Vous êtes sur la page 1sur 485

Cause - Condition - Concession - Contrast


Topics in English Linguistics

Bernd Kortmann
Elizabeth Closs Traugott

Mouton de Gruyter
Berlin · New York
Cause - Condition - Concession - Contrast
Cognitive and Discourse Perspectives

Edited by
Elizabeth Couper-Kuhlen
Bernd Kortmann

DE Mouton de Gruyter
G Berlin' New York 2000
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter GmbH & Co. KG, Berlin.

@ Printed on acid-free paper which falls within the guidelines of the

ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication-Data

Cause, condition, concession, contrast : cognitive and dis-

course perspectives I edited by Elizabeth Couper-Kuhlen,
Bernd Kortmann.
p. cm. - (Topics in English linguistics; 33)
Includes bibliographical references and index.
ISBN 3-11-016690-9 (alk. paper)
1. Grammar, Comparative and general-Clauses. 2. Gram-
mar, Comparative and general-Syntax. 3. Cognitive gram-
mar. 4. Discourse analysis. I. Couper-Kuhlen, Elizabeth.
II. Kortmann, Bernd, 1960- III. Series.

Die Deutsche Bibliothek - Cataloging-in-Publication-Data

Cause - condition - concession - contrast : cognitive and

discourse perspectives I ed. by Elizabeth Couper-Kuhlen ;
Bernd Kortmann. - Berlin ; New York : Mouton de Gruyter,
(Topics in English linguistics; 33)
ISBN 3-11-016690-9

© Copyright 2000 by Walter de Gruyter GmbH & Co. KG, 0-10785 Berlin
All rights reserv~d, including those of translation into foreign languages. No part of this
book may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopy, recording, or any information storage and retrieval
system, without permission in writing from the publisher.
Printing: Werner Hildebrand, Berlin - Binding: Liideritz &. Bauer GmbH, Berlin.
Cover design: Christopher Schneider, Berlin.
Printed in Germany.

Introduction 1
Elizabeth Couper-Kuhlen and Bernd Kortmann

I. Cause

The relevance of causality 9

Paul Georg Meyer

On the processing of causal relations 35

Leo G. M Noordman and Femke de Blijzer

Domains of use or subjectivity? The distribution of three 57

Dutch causal connectives explained
Henk Pander Maat and Ted Sanders

Causal relations in spoken "discourse: Asyndetic constructions 83

as a means for giving reasons
Christine Gohl

II. Condition

Constructions with if, since, and because: Causality, III

epistemic stance, and clause order
Barbara Dancygier and Eve Sweetser

On affirmative and negative complex conditional connectives 143

Estrella Montolio

Pre- and post-positioning of wenn-clauses In spoken and 173

written German
Peter Auer

Counterfactual reasoning and desirability 205

Noriko McCawley Akatsuka and Susan Strauss
vi Contents

III. Contrast

Adversative connectors on distinct levels of discourse: A 235

re-examination of Eve Sweetser's three-level approach
Ewald Lang

Viewpoints and polysemy: Linking adversative and causal 257

meanings of discourse markers
l~Cott A. Schwenter

The treatment of contrasts in interaction 283

Cecilia E. Ford

IV. Concession

Concessives on different semantic levels: A typological 313

Mily Crevels

Causal and concessive clauses: Formal and semantic relations 341

Ekkehard Konig and Peter Siemund

Concession implies causality, though in some other space 361

Arie Verhagen

Concessive patterns in conversation 381

Elizabeth Couper-Kuhlen and Sandra A. Thompson

"that's true, although not really, but still": Expressing 411

concession in spoken English
Dagmar Barth

From concessive connector to discourse marker: The use of 439

obwohl in everyday German interaction
Susanne Giinthner

Index 469

Authors' addresses 473


The theme of this book-endearingly referred to by its editors and

contributors as the CCCC or four-Cs volume-is a set of relations,
conceptual in nature but instantiated linguistically, which can be said to
hold typically between clauses or sequences of clauses in discourse.
Most, if not all, of our contributors will undoubtedly agree that each of
these relations can be realized or marked by different linguistic means,
e.g. by adverbials, particles, coordinating and subordinating conjunctions,
word order (see, for instance, the chapters by Barth, Dancygier/Sweetser,
Gohl, Montolio and Pander Maat/Sanders, which compare and contrast
different lexical connectives in the fields of causality, conditionality and
concessivity). Some of our contributors will undoubtedly maintain that
the C-relations can even hold in the absence of specific lexical or
syntactic markers (see, for instance, the chapters by Crevels, Gohl and
Meyer). Yet many of our contributors may disagree on whether these
relations are fundamentally semantic (see, e.g., the contribution by
Konig/Siemund) or fundamentally rhetorical/interactional in nature (see,
e.g., the chapters by Barth, Couper-KuhlenfThompson, Gohl and Meyer).
Those who think of the C-relations as semantic will be concerned to
represent their meaning (or the meanings of their markers) in a context-
independent, perhaps even a formal fashion, whereas those who see them
as basically rhetorical or interactive will address their interpretation (or
the interpretation of their markers) in specific-although perhaps
generalizable-contexts. Yet in this divergence-or rather, diversity-of
opinion we see one of the strong points of our endeavor. Indeed, the
novelty of the present volume lies not only in the cutting-edge research
which it presents but also in the fact that it embodies work at the frontier
of two very different approaches to language-cognitive linguistics and
discourse or interactional linguistics. By bringing these two traditions
together in one volume, we hope to initiate a dialogue in which the
respective bodies of work can be evaluated for their relevance to one
The contributions collected here have been grouped roughly into
sections according to C-relation in the order: cause, condition, contrast,
concession. However, since some chapters explicitly address the
relationship between more than one relation, the section boundaries are by
no means rigid. In fact, this permeability is a reflection of deep-lying
2 Elizabeth Couper-Kuhlen and Bernd Kortmann

affinities and wahlverwandtschaften between the C-relations, which make

themselves apparent in both cognition and discourse.
Cause, condition and concession have long been known to have a
special relationship with one another as adverbial relations of
circumstance (cf e.g. Halliday 1985, Thompson/Longacre 1985,
Kortmann 1997). At times they have even been conceptualized in terms
of one another, a conditional relation being seen e.g. as a hypothetical
variant of a causal relation, a concessive relation as an inoperant cause
(Harris 1988, Konig 1986, 1988). The conceptual similarity between
cause, condition and concession is also reflected in the fact that languages
may encode them in the same way (Kortmann 1997). All three relations
lend themselves to expression via syntactic subordination; moreover,
single subordinators may be polysemous between cause, condition or
concession (e.g., if can express both condition and cause, for both cause
and concession; see the chapter by Konig/Siemund in this volume).
Dancygier/Sweetser (this volume) explore how this functional overlap
can be accounted for with respect to the connectives if, since and
The relation of contrast, on the other hand, is typically expressed by
syntactic coordination and not unanimously thought of as an adverbial
circumstantial relation (Halliday 1985, Thompson/Longacre 1985). Yet
especially when it is considered from a pragmatic and/or discourse
perspective, contrast begins to have more in common with concession
(see e.g. Rudolph 1996) and even with cause and certain kinds of
condition. Sweetser (1990), for instance, applies her three-domain model
not only to causal, conditional and concessive sentences but also to
adversative sentences with the coordinator but (see Lang, this volume).
Moreover, as several of the discourse-based papers in this collection
argue, contrast is centrally implicated, e.g., in counterfactual
conditionality (Akatsuka/Strauss) and in concession (Barth, Couper-
Kuhlen/ Thompson). Contrast, specifically adversativity, is furthermore
involved in the discourse-marker use of Spanish si, a canonical
conditional conjunction (Schwenter) and in the discourse-marker use of
German obwohl, a canonical concessive conjunction (Giinthner). Finally,
contrast enters into wahlverwandtschaften with circumstantial C-rela-
tions, in particular with causality, in that justifications recurrently follow
contrasts in discourse, as Ford (this volume) shows. These findings
suggest then that the affinity between contrast on the one hand, and cause,
condition and concession on the other, has been underestimated in the
past. For this reason contrast has been included as a fourth C-relation
Introduction 3

Despite the division of our collected papers into sections, there are-
as might be expected-recurrent themes which cut across the C-relations.
One of the motifs in the cognitively oriented papers, for instance, is
Sweetser's (1990) seminal work on domains or levels of interpretation.
Noordmanlde Blijzer (this volume) adopt her distinction between content
and epistemic level and show how it will help account for differing
degrees of cognitive complexity in causal sentences. Other papers are
more critical of Sweetser's model: Lang (this volume) takes issue with
her claim that content, epistemic and speech-act levels of interpretation
can be determined independently of syntactic structure and, like Crevels
(this volume), fmds it necessary to supplement Sweetser's three domains
with a fourth, textual level. Given a fourth level, Crevels fmds the model
useful for investigating the formal means which typologically different
languages deploy in each domain. Pander Maat/Sanders (this volume), on
the other hand, reject Sweetser's model as a means of accounting for the
distribution and interpretation of specific causal markers in Dutch and
propose in its stead a notion of subjectivity.
Another recurrent motif in several of the chapters collected here is
Fauconnier's theory of Mental Spaces (1985, 1997). Dancygier/Sweetser
(this volume) show that important distinctions can be drawn based on the
ways in which causal and conditional conjunctions participate in the
configuration of mental spaces. Verhagen (this volume) uses a mental-
space model to account, e.g., for the relationship between concession and
(negated) causality, with which Konig/Siemund (this volume) are also
concerned. Yet AkatsukaiStrauss (this volume) are critical of a mental-
space account, because it neglects what they see as an inherent dimension
of counterfactual conditionality, namely that speakers are expressing a
stance towards the events in question as desirable or undesirable.
AkatsukaiStrauss advocate a subjective dimension in the analysis of
counterfactual conditionality and in this sense are in agreement with
Pander Maat/Sanders, who see subjectivity as responsible for the
behavior of specific causal connectives in Dutch discourse. It is perhaps
worth noting that both AkatsukaiStrauss and Pander Maat/Sanders take a
corpus-based approach to C-constructions and for this reason belong
more in the discourse than the cognitive tradition. Significantly, it is in
relying on real data rather than on introspection that these two studies
independently establish a need-above and beyond mental-space or
domain considerations-for taking the speaker's (or subject of conscious-
ness's) stance into account in the analysis of C-constructions, especially
for counterfactual conditionality and causality. In this sense they are
initiating the kind of dialogue between cognitive and discourse
4 Elizabeth Couper-Kuhlen and Bernd Kortmann

approaches to the C-relations which the editors have envisaged (cf. also
the chapters by Meyer and Konig/Siemund in this respect).
A second notion to be found in both cognitive and discourse-oriented
papers collected here is polyphony (Verhagen), or multiple viewpoints
(Schwenter). For Verhagen, the interpretation of concessive and negated
causal sentences as well as of all sentences involving epistemic C-
relations calls for the construction of two nearly similar mental spaces in
Fauconnier's theory-or, equivalently, for the assumption of polyphony in
Ducrot's sense of the term (1984, 1996). Schwenter also mentions
Ducrot's theory of polyphony as one way of conceptualizing viewpoints
and linguistic structure. He opts, however, for Roulet's (1984) distinction
between one viewpoint (monological) and two viewpoints (dialogical),
both of which can be found with either one speaker or two. Schwenter
finds that dialogicality is not only conceptually present in adversative
constructions but also (to a lesser degree) in 'exhortative' and epistemic
causals (i.e., ones which accompany exhortative speech acts and
inferential conclusions). The notion of polyphony or multiple viewpoints
then appears to be relevant to both cognitive and discourse-based
understandings of the C-relations. And it is clearly implicated in the
interactional model of concession which Couper-KuhlenfThompson (this
volume) propose, where multiple viewpoints often go hand in hand with
multiple speakers-although, as Barth (this volume) shows, they need not
do so.
Related to the dimension of multiple speakers is of course the question
of whether a piece of discourse-specifically, a C-construction-is
spoken or written. This is a recurrent motif in several of the discourse-
oriented papers collected here. Aside from questions of overall frequency
of use (see, for instance, Altenberg 1984, 1986 or Ford 1993), medial
considerations are relevant for C-constructions in at least two other ways.
One of these is with respect to the order of clauses, which Auer's chapter
in this volume investigates with respect to wenn-clauses in German (cf.
English 'when'/'if -clauses). A second issue relating to C-relations in
spoken and written discourse is the question of whether mediality has an
influence on the choice of C-marker. Montolio (this volume) argues that,
as in English, complex conditional constructions in Spanish such as a
condici6n de que ('provided that') and a menos que ('unless') express a
more restricted conditional relation than si 'if and are reserved for formal
written registers. And Barth (this volume) shows that the set of
concessive markers as well as the relative frequency of parataxis as
opposed to hypotaxis is significantly different when spoken English is
compared to written English.
Introduction 5

A further recurrent motif in the papers dealing with spoken discourse

is the use of C-connectives as discourse markers. This is an aspect of C-
relations which has been neglected in cognitively oriented studies. In fact,
the development of discourse markers from adverbial subordinators of
cause, condition, contrast and concession has only recently come to the
attention of discourse and interactional linguists (see e.g. Gohl/Giinthner
1999 for discussion of a causal discourse marker in German, and Lenk
1997 for discussion of contrastive discourse markers in English). In the
present volume it is the chapters by Schwenter on Spanish si and
Giinthner on German obwohl which deal with C-relations as discourse
markers. Whereas the obwohl phenomenon is comparable to paratactic
uses of although in English, the adversative and causal use of si has no
real equivalent with English if. But the evidence in both cases casts doubt
on developmental claims made in the literature to the effect that
conditional markers develop into concessive conditionals and from there
into concessives, while concessives do not develop any further (Konig
1986, 1988, Kortmann 1997). A line of development from conditionality
to adversativity and (paratactic) causality as with si, or from concessivity
to adversativity as with obwohl has gone unnoticed until now. Once
again, it is not insignificant that such insights have come from corpus-
based studies of spoken language. In fact, as Auer (this volume) points
out, a full understanding of spoken-language constructions can only be
achieved through corpus-based investigation. And this is another
recurrent theme in virtually all the discourse-oriented papers collected
The interactional linguistic papers in this volume (Auer, Barth, Ford,
Gohl, Giinthner, Couper-Kuhlenffhompson) go one step further with
respect to corpus-based language study. They take it as axiomatic that
spoken language is first and foremost a tool for social action and that in
order to be understood fully it must be examined in its original habitat,
i.e. in everyday interaction. Viewed from this perspective, the C-relations
can be thought of as ways of carrying out social actions-causality being
instrumental in providing justifications or accounts (Gohl), contrast and
concession in negotiating agreement/disagreement (Barth, Couper-
Kuhlenffhompson, Ford). C-constructions in tum can be conceptualized
as linguistic resources or practices for carrying out the actions in
question, e.g. for conceding a point or correcting a prior claim
(Giinthner). In an interactional perspective, C-constructions are seen as
especially appropriate for or adapted to situated interactional needs and
tasks: a German wenn-clause in post-position-as Auer shows-is not
merely a positional variant of a pre-posed 'one (and thereby freely
6 Elizabeth Couper-Kuhlen and Bernd Kortmann

interchangeable with it) but a tool 'designed' for or tailored to specific

contexts of usage.
A fmal motif-relevant to both cognitive and discourse-oriented
studies-is the relationship between asyndetic, paratactic and hypotactic
means of expressing C-relations. Although none of our chapters looks at
C-connectives diachronically, the implicational hierarchy for concession
that Crevels (this volume) offers in her typological study, namely Content
> Epistemic > lliocutionary > Textual, in conjunction with her marking
hypotheses has clear implications for language development. She
hypothesizes that the higher the semantic level, the more likely a
concessive relation is to be realized asyndetically. Moreover, if a
concessive relation is realized syndetically, the higher the semantic level
at which it is realized, the more likely it is that q rather than p will be
marked-where q-marking corresponds to parataxis and p-marking to
hypotaxis. The implicational hierarchy above corresponds, of course, to
what are thought to be increasing levels of cognitive complexity (see also
Noordmanlde Blijzer, this volume), whereas the cline asyndesis-
parataxis-hypotaxis corresponds to the unidirectionality hypothesis of
grammaticalization theory (HopperfIraugott 1993). Yet evidence from
spoken language does not fully conform to these hypotheses: asyndesis is
widespread, even for content-domain causality (Gohl, this volume) and
parataxis is more frequent than hypotaxis for the expression of
(epistemic) concessive relations (Barth, this volume). Furthermore, as
Schwenter's and Giinthner's chapters show, discourse markers which
originate from adverbial subordinators create parataxis out of hypotaxis
and thus reverse the unidirectional cline. In order to resolve such
contradictions we anticipate that more dialogue between cognitive and
discourse-oriented approaches to the C-relations will be necessary in the
Meanwhile the reader is invited to appreciate the state-of-the-art
research documented here for cause, condition, contrast and concession.
In line with the title of the series, the present collection focuses on
English; yet it does not eschew the world beyond. Indeed, the
contributions by Noordmanlde Blijzer and Pander Maat/Sanders (Dutch),
by Auer, Gohl, Giinthner and Lang (German), by AkatsukaiStrauss
(Korean, Japanese) and by Montolio and Schwenter (Spanish) serve, each
in their own way, as a valuable corrective on what would otherwise be a
perhaps excessively Anglocentric perspective. In this sense the volume
presents a picture of the C-relations which is enriched by cross-linguistic
Introduction 7

The editors would like to thank Verena Haser, Lieselotte Anderwald

and Manfred Krug for careful comments on earlier versions of the
contributions, Susanne Wagner for compiling the index, and Sabine
Conrad for checking all final versions for conformity to the publisher's
stylesheet. A word of heartfelt thanks also to Melitta Cocan for the
excellent job she has done in preparing the camera-ready copy.

Elizabeth Couper-Kuhlen, Konstanz Bernd Kortmann, Freiburg

Altenberg, Bengt
1984 Causal linking in spoken and written English. Studia Linguistica
38: 20-69.
Altenberg, Bengt
1986 Contrastive linking in spoken and written English. In: G. Tottie and
I. Baecklund (eds.), English in Speech and Writing. A Symposium,
13-40. Stockholm: Almqvist & Wiksell.
Ducrot, Oswald
1984 Le dire et Ie dit. Paris: Minuit.
Ducrot, Oswald
1996 Slovenian lectures/Conferences slovenes: Argumentative seman-
tics/Semantique argumentative. Ljubljana, ISH.
Fauconnier, Gilles
1985 Mental Spaces: Roles and strategies. Cambridge, MA: MIT Press.
Fauconnier, Gilles
1997 Cognitive Mappings for Language and Thought. Cambridge: Cam-
bridge University Press.
Ford, Cecilia E.
1993 Grammar in Interaction. Cambridge: Cambridge University Press.
Gohl, Christine and Susanne Gtinthner
1999 Grammatikalisierung von 'wei!' als Diskursmarker in der gespro-
chenen Sprache. Zeitschriftfur Sprachwissenschaft 18: 39-75.
Halliday, M. A. K.
1985 An Introduction to Functional Grammar. London: Edward Arnold.
Harris, Martin
1988 Concessive clauses in English and Romance. In: 1. Haiman and S.
A. Thompson (eds.), Clause Combining in Grammar and Dis-
course, 71-99. Amsterdam: Benjamins.
Hopper, Paul and Elizabeth C. Traugott
1993 Grammaticalization. Cambridge: Cambridge University Press.
8 Elizabeth Couper-Kuhlen and Bernd Kortmann

Konig, Ekkehard
1986 Conditionals, concessive conditionals and concessives: Areas of
contrast, overlap and neutralization. In: E. C. Traugott, A. ter
Meulen, 1. S. Reilly and C. Ferguson (eds.), On Conditionals, 229-
246. Cambridge: Cambridge University Press.
Konig, Ekkehard
1988 Concessive connectives and concessive sentences: Crosslinguistic
regularities and pragmatic principles. In: 1. A. Hawkins (ed.),
Explaining Language Universals, 145-166. New York: Basil
Kortmann, Bernd
1997 Adverbial Subordination. A typology and history of adverbial sub-
ordinators based on European languages. Berlin: Mouton de
Lenk, Uta
1997 Marking Discourse Coherence. Functions of discourse markers in
spoken English. Tiibingen: Narr.
Roulet, Eddy
1984 Speech acts, discourse structure, and pragmatic connectives.
Journal ofPragmatics 8: 31-47.
Rudolph, Elizabeth
1996 Contrast: Adversative and concessive relations and their
expressions in English, German, Spanish, Portuguese on sentence
and text level. Berlin: de Gruyter.
Sweetser, Eve
1989 From Etymology to Pragmatics. Metaphorical and cultural aspects
ofsemantic structure. Cambridge: Cambridge University Press.
Thompson, Sandra A. and Robert E. Longacre
1985 Adverbial clauses. In: T. Shopen (ed.), Language Typology and
Syntactic Description, Vol. 2 Complex constructions, 171-234.
Cambridge: Cambridge University Press.
The relevance of causality-

Paul Georg Meyer

"In der Natur gibt es keine Ursache und

keine Wirkung". (Mach 41901: 513)
"The law of causality, I believe, like
much that passes muster among philoso-
phers, is a relic of a bygone age, surviv-
ing, like the monarchy, only because it is
erroneously supposed to do no harm."
(Russell [1912] 1921: 180)
"The most fundamental fact of the
world" (KamIah 1991, title)
"The cement of the universe" (Mackie
1974, title)

This article discusses interrelations between relevance and causality from

different angles: first, it is shown that the very notion of causality is subject
to a relevance constraint. In communicating about causes people tend to
seek out one or at most a limited number of possible causes from a possibly
infinite number of logically admissible conditions. The criterion for select-
ing this limited number is obviously relevance. This interrelation between
causality and relevance is analysed by way of a critical application of rele-
vance theory, with the result that the conditions of relevance of causal
statements are more similar to felicity conditions of speech acts than to a
cognitive relevance principle.
Second, causality is described as a major device for creating relevance in
discourse. The relevance of causality in discourse cannot easily be over-
stated. Nevertheless, causality is only one among several principles of text
organisation, though obviously the most relevant one.

Causality is an "elusive" concept (Ziv 1988: 543). We all seem to know

a causal relation when we see one, but as soon as we try to explain why
it is a causal relation, or to communicate what it is precisely that we
have seen, we run into problems. Like time and space, causality is taken
for granted by everybody and fully understood, so it seems, by nobody.
And, what is most disconcerting to scientifically-minded people,
"knowledge of causes is possible without a satisfactory grasp of what is
involved in causation" (Anscombe [1971] 1975: 67). Scientists have
10 Paul Georg Meyer

consequently sought to eschew the notion in their considerations alto-

gether (see mottoes). Philosophers, on the other hand, partly drawing on
scientific results (Reichenbach 1991), have stressed its crucial impor-
tance for the functioning of the world (see mottoes).
Almost a hundred years after Mach's and Russell's verdicts, philoso-
phers, psychologists, social and cognitive anthropologists, and linguists
are still wondering about causality in their different ways. The reason
why the "law of causality", or at least the notion of causality, is still
going strong, is independent of its status in nature or science. Causality
is simply an indispensable principle in the organisation of people's eve-
ryday lives, both on the individual and the micro-group level, and, a
fortiori, also on the level of societies, states, and international relations.
When I clean the blades of my electric lawn-mower with my fingers,
I rely on the law of causality, which tells me that once the plug is pulled
from the socket, the engine cannot start and make the blades rotate
again. When I am late for a meeting, I will try to find a cause l for my
being late, and include it in the account that will presumably be ex-
pected from me; and my account will presumably be much more accept-
able if I am able to give a satisfactory explanation for my being late.
When people claim money from an insurance company, their claim pre-
supposes the law of causality, which means that there was a cause for
the damage and depending on what the cause was, the insurance com-
pany will or will not pay. When the UN sets up a tribunal to indict and
convict war criminals, the legitimacy of the procedures hinges on the
law of causality, on the belief that certain people are responsible for
certain crimes, that is, that their actions were causally involved in these
crimes in a crucial way.
The law of causality has not been allowed to survive because people
believed it could "do no harm" (Russell [1912] 41921: 180, cf motto). It
has been allowed to survive because without a notion of causality, most
of our material-technical, micro- and macro-social, economic, legal, and
international life and institutions would not make sense and would break
down. 2
It will be shown presently that these remarks have more meaning in
the context of this article than the usual conventional introductory asser-
tions about the outstanding importance of an article's subject matter
(Swales 1981). The first reason for this is that the very topic of this arti-
cle is the relevance of causality. Second, the examples adduced above to
show the relevance of causality in an intuitive sense also show that, oc-
casional scientific necrologues on the concept notwithstanding, causal-
The relevance ofcausality 11

ity seems to have its locus in the social sciences and humanities. Every-
day action, group cohesion on all levels, economic transactions of all
kinds, notions of obligation and responsibility, law as the formalisation
of the latter notions, and many other topics of social sciences and hu-
manities crucially depend on causality. We may thus take the above
examples as a first approximation to a concept of causality that is lo-
cated in a social rather than in a physical world. In other words, the uni-
verse of which causality is the cement, the world in which causality is
the most fundamental fact, is primarily a socially constructed universe, a
socially constituted world.
This article does not undertake to solve the problem of causality. Its
objective is much more modest. Based on observations made in the phi-
losophical, scientific, psychological, social scientific, and linguistic lit-
erature, I will defend a set of linguistically inspired theses concerning
the relationship between causality and relevance (cf Meyer 1983: 125-
126), putting it into the context of the present discussion and of an
emerging research programme.
I will begin with a (very) brief history of the concept of causality
(Section 1). I will then show how the history of the concept, the way it
has been discussed in the literature, and the problems it has raised, seem
to indicate a certain constraint on the notion of cause that is best de-
scribed by a notion of relevance (Section 2). Next I will ask whether this
notion of relevance is explicable in terms of the presently most influen-
tial pragmatic theory of relevance (Sperber and Wilson 1995 2) (Section
3), and discuss problems arising from such an attempt. Finally, I will
argue for a basically discursive explication of the notion of causality in
the framework of a coherence relations approach, establishing causality
as one of several primitive universal principles that govern the construc-
tion of coherence in discourse (Section 4).

