntax
of subordination". Text 5(1/2):85-122.
, (ed.). 1987. Coherence and Grounding in Discourse. Amsterdam:
John Benjamins.
Townshend, David J. and Thomas Bever. 1977. "Main and subordinate
clauses: a study in figure and ground". Bloomington: Indiana University
Linguistics Club.
Traugott, Elizabeth, Charles Ferguson, Judy Reilly, and Alice ter Meulen
(eds.). 1986. On Conditionals. Cambridge: Cambridge University Press.
Van Valin, Robert. 1984. "A typology of syntactic relations in clause lin-
kage". Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics
Society, 542-558. Berkeley: Berkeley Linguistics Society.
Wellander, Erik. 1947. Riktig svenska. Stockholm: Almqvist & Wiksell.
Wirth, Jessica (ed.). 1985. Beyond the Sentence: Discourse and Sentential
Form. Ann Arbor: Karoma.
Winter, Eugene. 1982. Towards a Contextual Grammar of English. London:
George Allen and Unwin.
Young, David J. 1980. The Structure of English Clauses. New York: St. Mar-
tin's.
SOURCES OF DATA
Anderson, Sherwood. 1958. Winesburg, Ohio. New York: The Viking Press.
Carlson, L. 1978. The TraveLeer Guide to Mexico City. Chicago: Upland
Press.
Farrell, J.G. 1973. The Siege of Krishnapur. London: Penguin Books.
Kernighan, Β. 1978. UNIX For Beginners. 2nd edition. Bell Laboratories.
Munro, H.H. 1977. The Best of Saki. Edited by Graham Greene. London:
Penguin Books.
Williams, Tennessee. 1975. Memoirs. Doubleday.
The grammaticization of coordination
Marianne Mithun
Of all syntactic constructions, we might expect the most basic and

universal to be coordination. Coordination seems so fundamental to
language that its form should be relatively stable over time and constant
across cultures. Yet there is surprising variety in the types of coordinating
constructions exhibited by different languages, and even in the degree to
which coordination is grammaticized at all. Some languages contain
elaborate sets of obligatory conjunctions, some indicate syntactic links mor-
phologically with non-finite tense or switch-reference markers, but many
languages contain no grammaticized markers of coordination at all.
Does this variety mean that coordination is a language-specific strategy?
Do speakers of different languages differ in the kinds of constituents they
conjoin, or the occasions on which they link them? In fact, if intonation pat-
terns are compared across languages, striking similarities emerge. Speakers
of very different types of languages combine elements intonationally under
essentially the same circumstances, with comparable frequency.
In what follows, the grammaticization of coordinating conjunctions
will be compared across typologically diverse languages. First, patterns of
coordination reflected in intonation will be compared. Next, the types of
grammaticized conjunctions used to signal coordination will be surveyed,
along with their diachronic sources and the processes involved in their gram-
maticization. Finally, the stimuli initially motivating the grammaticization
of the conjunctions will be explored.
1 COORDINATION BY INTONATION
A survey of spontaneous spoken texts in a variety of languages indicates

332 MARIANNE MITHUN
that coordination is normally signalled intonationally in two principal ways,

whether overt conjunctions are present or not. Coordinate constituents may
be combined with no intonation break, or they may be separated by 'comma
intonation', usually a pause and a special non-final pitch contour. These two
patterns characterize conjoined noun phrases and predicates as well as con-
joined clauses.
1.1 Coordinate noun phrases
Noun phrases joined with no intonation break typically designate a

single conceptual unit. Conjoined noun phrases of this type often refer to
the sorts of concepts designated by single lexical items or compounds in
many languages, such as 'parents' for 'mother and father'. By contrast,
noun phrases separated by comma intonation typically designate conceptual-
ly distinct members of some set. Each new conjunct introduces a separate
piece of information.
This distinction is characteristic of all types of languages. Compare the
sentences below from Gurung, for example, a relatively analytic, Sino-
Tibetan language of central Nepal. During a discussion of Gurung tradi-
tions, a speaker had reported that when girls are first born, they are given
a ragged dress. They wear ragged dresses until the age of three or four, when
they are given a cape and skirt. The cape and skirt together constitute a
costume, a replacement for the ragged dress. The speaker conjoined these
two nouns, 'cape' and 'skirt', with no pause.
(1) Gurung (Glover, 1974: 210.22)
jxa:é qxywí coló pî-m
then cape skirt give-NP
'Then (we) give (the girls) a cape and skirt'
When conjoined nouns identify conceptually distinct entities, however, they
are separated by comma intonation. In (2), the conjoined nouns refer to two
distinct groups of people, Gurungs and Nepalis.
(2) Gurung (Glover, 1974: 201.1)
dasaē ŋi tamü-maé-la, nxepa:li-la dasaẽ ...
Dasain we Gurung-PL-τορ Nepali-τορ Dasain
asoda mxaina-r pardi-m
Asod month-in fall-NP
THE GRAMMATICIZATION OF COORDINATION 333
'Dasain, for us Gurungs, and Nepalis, falls in the month of

Asod'
These intonation patterns are not restricted to analytic languages. Kam-
chadal, a Luoravetlan language of Siberia, is highly polysynthetic. Verbs
contain pronominal affixes referring to core arguments, and may incor-
porate nouns. In (3) below, the conjoined noun phrase 'wife and son' con-
stitutes a conceptual unit, the family of the hero. The nouns are joined with
no break in intonation.
(3) Kamchadal (Worth, 1961: 18.8)
ŋic pēc gam inxkinkīn
wife son not he found.
'He did not find his wife and son'
In (4), by contrast, the conjoined nouns refer to assorted, conceptually
distinct items. They are separated by comma intonation.
(4) Kamchadal (Worth, 1961: 41.15-16)
Qtkítxen qtxzenk mnil áŋqa cgízin:
throw on the route everything that there is
sápān, qéyulxïn, lxmin, csálēn
reindeer hides hides of reindeer calves sable furs fox skins
Throw everything we have out onto the path: reindeer hides,
reindeer calves' hides, sable furs, fox skins'
The two intonation patterns can be combined to form constructions
with internal structure. Note the conjoined noun phrase below from
Parengi, a mildly synthetic South Munda (Afroasiatic) language of India.
The first two nouns, 'bow' and 'arrow', constitute a conceptual unit, so they
are linked intonationally. The third noun, 'gun', represents a separate entity,
so it is separated from the others by comma intonation.
(5) Parengi (Aze and Aze, 1973: 324.107)
ona?-di amon-di, noli-di d'on-ay
bow-Foc arrow-Foc gun-Foc take-sp/IMP
'Bring me a bow and arrow, and a gun.'
1.2 Coordinate predicates
Similar intonation patterns characterize conjoined predicates. Concep-

334 MARIANNE MITHUN
tually unitary events are expressed in single intonation units, while those
consisting of conceptually distinct components are expressed in series of in-
tonation units. In the Parengi sentence in (5), the two future tense verbs,
referring to subparts of a conceptually unitary action, are combined intona-
tionally with no pause.
e-no?n d'ar-t-ay zum-t-ay
to-him grasp-FUT-sp eat-FUT-sp
'I will grasp him and eat him'
The verbs in (6) are conjoined because they describe causally related events.
The events are conceptually distinct, however, so the verbs are separated by
comma intonation.
no?n kuy alung ir-ru, din-ru?
he well inside jump-PAST die-PAST-uNDERGOER
'He jumped inside the well and died.'
These two intonation patterns can be combined to yield complex
predicates, similar to the complex nominals illustrated above. The Kam-
chadal sentence in (7), for example, contains two conjoined predicates, the
second of which also contains two conjoined predicates. The hero's coming
to Walen-Sinanewt is seen as one action, then his throwing himself on her
and seizing her constitute a second.
Waleŋn-Sinarŋéwtanke kólknen, qanaŋ kspensknan
To Walen-Sinanewt he came thus he threw himself seized
kinknen
her
'He came to Walen-Sinanewt, threw himself on her and seized
her'
Examination of connected discourse in a variety of languages indicates
that the types of predicates speakers conjoin are remarkably similar. Com-
pound predicates most often position a major participant for a subsequent
action, as in the three examples above. Not surprisingly, these are the very
sorts of concepts that typically exhibit special bonding or fusion in many
languages. The bond may take a variety of forms, such as serial verb con-
structions, verb compounds, and verb stems derived with andative affixes
meaning 'go and ...'
1.3 Coordinate clauses
Conjoined clauses exhibit the same kinds of intonation patterns as con-

joined nominals and predicates. Those conjoined with no intonation break
typically describe subparts of what is conceived of as a single event. One
clause typically sets the stage for the other by positioning a major partici-
pant. The Gurung sentence in (8), in which agents are moved into position
for their subsequent action, is typical.
(8) Gurung (Glover, 1974: 207)
sxĩ-ĩ xrẽsyo mxi gxrí gxrí aba xrõsa xrõsa-é
houses-EMPH each person one one now self self-of
mró xya:-í bin xrösa xrösa-é mró-r-bae mxa:rsi
field go CONJ self self-of field-in-ADJ marsi
mlxa-e na: tí-na
rice-of head pluck-DI
One person from each household went to their respective
fields and plucked of their own field a head of marsi rice.'
The Kamchadal sentence in (9), in which an instrument is obtained for a
subsequent action, is also typical.
Qaná:ŋ láŋen klin xkálax sítlxpket pémacx
kánkwatan
thus the girls took burning firebrands the boy burned all over
T h e girls took the burning brands and burned the boy all
over'
By contrast, clauses separated by comma intonation typically represent
conceptually distinct aspects of an action, event, or scene. The conjoined
clauses most often describe sequential actions as in (10)
Ktkil-ín lil, ktéwsiknen ína, méyen ína,
she threw the line was taken up he, in such a way he
kétcaknan, kriatqazúknen, knúqzuknen, kneyúqzuknen
was dried out, he became glad, he ate, he became satiated
336 MARIANNE MITHUN
'She threw down the cord, he was pulled up, dried out, and
became happy; he ate, he became satiated'
They frequently relate causes and effects, as in (11).
Kíma o:zózk hïnc mílkicen ésxanke ténaq,
I tomorrow not will go to the father again
nanqwátaxmaŋ kíma xkálan sítlxpqel
one will burn me with hot firebrands
'I will not go to my father again, or they will burn me with
hot firebrands'
Sometimes they describe simultaneous aspects of a scene or state.
(12) Gurung (Glover, 1974: 204)
km xra: kúdi mxaé-m, kwí laĩ kudi mxaé-m
some ferris swing play-NP some long swing play-NP
'Some play on the ferris wheel, and some play on the long
swing.'
2 FORMAL MARKERS OF COORDINATION
Despite their similarities in intonation patterns, languages vary strikingl;

in the degree to which coordination is structurally marked. A large number
of languages lack any morphological or lexical indications of conjunction
whatsoever. Many others, of course, contain highly grammaticized markers
of syntactic coordination. Yet not even these languages mark all types of
coordination uniformly or obligatorily. In many languages, only certain
kinds of coordination are overtly marked. In many, coordinating conjunc-
tions are optional. Interestingly, the diachronic sources of the markers are
often still transparent. This suggests that it might be possible to trace the
path along which formal syntactic coordination can develop over time. The
differences in system of marking coordination among languages might
represent different stages in the grammaticization of conjunction.
2.1 Sources of noun phrase coordination
A surprising number of languages contain no grammaticized markers

of noun phrase coordination at all. Juxtaposition and intonation alone are

considered sufficient to signal conjunction, as in the Gurung examples in (1)
and (2). Whether coordination is overtly marked or not in a language,
however, conjoined noun phrases are actually relatively rare in spontaneous
discourse, especially in topic position. They are considerably rarer than con
joined clauses. A casual perusal of almost any spoken texts in any language
will quickly bear this out. This rarity is not altogether inexplicable: several
factors converge to minimize the need for conjoined nominals in connected
speech.
Most important, speakers typically introduce only one major piece of
information into discourse at a time (Chafe, 1987). Conceptually distinct en
tities are introduced by distinct intonation units, often separate clauses, as
in the Parengi example below.
tangk le-taring-tu, ao?m le-dey:-tu
cooked rice we-cook-FUT vegetable we-cook-FUT
'We will cook rice and vegetables'
Once they have been introduced individually, sets of entities can be referred
to collectively by plural pronouns, so the need for conjoined noun phrases
is bypassed.
A number of languages with bound pronominal affixes exhibit a related
mechanism for identifying joint arguments without conjoined noun phrases.
One entity is first introduced with a full noun phrase and established as
topic. Thereafter pronominal reference is sufficient to identify it. When an
additional entity is involved jointly in subsequent events, only this new one
need be identified by a noun prase. A dual or plural pronoun is sufficient
to indicate the joint participation. An example of such a construction can
be seen in the sentence below from Kathlamet, a Chinookan language of
Washington state. The narrator has been describing the adventures of a
woman abandoned by her husband.
(14) Kathlamet (Boas, 1901: 158.9)
A'qa guä'nEsum qasxalō'kcaitx agā'xan
then always they two picked berries her daughter
Then she and her daughter always went picking berries'
The same strategy can be used to identify third person members of first or
second plural arguments. Since speaker and hearer are inherently given by
338 MARIANNE MITHUN
the discourse context, pronominal reference is sufficient to identify them.

Only the additional third person need be identified by a separate noun
phrase.
(15) Kathlamet (Boas, 1901: 36.14)
Ewa' tga'qaqstakuks aqo'lEktca agE kikal anta'Lxana
thus their heads are roasted my wife our smelts
T h e heads of our smelts (those of my wife and me) are
roasted'
In many languages, another construction provides a substitute for con-
joined nominals. A topic is established, then subsequently referred to pro-
nominally. If an additional entity participates jointly in some event, it is
mentioned in a comitative construction. The original topic remains the same.
Note the position of the companion in the sentence below from Jacaltec, a
Mayan language of Guatemala.
(16) Jacaltec (Craig, 1977: 32)
Xc-in to ilo' kin boj ix
ASP-I go to.see fiesta and/with CL/her
'I went to see the fiesta with her'
or: 'She and I went to see the fiesta'
In some languages, the coordination of noun phrases is usually either
expressed by juxtaposition or simply avoided, but particles do occasionally
appear that seem to link coordinate noun phrases overtly. In many cases, the
appearance of the particles is not random: they are used only when a set is
first established, or when the speaker wishes to specify that the set is closed
or complete. The Kamchadal sentence cited earlier in (3) lacked a noun
phrase coordinator, but it was preceded in the origial text by a sentence con-
taining one. Both can be seen in (17).
Czálkutq ktxílnen ŋic ēk pēc. Kílknen ínan késtanke.
Czalkutz remembered wife and son he went to his house
Then Czalkutz remembered his wife and son and went home'
Kólknen késtanke, ŋic pêc qam inxkinkïn
he came to the house wife son not he found
'He came to his house, but did not find his wife and son'
Sometimes, their optionality reflects a change in progress: the grammaticiza-

tion of coordination. Often in these cases, the origin of the developing coor-
dinating conjunctions is still transparent.
Frequently, the source of noun phrase conjunctions is a comitative con-
struction, of the type originally used to circumvent coordinate noun phrases.
An example of the early stages of such a development can be seen in Sarcee,
an Athapaskan language spoken in Alberta. Cook (1984) reports that noun
phrases are usually conjoined without any conjunctions at all in Sarcee,
whether the resulting set represents a conceptual unit, or a diversified list.
(18) Sarcee (Cook, 1984: 87)
dítòò dóó-?ī iná άάηί-là
own-father own-mother-DET she-told
'She told her father and mother'
(19) ìstlí gútsìs dóóní ìcī ctcùd, gīní
horse scalp gun I-capture, they say
' "I captured horses, scalps, and guns," they say'
Sarcee noun phrases may, however, be joined by means of the comitative/in-
strumental postposition ih plus ila.
(20) Sarcee (Cook, 1984: 87)
ditòò dóó ìhílά-άάsnì-lά-ά
own father own mother with she told
'She told her father and mother'
(21) Sarcee (Cook, 1984: 96)
tcágúá k'ìyìdjí istla ká-àhílà kàdīla-là
weasel coat legging shoe-with he-brought-out
'He brought out weasel coats, leggings, and shoes"
The postposition is in the process of being grammaticized as a conjunction.
Cook points out that the last "verb stem above, -lá 'to handle plural objects'
clearly suggests the plurality of the object NP. Internal as well as com-
parative evidence suggests that the conjunctive use of ih is a relatively new
development" (1984: 96).
The later stages of such a development have been documented in several
Kwa languages of West Africa by Lord (1973). She shows that noun phrase
conjunctions have a clear source in comitative verbal constructions. Note the
translations of the sentences below.
340 MARIANNE MITHUN
(22) Gã (Lord 1973: 288)

àyí kè tè tètè fà wónù
Ayi with Tete drink soup
'Ayi drank soup with Tete' or 'Ayi and Tete drank soup'
tètè nà àyí kè kòkó
Tete saw Ayi and Koko
Tete saw Ayi and Koko'
kòfí kè a 'má tsè dzí ówúlà àgó
Kofi and Ama father is Mr. Ago
'Mr. Ago is the father of (both) Kofi and Ama'
The particle ke originated as a comitative verb preceded by a subject and
followed by an obejct. Traces of this origin can still be seen in number agree-
ment. When the subject of a sentence is a conjoined noun phrase, the
predicate agrees with only the first noun of the noun phrase, the original
subject of comitative verb.
(23) mì tá
'I sit (SG)'
wo trá
'We sit (PL)'
mì kè lè tà
'I and he sit (SG)'
The contexts in which the particle is now used, as well as the behavior of
nominal conjuncts in focus constructions, indicate that ke has now moved
beyond its original status as a simple verb to a grammaticized syntactic con-
junction.
Nominal conjunctions also frequently develop from a second source, an
adverbial particle meaning 'also, too, as well. The original function of such
a particle is to point out a parallelism between otherwise separate entities.
In Cayuga, a Northern Iroquoian language of Ontario, the particle hni'
still has the meaning 'too' or 'also'. It appears in independent clauses, in-
dicating their connection with other information in the discourse. This use
can be seen below. A speaker was describing his day in the bush, explaining
that he had been scratched by branches. His listener commented sym-
pathetically and at some length on how unlucky he was. After a moment the
victim remembered an additional mishap.

(24) Cayuga (Reginald Henry p.c.)
Akitakrá hnï she nyó: n'atõ:tá:ke:
I fell also as far I came back
'I fell on the way back, too.'
On another occasion, a customer in a hardware store had just selected a
hammer. The clerk then asked whether he wanted anything else. The
customer replied,
Ehẽ', Enestanya'kthá' kV hnï tewakatōhwētsó:nih
yes one cuts board with it just also I want
'Yes, I want a saw, too'
In both cases, hnï links the content of the statement to the rest of the
discourse, but the link is semantic, not syntactic. The fall in (24) is not syn-
tactically coordinate with other misadventures in previous discourse, nor is
the saw in (25) coordinate with the hammer selected earlier.
Although conjoined noun phrases are relatively rare in spontaneous
Cayuga discourse, as in other languages, when they do occur, they are
systematically followed by hnï. The particle has been grammaticized as a
coordinating conjunction. Noun phrases linked with hnï form a syntactic
constituent. When they identify subjects, their predicates bear plural pro-
nominal prefixes.
It is not difficult to imagine how a particle of this type could develop
from an adverbial into a syntactic conjunction. Such a particle would often
occur in potentially ambiguous contexts. Consider the sentence below. One
neighbor had asked what the family was going to plant that year. The reply
was:
A:yẽ:' akwé:, onẽhẽ' osahe'tá' ohõn'atá'õ hni'
it seems all corn bean potato-guess also
'Oh, I guess everything, corn, beans, potatoes, and squash'
With pauses separating the nouns, the squash could be interpreted as an
afterthought, an addition to the original list. As pauses are shortened, the
particle appears to close the list.
In many contexts, hnï now functions unambiguously as a coordinating
342 MARIANNE MITHUN
conjunction, where all constituents have obviously been thought of before

the utterance. During the above conversation, the neighbor asked whether
they had planted potatoes the summer before. The friend replied that no,
they were afraid the potaties would rot, so they had only planted corn and
beans.
(27) Ne:' tshõ: ne' onẽhẽ' sahe'tá' hni9 õkwayēthwē hne:'
it only the corn beans also we planted CONTR
'No, we only planted corn and beans'
Corn, beans, and squash constitute a prototypical set in Iroquoian
agriculture. Intonation confirms that the gardener had both the corn and
beans in mind at the outset.
A guest had been watching three children play, and asked what their
names were. His host replied,
(28) Junior, Helen, Hercules hni9
Junior Helen Hercules also
'Junior, Helen, and Hercules'
The original meaning 'also' is no longer present here, either. Hni' has
become completely grammaticized as a coordinating conjunction for noun
phrases. It is systematically present when nominals are coordinated, and
signals no more than the syntactic link between them.
Note that hni' does not appear with conjoined verbs or clauses, unless
it is filling its original adverbial role. A man was stranded by the side of the
road with a car that would not run. A passerby stopped and asked what had
happened. The man replied:
(29) Tó:ka', Tho tshö: nhe:yóht ake'tré' atká:ta'
don't know there only so it is I drove it stopped
'I don't know. I was just driving along and it stopped'
No conjunction connects the verbs 'I drove' and 'it stopped';
Comparative evidence confirms that the development of hni9 from a
discourse particle to a syntactic conjunction is fairly recent. Other Iroquoian
languages have cognate particles, but these have not been grammaticized as
conjunctions. They still mean 'also', or 'too', and usually appear in indepen-
dent sentences.
Interestingly, one related language, Seneca, shows a parallel develop-
ment, but with a different particle, kho 'too'. Kho now has the same
distribution as Cayuga hni', following conjoined noun phrases, but other-

wise functioning as a discourse particle meaning 'also'. It shows a much
older relationship to a Siouan particle, Lakhota kho. The Lakhota particle
functions only adverbially with the meaning 'too, also, as well.
(30) Lakhota (Deloria, 1932: 3.13)
Thìpi-iyókheya wã él wóai khó
council-tipi a to carrying food too
škàpi
they were actively engaged
'Another thing else was that women were taking food to the
council tipi'
These two constructions, comitatives and 'also' adverbials, are by far
the most common source of noun phrase conjunctions. They are not the only
sources of coordinating conjunctions, however. Conjunctions may originate
as links between predicates as well.
2.2 Sources of predicate coordination
Many languages combine predicates by simple juxtaposition and ap-

propriate intonation, as in the Parengi sentences in (5) and (6) and the
Cayuga sentence in (24) above. There are also languages with special con-
junctions for coordinating predicates. A good example of the grammaticiza-
tion of predicate coordination comes from Nguna, a Melanesian language
of the Central New Hebrides. In Nguna, verb phrases that describe sub-
parts of a single action or event are conjoined with a special particle poo.
(A different particle conjoins nominals and clauses.)
(31) Nguna (Schütz, 1969: 3.8)
Λ ga vano poo tape na-peka seara
I NON-PAST go and get yam some
'I'll go and get some yams'
(32) Nguna (Schütz, 1969: 159.185)
...go e pae asa porau, poo paluse
and he begin sail and row
'... and he began to sail and row'
Poo is fully grammaticized as a coordinating conjunction. It is
344 MARIANNE MITHUN
obligatory between conjoined predicates, and adds little information beyond

identifying the structural link. It appears as an adverbial in other contexts,
however. In those situations, it adds a completive feature, or the meaning
'already'.
(33) Nguna (Schütz, 1969a: 27)
a poo punusi a
I COMPLETIVE see him
'I've already seen him'
Since conjoined predicates most often represent sequential actions, it is not
surprising that the completive particle should be exploited as a coordinating
conjunction.
2.3 Sources of clause coordination
Many languages indicate clause coordination by juxtaposition and in-

tonation alone, as was seen in the Kamchadal and Gurung sentences in (8)
- (12). Such languages do occasionally contain words that initially resemble
coordinating conjunctions. A closer look at the use of these words shows
that they are not in fact grammaticized conjunctions after all, but rather
sentence adverbials.
Tiwi, a non-Pama-Nyungan language of Australia, is a language of this
type. In his Tiwi grammar, Osborne remarks, "Coordination of clauses
within a complex sentence has to be effected in Tiwi by intonation alone, as
there is no form equivalent to English 'and'. The end of the sentence as a
whole is marked by a fall of intonation, while the non-final character of the
included clauses (other than the last) is signalled by a final rise of intona-
tion" (Osborne, 1974: 70). Osborne's textual material amply illustrates the
intonation patterns described earlier: conceptually unitary events are
presented with no break in intonation; conceptually more distinct actions are
separated intonationally.
There.are two Tiwi particles which might at first appear to be coor-
dinating conjunctions. One is ki 'then'. It can appear in intonationally con-
joined clauses, as below.
(34) Tiwi (Osborne, 1974: 70)
purewari, purewari, purewari, purewari
'They fought and fought and fought and fought
ki — juwuŋepuŋipa.
and then, he hit him in the eye'
The particle ki is more often used with intonationally independent clauses,
however. It specifies their temporal relationship to previous discourse, but
no particular grammatical relationship.
(35) Tiwi (Osborne, 1974: 90)
Looki ji-i-m-ani — mapetani
look he-LINK-do-REP dirty-M
Ki, ka i malala-ni Ki ji-ne-ri-mari
then here clean-M then he-LOC-LINK-get
'He kept looking — dirty water. Then here was some clear
water. He brought it back'
A second Tiwi particle that could be construed as a conjunction is apa.
(36) Tiwi (Osborne, 1974: 97)
tu-wente-rumuta apa, tuli
he-DUR-aim and spear he-stab
'He took aim, and he speared her'
Αρα is used in many other contexts, however, which could never be inter
preted as coordinate structures.
(37) Tiwi (Osborne, 1974: 96)
pu-ne-ri-kuwuntir apa kwanipiri
they-LOC-LINK-race Kwanipiri
They had a race to Kwanipiri'
Apa functions as a pause filler, a signal that the sentence is not yet over,
rather than as a formal marker of syntactic coordination.
The lack of a clear distinction between adverbials and clause conjunc
tions is not unusual among languages. Bogoras, in describing the
Luoravetlan languages, including Kamchadal, explicitly abandoned all hope
of distinguishing them. He remarked, 'On the following pages I give a list
of adverbs and conjunctions without attempting to differentiate between the
two groups. The meaning of many of the adverbial or connective particles
is so uncertain that a division seems hardly possible" (1922: 849). This situa
tion is typical of languages in which clauses and constituents are normally
combined by simple juxtaposition. Particles and clitics sometimes appear in
346 MARIANNE MITHUN
such constructions with meanings like 'also', 'then', 'and so', 'and now',
etc. These particles usually appear more frequently with separate sentences,
however. Their primary function is to provide a semantic or pragmatic link
to previous discourse, not to specify a syntactic one.
The fluidity of the boundary between discourse adverbials and syntactic
conjunctions is significant. The adverbial particles appear to be the source
of most clausal coordinating conjunctions. Consider the history of conjunc-
tion in modern Mohawk, a Northern Iroquoian language of Quebec, On-
tario, and New York State. Mohawk has a coordinating conjunction tanũ'
that conjoins clauses, not unlike English 'and'. It is not cognate to coor-
dinating conjunctions in any of the other Iroquoian languages, even closely
related ones, so it must have developed relatively recently. In fact, its source
is identifiable through historical documents.
In other Iroquoian languages, a particle ta is used to tie new informa-
tion to preceding discourse. It tends to appear at the beginning of paragraph-
like units, and may be translated 'and so', 'so then', 'so now', or 'now then'.
(38) Seneca (Mithun and Peterson, 1980: 111.15)