1. A brief history of causality

Talking of causality under a unitary label misleadingly suggests that the

label covers a unitary, homogeneous phenomenon. Breul (1997: 81-112)
has recently summarised the develoJ?ment of the notion of causality
from Aristotle via Hume into the 20 century. Among Aristotle's four
different kinds of causality (Metaphysics 5, 2, 1013a24ff), there is only
one'· (known as causa efficiens from the Latin translation) that is clearly
identifiable with a modern common-sense notion of causality. In the
12 Paul Georg Meyer

scientific and philosophical tradition of post-Renaissance Europe (e.g.

Hume), another notion of causality is brought forward, emphasising the
necessity and lawlikeness of the causal relation. Finally, the notion of
causality in that tradition dissolved into the mathematical notion of
function, as exemplified in Mach (1901) and Russell ([1912] 1921).
This notion is certainly far removed from everyday intuitive concepts of
causality. Strawson (1985) argued against a law-based notion of causal-
ity to cover everyday cases of causation:

A man, say, falls down a flight of stone steps as he begins the descent. The
steps are slippery and the man's mind is elsewhere. This is a sufficient ex-
planation of his fall. But of course not every preoccupied man falls down
every flight of slippery steps he descends. There is absolutely no question of
our formulating or envisaging exceptionless laws, ... , to cover all such
cases. (Strawson 1985: 131)

That is, all rational and enlightened people will assume that the
man's falling down the stairs is fully in accordance with some presumed
laws of nature and, in principle, fully explainable in such terms. And
yet, in ordinary language we do not expect such full "scientific" ac-
counts to be given, and indeed it is highly doubtful whether such full
causal accounts of singular events are useful, desirable or, for that mat-
ter, feasible.
For all practical purposes, we are content with much less than a list-
ing of all the conditions, regularities, probabilities, and laws of nature
that account for the event in question. A satisfactory explanation, in
fact, is not one that lists all this. A satisfactory explanation is concise
and concentrates on one or just a few causes. The problem that scientists
and philosophers have had with the notion of causality emerging from
this observation is that they cannot fmd a unique logical or mechanical
or otherwise physical characterisation of a state of affairs that happens
to be intuitively singled out as the "cause" of a certain event. Whatever
is named as "the cause" of something, does not seem to differ in any
significant way from other factors present in the situation in question.
There simply seems to be no logical or scientific basis for calling some-
thing a cause.
Logical analyses of the notion of cause have operated with necessary
and sufficient conditions. The two intuitively most plausible and (pre-
sumably for that reason) most influential defmitions have been Mackie's
(1965) "INUS condition" and the "counterfactual conditional", origi-
nally suggested by Hume and reintroduced into the modern debate by
The relevance ofcausality 13

David Lewis (1973). As we shall see in explicating what an INUS con-

dition is, the two are actually quite close to each other.
According to Mackie (1965) causes are best defined as "INUS" con-
ditions. An INUS condition is "an insufficient but necessary part of a
condition which itself is unnecessary but sufficient for the result. "
(Mackie 1965: 16).
Thus, to take Strawson's example quoted above, both the slippery
state of the stairs and the man's absentmindedness are causes of the
man's fall because they are
• insufficient: Slippery stairs or absentmindedness alone do not cause a
man to fall down a staircase, and would not have done so in that par-
ticular case.
• necessary: Had the man been more careful or had the stairs not been
slippery, the man would not have fallen.
• part: The two causes mentioned are only part of the set of conditions
causally involved in the man's fall. Thus, e.g., the man's trying to
descend the stairs in the first place is another condition on the man's
fall (that could also be called a cause on that account because it is an
insufficient but necessary part of the set of conditions leading to the
The set of such conditions taken as a whole is
• unnecessary for the result because the same effect might also have
been achieved by other causes, in an altogether different set of condi-
tions: Thus, e.g., the man might have fallen down the stairs because
somebody maliciously pushed him, or because he was drunk and the
steps were worn.
• sufficient (in that particular case) for the result because, after all, the
result was brought about.
According to Hume and David Lewis (1973) c is the cause of an
event e if and only if the counterfactual statement: "If c had not been
true, e would not have happened" is true. In other words, a cause is a
'conditio sine qua non'. The attentive reader will have noticed that in
explicating the notion of INUS condition above, we repeatedly had to
resort to counterfactual formulations: if the man had been more careful
or if the steps had not been slippery, he would not have fallen. Absent-
mindedness and the slippery state of the steps alone would not have
brought about the accident. The same accident could have been brought
about by altogether different causes, etc. It is thus clear that the INUS
definition of causality is only a more formalised and elaborate variant of
the counterfactual formulation of that definition.
14 Paul Georg Meyer

The problems with both these defmitions are manifold (Sosa 1975: 3-
4), but it is not my purpose to discuss them here. There is just one point
that I want to make: Mackie refers to a set of conditions that is unneces-
sary, but sufficient for the effect. The factor singled out as "the" cause is
not the whole set of conditions, but only an insufficient but necessary
part of this set. The problem that this definition raises is that it is not
very useful for an application of the concept of causality in the analysis
of everyday communication. The definition covers a number of condi-
tions that we would not call causes in a given situation. That is, Mackie
describes a necessary but not sufficient condition for calling an event a
cause. He describes a semantic precondition for a proposition to serve as
a causal explanation. Real events always have a large number of neces-
sary and sufficient conditions. Arguing that one circumstance was the
cause of a given event means foregrounding one insufficient and neces-
sary condition and stating it, and allocating all the other conditions that
jointly were unnecessary but sufficient to bring about the effect to the
background. Thus, the very problem of Mackie's definition is an indica-
tion of the basically pragmatic character of any reasonable notion of
causality that would be applicable in the analysis of discourse.
It is a situation that logical positivists and context-free semanticists
cannot easily cope with, but it is not unfamiliar to pragmatically ori-
ented linguists: reality is always understated and underdetermined by
the propositional content of utterances. The meaning of utterances al-
ways goes beyond what was said. There is always a set of meanings that
were meant but left unsaid. And there is always a background of knowl-
edge that is taken for granted, that is not problematised, that is simply
not at issue. If we believe Searle (1983: 141), this background contains
infinitely many propositions. And indeed, it is precisely this characteris-
tic of the set of sufficient conditions for a certain event that has led early
modem philosophers (such as Hume) to confine the notion of causality
to "lawful" instances and has driven 19th and 20th century scientists and
philosophers to despair.

2. What makes a cause a cause?

Let me try to illustrate this with Strawson's example (in fact, any exam-
ple of a causal relation would do). What is the set of conditions that is
jointly sufficient (though unnecessary) to bring about "the man's" fall
(for simplicity's sake let's call him Jones)? The two conditions men-
The relevance ofcausality 15

tioned by Strawson are in the foreground, of course: the steps were slip-
pery, and Jones was absent-minded. Apart from these, I have already
mentioned one further condition: the fact that Jones did try to descend
the stairs at the very time when the first two conditions were already
given. It seems that these three conditions, each of them independent of
the two others, were jointly sufficient for Jones' fall. But, as Strawson
(1985: 131) pointed out, "not every preoccupied man falls down every
flight of slippery steps he descends." So there must have been further
conditions that were necessary to make the set of conditions sufficient
for Jones' fall to occur: maybe if Jones had been wearing different shoes
(say, with non-slip rubber soles), he would not have fallen. Or, if there
had been a railing for him to hold on to, he would not have fallen. Or, if
his ankles had not been so stiff that day, he would not have fallen. Or,
perhaps, if he hadn't been drinking, in addition to his absentmindedness.
Or ...
On the other hand, we might also begin to question the most obvious
and plausible causes that were given: granted that the steps were slip-
pery, were they slippery enough for a man to slip? Granted that Jones
was absent-minded that day, was he absent-minded enough not to notice
the slippery state of the steps? That is, even the most readily accepted
explanations for an event are not undebatable. There is always a possi-
bility that the most plausible explanation is erroneous. Suppose that
Jones is absent-minded most of the time and the steps are always slip-
pery and he walks down these steps several times a day without slip-
ping. It is obvious that in this case we would still accept the explana-
tions given by Strawson to a certain extent, although we would have
evidence for the fact that the explanation is not sufficient. Even though
we would be willing to grant that there must have been other "causes"
for Jones' fall, the explanations "slippery steps" and "Jones' absent-
mindedness" are just too good to be discarded.
What this discussion is supposed to show is: as scientists and phi-
losophers found out long ago, there is no such thing as "the" cause or
"set of causes" of an event. Causality emerges as a discursive phenome-
non. Note that all the "causes" mentioned above could be adduced in
some kind of dispute over the "causes" of Jones' fall. But it depends
precisely on the kind of dispute as to which of the causes will be ad-
duced by whom.
Suppose Jones' wife had always been of the opinion that that stair-
case is dangerous and had always been warning Jones to use it (suppos-
ing there was an alternative available). In this case, the cause of Jones'
16 Paul Georg Meyer

fall that would be accepted by his wife would be the fact that Jones de-
scended the stairs in the first place. ("Did you have to use that danger-
ous flight of stairs? I told you one day you'd break your neck on it".) Or
that Jones descended the stairs while they were slippery ("Why couldn't
you wait until the steps had dried?") (Supposing, e.g., that they were
slippery because Jones' wife had just scrubbed them).3 For his wife, in
that case, the slippery state of the stairs would be taken for granted, just
like her husband's absentmindedness, which she has learnt to put up
with in long years of marriage.
Now take the case of an insurance company. Depending, of course,
on what kind of risk had been the object of the insurance, all kinds of
questions might be raised. Who was responsible for the slippery state of
the steps? Why did the flight of stairs have no railing? Was Mr. Jones'
absent-mindedness due to alcohol or drugs? What was the purpose of
his descent down the stairs? Was it on his way to work, for example?
Was Mr. Jones' stiflhess in the ankles due to an accident or some ill-
ness? Note that the insurance company's questions open up a much
wider background by considering what might provisionally be called
second- or third-order causes. The reason for this is that an insurance
company will tend to follow up causal chains until somebody is found
who can be blamed for an accident. Thus, for an insurance company or a
lawyer, the cause of Mr. Jones' fall might be the fact that the owner of
the house failed to replace a broken railing or to provide some kind of
slip protection measures for the dangerous stone steps, or the fact that
Mr. Jones' doctor prescribed some drug for him that affected his capac-
ity to keep his balance to a sufficient extent to cause, in connection with
the slippery state of the steps and Jones' absentmindedness, his fall
down the stairs.
That is, when considering the set of conditions jointly sufficient to
bring about an event, it is not sufficient to look at "first-order" causes
only. In any case, it is a question of focus and perspective what is re-
garded as "first-order cause". Take Mr. Jones' absentmindedness as an
example. Strawson said it "caused" Mr. Jones' fall. But looking more
closely at the actual event, we may be certain that innumerable other
"causes" intervened between Mr. Jones' absentmindedness and his fall.
Absentmindedness cannot immediately "cause" a fall. Before the fall
there must have been a slip. And it is difficult to ascertain at which point
in the causal chain leading to Mr. Jones' fall absentmindedness is sup-
posed to come in. Did absentmindedness cause him to slip, and the slip
cause him to fall? Or did he slip first, and did absentmindedness prevent
The relevance ofcausality 17

him from keeping his balance after the slip? Or did he slip and lose his
balance inevitably, and did absentmindedness prevent him from holding
on to something, which might have prevented his fall? Or was absent-
mindedness a limiting factor of all of Mr. Jones' movements that day,
and did it have disastrous effects all through the causal chain leading to
the accident?
Even if this was a real case, I doubt that all this detail could be ascer-
tained. It should be clear, however, that a distinction between first-order
and second- or third-order causes is impossible and has no systematic
status. What we regard as a cause is seldom the "immediate" cause (in
the sense that no intervening event between cause and effect can be
found) because our observations are seldom so fine-grained that "imme-
diate" causes would be noticed. Let us suppose that theoretically we
come to the conclusion that some basic muscular movement of Mr.
Jones immediately prior to his fall must have been the "immediate"
cause. Presumably neither Mr. Jones nor a by-standing observer would
have been able to observe and identify this "immediate" cause. Pre-
sumably this "immediate" cause was the inevitable result of some prior
events, conditions, and maybe of Mr. Jones' decisions, and these will
most likely be singled out as "the cause". We tend to look at events and
causal chains not from a mechanistic perspective, but from a perspective
that allows us to see and represent points of potential intervention (cf
Von Wright 1973). We look for conditions that could have been differ-
ent. And usually conditions can be different if people behave differently
or change conditions deliberately. Mr. Jones might not have used that
dangerous staircase at all that day. Mr. Jones could have been more
careful. The steps might not have been slippery (somebody could have
done something about that). So the insurance company, by opening up
an infinite space of conditions jointly sufficient for the accident, at the
same time limits the search to those states of affairs that are accessible
to human perception and intervention.
Let us now look at Mr. Jones himself How would he explain his fall
(supposing he survived it)? Again it would depend on a number of con-
tingent factors what would be a satisfactory explanation of the accident
for Mr. Jones himself So Mr. Jones might (most likely) blame the slip-
pery stairs, but it should be clear that he can only do so under certain
conditions. Suppose that Mr. Jones has been using these stairs for years,
they have always been slippery, he has never had an accident and has
never complained about the state of the steps. In that case it would be
somewhat difficult for Mr. Jones to blame the slippery steps. If he chose
18 Paul Georg Meyer

to do so, it would mean a "change of policy", so to speak. After years of

putting up with the slippery stairs, he would suddenly blame them for an
accident. Under a benevolent interpretation, this would mean his fall has
taught him a lesson, and he probably will have to do something about
the slipperiness of the stairs in the near future. Mr. Jones could also
blame his own absentmindedness, if he is willing to make such a face-
damaging admission. What is more likely (assuming Mr. Jones to be
vain like most people) is that he will blame some other condition that
does not figure in Strawson's account at all. He could resort to some
other explanation that would be more face-saving. "I was a bit stiff in
the ankles that moming" would do perfectly.
What this lengthy exposition of potential explanations for a fictitious
accident is meant to show is that explanations are not objective, based
on facts existing independently. In fact they are guided and informed by
people's interests. What explanations we are willing to give and willing
to accept as a "cause" is not so much a matter of fact, but in the best
case open to negotiation. Establishing a cause of an event is not so much
a fact-finding procedure, but rather a process of social interaction in
which diverging interests have to accommodated, from which different
consequences (even obligations) may arise (cf Hilton and Erb 1996:
303). It will thus be appropriate to look at the notion of causal explana-
tion in order to clarify the notion of cause. This will be done in the next
It should, however, be emphasised that the above account does not
imply that "the truth" is at the mercy of subjective interests. All the ex-
planations mentioned above that could be adduced for Mr. Jones' fall
might be, strictly speaking, true. Truth is a semantic concept, and se-
mantically, all the conditions mentioned above as possible causes of Mr.
Jones' accident are INUS conditions: they are insufficient but necessary
parts of the whole set of conditions leading to the accident. They differ
not in terms of truthfulness but in terms of relevance in certain discur-
sive contexts and for certain people.

3. The relevance of causality

What does it mean for an explanation to be relevant in a given context?

The notion of relevance plays a crucial role in several theories of prag-
matics (Grice 1975; Hom 1984; Sperber and Wilson 1986, 1995). But
before I relate the previous discussion to these theories, I would like to
The relevance ofcausality 19

explicate in my own terms what I mean by "relevance" here. What

makes an explanation relevant? A good explanation raises an issue that
was either
- outside the "normal course of events", unpredictable, and out of con-
trol. This explains the occasional meteorite examples in the literature
(Stegmuller [1969] 1983: 911). Or it was
- within control of people, that is, something that could have been
otherwise, a condition that was amenable to change, but still to a
certain extent unexpected and outside the normal course of events.
So when the slippery steps are adduced as an explanation of Jones'
fall, it is implicated that perhaps they should not have been slippery, that
something should be done to prevent people from slipping there. If
someone takes the slippery stairs for granted as a fact of life, then Jones'
absentmindedness or his using the steps in the first place will be fore-
Conditions that are absolutely unquestionable will never be men-
tioned as causes. Thus, the fact that Mr. Jones' body has weight, or, in
particular, that its specific weight is higher than the specific weight of
the atmosphere above the stairs and accordingly obeys the law of grav-
ity, would certainly not be accepted as an explanation for Jones' fall
except in circumstances (e.g., science fiction) where weight and gravity
are not to be taken for granted. Neither would the fact that Mr. Jones
was born be accepted as an explanation for his fall, although his birth is
an INUS condition on his fall, and if he hadn't been born, he couldn't
have fallen.
The closest equivalent to this informal notion of relevance for causal
explanations that I could fmd does not come from Gricean and neo-
Gricean pragmatics, but from speech-act theory. This should not be too
surprising if we accept that explanation or explaining is a speech act in
its own right. Like all illocutionary acts, it could have among its felicity
conditions a formulation reminiscent of Searle's classical formulation
(Searle 1969: 59)4

(1) Afelicitous explanation will adduce as a cause for a given event E

a condition C that is an INUS condition ofE in Mackie's (1965)
sense, where C is a proposition that would not have been true in
the normal course ofevents, or the truth of C is not obvious to the
hearer H.
20 Paul Georg Meyer

This formulation, as in all cases where Searle uses formulations of this

kind, has its problems, however. In particular, it is difficult to determine
what a "normal course of events" would be in each case, and what pre-
cisely is "obvious" to the hearer.
Our discussion above has shown something else. There is, strictly
speaking, no such thing as a "normal course of events"; and it is not
obvious what could be called "obvious to speaker or hearer"; but there
is, for each context of situation, a background (Searle 1983: 141).
Propositions which belong to the background are regarded as unques-
tionable, presupposed, imputed and taken for granted. The background
is never thematised except when communication breaks down or illocu-
tionary acts fail. What is part of the background in this sense cannot
felicitously be invoked as an explanation (like in most everyday con-
texts the fact that bodies have weight). We may thus stipulate:

(2) A felicitous explanation X will adduce as a cause for a given event

E a condition C that is an [NUS condition in Mackie's (1965)
sense and that is not part of the background (in the sense of
Searle 1983) for the description ofE orfor X

This formulation, I think, is quite analogous to what should replace the

corresponding felicity conditions in Searlean descriptions of illocution-
ary acts. A background fact cannot be felicitously made the content of
an assertion; it would be a typical case of a trivial assertion. Since ex-
planations are assertions, the above formulation provides only a special
case of such a felicity condition. Nor can background facts be made the
content of a directive: when you order a table in a restaurant, you don't
say you want chairs around it as well (unless it is a type of restaurant
where people usually sit or lie on cushions on the floor). And making
part of the background the content of a promise ("A happily married
man who promises his wife he will not desert her in the next week") "is
likely to provide more anxiety than comfort", as Searle (1969: 59) re-
Now this classical Searlean felicity condition, invoking obviousness
and the normal course of events, on most types of illocutionary acts, has
usually been equated with Grice's maxim of Relation ("Be relevant",
1975: 46-47), Horn's R-principle (1984) and Sperber and Wilson's
principle of RelevanceS (1986, 1995). It is doubtful to me whether this is
really the case. True, the Searlean felicity condition is approximately
consistent with colloquial meanings of the term relevance. But Grice's
The relevance ofcausality 21

maxim is a maxim of Relation, not of relevance (although its summary

uses the term). Grice's maxim of Relation has to do with the require-
ment that utterances should be related to the previous discourse and the
context of situation. In this way, it guarantees coherence of discourse.
But this is not what Searle had in mind when he referred to "the normal
course of events".
Hom's R-principle (1984), however, comes closer to the Searlean
notion because it incorporates half of the Gricean Quantity maxim as
well. In Gricean terms, utterly trivial statements or explanations violate
the Quantity maxim in that they give too little information. In Hom's
conception, this submaxim is combined with Grice's Relation maxim to
yield a complex R-principle, constituting a lower bound on how much
information should be given in an utterance.
Sperber and Wilson's (1986) conception of Relevance, although it
seems no more than a radicalisation of Hom's R versus Q duality, con-
stitutes yet another concept of 'relevance' because it includes Q- and R-
based considerations from the Hornian model in one formula. Relevance
is no longer one of several principles, but the only principle guiding
ostensive-inferential communication (Sperber and Wilson 1986: 155-
158). The R versus Q duality is converted into a quotient of "cognitive
effects" divided by processing effort. This means (counter to everyday
intuitions about relevance) that high processing costs of a message
(measurable in length, complexity, difficulty, amount of noise, etc.)
reduce the Relevance of that message.
Applying this conception to explanations, it is true that lengthy,
complicated, difficult explanations given in noisy surroundings or
through a fragile channel are certainly less acceptable to hearers. But
they may still be good, felicitous, relevant explanations in terms of the
above thesis (2).
This is not meant as an objection to Sperber and Wilson's theory-it
just points out a difference in conception. But there is another problem
that arises in connection with Sperber and Wilson's Relevance theory
when equating relevance as discussed here with their Relevance: The
Relevance of an event, Sperber and Wilson say (1986: 156-157), corre-
lates positively with the effect of that event on a given context of situa-
tion. What, now, is an effect? Doesn't this notion presuppose a notion of
causality? The event whose Relevance is to be measured is seen as a
cause having some effect on a situation. An event's Relevance is gauged
in terms of the relative effect of that event, seen as a cause. And it is a
cause that could be adduced in a felicitous causal explanation as defined
22 Paul Georg Meyer

in (2) above. It is a prerequisite for successful communication that an

utterance is not just an INUS condition for its contextual effect. It is the
very issue discussed by Sperber and Wilson that utterances should be
relevant causes of their contextual effects. If other INUS conditions of
that same effect (such as the Hearer's knowledge of the language used,
absence of noise drowning out the utterance, etc.) were more relevant in
a given situation, then the utterance itself would lose its relevance. The
problem is: combining Relevance theory and the relevance-based notion
of causality put forward here leads us into a vicious circle: we explicate
causality in terms of relevance only to find that relevance is explicated
in terms of causality!
I can presently see two ways out of this dilemma, without being sure
which of the two is to be preferred, and without being sure whether they
are actually mutually exclusive:
The first solution is less clear to me in all its detailed theoretical
implications, but there is some evidence in the literature in favour of it.
This solution would be to declare causality a semantic and cognitive
primitive (cf Ziv 1993: 21). This idea is actually an old one in linguis-
tics, first put forward formally, as far as I know, by postulating an
"atomic" semantic predicate CAUSE in generative semantics (McCaw-
ley 1968). The "linker" because is also contained in Wierzbicka's
(1991: 8, 1996: 70) lists of semantic primitives; though not in the origi-
nal one (1972: 16).
Lakoff's (1982: 163) idealised cognitive model (ICM) of causation
adds an interesting aspect to the discussion. Lakoff rejects the notion of
primitive in the traditional sense (1987: 279-280),6 replacing it by his
own notions of basic-level concept and image schema as the nearest
equivalents. Causation is obviously a basic-level concept in Lakoff's
terms. The interesting point is that basic level concepts have some char-
acteristics of traditional primitives in that they are "not put together by
fully productive principles of semantic composition" (ibd.). But they do
or may have internal structure. In this they differ from traditional primi-
tives. This internal structure is spelled out by Lakoff in idealised cogni-
tive models. The elements of Lakoff's model of causation do not men-
tion lawfulness or necessary and sufficient conditions at all. Similar to
Von Wright (1973), Lakoff regards agency and the transfer of energy as
essential features of prototypical causation. It should be noted that it is a
model of causation, not of causality, causation being understood as the
nominalisation of an agentive verb. It is plausible that prototypical cau-
sation is agentive; and it is quite conceivable that the abstract principle
The relevance ofcausality 23

of causality as used in discursive reasoning should be derived from

agentive causation. In this case, the cognitive experience of causality as
an abstract principle would be regarded as a notion that is gained by
abstraction from prototypical agentive causation.
There is also some justification to be found for the primitive analysis,
or rather non-analysis, of causality in the philosophical literature:
Anscombe ([1971] 1975) seems to advocate a view, summarised by
Sosa (1975: 4), that "causation is what it is and nothing else-and that
there is no analysis of causation that essentially involves conditionality
or lawfulness". And Sosa continues: "So far as I know no one has pub-
lished a successful analysis of causation by reference to conditionality
or lawfulness" (1975: 5). Note that Sosa, too, speaks of "causation"
rather than of"causality".
If it is true that causality is a primitive concept or very close to a
primitive concept of causation, we encounter a familiar phenomenon
that is often observable when we try to analyse a primitive concept.
Take the example of see. In accordance with Wierzbicka (1996: 78-79) I
would argue that it is a primitive concept. An obvious classical defini-
tion of see along the lines of genus proximum and differentia specifica
would of course be possible in terms of "visual perception" (Searle
1983: 61; Alm-Arvius 1993: 17). I argued in Meyer (1997: 121-122)
that visual, and presumably perception, too, cannot be defmed without
reference to the primitive concept see. It might just be that "analysing"
causality in terms of necessary and sufficient conditions and in terms of
relevance or lawfulness is a similar fallacy to "analysing" see in terms
of "visual perception".
It is quite conceivable that necessary and sufficient conditions are
conceptually secondary to causality; material implication as defmed by
means of truth tables in logic is an abstract notion far removed from the
conceptual lives of ordinary people. It is at least as far removed from
everyday concepts of if-then as an INUS condition is from everyday
causality. And, as we have seen above, relevance is possibly a notion
that cannot do without causality because it has to do with causal impact.
As this solution is compatible with Sperber and Wilson's account of
relevance and communication, and as Sperber and Wilson's account is
grounded in cognitive categories, it would mean that presumably causal-
ity is the cement of a cognitively appropriated world, on which both
perception and communication are based.
The other solution would be to develop a "causality-free" conception
of relevance. Maybe the formulation in (2) above is a step in that direc-
24 Paul Georg Meyer

tion. It would be a step "back" towards Grice, Hom, and Searle, away
from Sperber and Wilson. It would mean that relevance, and, in conse-
quence, causality, are categories of communication and interaction, not
of cognition (Edwards and Potter 1993: 24).
For the time being, I am inclined to pursue the second line of reason-
ing, interpreting causality as a discursive-interactional rather than a
cognitive phenomenon. There are two reasons for this: first, this line is
more in accordance with my own previous research on discourse
(Meyer 1975, 1983, 1996) and the role of causality in it. Second, it
seems to be the tum that psychological attribution research is taking at
the moment (Edwards and Potter 1993), so that further linguistic re-
search in that direction opens up promising channels for an interdisci-
plinary exchange of ideas.
As a textlinguist who has been workin~ in a coherence relations
framework (though under a different name), I have always been con-
vinced that causality has a major role to play in the explication of text
constitution and text coherence. But until recently I had tended to be-
lieve that causality was a principle "out there", in the world, being re-
ferred to and being made use of in discourse.
To see research in science, logic, psychology and social sciences
converge on the idea that causality is primarily located in discourse does
not come as a complete surprise, though, given the discursive-
constructionist tum in social sciences in general (Woolgar 1988; Latour
1987; Bazerman 1988) which has left none of the sacred traditional no-
tions of epistemology untouched. Expressions such as fact, truth, nature
or world cannot be used naIvely anymore after taking notice of that dis-
cussion (in which they invariably tum up in "scare quotes", and are ridi-
culed more often than seriously discussed).
Nevertheless, this "discursive turn" of the causality discussion sets
new tasks for discourse analysts as well. We cannot shift responsibility
for the notion of causality to scientists, psychologists, or philosophers
any more, but find ourselves as the branch of knowledge most immedi-
ately competent for a characterisation of that notion. While I remain
sceptical concerning social constructionism in general (Meyer 1997: 67-
74), I have allowed myself to be convinced by the overwhelming evi-
dence in the case of causality. It remains an open question, though,
which is the more appropriate approach, a cognitive or a discursive one.
My decision for a discursive approach is preliminary and inspired by
research-practical considerations.
The relevance ofcausality 25

4. Causality in discourse

So far I have considered the notion of causality as a concept external to

linguistics and found it subjected to relevance constraints. This led me
to adopt the perspective of causality as a discourse phenomenon in a
much more radical way than I had hitherto done.
Let me now start from the other end and see what can be said from
inside linguistics about interrelations between causality and relevance,
in particular from a textlinguistic perspective. Seen from this perspec-
tive, causality is not only subjected to relevance constraints, but also
creates relevance.
This can best be seen from the numerous examples, discussed con-
troversially in the literature, which show that asyndetic clause connec-
tions are often interpreted as causal connections, be they juxtaposing as
in (3) and (4), coordinating as in (5), or subordinating as the participle
construction in (6) (the examples are taken from a popularised social
science text):8

(3) People badly want a demonstration, a dramatisation of justice.

They need to have defined and reinforced for them what is right
and what is wrong.

(4) Penal methods by themselves will not put an end to crime. Even at
their fiercest, as in nineteenth-century England, ( ..) they did not

(5) The participants remain anonymous and there is no one upon

whom the authorities can pin the offence.

(6) The <accusatorial> system is more adaptable than the inquisito-

rial one, allowing for practical changes in treatment methods to
be introduced easily during the sentence.

It seems that causality is such a basic principle that very often there
is no need to draw special attention to it, and the causal relation between
the two clauses is a matter of implicature. Even the direction of causal-
ity is left to the reader to infer in such cases. While in (3) to (6), the sec-
ond clause gives an explanation of some kind relating to the first clause,
the causal relationship is reversed in (7). The second clause states a con-
sequence of the fact reported in the first clause:
26 Paul Georg Meyer

(7) In all the countries concerned the population is growing quickly.

All will be faced sooner or later with the spectre of overpopula-

The precise character of the explanations in (3) to (6) has also been the
subject of much debate, but is of little concern for the present contribu-
tion (see Dancygier and Sweetser; Pander Maat and Sanders [this vol-
Kortmann (1991: 118-121) has suggested a scale of informativeness
for adverbial relations. Causally based adverbial relations (concession,
condition, instrument, and cause) are to be found above a dividing line
in the middle of Kortmann's scale which separates more informative
from less informative adverbial relations, contrast being the only non-
causal relation figuring in the top group. Kortmann uses this scale in
explaining the semantic interpretation of free adjuncts and absolutes,
i.e., participle constructions in English. It could also be used for a rank-
ing of coherence relations. It is obvious, e.g., that a causal relation is
more informative than a temporal one because a causal relation implies
a temporal relation and only adds further information to this.
A standard analysis of examples (3) to (7) within the coherence rela-
tions approach would be to find specific coherence relations such as
"explanation" or "justification" to describe the relation between the two
clauses. Against such analyses much scepticism has been voiced, e.g.
recently by Dahl (1995: 259):

Causality in discourse is a rather fluid phenomenon whose relations with

what we would want to call rhetorical structure are of an indirect character.
The view I have argued against-that causality in discourse may be reduced
to a small set of rhetorical relations9-has to my knowledge never been
stated explicitly, but many treatments of discourse structure certainly give
the impression that such a reduction is possible.

I am not quite sure what is meant by "reduction" in the above quotation.

I agree with Dahl that the functions of causality in discourse are mani-
fold (see examples (3) to (7) above). But this is no reason why they
should not be capable of being reduced to one principle. There is one
thing that can be gained by an introduction of a principle of causality
into a coherence relations approach: the complexity brought about by
the "fluid" character of a fairly large number of different causally-based
coherence relations which are not easily delimited from each other may
be "reduced" by "reducing" them to a principle of causality. As was also
The relevance ofcausality 27

observed by Ziv (1993: 181), causality could be regarded as one of a

small number of basic universal reasoning principles that readers and
hearers use in interpreting clause sequences in texts.
I think that such principles of text organisation should have the fol-
lowing properties:
a. They stem from basic needs arising in human communication.
b. They therefore constitute reasons why we should need more than one
clause to say what we want to say, that is, motives to utter a text
rather than a single clause.
c. They represent ways in which clauses, sentences, or larger text
chunks can be relevant to each other.
d. They are not derivable from a further common principle.
Causality fulfils all the above criteria to a high degree:
a. It is a basic need in human discourse to explain, to justify, to reason
about causes, conditions and consequences. The close affinity be-
tween the primitive concept of causation and human action is no co-
incidence. People want to know about causes, reasons and conse-
quences because they need to act.
b. The most natural discourse strategy of explanation or stating conse-
quences is to add another clause.
c. The common use of asyndetic clause connections to explicate causal
relations in discourse shows that causality creates relevance on its
To apply criterion d. we need to know what the other principles are. In
my own work (Meyer 1983, 1996) I have identified five principles so
far that fulfil the above conditions: Topic, Time, Clarification, Causal-
ity, and Persuasion. 10 Among these, Causality seems to be the most
"relevant" one, both in terms of text frequency, and in number and di-
versity of coherence relations derivable from them. But the other princi-
ples can also be shown to be important organising principles in texts
that are not derivable from the principle of Causality.
When we speak or write about a certain topic, then usually there is
more to be said about this topic than can be accommodated in one
clause. Often several topics have to be related to each other. This is how
the different forms of topic development arise which in tum give rise to
a number of different coherence relations such as "constant topic",
"topic progression", "split topic", etc. (cf. Danes 1970).
When we narrate a story, there is more than one event to narrate, and
these events are primarily and minimally related temporally. This is how
time comes in as an organising principle of texts.
28 Paul Georg Meyer

Furthermore, when we describe something, there is often the need to

clarify what we are referring to, that is, to spell out in more detail what
we wish to say about our topic, to compare, to give examples, to para-
phrase. All this cannot be done in one clause and thus, specific coher-
ence relations such as elaboration and contrast arise.
And it is clear that very often persuasion, that is, the need to con-
vince the hearers rather than just make them comprehend, can be the
motivation for adding another clause, too, giving rise to coherence rela-
tions such as "qualification" or "evidence".
If what we said above is true, our five principles also represent five
ways in which clauses or sentences can be relevant to each other in the
Gricean sense: they represent five different ways to fulfil the maxim of
Relation in coherent discourse. But the five principles are not on an
equal footing here. 11 Kortmann's scale can be applied to them, yielding
different degrees of relevance, depending on the way in which a clause
relates to the rest of the discourse.
The lowest degree of relevance that two sentences can have to each
other is brought about by their being related through their topic or top-
ics. Very often this does not suffice for text coherence to be brought
about. When a topic is split up into subtopics, or when the text proceeds
from one topic to the next, this is only admissible insofar as the subtop-
ics or the newly introduced topics are relevant in terms of some global
purpose of the text, or, to put it in Grice's terms: contributions to a co-
herent discourse must be "appropriate to the immediate needs at each
stage of the transaction" (Grice 1975: 47).
It is obvious that clarification and persuasion devices are highly rele-
vant, even in Sperber and Wilson's sense, to the purposes of the com-
munication of which they are part. Clarification strategies aim at lower-
ing the processing costs of hearers or readers: they make the text more
easily comprehensible. But also in Grice's sense, it may be said that
clarification may be the way in which one clause is relevant to another
in discourse, serving the "immediate needs", at this "stage of the trans-
Persuasion devices are meant to heighten the probability of a contex-
tual effect of the message on the reader / hearer. They sometimes explic-
itly invoke relevance as an argument to impress the reader or hearer. In
Gricean terms, persuasion devices are applied at a certain stage of a
communicative transaction where scepticism is apparent on the part of
the hearer, or anticipated by a writer on the part of the reader.
The relevance ofcausality 29

When we narrate events that are in some temporal relation to each

other, these events must be also relevant to each other. This is why nar-
rated event sequences are so often causally connected as well. Two
events that are temporally related to each other do not necessarily make
a good story. But a causal chain of events is much more likely to be
accepted as suiting the "immediate needs" at this stage of the narration:
its narration may explain why something happened or point out conse-
quences of a certain event. More often than not, the question "What
happened next?" aims at learning about the consequences of a certain
event rather than about some causally unrelated event that happened to
happen next. And informing about the causes of a certain event fulfils a
basic need in human communication (see above). Thus, causality seems
to be the most relevant of the five principles mentioned. Explicating a
causal relationship in discourse is rarely questioned or regarded as
pointless. This is no real surprise given that, as we saw in the last sec-
tion, causality is constrained by relevance considerations. Causality is
thus much more intimately connected with relevance than mere tempo-
ral sequence (cf Kortmann's scale, 1991: 121).

5. Summary

In this paper I have discussed interrelations between causality and rele-

vance. I have tried to show from three different perspectives that there
are good reasons to regard causality as a primarily discursive phenome-
1. The relevance of a causal statement is not ascertainable objectively,
but only in a discourse in which human interests and social obliga-
tions play a role.
2. The relevance of a causal statement is best described in a speech-act
framework. Its conditions of relevance are more similar to felicity
conditions of speech acts than to a cognitive relevance principle.
3. The relevance of causality in discourse cannot easily be overstated.
Nevertheless, causality is only one among several principles of text
organisation, though obviously the most relevant one.
If this is reasonable (and this article, I hope, contains arguments to
show that it is), then social-psychological causal attribution research
(Edwards and Potter 1993) becomes more important for a characterisa-
tion of the concept of causality than logical analyses. This is all the
more compelling for linguists, like myself, working in a functionally
30 Paul Georg Meyer

oriented typological framework. Thirty years of experience with this

framework have shown that what is likely to emerge as an interesting
typological parameter for such a framework, or as a non-trivial seman-
tic-pragmatic universal, is more likely to be found in everyday linguistic
usage, everyday reasoning, everyday argument, than in logically puri-
fied abstractions of these.
Social scientists working on causal attribution have long begun to
look to linguistics for answers to some of their questions (Edwards and
Potter 1993; Hilton, Jaspars and Clarke 1990; Hilton and Erb 1996). It is
time that linguists began to understand these questions and provide
more satisfactory answers.


* This paper is dedicated to Ekkehard Konig on the occasion of his 60 th

I wish to thank the editors, Elizabeth Couper-Kuhlen and Bernd Kort-
mann, and, in particular, Verena Haser and Manfred Kmg (University of
Freiburg, Germany) for extensive (and partly devastating) comments on an
earlier draft of this paper. Although I failed to be convinced by some of their
arguments, they certainly were an invaluable help in reformulating and sup-
plementing my thoughts and in many cases provided me with further insights
into the problems concerned.
1. Some people might say that this would be a reason rather than a cause. I do
not wish to enter into the causes-reasons debate here, but if there is a distinc-
tion to be made, it would have to be between causes for events and reasons
for human actions (Beckermann 1977). Being late is not an action, so what-
ever causes my being late would be a cause rather than a reason.
2. It is perhaps interesting to note, in this connection, that the German word for
'reality', Wirklichkeit, is etymologically related to the verb wirken 'take ef-
fect' which in tum is etymologically related to English work). Literally, Wirk-
lichkeit could be translated as 'a coherent whole of things that take effect'.
3. I apologise for the somewhat stereotypic character of this whole example. I
could try to give a causal explanation for this, but I seriously doubt the poten-
tial relevance of such an explanation.
4. The formulation for promises (Searle 1969: 59), e.g., is: "It is not obvious to
bothS and H that S will do A in the normal course of events".
5. To distinguish Sperber and Wilson's concept of Relevance from others, in
particular the colloquial notion of relevance, I will capitalise the word when-
ever referring to Sperber and Wilson's principle.
6. I wish to thank Verena Haser and Manfred Kmg (University of Freiburg) for
pointing this out to me.
The relevance ofcausality 31

7. Other approaches with a similar set of basic assumptions, are Grimes (1975);
Mann and Thompson (1988); Hobbs (1983); Graustein and Thiele (1979)
and many others.
8. Klare, Hugh 1., Stress violence and crime, in: Mayne, Richard (ed.): Europe
tomorrow, London 1972,48-63.
9. Dahl uses Mann and Thompson's (1988) terminology, where "rhetorical
relation" corresponds to what is now called "coherence relation" in most of
the literature.
10. I use capital letters to indicate that I am talking about principles within a
framework that are not necessarily identical with their colloquial counterparts.
11. Grice himself (1975: 46) speaks about "different kinds and focuses of rele-
vance" whose exact nature is not made very clear. The maxim of relevance is
the least explicit one in Grice's account. Maybe the five principles discussed
here could contribute to a clarification.


Alm-Arvius, Christina
1993 The English verb see: a study in multiple meaning. (Gothenburg
studies in English 64.) Goteborg: University.
Anscombe, G.E.M.
1971 Causality and determination. Cambridge: Cambridge University :
1975 Causality and determination. In: Ernest Sosa (ed.), Causation and
conditionals, 63-81. Oxford: Oxford University Press.
Bazerman, Charles
1988 Shaping written knowledge: the genre and activity ofthe experimen-
tal article in science. Madison, WI: University of Wisconsin Press.
Beckermann, Ansgar
1977 Grande und Ursachen: zum vermeintlichen grundsiitzlichen Unter-
schied zwischen mentalen Handlungserkliirungen und wissenschaft-
lich-kausalen Erkliirungen. Kronberg: Scriptor.
Breul, Carsten
1997 Grammatik und Bedeutung der kausalen Satzverbiinde. (Linguisti-
sche Arbeiten 368.) Tubingen: Niemeyer.
Dahl, Osten
1995 Causality in discourse. In: Gert Rickheit and Christopher Habel
(eds.), Focus and coherence in discourse processing, 251-260. (Re-
search in text theory 22.) Berlin: de Gruyter.
Danes, Frantisek
1970 Zur linguistischen Analyse der Textstruktur. Folia linguistica 4: 72-
Edwards, Derek and Jonathan Potter
1993 Language and causation: a discursive action model of description
and attribution. Psychological review 100: 23-41.
32 Paul Georg Meyer

Graustein, Gottfried and Wolfgang Thiele

1979 Towards an analysis of English texts. Zeitschrift fur Anglistik und
Amerikanistik 27: 62-74.
Grice, Herbert Paul
1975 Logic and conversation. Syntax and semantics 3: 41-58.
Grimes, Joseph
1975 The thread 0.[ discourse. (Janua linguarum. Series minor 207.) The
Hague: Mouton.
Hilton, Denis 1., Joseph M.F Jaspars and David D.Clarke
1990 Pragmatic conditional reasoning: context and content effects on the
interpretation of causal assertions. Journal of pragmatics 14:
Hilton, Denis 1. and Hans-Peter Erb
1996 Mental models and causal explanation: judgements ofprobable cause
and explanatory relevance. Thinking and reasoning 2: 273-308.
Hobbs, Jerry R.
1983 Why is discourse coherent? In: Fritz Neubauer (ed.), Coherence in
natural-language texts, 29-70. Hamburg: Buske.
Horn, Laurence H.
1984 Toward a new taxonomy for pragmatic inference: Q-based and
R-based implicature. In: Deborah Schiffrin (ed.), Meaning, form and
use in context: linguistic applications, 11-42. Washington: George-
town University Press.
KamIah, Andreas
1991 The causal relation as the most fundamental fact of the world. Com-
ments on Hans Reichenbach's paper: the space problem in the new
quantum mechanics. Erkenntnis 35: 49-60.
Kortmann, Bernd
1991 Free adjuncts and absolutes in English: problems of control and
interpretation. London: Routledge.
Lakoff, George
1982 Categories: an essay in cognitive linguistics. In: Linguistic Society of
Korea (ed.), Linguistics in the morning calm: selected papers from
SICOL-1981, 139-193. Seoul: Hanshin.
1987 Women, fire and dangerous things: what categories reveal about the
mind. Chicago: Chicago University Press.
Latour, Bruno
1987 Science in action: how to follow scientists and engineers through
society. Milton Keynes: Open University Press.
Lewis, David
1973 Causation. Journal ofphilosophy 70: 556-567.
Mach, Ernst
1901 Die Mechanik in ihrer Entwicklung: historisch-kritisch dargestellt,
4th ed. Leipzig: Brockhaus.
Mackie, John Leslie
1965 Causes and conditions. American philosophical quarterly 2:
1974 The cement ofthe universe: A study ofcausation. Oxford: Clarendon.
The relevance ofcausality 33

1975 Causes and conditions. In: Ernest Sosa (ed.), Causation and condi-
tionals, 15-38.
Mann, William C. and Sandra Thompson
1988 Rhetorical structure theory: toward a functional theory of text or-
ganization. Text 8: 243-281.
McCawley, James D.
1968 Lexical insertion in a transformational grammar without deep struc-
ture. Papers from the Regional Meeting of the Chicago Linguistic
Society 4: 71-80.
Meyer, Paul Georg
1975 Satzverknupfungsrelationen. Ein Interpretationsmode II fur situation-
sunabhtingige Texte. (Tubinger Beitrage zur Linguistik. 61.) Tubin-
gen: Narr.
1983 Sprachliches Handeln ohne Sprechsituation: Studien zur theoreti-
schen und empirischen Konstitution von illokutiven Funktionen in
'situationslosen' Texten. (Linguistische Arbeiten. 135.) Tubingen:
1996 Textfunktion-Textstruktur-Textanalyse: Zur Linguistik des Sach-
textes. In: Wolfgang Bomer and Klaus Vogel (eds.), Texte im
Fremdsprachenerwerb: Verstehen und Produzieren, 21-44. (Tubin-
ger Beitrage zur Linguistik 418.) Tubingen: Narr.
1997 Coming to know: studies in the lexical semantics and pragmatics of
academic English. (Forum fur Fachsprachenforschung 35.) Tubin-
gen: Narr.
Reichenbach, Hans
1991 The space problem in the new quantum mechanics. Erkenntnis 35:
Russell, Bertrand
1912 On the notion of cause. Proceedings of the Aristotelian Society 13:
1921 On the notion of cause. In: Mysticism and logic and other essays,
180-208. London: Longman.
Searle, John Rogers
1969 Speech acts: an essay in the philosophy of language. Cambridge:
Cambridge University Press.
1983 Intentionality: an essay in the philosophy ofmind. Cambridge: Cam-
bridge University Press.
Sosa, Ernest
1975 Introduction. In: Ernest Sosa (ed.), Causation and conditionals, 1-14.
Oxford: Oxford University Press.
Sosa, Ernest (ed.)
1975 Causation and conditionals. Oxford: Oxford University Press.
Sperber, Dan and Deirdre Wilson
1986 Relevance: communication and cognition. Oxford: Blackwell.
1995 Relevance: communication and cognition. 2. ed. Oxford: Blackwell.
34 Paul Georg Meyer

Stegmtiller, Wolfgang
1983 Probleme und Resultate der Wissenschaftstheorie und Analytischen
Philosophie. Bd. 1: Erkltirung Begrundung Kausalittit. 2., verb. und
erw. Aufl. Berlin: Springer.
Strawson, P.F.
1985 Causation and explanation. In: Bruce Vermazen and Merril B. Hin-
tikka (eds.), Essays on Davidson: actions and events, 115-135. Ox-
ford: Clarendon.
Swales, John
1981 Aspects of article introductions. (Aston ESP research reports 1.)
Birmingham: University of Aston in Birmingham / Language Studies
Von Wright, Georg H.
1973 On the logic and epistemology of the causal relation. In: P. Suppes et
al. (eds.), Logic, methodology and philosophy of science, 293-312.
Amsterdam: North Holland.
1975 On the logic and epistemology of the causal relation. In: Ernest Sosa
(ed.), Causation and conditionals, 95-113. Oxford: Oxford Univer-
sity Press.
Wierzbicka, Anna
1972 Semantic primitives. (Linguistische Forschungen 22.) Frankfurt:
1991 Cross-cultural pragmatics: the semantics of human interaction.
(Trends in linguistics: studies and monographs 53.) Berlin: Mouton
de Gruyter.
1996 Semantics: primes and universals. Oxford: Oxford University Press.
Woolgar, Steve
1988 Science: the very idea. ChichesterlNew York: Tavistock/Methuen.
Ziv, Yael
1988 On the rationality of 'relevance' and the relevance of 'rationality'.
Journal ofpragmatics 12: 535-545.
1993 Causality in language and in discourse. Journal ofliterary semantics
22: 177-185.
1993 Conjunction and causality: pragmatics and the lexicon. Pragmatics
and language learning monograph series 4: 17-27.
On the processing of causal relations

Leo G. M Noordman and Femke de Blijzer

This study deals with how different kinds of causal sentences are under-
stood. The central hypothesis is that sentences that more directly reflect a
causal relation in the world and in the cognitive representation of this
situation are easier to understand than other causal sentences. This hy-
pothesis is tested in reading-time experiments: longer reading times reflect a
greater complexity in the processing of the underlying infonnation. Three
factors are discussed that detennine how directly a causal sentence matches the
cognitive representation of causality: content vs. epistemic relations, linear
order of cause and effect in the sentence, and conceptual order of cause and
effect. It is demonstrated that each of these factors makes an independent
contribution to the complexity of understanding causal sentences. In addition,
it is demonstrated that these factors have an effect independently of a fourth
factor, namely the causal constraint between cause and effect.