Tá: o:nẽ nä: kyõ'õ sẽ nõ'o:ta:' o:nẽ kyõ?õ
so then really QUOT three days then QUOT
wa:ayõ '
he arrived
'Now then, after about three days, it seems, a man appeared.'
(39) Oneida (Lounsbury ms: 2)
Tá: né: kati õnv tsha?káhewe' onv wahanéklate' laksá:.
and now when the time came when he was born boy
'And so when the time came for the baby to be born, ...
Early Mohawk grammars and texts recorded during the eighteenth and
nineteenth centuries indicate that ta once functioned in Mohawk as in the
other languages. The passage below is from a cosmology legend recorded at
the turn of the century.
(40) Mohawk (Hewitt: 1903: 282.5)
Ta', ne' ka'ti' wa'hi' ne' dji' neh iakotehiá:ron' -
so, the so then verily the where now she grew up
neñ ' eia 'tase' i 'ken, neñ ' wa 'hi'
now she is maid it is, now verily
'So then, of course, when she grew up and was a maiden,

then, of course, ...'
Ta is no longer used by most modern Mohawk speakers, although it is
familiar to many.
At the time the cosmology legend was recorded, the particle tanũ' had
not yet been grammaticized in Mohawk as a general conjunction. Other par-
ticles link sentences, clauses, and constituents in that text, but they are op-
tional. These also appear in independent sentences, which indicates that their
primary function was to supply a pragmatic, discourse level link.
A compound particle tahnũ, the ancestor of modern tanũ' does appear
in the cosmology text. (Tanũ' represents the combination of two particles,
the ta mentioned above, and some other particle, possibly related to nũ:wa
'now'. The two components have now fused for most speakers.) In the
cosmology text, tahnü' was still an adverbial, usually translated 'besides',
'moreover', or 'furthermore'. Like the other particles mentioned above, it
appears most often at the beginning of a new sentence or paragraph, an-
nouncing a separate, but somehow related, statement.
(41) Mohawk (Hewitt 1903: 258.15)
Enwa 'htka'we' ne ' dji ' katoñrie' 'se', tahnon"
it will cease the where I breathe besides
enkawis 'tohte' ne' kieron 'ke, neh ' tahnon"...
it will make it cold the my flesh on now besides
'My breathing will cease; besides that, my flesh will become
cold, and then, also, (the joints of my bones will become
stiff)'
Just within this century, tanũ has taken over the function of a gram-
maticized general conjunction, similar to English' and or French 'et' It ap-
pears systematically between conjuncts, and specifies syntactic coordination,
as in the remark below.
(41) Mohawk (Muriel Rice, p.c.)
A:ke ne tsi náhe' watyakv:?v, tanũ' katũhkárya'ks
oh the so long I was out and I am hungry
'I was out a long time, and I'm hungry'
The Kwa, Cayuga, Nguna, and Mohawk examples show that coor-
dinating conjunctions can enter languages at a variety of points. Some first
348 MARIANNE MITHUN
develop into noun phase conjunctions from comitative constructions or par-

ticles meaning 'also, too, as well'. Some develop into predicate constructions
from aspectual particles. Some develop into clause conjunctions from
discourse adverbial particles with meanings like 'and then', or 'and so'. Yet
in many if not most languages, particular coordinating conjunctions are not
limited to single type of constituent.
In Nguna, for example, the particle poo conjoins only predicates, but
another particle, go, can function adverbially to link new sentences to
previous discourse, it can conjoin full clauses, and it can conjoin noun
phrases. In the passage below, it functions as a discourse adverbial.
(42) Nguna (Schütz, 1969:122.11-15)
Go kanao pota sikai te too umai punusi pila-na
and man another one he PROG come see mother-his
poogi. Go mail-poogi te pano. Go pila-na maa te
night and morning he go and mother-his also she
pa-ki roara ateata raki na-vinaga. Go waina e liliu,
go-to garden their for his-food and when she return
'But at night, another man came to see the mother and left in
the morning. The mother, too, she left to go to the garden for
food. When she returned, ...'
In (44), it conjoins clauses.
(44) Nguna (Schütz, 1969: 240.39)
Eu munu na-maloku, go eu sale poogi
they drink kava and they dance night.
'They drank kava, and they danced at night'
It is obligatory, and adds no information beyond that of syntactic coordina-
tion. The conjoined clauses may represent sequential events, as in (44), or
simultaneous or generic ones, as in (45).
(45) Nguna (Schütz, 1969:271.39)
Na-rei ki na-vei-va-wota-ana maaga eu too ganikani asa,
people of chief PL they PROG eat in it
go eu too mari na-saisai-ana
and they PROG make meeting
'The chief's people eat in it and have meetings ...'
Go also appears obligatorily between conjoined noun phrases.

(46) Nguna (Schütz, 1969: 238.117)
Tama ni kanao e ga veea paa-kotovi
father of boy he INTR first pay
T h e father of the boy would first pay'
sua e pa-ki tama-na go pila-na
COMPLETION it to father-her and mother-her
'He [would] go to the [girl's] father and her mother'
This apparent multiplicity of functions could arise in two ways. Con-
junctions that coordinate a variety of types of constituents might either
develop all at once as general markers of coordination, or they might first
be grammaticized in one context, then spread through the language gradual-
ly to additional contexts.
3 THE GENERALIZATION OF COORDINATION
Instances of the spread or generalization of coordinating conjunctions

have in fact been documented. Recall that in Sarcee, a comitative postposi-
tion, ih 'with/and', is currently being grammaticized as a noun phrase con-
junction. Clauses are usually conjoined by simple juxtaposition. Cook
reports that 'conjoining is not a major syntactic device' (1984: 98). He notes,
however, ih is now sometimes being used to conjoin clauses as well. In this
context, it is affixed to mi-, a third person singular specified marker referr-
ing to the preceding clause, and yielding the combination mih(ila).
(47) Sarcee (Cook, 1984: 97)
tcúwa mih nánìstlús-í nínádìstlò
wait it with I-sew-NOML0 I go get
'Wait and I'll go get what I was sewing'
Cook also notes that Chipewyan has a particle t'a "which is virtually iden-
tical to Sarcee ih in its form class and function. Although they are not
cognates, their historical developments appear to have been parallel" (1984:
99).
As discussed earlier, the conjunctions coordinating noun phrases in
many Bantu languages have developed relatively recently from comitative
350 M A R I A N N E MITHUN
verbs or particles. In most of the languages, they have remained restricted

to this context. Weimer remarks:
in the vast majority of African languages, there seems to be a single simple word
for 'and', frequently a monosyllable ... The beginner is almost sure to assume that
he can use the same word to join verbs or sentences, to express combinations like
'they were eating and drinking'. That simply is not true in any African language to
which I have had sufficient exposure to find out (1973: 305).
Clauses are conjoined by other morphological devices.

There is nothing like a conjunction joining verbs or sentences. Nouns may be con-
joined, often in ways similar to English, but it is rare to find a language in which
verbs are conjoined; the constructions themselves indicate a following or
simultaneous action (1973: 365).
He points out, however, that

Akan is one of the very few languages in which a kind of verbal and clausal coor-
dinating conjunction is found. Nouns may be conjoined by / n a / or / n n a / 'and'.
The same conjunction is also used between verbs and clauses, sometimes
translatable as 'and' but frequently requiring the translation 'but' (1973: 372-3).
Coordinating conjunctions can thus originate as noun phrase links from

comitative constructions, then spread to predicates and clauses.
Not all general coordinating conjunctions spread form noun phrases to
clauses. They can also spread in the opposite direction. Recall the origin of
the Mohawk conjunction tanũ', from a discourse particle ta ('and so'),
whose function was to link paragraph-like segments. Texts from the beginn-
ing of this century document its development into a discourse particle tahnu'
('moreover, and'), whose function was to link sentences to previous
discourse. At that point, it was not used to conjoin noun phrases. In modern
Mohawk, however, tanũ' has become fully grammaticized as a regular
marker of syntactic noun phrase coordination as well. The sentence below
was the answer to the question 'What are you cooking?'.
(48) Mohawk (Muriel Rice p.c.)
0'wà:rũ tanũ' osahè:ta wakekhü:ni
meat and beans I food make
'I'm cooking meat and beans.'
The generalization of this Mohawk conjunction from clauses to noun
phrases yields an interesting result. It might have been assumed that the posi-
tion of conjunctions within coordinate constituents should point to

typologically interesting differences among languages. Languages do differ
in their placement of conjunctions. Compare English and, which appears
between coordinate constituents, with Latin -que, which follows the last con-
stituent. Differences in position can stem instead from the diachronic origins
of the markers themselves. Mohawk and Cayuga, both Northern Iroquoian
languages, are typologically similar in nearly every way. They have essential-
ly equivalent morphological structures and exhibit equivalent syntactic and
discourse patterns. Yet the Mohawk conjunction tanũ' 'and' appears bet-
ween coordinate nouns, as in (48), while the Cayuga conjunction hni''and'
follows them, as in (26), (27), and (28). The reason for this difference is
clear. The Mohawk conjunction developed from a discourse adverbial 'and
so', which preceded the new statement it linked to previous discourse. The
Cayuga conjunction developed from a particle 'as well', which followed the
additional information added to a list.
4 THE SUDDEN EMERGENCE OF GRAMMATICIZED CONJUNC-

TIONS
Coordinating conjunctions arise from a variety of sources, at a variety

of points in the grammar, and spread in a variety of directions. There does
not appear to be a universal path of development along which they necessari-
ly evolve.
A surprising number of coordinating conjunctions do share one
characteristic, however: their youth. Both internal and comparative evidence
indicate that conjunctions have been grammaticized quite recently from all
of the sources mentioned above, in languages all over the world. The evolu-
tion of noun phrase conjunction from comitative constructions in Sarcee
and West African languages are both still in progress. The grammaticization
of noun phrase conjunction in Cayuga from an additive particle is relatively
new. The development of the general coordinating conjunction in Mohawk
can be dated through documents to the early part of this century, a time
when Mohawk speakers were becoming widely bilingual, first in French,
then in English.
Several facts suggest that this bilingualism may be a significant factor.
An astonishing number of coordinating conjunctions have been recently
borrowed into languages that previously had none. Bogoras (1922: 881)
352 M A R I A N N E MITHUN
noted the presence in Kamchadal of local Russian conjunctions i, dai 'and',

je 'but', potom 'after that', etc. Osborne's Tiwi texts, like many other texts
from Australian Aboriginal languages, show a surprising English loanword:
'and'. The vast majority of coordinating conjunctions in languages
throughout Mexico are borrowed from Spanish. Suárez noted that "in most
of these [Mixe-Zoque] languages coordinating particles have been borrowed
from Spanish, but in spite of that, coordination through mere juxtaposition
(with different meanings according to context) is still very common" (109).
In Tequistlatec-Jicaque languages, "constituents of the clause and clauses
may be linked by coordinating particles; in Coastal Chontal some of these
particles are native, but in Highland Chontal all particles with this function
are borrowings from Spanish" (115). "Coordination is made largely
through juxtaposition in Huixtan Tzotzil. In Tojolabal, the same mechanism
is found, althought there are coordinating particles borrowed from
Spanish" (120). In Huave, "in most cases coordination is marked with par-
ticles borrowed from Spanish, and the constructions with a reduced second
clause match the Spanish patterns so closely that these have probably been
imitated too" (132). Grammars of other Meso-American languages docu-
ment the same situation. South American Indian languages in contact with
Spanish, such as Guaraní and Quechua, exhibit the phenomenon as well.
(See, for example, Cole 1982: 78-80.)
It may not be a coincidence that the source languages for these conjunc-
tions have literary traditions. Literacy itself may contribute to the develop-
ment of grammaticized conjunctions. Loogman notes, for example:
Careful Swahili writers do not commonly interpolate na as a conjunction connec-
ting clauses; however there is an increasing tendency to use na in this way, especially
in too literal translations from English texts or in essays written under the direction
of English speakers (1965:282).
Weimers recounts an interesting anecdote about Vai, a Northern Mande

language of Liberia and Sierra Leone. Clauses are usually joined in Vai by
simple juxtaposition alone, whether coordinate or subordinate. There is,
however, a particle âmù 'and then, and so' that can occur optionally be-
tween clauses. Weimers writes,
The appropriateness of simple parataxis, and the optionality of the conjunctive
/ á m ù / , are vividly demonstrated by a conversation (in English) with Fr. Kandakai
after we had heard a little of a Vai news broadcast on a car radio. He mentioned
that the particular announcer we had been listening to was well known for his
sprinkling /ámu, ámu, ámu/ liberally through his news reports, when in most cases
it was unnecessary; the effect was clearly considered rather amusing. (1976: 130)
Radio news reporting is especially likely to influenced by a literary European

style. The Vai are also well known for their own indigenous writing system.
(See, for example, Scribner and Cole, 1981)
Cherokee, an Iroquoian language spoken in North Carolina and
Oklahoma, now has several devices for indicating coordination. One, ale has
clear cognates in the other Iroquoian languages meaning 'again'. In
Cherokee, it can now be used to coordinate clauses as well as nominals or
predicates, and is easily elicited by request for translations from English. It
is rare, however, in spontaneous spoken language. Pulte and Feeling (1975:
343) note that "it is used in relatively formal speech styles in Cherokee and
in written style." Speakers comment that it sounds Biblical. (Scancarelli
p.c.) Recall that the Cherokee are not only generally bilingual in English, but
they have also had a well known tradition of general literacy in their own
syllabary.
It is intriguing that so many languages could exist for so long without
grammaticized coordinating conjunctions, then suddenly need them enough
to grammaticize them spontaneously from any available source, internal or
external. This suggests that grammaticized coordination might afford some
power or quality of expression impossible with intonation and discourse par-
ticles alone, but that this extra power may not always be felt to be cognitively
essential.
The linking of noun phrases by intonation alone can indeed indicate a
conceptual link, but the precise nature of this link can be quite vague. Series
of noun phrases may indicate sets as in the sentence below from Kupia, an
Indo-European language of India, recorded by Christmas and Christmas
(1973).
(49) abbo-si ayya-si uT-l-a
father-poss 3 mother-poss 3 waken-PAST-they
'his father and mother woke up'
Simple juxtaposition can also link alternatives, however.
(50) Do9gr-e gec-a ker-a, wagg-o, welugu-lu, kicco
mountain-to go-pp do-pp tiger-PL bear-PL what
jantu Di:s-ile kuDa Do9gr-e tinto
wild animal appear-when even mountain-on from
354 MARIANNE MITHUN
peTT-a der-a je-yula

hit-pp bear-pp come-they will
'Having gone up the mountain, they'll shoot tigers or bears, or
whatever wild animal happens to appear, and come carrying it
down from the mountain' (10.12)
It may also indicate apposition, quite a different type of link.
(51) ja-le taruwate jewe-ka ga:Nw-i-ca te:r
become-if then her-to village-in-ones female
bo:da-lu, well-ela mantiri na:nsu-l, gec-ula
child-PL big-PL old woman person-PL go-they will ...
'Then the women of the village, the important senior women,
will go to her ...' (28.15)
Most languages possess devices to disambiguate these relationships when
necessary, such as comitative constructions, or particles like 'as well' or 'in
addition'. The formal grammaticization of conjunction, however, provides
systematic overt disambiguation.
The linking of predicates and clauses by intonation alone can also mir-
ror a conceptual link, but again, the precise nature of this conceptual link
can vary considerably. The most common type of relationship between in-
tonationally linked predicates or clauses in narrative is one of sequence, 'and
then'. The examples below come from the Kathlamet texts recorded by Boas
(1901).
(52) Ilgë 'gElga, iLgēxē 'ma
he took him he threw him away
'Then the man took him and threw him away' (105.13)
Also very frequent, and not always distinguishable from sequence, is conse
quence ('and so'), or cause.
(53) noxoā 'xax tiā 'xatakoax, qatciō 'xax iä 'lXam,
nothing became his reason he did it his town
break break break break

'Then he lost his senses and destroyed the whole town' (33.4)
(54) Niktcā 'xamx qatcuxāwalōtō '-itx tiä 'cuXtikc
he cried all the time he made them unhappy his relatives
'He cried, because he made his relatives unhappy' (32.9)
Juxtaposed clauses may describe simultaneous events or states.

(55) 'lak aë'Xt 'nam,
dried salmon one dish
'LkLiX aē 'Xt 'nam
dried and broken salmon bones one dish
'Dry broken salmon was in one dish, and pounded salmon
bones were in the other' (34.5)
They may, instead, indicate a contrast.
(56) Kë'nuwa nëct igaxElā ' lEmtck
try she was pulled not she moved
'They tried to pull her, but she did not move' (223.17)
They may specify a purpose.
(57) A'yaq aqa qaLxE'laqL, aLagā'tpqa
quick then open the door they may come in
'Now open the door that they may enter' (37.10)
Frequently, juxtaposed clauses represent successive elaborations. A speaker
introduces an idea, then, in successive clauses, adds more information.
(58) Igō 'ya, igagE 'loya cana
she went, she went to gather fern roots (225.15)
Often juxtaposed clauses are not actually coordinate pragmatically; some
would be rendered in other languages by subordinate clauses. These may be
adverbial, as below.
(59) Aqō 'watiX ictö 'qoya, igë 'kim ē 'yalXt ...
several times they slept he said his elder brother
'After a few days, his elder brother spoke' (105.4)
They may represent complements.
(60) O:, tgā ' aquiuqoā 'nimx ikEnuwakcō 'max
Oh it is forbidden he is laughed at the Thunderbird
Oh, it is forbidden to laugh at the Thunderbird' (232.17)
(61) Anxku Lē 'tcguama, tiā 'maq atktëlö 'xoa tgā 'xēqLax
I will tell them shoot they will do them the hunters
'I will tell the hunters to shoot them' (233.12)
356 MARIANNE MITHUN
Prior to the grammaticization of clause conjunction, the relationships

between juxtaposed clauses are usually interpreted from context. The gram-
maticization results in a systematic specification of the precise nature of link.
It has another effect as well. The link is no longer merely semantic or
pragmatic: it is also syntactic. While discourse adverbials relate the content
of material within their scope to previous information, grammaticized con-
junctions specify syntactic structure and overtly signal sentence boundaries.
The cooccurrence of grammaticized conjunction and contact with
literary languages is probably no accident. As is by now well known, written
language differs from spoken language in a number of pertinent ways. Chafe
(p.c.) has discovered that within English there are substantial differences
between the use of conjunctions in spontaneous spoken discourse and in
planned written texts. Speakers use more sentence-initial coordinators, link-
ing new sentences to previous discourse, than writers (32 versus 0.9 per 1000
words). This is reminiscent of the use of the pre-grammaticized discourse
adverbials described above, whose function is to link information
pragmatically to previous ideas, rather than to link structures syntactically.
Within clauses, speakers conjoin significantly fewer constituents than
writers (9.9 versus 23.8 per 1000 words). Speakers, under constraints of
memory and production time, typically produce syntactically simpler con-
structions (Chafe, 1987). They tend to introduce only one important piece
of information per intonation unit. Writers, by contrast, have the luxury of
time to produce long, elaborate sentences, adding new information to earlier
statements or rephrasing at will. They may pause as they write to assemble
new thoughts without being forced to end their sentences each time. Accor-
dingly, written sentences are typically longer and packed more tightly with
information (Chafe 1985).
Speakers do have an important resource at their disposal that writers
lack. Intonation, with fine ranges of rhythm and pitch, and varying degrees
of pause length and volume, can provide structural cues to hearers that
punctuation in written language can only weakly imitate. The overt marking
of grammatical relationships in written language is functional, guiding
readers through highly complex structures.
As Chafe (1985), Pawley and Syder (1983), and others have demon-
strated, many of the special characteristics of written language represent an
adaptation to the conditions under which it is produced and processed. The
variability in coordinating constructions across languages suggests more.
The sudden grammaticization of coordination in languages all over the
world, at times when many of their speakers have been exposed to Indo-
European languages with literary traditions, indicates that written language
can, in turn, exert an influence on spoken language. Perhaps such influences
are first-felt in'the speech of writers and readers, possibly in marked formal
settings. The power of these influences, furthermore, is astonishing. In many
of these languages, speakers who now use newly created or borrowed con-
junction's are not themselves literate, nor even bilingual in a language with
a literary tradition.
5 CONCLUSION
Although the intonational linking of concepts seems to be universal in

spoken discourse, the grammaticization of coordination is not. In some
languages, coordination is not formally marked at all, while in others, it is
marked systematically and obligatorily. Even the development of formal
systems for marking coordination is not uniform across languages. Coor-
dinating conjunctions may originate in a number of different areas of the
grammar, from a number of different sources, and spread in a variety of
directions.
What seems like such a basic device is not only highly variable in form
from one language to the next, but it is also easily borrowed. Exposure to
language with written traditions, or, even more, exposure to literacy itself,
may provide a stimulus for the overt marking of grammatical structures,
and, eventually, the grammaticization of coordinating conjunctions. This
grammaticization not only results in the systematic marking of distinctions
often previously left vague, but perhaps also in the heightened identity of the
sentence as a fundamental unit of linguistic structure.
ABBREVIATIONS
DI = discovery; for further abbreviations, see list on p. vii.
REFERENCES
Aze, Richard and Trish. 1973. "Parengi Texts". In Patterns in Clause,
Sentence, and Discourse in selected languages of India and Nepal: Texts.
358 MARIANNE MITHUN
Vol III, R. Trail (ed.), 213-362. Kathmandu: University Press.