1. Causality in cognition

Causality is an important ordering principle of human perception and hu-

man experience, and thus a central category in human cognition. It is fun-
damental both to the representation of human knowledge and to other cog-
nitive processes like predicting, explaining, and comprehending.
Our claim is that sentences that more directly reflect causality are easier
to understand than other sentences. We should therefore be explicit about
what we mean by understanding, and about how we conceive the cognitive
representation of causality. From a psycholinguistic point of view, we con-
sider the process of understanding as one in which the reader constructs a
representation of the information in the discourse. An important aspect of
that representation is that it is coherent, since the consecutive sentences in
a discourse are related to each other. Accordingly, the representation
contains relations between the sentences, or rather between the
information units that are expressed by the sentences. But that representa-
tion is also a representation of something else, i. e., of the world. Therefore,
the representation has relations with the world. The representation may be
related to the world in terms of a number of concepts, such as truth~n
which semantic theories have in general concentrated-possibility, and
36 Leo Noordman and Femke de Blijzer

plausibility. The distinction between relations in the representation and

relations of the representation to (a model of) the world is related to the
distinction made by Guenthner (1989) between D-relations (for discourse)
and T-relations (for truth). A similar distinction is made in theories about
psychological representations (Johnson-Laird 1983). In this conception,
understanding implies that the reader not only constructs a coherent repre-
sentation but also that the reader implicitly evaluates whether the informa-
tion in the discourse corresponds to the world, for example, whether it is
true or false, plausible or possible. Accordingly, understanding causal sen-
tences requires matching the information in the clauses with a cognitive
representation of the world, in this case with a cognitive representation of
the causality in the world.
This leads to the question of how causality is cognitively represented.
We will not develop an elaborate theory of the representation of causality.
We only assume that causal relations are represented as ordered cause-
effect pairs where the cause temporally precedes the effect. The represen-
tation of causal relations originates in our experience in the world. We
observe co-occurrences of events in the world. On the basis of these co-
occurrences we deduce causal relationships. We have a strong tendency to
interpret sequences of events in terms of causal relations, even when there
is no real causality involved (Michotte 1954). A child learns that when she
drops a glass, the effect is that it breaks into pieces. A child also learns to
use causal schemes in reasoning processes: if the glass is broken, it proba-
bly has fallen. The experience of causal relations between events in the
world is fundamental to the conceptual representation of causality. Given
the fact that causal relations derive from our experience of the world,
where causes temporally precede effects, we assume that causal relations
in our representation of the world reflect this experience and that they are
represented as cause-effect pairs with the cause preceding the effect.

2. Relevant factors in the processing of causal sentences

2.1. Content vs. epistemic relations

Causal sentences can be characterized with respect to a number of factors,

which presumably affect their processing. One factor is the distinction be-
tween content relations and epistemic relations (Sweetser 1990), some-
times also identified as semantic vs. pragmatic relations (Sanders,
Spooren, and Noordman 1992). A sentence that expresses a causal content
On the processing ofcausal relations 37

relation describes a real-world causality between two events or states in

the world. An example is (1):

(1) Because John worked hard, he passed the exam.

An epistemic causal relation expresses a conclusion by the writer/speaker

that is based on a causal relation in the world. An example is (2):

(2) Because John passed the exam, he must have worked hard

This sentence expresses that the speaker's knowledge of John's passing

causes the conclusion that he worked hard. A relation is a content relation
if the two clauses are related because of their propositional content. A
relation is an epistemic relation if the clauses are related by the speaker's
What can we expect with respect to the difference in processing be-
tween content and epistemic causals? Earlier, it was said that understand-
ing requires matching the information in the sentence with a model of the
world. Since a content causal sentence directly reflects a state of affairs in
the world, it may be easier to process than an epistemic causal sentence,
which does not directly express real-world causality. One can argue that an
epistemic causal expresses a real-world causality in a derived way. In fact,
an epistemic relation reflects a line of reasoning that is allowed by the co-
occurrence of events or situations in the world. The justification of that
reasoning is the contingency of events in the world. So, in this sense, an
epistemic relation is based on an underlying content relation. In under-
standing an epistemic relation, the reader has to check the possibility of the
underlying content relation in the world. Understanding an epistemic rela-
tion implies understanding the underlying content relation. We will illus-
trate this with two examples.
In sentence (2), the epistemic conclusion is justified by the underlying
content relation. The reason that one may infer from "John passed the
exam" to "he worked hard" is that ifyou work hard, you pass the exam.
In (3), there is the same underlying content relation: if you work hard,
you pass the exam.

(3) John worked hard, so he must have passed the exam.

The content relation is used in an act of concluding, which is expressed by

the connective so.
38 Leo Noordman and Femke de Blijzer

According to this analysis, content relations reflect experience in the

real world and epistemic relations are derived from these content relations.
Epistemic relations express real-world causality in a less direct way. One
might predict that processing an epistemic relation requires more time than
processing a content relation.
There is some support in the literature for this prediction. Noordman
(1979) asked readers to read conditional sentences that expressed content
and epistemic causal relations. An example of the first type is: "If John is
ill, he is not going to work". An example of the second type is: "If John is
not going to work, he is ill". The readers had to press a response button as
soon as they understood the sentence. The time was measured from the
moment the sentence was presented until the participant pressed a button.
Sentences expressing a content causal relation were processed 286 msec
faster than sentences expressing an epistemic causal relation. So, sentences
that express the causal relation more directly are processed faster than
sentences that express the causal relation in a less direct way. Similar re-
sults have been obtained by Traxler et ale (1997) for the understanding of
causal sentences and diagnostic sentences. In their experiment, these two
types of sentences corresponded to content and epistemic causals. The
interpretation by Traxler et ale is that diagnostic statements need a more
complex mental representation than causal statements; diagnostic state-
ments require the representation of an embedding phrase like: "someone
believes that. ..".
These data show that content and epistemic causal sentences are proc-
essed in a different way. A possible interpretation is that understanding
causal relations that describe events in the real world is easier than under-
standing causal relations that express a reasoning process. But we may not
yet derive this conclusion. When we scrutinize the sentences that are used
in the experiment by Noordman, it is clear that the difference between
content and epistemic relations is not the only difference between the sen-
tences. There are other factors that might play a role in the processing of
these causal sentences. One factor is what we call the conceptual order of
the clauses. Another factor is the linear order of the clauses. We will first
discuss these factors: conceptual order in Section 2.2. and linear order in
Section 2.3. Finally, we will discuss a factor that will be called causal con-
straint (Section 2.4.). In Section 3, we will present a reading-time experi-
ment in which these factors are investigated.
On the processing ofcausal relations 39

2.2. Conceptual order

Consider sentences (1) and (4):

(1) Because John worked hard, he passed the exam.

(4) John passed the exam, because he worked hard

Both sentences express a cause-effect relation. Conceptually, the direction

in which the causality is expressed is from cause to effect. In the concep-
tualisation underlying sentences (1) and (4), the effect is derived from the
cause. We will refer to this order as the conceptual cause-effect order.
There is a correspondence between the antecedent (John worked hard) and
the consequent (he passed the exam) in these sentences on the one hand,
and the cause and effect in the real world on the other hand. What con-
ceptually is the cause (working hard) is expressed as the antecedent in the
sentence; what conceptually is the effect (passing the exam) is expressed
as the consequent in the sentence. In these causal sentences, we define the
antecedent as the starting point in the causal deduction that is expressed by
the sentence and the consequent as the end point. In a content sentence, the
antecedent is the cause and the consequent is the effect, as in (1). In an
epistemic sentence, the antecedent is the argument and the consequent is
the claim, as in (2). Sentences that have the conceptual cause-effect order
exhibit what we call conceptual correspondence: they indicate that on the
basis of the cause the effect is derived. This order corresponds to the order
ofthe events in the world.
Sentence (2) expresses a effect-cause relation.

(2) Because John passed the exam, he must have worked hard

Conceptually, the cause is derived from the effect. The effect is considered
as a sign for the cause. (Therefore it was identified as an epistemic relation
in the first place). Sentence (2) has the conceptual effect-cause order.
There is an incongruence between what is cause and effect in the real
world and what is expressed in the sentence as antecedent and consequent.
It should be noted that the distinction between conceptual cause-effect
order and conceptual effect-cause order is not the same as the distinction
between content and epistemic relations. In fact, sentence (3) expresses an
epistemic relation, but the effect is derived from the cause.
40 Leo Noordman and Femke de Bli.Jzer

(3) John worked hard, so he must have passed the exam.

Therefore, sentence (3) has the conceptual cause-effect order. On the other
hand, the two distinctions are not completely independent of each other, as
is shown in Table 1. Content relations only occur in the conceptual cause-
effect order, because content relations directly reflect real-world causality.
Epistemic relations can have either conceptual cause-effect order or effect-
cause order.

Table 1. Examples of content and epistemic relations, in different conceptual and

linear orders.

conceptual order
cause-effect effect-cause
linear cause-effect effect-cause cause-effect effec~cause
content Because John John passed the -
relation worked hard, he exam, because
passed the he worked hard.
epistemic John worked John must have John must have John passed the
relation hard, so he must passed the worked hard, exam, so he
have passed the exam, since he since he passed must have
exam. worked hard. the exam. worked hard.

Since sentences with the conceptual cause-effect order reflect causality in

the real world more directly than sentences with the conceptual effect-
cause order, and since we assume that the more directly a sentence
matches a situation in the world, the easier it is to process, we predict that
sentences with cause-effect order (such as (1)) are easier to process than
sentences with effect-cause order (such as (2)). We prefer deriving effects
from causes to deriving causes from effects.
The basis for this assumption is our sensorimotor experience in the
world. We learn the notion of causality and we learn causal relations by
acting in the world, by observing that causes precede effects. If the cause-
effect order is conceptually more fundamental than the effect-cause order,
it seems likely that we prefer to reason from cause to effect instead of from
effect to cause.
Support for the assumption that conceptual cause-effect order is more
fundamental than effect-cause order is obtained by Tversky and Kahne-
man (1982). They found that reasoning from cause to effect is easier than
from effect to cause, even in situations in which the likelihood of the cause
given the effect is the same as the likelihood of the effect given the cause.
On the processing ofcausal relations 41

Tversky and Kahneman conclude from their studies that people "infer
effects from causes with greater confidence than causes from effects"
(page 118). People make use of cause-effect schemas, and causal infer-
ences that follow the normal cause-effect sequence are easier to make than
diagnostic inferences that reverse this sequence.

2.3. Linear order

The conceptual order of the clauses should be distinguished from the lin-
ear order. Both sentences (1) and (4), repeated below, express the concep-
tual order of cause-effect.

(1) Because John worked hard, he passed the exam

(4) John passed the exam, because he worked hard

The sentences differ in the linear order of the clauses. In (1), the first
clause expresses the cause and the second clause the effect. In (4), the first
clause is the effect and the second clause is the cause. Sentence (1) thus
has an order which is iconic with the world; sentence (4) is non-iconic.
What can we predict with respect to the processing of causal sentences?
An obvious prediction is that an iconic order facilitates processing. The
reason is similar to the one given in the section on conceptual order. If the
order of the clauses corresponds to the order of causality in the world (or
more specifically, to a model of causality), the matching process between
language and knowledge is easier (and thus the understanding of the sen-
However, on the basis of the literature one might make a different pre-
diction. Magliano et al. (1993) argue that inferences are made only about
causes and not about effects. They argue that readers make inferences
about causes because these explain the current situation (and the sentence
that expresses this situation). The inference forms a backward link. For-
ward inferences tend not to be made, because, at any point in a text, there
are numerous inferences one can make about the possible effects of the
situation expressed in the current sentence. Lexical decision data did in-
deed support the conclusion that inferences about causal antecedents are
made but inferences about effects are not. However, in this experiment the
causal constraints were rather weak (.30 and .26). We predict that in a
situation in which the cause strongly determines the effect and the effect
42 Leo Noordman and Femke de Blijzer

determines the cause, iconicity will facilitate processing of a causal sen-

tence. This prediction rests on the assumption that the cause-effect order
reflects the structure of our causal knowledge. These predictions were
tested by Noordman, Vonk and Meyer Viol (Noordman and Vonk 1998;
Meyer Viol 1984). In the experimental texts, a causal relation was ex-
pressed in two different orders. The following fragment is an example:

"In order to earn some money, John was cutting out weeds in his mother's
garden. It was a tough job, because the stinging-nettles were a metre high.
After two hours, he took a short break. The job was half finished. The sun
stood high in the sky and it was sweltering hot. He wiped away the sweat from
his forehead with his hand. He had touched the stinging-nettles with his hand.
His hand itched terribly."

In this fragment the cause sentence ("He had touched the stinging-nettles")
preceded the' effect sentence ("His hand itched terribly"). In the other ver-
sion of the same fragment, the cause sentence and the effect sentence were
reversed ("His hand itched terribly. He had touched the stinging-nettles").
It should be noted that the causal relations in both conditions have con-
ceptual cause-effect order. If we make the causality of the relation explicit,
we obtain the sequences: "because he had touched the stinging-nettles, his
hand itched terribly", and "his hand itched terribly, because he had
touched the stinging-nettles".
The materials in this experiment were constructed in such a way that,
first, the effects could very well be predicted on the basis of the causes
and, second, that the causes could very well be predicted on the basis of
the effects. This was achieved by two separate pilot studies to test and
improve the materials. The experiment showed that a "cause sentence
speeds up the processing of a subsequent consequence sentence, but a
consequence sentence does not speed up the processing of a subsequent
cause sentence... What this experiment demonstrates is a kind of iconicity
between cognitive structure and language" (Noordman and Vonk 1998:
205). If the linear order of the clauses is iconic with our model of causality
in the world, processing the sentence is speeded up. This has been demon-
strated in texts in which the effect is highly predictable given the cause,
and the cause is highly predictable given the effect. In the experiment of
Section 3, reported more completely in de Blijzer (1999), predictability
will no longer be held constant, but will be an independent variable.
Summarizing, we have made three distinctions: content vs. epistemic
relations, conceptual order, and linear order. In principle, these three dis-
tinctions yield a 2x2x2 matrix of eight different kinds of causal sentences
On the processing ofcausal relations 43

(see Table 1). As has already been noted, two cells are empty. A content
relation always has cause-effect order. If the conceptual order is effect-
cause, the relation is always epistemic. In Section 2.1, we discussed an
experiment that compared two kinds of sentences that differ on all three
factors. What we wish to do in the remainder of this paper is to tIy and
find evidence for the factors separately. Also, we would like to further
investigate a fourth factor, the predictability of causes and effects.

2.4. Causal constraint: the predictability ofcauses and effects

A fourth factor that may affect the processing of causal sentences is the
predictability of causes and effects in causal relations. How likely is the
effect given the cause and how likely is the cause given the effect? In par-
ticular, if a speaker draws a conclusion on the basis of an epistemic rela-
tion, it is crucial that there is a solid ground for drawing that conclusion
and that there are few possible alternatives for the conclusion. More spe-
cifically, when the derivation is from cause to effect, it is crucial that there
are few conditions that prevent the cause from having an effect. Following
Cummins et al. (1991), we will call such conditions "disabling condi-
tions". When the derivation is from effect to cause, it is crucial that there
are few other causes. We will call these causes "alternative causes". In the
previous experiment these factors were held constant.
Sentences (5) and (6) illustrate that disabling conditions affect the
plausibility of a sentence that expresses a derivation from cause to effect
(i.e., where the effect is derived from the cause).

(5) The woman must have got a sunstroke, since she was in the sun for
a long time.

(6) The girl must have got wet, since she jumped into the river.

In (5) it is quite likely that there are disabling conditions which could pre-
vent the effect (e.g., wearing a hat, sitting in front of a ventilator, throwing
water over your head once in a while or only being in the sun when it is
not too hot). So when a person has been in the sun for a long time, it is not
at all certain that the effect occurs and that this person does indeed get a
sunstroke. This line of reasoning, expressed by sentence (5), is rather im-
plausible. In (6), however, it is rather unlikely that·there are disabling con-
ditions that could prevent the effect. If someone jumps into the river, it is
44 Leo Noordman and Femke de Blijzer

highly likely that this person gets wet. Therefore, when we know the cause
we can easily draw the conclusion that the effect must have taken place.
When we derive the cause from the effect, the plausibility of a sentence
is affected by alternative causes, as is illustrated by sentences (7) and (8).

(7) The woman got a sunstroke, so she must have been in the sun for a
long time.

(8) The girl got wet, so she must have jumped into the river.

In sentence (7) it is unlikely that there are alternative causes for a sun-
stroke. Therefore, if someone has a sunstroke, we can easily draw a con-
clusion about the cause: this person has been in the sun for a long time.
This line of reasoning is very plausible. In sentence (8), however, it is
likely that there are alternative causes for being wet (e.g., rain, snow, tak-
ing a shower or a bath, a bucket of water). So, when a person is wet, it is
not at all certain that this person has jumped into the river. The line of
reasoning is not very plausible. It should be noted that these observations
do not regard the acceptability of the sentences as linguistic entities (they
all are acceptable), but the plausibility of the sentences as descriptions of
the world.

2.4.1. Causal constraint in logical deduction

Cummins (1995) and Cummins et al. (1991) have shown that disabling
conditions and alternative causes affect logical reasoning. For example,
people have difficulty with the following deduction:

(9) If you liein the sun for a long time, you get a sunstroke. Mary has
been lying in the sun for a long time. So Mary will get a sunstroke.

Relatively many subjects judged such a conclusion incorrect, although it is

a logically valid deduction. The interpretation is that there are many dis-
abling conditions. On the other hand, if there are few alternative causes,
relatively many subjects made the logically incorrect deduction from ef-
fect to cause:

(10) If you lie in the sun for a long time, you get a sunstroke. Mary got a
sunstroke. So Mary must have been lying in the sun for a long time.
On the processing ofcausal relations 45

Many subjects judged such a conclusion correct, because there are few
alternative causes. Cummins et al. showed that disabling conditions and
alternative causes affect judgments on correct and incorrect conclusions.
But in this study we are not interested in true/false judgments concerning
logical conclusions. We are interested in the understanding of causal sen-
tences. The question then is whether disabling conditions and alternative
causes also affect the understanding of causal sentences. Earlier we argued
that language understanding implies an implicit evaluation of the sentence
on the basis of a model of the world. Therefore, it is quite likely that the
understanding of causal sentences is sensitive to factors that affect the
validity of conclusions, and therefore that the understanding of causal
sentences is sensitive to disabling conditions and alternative causes. For
this reason, we incorporated causal constraint, operationalized in terms of
disabling conditions and alternative causes, as a variable in the experiment
to be reported in Section 3.

2.4.2. Causal scenarios

Sentences (5) to (8) illustrate two causal scenarios. The scenario underly-
ing (5) and (7)-if one lies in the sun for a long time, one gets a
sunstroke-has many disabling conditions (DC) and few alternative
causes (AC); the scenario underlying (6) and (8)-if one jumps into the
river, one gets wet-has few disabling conditions and many alternative
causes. These two kinds of scenarios will be used to study the role of
causal constraint. We collected a set of 71 causal scenarios. In order to
find appropriate sentences for the experiment, we presented these
scenarios to twenty participants in two judgment tasks. In one task,
participants had to assume that the causal event occurred and estimate on a
four-point scale how certain the effect was. For example: Suppose that the
girl jumps into the river. How certain is it that she will get wet? The higher
the score, the higher the causal constraint, the smaller the impact of
disabling conditions. The second task was just the reverse. Participants had
to assume that the effect event had occurred and to estimate on a four-
point scale how certain the cause event was. For example, suppose the girl
has got wet, how certain is it that she jumped into the river? The higher the
score, the higher the causal constraint, the smaller the impact of alternative
causes. For each scenario, the mean DC (disabling conditions) and AC
(alternative causes) scores over the twenty participants were computed as
well as the mean difference between the AC scores and the DC scores
46 Leo Noordman and Femke de Blijzer

(AC-DC). The 24 scenarios with the highest negative AC-DC difference

scores and the 24 scenarios with the highest positive AC-DC difference
scores were selected for the experiment. These groups of scenarios
constituted the two categories of the experimental materials: high on
disabling conditions and low on alternative causes on the one hand, and
low on disabling conditions and high on alternative causes on the other

3. Experimental study: the processing of causal sentences

The aim of this experiment was to find out whether the processing of con-
tent relations differs from that of epistemic relations and whether the proc-
essing of relations that have cause-effect conceptual order differs from
relations that have effect-cause conceptual order. Linear order was kept
constant: in all cases the first clause mentioned the effect and the second
clause the cause.
The experimental conditions in this experiment are illustrated in Table
2. The experiment was conducted in Dutch, but the table presents English
translations as well. There is only one type of content sentence: a sentence
type that expresses a cause-effect conceptual order. There were three types
of epistemic sentences. The fact that a sentence expresses an epistemic
relation was marked by the modal must have (in Dutch zal we/), or by the
conjunctions so or since (in Dutch: dus and want). In sentence type 3, the
epistemic character was indicated only by the conjunction so. The four
sentence types were applied to the two categories of scenarios.
On the processing ofcausal relations 47

Table 2. The experimental conditions according to type of relation (content vs.

epistemic), conceptual order (cause-effect vs. effect-cause) and causal
constraint (scenarios low on disabling conditions/high on alternative
causes vs. high on disabling conditions/low on alternative causes).

Causal constraint: low disabling conditions / high alternative causes

Relation Conceptual
type order
1 content cause-effect Het meisje is nat geworden, omdat ze in de rivier is
The girl got wet, because she jumped into the river.
2 epistemic cause-effect Het meisje zal weI nat zijn geworden, want ze is in de
rivier gesprongen.
The girl must have got wet, since she jumped into the
3 epistemic effect-cause Het meisje is nat geworden, dus ze is in de rivier
The girl got wet, so she jumped into the river.
4 epistemic effect-cause Het meisje is nat geworden, dus ze zal weI in de rivier
zijn gesprongen.
The girl got wet, so she must have jumped into the

Causal constraint: high disabling conditions / low alternative causes

Relation Conceptual
type order
1 content cause-effect De vrouw heeft een zonnesteek, omdat ze lang in de zon
heeft gezeten.
The woman got a sunstroke, because she was in the sun
for a long time.
2 epistemic cause-effect De vrouw zal weI een zonnesteek hebben, want ze heeft
lang in de zon gezeten.
The woman must have got a sunstroke, since she was in
the sun for a long time.
3 epistemic effect-cause De vrouw heeft een zonnesteek, dus ze heeft lang in de
zon gezeten.
The woman got a sunstroke, so she was in the sun for a
long time.
4 epistemic effect-cause De vrouw heeft een zonnesteek, dus ze zal wei lang in
de zon hebben gezeten.
The woman got a sunstroke, so she must have been in
the sun for a long time.

The causal relations were expressed by sentences without context. That is

a reasonable first step to find out the extent to which different factors play
a role in understanding, e.g., the difference between content and epistemic
relations and conceptual order. Further research will be required, for ex-
48 Leo Noordman and Femke de Blijzer

ample, to find out whether the context can overrule the effects of these
What kind of reading task is appropriate in this situation? We are inter-
ested in understanding causal relations. Understanding requires that the
sentence is related to a model of the world. Accordingly, when readers
understand a sentence, they should be able to evaluate whether the causal
relation expressed by the sentence is a possible causal relation in the
world. In the experiment, that was the task the readers were instructed to
perform: Is a causal relation possible between the two clauses? Since this
task requires that there are also impossible causal relations, so-called filler
items are incorporated in the experiment that expressed impossible rela-
tions (see Section 3.3.)

3.1. Procedure

The sentences were presented using a moving window paradigm. In this

paradigm, the sentences are presented in three parts: the first clause, the
conjunction and the second clause. Only one part is visible at a time; in the
other parts, the letters are replaced by hyphens. In the first presentation,
only the letters of the first part are visible; the letters of the other parts are
hyphens. As soon as the participant has pressed a button, the reading time
for the first part is registered and the hyphens of the second part are re-
placed by letters. In this way, the three parts were read consecutively and it
was possible for the reading times for the three parts to be measured sepa-
rately. The reading times for the third part are especially interesting, be-
cause at that moment and only then can the causal relation be understood
and does it become clear what kind of relation was expressed by the sen-
tence. Consequently, differences in processing time are expected to occur
in the third part. In this experiment, the third part is always the causal

3.2. Predictions

The four types of sentences (see Table 2) define the four conditions in this
experiment. The predictions for the reading times for the identical second
clauses (excluding the conjunction) in conditions 1, 2, and 3 are as fol-
lows. On the basis of conceptual order, the reading times for conditions 1
and 2 are expected to be shorter than for condition 3. The reading times for
On the processing ofcausal relations 49

the content condition 1 should be shorter than for the epistemic condition
2. This leads to a predicted increase in reading times for the conditions 1 to
2 to 3. The second clause in condition 4 is two words longer than in the
other conditions, so these reading times cannot be compared to the reading
times for the other second clauses, but the total sentence in condition 4 has
the same length (in terms of words and letters) as in condition 2. Based on
conceptual order, the reading time for sentences in condition 2 is predicted
to be shorter than in condition 4.
It is likely that these predictions are qualified by the causal constraint:
disabling conditions may make the cause-effect order more difficult, and
alternative causes may make the effect-cause order more difficult. More
specifically, for the scenarios that have low DC scores and high AC
scores, the low DC scores will make the cause-effect items easier, and the
high AC scores will make the effect-cause order more difficult. For the
scenarios that have high DC scores and low AC scores, the high DC scores
will make the cause-effect items more difficult and the low AC scores will
make the effect-cause items easier. So the difference between conditions 1
and 2 on the one hand and 3 on the other hand will be greater for the sce-
narios with low DC scores and high AC scores than for the scenarios with
high DC scores and low AC scores. In other words, an interaction is pre-
dicted between conceptual order and the two categories of causal con-
straint. No effect of causal constraint is expected on the comparison be-
tween content and epistemic sentences (condition 1 versus 2), because
both sentence types have the conceptual cause-effect order.
The effect of causal constraint can be investigated in an additional way.
The hypothesis is that the cause-effect order becomes more difficult when
the DC scores are higher. Consequently, there should be a positive corre-
lation between the DC scores and the reading times for the causal clauses
of conditions 1 and 2. The effect-cause order is supposed to become more
difficult when the AC scores are higher. Consequently, there should be a
positive correlation between the AC scores and the reading times for the
causal clauses of conditions 3 and 4.