Boas, Franz. 1901. Kathlamet Texts. Bureau of American Ethnology,
Bulletin 26. Washington: Government Printing Office.
Bogoras, Waldemar. 1922. "Chukchee." Handbook of American Indian
Languages, Part 2. Bureau of American Ethnology Bulletin 40:631-903.
Chafe, Wallace. 1985. "Linguistic differences produced by differences be-
tween speaking and writing.'' In Literacy, Language, and Learning.,
David R. Olson, Nancy Torrance, and Angela Hildyard (eds.). Cam-
bridge: University Press.
. 1987. "Cognitive constraints on information flow". In Coherence and
Grounding in Discourse, Russell Tomiin (ed.), 21-51. Amsterdam: John
Benjamins.
Christmas, R.B. and J.E. Christmas. 1973. "Kupia texts". In Patterns in
Clause, Sentence, and Discourse in selected languages of India and Nepal:
Texts., Vol III., R. Trail (ed.), 1-108. Kathmandu: University Press.
Cole, Peter. 1982. Imbabura Quechua. Lingua Descriptive Studies. Amster-
dam: North-Holland.
Cook, Eung-Do. 1984. A Sarcee Grammar. Vancouver: University of British
Columbia Press.
Craig, Colette Grinevald. 1977. The Structure of Jacaltec. Austin: Universi-
ty of Texas Press.
Deloria, Ella. 1932. Dakota Texts. American Ethnological Society Publica-
tion XIV. New York: G.E. Stechert.
Glover, Warren W. 1974. Sememic and Grammatical Structures in Gurung
(Nepal). Norman: Summer Institute of Linguistics, University of
Oklahoma Press.
Hewitt, J.N.B. 1903. Iroquoian Cosmology, First Part, Mohawk Version.
Bureau of American Ethnology, Twenty-first Annual Report: 255-338.
Lord, Carol. 1973. "Serial verbs in transition." Studies in African
Linguistics. Vol 4 (3):269-96.
Mithun, Marianne and Myrtle Peterson. 1980. "The Stonecoat" In Nor
thern Iroquoian Texts, M. Mithun and H. Woodbury (eds.), IJAL-NATS
Monograph No. 4. 110-122. Chicago: The University of Chicago Press.
Osborne, C.R. 1974. The Tiwi Language. Canberra: Australian Institute of
Aboriginal Studies.
Pawley, Andrew and Frances Hodgetts Syder. 1983. "Natural selection in
syntax: Notes on adaptive variation and change in vernacular and literary
grammar". Journal of Pragmatics 7:551-579.
Pulte, William, and Durbin Feeling. 1975. "Outline of Cherokee gram-

mar." Cherokee-English Dictionary. Tahlequah: Cherokee Nation of
Oklahoma.
Schütz, Albert J. 1969. Nguna Texts. Oceanic Linguistics Special Publica-
tion No. 4. Honolulu: University of Hawaii Press.
. 1969a. Nguna Grammar. Oceanic Linguistics Special Publication
No. 5. Honolulu: University of Hawaii Press.
Scribner, Sylvia and Michael Cole. 1981. The Psychology of Literacy. Cam-
bridge: Harvard University Press.
Suárez, Jorge A. 1983. The Mesoamerican Indian Languages. Cambridge
Language Surveys. Cambridge: Cambridge University Press.
Weimers, William E. 1973. African Language Structures. Berkeley: Univer-
sity of California Press.
— . 1976. A Grammar of Vai. University of California Publications in
Linguistics 84. Berkeley: University of California Press.
Worth, Dean Stoddard. 1961. Kamchadal Texts Collected by W. Jochelson.
The Hague: Mouton.
The discourse function of clause-chaining
John Myhill J u n k o Hibiya
University of Michigan University of Pennsylvania
Sophia University
INTRODUCTION
Linguists looking at discourse phenomena from a typological point of

view have recognized for some time that clause-chaining constructions are
crucial to our understanding of the discourse motivation for clause-
combining and subordination*. Largely on the basis of data from Western
Indo-European languages, it has been traditionally assumed that subor-
dinate and non-finite clauses must be associated with backgrounding rather
than foregrounding; Givón (1987:12) suggests that the existence of clause-
chaining discourse calls this assumption into question:
We have taken for granted for too long that there must be a strong correlation
between main-finite clause syntax and the foregrounding function in discourse. In
the main, our older faith in this correlation hinged primarily on Indo-European
facts of grammar, and relatively little on cross-linguistic studies of the matching of
grammar with discourse... In the area of grammatical studies, we have the major
phenomenon of clause-chaining discourse, which seems to go against the grain of
our earlier Euro-centric assumptions.
In spite of the importance of clause-chaining as a discourse phenomena,

description of its function thus far has been anecdotal rather than empirical
in nature and so comments on the theoretical implications of this function
have necessarily been speculative. In this paper, we investigate and discuss
in detail the discourse function of clause-chaining constructions in two
unrelated non-Indo-European languages, Japanese and Soddo, using quan-
titative analysis of language production data. We determine that these con-
structions are in fact not associated with foregrounding function in the
clause in which they appear, contrary to what Givón suggests. In Section 1,
we outline the methodology which we use to be able to describe objectively
362 JOHN MYHILL AND JUNKO HIBIYA
the discourse function of a particular construction. In Section 2, we analyze

the discourse function of a clause-chaining construction in Soddo, and in
Section 3 we do the same with three such constructions in Japanese.
1 METHODOLOGY
All too often, studies of discourse analysis have taken as their objectives
a single, neat, notional definition which accounts for a wide variety of
phenomena. We need not recount here all the definitions of 'topic', 'new in-
formation', 'contrast', etc., which have been advanced, only to be shown to
be circular, vague, and/or language-specific.
The foregrounding/backgrounding distinction, however, as outlined in
Hopper (1979) and Hopper and Thompson (1980), is not of this type.
'Foregrounding' consists of a number of different objectively definable fac-
tors, grouped together because they tend to cooccur in discourse but having
independent effect as well. Given this understanding of grounding, it is clear
that clauses may be more or less foregrounded, by virtue of having more or
fewer foregrounding features. There is some danger, however, that groun-
ding may be seen to be independent of these features, so that they are given
some sort of circular notional definition designed to account completely for
the distribution of a particular form in a particular language, as has been
done with other concepts used in discourse analysis. This could happen if
some linguist determined that a given form in a given language is in and of
itself used for foregrounding, so that foregrounding would be defined simp-
ly to account for the facts of that language; this has happened to such con-
cepts as, e.g. 'topic' (equated with the Japenese particle -wa) and 'old infor-
mation' (equated with English pronouns), and has effectively put an end to
the use of these concepts as tools of scientific research.
The position we take here is that it is simply not possible to come up
with an objective definition which will account for all and only the occur-
rences of a given form. Individual forms are not simply associated with
'foregrounding' or 'backgrounding' but rather with particular features of
foregrounding and backgrounding; a certain form may be associated with
the presence of one particular foregrounding feature and the absence of
another.
We will determine the effect of these feature upon the use of a given
clause-chaining construction by using multivariate statistical analysis. The
THE DISCOURSE FUNCTION OF CLAUSE-CHAINING 363
discussion of the effect of a given foregrounding feature will consist of a

presentation of statistical data demonstrating the effect of that feature and
an illustrative example which neatly fits our discussion of this effect.
However, the reader will not have to look hard to find cases where our
claims as to the effect of one factor are counterexemplified by our data il-
lustrating the effect of another factor1. This is inevitable, given our assump-
tion that there is no one-to-one correspondence between a given form and
any single objectively defineable criterion. Nevertheless, these examples do
indicate general trends.
1.1 Definition of clause-chaining
Clause-chaining will here be defined as the use of non-finite forms not

headed by a conjunction with temporal or circumstantial meaning. Clauses
headed by dependent or independent forms meaning e.g. before, after, if,
when, because, etc., are therefore considered not to constitute clause-
chaining2. Clause-chaining may occur in English, although it generally has
a bookish ring to it, as in (1):
(1) Sitting down, taking out a pencil, he began to write.
Here sitting and taking are non-finite, clause-chaining forms, while began
is not.
1.2 Coding
We coded for the presence of a number of foregrounding features,

namely continuation of same subject, chronological sequencing, perfective
aspect3, and human subject. We then made studies of the correlations be-
tween each of these features and specific clause-chaining forms.
Each individual token in the data base is associated with a particular
predicate. We coded for whether or not that predicate was realized as a cer-
tain form. The predicate coded for in each token will be referred to as the
'present clause'. For each token, we coded the present clause, the immediate-
ly preceding clause, and the immediately following clause for a number of
factors. This is exemplified in (2) and Table 1:
(2) Tom walked into the room. He still had a headache. Sitting
1 2 3
down, taking out a pencil, he began to write. The pencil broke.
4 5 6
Table 1
Form Preceding Following Preceding Present Following

subject subject aspect aspect aspect
1) Not chaining same PFV IMPFV

2) Not chaining same same PFV IMPFV PFV
3) Chaining same same IMPFV PFV PFV
4) Chaining same same PFV PFV PFV
5) Not chaining same different PFV PFV PFV
6) Not chaining different PFV PFV
Sequenced with Following Preceding Present Following

preceding sequenced subject subject subject
1) No Human Human
2) No No Human Human Human
3) No Yes Human Human Human
4) Yes Yes Human Human Human
5) Yes Yes Human Human Non-huma
6) Yes Human Non-human
In each of our examples, we put numbers under each of the tokens coded
for, as in (2). Only 3 and 4 use the chaining form, in this case the present
participle. The same subject is continued until the last clause: thus all the
subject continuation data is 'same' except for 'preceding subject' for token
6 and 'following subject' for sentence 5. All the clauses are perfective except
sentence 2; therefore, 1 has a following imperfective aspect, 2 has a present
imperfective aspect, and 3 has a preceding imperfective aspect, and all the
other aspects are perfective. 2 and 3 are not chronologically sequenced with
the immediately preceding clause, so 1 and 2 are coded as not having a

following sequenced clause, and only 6 has a non-human subject, so 5 is cod-
ed as having a following non-human subject.
The data contained a number of habitual narratives, e.g. When the river
rises, it tears up trees by their roots and carries them off. The second two
clauses would be coded as imperfective (because they are habitual) but se-
quenced.
A number of different clause types were not coded. These were relative
clauses, sentential complements to verbs, infinitives, and nominalizations. If
and because clauses were excluded from the variable rule runs, but they were
counted in terms of coding for preceding or following clause.
2 SODDO
Soddo is a verb-final Semitic language spoken in Southern Ethiopia 4 . In

this language, there is a basic division in verbs between the 'Subordinate'
form and the 'main' clause form. A sentence must end with a main clause
form; the subordinate form cannot stand alone, and it is accompanied by
either the suffix -m or by a conjunction with a meaning such as / / o r when.
The subordinate form with the -m suffix and with no conjunction, which we
will refer to as the 'm-form', is by the definition used here a clause-chaining
form. It is exemplified in (3):
(3) att amora mättam-goy bali akako anat lalä
a crow came -when on-Ali Akako top-of-head on
1
yänäbbäri qəb -abäläm nässam bärräro.
that-was-the butter he-snatched-M he-took-M he-flew-away
2 3 4
'A crow came and snatched the butter that was on Ali
Akako's head and he took it and flew away' (Leslau 1968:49)
We will put an M in the interlinear gloss under the verbs which show
the clause-chaining use of the m-form. The conjunction goy 'when' accom-
panies the subordinate verb in clause 1, while the m-form is used without a
conjunction in clauses 2 and 3, and the sentence ends with a finite verb in
clause 4. Only 2 and 3 constitute occurrences of the m-form.
366 J O H N MYHILL A N D JUNKO HIBIYA
2.1 Factors affecting the use of the m-form
The m-form is not particularly common in the texts we studied, occur

ring in only 81 out of 537 clauses. It is apparent from even a superficial ex
amination of the Soddo texts that the use of the m- form is restricted to nar
rative sections of the texts. By 'narrative sections', we refer to portions of
the text which contain a relatively high concentration of foregrounding pro
perties, such as chronological sequencing and continued subjects. A nar
rative section does, of course, contain some backgrounded (e.g., stative,
habitual, etc.) clauses; however, if a number of stative or habitual clauses
are strung together, this indicates that the narrative has ended and a descrip
tive passage has begun.
Although the m-form occurs only in narrative sections, we will see that
it is not itself associated with following foregrounded clauses; it is also
favored when the present clause involves some sort of switch (in terms of
subject or aspect) from the preceding clause. It is disfavored when the im
mediately preceding clause has a finite verb.
2.1.1 The m-form and following foregrounded clauses
The association between the m-form and following foregrounded

clauses is reflected in the data in several ways. If the following verb has the
same subject as the present verb, the m-form is much more likely to be used,
as shown in Table 2 5 .
Table 2
M-form N %M PROB
Same subject following 75 176 43 .86 ρ<.001

Different subject following 6 361 2 .14
The categories in the column on the left (e.g., 'same subject following') represent each of the
environments coded for. The column under 'M-form' indicates the number of occurrences of
the m-form in each particular environment, e.g., the m-form occurred 75 times when the subject
was the same as the subject of the following clause. The column under 'N' represents the total
number of occurrences of that environment in the data, e.g., there were 176 clauses in the data
which had the same subject as the following clause. The column under '%M' indicates the fre
quency with which the m-form occurred in that particular environment; for 'same subject
T H E DISCOURSE FUNCTION OF CLAUSE-CHAINING 367
following', this was 75 out of 176 times, or 43% of the time. PROB indicates the effect of a
factor on the probability of 'rule application' (here use of the m-form) as estimated by a
multivariate analysis. An occurrence of the m-form is coded as a 'rule application' and an oc-
currence of any other verbal form (subject to the restrictions mentioned above) is coded as a
non-application. This coding is purely for convenience and does not reflect any commitment
to regarding using the m-form as a 'rule'. Factor effects vary between 0 and 1, with PROBs
which are higher than .5 favoring rule application and PROBs lower than .5 disfavoring rule
application. The PROB values measure the effect of a factor; however, they are not based upon
N-size, and hence do not directly indicate the likelihood that the result could have come about
due to chance. This is indicated by the p-value, the number to the right of the PROB (here
'p<.001'), which indicates the likelihood that the result shown could have happened due to
chance; this number refers to the distinction between the factor in the row and the factor in
the row directly below. Here the likelihood is less than 0.1%; cases where the likelihood is
higher than 5% are reported as NS, which means not significant. The PROB values, not the
percentages, are the true measure of the effect of a factor (see discussion in Section 2.3).
If the following clause is in chronological sequence with the present

clause, the m-form is again favored (see Table 3):
Table 3
M-form N %M PROB
Following clause sequenced 48 131 37 .62 p<.001

Following clause not sequenced 33 406 8 .38
If the following clause is perfective, the m-form is again favored (see Table
4). Here, however, the difference is not statistically significant; we return to
discussion of this below.
Table 4
M-form N %M PROB
Following clause perfective 22 56 39 .58 NS

Following clause not perfective 52 446 12 .42
2.7.2 The m-form and discontinuity with the preceding clause
The m-form is also favored by a discontinuity with the preceding clause,

either in terms of the subject or in terms of aspect. A switch of subject from

the preceding clause favors the m-form, as shown in Table 5.
Table 5
M-form N %M PROB
Different subject
from preceding clause 47 349 13 .60 ρ <.025
Same subject
as preceding clause 34 188 18 .40
It is also favored when there is an aspectual shift between the present clause
and the preceding clause, that is, either the preceding clause is perfective and
the'present clause is imperfective or vice versa (see Table 6)6. The difference
heve is not statistically significant, however; we return later to discussion of
this.
Table 6
M-form Ν %M PROB
Aspectual switch 15 48 31 .62 NS

Other 88 489 13 .38
We have seen thus far that when there is a discontinuity between two clauses,
a switch in aspect or subject, the m-form is likely to be used in the second
clause, and that it is also likely to be used in clauses preceding foregrounded
clauses. Thus the ideal time to use the m-form is in the second clause of a
sequence with a meaning such as Bill was being obnoxious, I got mad (M),
I hit him.
2.1.3 The m-form and position in sentence -
If the last verb coded for was finite, the m-form was disfavored; the m-form
was likely when preceded by one, two, or three non-finite clauses. Because
the overwhelming majority of finite verbs are sentence final, this means that
the m-form is disfavored in the first clause of a sentence. This is shown in

Table 7 and exemplified in (4)7.
Table 7
M-form N %M PROB
1-3 non finite verbs before 38 185 21 .59 p<.025

First after finite verb 43 352 12 .41
(4) ənsittatä imdr taywäta yənnässämam iga bäge

the-women sun rising-before getting-up water in-house
bällelä sammahənnäma yətormam yamätämayyan
if-not-being their-water-jars take to-get-water
yalfəman, bäzzibbärma -fwälä bunnahənnäma yeqwälämam
they-go they-returned-after their-coffee they-roast-M
yəwäqtəman...
they-pound
T h e women rise before the sun appears, and if there is no
water in the house, they take their water jars and go to get
water. After they return, they roast their coffee, pound it...'
(Leslau 1968:71)
Bäzzibbärma and yeqwälämam are identical in terms of the other factors
coded for; both continue the preceding subject and have the same subject
following, both are sequenced and followed by sequenced clauses. However,
bäzzibbärma follows a finite verb while yeqwälämam does not, and the m-
form only occurs with the latter.
2.2 The effect of combining factors
A text such as the one we used for Soddo may be roughly divided into
narrative and descriptive sections. The m-form is entirely restricted to nar-
rative sections of the text, but the factors favoring its use can clearly occur
in non-narrative sections; in fact, 'different subject preceding' (see Table 5)
is obviously more likely in non-narrative sections. Nevertheless, the occur-
rence of one of these factors in a non-narrative section does not occasion (or
even favor) the use of the m-form. Rather, it is necessary that a number of
these factors hold simultaneously for the m-form to be used.

We have seen that the use of the m-form is favored, though not
significantly, by a following perfective verb (Table 4) and an aspectual
switch from the preceding verb (Table 6). When both of these factors hold
at the same time, the m-form is favored very strongly, as shown in Tables
8 and 9.
Table 8
M-form N %M PROB
Aspect switch, perfective after 14 24 58 .82 p<.025

No switch, perfective after 8 32 25 .45 NS
No switch, imperfective after 58 457 13 .39 NS
Aspect switch,
imperfective after 1 24 4 .30
Table 9
M-form N %M PROB
Aspect switch, perfective after 14 24 58 .73 p<.001

Other 67 513 13 .27
We see in Table 8 that the m-form is most favored (PROB = .82) when there
is an aspect switch with the preceding clause and when the following verb
is perfective; the difference between this factor and the next highest is signifi-
cant at the .025 level. The difference between the other three factors, having
probabilities of .45, .39, and .30, is not significant; thus all three of these
are combined in Table 9, which shows that the difference between the factor
most favoring the m-form and the other factors is significant at the .001
level.
It is certainly possible for a single isolated perfective clause to occur in
the middle of a descriptive section; such a clause would create two aspect
switches, from imperfective to perfective and then back. However, Table 8
shows that this single perfective clause would have no effept on the use of
the m-form; the clause preceding the perfective would be 'no aspect switch
(imperfective and preceding imperfective), perfective after', with a PROB

value of .45, while both the perfective clause and the clause following the
perfective clause would be 'aspect switch, imperfective after', with a PROB
value of .30. The environment favoring the m-form, 'aspect switch, perfec-
tive after', can only occur if there are two perfective clauses in close proximi-
ty, which would only be likely to happen in a narrative section.
The situation is the same for the other factors favoring the use of the
m-form. A shift of subject is quite common in a descriptive passage (in fact,
it is far more common than a continuation of the same subject); never-
theless, the m-form is not possible in such cases. In fact, Table 10 shows that
a subject switch only favors the m-form if the following subject is the same
and the following verb is sequenced, and this will only happen in a narrative
passage.
Table 10
M-form N %M PROB
Same subject after, seq after

Different subject before 26 38 68 .95 p<.001
Same subject before 20 45 44 .83
Same subject after, unseq after
Different subject before 17 51 33 .83 NS
Diff subject after, seq after
Diff subj after, unseq after
Table 10 shows that a preceding different subject only favors the m-form
when the following subject is the same and the following clause is sequenced;
the difference is .95 to .83, which is significant at the .01 level. In all other
cases, there is no effect of same or different preceding subject (.83 vs . .72,
.23 vs. .23, .10 vs .08).
When the non-significant differences in Table 10 are collapsed, the
results are as shown in Table 11 :
Table 11
M-form N %M PROB
Sequenced, same subject after

different subject before 26 38 68 .89 ρ<.001
Same subject after, other 49 138 36 .65 p<.001
Different subject after 6 361 2 .06
We see here that the m-form is most strongly favored when the following
clause is sequenced and has the same subject and when there is a switch of
subject from the preceding clause. The m-form is less favoured when the
following subject is the same but one or both of the other two factors is not
present (i.e., the following clause is not sequenced and/or there is no switch
of subject from the preceding clause) and the m-form is strongly disfavored
when the following subject is different, occurring only 6 out of 261 times.
If the following subject is different, the m-form is disfavored so strongly
that no other factor is important. The difference between each of these fac
tors is significant at the .001 level.
(5) illustrates how these factors affect the use of the m-form.
(5) ähäk bäduläti ge yänäbbäri golä, yägäbsənna
now of-dulat house that-was parched-grain of-barley-and
yäməššərä näbbär, tätəkk yatilläfä
of-lentils it-was extremely that-surpassed
1
yəčäm näbbär. fäyyä qinnahum bäbälhw-kaččä wägra
was-delicious well I-doing-M I-ate -after satiated
2 3M 4
mänka qärräski. "ay ahumma bägärrawh zi
somewhat I-began well indeed if-I-am-satiated this
5
yəmäsəl qolä täbäla wäbäl
that-resembles parched-grain I-shall-not-eat to-mean
ädäbəll? ähäk mən čoňňä?" bakkum täləbbəddi
would-it-not now what better I-thought-M with-heart
6M
wämməkakär qärräski. yädənq bissäbkw -kaccänna attə kit

to-take-counseI I-began much I-thought-after about two
7 8
bərəlle yägärä yin yəmässI atabät yatmittət
glasses of-hen eye-color similar finger that-makes-suck
yägäbs səkär bäsäččäwh-fwälä attə bəlat agäňňähi. Zim
of-barley beer I-drank -after a trick I-found this
9 10
qoli bäkisəddi w assəgəm duläti yalqə-goy
parched-grain in-pocket to-put-in dulat when -finished
tägeyi näwäläf näbbar, bäzarrat bämäsoborqi
to-house to-go was so-then from-food-basket
11
qolä ätäbtom bäkotoddi yäwəěta kis wassog
parched-grain I-taking-M into-my-coat of-inside pocket to-put
12M
qärräski.
I-started
13
'Now the parched grain (it was barley and lentils) that was in
the house of the dulat (clan assembly) was extremely delicious.
I did well, and after I ate, I began to be somewhat full. "Well,
if I get full now, does it not mean that I will not keep eating
this good parched grain? What must I do?" I thought, and I
began to take counsel with my heart. After thinking much and
drinking one or two glasses of delicious beer that had the color
of a hen's eyes and that made one suck the fingers, I found a
plan. That was to put the parched grain in my pocket to take
home after the dulat was over. So I started to take grain from
the food basket and put it in the inside pockets of my coat'
(Leslau, 1968:182-3)
The first two clauses are background orientation clauses, and therefore
do not use the m-form, although they do have the same subjects. Then the
clauses begin to have more foregrounding features; verbs 3 through 5 all
have the same subject and are perfective. The ra-form is only used on the
first of these {qinnahum), which involves a switch both in subject and in
aspect from the preceding clause (from imperfective to perfective). Next
there is an interruption, as the speaker interjects what he was thinking, and

then the narrative starts up again. Clauses 6 through 10 again have the same
subject and are in chronological sequence, and again only the first clause
uses the ra-form (bakkum). There is then another interruption in the nar-
rative sequence, as the speaker describes his plan in clause 11. Then the nar-
rative resumes again, with 12 and 13 having the same subject and being in
sequence. The ra-form is used in 12, which has a shift of both subject and
aspect from the preceding clause.
2.3 The apparent correlation between the m-form and foregrounding
Givón (1985:12) calls into question the assumption that there is a

necessary correlation between subordination and backgrounding, and he
points to clause-chaining as a particular case where non-finite forms are used
with foregrounding function. Examples of the use of the ra-form such as (3)
seem to support this argument; both uses of the m-form here are perfective
and have the same subject as the preceding clause. This argument appears
to be further supported by the quantitative data in Table 12.
Table 12
M-form N %M PROB
Sequenced with preceding verb 33 133 25 .54 NS