3.3. Materials

As was explained in Section 2.4.2., the materials in the experiment con-

sisted of two groups of 24 scenarios each: the group of scenarios with high
DC scores and low AC scores, identified as high DC/lowAC and the group
of scenarios with low DC scores and high AC scores, identified as low
50 Leo Noordman and Femke de Blijzer

DC/high AC. For each scenario, four sentences were formulated corre-
sponding to the four experimental conditions. Four lists of materials were
composed, by pairing each of the 48 scenarios with a different experi-
mental condition in each list, so as to balance the lists with respect to the
scenarios and experimental conditions. Although sentence length does not
affect the way the hypotheses are tested, it is advisable to keep the sen-
tence length as constant as possible in order to reduce experimental noise.
The mean lengths of the causal clauses for the low DC/high AC and the
high DC/low AC scenarios were 8.25 and 8.33 syllables respectively. The
mean lengths of the complete sentences were also rather similar: 15.21 and
16.21 syllables respectively.
Given the fact that the task was to judge whether the causal relation
was possible, and all the experimental sentences required an affirmative
answer, filler sentences requiring a negative answer were needed. The
filler sentences expressed two clauses between which no causal relation
was possible. The clauses themselves made perfect sense and were quite
possible, so that the negative answer could only be based on the impossi-
bility of a causal relation, as it should be. An example is: "The letters have
become blurred because someone held the book upside down", and "Her
nose was bleeding because she had a hair cut". There were 32 filler sce-
narios. The filler scenarios were incorporated in the four lists and were
balanced with the experimental conditions in the same way as the experi-
mental scenarios.
Each of the four lists was presented to 8 participants. The participants
were students ofTilburg University, between 18 and 27 years of age.

3.4. Results and discussion

We are mainly interested in reading times. Since the data of participants

with many errors are not reliable, the reading time data of participants
whose error rate was higher than 5 % were not analysed, except in one
group in which two participants had an error rate of 7.5 %. Each reading
time of a particular reader for a particular item that wa~ extremely high or
extremely low in comparison to the average of the participant and the av-
erage of the item was replaced by an estimate based on the average of the
reader and the average of the item. In a similar way, reading times for
incorrect answers were replaced, since they are not informative. The per-
centage of substitutions was 3.1 %.
The reading times for the causal clauses are presented in Table 3.
On the processing ofcausal relations 51

Table 3. Reading times (msec) for the causal clauses.

1 2 3 4
content epistemic epistemic epistemic
cause-effect cause-effect effect-cause effect cause + modal
low disabling conditions / high alternative causes
1567 1730 2029 1985

High disabling conditions / low alternative causes

1380 1489 1620 1672

1473 1610 1824 1829

The reading times for the clauses in condition 1 are a significant 137 msec
shorter than for clauses in condition 2: F 1(1,28)=8.44, p<.Ol~
F2(1,40)=8.82, p<.Ol, supporting the hypothesis that content relations are
easier to process than epistemic relations. The reading times for clauses in
condition 2 are 214 msec shorter than for clauses in condition type 3:
F 1(1,28)=27.72, p<.OOl~ F2(1,40)=13.37, p<.OOI, supporting the hypothe-
sis that conceptual order plays a role: cause-effect relations are easier to
process than effect-cause relations.
There was a significant interaction between the experimental conditions
and the causal constraint, just as predicted. The difference between cause-
effect and effect-cause conceptual orders is greater for low DC/high AC
scenarios than for high DC /low AC scenarios. Specifically, the difference
between clauses in conditions 1 and 3 is greater for the low DC/high AC
than for the high DC/low AC scenarios: F 1(1,28)=7.53, p=.OI~
F 2(1,40)=4.66, p<.05. A similar pattern is observed for the difference be-
tween clauses in conditions 2 and 3, although the F 1 shows only a trend
and the F2 is not significant: F 1(1,28)=3.69, p=.065~ F2(1,40)=2.03, p=.16.
These interactions support the hypothesis that the cause-effect order is
more difficult when there are more disabling conditions and that the effect-
cause order is more difficult when there are more alternative causes.
As predicted, the difference between the clauses in conditions 1 and 2
is constant for the two categories of causal constraint. That should be the
case, since sentences in both conditions express cause-effect relations. One
prediction was not confirmed. The reading time for the complete sentence
2 was not shorter than for the complete sentence 4, as emerges from the
data in Table 4.
52 Leo Noordman and Femke de Blijzer

Table 4. Reading times (msec) for the complete sentences in conditions 2 and 4

2 4
epistemic epistemic
cause-effect effect-cause
low disabling conditions / high alternative causes
3770 3709
high disabling conditions / low alternative causes
3621 3441
3695 3575

The basis for this prediction was that the conceptual order in sentence 2
but not in sentence 4 corresponds to the order of cause-effect in the world.
The prediction was justified by the fact that sentences 2 and 4 had the
same length. In sentence 2, the modal must have occurred in the first
clause; in sentence 4 it occurred in the second clause. The assumption was
that the effect of must have is constant. But this assumption appears to be
wrong. In fact, the reading times for the second clauses in condition 3 and
4 did not differ. This is surprising, since the Dutch clause in 4 (ze zal weI
lang in de zon hebben gezeten; "she must have been in the sun for a long
time") is two words longer than clause 3 (ze heeft lang in de zon gezeten;
"she was in the sun for a long time"). Interestingly, the same difference in
length occurred (in the original Dutch examples) in the first clauses of
conditions 2 (De vrouw zal weI een zonnesteek hebben; "The woman must
have got a sunstroke") and 4 (De vrouw heeft een zonnesteek; "The
woman got a sunstroke"). Here the longer clause did indeed require a
longer reading time; the reading times for the first clause in the conditions
2 and 4 were 1464 msec and 1130 msec respectively.
The interpretation is that must have in the second clause (condition 4) is
an indication that the clause expresses a conclusion on the part of the
writer. So must have facilitates the epistemic interpretation. If must have is
absent (in condition 3), the reader has to infer on the basis of so and hislher
knowledge of the world that the sentence expresses an epistemic conclu-
sion. A similar effect has been observed by Cozijn (in preparation) for the
conjunction because. The causal relation between two clauses can, but
need not be expressed by the conjunction because. If there is no conjunc-
tion, the sentence is shorter. But Cozijn found that the presence of because
speeds up reading, counteracting the increase in length. The facilitating
effect of must have is not expected for the first clause (conditions 2 and 4).
The first clause of condition 4 is initially interpreted as a content relation,
and during the first clause there is not yet any evidence that the sentence
On the processing ofcausal relations 53

expresses an epistemic relation, which otherwise would slow down the

processing ofthis first clause in condition 4.
Now we return to the predicted difference in reading time for the
complete sentences in conditions 2 and 4. The modal must have in the first
clause increases the reading time for sentence 2 and this cancels the
advantage of the conceptual order, leading to a reading time as long as that
for sentence 4. Since the effect of the conceptual order is smaller for the
high DC/low AC scenarios than for the low DC/high AC scenarios, the
increase due to the extra must have for the high DC/low AC scenarios is
greater than the advantage of the conceptual order. For the high DC/low
AC scenarios, this results in a significantly longer reading time for
sentence 2 than for sentence 4 (Fl(1,28)=4.97, p<.05; F 2(1,20)=4.86,
Finally, it was predicted that disabling conditions make the cause-effect
order more difficult, and that alternative causes make the effect-cause or-
der more difficult. This implies that the cause-effect sentences (conditions
1 and 2) should be easier for the low DC/high AC scenarios than for the
high DC/low AC scenarios, and that effect-cause sentences (conditions 3
and 4) should be easier for the high DC/low AC scenarios than for the low
DC/high AC scenarios. However, all the reading times of the high DC/low
AC scenarios were shorter than those of the low DC/high AC scenarios: F 1
(1,28)=39.89, p<.OOl; F2(1,40)=13.60, p<.OOl. This can be explained in
terms of the linear order of the clauses. It should be remembered that the
effect clause was always presented as the first clause and the causal clause
as the second clause. If the reader has read the effect clause, and tries to
anticipate the subsequent causal clause, the reading time for the causal
clause will be shorter if there are only few causes than if there are many
causes. This is exactly what the results show. So there is an effect of the
order of presentation of the sentences. Although we kept the order of pres-
entation constant, these results do show a linear effect in the processing of
the causal relations. The interesting result is that the effect of linear order
apparently overrules the effect of causal constraint within each of the con-
ceptual orders.
An independent confirmation of this result is obtained from the correla-
tions. It was predicted that there would be a positive correlation between
the DC scores and the reading times for the cause-effect order (conditions
1 and 2), and a positive correlation between the AC scores and the reading
times for the effect-cause order (conditions 3 and 4). But, as Table 5 dem-
onstrates, both orders correlate positively with the number of alternative
54 Leo Noordman and Femke de Blijzer

Table 5. Correlations between clause reading times (msec) and scores for
alternative causes (AC) and disabling conditions (DC) respectively.

1 2 3 4
content epistemic epistemic epistemic
cause-effect cause-effect effect-cause effect-cause + modal
AC .41* .37* .51 * .50*
DC -.01 .00 -.08 -.01

* p<.Ol
This makes perfect sense, since the effect clause was presented first, lead-
ing the reader to anticipate the cause sentence. The greater the predictabil-
ity ofthe cause sentence, the shorter, in fact, the reading time.

4. Conclusion

In our investigation of the processing of causal relations, a first distinction

was drawn between content causal and epistemic causal relations. Experi-
mental evidence from the literature (Section 2.1.) showed that processing
epistemic relations is more complex than processing content relations. But
it was argued that several factors were involved in the processing of the
causal relations that were investigated. Besides the distinction between
content and epistemic relations, distinctions between callSal relations can
be made on the basis of the correspondence of the conceptual order of the
clauses and a model of the world, on the basis of the correspondence of the
linear order of the clauses and a model of the world, and on the basis of
causal constraint between the cause and the effect. In this study, empirical
evidence was obtained for the effect ofthese four factors separately.
First, epistemic relations require more processing time than content re-
lations, as was indicated by the results of the experiment described in Sec-
tion 3. Understanding a sentence that describes a relation between states of
affairs in the world is easier than understanding a sentence that expresses a
reasoning about that relation. The explanation is that an epistemic relation
implies a content relation. The basis for an epistemic causal relation (a
reasoning) is a cause-effect relation in the world.
Second, there is an effect of conceptual order, as was indicated by the
results of the same experiment. Processing a causal relation from cause to
effect is easier than processing from effect to cause.
On the processing ofcausal relations 55

Third, the experiment also showed that the causal constraint has an ef-
fect. Processing from effect to cause is easy for scenarios that score low on
alternative causes. Processing from cause to effect is easy for scenarios
that score low on disabling conditions. But in both cases, the effect of
content vs. epistemic and the effect of conceptual order was observed.
Fourth, there is an effect of linear order. Presenting the cause first fa-
cilitates processing, as was indicated by the results of the experiment re-
ported in Section 2.3. An effect of linear order in processing was also
demonstrated in the experiment described in Section 3. In that experiment,
the effect clause was always presented as the first clause. The reading time
for the causal clause was shorter as it was the more predictable. Appar-
ently, readers try to anticipate the clause coming next.


We gratefully acknowledge many valuable contributions by Wietske Vonk

during all phases of this research and many valuable comments on an ear-
lier version of this paper by Lieselotte Anderwald, Elizabeth Couper-
Kuhlen, Verena Haser, and Bernd Kortmann.


Blijzer, F. de
1999 Causale scenario's. Het effect van de conceptuele volgorde en de
sterkte van de oorzaak-gevolg koppeling op de verwerking van seman-
tische en epistemische relaties [Causal scenarios. The effect of concep-
tual order and causal constraint on the processing of semantic and
epistemic relations]. MA thesis, Tilburg University, The Netherlands.
Cozijn, R
in preparation Integration and inference processes in understanding causal
relations. Ph.D. dissertation, Tilburg University, The Netherlands.
Cummins, D.D.
1995 Naive theories and causal deduction. Memory and Cognition 23: 646-
Cummins, D.D., T. Lubart, O. Alksnis and R Rist
1991 Conditional reasoning and causation. Memory and Cognition 19: 274-
Guenthner, F.
1989 Discourse: understanding in context. In: H. Schnelle and N.O. Bernsen
(eels.), Logic and Linguistics. Research Directions in Cognitive Science,
127-142. (European Perspectives 2.) Hove: Erlbaum.
56 Leo Noordman and Femke de Blijzer

Johnson-Laird, P.N.
1983 Mental Models. Cambridge: Harvard University Press.
Magliano, J.P., W.B. Baggett, B.K. Johnson and A.C. Graesser
1993 The time course of generating causal antecedent and causal conse-
quence inferences. Discourse Processes 16: 35-53.
Meyer Viol, W.P.M.
1984 Foregrounding and causality in inference making. M.A. thesis, Max
Planck Institute for Psycholinguistics, Nijmegen, The Netherlands.
Michotte, A.
1954 La Perception de la Causalite. Louvain: Publications Universitaires de
Noordman, L.G.M.
1979 lnferringfrom Language. Berlin: Springer.
Noordman, L.G.M. and W. Vonk
1998 Memory-based processing in understanding causal infonnation. Dis-
course Processes 26: 191-212.
Sanders, T.J.M., W.P.M. Spooren and L.G.M. Noordman
1992 Toward a taxonomy of coherence relations. Discourse Processes 15: 1-
Sweetser, E.
1990 From Etymology to Pragmatics. Metaphorical and Cultural Aspects of
Semantic Structure. Cambridge: Cambridge University Press.
Traxler, MJ., A.J. Sanford, J.P. Aked and L.M. Moxey
1997 Processing causal and diagnostic statements in discourse. Journal of
Experimental Psychology: Learning, Memory, and Cognition 23: 88-
Tversky, A. and D. Kahneman
1982 Causal schemas in judgments under uncertainty. In: D. Kahneman, P.
Slovic and A. Tversky (008.), Judgment under Uncertainty: Heuristics
and Biases, 117-128. Cambridge: Cambridge University Press.
Domains of use or subjectivity? The distribution
of three Dutch causal connectives explained*

Henk Pander Maat and Ted Sanders

This paper proposes a new account of the differences in meaning between

the three most frequent causal connectives in Dutch: daardoor, daarom
and dus. In an earlier corpus-analytic study, it was found that a distinction
between domains, such as the one proposed by Sweetser (1990), cannot
account for the distribution of these connectives.
Alternatively, the difference between the connectives is conceived of
here in terms of subjectivity or, more precisely, in terms of the distance
between the speaker and the Subject of Consciousness (SOC) responsible
for the causal relation. In such a subjectivity account, the three connectives
can be characterized as follows. In causal relations expressed by daardoor
'as a result' there is no SOC present because the causality is located outside
of any SOC. In relations expressed by daarom 'that's why' an SOC is
involved, but it is located at a certain distance from the Speaker~ in the case
of dus 'so', this distance is small or even absent. Hypotheses derived from
the subjectivity approach are supported by a new corpus-analytic study.

1. Causal connectives

The expression of causal relations in language is crucial for

communication. The conceptualization of causality is a fundamental
aspect of human cognition, and without much risk of overstatement it can
be claimed that every language has specific linguistic means to express
causal relations. When language users want to relate two discourse
segments, say clauses, in a causal way, as in (1) below, they can use
causal connectives like English because, so and since" or lexical cue
phrases like English as a result, that's why and on the grounds that.
Although all these linguistic means may express causality in one way or
another, it is clear that they cannot be used interchangeably. In English, it
is perfectly fme to use As a result / That's why / So to connect the
segments of (1), but it is impossible to use As a result to connect the
segments of (2). (This impossibility is expressed by the symbol #, which
signals an uninterpretable sequence; this symbol should not be confused
with *, which signals an ungrammatical sequence.) Also, it is at least
58 Henk Pander Maat and Ted Sanders

doubtful whether That's why could be used in (2), as indicated by the ?,

whereas ~~o would fit neatly.

(1) The neighbours suddenly left for Paris last Friday. As a result /
That's why / So they are not at home.

(2) The lights in the neighbours' living room are out. # As a result / ?
That's why / So they are not at home.

In other words, there are restrictions on the use of a cue phrase and
connectives. Ever since Halliday and Hasan's (1976) seminal work, it has
repeatedly been argued that the lexicon of connectives and cue phrases is
ordered according to the type of relation they can express (e.g. Martin
1992; Knott and Dale, 1994). For instance, temporal connectives can be
distinguished from causal connectives. Yet, as the examples (1) and (2)
show, the restrictions on the use of connectives and cue phrases are also
more subtle, i. e. they also hold within the class of causal relations as such.
Not all causal markers express the same type of causal relation. How then
can these restrictions be described?
In this chapter we report a study on the meaning and use of three
Dutch causal connectives: daardoor, daarom and dus. All three
connectives express causality in a 'forward' direction, that is, cause
precedes consequence. These connectives can best be translated in
English by means of the phrases we used in examples (1) and (2):
daardoor is similar to as a result / as a consequence, daarom can best be
translated as that's why and dus is quite similar to so. In (3) and (4) we
have translated examples (1) and (2) in Dutch.

(3) Ze zijn afgelopen vrijdag plotseling naar Parijs vertrokken.

Daardoor / Daarom / Dus zijn de buren niet thuis.
'They suddenly left for Paris last Friday. As a result / That's why/
So the neighbours are not at home. '

(4) Bet licht in de woonkamer is uit. # Daardoor /? Daarom / Dus zijn

de buren niet thuis.
'The lights in their living room are out. # As a result / ? That's
why/ So the neighbours are not at home. '

All three connectives seem to fit in (3), although they do express different
relations. Daardoor expresses a simple cause-consequence relation in
which the second segment (S2) might even be the unintended result of the
Domains o.fuse and subjectivity 59

first segment (S 1) ("They had to go to Paris"); with daarom the sequence

can be interpreted as a so-called volitional relation in which the first
segment contains the reason for an intentional action in S2 ("Going to
Paris is their reason for leaving home") and with dus S2 is a conclusion
based on S1 ("I conclude they are not there on the grounds that (I know)
they had to leave"). In (4), which can only be interpreted as a conclusion,
dus fits in very well, whereas daardoor leads to an unacceptable
sequence and daarom is at least odd (indicated by a'?').
These three connectives were selected because they are the most
frequently used ones expressing forward causality in Dutch (Uit Den
Boogaart 1975). Second, the differences between them seem to overlap
with distinctions put forward in recently developed theoretical proposals.
For instance, several linguists studying adverbial clauses distinguish
between content and epistemic layers of meaning (Kortmann 1997 for an
overview). Sweetser (1990) introduces a three-level approach to account
for differences in meaning and use of connectives. She distinguishes
between the content, epistemic and speech act domains. This type of
functionally and cognitively oriented approach seems attractive and
promising for the three Dutch causal connectives as well.
In this chapter we will first put such an approach to the test. We
conclude- that a domain theory does not account for our data. Still, the
testing of this hypothesis does give rise to an alternative account in terms
of subjectivity, which we will develop in more detail.

2. The distribution of three Dutch causal connectives:

Domains of use?

2.1. Content, epistemic and speech act domains

Sweetser (1990) has formulated an influential proposal on the

classification of connectives, which we will call the three-domain theory.
She demonstrates ambiguity in the use of conjunctions and claims that her
theory explains these ambiguities. Examples (5)-(7), for instance,
illustrate that English because has three readings: a content reading, an
epistemic and a speech act reading.

(5) John came back because he loved her.

(6) John loved her, because he came back.

60 Henk Pander Maat and Ted Sanders

(7) What are you doing tonight, because there's a good movie on.

In (5) John's love was the real-world cause of his coming back. (6)
means that my knowledge of John's return leads to the conclusion that he
loved her. An adequate paraphrase of (7) would be "I am asking what
you are doing tonight because there is good movie on". Distinctions like
these have been used by many linguists to account for differences in the
use of connectives and cue phrases (e.g. Halliday and Hasan 1976;
Martin 1992; Knott and Dale 1994), even though there are clear
differences in terminology and definitions (Bateman and Rondhuis, 1997;
Moore and Pollack 1992; Oversteegen 1997; Sanders and Spooren
How do such distinctions apply to the three Dutch causal connectives
introduced earlier? If we reconsider the two examples, (3) and (4), it was
already noted that in (3) all three connectives seem to fit in, although they
do express different relations. Daardoor expresses a content (Cause-
Consequence) relation, just like daarom (although this connective gives
rise to a volitional, a Reason-Action relation), and with dus the sequence
is of an epistemic nature: S2 is a conclusion based on S1. In (4), which
can only be interpreted epistemically, dus fits in very well, whereas
daardoor leads to an unacceptable sequence and daarom is at least odd.
Hence, at first sight, the domain theory seems relevant for the three
connectives. A first conclusion might be that daardoor can be described
as a connective used exclusively for content relations.
Similar conclusions of domain-specificity are reported for causal
connectives in other languages. For instance, it has been suggested that
German denn can only be used to express epistemic relations (among
others, Giinthner 1993; Keller 1995), just as French donc (Rossari and
Jayez 1996), Hungarian hat (Nemeth 1995) and English since are
supposedly impossible to use in the content domain (Sweetser 1990).
The attraction of the domain theory becomes even more apparent if we
consider the similarity between the domain distinctions and the categories
of coherence relations that have been proposed in cognitively inspired
work on relation categorization. In Sanders, Spooren and Noordman
(1992), for instance, it is shown how the set of coherence relations that
can hold between discourse segments can be decomposed in more basic
categories by making use of only four parameters. One of these
parameters, the Source of Coherence, which distinguishes between
semantic and pragmatic relations, shows great resemblance to Sweetser's
content-epistemic-speech act distinction (Sanders 1997; Knott 2000).
Given the 'basicness' of these conceptual distinctions it is attractive to
Domains ofuse and subjectivity 61

relate them to restrictions on the use of connectives. In such a view, the

lexicon of connectives would reflect the basic distinctions between
coherence relations in a systematic way.
No matter how attractive such a view, the categorization of these
linguistic devices is as yet not based on distributional data from actual
language use. In fact, none of the above ideas on 'domain-specificity' of
connectives and cue phrases are based on authentic occurrences of the
phrases in text corpora. By contrast, recent corpus studies do reveal that
existing categorization proposals, including that by Sweetser (1990),
cannot account for the data of connective distribution in a straightforward
way (Degand 1996 and Pander Maat 1998 and 1999). In an exploratory
cross-linguistic study, Knott and Sanders (1998) did find specific phrases
expressing the semantic-pragmatic distinction, but these tended to be
lexical signals or idiom chunks of the type on the grounds that
(exclusively semantic) and it follows that (exclusively pragmatic). The
most frequently used connectives did not seem to be very specific in
meaning and use.
Therefore, current suggestions on the 'domain-specificity' of
connectives are to be considered as hypotheses to be tested further in
empirical research of language in use. This was exactly the purpose of a
corpus study we conducted (Pander Maat and Sanders 1995): to
investigate to what extent the actual distribution of Dutch daardoor,
daarom and dus can be described in terms of the domains mentioned.

2.2. An empirical test ofthe domain theory

The use of dus, daarom and daardoor was studied in a corpus of 150
fragments, 50 for each connective. The corpus consisted of different text
types, argumentative / persuasive as well as descriptive / informative. It
contained annual reports and publicity brochures of Dutch companies,
public information brochures, articles from the readers' opinion page of a
Dutch quality paper (De Volkskrant 1994) and news articles from this
same paper.
Our methodology involved three steps. We first determined the
possible relational interpretations of these fragments without connectives.
Then we counted how often each connective expressed a certain relation.
In the final step, we checked whether the original connective could be
replaced by another. This substitution method is a way of testing
semantic intuitions (Knott and Dale 1994; Knott and Sanders 1998). The
questions are: Does substitution lead to a sequence which is still
62 Henk Pander Maat and Ted Sanders

acceptable? And, if acceptable, does the relational interpretation change

as a result of this substitution?
Here is a summary of our findings, formulated in terms of the relations
the three connectives can and actually do express.