Not sequenced with
preceding verb 48 404 12 .46
Perfective 21 61 34 .56 NS
Not perfective 60 476 13 .44
Different subject
from preceding clause 47 349 13 .60 p<.025
Same subject as
preceding clause 34 188 18 .40
The m-form is in each case more common in foregrounding environments;

it is more common in sequenced clauses than in unsequenced clauses (25 %
to 12%), in perfective clauses than in imperfective clauses (34% to 13%),

and when the subject is continued from the preceding clause compared to
when it is switched from the preceding clause (18% to 13%).
Nonetheless, the multivariate analysis showed that these percentages are
misleading; sequencing and perfectiveness of the present clause do not have
a significant effect on the use of the m-form, while having the same subject
as the preceding clause actually disfavors the m-form (see Tables 5 and 11).
In spite of examples like (3) and the percentages reported in Table 12, the
m-form is not favored by foregrounding features in the present clause.
Why should the percentages be misleading here? The reason is that the
m-form clearly is favored by foregrounding features in the following verb;
this is confirmed by the multivariate analysis reported in Tables 2, 3, and
8-11. It is simply a fact of the organization of discourse that foregrounding
features come in clusters; the more foregrounding features a verb has, the
more foregrounding features neighboring verbs have. Table 13 shows typical
data (taken from the Soddo data base) illustrating this.
Table 13
Preceding subject
Same Different
Following subject
Same 85 82
Different 94 240
%Same 48% 25%
Following verb
Sequenced 58 59
Unsequenced 121 263
%Sequenced 32% 15%
When the subject of a clause is the same as that of the preceding clause, the
subject of the following clause is the same 48% of the time and the following
clause is sequenced 32% of the time. There is thus a fairly good chance that
the following clause will be foregrounded and therefore also a fairly good
chance that the m-form will be used in the present clause; but this use of the
m-form is caused by the foregrounding of the following verb, and only in-
cidentally correlated with the continuation of the preceding subject. On the

other hand, when the present clause has a different subject than the
preceding clause, the chance that the following clause will be foregrounding
is much lower (25% following same subject and 15% following sequenced
verb), so that the chance of using the m-form in the present clause is much
lower.
The data from Soddo show that the m-form is not strongly associated
with either backgrounded or foregrounded clauses; Table 12 shows that se-
quencing, lack of sequencing, completedness, and incompletedness all have
no effect upon the use of the m-form. The m-form is therefore neither as
foregrounded as main clause verbs nor as backgrounded as other subor-
dinate forms. In this connection, it does seem that the m-form is generally
less dispensible to the narrative (and hence less backgrounded) than are verbs
with lexical subordinators (with meanings such as 'after', 'while', 'during',
etc.). For instance, in (5), verb 6 is in the m-form, and this first mention of
thinking is indispensible to the narrative. On the other hand verb 8 mentions
again this action of thinking, is therefore redundant and dispensible, and the
subordinator meaning 'after' is used, while the action of drinking beer
reported by verb 9 is irrelevant to the rest of the narrative and again there
is a lexical subordinator. Unfortunately, it is not possible to objectively code
for the dispensibility of a given predicate, and so we cannot quantitatively
substantiate this claim.
Our most important finding for Soddo, then, is that the m-form is used
in clauses which are followed by foregrounded clauses. Because foregroun-
ding features tend to cluster, so that texts may be divided into narrative and
descriptive sections, clauses preceding foregrounding clauses are likely to
have foregrounding features. That explains the correlation shown in Table
12; foregrounding features in a clause favor the appearance of the m-form
in not that clause, but rather in the preceding clause.
3 JAPANESE
In this section, we will examine the function of three different clause-

chaining forms in Japanese. We will refer to these as the 'stem', the 'gerund'
(characterized by the suffix -té), and the '-to' form 8. Our analysis shows that
the most basic division between these forms is between the stem and gerund
on the one hand and the -to form on the other. The first two are associated
(at the probabilistic level) with continuing the same subject in the following
clause, while the third is associated with switching subjects in the following
clause. Other differences between the forms are considerably more complex;
the use of the stem and the gerund will be discussed in Section 3.1 and the
use of the -to form will be discussed in Section 3.2.
3.1 The stem and the gerund
In Section 3.1.1, we describe the function of these forms in marking

grounding, while in Section 3.1.2 we discuss their positional characteristics9.
3.1.1 Marking grounding
Both the stem and the gerund are associated with continuing the same
subject in the following clause. This is shown in Table 14.
Table 14
Gerund N Gerund % PROB
Same subject following 41 220 19 .73 p<.001

Different subject
following 9 268 3 .27
Stem N Stem% PROB
Same subject following 87 220 40 .68 p<.001

Different subject
following 35 268 13 .32
This pattern is exemplified in (6)10. Hirata (the protagonist) is a swimmer on

a high school swimming team, Tanimura is his coach, and they are taking
a bus to participate in a meet; for later reference, each sentence (defined or-
thographically) is marked with a number before it (e.g. S1 for the first
sentence), while each verb is numbered and categorized directly below the
verb:
(6) (S1) Tanimura wa amidana kara baggu o oroshi,

T. rack from bag take-down
1-STEM
denkikamisori o toridashite ago ni ateta, (quote). (S2)
electric-razor took-out chin put
2-GERUND 3-FIN
Minna ga ore o mite dotto waratta. (S3) Sharyo no hashi no
everybody me look burst-into-laughter car end
4-PFV-GERUND 5-PFV-FIN
kyaku made ga, nanigoto ka to tachiagatte kochira o
passenger even what's-the-matter stand-up here
6-IMPFV-GERUND 7-IMPFV-FIN
mita, (quote) (S4) Ore wa damatte utsumuita. (S5) Hai
looked I keep-quiet looked-down yes
8-IMPFV-GERUND 9-IMPFV-FIN
to henji o sureba surude, kakko tsukerunjanee to tachimachi
answer if-do don't-be-pretentious in-a-moment
gyakushyuushitekuru Tanimura no sokoiji no warusa o ore wa
counterattack T. maliciousness I
omoishirasarete iru. (quote) (S6) Tanimura wa hosoi surudoi
was-aware T. thin sharp
10-PFV-FIN
me o hikarase, iyami tappurini mimawasu. (S7) Minna
eye flash irony full-of look-around everybody
11 -PFV-STEM 12-PFV-FIN
hanajironde utsumuita. (quote) (S8) Tanimura no ore no
look-daunted looked-down T. my
13-IMPFV-GERUND14-PFV-FIN
seikaku o misukashitayoona monoii ni, ore wa kuchibiru o
character read-mind way-of-talking I lips
kanda. (S9) Hitori de oyoidari, renshuu no toki ni wa
bit alone swim-or practice time
15-FIN
shinkiroku o dasunoni, taikai ni nari, uoo to iu kansei
record make-though meet become sound like applause
o kiitadake de agatte shimau ore no warui kuse ni Tanimura
here-only get-nervous my bad habit T.
wa akirehateteita. (quote) (S10) Tanimura wa magao de

was-disgusted T. serious-look
16-IMPFV-FIN
kuchi o togarase, hana ni shiwa o yoseta. (S11) Ore wa tada
mouth purse nose wrinkles made I only
17-PFV-STEM 18-PFV-FIN 19-IMPFV-STEM
utsumuki, kyooshuku shiteita. (quote) (S12) Ore no itteiru
look-down felt-discouraged I attend
20-IMPFV-FIN
jishuukan kookoo wa aichiken de mo yuusuu no
J. school A.-prefecture in even prominent
shingakukoo de suieibu ga yowaku, ore wa itsumo moto
good-school be swimming-team weak-be I always ex-
21 -IMPFV-STEM11 22-IMPFV-STEM
orinpikku senshu de aru Tanimura ga koochi shiteiru
Olympic swimmer be T. coach doing
toyohashikoogyoo kookoo ni oyogi ni itteita.
T. high-school in swim to would-go
23-IMPFV-FIN
Tanimura took the bag down from the rack, took out his
electric razor, and put it to his chin.
"Hirata, what's wrong with you? Your eyes are teary! Are
you horny?"
Everybody looked at me and laughed loudly. Even the
passengers at the end of the car stood up wondering what the
matter was and looked at us.
"How're you feeling? We're relying on you; the only
chance we have to get to the finals is you in the 1,500 meter
freestyle. Do you understand?"
I kept quiet and turned my eyes downward. I knew very
well that Tanimura was so nasty that if I said I would try he
would tell me not to be pretentious.
"There's no use taking guys who can't possibly win just
to fill out the team."
Tanimura flashed his sharp thin eyes and looked around
in a hostile way. Everybody felt insulted and turned their eyes
away.
"Hirata! Don't be too nervous! If you swim like you do

in practice, you'll win."
I bit my lip at Tanimura's insulting manner. Tanimura
was disgusted with me because I could break records when I
swam alone or in practice but I would get nervous in front of
crowds at meets.
"Be confident. Be confident. Everybody's nervous.
Whoever's confident will win."
"Okay."
"Why are you so timid when you're so big?"
Tanimura puckered his lips in a complaining manner and
wrinkled his nose. I just kept looking down and was ashamed.
"By the way, what are you going to do about college?"
"There's still six months left."
"How about going to Nittaidai? My friend wants to have
you there. You can get a sport scholarship."
"Uh-huh."
"If you get serious, it's not too late for the Munich
Olympics"
"Uh-huh."
"Do you have any other ideas?"
"Yeah. I want to go to Waseda."
"Waseda??"
The high school I go to, Jishuukan, is one of the most
competitive schools in Aichi prefecture and its swimming team
is weak and I would always go to swim at Toyohashikoogyoo
High School, where Tanimura, a former Olympic swimmer,
coaches' (Tsuka, 1980:8-10)
We see here that the stem and gerund are associated with continued sub-
jects in the following clause. Essentially, each sentence has only one subject,
and the sentence is continued, using the stem and gerund, until there is a new
subject. Table 15 illustrates this.
Table 15
Sentence Clauses Subject
1 1,2,3 Tanimura
2 4,5 Everybody
3 6,7 The passengers at the end of the car
4 8,9 I
5 10 I
6 11,12 Tanimura
7 13,14 Everybody
8 15 I
9 16 Tanimura
10 17,18 Tanimura
11 19,20 I
12 21,22,23 The school, I
Same subject following, gerund or stem 1,2,4,6,8,11,13,17,19,21
Same subject following, finite 9,16
Different subject, gerund or stem 22
Different subject, finite 3,5,7,10,12,14,15,18,20
There are only two cases where a new sentence is started even though the
same subject is continued. These are between clauses 9 and 10 and between
clauses 16 and 17. Clauses 16 and 17 are separated by a quote; 9 and 10 (ut-
sumuita and iru) are, respectively, perfective and imperfective, and we will
see that this environment disfavors both the gerund and the stem.
Although both the gerund and the stem are favored by the same follow-
ing subject, they are favored for different reasons. The gerund is favored
when the following verb has foregrounding features, and one of these
features is continued subject. The stem, on the other hand, is favored when
the present and following clauses are parallel constructions, either both
foregrounded or both backgrounded, so that there is no great discontinuity
between the two; one way in which two clauses may be parallel is if they have
the same subject. The difference between the two forms is brought out by
other statistics (see Table 16), these relating to aspect.
Table 16
Perfective in
following clause 26 182 14 .61 p<.005
Imperfective in
following clause 24 306 8 .39
Stem N Stem% PROB
Aspect same in
following clause 103 332 31 .62p<.001
Aspect different 19 156 12 .38
In the upper part of Table 16 we see that the gerund is favored when the
following verb is perfective, because completedness is a foregrounding
feature. On the other hand, the lower part of Table 16 shows that the stem
is favored when the present and following verb have the same aspect, either
both perfective or both imperfective12.
When the present and following clauses are both perfective, both the
gerund and the stem are favored, the former because the following verb is
perfective and the latter because the present and following verbs have the
same aspect. In (6) there are six clauses which are perfective and followed
by perfective clauses with the same subject; these are 1, 2, 4, 6, 11, and 17.
The stem is used in 1, 11, and 17, while the gerund form is used in 2, 4, and
6. On the other hand, when the present clause is imperfective and the follow-
ing clause is perfective, only the gerund is favored, because the following
clause is perfective but the two clauses do not have the same aspect. So 8
and 13 are both imperfective and followed by perfective clauses and both use
the gerund, while 19, 21, and 22 are all imperfective and followed by im-
perfective clauses and all use the stem. Finally, when the present verb is
perfective and the following verb is imperfective, neither form is favored, so
in clause 9 the finite form is used although the following subject is the same
The gerund is to some extent similar to the Soddo m-form in its usage;
it is not itself associated with foregrounding but rather with following
foregrounded clauses. It is therefore limited for the most part to narrative

sections of the text; there are however some differences from the m-form,
which we will discuss later.
The stem, on the other hand, is not associated with following
foregrounded clauses but rather with following clauses which are parallel to
and continuous with the present clause. Therefore, it may link foregrounded
clauses to each other or backgrounded clauses to each other. It is not nor-
mally used preceding transitions from backgrounding to foregrounding or
vice versa or any similar kind of discontinuity.
This tendency for the stem to be used to link backgrounded clauses is
demonstrated by another factor affecting its use: it is favored when the pre-
sent and following clauses both have non-human subjects, even when these
subjects are different. Human subjects in the unmarked case are continued
as subjects, and attention in a narrative is normally focused on human par-
ticipants, so that a switch to or from a human subject involves a switch of
attention and a disruption of continuity. On the other hand, non-human
subjects are not centers of attention, and usually do not continue as subjects;
so, switching from one non-human subject to another does not generally in-
volve a discontinuity. This is shown in Table 17 and (7).
Table 17
Stem N Stem% PROB
Subject and following

subject non-human 43 111 39 .65 p<.001
Other 79 377 21 .35
(7) Shibaraku hikui koe to kusukusuwarai ga tsuzuita ga,

for-a-while low voice and giggling were continuing
1-FINITE
sore mo tsukare to yoi de suguni osamari,,
that fatigue and drunkenness from soon dying-down
2-STEM
koichijikan mo suruto, suyasuyato neiki ga kikoeru
about-an-hour passed peaceful sleeping-breath can-be-heard
3-TO
dake ni natta. Ore wa myooni ikigurushikute

only became I oddly feel-like-suffocating
4-FIN 5-GERUND
netsukezu, itsumademo wakanoura no
not-being-able-to-go-to-sleep forever W.
6-GERUND13
namioto ga mimi ni tsuiteshiyooganakatta. Nando mo
sound-of-wave ear keep-on-hearing many-times
7-FIN
negaeri o uttewa sonotabi ni yami no naka de
turning-around made each-time darkness in
8-GERUND
muishiki ni Shigeru o sagashiteiruto (quote) Shigeru ga
unconsciously S. was-looking-for S.
9-TO
koe o koroshite yobikaketekita. (quote) Jibun no beddo kara
voice suppress addressed his bed from
10-GERUND 11-FIN
oritekita Shigeru wa, ore no mimi ni kuchibiru ga
came-down S. my ear to lips
furesoo ni naru gurai kuchi o chikazukete, soo
about-to-touch become almost mouth approach so
12-GERUND
sasayaita. Shigeru no atsui iki ga ore no mimi ni kakaru.
whispered S. 's warm breath my ear on is-failing
13-FIN 14-FIN
Ore wa karadajuu no kinniku ga koochokusuru no o kanji,
I all-over-body muscle become-tense that feel
15-STEM
sugu ni wa henji o suru koto mo dekinakatta. Asa no jikan o
soon answer couldn't-even morning time
16-FIN
kinisuru ore ni, Shigeru wa yoyuu tappuri no tokoro o
worry I S. easy full-of being
misenagara ashibaya ni aruita. Dekita bakari no
show-while quickly walked done just
17 18-FIN
rikujokyoogijoo wa shizumarikaeri, maatarashii konkuriito

athletic-field be-completely-quiet brand-new concrete
19-STEM
no sutando ga, tsukiakari ni aoku hikatte ita. Oretachi wa
stand moonlight in blue shining was we
20-FIN
antsuukaa no gurando ni ori, koshi o oroshita.
all-weather ground to go-down hip put-down
21-STEM 22-FIN
Rengairo no torakku wa kataku tsumetakatta,
brick-color track be-hard was-cold
23-STEM24-FIN
Their low voices and giggling continued for a while, and
died down soon from fatigue and drunkenness, and after
about an hour had passed, only the sound of their sleeping
was heard.
Having trouble breathing for some reason, I kept hearing
the sound of the waves of Wakanoura. Turning around many
times, I was unconsciously groping for Shigeru in the dark
(each time), and then "Hirata, Hirata, are you awake?"
Shigeru called to me in a suppressed voice. "Why don't we go
outside?"
Shigeru, who came down from his bed, brought his mouth
close to my ear so that his lips almost touched my ear, and
whispered this. Shigeru's breath fell upon my ear. I felt my
body become tense, and I couldn't even answer for a moment.
"It's okay not to attend the opening ceremony. None of the
athletes attending the meet tomorrow will show u p . " In con-
trast to my anxiety about the opening ceremony the next day,
Shigeru appeared unconcerned and walked quickly.
The athletic field, which had just been finished, was
deathly silent, and the brand new concrete grandstand shone
blue in the moonlight. We went down to the artificial surface
and sat down. The brick-color track was hard and cold'
(Tsuka, 1980:26-7)
We will for the moment disregard the two tokens of the -to form of the
verb (clauses 3 and 9), which, as we have noted, is associated with a subject
switch in the following clause; this form will be discussed in Section 3.2. The
subject switches in (7) are listed in Table 18.
Table 18
Present clause Following clause Form in present clause
2-voice and laughter 3-hour stem

4-sleeping sound 5-I finite
13-Shigeru 14-breath finite
14-breath 15-I finite
16-I 17-Shigeru finite
18-Shigeru 19-field finite
19-field 20-stands stem
20-stands 21-we finite
22-we 23-track finite
There are two examples of consecutive non-human subjects, 2-3 and 19-20,
and both use the stem. On the other hand, when either of the subjects is
human, there is a finite form and the sentence ends, whether the human is
in the present clause (13-14, 18-19, and 22-23), the following clause (4-5,
14-15, and 20-21) or both (16-17).
Both of the sentences using the stem to link clauses with consecutive dif-
ferent non-human subjects are in backgrounding passages. This is clearly the
case in the sentence with clauses 19 and 20. The sentence with clauses 1-4
clearly has a backgrounded feeling, although there is some temporal sequen-
cing; this backgrounding is doubtless due to the non-humanness of the sub-
jects.
3.7.2 Positional characteristics
The other important difference between the stem and the gerund is that
the stem occurs earlier in both sentences and narrative sequences. Table 19
shows that the stem is favored in the first clause of a sentence.
Table 19
Stem N Stem% PROB
1 93 267 35 .65 p<.001

2,3,4 29 221 13 .35
The gerund is actually more likely to occur later in a sentence, although the
difference here is not statistically significant (see Table 20):
Table 20
1 28 267 10 .42 NS
2 12 155 8 .39 NS
3 8 49 16 .57 NS
4 2 17 12 .62
This is illustrated in (8):
(8) Hara wa tarumi, atama wa hageagari, akajimita karada ni

belly be-flabby head be-bald filthy body
1-STEM 2-STEM
abara ga ukide, kao no kawa ga mukete madara ni
rib be-sticking-out face skin be-peeling mottled
3-STEM 4-GERUND
natte ita.
was
5-FIN
'His belly was flabby, his head was bald, the ribs stuck
out from his filthy body, the skin on his face was peeling, and
(his face) was mottled' (Tsuka 1980:12)
This is a typical backgrounding sequence, with a series of non-human sub-
jects, subject switches, and imperfective verbs, and the stem form is ap-
propriately used in clauses 1, 2, and 3. But in clause 4, the gerund form is

used, even though the following subject is different and the following verb
is imperfective. This is because of the tendency for the gerund to replace the
stem form later in the sentence.
Just as the stem tends to occur as the first verb of a sentence (see Table
19), it also tends to be used in the first clause of a narrative sequence. The
gerund, on the other hand, tends to occur later in both sentences (see, Table
20) and narrative sequences. The gerund is therefore disfavored for perfec-
tive verbs when the preceding verb is imperfective; such a perfective verb is
commonly the first verb of a narrative sequence, and the gerund is not usual-
ly used here. This is shown in Table 21.
Table 21
Preceding imperf.,
present perfective 5 78 6 .36 p<.025
Other 45 410 11 .64
Example 6 demonstrates this pattern; the verbs in 2, 4, and 6 are all perfec-
tive and preceded by perfective verbs, and all use the gerund, while 11 and
17 are perfective but preceded by imperfective verbs and use the stem.
There is also what appears to be a stylistic restriction against using con-
secutive stem forms (see Table 22).
Table 22
Stem N Stem% PROB
Stem form in
following clause 12 116 10 .33 p<.001
Other 110 372 30 .67
No such restriction was found on the use of the gerund form.

Finally, the stem form was less common when the present clause was
unsequenced and the following verb was sequenced, as shown in Table 23.
Table 23
Stem N Stem% PROB
Unsequenced, next
clause sequenced 6 89 7 .38 p < . 0 5
Other 116 399 29 .62
However, a similar pattern emerged with the gerund1
Table 24
Unsequenced, next
clause sequenced 4 89 4 .38 p < . l NS
Other 46 399 12 .62
The effect of this factor on the use of the gerund was not statistically signifi-
cant (chi-square = 3.29, p < . 1); nevertheless the effect on the use of the stem
form was only significant at the .05 level (all the other effects were signifi-
cant at least at the .025 level) and the probabilities for the effects on the two
forms are the same. We may conclude that the effect of this factor is similar
on the stem form and the gerund. The reason for this effect is that there is
a very strong tendency for sentences to end when there is a sequenced clause
followed by an unsequenced clause; this is shown in Table 25:
Table 25
Verb in Verb in
present clause following clause Sentence-final Other S-F%
sequenced sequenced 26 56 32
sequenced unsequenced 36 55 40
unsequenced sequenced 71 18 80
unsequenced unsequenced 108 118 48
390 JOHN MYHILL AND JUNKO HIB1YA
Unsequenced clauses followed by sequenced clauses are sentence-final 71 out

of 89 times (80%), while in all other situations sentence-final verbs occur less
than 50% of the time. So neither the stem nor the gerund normally occurs
in unsequenced clauses followed by sequenced clauses.
To review, neither the gerund nor the stem is itself associated with
foregrounding. The gerund, like the m-form, is associated with following
foregrounded clauses. As with the m-form, the clause containing the gerund
is not itself associated with either foregrounding or backgrounding. The
stem is associated with parallelism between the present and following
clauses, with either both being foregrounded or both being backgrounded;
it is thus favored by following same subject and consecutive perfective
clauses (characteristic of foregrounded passages) but also consecutive non-
human subjects and imperfective clauses (characteristic of backgrounded
passages). So both of these forms, like the m-form, are not as foregrounded
as main clause forms but also are not strongly backgrounded either.
3.2 The -to form
The -to form is generally associated with foregrounding properties in

the present clause and backgrounding properties in the following clause.
Unlike the other clause chaining forms analyzed thus far, the -to form is
associated with different subjects in the following clause. This is shown in
Table 26:
Table 26
-to N %-to PROB
Different subject following 20 254 8 .66p<.01

Same subject following 8 217 4 .34
Another backgrounding feature associated with the clause following the -to
form is that the -to form is favored when the following clause is imperfective
and the present clause is perfective.
Table 27
-to . N %-to PROB
Present clause perfective,

next clause imperfective 9 77 12 .65 p<.025
Other 19 394 5 .35
Thus the -to form is favored when the following clause has backgroun-
ding features. It is also favored when the present clause has foregrounding
features, such as when it is perfective (see Table 27), when it has the same
subject as the preceding clause, and when it is temporally sequenced with the
preceding clause (see Table 28).
Table 28
-to N %-to PROB
Same subject in
preceding clause 18 207 9 .67 p<.005
Different subject 10 263 4 .33
Sequenced with
preceding clause 24 168 14 .81 p<.001
Not sequenced 4 303 1 .19
It appears from this data, then, that the -to form, unlike the gerund,
the stem, or the Soddo m-form, is associated with foregrounding in the pre-
sent clause and backgrounding in the following clause. But when we look at
the individual examples, an entirely different picture emerges; the clause
following the -to form is in each case clearly more indispensible to the nar-
rative (in spite of its backgrounding features) than the clause with the -to
form13. As noted above, we do not know of an objective way to measure 'in-
dispensibility'; we will therefore only be able to demonstrate this point here
with specific examples and not quantitative data. Example 9 illustrates the
use of the -to form:
(9) nijuppunhodo de michi wa migi ni ukaishihajime,

about 20 minutes after route right to start-to-turn
1-STEM
chiisana ike o toorikosuto shio no kaori ga tadayottekita.
small pond pass ocean smell was-coming-in
2-TO 3-FIN
Me no mae ni hiraketa umi wa odayakade nami mo naku,
eye front opening-up sea calm wave even not-be
4-STEM 5-STEM
kagami no yooni shizumarikaetteita.
mirror like calm-down
6-FIN
'After about 20 minutes, the route started to turn to the

right, passed a small pond, and then the smell of the ocean
was coming in. The ocean opening up in front of our eyes was
calm, without a single wave, and was like a mirror'
This passage occurs during a bus trip to an ocean town. If we compare
clauses 2 and 3, it is clear that 2 has more foregrounding features. 2 is perfec-
tive, while 3 is imperfective. 2 has the same subject as the preceding clause,
while 3 does not. Nevertheless, it is clear that 3 is much more important to
the narrative than 2; it is of no importance to the story that a small pond
was passed, but the introduction of the ocean is a significant event.
Therefore -to is used in clause 2 to indicate that the following clause, though
lacking foregrounding features, is in fact foregrounded.
10 and l1 give other examples of the use of -to14; the new subject follow-
ing the -to clause in example 10 is a character with whom the narrator is in
the process of becoming romantically involved, making his introduction
more important to the narrative:
(10) Okazu no sara de afuresooni natta bon to, sore ni
main-course dish too-full become tray and it on
norikiranakatta meshi o oomori ni yosotta donburi o
not-fit-in rice heaping put-on bowl
ryoote ni mochi, shibaraku urouroshiteitaga, youyaku
both-hands have for-a-while was-wandering-around finally
2-FIN
aiteiru seki o mitsuketa. Hitori, mokumokuto tabeteiruto,

empty seat found alone silently is-eating
3-FIN 4-TO
itsunomanika Shigeru ga yoko ni taiteita.
don't-know-when S. side was-standing
5-FIN
' With hands full carrying the tray with my main course
dishes and the heaping bowl of rice which didn't fit in the
tray, I wandered around for a while and finally found an emp-
ty seat. I was eating alone silently; (suddenly) Shigeru was
standing right next to me. (Tsuka 1980:21)
(11) Oto ko no naka ni hitori, Yumiko no sugu tonari ni jindori
guys among one Y. right next sit
koohii ni satoo o iretari, mizu o torikaetarishite yatarato
coffee in sugar put-in water change really
kobi o utte iru no ga ita ga, yoku chuuishite miruto
flirting is one was and carefully look
FIN TO
annojoo Nakamura no yatsu datta.
not-surprisingly N. jerk was
FIN
4
Among the guys, there was one who sat right next to
Yumiko, and was really flirting with her by putting sugar in
her coffee and getting her water, and when I looked carefully,
it turned out not surprisingly to be that jerk Nakamura.'
(Tsuka 1980:21)
In each case, the -to clause has more foregrounding features than the
following clause. In 10, the -to clause has subject continuation from the
preceding clause while the following clause does not, while in 11 the -to
clause is perfective while the following clause is imperfective. Nevertheless,
it is clear that the -to clause is always less significant to the narrative than
the following clause.
The function of the -to clause, then, is to indicate that the relative
grounding of consecutive clauses is exceptional and that, although the -to
clause has foregrounding features which the following clause lacks, the
former is actually backgrounded to the latter. In fact, the clause following
the -to clause does tend to have one foregrounding feature, that of
chronological sequencing.
Table 29
-to N %-to PROB
Sequenced clause following 21 165 13 .73 p<.001