• Daardoor can only express relations of the content non-volitional

• Dus can express content volitional, epistemic, summary and
paraphrase relations, but not content non-volitional relations. It most
often expresses epistemic relations.
• Daarom can express content and epistemic relations. It most often
expresses content volitional relations.

Below are prototypical examples from the corpus.

(8) Een campagne in de VS duurt in de eerste plaats zeer lang. Een

kandidaat dient eerst de voorverkiezingen te winnen voordat hij
officieel wordt gekandideerd Een presidentiele campagne duurt
daardoor al gauw anderhalfjaar.
'In the first place, a campaign in the US takes a very long time. A
candidate has to win the pre-elections before he is officially
nominated. Daardoor 'as a result' a presidential campaign easily
takes one and a half years. '

(9) Vaste klanten zijn voor de Bijenkorf van vitaal belang: zij
besteden per jaar twee maal zoveel in de Bijenkorfwinkels als
andere klanten. Daarom heeft de Bijenkorf aan de Vaste
Klantkaart een aantal voordelen verbonden
'Regular customers are of vital interest to the Bijenkorf: in a year
they spend twice as much as other customers do. Daarom 'that's
why' the Bijenkorf has added a number of advantages to the
Regular Customer-Card. '

(10) Er blijkt weer een toename van het aantal besmettingen met het
HIV-virus te zijn. Ais AIDS-verpleegkundige zie ik de gevolgen
ervan. AIDS-preventie, daar komt het dus op aan.
'There appears to be an increase of infections with the HIV-virus.
As an AIDS-nurse, I see the consequences. Dus 'so' prevention of
AIDS, that's what it's all about. '
Domains ofuse and subjectivity 63

These results imply that only the use and meaning of daardoor can be
described in terms of relational domains, but even there we need the
additional criterion of volitionality, that is we need to determine whether
the relation concerns a reason for an intentional action or not. Daardoor is
restricted to the content domain, more specifically to relations of the non-
volitional content type. The difference between daarom and dus is
difficult to describe in terms of domains: both connectives regularly
express both kinds of relations. Hence we are faced with the situation that
the purported distinction between content and epistemic causality remains
unmarked in the field of most common Dutch connectives, while the
distinction between volitional and non-volitional relations is marked
linguistically by commonly used causal connectives. Does this imply that
(non-)volitionality has to be considered a more fundamental criterion than
the distinction between content and epistemic meaning?
In proposals for the categorization of coherence relations, volitionality
is indeed considered relevant but it is treated as a second-order distinction
within the content domain relations. For instance, Sanders et al. first
distinguish between semantic (equivalent to content) and pragmatic
(roughly equivalent to epistemic and speech act) relations, and then
separate volitional from non-volitional relations and epistemic relations,
as do Mann and Thompson (1988). Similarly, in the linguistic literature
on connectives and adverbial subordination, the first relevant distinction is
between content and epistemic.
It seems we might have to allow for the idea that, as far as the
typology of causality is concerned, a distinction of the type ["volitional]
may be as fundamental as the one between content and epistemic
domains (Stukker, Sanders and Verhagen 1999). Adding to this the
observation that volitional and epistemic relations are often lexicalized by
the same connectives, it makes sense to ask what the common
denominator in these relations might be.

3. Subjectivity in coherence relations and connectives

So far, the only clear contrast we have seen in the data from actual
language use is the one between daardoor on the one hand, and dus and
daarom on the other: daardoor only expresses non-volitional relations,
while daarom and dus may express both volitional and epistemic
relations. This contrast constitutes a problem for the 'domain theory',
since it cuts right through the content domain. Hence our task is to
64 Henk Pander Maat and Ted Sanders

develop a conceptualization that adequately captures the nature of the

contrast between non-volitional and other causal relations.

3.1. Subjectivity

The question now is: what is it that 'epistemic' causality has in common
with 'volitional' causality, so that speakers of Dutch easily use the same
vocabulary to express these two relation types, but not to express
relations of non-volitional causality? We want to suggest that what
epistemicity and volitionality have in common is that both crucially
involve an animate subject, a person, whose intentionality is
conceptualized as the ultimate source of the causal event, be it an act of
reasoning or some 'real-world' activity. That is, it seems that to us
humans a very fundamental distinction is the one between events
ultimately originating from some mind, and events that originate from
non-intentional causes; between causes that are crucially located in a
subject of consciousness, and those that are located in the inanimate
world. This distinction is so fundamental that it shows up in similar ways
at different linguistic levels and is often the only one marked explicitly by
means of some linguistic form (Verhagen 1995 and other contributions to
Stein and Wright 1995).
We will use the notion of subjectivity to specify this idea further. In a
sense, every linguistic utterance is subjective, that is, connected to the
point of view of some 'subject'. However, this use of the concept is not
very informative. A more interesting question regarding this notion is
whether this 'subject' is included in the semantic characterization of the
utterance. Langacker (1990), who applies the notion of subjectivity to
several linguistic phenomena, distinguishes between three situations (Pit
1997). First, the ground-the term he uses to refer to the speech event,
its participants and its immediate circumstances-may be entirely
external to the semantics of the utterance. This situation is exemplified by
(11) below. Second, the ground may be included in the scope of
predication as an offstage, unprofiled reference point. This is the case
when deictics like yesterday, tomorrow, etc. are used. Another example
is (12) below, in which the modal adverb probably invokes the present
speaker as the source for the probability judgement. Third, the ground
may be put onstage, as in (13). In this case, the ground is more or less
,objectified', that is, it is made part of the situation referred to in the
Domains ofuse and subjectivity 65

(11) Peter is in Paris.

(12) Peter is probably in Paris.

(13) I think Peter is in Paris.

In an overview of the different accounts of subjectification and

perspective, J. Sanders and Spooren (1997) refer to sentences (12) and
(13) as instances of subjectification, by which they mean that the present
speaker accepts responsibility for the propositional content of the
utterance, and possibly also for its form. However, a piece of information
may also be connected to another subject ofconsciousness (from now on
SOC) than the present speaker. In this case, Sanders and Spooren speak
ofperspectivization. In (14), Peter is the SOC.

(14) Peter wants to go to Paris.

Normally, perspectivization requires indicators like verbs of cognition,

perception and evaluation, such as want in (14), as has been shown in
recent cognitive linguistic work on perspective and mental spaces
(Fauconnier 1994; Fauconnier and Sweetser 1996; J. Sanders and
Redeker 1996). In the following, we will analyze the role of a subject of
consciousness in relational interpretations. To determine to what degree a
certain coherence relation may be called subjectified or perspectivized,
we need to inspect both the related discourse segments and the relation

3.2. The similarity ofvolitional and epistemic relations

Our task now is to identify the common characteristics of volitional and

epistemic relations. We start, however, by characterizing non-volitional
relations. These relations are presented as strictly objective. Not only does
the relation concern two real world situations, the relation itself is also
construed as an objective fact. Of course the speaker is committed to her
statements (i.e. the speaker cannot subsequently deny them), but the
crucial point is that the speaker is not involved in the events being
described, nor in constructing a causal relation between them. See
fragment (15), which corresponds to the first situation distinguished by
Langacker, in which the ground is external to the semantic representation
of the utterance.
66 Henk Pander Maat and Ted Sanders

(15) The sun rose. As a result, the temperature went up.

By contrast, consider the following epistemic relation:

(16) It was a small case, completely rusty. So it was made out ofiron.

The causal relation in (16) does not relate real-world situations, but two
beliefs of an SOC (Knott 2000). In this case the SOC is a concluder, that
is somebody making inferences. In (16), the concluder is identical to the
speaker. And the relation between these two beliefs is again situated
within the causal domain of this SOC. Hence, the contrast between non-
volitional and epistemic relations seems rather clearcut in that the
representation of epistemic relations assigns a very prominent role to an
SOC, while such a participant is entirely absent in our interpretation of
non-volitional relations.
As was indicated above, the distribution of causal connectives in
Dutch suggests a similarity of volitional and epistemic relations: dus and
daarom both express volitional and epistemic relations, whereas
daardoor encodes non-volitional relations. How should we conceptualize
this similarity? In our View, the representation of a volitional relation also
involves an SOC, namely the actor of the action in the second segment.
Consider fragment (17).

(17) It was eight 0 'clock. Claire put on the television.

Let us assume a volitional interpretation of (17), in which the time is the

reason for Claire to put on the television. The first thing to note about this
reading is that, although the speaker of (17) is taken to be committed to
the truth of the first segment, the cause for Claire's action is not the
objective fact of the time being eight o'clock, but Claire's mental
representation of this information. Likewise, although the speaker posits
the existence of the causal relation in the outside world, the causal
relation between reason and action is at the same time situated within
Claire's mental domain, since it is constituted by her decision-making
process, a process that she will generally be conscious of
The same goes for the action description in the second segment: this
description is valid both in the real world as conceived of by the speaker
and in the domain of the actor's beliefs. That is, normally describing an
action conveys not only the action taking place in reality as viewed by the
speaker, but also the action which the actor thinks she is performing.
Usually, this duality remains unnoticed, since in both domains the same
Domains ofuse and subjectivity 67

action is represented. The following fragment, inspired by Fauconnier

(1994,150-151) illustrates both the possibility of such conflicts and their
counterexpectational nature.

(18a) Calvin is drawing an elephant, but in fact it looks more like an


(18b) Calvin thinks he is drawing an elephant, but in fact it looks more

like an ant-eater.

(18b) sounds more acceptable than (18a) because a divergence between

speaker perspective and actor perspective is more appropriate when
marked by a perspective indicator (thinks) in the action description itself
To sum up, the representation of volitional (content) relations always
involves both the mental domain of the speaker and the mental domain of
the actor. In general, the speaker adopts the actor perspective and in this
case, the actor is the SOC. In other words, the actor is the default SOC
for volitional relations. This holds for the relation as such, but also for the
segment containing the action.
We have already seen the clear difference in subjectivity between non-
volitional andepistemic relations, which was constituted by the presence
of an SOC in epistemic relations and its absence in non-volitional
relations. In the unmarked case the SOC, the concluding subject, is
identical to the present speaker. However, it is certainly possible to
represent the reasoning of a subject other than the speaker in epistemic
relations, as (19a, b) show.

(19a) The lights in the neighbours' living room were out. So they were
not at home.

(19b) Harry saw that the lights in his neighbours' living room were out.
He concluded they were not at home.

Subsequently we have argued that the presentation of volitional relations

involves an SOC as well, viz. the actor choosing a certain course of action
for a reason referred to in the first segment. This SOC, who is an actor in
volitional relations, is typically present in the second segment. Generally,
there is a certain distance between this SOC and the present speaker.
However, this distance decreases when the actor is referred to by I. The
distance also decreases when the difference between the point in time at
which the action is performed and the moment of speaking is reduced.
68 Henk Pander Maat and Ted Sanders

For these reasons the distance between the SOC and the present speaker
is larger in (20b) than in (20a).

(20a) / am tired, so / am leaving.

(20b) He was tired, so he left.

In conclusion, volitional and epistemic relations are similar in that their

representations involve an SOC. In both relations this SOC may be more
or less distant to the present speaker. However, epistemic relations seem
to have a built-in preference for speaker SOCs, a preference lacking in
volitional relations.

3.3. 'Dus ' and 'daarom ' differ in terms of subjectivity

In section 2 it was found that, although daarom and dus may both
express volitional and epistemic relations, dus is much more frequent in
epistemic relations than it is in volitional relations, while daarom shows a
slight preference for volitional relations. The hypothesis we want to
explore below is that this differential distribution is to be explained by a
difference between the two connectives with regard to the distance they
express between the SOC involved and the present speaker. More
specifically, our claim will be that daarom encodes a larger distance than
dus does. We will now derive some specific predictions from this general
claim, so that they can be put to the test in a corpus study.
Our first prediction presents itself in a rather straightforward way.
When dus really encodes a smaller distance between the SOC and the
speaker, the subject responsible for the action or the conclusion in dus-
segments should more often be identical to, or close to the speaker than
the actors and concluders in daarom-segments are. In order to test this
claim, we distinguish between four types of sacs, ranging from minimal
(1) to maximal (4) distance between the SOC and the speaker.

(1) Implicit speaker SOC included in the semantic representation of the

Example: Baking cakes is fun.

(2) Explicit speaker SOC, referred to by /; this includes cited speakers.

Example: / bakedfour cakes or "/ bakedfour cakes", Calvin said.
Domains ofuse and subjectivity 69

(3) Explicit pronominal or nominal third-person SOC.

Example: He / Calvin bakedfour cakes.

(4) Contextually recoverable third-person participants invoked by a

passive construction.
Examples: Calvin loved baking cakes. So every birthday at least
four were baked (recoverable actor in the second segment).

The SOC-speaker distance is smallest for implicit speaker sacs.

Following Langacker (1985, 1990), we view explicit speaker references
as expressing a certain distance between the present speaker and the first-
person participant referred to in the utterance. Clearly, third-person sacs
are further removed from the speaker than first-person ones. Finally,
recent work on the passive shows that this construction indicates non-
identification with the actor, when compared with the corresponding
active construction (Comelis 1997). Therefore, implied agents in passive
constructions are placed at the bottom of our list: there is maximal
distance between the SOC and the speaker.
Our second prediction needs some additional introduction. As stated
earlier, we claim that dus prefers speaker sacs. If this is correct, dus-
fragments with third-person sacs would need additional marking,
indicating a non-default soc. Consider the following example.

(21) Het sloeg vijfuur. Dus pakte Frits zijn aktentas.

'The clock struck five. Dus 'so' Fritz took his briefcase.'

In this example, there are two possible interpretations. In the first one,
Fritz decides to go home on the basis of his knowledge of the time and his
knowledge of the need to go home at five. Under this interpretation, Fritz
is the SOC. In the second interpretation, the speaker presents Fritz'
behavior as an habitual action: always when the clock strikes five, Fritz
forgets all about his work and gets ready to leave. Under this
interpretation, the speaker is the SOC, because Fritz does not necessarily
consciously decide to leave work, he just does what he does every day.
If the segments in example (21) are connected by dus, the second
interpretation is preferred. If we want listeners to construct the first
interpretation, and we still use dus, we would need to mark the first
segment as being presented from the perspective of the non-default SOC:
the actor, not the speaker. Examples containing such non-default SOCs
are (21a) and (21b). In (21a) the SOC is explicit in the first segment; in
70 Henk Pander Maat and Ted Sanders

(21 b) it is contextually available on the basis of a segment m the

preceding discourse.

(21 a) Frits hoorde de klok vijfuur slaan. Dus pakte hi} zi}n aktentas.
'Fritz heard the clock striking five. Dus 'so' he took his briefcase.'

(21b) Frits keek op zi}n horloge. Het was al laat. Dus pakte hi} zi}n
'Fritz looked at his watch. It was late already. Dus 'so' he took his
briefcase. '

If this line of reasoning is correct, our prediction should be that in

comparison with volitional daarom-fragments, volitional dus-fragments
more often have a first segment in a third-person actor perspective or
contextually imply such an actor. However, this prediction only applies to
fragments with third-person actors: when the actor is the speaker, the first
segment is in the speaker-actor perspective by default, even when this is
not made explicit. The prediction of perspective continuity in third-person
volitional dus-fragments should be valid for third-person epistemic
fragments too. However, since third-person epistemic fragments are rare,
it remains to be seen whether the continuity hypothesis can be tested for
epistemic relations.
These are then the two hypotheses to be tested in corpora.
1. The SOC-speaker distance hypothesis: In dus-fragments, the distance
between SOC and speaker is smaller than in daarom-fragments.
2. The continuity hypothesis: In third-person dus-fragments, the first
segment will appear more often in the perspective of the actor (in
volitional relations) or concluder (in epistemic relations) than in third-
person daarom-fragments.

4. Corpus analysis

4.1. Corpus

25 dus- and 25 daarom-fragments were selected at random from a

corpus of a Dutch quality newspaper, the Volkskrant (1995). The corpus
was relationally stratified in the sense that the number of volitional and
epistemic relations for each of the connectives was equal. Three types of
fragments were not included in the corpus.
Domains ofuse and subjectivity 71

- Fragments with another connective immediately following dus or

daarom (e.g. dus omdat ... )
- Fragments in which the connective did not connect full sentences (e.g.
"Een mooie dus dure boot"/ 'A beautiful therefore expensive boat')
- Daardoor-fragments in which the connective did not refer to the
preceding segment as a whole, but rather to parts of it.

4.2. Results: volitional relations

For reasons of exposition, we start with the results regarding the second
hypothesis, the continuity hypothesis. In the analysis, we determined the
perspective of the first segment in the third-person volitional relations for
both connectives (daarom: n=2I; dus: n=I4). The results (see Table 1)
show that the proportion of fragments characterized by an explicitly or
implicitly continuous perspective for dus is significantly larger than that
for daarom (Chi =15.26, p < .001, df=2).

Table 1. The perspective of the first segment in third-person volitional daarom-

and dus- fragments: frequencies and column percentages.

81 has: daarom dus

Explicit actor perspective 10 (48%) 9 (640/0)
Contextually implied actor perspective 5 (36%)
Non-actor perspective (discontinuous) 11(52%)
Total 21(1000/0) 14 (1000/0)

Fragment (22) is a typical continuous dus-fragment, where he is both the

actor in segment (2a) (where segments are rougWy equivalent to clauses)
and the participant responsible for the evaluation in (Ib) and (Ic).

(22) (Anna Enquist debuteerde in 1991 met Soldatenliederen. Daar

had Theo Sontrop voor gezorgd. )
(Ia) Ze had eens een paar gedichten naar Maatstaf gestuurd, (Ib) en
daarover was hi} laaiend enthousiast geweest-(lc) vond haar
poezie meteen af (2a) Dus hi} schreefhaar meteen ofze nog meer
had, (2b) en dat werd toen die bundel.
'(Context: "She" refers to Anna Enquist, a Dutch writer. "He"
refers to Theo Sontrop, the Dutch editor of a literary journal called
72 Henk Pander Maa! and Ted Sanders

(la) She had sent some poems to Maatstaj, (lb) and he had loved
them-(lc) found her poetry had an immediate perfection. (2a)
Dus 'so' he wrote to ask her if she had any more (2b) and this
became her first collection of poems. '

The discontinuities in volitional daarom-relations predominantly occur

when the first segment of the relation contains the judgement of the
author. In (23), for instance, the author's judgement in (3) is marked by
"what is certain is that", whereas China is the actor in (4).

(23) (Is Noord-Korea bezig een kemwapen te ontwikkelen?)

(la) Voor aile duidelijkheid: (lb) het bewijs dat Noord-Korea met
dit soort snode plannen bezig is, is niet geleverd. (2a) De
Amerikaanse inlichtingendiensten geloven van wei, (2b) de
ministeries van Buitenlandse Zaken en Defensie in Washington
houden zich op de vlakte. (3) Vast staat dat een in een het nauw
gedreven No 0 rd-Korea met atoomwapens in Azie een
kemwapenwedloop zal veroorzaken. (4) Met name de regionale
kernmogendheid China spant zich, op Amerikaans aandringen,
daarom in de oude bondgenoot in te tomen.
'(Is North Korea working on the development of a nuclear
(1a) For the sake of clarity: (1b) the evidence that North Korea is
working on these infamous plans is not available. (2a) The
American intelligence agencies believe they are doing so; (2b) the
Department of Foreign Affairs and the Defense Ministry in
Washington remain silent. (3) What is certain is that a hard-pressed
North Korea can cause a nuclear arms race in Asia. (4) Daarom
'that's why' the regional nuclear power China, following American
suggestions, is making great efforts to restrain the former ally'.

As Table 1 shows, dus-fragments regularly have contextually recover-

able third-person participants, i.e. the participant is inferred on the basis
of the discourse environment. An example is fragment (24).

(24) (1) Maar al te vaak wordt het beeld van het buurland bepaald
door de media. (2a) Berichten over positieve zaken zijn geen
nieuws, (2b) dus concentreert men zich op het negatieve.
'(1) All too often the image of a neighbouring country is
determined by the media. (2a) Information about positive things is
not news, (2b) dus 'so' one concentrates on what is negative. '
Domains ofuse and subjectivity 73

In the volitional interpretation of (24), representatives of the media are

taken to be responsible for the evaluation ('that something is not regarded
as news') in (2a). That is, the participant of (2a) is to be recovered from
the preceding sentence. The author may very well disagree with (2a) and
disapprove of the action in (2b), which is motivated by it. Given the larger
context (a letter to the editor by the German ambassador in the
Netherlands pleading for mutual understanding between neighbouring
countries), this is the only plausible interpretation. This concludes our
discussion of the continuity hypothesis for now.
The results for the SOC-speaker distance hypothesis show that dus-
actors are indeed closer to the author than daarom-actors, see Table 2, in
which the possible participants are ordered from minimal to maximal
distance to the speaker (see section 3.3 for this identification hierarchy).
In the majority of cases, the speaker is the dus-participant, where the
speaker can be the author of the fragment as well as the cited speaker,
and usually appears in the text as I (occasionally we). By contrast,
daarom-participants are predominantly realized as a (nominally
indicated) third person. The difference in Table 2 is significant (Chi2 =
10.99, P < .05 one-tailed, df = 2), thereby supporting the second

Table 2. The identity and realization of actor-participants in the volitional

daarom- versus dus-fragments: frequencies and column percentages.

Identity and realization daarom dus

Speaker explicit 4 (160/0) 11 (480/0)
Third person 13 (560/0) 9 (360/0)
Pronominal 2 6
Nominal 11 3
Contextually recoverable third-person agent 8 (32%) 5 (20%)
Total 25(100%) 25(100%)

We will now illustrate the data regarding the SOC-speaker distance

hypothesis. In daarom-fragment (23), the actor in (4) is a third-person
nominal (China). By contrast, dus-fragments more often have first-person
actors, as in (25). And in (24), there is a third-person participant
recoverable from the context.

(25) "In '68 is de cultuur daar verbannen en gedegradeerd tot 1,3

miljard spugende Chinezen die geen enkel gezag meer kunnen
opbrengen voor schoonheid. Zakken afval en stront gaan zo het
raam uit. Dat gaat mij te ver, dus neem ik de tweede klas ... "
74 Henk Pander Maat and Ted Sanders

'''In ' 68 culture was banned from there and degraded to 1,3 billion
spitting Chinese who cannot bring themselves to any respect for
beauty. Bags full of trash and shit simply go out of the window.
That is too much for me, dus 'so' I take the second class. '"

As shown in Table 2, dus and daarom show a clear difference with

respect to pronominal or nominal references to third-person actors:
daarom prefers nominal actors, while dus prefers pronominal actors (e.g.
fragment PZ», the difference being significant (Chi2 = 6.05, P < .05
two-tailed , df = 1). This indicates that dus-actors tend to have a higher
level of accessibility (Ariel 1988) than daarom-actors. This matches with
the fact that dus-actors are more often responsible for the first segment
than daarom-actors.

4.3. Results: epistemic relations

The SOC-speaker distance hypothesis was also investigated for epistemic

relations, with a method identical to the volitional study. The results are
presented in Table 3.

Table 3. The identity and realization of concluder-participants in the epistemic

daarom- versus dus-fragments: frequencies and column percentages.

Identity and realization daarom dus

Speaker implicit 14 (56%) 25 (100%)
Speaker explicit 2(12%)
Third-person nominal 9(32%)

All dus-participants are implicit speakers, while in daarom-connections

the second segment also contains explicit speakers and (nominally
indicated) third-person participants. There is clearly more distance
between participants (concluders) in the second segment and the author
for the daarom-fragments than for dus-fragments (Chi2 = 10.98, df= 1, p
< .001, two-tailed), thereby supporting the second hypothesis.
Since there are no third-person epistemic relations containing dus, we
cannot test the continuity hypothesis for epistemic fragments. Instead, we
carried out an exploratory study of the subjectivity relations between both
segments for the entire corpus of epistemic relations. For the purposes of
this analysis, all elements indicating deontic (have to) or epistemic
modality (must, I think, probably) and other lexical elements with
Domains ofuse and subjectivity 75

evaluative meanings (e.g. important, surprising) were considered as

indicators of subjectivity (Lemke 1998).
The results of this analysis are summarized in Table 4; illustrative
fragments follow below. Daarom and dus show different perspective
configurations (Chi2 = 11.60. df= 3, P < .01)

Table 4. Subjectivity configurations in epistemic daarom versus dus-fragments.

Configuration daarom dus

SI subj - S2 subj (continuous) 15 (600/0) 15 (60%)
SI subj-l - S2 subj-2 (discontinuous) 3 (12%)
SI obj - S2 subj 7 (280/0) 3 (12%)
SI obj - S2 obj 7 (28%)
Total 25 (100%) 25 (100%)

S1, S2 refer to the first and second segments connected by a connective.

Subj means Subjective and obj means Objective.

In the first and most frequent configuration, the same partIcIpant is

responsible for both segments. This is illustrated by fragment (26) (first-
person SOC) below.