Not sequenced clause following 7 306 2 .27
Thus the -to form is most favored in the fairly exceptional circumstance
where the following clause has a different subject and an imperfective verb
but is still chronologically sequenced15.
3.3 Comparison of the -to form and other non-finite forms
The -to form differs sharply in function from the other non-finite forms
studied here. It is associated with following switch subjects, while the stem
and gerund forms are associated with following continued subjects. -To is
favored when the present clause is perfective and the following clause is im-
perfective, precisely the environment where neither the stem nor the gerund
is favored. Unlike the stem or the gerund, it is associated with a highly mark-
ed distribution of grounding features; so it is not surprising that it is the
rarest of the three forms, occurring only 28 times in the 500 clauses coded,
while the stem form occurred 122 times and the gerund 50 times.
The -to form does share one characteristic with the gerund, the stem,
and the m-form: it is neither maximally backgrounded nor maximally
foregrounded. It is backgrounded in that its clause is less important to the
narrative than the following clause. On the other hand, it also has more
foregrounding features than any of the other forms investigated here, so it
is by no means fully backgrounded.
4 CONCLUSION
The major findings of this study may be summed up as follows:

(i) Givón's suggestion that clause-chaining forms run counter to the
T H E DISCOURSE FUNCTION OF CLAUSE-CHAINING 395
assumption that non-finite forms must be backgrounded is not sup-

ported by the data here. The non-finite forms we have investigated
are not associated with foregrounded clauses. They are primarily
associated with following foregrounded clauses or following
clauses with characteristics parallel to those of the present clause,
(ii) However, these forms are not particularly associated with clauses
having overwhelming backgrounding characteristics either. In
terms of grounding, they occupy a position between fully finite
forms and other subordinate forms.
In this paper, we have worked toward an understanding of grounding
based upon statistical analyses of a particular kind of construction, clause-
chaining forms. The criteria we have used in coding (same or different sub-
ject, sequencing, etc.) may be applied to the study of any form in any
language. They are not the only criteria we could have applied nor even
necessarily the most revealing16; however, they do allow for a precise
description and comparison of the discourse function of these constructions.
As investigation proceeds, it will eventually become possible to classify con-
structions from different languages according to their discourse function in
a more precise way than has been possible to date. Considerable empirical
and descriptive work of this sort is required before a comprehensive theory
of discourse can be developed.
ACKNOWLEDGEMENT
We thank Shoichi Twasaki and Sandra Thompson for their helpful com-
ments on earlier drafts of this paper.
NOTES
For the abbreviations used in this article see list on page vii.
1. Our examples thus have a different objective than do the invented examples of linguists
such as Kuno (1973), which are accompanied or not by asterisks to indicate whether they
are acceptable. Examples using asterisks are supposed to prove one thing or another. Our
own examples prove nothing but are rather for illustration. We rely instead on tables with
data to prove our points.
2. Somewhat more problematic are cases such as the Telugu 'conditional' (Lisker, 1963), a
396 JOHN MYHILL A N D JUNKO HIBIYA
special non-finite form used in protasis clauses apparantly without a lexical head; in such
cases, it is likely that the conjunction has historically fused onto the clause-final verb,
creating what is synchronically a bare clause. We will leave open here the status of such
constructions; the languages we will discuss in detail here have no such ambiguous cases.
3. Since our purpose here is cross-linguistic comparison, we did not code for morphological-
ly marked aspect in the languages studied but rather for universally applicable conceptual
aspectual categories. Perfectivity is however quite difficult to code in an objective
fashion; definitions such as that in Comrie (1976) cannot be applied to naturally occurring
data without having to make an extremely large number of arbitrary decisions, and the
same may be said of other definitions which have been devised in order to rationalize ar-
tificial data rather than analyze production data. We have given considerable thought to
the problem of finding criteria which may objectively applied to as high a proportion of
naturally occurring data as possible and which also reflect the foregrounding/backgroun-
ding distinction. The most satisfactory system we could find was to code all predicates
which were habitual, stative, or progressive as imperfective; all others were coded as
perfective. Actions resulting in states were coded as perfective.
4. The data for Soddo were taken from Leslau (1968), a colletion of texts in which a speaker
describes life in the area of Ethiopia where he lives. The grammatical terminology we use
is also taken from Leslau, except for the term 'm-form', which we have coined.
5. Clause-chaining forms in many languages are marked for whether the following subject
is the same or different from the present subject. This is not the case in Soddo or
Japanese, although, as will be seen, each of the their clause-chaining forms shows a strong
statistical preference for either same subject or different subject following
6. It did not matter whether the switch was from perfective to imperfective or from im-
perfective to perfective.
7. The longest chain in the Soddo data was five clauses long, and the m-form was not used
after more than three non-finite clauses.
8. The Japanese data were taken from Tsuka (1980), a popular novel. We are here using the
terminology employed in Jorden (1963). Kuno (1973) refers to the gerund as the 'gerun-
dive' and the stem as the 'continuative' but reverses these terms in Kuno (1978). The
'gerund' is fully verbal in terms of the case-marking of its arguments. The '-to form' con-
sists of the present tense form of the verb plus the affix -to; we refer to this form as non-
finite because it cannot stand alone and it has invariant present tense form regardless of
whether it has present or past meaning.
9. The stem form is also more associated with writing style while the gerund is associated
with colloquial speech; however, as our data base is limited to written Japanese, we will
not discuss the stylistic distinction between these forms here.
10. Some of the Japanese examples have a number of quotes, which were ignored for the pur-
pose of coding. These are left out of the Japanese data and replaced by (quote). To allow
readers to follow the example, however, we have put the quotes back into the translation.
11. Kuno (1978) claims that the gerund has the meaning of temporal or logical sequencing
while the stem is used for neutral coordination. Kuno's analysis is consistent with the fin-
dings reported here in that the gerund is indeed associated with chronological sequencing
T H E DISCOURSE FUNCTION O F CLAUSE-CHAINING 397
in the following clause while the stem indicates the two clauses are parallel constructions.
12. Kuno (1973:199) claims that "the SI -te S2 construction implies that SI has taken place
before S2 does" so that the gerund should always be followed by a temporally sequenced
clause. This is clearly not the case; counterexamples in our data were namerous, including
clauses 6, 8 and 13 in example 6, clause 10 in example 7, and clause 4 in example 8.
13. Kuno (1973:194) observes that the construction with -to "carries with it the connotation
of suspense and surprise". The sentence following the -to clause provides the surprise,
and it is this feature of surprise which makes it indispensible to the narrative and hence
foregrounded. In connection to this, the -to form is favored when the preceding verb is
imperfective:
-to- N %-to PROB
Preceding imperfective 19 288 7 .68 ρ<.005
Preceding perfective 9 183 5 .32
This is because when the preceding verb is imperfective, there is less action immediately
going on, and therefore more possibility of a surprise.
14. Kuno (1973:193-4) claims that the -to form must either be "a general statement (which)
represent a habitual or logical antecedent-consequent relationship (or the clause following
the -to clause) must represent an event that the speaker could observe objectively." 11
clearly does not fall into either of these categories. This example also contradicts Kuno's
claim that the -to form is associated with surprise (see note 13), considering that the nar
rator specifically mentions that he was not surprised that it turned out to be Nakamura.
15. The -to form is also favored in the first or second clause of a sentence, disfavored in the
third, and does not occur at all later than the third clause:
-to- Ν %-to PROB
First or second 27 422 6 .71p<.05
Third 1 49 2 .29
16. For example, it would be desirable to have an objective test showing the increased
discourse salience of the clause following the Japanese -to form; our discussion of the
function of this form would be strengthened considerably by quantitative evidence of this
sort.
REFERENCES
Comrie, Bernard. Aspect. Cambridge: Cambridge University Press.

Givón, Talmy. 1987. "Beyond foreground and background". In Coherence
and Grounding in Discourse (= Typological Studies in Language, 11), R.
Tomiin (ed.), Amsterdam: John Benjamins.
Hopper, Paul J. 1979. "Aspect and foregrounding in discourse". In
Discourse and Syntax ( = Syntax and Semantics, 12) Talmy Givón (ed.),
213-41. New York: Academic Press.
— , and Sandra A. Thompson. 1980. 'Transitivity in grammar and

discourse". Language. 56:251-99.
Jorden, Eleanor Harz. 1963. Beginning Japanese Part 2. New Haven: Yale
University Press.
Kuno, Susumu. 1973. The Structure of the Japanese Language. Cambridge,
Mass.: The MIT Press.
-—. 1978. "Japanese: A characteristic OV language". In Syntactic
Typology, W. Lehmann (ed.), 57-138. Sussex: The Harvester Press.
Leslau, Wolf. 1968. Ethiopians Speak: Studies in Cultural Background.
Berkeley: University of California Press.
Lisker, Leigh. 1963. Introduction to Spoken Telugu. New York: American
Council of Learned Societies.
Tsuka, Kouhei. 1980. Itsumo Kokoro ni Taiyoo o. Tokyo: Kadokawa Press.
Nominalization and assertion in scientific
Russian prose
J o h a n n a Nichols
University of California, Berkeley
1 INTRODUCTION
A crucial question for the study of clause combining in discourse is which

predicates in a chunk of text will surface as main, or finite, verbs and which
will surface as chained, nominalized, or otherwise nonfinite verb forms. I
take it as a noncontroversial assumption that finite verbs in main clauses,
and perhaps indicative verbs in general, carry what can be called the
pragmatics of assertion; that non-indicative finite verbs such as conditionals
carry what can be called the pragmatics of contingent assertion; and that
nonfinite forms lack any such pragmatics but are capable only of presup-
posing, alluding, etc. In particular, it seems uncontroversial to claim that
nominalized verb forms presuppose but do not assert. It follows that the
choice of main vs. non-main, finite vs. nonfinite, verb forms should be at
least partly predictable from information about what was said in the prior
context, what the speaker (or writer) presupposed, and what point she in-
tended to make with a paragraph. 1 It would also seem to follow that a
translation of scientific prose from Russian to English — especially since
these are languages with shared ancestry, similar sociopolitical contexts, and
similar informational functions within similar (and partly overlapping)
scientific communities — should be able to mechanically transfer to English
the Russian choices of nominalized vs. finite verb form, with perhaps some
heaviness of style but no effect on content.
Scientific Russian usage belies these claims. The Russian verbal noun
cannot be used to assert; but it is systematically used in contexts where the
best English translation uses assertion, or at least a finite verb. This paper
surveys examples of that usage of verbal nouns, collected during the process
400 JOHANNA NICHOLS
of translating Gamkrelidze and Ivanov (1984). The analysis of those ex-

amples shows that the choice of assertion vs. non-assertion, and the choice
of finite vs. nominalized verb form, is at least partly based on language-
specific inventories of grammatical categories. It also shows that the rela-
tions among reader, writer, text, and knowledge differ in Russian and
English scientific prose, and these differences explain some of the dif-
ferences in use of nominalizations. Specifically, as will be argued below, the
Russian text is not so much a communicative contract between writer and
reader as it is a gnomic statement of available knowledge.
Scientific, or more generally expository, prose has received little atten-
tion from discourse grammarians. Since it lacks plot and a time line, it is
quite different from the narrative text that has provided most of our
understanding of discourse organization and categories. Going back to the
early distinction of Benveniste (1971), scientific prose can be classified as
speech rather than narrative, since it abounds in speech categories such as
first persons and perfects. The present paper will have no more to say about
a general theory of scientific prose as text.
2 ANALYSIS
There is a fairly standard cognitive-functional view wherein grammatical

categories signal (or are grammaticalizations of) discourse functions which
themselves answer to cognitive and processing needs. On such a view, many
features of discourse and grammar have to do with the speaker's monitoring
of the hearer's frame of mind, knowledge, and experience (saliently, his ex-
perience of the prior stretch of discourse). This paper assumes that general
framework as an analytic strategy, and indeed that framework emerges as
a useful one for describing the differences between Russian and English
usage. But let it be clearly noted at the outset that such a framework con-
tributes little to our understanding of the Russian facts studied here; it is of
value insofar as it lets us state how Russian differs from English. At least
for the phenomena studied here, Russian seems not at all to grammaticalize
the cognitive and discourse parameters that figure prominently in the receiv-
ed view. If Russian were described in its own terms, those terms would prob-
ably not involve reference to what the hearer is assumed to know, to opening
and closing mental files, and so on. This finding does not undermine the ade-
quacy of the received view, but it does seem to call its universality into ques-
tion.
NOMINALIZATION AND ASSERTION IN RUSSIAN 401
Three cognitive-functional parameters figure in the analysis below:

definiteness, new participants, and information increments. The received
view for English, and my findings for Russian, will be briefly stated here
before the examples are surveyed in Section 3.
2.1 English
Definiteness (see e.g. Givón, 1984:399ff., Foley and Van Valin,

1985:285): English definiteness has to do with the writer's assessment of
what the reader knows, has retained from previous context, etc., and hence
with what she can expect him to identify. We could represent the presupposi
tion carried by the in its various functions as  assume you know which
one', a formula which reflects the centrality of the personalized speaker and
personalized hearer to the received view of English definiteness.
New participants. I will simply assume that the introduction of new issues
into scientific prose is functionally analogous to the introduction of the new
participants into narrative prose. The indefinite article in English cannot by
itself gracefully introduce important new participants into the discourse.
New participants, at least potentially salient ones, are best introduced with
motion and stance verbs, typically with presentative constructions (e.g.
Hetzron, 1975). This follows from the next principle.
Information increments. English prefers to introduce no more than one
chunk of new information per clause (see Givón, 1984:258ff.). To judge
from the corpus for this study, nominalizations do not count as clauses for
purposes of distributing new information: in some of the examples below,
what is wrong with literal English translations is that the nominalization car
ries an important chunk of new information which encumbers the main
clause in English (although not in Russian).
2.2 Russian
Definiteness: Russian has no articles and no grammatical category of

definiteness. It does no systematic, obligatory monitoring of the writer's
assessment of the reader's knowledge and experience of the text. It does have
a few lexical items which carry meanings related to definiteness: dannyj, lit.
'given', often close to 'the, this'; sootvetstvujuščij 'corresponding', often
'the, the relevant, the one at issue, belonging to the most salient recently
mentioned noun'; opredelennyj, lit. 'definite, determined', often 'a, one or
402 JOHANNA NICHOLS
another, a particular (but unspecified or unspecifiable), some specific',

izvestnyj, lit. 'known', often 'a certain, a particular'. However these tend to
have to do with speciflability rather than with (the writer's assumption of)
the speaker's knowledge. Furthermore, recent works (Timberlake, 1975;
Klenin, 1983; Nichols, 1981) have shown that Russian morphosyntax is sen
sitive to a parameter reminiscent of definiteness at a number of points; but
such 'definiteness' has to do with quantification, individuation, discreteness,
etc., and not with the hearer's knowledge or speaker's assessment thereof.
There is one lexical device in the realm of definiteness, etc. that is used
systematically and frequently in scientific Russian: the abbreviation sr. 'cf.'
(it abbreviates Russ. sravni 'compare [imper.]'). As examples in 3.4 below
show, the abbreviation carries -a presupposition about general knowledge
and truth rather than about the reader's knowledge (or the writer's assess
ment thereof). This meaning is consistent with the contention that the Rus
sian text is not so much a communicative contract between writer and reader
as it is a statement of truth or general knowledge.
New participants. Russian has a VS presentative construction whose
typical function in narrative discourse is to introduce a new participant. This
construction appears to be lacking in scientific prose, however. As several
examples below will show, Russian is perfectly capable of inserting a new
participant into the discussion as an argument of an ordinary SV clause,
where literal English translations are awkward.
Information incrementation. Russian also appears to employ a one-
chunk-per-clause constraint of sorts. But in Russian, unlike English, a clause
can both introduce a new issue and make a new point about that issue, as
several examples below demonstrate. In English, the new issue counts as a
chunk of new information and overloads the clause. In spoken Russian, and
hence when a scientific paper is read aloud, intonation contours function
conspicuously to break clauses down into smaller information units, and
there is a rigid one-chunk-per-contour constraint. But the printed text, which
does not show intonation, contains clauses that would be perceived as
overloaded in English.
2.3 Methodological observations
The following example (Gamkrelidze and Ivanov 1984:lxxx) is an entire

ly typical Russian sentence which bristles with nominalizations. Nominaliza-
tions are in boldface.
(1) Ob" jūsnenie fonemnyx sootvetstvij mezdu jazykami

explanation of phonemic correspondences between languages
obščnosťju ix proisxoždenija iz opredelennoj
by commonality of their descent from particular
isxodnoj jazykovoj sistemy predpolagaet neobxodimosť
source linguistic system assumes necessity
rekonstrukcii  s ceľju izučenija vozniknovenija
of reconstruction of it with goal of study of rise
i putej preobrazovanija istoriceski zasvideteľstvovannyx
& paths of transformation of historically attested
rodstvennyx jazykovyx sistem
related linguistic systems
Literally: '(An) explanation of phoneme correspondences be
tween languages by means of (the) commonality of their des
cent from an original linguistic system assumes (the) necessity
of (a) reconstruction of it [the original system] with the goal of
(a) study of the rise and paths of transformation of historically
attested cognate linguistic systems.'
Free translation: If phonemic correspondences between
languages are to be explained as due to a common descent
from an original system, then that system must be
reconstructed in order that the origin of the attested cognate
languages and the changes undergone by them may be
studied.'
The literal translation has many stylistic faults, and in addition it con
tains one major error of content: it fails to get across the main point of the
sentence, which is that reconstructions must be made. Even if we streamline
the original sentence, no English version which contains nominalized
necessity or a synonym of it communicates that point unambiguously:
a. Such an explanation carries an obligation to reconstruct ...
b. Explaining correspondences in this fashion implies the necessity
of reconstruction ...
 An explanation of this type presupposes the necessity of
reconstructing ...
(a) comes closest, because carry an obligation is a fixed expression
404 JOHANNA NICHOLS
(specifically, it involves a lexical function: see Meľčuk, 1982) which

amounts to a complex verb and can be said to assert obligation as much as
carry. (b) and (c) fail completely to communicate the intended content. Only
by asserting 'necessity', which can be done by using must as predicate of the
main clause, can English get across the point of the Russian sentence.
The main claim of this paper is that English-Russian differences reveal
ed in examples like (1) are due to the differences in definiteness, treatment
of new participants, and information increments discussed above. It is not
the case that Russian nominahzations carry assertion. Russian and English
nominahzations are structurally alike, and both represent a part of speech
that cannot assert. In the Russian sentence, the verb predpolagaet 'assumes,
presupposes' is asserted and the noun neobxodimosť 'necessity' is presup
posed; in English, If... then conveys, but does not assert, the content of
predpolagaet 'assumes' while must is asserted and conveys the content of
neobxodimosť 'necessity'.
It is the discourse categories that force English to redistribute assertion.
Definiteness is involved in the following fashion. In the literal translation,
necessity is a noun and hence forces the writer to choose between the, a and
0. Now, it is my observation that nominahzations in English are most often
definite.2 Definiteness is favored when the nominalization has a subject or
object phrase with of, and it is favored when the nominalization functions
as object. Indefiniteness is favored only by nominahzations that are highly
lexicalized (such as explanation, goal, study in (1)). These principles, and
others doubtless involved, must follow from regular rules of English gram
mar which act together to create a statistical propensity for definiteness on
nominahzations. This means that the speaker has little freedom to choose
between definiteness and indefiniteness for nominahzations. And in the case
of (1), the (virtually obligatory) definiteness of necessity carries the incorrect
presupposition that this necessity has been mentioned before, is inferrable,
or is otherwise known information. This precludes interpreting necessity as
the main and new point of the sentence.
The constraint on new participants can be discerned for (1) in the fact
that, even if necessity could be made indefinite ('The explanation ... assumes
a necessity to ...'), it could not be interpreted as referring to a newly in
troduced salient issue. Even a necessity implies that necessity has been at
issue. (A clearer example of this effect is (4) below.) Now, there is nothing
about English nominahzations per se that presupposes prior mention of their
category. Rather, this is a secondary effect: since new participants cannot be
introduced as ordinary indefinites, use of an ordinary indefinite entails that

the participant or its category is not new; since indefiniteness precludes inter
preting the referent as old information, it must be its category that is inter
preted as old.
The constraint on information incrementation plays a very clear role in
(1), where the literal translation is almost uninterpretable precisely because
each nominalization contributes potentially salient information and the
sentence is greatly overloaded.
In Russian, the absence of the English new-participant constraint means
that a verbal noun carries no secondary presupposition of prior mention in
the text. The absence of definiteness in Russian means that a nominalization
carries no presupposition about the hearer's knowledge or familiarity with
the text. It can be used to name events, etc. previously asserted, and it can
be used with equal ease to introduce newly mentioned topics. Importantly,
it can be used to make statements which are the writer's original claims,
presented for the first time. Since claims articulated in this fashion are not
asserted, the effect of using a verbal noun for new claims is to blur the
distinction between writer's opinion, analysis, findings, etc. and general,
received knowledge or truth. (General knowledge may of course be explicitly
indicated by giving bibliographical references.) In this respect the Russian
text resembles a statement of general truth which the reader may learn,
rather than a communication from a writer to a reader which is intended to
be experienced as a text by the reader.
It is important to make clear what this study does and does not purport
to be. It is limited to examples drawn from only one text — a text recognized
by all native speakers consulted as an example of well-written scientific Rus
sian, but nonetheless only one text. The Russian-English comparison is
based on differences that emerge from contrasting particular examples in the
text with their possible English translations. The Russian corpus results from
a survey; the English data is contrastive only, and arrived at by my own in
trospection. Hence the study is offered as an analysis of Russian but not as
an analysis of English; English figures in this study only heuristically.
A final caveat is that this study does not distinguish the more or less lex-
icalized verbal nouns from the more syntactically formed ones which func
tion as true nominalizations. The two types pattern alike, as is consistent
with the claim that it is not properties of nominalizations but grammatical
and discourse categories of nouns that produce the differences in English
and Russian nominalizations. Furthermore, Section 3.6 shows that
406 JOHANNA NICHOLS
nominalizations and ordinary underived nouns pattern alike in the relevant

respects. Additional evidence that they pattern alike comes from examples
such as the following (Gamkrelidze and Ivanov 1984:lxxxiv), where the
highly lexicalized analiz 'analysis' and the clearly syntactic privedenie 'bring
ing' are coordinated and treated as parallel.
(2) Takoj analiz tradicionnyx indoevropejskix rekonstrukcii
such analysis of traditional Indo-European reconstructions
i privedenie ix v sootvetstvie s tipologiceski
& bringing of them into correspondence with typologically
verojatnymi sistemami mozet vyzvať suscestvennyj
plausible systems can evoke essential
peresmotr ètix rekonštrukcii.
reconsideration of these reconstructions
This analysis of traditional reconstructions of IE, and their
adjustment to typologically plausible systems, can lead to a
fundamental overhaul of those reconstructions.'
3 EXAMPLES
The corpus for this survey was collected semi-systematically: for a