(26) (1a) Drugs verwoesten mensenlevens, (lb) dus moeten drugs

strafrechtelijk bestreden worden.
'(1a) Drugs ruin human lives, (1b) dus 'so' drugs must be
combatted by means of criminal law. '

The second configuration is discontinuous and occurs only in daarom-

fragments. In fragment (27) the first sentence clearly adopts the
perspective of the Russian citizen, while the author takes full
responsibility for the second segment.

(27) (1a) Gewone burgers voelen zich kwetsbaar, (1b) overgeleverd

aan gangsters, gewone misdadigers, corrupte bureaucraten en de
mafia-ric) terwijl hooghartige bureaucraten hun zakken vullen.
(2) Daarom is het niet tegenstrijdig dat de kiezers zich uitspraken
voor de nieuwe grondwet en tegelijkertijd voor Zjirinovski
'(1a) Ordinary citizens feel vulnerable, (1b) being at the mercy of
gangsters, common criminals, corrupt bureaucrats and the mob-
(Ic) while arrogant bureaucrats fill their pockets. (2) Daarom
76 Henk Pander Maat and Ted Sanders

'that's why' it is not contradictory that the voters voted for the new
constitution and at the same time for Zjirinovski. '

In the third configuration the concluder draws a conclusion on the basis of

an objective, factual statement: see (28). However, the most pronounced
difference between the epistemic daarom- and dus-fragments is found in
second segments which follow objective first segments. Daarom-
fragments with a factual S 1 always have an evaluation as their second
segment, as in (29). By contrast, dus-fragments more often contain
objectively formulated second segments, as in (30).

(28) (1) De VPRO-quiz was echter een open quiz. (2) Het stond
iedereen vrij om allerlei feiten op te zoeken in encyclopedie of
bibliotheek. (3) Of om overleg te plegen met familie, vrienden of
collega's. (4) Het hebben van een hoge score impliceert dus niet
noodzakelijkerwijs dat men ook veel verstand van wetenschap
'(1) However, the VPRO-quiz was an open quiz. (2) Everybody
was allowed to look things up in an encyclopedia or a library. (3)
Or to consult family, friends or colleagues. (4) Dus 'so' having a
does not necessarily mean that one knows a lot about

(29) (la) Hi} had ervaring op die circuits, (lb) ik niet. (2) Hi} maakte
daarom theoretisch meer kans om punten te halen.
'(la) He was experienced on those circuits, (lb) I was not. (2)
Daarom 'that's why' in theory he had a better chance to score
points. '

(30) (1) Het punt is echter dat de verdachten die schuldjuist met klem
ontkennen en daarom bi} het hof hoger beroep hebben
aangetekend tegen hun veroordeling bi} de rechtbank. (2) Die
strafis dus nog niet definitief
'(1) The issue, however, is that the suspects vehemently deny their
guilt and have lodged an appeal against their conviction by the
court. (2) Dus 'so' this sentence is not final yet.'

Apparently, daarom is incapable of conveying an epistemic interpretation

in an entirely factual discourse passage. By itself: it fails to convey
subjectivity in the interpretation of the second segment and of the relation
Domains ofuse and subjectivity 77

This result fits very well with our account, which claims that dus and
daarom encode different speaker-SOC distances. In some contexts, no
SOC is present, either explicitly or implicitly (in the form of, for instance,
subjective elements). In these situations the use of dus is unproblematic.
The reason is that the processing instruction constituted by dus, i.e. 'find
an SOC close or identical to the speaker', can always be carried out
successfully because every utterance presupposes the presence of a
speaker. However, the instruction encoded by daarom, i.e. 'fmd an SOC
at a certain distance from the speaker' presents problems in situations
where the utterance itself does not present an SOC at all. After all,
selecting the speaker is impossible. In order to support selection of the
speaker, daarom needs linguistic elements indicating his presence,
though not referring to him as such.

5. Conclusion

In this paper we have adopted an empirical, corpus-analytic approach to

investigate the meaning and use of three frequent Dutch causal
connectives. It can be concluded that the lexicon of connectives cannot be
described in terms of distinctions well-known from accounts like
'domains' (Sweetser 1990) or layers of meaning, such as content,
epistemic and speech act.
Alternatively, we have proposed considering the difference between
the connectives in terms of subjectivity or, more precisely, in terms of the
distance between the speaker :md the SOC responsible for the causal
relation. In such a subjectivity account, the three connectives can be
characterized as follows. In causal relations expressed by daardoor 'as a
result' there is no SOC present because the causality is located outside of
this SOC. In relations expressed by daarom 'that's why' there is a certain
distance between speaker and SOC, whereas in the cases of dus 'so', this
distance is small or even absent. We have shown that this approach
accounts for the actual distribution of the connectives in newspaper
The subjectivity approach may also have.further advantages. For one
thing, it points to the commonality between connectives on the one hand
and other linguistic phenomena which have been treated in terms of
concepts similar to subjectivity. For instance, it has long been recognized
that deictic reference is essentially relational in nature, in that it does not
primarily characterize the referent but the relation between the referent
and the contextual framework of the present utterance (Hanks 1992). The
78 Henk Pander Maat and Ted Sanders

relational dimensions at issue include the distance between the referent

and the speaker, both spatially (e.g. this vs. that; here vs. there) and
temporally (now vs. then), and the degree of active involvement in the
utterance event (1 vs. you vs. he/she). If the SOC of the causal relation
can be considered a kind of referent, the difference between dus and
daarom is conceptually parallel to the one between, say, here and there.
Alternatively, dus and daarom may be taken to encode different degrees
of speaker involvement in the causal relation (Degand and Pander Maat
Another phenomenon which has been treated in terms of subjectivity is
modality (Lyons 1977: 797-809; Coates 1983: 32-37; Nuyts 1992). In a
corpus study comparing Dutch modal adverbs (English gloss: probably)
to constructions involving modal adjectives (probable), Nuyts (1994: 45-
52) found that adverbs can only be used to express modal qualifications
"performatively", that is as expressing an epistemic commitment of the
present speaker, while adjectives may be used both performatively and
"descriptively", that is as reporting objective probability assessments or
epistemic evaluations by other people.
In other words, a distinction in terms of subjectivity is not only
lexicalized in the field of connectives but also in other semantic domains
like deixis and modality. In all three fields, speakers have to determine
their position vis-a-vis, or their involvement with, the entities and
propositions presented in their discourse. Language, then, seems to show
a systematic sensitivity to this expressive need.


* We wish to thank Bernd Kortmann, Elizabeth Couper-Kuhlen and Alistair

Knott for comments on an earlier version of this paper, and Alistair Knott
for correcting our English. Needless to say, all remaining errors are ours.
1. Our claim is that the difference between dus and daarom does not lie in the
intrinsically volitional or epistemic meaning of the connectives. So we
should be able to predict differences between the two connectives occurring
within the classes of volitional and epistemic relations respectively.
2. The difference between pronominal and nominal references to third-person
actors for dus and daarom was tested by a two-tailed test, since there was no
hypothesis concerning this difference.
Domains ofuse and subjectivity 79

Ariel, Mira
1988 Referring and accessibility. Journal ofLinguistics 24: 65-87.
Bateman, John A. and Rondhuis, Klaas-Jan
1997 Coherence relations: towards a general specification. Discourse
Processes 24: 3-49.
Coates, Jennifer
1983 The semantics ofmodal auxiliaries. Beckenham: Croom Helm.
Cornelis, Louise H.
1997 Passive and perspective. Amsterdam!Atlanta: Rodopi.
Degand, Liesbeth
1998 Het ideationele gebruik van want en omdat. [The Ideational Use of
want and omdat]. Nederlandse Taalkunde 4: 309-326.
Degand, Liesbeth and Henk Pander Maat
1999 Scaling causal relations in terms of speaker involvement. In: Levels
of Representation in Discourse, 45-53. Working notes of the
International Workshop on Text Representation. University of
Fauconnier, Gilles
1994 Mental spaces. Cambridge: Cambridge University Press.
Fauconnier, Gilles and Eve Sweetser (eds.).
1996 Spaces, worlds and grammars. Chicago: University of Chicago
Gtinthner, Susanne
1993 "...weil-man kann es ja wissenschaftlich untersuchen"-
Diskurspragmatische Aspekte der WortstellUng in WEIL-Satzen.
Linguistische Berichte 143: 37-59.
Halliday, M.A.K. and Ruqaiya Hasan
1976 Cohesion in English. London: Longman.
Hanks, William F.
1992 The indexical ground of deictic reference. In: Alessandro Duranti
and Charles Goodwin (eds.) Rethinking context. Language as an
interactive phenomenon, 43-76. Cambridge: Cambridge University
Keller, Rudi
1995 The epistemic weil. In: Dieter Stein and Susan Wright (eds.)
Subjectivity and subjectivisation. Linguistic perspectives, 16-30.
Cambridge: Cambridge University Press.
Knott, Alistair and Robert Dale
1994 Using linguistic phenomena to motivate·a set of coherence relations.
Discourse Processes 18: 35-62.
Knott, Alistair and Ted Sanders
1998 The categorization of coherence relations and their linguistic
markers: an exploration of two languages. Journal of Pragmatics
30: 135-175.
80 Henk Pander Maat and Ted Sanders

Knott, Alistair
2000 Semantic and pragmatic relations and their intended effects. In:
Ted Sanders, Joost Schilperoord and Wilbert Spooren (eds.),
Cognitive approaches to coherence. Amsterdam: Benjamins.
Kortmann, Bernd
1997 Adverbial subordination. A typology and history of adverbial
subordinators based on European languages. BerlinlNew York:
Mouton de Gruyter.
Langacker, Ronald W.
1985 Observations and speculations on subjectivity. In: John Haiman
(ed.), Iconicity in syntax, 109-150. Amsterdam/Philadelphia:
1990 Subjectification. Cognitive Linguistics 1: 5-38.
Lemke, Jay L.
1998 Resources for attitudinal meaning. In: Functions ofLanguage 5: 33-
56. Amsterdam: Benjamins.
Lyons, John
1977 Semantics 2. Cambridge: Cambridge University Press.
Mann, William C. and Sandra A. Thompson
1988 Rhetorical structure theory: Toward a functional theory of text
organization. Text 8: 243-281.
Martin, J. R.
1992 English text. System and structure. Philadelphia: Benjamins.
Moore, Johanna D. and Martha E. Pollack
1992 A problem for RST: The need for multi-level discourse analysis.
Computational Linguistics 18: 537-544.
Nemeth, Enik6
1995 On the role of pragmatic connectives in Hungarian spoken
discourses. In: B. Warvik, S-K. Tanskanen and R. Hiltunen (eds.),
Organization in discourse. Proceedings from the Turku conference,
393-402. Anglicana Turkuensia 14.
Nuyts, Jan
1994 Epistemic modal qualifications: on their linguistic and conceptual
structure. Antwerp Papers in Linguistics 81. Antwerpen:
Universiteit Antwerpen.
1992 Subjective vs. objective modality: what's the difference? In:
Fortescue, Michael D., Peter Harder and Lars Kristoffersen (eds.),
Layered structure and reference in a functional perspective, 73-99.
Amsterdam: Benjamins.
Oversteegen, Leonoor E.
1997 On the pragmatic nature of causal and contrastive connectives.
Discourse Processes 24: 51-85.
Pander Maat, Henk
1998 Classifying negative coherence relations on the basis of linguistic
evidence. Journal ofPragmatics 30: 177-204.
1999 The differential linguistic realization of comparative and additive
coherence relations. Cognitive Linguistics 10: 147-184.
Domains ofuse and subjectivity 81

Pander Maat, Henk and Ted Sanders

1995 Nederlandse causale connectieven en het onderscheid tussen
inhoudelijke en epistemische coherentie-relaties. [Dutch causal
connectives and the distinction between content and epistemic
coherence relations.] Leuvense Bijdragen 84: 349-374.
Pit, Mirna
1997 Subjective and objective construal in coherence relations. Jan Don
and Ted Sanders (eds.), Utrecht Institute of Linguistics OTS
Yearbook 1997, 45-71. Utrecht: UiL OTS.
Rossari, Corinne and Jaques Jayez
1996 Donc et les consecutifs. Des systemes de contraintes differentiels.
Lingvisticte Investigationes XX: 117-143. Amsterdam: Benjamins.
Sanders, Jose and Gisela Redeker
1996 Perspective and the representation of speech and thought in
narrative discourse. In: Gilles Fauconnier and Eve Sweetser (eds.),
Spaces, worlds and grammars, 290-317. Chicago: University of
Chicago Press.
Sanders, Jose and Wilbert Spooren
1997 Perspective, subjectivity and modality from a cognitive linguistic
point of view. In: Wolf-Andreas Liebert, G. Redeker and L. Waugh
(eds.), Discourse and perspective in cognitive linguistics, 85-112.
Amsterdam: Benjamins.
Sanders, Ted
1997 Semantic and pragmatic sources of coherence. On the categorization
of coherence relations in context. Discourse Processes 24: 119-147.
Sanders, Ted lM., Wilbert P.M.Spooren, and Leo G.M. Noordman
1992 Toward a taxonomy of coherence relations. Discourse Processes 15:
Sanders, Ted and Wilbert Spooren
1999 Communicative intentions and coherence relations. In: W. Bublitz,
U. Lenk and E. Ventola (eds.), Coherence in spoken and written
discourse, 235-250. Amsterdam/Philadelphia: Benjamins.
Stein, Dieter and Susan Wright (eds.)
1995 Subjectivity and subjectivisation. Linguistic perspectives.
Cambridge: Cambridge University Press.
Stukker, Ninke, Ted Sanders and Arie Verhagen
1999 Waar een wil is, is geen wet. De categorisering van causale relaties
binnen en tussen zinnen. [When there is a will, there is no law. The
categorization of causal relations within and between sentences.]
Gramma / TIT 7: 66-86.
Sweetser, Eve E.
1990 From etymology to pragmatics. Cambridge: Cambridge University
Uit den Boogaart, Pieter C.
1975 Woordfrequenties in geschreven en gesproken Nederlands. [Word
frequencies in written and spoken Dutch. ] Utrecht: Oosthoek,
Scheltema and Hoekema.
82 Henk Pander Maat and Ted Sanders

Verhagen, Arie
1995 Subjectification, syntax and communication. In: Dieter Stein and
Susan Wright (eds.), Subjectivity and subjectivisation, 103-128.
Cambridge: Cambridge University Press.
Causal relations in spoken discourse: Asyndetic
constructions as a means for giving reasons*

Christine Gohl

When speakers produce utterances offering reasons for a previous

utterance, they may do so either explicitly by linking these utterances, for
example, with a conjunctive element, or implicitly, i.e. by means of
asyndetic linkage. The analysis of asyndetically linked constructions poses
several problems regarding their interpretation as causally related
utterances. The lack of a conventionalized element signaling the causal
relationship raises questions concerning what counts as a causal relation
from the participants' perspective and how analysts can identify and
describe these unmarked constructions. After a discussion of general issues
with respect to asyndetic causal constructions, different environments in
which these occur are investigated. It will be argued that these
environments, constituted by different actions which call for accounting,
constrain the interpretation of an asyndetically linked construction.

1. Introduction

Earlier research on causal relations has dealt mainly with explicit lexical
markers, above all with subordinating conjunctions. More recent studies
on causal relations in spoken English draw attention to the paratactic use
of because (e.g. Ford 1994; Ford and Mori 1994; Couper-Kuhlen 1996;
Schleppegrell 1991), while studies on causal conjunctions in spoken
German focus, for example, on weil ('because') and the use of main
clause constructions in weil-clauses (e.g. Giinthner 1996; Pasch 1997;
Scheutz 1998; Uhmann 1998). This focus on the use of conjunctions
reflects the fact that they are indeed an important means for signaling
causal relations. That is not to say, however, that they are the most
frequent means of signaling causal relations in spoken interaction.
Especially in informal talk, a frequent device is the use of asyndetic
linkage. In these cases, speakers produce a reason for a previous action
without explicitly linking (e.g. via a causal connector) the utterance that
carries out the action to the utterance that is used to provide the account.
Although one can find various remarks in the literature on such asyndetic
causal structures (e.g. Blakemore 1987: Ch. 4.2.), they have not been
84 Christine Gohl

investigated empirically so far. In particular, there is a lack of research on

the use of such constructions in spoken interaction.
In this paper, I shall analyze asyndetic causal constructions in spoken
German discourse. This involves dealing with the following questions:
(1) What is a construction, i.e. in which cases can two (or more)
utterances be considered a structure in the sense that they form a
cohesive relation? (section 2)
(2) How is this cohesiveness signaled? How can we distinguish
adjacent utterances with no specific coherence relation from
adjacent utterances with a specific, for example, a causal relation?
(section 2)
(3) What counts as a causal relation in spoken interaction? How does
the term 'causal relation' relate to the notion of 'account' as it is
used in conversation analysis (e.g. Heritage 1984: 250-251, 266-
273)?1 (section 3)
(4) If there are no conventionalized elements present to signal the
causal relationship-how is it still possible to identify asyndetic
constructions as structures involving a causal relation, and how
can we arrive at a plausible interpretation of these constructions?
(sections 4.1-4.6)
With regard to (4), the analysis of the data will show that accounts
typically occur after particular actions. These actions, it will be argued,
establish constraints on the interpretation of an asyndetic construction.

2. Signalling connectedness in asyndetic constructions

Cohesiveness between utterances is typically signaled in some way.

Utterances that form a cohesive unit-also termed 'construction' in the
following-are connected either by explicit lexical marking or by more
implicit means, such as prosody. Dealing with asyndetic constructions,
i. e. with structures that are not connected by a grammaticalized element,
for example, a conjunction, an adverb, or a preposition, raises the
question: how is connectedness signaled in these cases?
A first indication as to the connectivity of asyndetically linked
utterances is their sequential proximity.3 By way of illustration, consider
excerpt (1). The participants are arguing about smoking: Erik and
somebody else want to smoke. Some other family members, however, are
still busy eating.
Causal relations in spoken discourse 85

(1) SMOKING (Schwab 4; 49:52)

lMaria: < <h> uwele ma: gsch no () Iguck da hat==s no
knoch[ele zum abnage. >
2Erik: [nix kriegt er me: mir musset rauche. ()
3?Erik: zum wohl. ((Anstoj3en von Gliisern))
4?: [(4 Silben)]
5Anna: fda kommt me tota:l untererniihrt kommt mer hoim,
6 bloj3 weil die rauche wollet.

, 1Maria: uwe dear do you want anything else (.) look there' re
bo[nes to pick. 4
2Erik: [he'll get nothing more we have to smoke. (.)
3 ?Erik: cheers. «clinking of glasses))
4?: [(4 syllables)]
5Anna: [we'll come home totally undernourished,
6 just because these guys want to smoke. '

The tum produced by Erik in line 2, nix kriegt er me: mir musset rauche
'he'll get nothing more we have to smoke', consists of two successive
tum-constructional units which together form a cohesive unit or
construction. The cohesiveness between these two units is of course
supported by the fact that a causal relation can be construed between
them. Yet this causal relation would not be sufficient to create a
construction, were the two units not in sequential proximity.
However, it is not only units produced by one speaker, as in the case
above, which may form a construction, but also units produced by two or
even more than two speakers. The latter case is rare in my data, but it
does exist (cf (14) CAT II below). The interactive production of
constructions by two speakers occurs more often (cf (10) CAR below).
Yet the majority of causal constructions in my data are produced by a
single speaker.
Another means of signaling connectedness is via the prosodic design
of two utterances, in particular the pitch level at the end of the first
utterance. There are two possibilities for signaling connectivity: (i) the
two utterances are produced in one intonation unit, (ii) the two utterances
are produced as two intonation units with the cadence of the first unit
signaling continuity. 5 The latter is found most often in my data in cases
where a causal construction is produced by a single speaker. In these
instances, the fmal pitch movement of the first unit does not reach the
bottom of a speaker's pitch range, i.e. does not end in 'period intonation'
86 Christine Gohl

(Chafe 1988: 2) (transcription symbol: .). It either just slightly falls

(transcription symbol: ;) or rises (transcription symbol: ,) or stays level
(transcription symbol: -), i.e. ends with 'comma intonation' (Chafe 1988:
2). Although in most cases the causal constructions in my data comprise
two intonation units, there are also cases in which a single speaker
produces both parts within one intonation unit. Excerpt (1) SMOKING
illustrates this. The speaker uses a single intonation phrase for the
production of the two units making up the asyndetic construction: nix
kriegt er me: mir musset rauche 'he'll get nothing more we have to
smoke'. The second part, mir musset rauche., justifies the action encoded
in the first part nix kriegt er me:, i.e. Erik's rejection of Maria's offer to
In excerpt the speaker uses two intonation units for the production of
one construction. The participants in this segment are the same ones as in
(1). They are talking about going to church after having had their
Christmas dinner. It is quite late in the evening.

(2) CHURCH (Schwab 3; 1:31:39)

Erik: gott sei dank muft 11 () < <all> Jheut nimme in(d) larch. >
i ddt schnarche.
Uwe: des isch ebe des was [(5 Silben)
Kai: [da: kommsch eh net
zum schnarche -
mir werdet eh schtande. ()
Maria: (nei)nei mir ganget so friih daft me net

'Erik: thank god I don't have to go to church today anymore.

I would snore.
Uwe: that's what [(5 syllables)
Kai: [you won't get round to snoring
there anyway -
we'll have to stand anyway.
Maria: (no)no we'll go so early that we won't have to stand.'

Kai's utterance contains a construction consisting of two intonation units.

The first part of the construction (da: kommsch eh net zum schnarche-
'you won't get round to snoring there anyway-') is accounted for in the
Causal relations in spoken discourse 87

next intonation unit (mir werret eh schtande 'we'll have to stand

A further hint as to the connectedness of two (or more) utterances is
the use of particular modal particles. In my data, particles that occur in
the second utterance are, above all, doch, ja, halt and eben. The literature
on German modal particles is abundant, and there is as much agreement
as disagreement on the meanings and functions of particular particles.
Generally speaking, there are two main functional components of modal
particles according to the literature (Hartmann 1986): (1) they signal the
speaker's attitude towards the proposition of the utterance; (2) they
connect the present utterance with the preceding utterance. The latter
point is much less controversial than the former and is the one which
concerns us here: the modal particles that occur in my data help to signal
the causal relationship, not in the sense that they convey a causal
meaning, but that they underline the connection between the units in
question (cf example (14) CAT II below).
The collection of asyndetic causal constructions in my corpus includes
constructions which exhibit at least one of the features discussed in this
section, i.e. sequential proximity, a single intonation contour or continuing
intonation at the end of a first contour, or the use of a modal particle.
Usually at least two of these characteristics occur together. Sequential
proximity, for example, usually co-occurs with the use of continuing
intonation or the use of a particle or both.
I will now specify my use of the terms 'causal relation' and 'account'
and come back to excerpts (1) and (2).