3C page stretch of Russian text (Gamkrelidze and Ivanov, 1984:623-53)
I monitored all examples of nominalizations which turned out to require
translations as English finite verbs for reasons other than purely stylistic
considerations such as 'heaviness', 'clumsiness', etc. I also continued to
monitor interesting examples throughout the text. The completed analysis
was then tested on a twenty-page stretch of the then-untranslated first
volume of the same work, and on selected passages in the scientific Russian
of other authors. Many examples were discussed with native speakers ex
perienced in reading and writing expository prose. All of these tests show
that the usage documented here is systematic and that the analysis given
below accounts for all nominalization tokens encountered. Relative frequen
cies of the various functional types of nominalizations have not been
tabulated, since the point of interest is not the frequency of the phenomenon
at issue but the fact that it occurs at all.3
Examples are presented in the following format. First there is a section
and page citation (the page number is that of the Russian original). The rele-
vant sentence is given in Russian (transliterated) with glosses and a transla

tion which renders the verbal noun literally; the verbal noun is boldfaced in
the Russian and the translation. (Footnote numbers have omitted from the
examples, and minor typographical changes have been made.) Then, follow
ing the notation 'Context', the same sentence is given (boldfaced) in a less
literal translation which conveys the meaning better; it is given with a
paragraph or so of preceding and/or following context. (Context is not given
for examples (3) and (14).) For some examples, lengthy preceding or follow
ing context is summarized in square brackets. Each example is followed by
a discussion of it.
3.1 Verbal nouns which presuppose and are definite
Verbal nouns are extremely frequent in Russian scientific prose. A good

number of them function exactly as they do in English and can be rendered
literally in a smooth and accurate English translation.
(3) G and I 4.2.1.3 (p. 650)
Semantika privedennyx obrazovanij na koren' *ueį- / *ui-,
semantics of-cited formations on root
 i značenie samogo kornja 'viť', 'plesti', deleat ves'
like meaning itself of-root twist weave makes highly
verojatnym predpoloženie ob obrazovanii indoevropejskogo
plausible assumption of formation of-Indo-European
slova dija 'vinograda' 'vina'  'ploda v'juscegosja
word for grape wine as fruit of-winding
rastenija' ot togo ze kornja *ueį~ ...
plant from the same root
T h e semantics of the above-mentioned formations from *ueį-
/ *ui֊, as well as the very meaning of the root 'weave, plait,
twist', renders entirely plausible the claim that the Indo-
European term for 'grape' and 'wine' is a formation from the
same root *ueį֊ in its various vowel grades — 'fruit of twining
plant'.
Discussion: The claim for this etymology was cited one page back,
characterized as 'long since proposed', and supported by references and
cognates.
Another example of this type is vyvod 'conclusion', cited in (9) below.
408 JOHANNA NICHOLS
3.2. Verbal nouns used for first mention in the text
In contrast to the preceding type, many nominalized verbs have no prior

mention and cannot be correctly translated with an English definite article.
(4) is an example.
(4) G and I 4.1.3.3 (p. 623)
Odnako po novejsim dannym paleobotaniki ottesnenie
however by latest data paleobotany-GEN pushing back
buka na zapad i prodviženie ego na sever ...
beech-GEN to west and moving forth its to north
proisxodilo ne ranee Subboreaľnogo perioda (...)
occurred not earlier than-Subboreal period
'However, according to recent paleobotanical data, (the?)
displacement of the beech to the west and its advance north-
ward ... occurred no earlier than the Subboreal ...'
or: '... a displacement of the beech toward the west and an ad
vance to the north took place no earlier than the Subboreal'
Context:
In historical times the eastern limit of the beech (Fagus
silvatica L.) has held closely to a line running from the Baltic sea
approximately along the Visla river southward to the middle
Danube. In fact this distribution has given rise to one of the basic
arguments advanced by proponents of a European homeland for
Proto-Indo-European (see Wissmann 1952; Krogmann 1954;
1955; Lane 1967 and references therein).
However, recent paleobotanical research has shown that the
limits of the beech range moved west and north to their historical
European locations no earlier than the Subboreal period (late
fourth to first millennia B.C.): see Nejštadt 1957:305-11. In addi
tion, varieties of the eastern beech, Fagus orientalis, have always
been found in the Near East, from Asia Minor to the
Transcaucasus and northern Iran.
Discussion: Nowhere earlier in the text is there any discussion of move
ment of the beech boundaries. The first discussion of the range of the beech
begins with the passage quoted here. For English, use of verbal nouns ('a
withdrawal ... and an advance ... took place') is possible here only if the
noun can be made indefinite. This version sounds less good in context,
however, partly because it is still not identified as a new issue and partly
because (due to English information-incrementation constraints) the
relatively empty verb 'took place' tends to shift the focus of the predication
to the time adverbial 'no later than the Subboreaľ, leaving the verbal noun
to suggest (incorrectly) that movement of the beech lines has been at issue
in the preceding text.
(5) G and I 4.1.13.2 (p. 645)
Predpoloženie o sv jazi osnov *khr-n- 'kiziľ, 'višnja'
assumption of link of-roots cornel cherry cherry
i *kher- 'rasti, 'korm' stavit vopros o formaľnom
and grow fodder raises question of formal
sootnosenii veljarnogo *kh- v osnove khr-n- i palataľnogo
correlation of-velar in base and palatal
h h
k v korne *k her-.
in root
'(An) assumption of a connection between the roots *khr-n-
'cornel cherry, cherry' and *kher- 'grow'; fodder' raises the
question of formal relatedness of the velar *kh in *khhr-n- and
the palatal *kh in *khher-.-'
or: '(An) assumption that the roots ... are connected raises the
question of whether the velar ... and the palatal ... are formal
ly related.'
Context:
[4.1.13.1. Cognates for 'cherry' (only)]
[4.1.13.2. Range of cherry and cornel cherry. Cornel cherry as
livestock fodder in earlier IE traditions]
The roots *khr-n- 'cornel cherry, cherry' and *kcher- 'grow;
fodder' may be connected, a possibility which raises the question
of whether the velar of the first root corresponds formally to the
palatovelar of the second. The two forms can be reduced to a
single source form *kher- if we assume neutralization of the
palatality feature before the *r of the zero-grade root form:
*khher- (Lith. sérti) ~ *khr-n- (Lith. Kirnis) (see 109ff. above for
this neutralization, reflected in such forms as Lith. kárvė 'cow':
stirna 'roe deer').
410 JOHANNA NICHOLS
Discussion: Nothing has been said before about the possibility that the
two roots are related. Specifically, no such assumption (predpolozenie) has
been made explicitly.
3.3 The truth of the presupposition made by the verbal noun is either being
established or being questioned
Presentation of the writer's original claims and arguments is one of the

clearest places calling for assertion in English. The following examples show
that Russian can use nominalizations even in this kind of context.
(6) G and I 4.1.12.6, p. 643
Drevnegermanskij mifologičeskij motiv, v kotorom 'jabloki
ancient Germanic mythic motif in which apples
bessmertija' svjazany s bogom Poèzii Bragr (...) motzet
of-immortality connected with god of-Poetry may
služiť ukazaniem na ispoI'zovanie jablok dlja
serve as-indication to use of-apples for
izgotovlenija rituaľnogo napitka ...
preparation of-ritual beverage
'(The? An?) ancient Germanic mythic motif in which 'apples
of immortality' are linked with the god of poetry Bragr ...
may serve as evidence for use of apples in preparing a ritual
beverage...'
Context:
[Golden apples which confer immortality in several IE traditions]
The ancient Germanic mythic motif linking the 'apples of im
mortality' with the god of poetry Bragr (whose name is
etymologically related to *bhreu- 'ferment (of beverages); brew':
cf. OE breowan, Engl. brew, OHG briuwan, Ger. bräuen, and
also Lat. ferctum 'sacrificial cake made of groats, butter, and
honey', Lith. biř'gelas 'beer', etc.) may be evidence that apples
were used to prepare a ritual beverage that was the functional cor
respondent to wine, beer, and other intoxicating beverages.
Discussion: This use of apples has not been mentioned before; it is the
authors' original analysis. The close translation, which uses a verbal noun,
is not ungrammatical if inserted into the larger context, but (due to the new-
participant constraint) it suggests that this use of apples has been at issue
before in the text and hence obscures the originality of the claim.
(7) G and I4.1.12.3 (p. 639)
Takoj vyvod podderzivalsja otsutstviem nadeznogo
such conclusion was supported by-absence of-reliable
sootvetstvija v kakom-libo drugom drevnem dialektnom
correspondence in whatever other ancient dialect
areale.
grouping
This conclusion has received support from the absence of a
reliable correspondence in any other ancient dialect group.'
Context:
The restriction of *ablu֊ / *aplu֊ to the Ancient European
dialects could be taken as evidence that the word appeared in this
area at a relatively late stage of Indo-European dialect groupings.
Such a conclusion has drawn support from the absence of any
reliable correspondent in any other ancient dialect group.
Discovery of a cognate to the Ancient European form in any
other dialect group would enable us to project the word 'apple'
back to an earlier dialect stage. A Hittite form sam(a)l uųanza 'ap
ple, apple tree' is of particular interest here.
[Discussion of the Hittite form. Its correspondence to the
other IE forms. Reconstruction of a single PIE form.]
Discussion: Absence requires the here in English, although it has not
been mentioned before. That there is no reliable cognate outside of Ancient
European has not been explicitly stated in the foregoing, but it is inferrable
from the explicit statement (two sections back) that a word for 'apple' is at
tested in Ancient European (=Baltic, Slavic, Germanic, Italic, Celtic).
However, the subsequent paragraphs go on to falsify this assumption and
argue that the Hittite word is cognate.
(8) G and I 4.1.14.1 (p. 645)
Bolee drevnee značenie 'tutovoe derevo' i sootvetstvenno
more old meaning mulberry tree and respectively
'plody tutovogo dereva' možno videť v rannem greceskom
fruits of-mulberry tree can see in early Greek
412 JOHANNA NICHOLS
znacenii formy móron i  proizvodnyx  Gomera ...

meaning of-form and its derivatives in Homer
'The/an earlier meaning 'mulberry tree' or 'mulberry' can be
seen in the early Greek meaning of móron ['mulberry (fruit or
tree)'] and its derivatives in Homer...'
Context:
[IE protoform for 'mulberry'; cognates, some glossed
'blackberry', without comment]
The dialect distribution of the form (Greek, Armenian,
Italic-Celtic) testifies to its antiquity in Indo-European in the
meanings 'mulberry' and 'blackberry'.
That the meaning 'mulberry' is the older one can be seen in
the early Greek meaning of móron and its derivatives in Homer,
as well as in the archaic opposition of -s and -m forms denoting
respectively the plant and the fruit in Latin; moreover, Lat. morus
means only 'mulberry tree' and not 'blackberry plant' (while
morum means both 'mulberry' and 'blackberry'), another ancient
word rubus (from PIE *rdh-o-s, see Schulze 1933: Skt. várdhati
'grows', vrddh-á 'full-grown') denoting the blackberry plant.
Semantic extension of 'mulberry' to 'blackberry' could have been
based on the similar form and color of the two berries.
Another possible translation of the sentence would be:
'The meaning 'mulberry' is the older one, as can be seen in the early
Greek meaning of móron and its derivatives in Homer, ...'
Discussion: This example is structurally different from the preceding
ones, in that it is not the verbal noun itself but its modifier 'older' which
must be made into the predicate when the nominalized clause is asserted in
English.
The question of which meaning is older is raised here for the first time.
The passage presents the authors' conclusion, which is an original claim: the
PIE meaning of the word is 'mulberry', and semantic discrepancies among
the reflexes can be accounted for. The literal English translation cannot be
interpreted as a first announcement of this new claim.
(9) G and I 4.2.1.7 (p. 653)

Vyvod ob obšceindoevropejskom xaraktere perednea iatskogo
conclusion of PIE character of SW Asian
migracionnogo nazvanija 'vina' i ètimologiceskoj svjazi
migratory term wine-GEN and etymological link
ego s pervicnym indoevropejskom kornem stavit vopros
its with original Indo-European root raises question
 pricinax otsutstvija sootvetstvujuscix slov v znacenii
of reasons of-absence of [these] words in meaning
'vina' i 'vinograda' v rjade drevnix indoevropejskix
wine and grape in number of-early Indo-European
dialektov ...
dialects
The conclusion of a PIE nature of the Southwest Asian
migratory term for 'wine' and its etymological connection to a
native IE root raises the question of the reasons for the
absence of words meaning 'wine' and 'grape' in a number of
early IE dialects...'
Context:
[Reflexes of PIE 'wine'; none are included from Tocharian or
Indo-Iranian, but that fact is not commented on; the list takes up
a whole page, so the absence is hardly conspicuous.]
[PIE 'wine' is native IE and spread as a migratory term to other
languages of ancient Near East.]
[Rise of viticulture centered in ancient Near East]
[Religious and cultural significance of wine in early IE traditions]
[The next section, with its headings, begins as follows:]
4.2.1.7. Absence of the Indo-European term for 'wine' and
'grape' in Indo-Iranian. The cult of the divine beverage
soma/haoma.
The Indo-European character of the Southwest Asian
migratory term for 'wine' and its etymological connection to a
native Indo-European root raises the question of why the cognate
words for 'wine' and 'grape' are missing from a number of early
Indo-European dialects such as Indo-Iranian and Tocharian. It
414 JOHANNA NICHOLS
may be assumed that the place of wine as a cultic and everyday

beverage was being taken in the Indo-Iranian tradition by other
intoxicating beverages, made from plants other than grapes which
replaced grapes in the new ecological conditions where the early
Indo-Iranians lived.
Discussion: English absence must again be definite, although this
absence has not been mentioned before. The lack of an Indo-Iranian and
Tocharian cognate is a new issue, announced here. (The mention in the sec
tion heading does not constitute an occurrence in the text.) The importance
of this fact — and evidently the fact itself — are first pointed out by the
authors, and are focal to their overall argument. An attentive reader might
have noticed the absence of Indo-Iranian and Tocharian cognates, but few
will have have done so in view of the extensive presentation of cognates from
the other stocks and the assertion that the wide dialect distribution of
reflexes is evidence for PIE status of the protoform. Also, cognates to the
root (connected with 'weave' and ultimately descriptive of the grapevine)
were adduced from Indo-Iranian; it is only the meanings 'wine', 'grape' for
these cognates that are lacking there.
(Note, in contrast, that the first word in the passage, vyvod 'conclu
sion', presupposes a known fact much as verbal nouns do in English: that
the word is native Indo-European and a migratory term have both been
established earlier in the same section.)
3.4 Examples with sr. 'cf.'
(10) G and I 6.5.6 (p. 707)

sr. nevozmožnosť rekonstrukcii s nesomnennosťju
cf. the impossibility of a reconstruction with certainty
obsceindoevropejskogo special'nogo nazvanija (goncarnogo
Proto-Indo-European-GEN special(ized) name-GEN pottery-ADJ
kruga'
wheel-GEN
'cf. the impossibility of a reconstruction with certainty of a
Proto-Indo-European specialized term for 'potter's wheel'.
Context:
The absence of any other clearly Proto-Indo-European words for
pottery making may be due to loss of the common words for pot
tery and potting implements in the daughter stocks and their
replacement by new terms (for dialect pottery terms see Trubačev
1966:173-308); for instance, it is impossible to reconstruct with
certainty a Proto-Indo-European term for the potter's wheel.
[footnote: 'Unless we regard the Indo-European words for
'wheel' (examples) as words for 'pottery wheeľ ...]
Discussion: That a reconstruction for 'potter's wheel' is impossible has

not yet been mentioned. The abbreviation sr. indicates that this fact is not
(or is not presented as) the authors' original conclusion but that the informa
tion is generally known or available to the scientific community. If cf. is used
in the English translation of this sentence, the noun must be definite; but
such definiteness suggests that the fact was mentioned before and is
subscribed to by the authors, neither of which is actually the case. A finite
clause is therefore the most accurate translation.
(11) G and I 5.5.2, p. 512

... sr. prinesenie v zertvu kogtej rysi ili medvedja
...cf. bringing in sacrifice claws-GEN lynx-GEN or bear-GEN
'cf. (the) sacrificing of lynx or bear claws'
...sr. takže yozmožnyj semanticeskij sdvig sr.-irl. lug 'rys''
... cf. also possible semantic shift Mid.-Ir.[GEN] lynx
'cf. also (the/a) possible semantic shift in Middle Irish lug'
Context:
The lynx is of minimal mythological and ritual significance
in most ancient Indo-European traditions. In northern traditions
the lynx is a functional replacement for the large predators not
found in those regions. In East Slavic burial rites for kings, the
role of the leopard in other Indo-European traditions is taken by
the lynx (a sacrifice of lynx or bear claws was made); in East
Slavic folklore the epithet ljutyj z v e r ' refers to lions, leopards, or
lynxes (Ivanov and Toporov, 1974:59). In Lithuanian, the word
for 'lynx' can also be applied to tigers and leopards (Būga
1958-61.11.549); MIr. lug 'lynx' may also have undergone a
416 JOHANNA NICHOLS
semantic shift; Latvian folk songs glorify lynx pelts (lūšu

kazuocini).
Discussion: The only preceding reference to the Middle Irish word was
its inclusion (earlier on the same page) in a list of IE words for 'lynx';
nothing was said about a semantic shift (for that or any other form in the
list). In English, cf. requires definiteness, which is inappropriate here
because the reader knows nothing about the sacrifice of claws or the seman
tic shift. The effect of the two clauses with sr. in Russian is to indicate that
these are facts known to folklorists and etymologists, without presupposing
that the reader knows anything about them.
3.5 Other explicit devices for presupposing.
It is of course possible to indicate in Russian that something has been

mentioned before in the text. The various devices with this function indicate
that the reader is expected to recall the prior mention, just as their analogs
do in English. Where Russian differs from English is not in being unable to
signal such meanings, but rather in not obligatorily monitoring them. One
obvious device for indicating prior mention is cross-reference to earlier men
tions in the same text. An example is (5) above, where sootnošenie 'correla
tion, correspondence' (sc. of velars and palatovelars in PIE) refers to
something that has been established earlier in the same work. That fact is
signaled not by the use of the verbal noun, which is ambiguous (as shown
by the preceding examples in the same sentence), but by the reference to p.
109 of the same work.
(12) G and I 4.1.12.3, footnote 44

Ves'ma xarakteren fakt proniknovenija v tjurksie
highly characteristic fact of-penetration into Turkic
jazyki (eščce v obšcetjurkskij period)
languages still/back in Proto-Turkic period of
'drevneevropejskogo' slova, cto možet svideteľstvovať
of Ancient European word which can testify
o nalicii v opredelennoe vremja kontaktov mezdu ètimi
of presence in the-specified time of-contacts between these
jazykami.
languages
T h e penetration of an Ancient European word into Turkic

languages as early as Proto-Turkic times is highly significant
and testifies to contacts between the two language groups at
the time in question.'
Context:
The hypothetical *amlu- 'apple' underlying the Ancient
European cognates finds real-world attestation in Proto-Turkic
*alma- 'apple', which many Turkologists have considered a bor
rowing from an unknown Indo-European source (Sevortjan
1974:138; G. Morgenstierne, p.c.). Borrowing from Iranian into
Turkic is to be ruled out on phonetic grounds: Iranian has -r-,
Turkic -1- (on the Iranian forms see below). In Turkic, Ancient
European *amlu- yields attested alma as a result of natural and
minor phonetic adjustments. Thus we can establish a source for
[or the source of] Proto-Turkic *alma- 'apple' with a sufficient
degree of certainty. The penetration of an Ancient European
word into Turkic at the level of Proto-Turkic testifies to contacts
between the two groups at the time of Ancient European and
Proto-Turkic.
or: 'The fact that an Ancient European word entered Turkic ... is
significant, as it testifies...'
Discussion: This is the entire footnote and the entirety of the discussion
of Turkic in this section. Entry of an IE word into Turkic has not been ex
plicitly asserted prior to the last sentence, but is could easily be inferred from
the preceding sentences quoted here. The inclusion of fakt in the Russian
passage makes it unambiguous that the entry of the word into Turkic did
take place. In this respect the verbal noun can be said to presuppose. Note,
however, that the explicit claim for the entry is being made here for the first
time; this is then simultaneously a presupposition and a first claim of a new
conclusion. Translation with an English (definite) verbal noun is possible
here because the conclusion is sufficiently obvious from the preceding con
text that the reader can be expected to share it; but it is slightly underhanded
because the text does not really contain the assertion. Perhaps more impor
tantly, the verbal noun in English suggests that the conclusion is not an
original one presented here (which it is) but a more generally held opinion.
Thus Russ. fakt here does not reflect the writer's assumption that the
418 JOHANNA NICHOLS
reader knows something, but rather tells the reader that this is a fact.4
3.6 Similar behavior of nouns which are not deverbal
So far we have seen that Russian verbal nouns can be used where English
prefers finite verbs and especially assertions, and that the Russian verbal
noun differs from the English one in carrying no presuppositions about text
structure and the reader's knowledge of it. The next two examples show that
this behavior is not limited to verbal nouns, but is found in nouns generally.
This is not surprising, since it is not nominalizations but the grammatical
and discourse categories of nouns that differ in the two languages.
(13) G and I 4.1.14.1 (p. 646)
Indoevropejskij xarakter slova v znacenii 'tutovogo dereva
Indo European character of-word in meaning mulberry tree
s temnymi plodami', 'temnýj plod tutovogo dereva'
with dark fruits dark fruit of-mulberry tree
možet byť podķreplen vozmoznost'ju ètimologiceskoj svjazi
can be confirmed by-possibility of-etymological link
ego s kornem *mer-, *mor= 'temnyj', 'černyj'
of-it with root dark black
(Pokorny, 1959:734): ...
T h e Indo-European character of the word in the meaning
'dark-fruited mulberry tree', 'dark fruit of mulberry tree' can
be confirmed by the possibility of its etymological connection
to the root *mer-, *mor- 'dark', 'black' (Pokorny. 1959:734)
Context:
[PIE word for 'mulberry'. Cognates means variously
'mulberry' and 'raspberry'; 'mulberry' as the older meaning.]
The meaning 'dark-fruited mulberry tree', 'dark-colo -d
fruit of mulberry tree' is of [Proto-JIndo-European age, as is con
firmed by its possible etymological connection to *mer-, *mor-
'dark, black' (Pokorny 1959:734): Hom. Gk. mórussõ 'blacken,
make dirty', memorugménos 'dirty', mórukhos 'black (with
soot)'.
Discussion: Like (8), this example also involves not just a verbal noun
but an NP with a modifier, and the modifier seems to call for assertion in
English. The sentence asserts that the meaning 'dark' goes back to PIE as
a component of the meaning 'mulberry (tree)'. That the focus is on the ele
ment 'dark' is hard to capture in English; one way to render it might be to
use non-restrictive modifiers:
The meaning 'mulberry, a tree with dark fruit', 'fruit of the
mulberry tree, dark in color' ...
xarakter 'character' (i.e. 'age', 'time depth' in this context) happens to
be a root abstract noun rather than one derived from a verb. It can be
translated as asserted here on first mention, just as verbal nouns are in the
examples given above.
(14) G and I 4.2.1.3, note 64 (note 2 on pp. 650-51, cited portion
from p. 651)
Pri vsej razvitosti vinogradarstva i vinodelija v
PREP all developedness of-viticulture and winemaking in
drevnejsem Zakavkaťe starye, iskonno kartveľskie slova dlja
earliest Transcaucasus old native Kartvelian words for
osnovnyx ponjatij svjazannyx s kuľturoj vinodelija,
basic concepts connected with culture of-winemaking
mogli byť vytesneny zaimstvovanijami iz drugix jazykov ...
could be displaced by-borrowings from other languages
Given the considerable development of viticulture and
winemaking in the ancient Transcaucasus, ancient, native Kart
velian terms for the basic concepts of winemaking culture
could have been displaced by loans from other languages.'
or: '...if there had been earlier native Kartvelian terms... they
might well have been displaced by borrowings ...
or: '...there could have been earlier native Kartvelian terms...
which were displaced by borrowings ...'
Discussion: This passage follows a discussion of Kartvelian words that
can be traced to IE. This is the first reference to a potentially earlier, native
stratum, an issue important enough to fall under the new-participant con
straint. Hence the single word is best translated by a whole subordinate
clause.
420 JOHANNA NICHOLS
Slova 'words' is not a deverbal noun and not even an abstraction, but
it behaves the same way as deverbal nouns used for first mention. Like
deverbal nouns it lends itself to an English translation using a finite verb
form.
3.7 A formulaic example
Corresponding to English 'Even granting that...', 'Even if we assume

that...', 'Assuming for the sake of argument that...', etc., Russian regularly
uses a chain of at least two nominalized verbs in a formula consisting of the
preposition pri 'in connection with' plus the prepositional case form of
dopušcenie 'admission, assumption' plus one or more other verbal nouns.
(15) G and I Vol. I, 1.1.1 (p. 7)
Odnako, daze pri dopuščenii vozmožností naličija
however even admission of possibility of presence
obšceindoevropejskogo *b v rassmotrennyx vyše formax, ...
of PIE in examined above forms
'However, even granting the possibility of the existence of PIE
*b in the forms just examined, ...'
Context:
'However, even if we grant that there may have been a Proto-
Indo-European *b in the forms just examined, we cannot fail to
note the striking quantitative discrepancy between forms with
Proto-Indo-European *b and those with *d and *g. Based on our
count of Pokorny, *d and *g occur over 250 times each in Proto-
Indo-European words.'
or: 'However, even granting that there may have been a Proto-Indo-
European *b in the forms just examined, ...'
Discussion: This example follows a discussion of the three known
cognate sets which may reflect Proto-Indo-European *b, during which it is
stated that all three are suspect. The three verbal nouns, 'admission',
'possibility', and 'presence', have affinities both to those which presuppose
and are definite, discussed in Section 3.1 above, and to those used for first
mentions, discussed in Section 3.2. There has been no explicit reference to
admission of the opposed viewpoint (namely that the sets are solidly Proto-
Indo-European), the possibility that the sets contain PIE *b, or the presence
of PIE *b in them; but of course these three things are very obviously im
plicit in the passage. The formula pri dopuscenii + verbal noun 'if we admit'
appears to be appropriate for introducing a discussion of the shortcomings
of an opposed view regardless of whether there has been prior reference to
the existence or the content of such a view.
3.8 Further evidence
Further support for the analysis given above comes from the English
writings of Roman Jakobson, which use the words and most of the grammar
(up to the sentence level) of English but rely on Russian expository devices.
For example (Jakobson 1971:135, boldface added):
2.51 E n E ns /E s ) EVIDENTIAL is a tentative label for the verbal
category which takes into account three events — a narrated
event, a speech event, and a narrated speech event (E ns ), namely
the alleged source of information about the narrated event. The
speaker reports an event on the basis of someone else's report
(quotative, i.e. hearsay evidence), of a dream (revelative
evidence), of a guess (presumptive evidence) or of his own
previous experience (memory evidence). Bulgarian conjugation
distinguishes two semantically opposite sets of forms: "direct nar
ration" (Ens = Es) vs. "indirect narration" (Ens =/ E s ). To our
question, what happened to the steamer Evdokija, a Bulgarian
first answered: zaminala "it is claimed to have sailed", and then
added: zamina "I bear witness; it sailed." (Cf. H. G. Lunt on the
systematic distinction made in the Macedonian verbal pattern be
tween "vouched for" and "distanced" events.)
This paragraph constitutes a complete subsection and is hence self-con
tained. The transition between 'Bulgarian conjugation distinguishes...',
which describes a verbal category, and 'to our question ...', which illustrates
its operation, is too abrupt in English (although its literal translation Na nas
vopros ... would be exactly right in Russian). This is because the nominaliza-
tion question introduces a new issue without identifying it as new. The effect
is to presuppose that the reader knows something that he does not in fact
know. The nominalization is not the sole problem with this sentence, in
which the initial To and the complementation after question calque Russian
422 JOHANNA NICHOLS
grammar, but is the source of the transition problem.5

Later in the same sentence, a Bulgarian is used to introduce a new par
ticipant, again in violation of the English new-participant constraint but
consistent with Russian usage. The effect in English is to suggest that
Bulgarians have been at issue before in the text. Overall, the use of Bulgarian
in this passage is analogous to the use of slova in (14) above. Both are con
crete nouns, but both illustrate the same problems as verbal nouns do.
Published translations from English into Russian provide more
evidence for differences in English and Russian nominalization. In the
following passage, a Russian verbal noun has been used to translate an
English infinitive.
(16) G and I p. 957, translating Kuhn 1970
Resenie otvergnuť odnu paradigmu vsegda predpolagaet
decision to reject one paradigm always presupposes
odnovremenno resenie o prinjatii drugoj ...
simultaneously decision about accepting other
(original) The decision to reject one paradigm is always
simultaneously the decision to accept another ...'
The infinitive prinjať 'accept' would have been perfectly grammatical in
place of the nominalization  prinjatii, as is shown by the parallel use of the
infinitive otvergnuť in the first clause (where an infinitive must be used
because this verb happens not to form a verbal noun). The choice of verbal
forms is determined by semantic factors: the infinitive is favored in contexts
where a specified subject actually performs an action, while the verbal noun
is favored where there is no specified subject and/or the action is not realis
in modal type.6 Hence the infinitive is required in (17), possible and better
than the verbal noun in (18), and less good than the verbal noun in (16).
(17) On rešil prinjať novuju paradigmu
*o prinjatii novoj paradigmy
he decided accept new paradigm
'He decided to accept (infinitive/*verbal noun) the new
paradigm'
(18) On prinjal resenie prinjať novuju paradigmu
?o prinjatii novoj paradigmy
he took decision accept new paradigm
'He made (lit. 'took') a decision to accept the new paradigm'
A semantic factor favoring the verbal noun over the infinitive where the sub
ject is unspecified and/or the verbal noun not referential is entirely consis
tent with the use of Russian verbal nouns in the absence of prior mention
or definiteness.7 The strength of this factor is shown by the fact that in (16)
it overrides the influence of the infinitive of the English original and the im
portant parallelism of the two clauses.
4 CONCLUSION
It has been argued above that Russian lacks grammatical devices which
monitor the speaker's assessment of the hearer's knowledge or experience,
and that even the abbreviation sr. 'cf', used frequently with verbal nouns,
makes a presupposition of truth or general knowledge rather than a presup
position about reader's knowledge. In both respects Russian differs from
English. Because English verbal nouns, or more precisely the articles they re
quire, add presuppositions about the reader's knowledge (including the
reader's knowledge of the text), they are often inappropriate translations of
Russian verbal nouns. The result is that English must use a finite verb form,
hence the pragmatics of assertion, where Russian uses a verbal noun which
carries very different pragmatics.
In English, a salient new participant or issue is best introduced in a
presentative or similar clause, which puts it into the scope of a separate ver
bal assertion; this seems to be true for Russian narrative but evidently does
not carry over to scientific prose, where a salient new issue can be intro
duced as an ordinary clause actant. Hence nominalizations, like any other
type of noun, can introduce salient new issues in Russian, while the best
English translations denominalize them so as to restore the separate asser
tion required by English.
Whatever the exact form of the information-incrementation rules of
English and Russian, it is clear that Russian nominalizations do not violate
this rule when their close English translations do. This particular difference
could be handled by a purely formal analysis: we could say that the Russian
nominalization is some kind of S (and hence constitutes an independent in
formation unit), while the English nominalization is some kind of Ñ or NP
(and hence is not an independent information unit). This formal difference
would allow us to state the information-incrementation rule identically in the
two languages: one chunk per clause (so the Russian nominalization, being
424 JOHANNA NICHOLS
a clause, can carry its own chunk). Alternatively, we could give a functional
analysis whereby English and Russian make use of different information
units: the clause in English, the intonation unit (smaller than a clause) in
Russian. The two analyses are not mutually exclusive. The functional one
receives support from the fact that, independently of nominalizations, Rus
sian intonation behaves very differently from English intonation, and does
indeed segment utterances into roughly phrase-sized information units.
In summary, the discourse factors of definiteness, treatment of new
participants, and information incrementation are responsible for differences
in use of nominalization in Russian and English. To at least this extent, then,
the choices as to which clauses will be nominalized and which will be finite,
and hence the choices as to which clauses will be asserted and which will not,
are not made simply on the basis of the message to be conveyed. They are
also influenced by the code used. (For message and code see Jakobson
1971:130-33.)
But if assertion, presupposition, and the like are imposed by the code,
they are nonetheless also part of the message conveyed. Thus the code deter
mines the message to some extent. This means that differences in the Russian
and English inventories of grammatical categories give rise not only to dif
ferences in the form taken by the message, but also to differences in the con
tent it conveys.
What has been said above explains why Russian can use nominaliza
tions in certain places. But it does not explain why it should choose to do
so in the following example:
(19) G and I p. lxxxiv
Odnostoronnosť i ograničennosť klassiceskogo
one-sidedness & restrictedness of classical
sravniteľno-istoriceskogo indoevropejskogo jazykoznanija
comparative-historical  linguistics
zakljucalis' v tom, cto rekonstruiruemaja model'
consisted in the fact that reconstructed model
obsceindoevropejskogo jazyka javij alas' lis' rezuľtatom
of PIE language was only result
vnesnego sravnenija ...
of external comparison
'(The) one-sidedness and restrictedness of classical historical-

comparative Indo-European linguistics lay in the fact that its
reconstruction of PIE was the result of exclusively external
comparison...'
or: '(The) one-sided and restricted nature of ...
Free translation: 'Classical comparative-historical Indo-Euro
pean linguistics was one-sided and restricted, since its recon
struction of PIE was based only on external comparison ...'
The sentence begins a paragraph which shifts to a topic not yet mention
ed in the book: the inadequacy of the received PIE reconstruction. No con
straints on definiteness, new participants, or information incrementation bar
the new topic from nominalization. But this is not merely one more new
topic; it is the fundamental thesis of the introduction and the reason for
writing the entire book. Here above all the English norm would call for
assertion. Yet the only mark of salience in Russian is the use of a position
of rhetorical prominence (paragraph-initial in a section-final paragraph).
The possibility of nominalizing this example is given by the grammar, but
the decision to nominalize it lies outside of grammar and within rhetoric and
the sociology of scientific communication. Writing in the Russian tradition,
the authors present their central claim not by making explicit its originality
and novelty, but by embedding it in a discussion proceeding from fundamen
tals, so that it is made to follow automatically from, and stand among, the
most elementary truths of linguistics. It thus becomes inevitable (in the sense
of Du Bois 1986 — who, however, is not describing expository prose) and
inextricable from general truth. This expository strategy is consistent with
the view of the Russian scientific text as gnomic rather than personal.
The grammatical and rhetorical differences between English and Rus
sian described above are entirely consistent with more general features to be
observed in the respective canons of scientific writing. If the phenomena in
vestigated here suggest that the Russian scientific text is a gnomic statement
of the knowledge available to the scientific tradition, this is also consistent
with its use of the depersonalizing editorial we, a fairly uniform style and
level, a favored expository strategy consisting of a statement or generaliza
tion followed by illustration, and such features of synthesis as lengthy
surveys of literature. If the English text is a communicative contract from
426 J O H A N N A NICHOLS
a personal writer to a personal reader, this is echoed in the fact that editorial
we is obsolescent, style is personal and variable, the favored expository
strategy is argument designed to lead the reader through a thought process
designed to inculcate certain opinions, intuitions often figure as objects of
analysis, and whole papers can be devoted to presenting hypotheses and
showing that they are appealing, rather than testing them. The sociology of
the possession and communication of knowledge is bound up with the dif
ferences in scientific texts produced in each tradition, and it is echoed in the
grammatical categories of each language. Thus the code and the coded
message combine to give a language-specific stamp to the broader endeavor
of scientific communication. The differences in Russian and English use of
nominalization and assertion are not caused by the broad cultural difference
(they are caused by language-specific inventories of grammatical categories),
and they cannot be shown to create it; but they are consistent with it.
ACKNOWLEDGMENT
Some of the translation utilized here was done in Tbilisi, Republic of Georgia as a participant
in the 1984 Exchange of Senior Scholars between the American Council of Learned Societies
and the Soviet Academy of Sciences, sponsored by the International Research and Exchanges
Board, New York. I am grateful to the Oriental Institute of the Georgian Academy of Sciences
and its director Thomas V. Gamkrelidze for facilities and hospitality. I thank Arkady Alexeev,
Boris Gašparov, Olga Hughes, and Igor Meľčuk for their discussion of these and other ex
amples. This paper has benefited from comments by Christian Lehmann, Richard Schupbach,
Sandra Thompson, and Kenneth Whistler, and my understanding of text structure has profited
from discussions with Boris Gašparov and Alan Timberlake. None of these people necessarily
endorses my entire analysis.
NOTES
For the abbreviations used in this article see list on page vii.
1. For purposes of pronominalization in this paper, I assume all speakers or writers are she
and all readers or hearers are he.
2. The following example from English linguistic writing illustrates this claim:
The unacceptability of (16b) supports the claim that that/those does not replace
just the plus contiguous material but only the plus N. (McCawley, 1985:12-36n)
The claim in question has not been made previously in the text; it is made only here, in
the complement clause to claim, *supports a claim that... is impossible for reasons that
apparently have to do with the author's presupposition of the truth of the complement
(rather than the audience's familiarity with it); the indefinite article becomes possible only
if the truth is disclaimed (would support a claim that...). In this example, the serves to
announce rather than presuppose, and it focuses on truth rather than knownness or iden-
tifiability. These differences from what is usually said about the presumably have to do
with the fact that the verbal noun claim can refer not only to a proposition but also to
an event.
3. In addition, such a survey would be meaningless because we have no comparable survey
of the syntactic functions of all Russian verbal nouns (the examples discussed here are a
small subset of the verbal nouns found in this text). A survey of all of them might well
prove interesting, however, because the syntactic relation of a verbal noun might affect
whether it could be in the scope of pragmatic operators on the main verb of its sentence,
which in turn might have some bearing on the relative felicity of assertion and nominaliza֊
tion in English translations.
4. In this respect the example is reminiscent of that discussed in note 2, where English
definiteness announces, rather than presupposing, the existence of the claim.
5. The sentence in question is of course not only awkward but incoherent as well: it implies
that the Bulgarian speaker freely used both evidential and non-evidential forms to refer
to the same event, which contradicts Jakobson's analysis of evidentiality. Jakobson's
question ('What happened...?') to the Bulgarian must have used the -/- form that is the
sole past tense of Russian but marks the evidential in Bulgarian. The Bulgarian's first
answer echoes the (unintended) evidential form, and the second is a correction.
6. To a native speaker consulted, the most salient difference is stylistic: the nominalization
in the second clause sounds more formal and is hence more appropriate for scientific
prose. I think this stylistic effect is the secondary result of a distributional fact:
unspecified subjects and irrealis verbs of relevant types are commonest in formal prose,
so the verbal noun they favor is associated with a formal style.
7. More accurately, the verbal noun is neutral in these regards and the infinitive requires
specification of its subject, either through syntactically controlled anaphoric binding or
— in the absence of discourse — through situational deixis. The following example can
be spoken by someone who has just entered a meeting and has no knowledge of the
preceding discourse:
Nu, cto — prinimať novuju paradigmu ili not?
well what accept-INF new paradigm or net
'Well, what do you think — shall we accept the new paradigm or not?'
Because there is no possible discourse deixis or syntactic control, the subject referent is
determined by the situational context. The fact that the infinitive carries obligatory sub
ject deixis makes it possible for a clause headed by an infinitive to constitute an indepen
dent utterance (as this example does). In contrast, the verbal noun carries no subject
deixis, and there is no way that prinjatie novoj paradigmy 'acceptance of (the) new
paradigm' can function as an independent sentence without the addition of a verb.
428 JOHANNA NICHOLS
REFERENCES
Benveniste, Emile (1959)1971. "Correlations of tense in the French verb".

Problems of General Linguistics, 153-61. Coral Gables: University of
Miami Press.
Du Bois, John W. 1986. "Self-evidence and ritual speech". In Evidentiality:
The Linguistic Coding of Epistemology, W.L. Chafe and J. Nichols,
(eds.), 313-33. Norwood, NJ: Ablex.
Foley, William Α., and Robert D. van Valin. Jr. 1985. "Information
packaging in the clause". In Language Typology and Syntactic Descrip
tion, I: Clause Structure, Timothy Shopen (ed.), 282-364. Cambridge:
Cambridge University Press.
Gamkrelidze, T.V., and V.V. Ivanov. 1984. Indoevropejskij jazyk i in֊
doevropejcy. Tbilisi: Tbilisi University Press and Georgian Academy of
Sciences. [English translation by Johanna Nichols: Indo-European and
the Indo-Europeans. Berlin: Mouton-De Gruyter.]
Givón, Talmy. 1984. Syntax: A Functional-Typological Introduction. Vol.
I. Amsterdam/Philadelphia: Benjamins.
Hetzron, Robert. 1975. "The presentative movement; or, why the ideal
word order is V.S.O.P." In Word Order and Word Order Change, C. Li,
(ed.), 346-88. Austin: University of Texas Press.
Jakobson, Roman. 1971. "Shifters, verbal categories, and the Russian
verb". In Selected Writings of Roman Jakobson II:130-47. The Hague:
Mouton.
Klenin, Emily. 1983. Animacy in Russian: A New Interpretation. Columbus:
Slavica.
Kuhn, Thomas S. 1970. The Structure of Scientific Revolutions. Chicago:
University of Chicago Press.
McCawley, James D. 1985. "More About English Syntax than you Probably
Want to Know." MS, University of Chicago.
Meľčuk, I.A. 1982. "Lexical functions in lexicographic description." BLS
8:427-44.
Nichols, Johanna. 1981. Predicate Nominals: A Partial Surface Syntax of
Russian. (UCPL 97.) Berkeley-Los Angeles: University of California
Press.
Timberlake, Alan. 1975. "Hierarchies in the genitive of negation." SEEJ
(19)2:123-38.
INDEX OF LANGUAGES
The indexes were compiled

by Cheng Luo
Abkhaz 186, 195 British 77

Adamawa languages 256 contemporary 77, 96
Akkadian 194 conversational 307
Arabic 79 modern 78, 94
Armenian 412 Old 410
Aztec 309 spoken 1-25, 153-4, 158
written 22ff., 275
Baliic 411 European languages 62
Bambara 184f Ancient 411, 417
Bantu languages 349
Bulgarian 421, 427 French viiff, 80ff, 84ff, 92, 96f, 203, 347,
351
Catalan 82-6, 96 early 82
Old 83 Middle 80ff
Cayuga 340-3, 347, 351 Old 80ff, 85
Celtic 411 spoken 135-75
Cherokee 353
Chinese 148, 153
Chipewyan 349 Gã 340
Choctaw 191 Gbaya 256
Chontal 352 German viiff, 90, 95f, 101-29, 145, 149,
Cretan 97 153, 173f, 188, 202, 209, 211, 285, 312,
410
Danish 96 Old High 410
Dutch ix, 101-29, 285, 312 Germanic 411
Ancient 410
English vii, viii, 56, 66f, 71ff, 81, 83ff, Godie viii, 220, 248-73
90, 106, 137-43, 145, 148f, 152, 154-5, Greek vii, 79, 145, 412, 418
158f, 161, 173f, 187f, 192, 196f, 202, (Classical 79
221, 240, 282, 285, 287, 310, 319ff, 347, modern 79, 97
351f, 356, 362f, 399-412, 414-26 Guarani 352
American 2, 153 Gurung 332, 335ff, 344
430 INDEX OF LANGUAGES
Hebrew Melanesian (languages) 343

Biblical 254 Meso-American languages 352
Hittite 184f, 188, 219, 411 Mohawk 346-7, 350-1
Hixkaryana 186
Hua x, 49-69, 231 Neo-Melanesian 242
Huave 352 Newari viii, 29-46
Huixtan Tzotzil 352 Nez Perce 45
Nguna 343-4, 347ff.
Indo-European 200, 202, 206, 220, 353, Nyabwa 250
357, 361,409, 413f,416ff.
Proto- 408, 411ff, 418f.,425 Occitan 83ff
Indo-Iranian 413f (dialects) 78, 82
Iranian 417 modern 84
Iroquoian languages 340, 346, 351, 353 Oneida 346
Italian x, 82, 85ff, 89f, 92, 95, 138-43,
161, 172f, 188, 191f, 201, 209f, 212 Papuan New Guinea
dialects 78, 83 languages 30, 49, 227
Italic 411 Parengi 333f.,337, 343
Italic-Celtic 412 Pascal 50
Portugese 78, 83, 85, 89f, 210, 212
Jacaltec 197-8, 338 medieval 94
Japanese viii, 162f, 170, 175, 196, 198, Provençal 95
200, 361f, 376-94, 396 literary 82
Old 83
Kaingang 254
Kamchadal 333ff, 338, 344f, 352 Quechua 191ff, 196, 198f, 201, 206, 219,
Kartvelian 419 352
Kathlamet 337-8, 354f
Kobon 184f, 189ff, 195f, 205-6 Rhetian (dialects) 78
Kru 256, 267 Romance languages ix, 71, 75-97, 148,
Kupia 353 163, 170, 199
Kwa languages 339, 347 early 80
modern 82
Lakhota 343 Proto- 78f
Latin viif, 79ff, 85ff, 92ff, 142, 184f, 187- Rumanian 78f, 82ff, 86f, 92
8,191,195ff, 210-13, 215, 221, 351, 410 Russian x, 142, 172, 399-427
Christian 79
Classical ix, 76-78, 93, 197 Sacapultec 135
Vulgar viiif, 137, 152 Sanskrit 203
Lithuanian 409f, 415 Sarcee 339, 349, 351
Luoravetlan languages 345 Scandinavian languages 102, 128
modern 108
Macedonian 421 Seneca 342, 346
Malayalam 221 Siouan (languages) 343
Maya (languages) 7 Slavic 411
INDEX OF LANGUAGES 431
Soddo viii, 361f, 365-76, 382, 396 Tequistlatec-Jicaque

Spanish 79, 82ff, 89f, 93, 137, 152, 171ff, languages 352
175, 352 Tibeto-Burman languages 30
American 95 Tiwi 344-5, 352
Argentinian 79 Tocharian 413f
contemporary 85 Tojolabal 352
medieval 94 Turkic
Old 78, 83 Proto- 417
Swahili 352 Turkish 195, 200
Swedish 108, 110
Vai 352-3
Tamil 219
Tauya x, 66, 227-46 Walbiri 184f
Telugu 395 Wobe 254
INDEX OF NAMES
(excluding self-citation)
Aissen, J., 220 Bowley, C C , 319

Akmajian, Α., 172 Brentano, F., 162, 173
Alcına, F.J., 86,90 Bresnan, J., 172
Algeo, J.E., 83,85,89,94 Brettschneider, G., 202, 210, 221
Alving, Hj., 108 Brockett, C S . 319
Andersson, L-G., 101, 110, 311ff, 318, Brøndal, V., 220
321 Brown, J.F., 319
Aze, R., 333ff Burnham, J.M. 72, 94
Aze, T., 333ff
Callow, J., 318ff
Bailard, J., 173 Carden, G., 322
Barnes, ., 172 Carvalho, P. de., 186, 200, 216, 221
Baschewa, E., 104, 107, 118f, 122, 124, Chafe, W., viiif, 275, 305, 307, 315, 337,
126 356
Battaglia, S., 89 Chomsky, Ν., 172
Beekman, J., 318ff Christmas, J.E., 353
Behaghel, vii Christmas, R.B., 353
Bennett, J., 73,93 Christel, Α., 208
Bentinck, J., 250 Clancy, P., 170
Benveniste, E., 400 Clark, E., 152
Bergqvist, B.J., 108 Cole, M., 353
Berkeley, 64 Cole, P., 201,206,352
Berry, M., 319 Coleman, R.G., 197
Berstein, ., 94 Comrie, ., 29, 173, 197, 221, 396
Biraud, M., 213 Cook, E-D., 339,349
Bird, C.S., 184 Cooreman, Α., 215
Blake, J., 240 Corominas, J., 79
Blanche-Benveniste,  , 172 Coromines, J., 83, 85
Blecua, J.M., 86,90 Craig, C G . , 197,338
Bloomfield, Լ., 300 Crothers, E., 320
Boas, F., 337,354 Crystal, D., 25
Bogoras, W., 345,351
Bolinger, D., 106, 142, 192, 316 Damourette, J., 172
Bolkestein, Μ.Α., 186,221 Danielsson, E., 115f
Bossuyt, Α., 107f Davies, J., 184f, 190
Bourciez, E., 79 Davison, Α., 101,110, 204, 221, 312, 318
434 INDEX OF NAMES
De Boor, ., 108 Green, G.M., 101, 106, 194, 220

de Dardel, R., 78,94 Greviss, M., 89f, 96
Debrabandere, L, 107 Grimes, J.E., 25, 268, 275, 318ff
Delbrück, ., 213
Deloria, E., 343 Haase, Α., 97
Den Besten, H., 129 Haegeman, L., 110
Dik, S., 195,220 Haider, H., 129
Dillon, G.L., 275, 313f Haiman, J., x, 18, 71f, 74,76, 91, 94, 101f,
Dittmar, N., 94 185, 188, 206, 215, 220, 228, 245, 251,
Dixon, R.M.W., 186,221 318
Downing, P., 170 Hale, Α., 33
Drossard, W., 202 Hale, K.L., 184f
DuBois, J., 135, 173,425 Hale, M., 33
Halliday, M.A.K., 2, 25, 91, 104, 275f,
Ellis, W., 290 280-4, 286, 298, 303-5, 313, 318ff
Emout, Α., 77,88,94 Hammarström, E., 107
Handke, J., 101,318
Farrell, J.G., 279 Harmer, L.C, 85
Faucher, E., 104f, 112, 115, 129f Harris, M., ix, 220
Fauconnier, G., 74, 95 Hasan, R., 2, 25, 275, 313, 318ff
Feeling, D., 353 Haudry, J., 185
Fillmore, C., 158f Hawkins, J., 173
Fleischman, S., 174 Helander, H., 200
Fleischman, ., 104 Henry, R., 341
Foley, W., 68, 102, 128, 152, 185, 189, Herman, J., 78,95
191, 193,197, 204, 206, 215f, 220f, 279, Hermodsson, L., 110f
311, 318f, 401 Hetzron, R., 161,174,401
Ford, , 307 Hewitt, B.G., 97
Franklin, K.J., 30 Hewitt, J.N.B., 346-7
Fries, P.H., 305,309,313 Hibiya, J., viii, 36, 185
Hinds, J., 268,317
Gamkrelidze, T.V., 400, 402, 406 Hobbs, J., 320
Geerts, G., 116 Hoff, F., 188,206
Geis, M., 119 Holton, D., 97
Genetti, C , viii Hooper, J., 106,280
Gerritsen, M., 108 Hopper, P., 45,362
Gildersleeve, B.L., 77,80 Horace, 95
Givón, T., 29, 36, 38, 46, 186, 206, 215, Horacek, ., 104, 107
221, 254, 312ff, 361, 374, 394, 401 Huddieston, R.A., 319,321
Glover, W.W., 332, 335f Hymes, D., 25
Goddard, P.R., 319
Golkova, E., 275 Ivanov, V.V., 400, 402, 406, 415
Gossen, G.H., 7
Gougenheim, G., 85 Jacobs, J., 120,124
Gratrix, C., 254 Jacobson, R., 421,424,427
INDEX OF NAMES 435
Jansen, F., 108,122 Mackenzie, J.L., 195,220

Jayaseelan, Κ.Α. 221 Mallinson, G., 240
Jespersen, ., vii, 68, 216, 279, 311 Mann, W.C., 290, 319f
Johnson-Laird, P.N., 110 Manoliu-Manea, M., 86f
Jorden, E.H., 396 Marchese, L., viii, 220
Marouzeau, J., 221
Kaufmann, G., 104, 125 Martin, J.R., 318,320
Kiparsky, P., 159,174 Martin, R., 80,82
Klare, J., 80f Marty, Α., 162, 173
Klavans,J., 172 Mathesius, V., 173
Klenin, E., 402 Matthiessen, C., ix, 91f, 102, 104, 186,
Koffka, ., 290 220f
Köhler, W., 290 McCawley, J.D., 426
Kolb, H., 105 McCray, Α., 318
König, E., ix, 71f, 74f, 77, 81, 83, 90f, Mchombo, S.A., 172
94, 220, 245 Mcintosh, Α., 319
Kopesec, M., 319f McKaughan, H., 30
Kroll, ., 25 McKeown, K.R., 320
Kruisinga, E., 318,321 McLendon, S., 7
Kuno, S., 163, 175, 395ff Meillet, Α., 214
Kuroda, S-Y., 162,175 Melander, J., 153, 172
Meľčuk, I.A., 404
Labov, W., 252 Metrich, R., 124,126
Lakoff, G., 194 Milewsky, T., 189, 197
Lakoff, R., 318 Miltschinsky, M., 95
Lal, J., 33 Mitchell, ., 72, 77, 93f, 96
Lambrecht, ., ixf Mithun, M., ix, 7, 221, 275, 346
Lehmann, C , ixf, 71, 74, 76, 80, 90ff, Muir, J., 319
102, 104, 127, 245, 251, 276 Munro, P., 206,318
Lenerz, J., 129 Myhill, J., viii, 36
Leslau, W., 373,396
Li, C , 142 Napoli, D.J., 154
Link, C , 254 Nichols, J., ix
Lisker, L., 395 Norton, F.J., 85
Lodge, G., 77,80 Noss, P., 256
Löfstedt, E., 209 Nseme, 252
Longacre, R.E., 110, 186, 207, 209, 211,
263, 268, 275, 280f, 286, 305, 307, 313, Okamoto, S., 175
315, 318ff Olson, M., 68
Lord, C , 220 Osborne, C.R., 344-5
Lunt, H.G., 421
Lyons, J., 310 Paul, H., 220f
Pawley, Α., 25
Macaulay, M., 172 Payne, J.R., 318
MacDonald, L., x, 66, 185 Pernicone, V., 89
436 INDEX OF NAMES
Peterson, M., 346 Suárez, J.Α., 353

Pichon, E., 172, 174 Svartvik, J., 15
Pike, E., 320 Syder, F.H., 25,356
Pike, ., 320 Szantyr, Α., 208,221
Pinkster, H., 185,206,211
Plato, 64 Taglicht, J., 124
Posner, R., 318 Talmy, L., 318
Prince, E., 173 Tao, L., 221
Pulte, W., 353 Tekavčič, P., 83,95
Thomas, F., 77,88,94
Quirk, R., 93,279,309,318 Thompson, S., ix, 18, 45, 91f, lOlf, 104,
106, 110, 142, 173, 186f, 197, 207, 209,
Ramsey, F., 66 211, 220f, 251f, 275, 280, 290, 305, 307,
Rando, E., 154 313, 315, 319, 362
Rappaport, G., 318 Timberlake, Α., 402
Reis, M., 129 Todd, T.L., 191
Reinhart, T., 146,252 Tomlin, R., 275
Rice, M., 347,350 Touratier, C , 220
Riverola, J.L., 83,85, 90
Rohlfs, G., 79,82,87
Vallette, 252
Rosén, H., 201
Van de Velde, M., 104
Rude, N., 45
Van der Auwera, J., ix, 91, 220
Rutherford, W., 318
Van der Horst, J.M., 107f
Van Es, G.Α., 123
Sapir, E., 135
Van Oosten, J., 149, 152
Sasse, H J . , 4, 175
Van Valin, R., 68, 102, 128, 152, 185,
Sandfeld, ., 71
189,191,193,197, 204ff, 215, 217, 220f,
Schelfaut, F., 104, 108, 115f
279, 311, 318f, 401
Schieb, G., 123
Vincent, N., 97
Schiffrin, D., 10f, 13f, 16, 307, 318
Von Wartburg, W., 80,96
Schmerling, S.F., 175
Schmidt, C., 108, 122
Schmitt Jensen, J., 89 Wandruszka, U., 163, 175
Schourup, L., 10, 14ff Wehr, ., 172f, 175
Schuetze-Boburn, S., 312 Wellander, E., 310ff, 321
-Schütz, A.J., 343-4,348-9 Weimers, W.E., 350,352
Schwyzer, vii Wendt, H.F., 195
Scott, F.S., 319 Wiesemann, 252, 254
Scribner, S., 353 Wilmet, M., 80,82
Seiler, H., 221 Winter, E., 275, 305, 312f
Slotty, F., 218 Woodcock, E.C., 76f, 80
Smaby, R., 318 Worth, D.S., 333ff, 338
Sopher, H., 318
Strawson, P., 66 Young, D.J., 319
Strevens, P., 319
Stump, G., 172 Zwicky, Α., 119, 172
INDEX OF SUBJECTS
ablative, 199, 202, 228 apposition, 181, 185, 282f, 300, 354
absolutives, 29ff, 34, 39, 44, 46, 195, articles, 145,401,408
197, 219 aspect, 34, 185, 193, 195, 200, 204f, 248,
accusative, 196, 198f 366, 381f
actant, 193, 197, 204ff, 423 imperfect, 158, 364f, 368, 370f, 375,
activation, 144-7, 169 381f, 387f, 392ff
additive, 351 perfective, 30, 36, 248f, 255, 257,
adjoined clause, 185f, 188 264, 363f, 367f, 370f, 373ff, 381f,
adjunct, 197, 214, 264, 280 388, 390ff
benefactive, 205 pluperfect, 158
predicative, 206 progressive, 30, 34
privative, 205 aspect switch, 368, 370f, 373f
adposition, 197f, 201 aspectual particle, 348
adverbial, 89, 149, 151f, 181, 199, 205, assertive, 193f, 229, 399-426
212, 216f, 280, 286, 340f, 343ff, 347, asyndesis, ix, 210-4, 216, 218
355 attributes, 181, 199
discourse, 346, 348, 351, 356 auxiliary, 31, 197, 201
locative, 137 conditional, 248
sentence, 344 future, 248
time, 409 negative, 248
adverbial clause, 5, 33, 71, 103, 105f, perfect, 248, 263
109f, 122,128,190, 211, 213, 219, 279f, sequential, 247ff
285f, 309, 312f
afterthought, 4ff, 9, 24, 341 background(ing), 36, 293ff, 297f, 307,
agent, 136 361f, 374, 383, 386f, 390f, 393ff
agentive subject 135 features of, 362, 390f, 395
agreement, 340 backgrounded clause, 366, 373, 376, 381,
anacoluthon, 105, 115, 128 383
anaphora, vii, 112ff, 147, 205, 244, 313 benefactive, 231
anaphoric pronoun, 198, 211, 243f
andatives, 335 cadence, 1f
anticipatory subject canonicity, 109f
desinence, 50, 52 case marking, 29, 32, 34-9, 44, 46
antithesis, 56f, 67, 295-7, 316 categorical judgement, 162
antithetical adversative, 57, 67, 125 causal clause, 71f
apodosis/consequent, 50, 66, 71, 106f, causatives, 192, 196, 201f, 215
109, 112ff, 118ff, 128 central position, 186
438 INDEX OF SUBJECTS
circumstance (clause), 277, 284, 294-5, conditional, 76ff

297, 305f, 308, 315 coordinative, ix, 215, 331, 336, 339,
clause chaining, viii, , 29f, 34, 46, 49f, 341ff, 346ff, 350ff, 356
185, 361-95 subordinative, 185, 193, 215
discourse function of, 361-95 temporal, 76f, 85, 90
forms of, 363, 365 connectives, viii, 2, 6ff, 22f, 71, 105,
clause integration, 101-29 210ff, 216, 277, 304f, 345
clause linkage, 182, 189, 192, 203, 206, concessive, 72
214, 216-9 coordinating, 10-17
typology of, 218-9 subordinating, 17-21
cleft sentence, 135-71,201,217 constraints, viii, , 146, 148, 151, 154,
climax, 263ff 186, 194, 200, 303
cohesion, 209, 211 definiteness, 425
comitative construction, 338f, 340, 343, information increment, 405, 409, 425
348ff, 354 information structure, 143, 153
complement clause, 30, 105, 203, 206, new-participant, 404, 410, 419, 422,
227ff, 234-9, 240-1, 243f, 365 425
completive particle, 344 pragmatic, 153
concessive (clause), ix, 71-93, 102f, 106- word-order, 142
10, 127ff, 183, 277, 294, 297 contrastive stress, 124
antecedent, 72ff, 76f, 81, 89f, 92 control, 186, 206f, 215f
canonical, 124-5 control verb, 207f
consequent, 72ff, 81, 90, 92 coordinate clause, 229f, 233, 240, 243, 253
factual, 77, 86, 88 ccordination, viiff, 11, 18, 78, 91, 93,
non-canonical, 125-7 101f, 181f, 240, 276, 281ff, 300, 303,
conditional (clause), ix, 50ff, 59, 61, 65ff, 309-10, 316, 331-57
71, 73f, 88, 90f, 102f, 106-10,127ff, 158, clause, 335-6, 344-9
188, 254, 277, 283, 295-7, 303-4, 399 noun phrase, 332-3, 336-43, 350f
canonical, 111-7, 122 predicate, 333-5, 343-4
canonical concessive, 118-22 copula, 148
causal, 68 coreferentiality, 29, 31f, 39f
concessive, 66, 68, 71ff, 76, 78, 85ff, correlative clause, 186, 188, 205
103, 106-10, 117-23, 126ff correlative diptych, 185, 205, 218f
counterfactual, 52, 158 cosubordination, 68, 185, 191
given, 52
hypothetical, 52 dative, 152, 197, 202
non-canonical, 116-7, 117 declarative, 115, 314
non-canonical concessive, 122-3 definiteness, 145, 148, 153f, 156, 250,
conjunct participle, 185, 195, 206 401f, 404f, 407, 414ff, 420, 423
conjunction, viii, 6, 32, 74, 80, 84, 86ff, deictic adverb, 137
92, 103, 109, 111, 123, 126, 188, 198, deictic pronoun, 147, 211
210, 212f, 216, 240, 254, 304, 331, 336f, deletion, vii, 232
339-41, 342ff, 349ff, 356, 363, 365 dependency, x, 102, 181ff, 194, 243
clause, 345 desententialization, 182f, 193-200, 204f,
concessive, 75ff, 79, 82ff, 90, 93 213ff, 217f
INDEX OF SUBJECTS 439
desideratives, 202-4, 215 of coordination, 331-57

different subject, 36, 49, 206f, 215, 249, of Nucleus-Satellite
252, 366, 368f, 371f, 374, 376f, 381, relations, 307f, 317
383, 386, 388, 391, 394f of rhetorical units/
direct quote, 59, 63ff, 283 organization, ix, 299, 301f, 309
discourse particles, 353 of serial verbs, 191
disjunct, 80 habitual aspect, 192
dummy subject, 137 human subject, 363, 383
hypotaxis, vii, 92, 101, 104, 106, 182,
elaboration, 288, 290f, 294-5, 297, 307 185, 205, 210, 275, 282f, 287, 290f, 300f,
ellipsis, 90, 104, 313 303f, 307ff, 312ff
embedding, vii, ix, , 92,102ff, 127,182, elaborating, 283
214, 217f, 276, 278-82, 284ff, 300, 303, enhancing, 91, 93, 283-4, 286, 299,
310-12, 317, 425 301ff, 305, 308, 311
enablement, 290-1 extending, 283
endocentric construction, 104, 181, 301 grammatical marking of, 304-5
episode boundaries, 8, 25, 33, 35, 37-8
ergatives, viii, 29ff, 34ff, 39, 46, 231, 236 iconicity, 50, 116, 128
event-reporting, 160-4, 171 tense, 229f, 243
existential sentence, 18, 138, 148f, 154, identifiability, 144-7, 169
159 illocutionary force, 101, 193f, 200, 314-5
exocentric construction, 301 imperative, 53, 122
implicative verb, 186
focus, 119,124,138,140ff, 149,155,163ff, inconsequential clause, , 51-69, 227ff,
168f, 250, 264, 266, 340 232-45
focus particle, 107, 119ff, 126, 128 independent clause, 68, 91,189,193,195,
focus pronoun, 249-50 227, 239f
foreground(ing), viii, 36, 104, 249, 264, desinence of, 51
266, 361f, 374ff, 383, 390f, 394 indicative, 51, 76f, 87ff, 114, 158, 194,
features of, 362f, 366, 373, 375f, 228, 399
381f, 390ff future, 51
foregrounded clause, 366, 368, 376, 381, infinitive, 85, 192, 203, 207ff, 213, 219,
383, 395 304, 365, 422f
information increment, 401f, 404, 423f
gapping, 204-5 instrumental, 231, 339
genitive, 196f interrogative, 75, 229, 231, 284f
gerund, 63ff, 67, 198f, 212, 376-91, 394 interrogative pronoun, 232, 234, 237, 238f
given (information), 67f, 312ff intonation, 2, 6, 101, 124, 139, 142, 192,
government, 185f, 190, 192, 197, 200f, 331-7, 342ff, 353f, 356, 402, 424
215 comma, 6, 9-11, 13, 17, 19, 22f, 25,
grammatical operator, 214f 332ff
grammatical relation, 181 period, 6ff, 13, 17, 19, 24f
grammatic(al)ization, viii, , 88f, 136, intonation break, 333, 335
183, 201,214f, 217f, 275f intonation peak, 1, 7, 155
of conjunction, 336 intonation units, 334, 337, 356, 424
intransitive clause/verb, viii, 29, 31f, 34ff, nominatives, viii, 196,198, 236f, 239, 244
39f, 44, 139f non-agentive subject, 152
inversion, ix, 90,140, 142f, 148,163,170f non-compositionality, 159
non-control verb, 208
jussives, 76, 83, 88 non-dependency, 181
juxtaposition, 127f, 156, 337f, 343ff, 349, non-final clause/verb, 30f, 36
352f, 355f normative grammar, 89
nucleus, 92, 288, 290, 293, 299, 300,
left-dislocation, 104, 136, 155, 166f, 187, 302ff, 309, 315
214, 227, 237-45
locatives, 39f, 44, 152 object clause, 105, 185, 199, 206f, 215,
282
main clause, x, 5,18, 31f, 68, 71ff, 76, 78, object complement, 59, 61
81, 91f, 115, 127f, 138, 157, 160, 183, oblique cases, 196
186ff, 190, 194, 204f, 208, 213f, 216, old/known information, 312ff, 362, 404f,
227f, 243f, 248, 251, 255, 311, 313, 399, 415
401 optative construction, 122
main clause phenomena, 101, 106
main verb, 185f, 192, 203f, 209, 214f parataxis, vii, 66, 78, 80, 82, 89, 92f, 101,
marginal position, 186f 127, 182, 184f, 205, 210f, 214, 282ff,
markedness, 109ff, 116f, 119, 124f, 128, 300ff, 316, 352
140, 147ff, 171,299 asyndetic, 182
matrix clause, ix, 186,193, 201, 206, 209, syndetic, 182
213, 216f passive, 195
medial clause, 36, 49ff, 58, 60, 66, 189, pause, viif, 1, 8f, 22, 238f, 243ff, 262, 332,
195 334,341,345,356
medial desinence, 50 pitch, 8f, 14, 138, 332, 356
medial verb, 49, 58, 64, 185, 190, 205-6 falling, 4, 6, 9, 344
coordinate, 49, 54, 67f rising, 9, 344
subordinate, 49, 51f, 67 plurality, 339
modal, 87ff, 197, 200f, 203, 219, 242 polarity, 67, 194, 197, 200
mood, 67, 76, 80, 87, 89, 101, 185, 1931՝, positional variability, 186
200, 233, 243, 314 possesives, 142, 151, 159, 196
irrealis, 158 pragmatic accessibility, 165-6, 169f
realis, 422 predication, 125, 158, 182, 190,195,197,
motivation, 292-3, 297f, 306 218
present participle, 364
negative, 248, 264 presentational structure, 146-56, 160ff
new information, x, 33, 147f, 155, 163, presentative construction, 401f, 423
250, 312, 314, 346, 401f, 404 presupposition, 33, 61, 120, 124, 137,
new participants, 401f, 422ff 139, 141, 149, 155, 160, 164, 313, 399,
nominality, 197f, 200, 214 401f, 404f, 410, 416-8, 420f, 423f
nominalization, x, 193, 197, 201, 214, prolepsis, 208f, 215
218f, 280, 309, 365, 399-426 pronominalization, 316-7
INDEX OF SUBJECTS 441
backward, 316-7 215, 249, 252, 363f, 366, 368f, 371ff,

proposition, 67, 72f, 80, 106, 110, 112, 377, 381f, 391f, 395
120,128,137,139,141,146ff, 151,155, satellite, 92, 288, 293, 299, 300, 302ff,
161ff, 167ff, 182, 204, 207, 209, 211, 310, 312, 314f
277, 295, 298, 315 sententiality, 200
interlacing of, 204, 214ff sequencing, 33, 35-7, 186, 247-66, 375f,
protasis/antecedent, 50, 52ff, 58, 66f, 71, 389-90
106f, 112ff, 119ff, 128 chronological/temporal, 249, 363f,
purpose (clause), 187f, 207, 277, 283, 366f, 374, 386, 394
295-7, 304, 306f, 313, 355 serial verbs, 191, 195f, 216, 334
sociation, 181ff, 194
quantifier, 8 solutionhood, 295-6, 297, 306
stress/accent, 140, 142f, 147
raising, 208f, 215 subject clause, 105, 198, 215, 282
reason (clause), 277, 282f, 288, 309 subject continuation, 364, 366, 380f, 393f
redundancy, 149 subject focussing, 251
referentiality, 146, 423 subject identity, 207
register, 89, 92, 138 subject prominence, 142
relative clause, 5, 18, 20-21, 25, 65, 141, subject switch, 371ff, 377, 386f, 394
148,155,158,185,187f, 190,198f, 227- subjectivity, 63ff
9, 251-3, 254f, 285, 365 subjunctive, 68, 76, 80, 87ff, 112f, 194
headless, 207 concessive, 80f, 84, 89
non-restrictive, 194, 283 future, 51
restrictive, 141, 158, 279, 282, 287 jussive, 83
relative pronoun, 21, 23, 25, 137, 153, subordinate clause, ix, 19, 32, 51, 68, 71f,
156, 199, 212 102, 106, 115, 127, 138, 157, 160, 183,
repetitive clause, 252, 264, 266 194f, 197ff, 204ff, 211, 216, 218, 227-
reported speech, 283 30, 233, 236, 239-40, 243f, 248, 250ff,
resumptives, 103, 116, 237ff, 241ff 254,262,275,285f, 310ff, 314, 317, 355,
rheme, 195 361, 419
rhetorical nesting, 304 hierarchical downgrading, 183-91,
rhetorical prominence, 425 194, 214f, 217, 251
rhetorical relations, 289, 298, 300ff, 317 integration of, ix, 101-29,183-92, 214f
elaboration, 297f subordination, viiff, 18f, 80, 87, 91, 101f,
enhancing, 297f, 301, 303 104, 106, 115f, 181, 192, 203, 210, 214,
listing, 288f, 297, 300f, 308 227-37, 251, 275, 282, 305, 309-12, 361,
nucleus-satellite, 289f, 297ff, 305, 374
308, 316f substitution, 313
typology of, 297-8 superordinate verb, 201, 203, 208f
rhythm, 356 switch reference, 49f, 206, 215, 249, 331
root transformation, 106 syndesis, 210-4
syntactic level, 189-92, 214
saliency, 161, 401, 405, 423, 425 syntagm, 181ff
same subject, viii, 30f, 34, 36, 54, 206ff, subordinate, 183, 186, 190, 209, 213
syntagmatic interweaving, 91, 208 topicality, 29, 33f, 38-44, 46

systemic grammar, 280 topicalization, 251
transitive clause/verb, viii, 29, 31, 34ff,
tagmemic grammar, 280 39, 135
temporal (clause), 206, 251, 254f, 263, transitivity, 33f, 37, 44, 46, 136, 284
277, 284, 295 V-2 feature, 102-6, 108, 115
tense, 67, 101, 156, 185, 193, 195, 200, valency, 202, 206, 216
204f, 233, 243, 248, 251, 260, 331 verb serialization, 190, 219
aorist, 229 verbal noun, 407-14, 417ff, 422f
future, 51,229 V-late placement, 102,105,115,117,123
present, 137, 157 vocative, 53
past, 157ff, 248 volitives, 76,79,81,83,88
remote, 257, 259f volumn, 8f, 14, 356
tense concord, 157
text relations, 287-309 wh-clause, 285
thematic clause, 313 wheel-spinning, 7ff, 24
theme, 139f, 195 word order, 30, 62, 91, 102ff, 114f, 117,
thetic judgement, 162f 119, 127, 139, 142f, 195, 248, 251, 257,
tone, 124, 138, 140, 155, 265 266, 284f, 312
topic, viii, 15,29, 33, 45,103,135f, 138ff, choice of, 109-11, 124
143, 146ff, 153, 155f, 161ff, 187f, 231f, integrative, 107f, 111, 115, 117ff,
234f, 238f, 243, 249-50, 255, 337f, 362, 122ff, 126ff
405, 425 non-integrative, 107f, 111ff, 115ff,
anaphoric processing, 43 126ff
cataphoric processing, 43, 46 resumptive, 107f, 111, 115, 117f,
topic accessibility, 149, 151, 153 121f, 124, 126f
topic continuity, 33, 38-9, 43-4

Clause Combining in Grammar and Discourse

Uploaded by

Copyright:

Available Formats

You might also like

Clause Combining in Grammar and Discourse

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Clause Combining in Grammar and Discourse

Uploaded by

Copyright:

Available Formats

CLAUSE COMBINING IN GRAMMAR AND DISCOURSE

TYPOLOGICAL STUDIES IN L A N G U A G E (TSL)

A companion series to the journal "STUDIES IN L A N G U A G E "

Honorary Editor: Joseph H. Greenberg

Volumes in this series will be functionally and typologically oriented, cove-

John Haiman and Sandra A. Thompson (eds)

CLAUSE COMBINING IN GRAMMAR AND DISCOURSE

JOHN BENJAMINS PUBLISHING COMPANY

List of abbreviations vii

The discourse function of clause-chaining 361

ABL ablative DAT dative

BEN benefactive FEM feminine

INDEF indefinitive PP past participle

MASC masculine Q interrogative

NEG negative RECIP recipient

Traditional and modern grammarians alike have restricted what they

course. Mithun shows that the coordinating conjunction, both between

tically well-formed but pragmatically unacceptable monoclausal W P VP

provide a field of inquiry where functional motivation for grammatical

Intonation units and clauses

When one listens carefully to spoken language, or even when one

are being focussed on at different times, but also — of special interest to us

Linking with intonation alone

There is a major distinction to be made between intonation units ending

Afterthoughts. Quite often in spoken language a period intonation will

Wheel-spinning. A second, not at all rare situation in which one intona­

spinning that occurs often and unremarkably in colloquial speech in examples

Crossing episode boundaries. In the next example, intonation units (b)

Comma intonation: more to come. The consistent function of the falling

Linking with 'coordinating' connectives

If temporal sequencing is the most common relation associated with the

common use of or in spoken English, as also in:

(29) a. ... I've had headaches maybe twice before in my life.

like. This word, frequently used in spoken English as a hedge, is also

well. This connective signals a shift in the flow of information, a partially

c. ... well I wasn't really much at all.

then. Sometimes then as a connective straightforwardly signals temporal

returns to the topic of fainting with:

oh. As a connective oh functions similarly to to be sure: "what follows is

Linking with 'subordinating' connectives

data, constituting only 3.5% of all the linkages examined.

We began by looking at cases in which the linkage between two intona­

A comparison with written English

One of my purposes in constructing this tentative picture of linkages in

d. a subject might initially use each property separately to try to

We have seen that the flow of ideas is manifested in spoken language in

and episode boundaries. In contrast, a comma intonation at the end of the

. 1987b. "Punctuation and the prosody of written language." Technical

This paper examines an alternation in case marking of subject NPs in

Newari is an ergative language with dominant SOV word order although

This is consistent with my informant's intuition about such sentences,

Examples were chosen based on the criteria mentioned in the discussion

Table 1. Number of clauses in examples

Case marking and position

Table 2. Case marking on subject NP

Table 2 shows that the distribution of ergative and absolutive morphemes

Table 3. Position of clause controlling case marking

Table 4. In-sequence/ out-of-sequence

Table 5. Distance from episode boundaries

Wheel-spinning. A second, not at all rare situation in which one intona

We began by looking at cases in which the linkage between two intona

coordinate or same-subject coordinate, are obligatorily followed by an an

quential of (27b), depending on whether the quote was presented as a subjec