3. What constitutes a causal relation and an account?

In the case of asyndetic linkage, it is almost impossible to specify the

coherence relation that exists between two adjacent utterances without
looking at the broader sequential and contextual environment of the
utterances in question. This means that the specific relation between the
utterances is achieved via the interaction of their meaning and contextual
aspects (linguistic and non-linguistic context). Semantics alone is not
This can be nicely illustrated by looking at excerpts (1) and (2) again.
In both cases, the causal relation between the two units can be
reconstructed easily by taking into account the situational context and
encyclopedic knowledge: in (2), the speaker presents the fact that people
will probably have to stand in church as a reason for his claim that
88 Christine Gohl

snoring, i.e. falling asleep, will not be possible; it is not possible because
standing is not a good position for falling asleep. In (1), the knowledge
that is required in order to discover and make sense of the causal relation
springs from cultural conventions rather than from knowledge of, for
example, particular biological constraints upon human beings. The
speaker uses the argument mir musset rauche 'we have to smoke' as an
account for the production of the sentence nix kriegt er me: 'he'll get
nothing more:'. For reasons of politeness one should not smoke while
someone else is eating. Therefore, mir musset rauche 'we have to smoke'
justifies the first part of Erik's utterance (nix kriegt er me: 'he'll get
nothing more') by giving a reason for why he says that Uwe should not
have anything else to eat.
The analysis of these examples brings us back to question (3): what
counts as a causal relation in spoken interaction? When is a construction
a causal construction? When is it a construction with a different, e.g.
temporal relation? Which constructions are used by participants in
communication as a means for producing a causal relationship? How do
we know that participants understand a relation between two or more
adjacent utterances as causal? And how can we, from an analytic point of
view, give a plausible interpretation of this relation that goes beyond
intuitive judgment?
One possibility is to look at the way speakers treat and recipients
respond to the utterance in which the construction in question occurs. In
these cases, we can make use of the 'next-tum-proof-procedure', i.e. use
the participants' own interpretations as strong confirmation of the analytic
interpretation (cf Bergmann 1991: 218; Sacks, Schegloff, and Jefferson
1974: 728-729). There are several instances in my data in which a next
speaker refers to a prior asyndetic causal relation with an utterance that
makes it clear that he or she has understood the preceding construction as
containing a causal relation. In these cases, next speakers (or same
speakers in their next tum) use a causal conjunction, a causal adverb, or a
causal phrase like das ist der Grund warum 'that is the reason why' with
respect to the same relation that has been constructed asyndetically
before. These lexical elements or phrases give the analyst a clear
indication that the causality is indeed present for the participants.
Consider excerpt (1) SMOKING again: Erik's utterance consists of a
dispreferred second action (nix kriegt er me: 'he'll get nothing more')
and a following account for this dispreferred action (mir musset rauche
'we have to smoke' ): Erik, although not the addressee of Marie's offer,
rejects her offer to Uwe by implying that if Uwe continues eating, he
(Erik) and the other smokers will not be able to smoke. Anna, in her tum,
Causal relations in spoken discourse 89

takes up both components of Erik's utterance, eating and smoking, and

connects these with the causal conjunction weil 'because': da kommt me
tota:l untererniihrt kommt mer hoim, blofJ weil die rauche wollet 'we'll
come home totally undernourished, just because these guys want to
smoke'. She thereby displays her understanding of the two components
as entities that are causally related in this particular context and displays
her understanding of Erik's previous utterance as a construction involving
a causal relation, although Erik did not use a causal connective.
Unfortunately, cases like these are rather infrequent in my data, which
makes it necessary to look for further regularities in the sequential
environment of asyndetic causal constructions. These regularities, as I
will show in the following analysis, are to be found in the kinds of actions
after which accounts occur. Second parts of causal constructions have in
common that they are accounts: they explain or justify a previous
conversational move, i. e. the action carried out by the first utterance.
Occasionally, the account relates to some non-linguistic action or to a
premise which is reconstructable only in connection with the context. The
vast majority of cases in my data, however, are those in which the causal
unit refers to a linguistic utterance or to implications that can be derived
directly from this utterance.
Part of the definition of accounts adopted here is covered by the notion
of 'account' as it has been employed in conversation analysis (e.g.
Atkinson and Drew 1979: 58-60; Heritage 1984: 250-251; 266-273;
Heritage 1988; Levinson 1983: 306-307, 334;), i.e. actions justifying
"disaffiliative (dispreferred) second actions to invitations, requests and
the like" (Heritage 1984: 272). Viewing causal relations in such a way
means including various types of utterances which do not seem to
establish causal relations at first glance. Consider the following excerpts
from Heritage (1984: 250). In excerpt (3), J asks a question about a train
that is supposed to go on a boat. In (4), the participants are talking about
a child: M wants to know whether the child is alright. J, however, does
not know anything about the child's welfare because the child has not
returned, i. e. J has not met with or talked to her yet.

(3) (Heritage 1984: 250)

J: But the trai: n goes. Does th 'train go 0: n th 'boa: t?
M: .h .h Doh I've no idea:. She ha: sn 't sai: d.

(4) (Heritage 1984: 250) «Concerning a child's welfare»

M: '8 alri:ght?,
J: Well 'e hasn ' c 'm ba-ack yet.
90 Christine Gohl

While it is easy to describe the causal relation produced by M in (3)

because the account-portion She ha: sn 't sai: d. refers to the preceding
action encoded in the utterance .h .h Doh I've no idea:., the situation in
(4) is a different one. Here, the account refers to an action that is not
encoded linguistically, but is constituted by the act of accounting itself,
namely J's inability to provide M with the information she has asked for.
Still, a causal relation is present, since Well 'e hasn' c 'm ba-ack yet.
provides a reason for another action, namely for the production of a
dispreferred second, in this case, an answer lacking the information
requeste d.
This understanding of accounts has to be extended, though. As will
become clear in the course of my analysis, utterances explaining or
justifying a previous action occur not only after dispreferred second
actions, but also after particular first actions.

4. The production of accounts after particular actions

4.1. Analyzing causal talk

The identification of accounts produced via asyndetic causal construction

requires attentive data analysis, starting with the inductive search for
regularly occurring patterns in the original spoken data together with
transcript work. The features described in sections 2 and 3 prove helpful
in recognizing these action structures. In addition, I have made use of
Ford's fmdings with respect to the environments in which explanations
and justifications regularly emerge, either marked by because (1994 and
this volume) or unmarked (this volume). However, her finding that
accounts are frequently produced in contexts involving some kind of
contrast is only applicable to part of the material analyzed here. The
occurrence of accounts, it turns out, is not directly linked to the linguistic
structure which a previous utterance displays, but to the kind of action
accomplished by a previous utterance.
The next step in the analysis is to look for deviant cases, i. e. for cases
in which participants treat the non-production of an account as
problematic or in which they explicitly ask for the provision of an account
at a point when an account is due. These cases are typically less common
because they deviate from normatively required behavior (Heritage 1988:
133). Consider example (5), which is taken from a radio phone-in. The
context is as follows: the caller has problems with his mother, who
interferes in his relationship with his girlfriend. The psychologist is trying
Causal relations in spoken discourse 91

to convince him that his problem is his own fear of coming to a decision:
she advises him to stand by his girlfriend and speak openly to his mother.
The caller, however, rejects this by saying that his mother does not want
him to see his girlfriend at all (line 9):

(5) DECISION (18/4; 15:30)

1Psychologin: mhm .h [abe darumJ gehts,
2Anrufer: [aberwenn aber (ich)]
3Psychologin: ob da: ne entscheidung (-) wirklich
moglich ist;
4 (2.5)
5Anrufer: ja mh
6 0.0)
7Anrufer: ich weifi es seIber aber==aber ich: :,
8 (0.5)
((10 intonation units left out, in which the caller raises objections))
9Anrufer: .hh[meine] mutter will==aber nicht
da:fi==ich zu meinerfreundinfahr.
10Psychologin: [del ]
11 (1.0)
12Psychologin: aha:;
13 (0.5)
14Psychologin: und:
15 (0.5)
16Psychologin: warum will die mutter das nicht?
17 (-)
18Anrufer: ja sie sagt h.
19 meine freundin hat drei kinder,

'lPsychologist: mhm.h [but that's] the crucial point,

2Caller: [but if but (I)]
3Psychologist: if a decison (-) is really possible;
4 (2.5)
SCalIer: yeah mh
6 (1.0)
7Caller: I don't know but I,
8 (0.5)
((10 intonation units left out, in which the caller raises objections))
9CalIer: 'hh[my] mother doesn't like me to see my girlfriend.
10Psychologist: [del ]
11 (1.0)
92 Christine Gohl

12Psychologist: aha:~
13 (0.5)
14Psychologist: and
15 (0.5)
16Psychologist: why doesn't she like it?
17 (-)
18Caller: well she says h
19 my girlfriend has three children,'

Here the account is only produced after the psychologist has explicitly
asked for it (line 16). Between the caller's utterance (line 9)-an
objection to the psychologist's insisting that a decision is necessary-and
the psychologist's question in line 16, there are three long pauses and two
short utterances which do not pursue any further topical talk. That is, the
psychologist can be heard as pursuing an account from her caller by
providing him multiple opportunities at which he can provide such an
The analysis of such deviant cases leads to the conclusion that
"account giving is not merely an empirically common feature that is
associated with unexpected or unlooked for actions, but is a normatively
required feature of such actions. Since failures to provide accounts attract
either overt pursuits of them or sanctions, we can conclude that the giving
of accounts in such contexts is itself a morally accountable matter"
(Heritage 1988: 135).
In sections 4.2.-4.6., different environments will be described in which
accounts, realized by asyndetic linkage, regularly occur in my data. The
environments will be presented in the order of their frequency of
occurence: accounts after dispreferred seconds are most frequent, then
accounts after assessments, requests, and finally after complaints and

4.2. Accounts after dispreferred second actions

Various conversation analytic studies (e. g. Atkinson and Drew 1979, Ch.
2~ Heritage 1984: 265-269~ Pomerantz 1984) have shown that particular
actions, often organized as first parts of adjacency pairs, make a second
action relevant, i. e. these actions call for a second action by virtue of their
sequential implications. These second actions have alternative, but
nonequivalent, realizations: one realization is preferred, the other
dispreferred. After requests, suggestions, offers and invitations, an
Causal relations in spoken discourse 93

acceptance is the preferred and unmarked second action. A refusal, or no

reaction at all, represents a dispreferred second. After assessments, an
agreement is the preferred response, disagreement the dispreferred one.
Dispreferred seconds' exhibit several common features, including delays,
prefaces and-as is relevant in the following-accounts. 12
Example (6) illustrates the provision of an account after a dispreferred
second following a suggestion. Four family members are talking about
how to handle the meat they are having for dinner. Anne, the daughter,
suggests to her father (lines 1 and 2) that he use the scissors that are on
the table to cut the meat.

(6) SCISSORS (Schwab 8; 14:19)

lAnne: nimm halt die () schere;
2 fiir was haschn die;
3 (1.5)
4Fritz: da komm i mit der schere nicht durch;
5Erik: des sind doch knoche;
6Fritz: jetztet;
7 .hh
8Anne: ja grad deshalb hab i dacht,
9Erik: des kannsch beim e pock-pock;
10 da kannsch dknoche no durchschnipple;

,1Anne: take the (.) scissors;

2 what do you have them for;
3 (1.5)
4Fritz: I can't get through with these scissors;
5Erik: there are bones;
6Fritz: now;
7 .hh
8Anne: that's exactly why I thought,
98Erik: you can do that with a pock-pock 14 ;
10 there you can snip the bones;'

With his utterance in line 4, da komm i mit der schere nicht durch 'I
can't get through with these scissors', Fritz rej ects Anne's suggestion
nimm halt die () schere; 'take the (.) scissors;'. The preferred, and thus
structurally unmarked, second part following a suggestion would be its
acceptance. In such a case no further talk would be necessary. In (6),
however, Fritz offers an account for refusing to follow Anne's
suggestion, i. e. for why he does not put it into action. In providing a
94 Christine Gohl

reason for not taking up his daughter's suggestion, Fritz need not make
his refusal explicit: the action is evident although non-verbal. In this case,
asyndetic linkage is the only possibility for the production of the account
because there is no previous utterance or implication to which a
conjunctional clause could be linked.
Accounts occur after various kinds of such dispreferred second
actions. In excerpt (7), from a doctor-patient interaction, an account
follows a request for information that cannot be fully met by the second
The doctor is a specialist in the field of reproduction technological
medicine and the patient, a woman with a hormonal disorder who is
seeking advice about problems that might occur if she wants to get

(7) FINDINGS (2; 5-16)

1Doctor: frau bade sie sind vom frauenarzt zu uns
geschickt worden wegen hormoneller sto: rung steht hier;
2 konnen sie mir da noch was niiheres dazu schon sagen,
3Patient: eigentlich vom hausarzt;
4Doctor: mhm,
5 (-)also ich hab angefangen (---) mit etwa zwanzig,
6Patient: fing das bei mir an mit der pille,
7Doctor: mhm,
8 ich hatte diane fiinfunddreifiich bekommen,
9 aber da hab ich von meinem ersten
frauenarzt die befunde noch nich;
10 [derwar] einfach noch im [urlaub;]
llDoctor: [mhm,] [mhm,]
12Patient: die schick ich dann noch nacho

'lDoctor: mrs bade you were sent to us by your

gynecologist because of a hormonal disorder is written here;
2 could you tell me a little bit more about this now,
3Patient: by my family doctor, actually;
4Doctor: mhm
5 (-) well I started to (---) at the age of twenty,
6Patient: I started to take the pill,
?Doctor mhm
8 I was given diane fiinfunddreillig, (name of the product)
9 but I don't have the fmdings from my first gynecologist yet;
10 [he was] still on [holiday;]
Causal relations in spoken discourse 95

11Doctor: [mhm,] [mhm,]

12Patient: I will send them later. '

In lines 1 and 2, the doctor is asking the patient for some information
about her problem. After correcting one of the doctor's assumptions (line
3), the patient starts reporting on earlier treatment, thus producing the
answer that has been made relevant, i. e. sequentially necessary by the
doctor's request for information. She cannot completely comply with the
doctor's request, however (line 9). Immediately following this partly
dispreferred second action, she produces an account, explaining why she
does not have the findings from her first gynecologist yet: der war
einfach noch in urlaub; 'he was still on holiday;'. i.e. there was no way
of collecting the findings or having them sent.
The excerpt in (8) presents an interactive construction of a sequence
with the dispreferred-second-account structure. This construction shows
that recipients are sensitive to dispreferred actions even if a request for
information and a following dispreferred action occur within the context
of a narration and are embedded in a reported speech situation. Five
family members are talking about a television program on the fight
against crime. In the passage transcribed, the topic is undercover agents.
Fritz is telling the story of a man he knows personally because he used to
be a customer in Fritz's shop.

(8) UNDERCOVER (Schwab 3; 1:21:52)

1Fritz: (( ))
2 der war bei der polizei.
3 ie g,J ludwigsburg.
4 (1.5)
5 und is dann in=en g/ untergrund gegangen.
6 (1.5)
7 unlangst waren ersch e: die eltern da, (---)
8 un na habe gfragt ja: eh:,
9 wie gehts ihrem sohn und so e::,
10 < <p> sagt er () wir wissen nichts, >
11Erik: hajaja
12 die krieget.h [die krieget () en vollstandig neue lebenslauf,
13Fritz: [< <f> wir wissen nichts, () >
14Erik: die krieget en dreifufJzig s e:, ()
15 un=e stattliches bankkonto,
96 Christine Gohl

16 (1.5)
17 na diirfe se 10s lege.

, 1Fritz: « »
2 he used to work as a policeman.
3 ie gJ, ludwigsburg. «name of policestation and town»
4 (1.5)
5 and then he went underground.
6 (1.5)
7 only recently his parents came in, (---)
8 and then I asked them well: eh:,
9 how is your son these days e::,
10 «p> he said (.) we don't know anything,>
11Erik: ohyeah
12 they get .h [they get (.) a completely new personal record,
13Fritz: [<<f> we don't know anything, (.»
14Erik: they get a three fifty s e, ((name of a car» (. )
15 and a large bank account,
16 (1.5)
17 and then they can get started. '

While Fritz is reporting on the conversation he had with the parents of the
former policeman (lines 8-13), Erik comes in (lines 11-12) and produces
an account for the parents' inability to give Fritz the information he had
asked for. Erik's utterance explains the production of a dispreferred
second by the characters in Fritz' story: the parents are not able to answer
Fritz's question wie gehts ihrem sohn und so e::, 'how is your son these
days e::,' because their son, after having gone undercover, was given a
completely new identity (die krieget () en vollstandig neue lebenslauf,)
and various other things that cut him off from his former life (lines 14-
The accounts offered after dispreferred seconds in examples (6), (7),
and (8)-whether they follow requests for information (excerpt (7) and
(8» or suggestions (excerpt (6»-all refer to some kind of inability to
"carry out the proposed or required conversational action" (Heritage
1988: 136). In (6), the speaker accounts for why he is not able to use the
scissors for cutting the meat as Anne has suggested; in (7) the patient
accounts for her inability to completely comply with the doctor's request
for information, and in (8) a third speaker accounts for the inability of two
other-distant-speakers to carl)' out the action that has been called for
by the question asked, namely to provide an answer.
Causal relations in spoken discourse 97

4.3. Accounts after assessments

Accounts are not only offered after dispreferred seconds; they also
emerge regularly after assessments. 15 These assessments 0 f en t ·
lexical markers or phrases that make them 'strong assessments' in the
sense of what Ford has described as 'strong evaluation' (Ford 1994:
548). In German, some of these lexical markers are: besonders
'especially', wirklich 'really', sehr 'very', ganz 'totally', as well as
particular adjectives like abscheulich 'dreadful', wunderbar 'wonderful',
genial 'brilliant' or adjectives in their superlative form, e.g. am
schlimmsten 'worst' and nominal expressions like da war die holle los
'all hell broke loose' .
Accounts after strong evaluative assessments are more frequent in my
data than accounts after assessments that do not express a strong
evaluative stance. This observation leads to the conclusion that it is not
particular lexical items that are responsible for the production of an
explanatory unit, but the action carried out by the evaluative statement,
the assessment.
Let us look at transcript (9). The participants are talking about the
family's cat. The starting point for this discussion is Marie's utterance.
She thinks that the cat is sad because it did not get anything for

(9) CAT I (Schwab 3; 0:07:45)

lMarie: i mochtja nix (sage aber) die katz isch ganz traurig-
2 die hat nix kriegt.
3Fritz: <<f> grad eben ha () (ha n) neido>, ()
4 <<p> des hat er net emol aguckt>.
5Erik: des isch doch dere katz ihr erschtes
weihnachte -
6 die weifi gar net dafi mer do [was kriegt== -
7?Marie: [was los isch
8Erik: etz niichsch jahr () da muj3 me re
vielleicht was gebe;

'lMarie: I don't want to (say) anything (but) the cat is really sad -
2 she didn't get anything.
3Fritz: «f> just now I gave her something>, (.)
4 «p> she didn't even look at it>.
5Erik: it's the cat's first Christmas, you see -
6 she doesn't know that [you get something at Christmas= -
98 Christine Gohl

7?Marie: [what's going on

8Erik: next year you'll probably have to give her something;'

With her utterance die hat nix kriegt 'she didn't get anything' Marie
gives an account for her assessment die katz isch ganz traurig- 'the cat is
really sad-' .
The following excerpt illustrates the interactive production of an
assessment-account sequence. The participants, mother and son, are
talking about the car the son has used to drive to his parents' home. This
car is obviously not his own, but belongs to the company he works for.

(10) AUTO (Schwab 3; 0:27:29)

lMarie: < <cresc> hasch du heut mit dem schone lauto
Jahre> J diirfe;
2Erik: ja:: i mu.f3 ja was darschtelle oder?

'IMarie: «cresc> were you allowed to drive> that

beautiful car today;
2Erik: yeah:: I have to keep up appearances, haven't I?'

With her utterance, Marie is doing two things that call for a second action
by the recipient. She is asking a question and is also making an
assessment of the car the son has used to visit his parents: mit dem
schone auto 'that beautiful car'. The content of this question is marked on
the prosodic level as something extraordinary: first, during its production
the utterance increases in volume; second, there is remarkable pitch
movement on auto Jahre, with the pitch falling from a very high level.
Marie's question - which constitutes the first part of an adjacency pair
- calls for an answer. This is produced by Erik immediately after Marie
has finished her tum (ja:: 'yeah::'). The assessment, however, requires a
separate acknowledgment. This is provided when Erik gives a reason for
what Marie has said (line 2). Note that Erik not only reacts to the content
of Marie's utterance, but also to its prosodic design, which conveys a
somewhat surprised and approving attitude towards the fact that Erik was
allowed to drive this beautiful car. This attitude, however, has a slight
ironic undertone, which is taken up by Erik in his utterance: both Marie's
question and Erik's answer are thus produced in a teasing, slightly ironic
style-judging from content and, especially in Marie's utterance, from
the prosody.
Causal relations in spoken discourse 99

4.4. Accounts after requests

Asking a question is a particular form of requesting something from the

recipient, namely the provision of information (cf example (7)
FINDINGS, in which the doctor asks for detailed information about the
patient's medical history).
Consider now example (11). The participants are discussing the
appropriate time for smoking and the quality of the cigars a family
member has brought along. Erik was offered one of these cigars on an
earlier occasion. He now reports on the bad effects it had on him.

(11) CIGARS (Schwab 3; 0:0:37)

lMarie: do hocksch na un rauchsch st[att daj3 de unsre esse/
2Erik: [ja sind die
zigarre scho was,
3 also mir isch ganz ubel denoch gwase.
4Marie: er==hot==doch==gsagt die beschtn wo==s gibt-
5 (1.5)
6Fritz: was
7Erik: die zigarre,

'IMarie: then you will sit down and smoke

in[stead of our eating our foodl
2Erik: [are these cigars okay,
3 I felt absolutely sick after smoking one.
4Marie: he said the best ones you can get. -
5 (1.5.)
6Fritz: what
?Erik: the cigars,'

In line 3, Erik gives a reason for asking the question ja sind die zigarre
scho was? 'are these cigars okay?'. He asks if the cigars are okay,
because he felt sick after smoking one of them, implying with his
question that the cigars might not be okay. By offering an account, Erik is
giving the recipients some insight into the motive he has for asking the
In the speaker makes a request in the form of an imperative. The
participants are talking about the food, the meat they are having for dinner
in particular.
100 Christine Gohl

(12) MEAT (Schwab 8; 15:18)

1Erik: tausche mer <<p> deins isch kleiner.>
2Kai: gar net -
3 [fwe: noi nemm du i will blo) e schtii[ck; ==
4Kai: [nein (2 Silben) esse.
5Uwe: i will doch net solche brocken.

'IErik: let's swap «p> yours is smaller.>

2Kai: not true -
3Uwe: no you take it Ijust want a pie[ce;=
4Kai: [no (2 syl.) eat.
5Uwe: 1 don't want such big pieces. '

Erik asks Uwe if he wants to swap pieces of meat. Immediately after this
request, he offers a reason for requesting: deins isch kleiner 'yours is
smaller'. In his next tum (line 3), Uwe refuses to accept Erik's request
(noi nemm du 'no you take it') and accounts for the production of this
dispreferred second action in lines 3 and 5: i will blo) e schtiick; 'I just
want a piece', and i will doch net solche brocken 'I don't want such big
pieces' .
Excerpts (11) and (12) show that requests-in these cases formulated
as a question and as an imperative-are followed by accounts. Requests
are face-threatening activities because they put some obligation on the
recipient he or she has to do something for the speaker, either in the form
of a verbal action (provide the information requested) or in the form of a
non-verbal action (do the thing requested). This format makes requests
conversationally sensitive actions-and these are likely to be followed by

4.5. Accounts after complaints and reproaches

A common feature of activities like complaining and reproaching is their

moral bias. "In these activities interlocutors focus on an infringement of
expectations concerning situatively appropriate behaviour and in doing
so, demonstrate their own orientation to moral rules and values"
(Giinthner 1994: 2). Reproaches make second actions by the person the
reproach is directed at relevant, i.e. they call for justifications or apologies
(Giinthner 1993: 3-6). In the data analyzed here, however, reproaches are
not only followed by such second actions (compare (13) below, where
Fritz denies the action he has been reproached for). First speakers, too,
Causal relations in spoken discourse 101

i. e. the persons who are doing the reproaching, offer accounts for
reproaching someone or complaining about something.
In (13) the topic is once again the family cat. Anna is blaming Fritz for
having given his cold to the cat.

(13) COLD (Schwab 3; 1:49:38)

JAnna: mensch jetz hasch die katz scho angschteckt mit deiner
2 der nie:st scho de ganze d[a:g;
3Fritz: [des hat er heut morge scho gmacht.

'lAnna: good heavens you've given the cat your cold=

2 he's been sneezing all d[ay
3Fritz: [he did it this morning already.'

From the external, visible fact der nie:st scho de ganze dag 'he's been
sneezing all day', Anna draws the conclusion that Fritz, who has a cold,
must have given it to the cat. She formulates this conclusion as a
reproach: both the content of this utterance and the way it is expressed
(e.g. the term mensch, which is very often used at the beginning or at the
end of an accusatory utterance) give it its reproachful character. Right
after this unit, Anna offers an explanation and a justification for the
conclusion and the reproach: the cat's been sneezing all day. 16
Reproaches are face-threatening activities, i.e. activities which are
likely to threaten both the face of the person the reproach is directed at
and the social relationship between the parties. Offering an account after
such an action makes a reproach more understandable because it reveals
the reasons behind it and gives the person blamed the chance to react to
the reproach in a way which is more customized than a mere refusal
would be: he or she can exculpate him- or herself by referring to the
reason for the reproach. This happens in line 3 in (13): Fritz makes it
clear that he refuses to take responsibility for the cat's cold because the
cat has been sneezing all morning. He implies that it must have caught
the cold elsewhere.
It will not come as a surprise that accounts after reproaches emerge
not only after a speaker has blamed some other participant for something,
but also after complaints about a third party. Consider the following
transcript. The content is the same as in CAT I: three family members are
talking about the meal they have just had and are considering the 'poor
cat', who seems to be sad because it did not get anything.
102 Christine Gohl

(14) CAT II (Schwab 3; 0:07:45)

lMarie: i mochtja nix (sage aber) die katz isch ganz traurig-
2 die hat nix kriegt.
3Fritz: <<f> grad eben ha () (ha n) neido>, ()
4 <<p> des hat er net emol aguckt>.
5Erik: des isch doch dere katz ihr erschtes weihnachte -
6 die weifJ gar net daft mer do [was kriegt== -
7?Marie: [was los isch;
8Erik: etz niichsch jahr () da muj3 me re vielleicht was gebe;

'lMarie: I don't want to (say) anything (but) the cat is really sad-
2 she didn't get anything.
3Fritz: «f> just now I gave her something>, (.)
4 «p> she didn't even look at it>.
5Erik: it's the cat's first christmas you see -
6 she doesn't know that [you get something at christmas= -
7?Marie: [what's going on;
8Erik: next year you'll probably have to give her something;'

With his utterance des hat er net emol aguckt 'she didn't even look at it',
Fritz complains about the cat's behavior with respect to his action: he has
given the cat some food, but the cat did not appreciate it. Erik and Marie
account for this behavior with the utterances that follow. Thus, once
again, the causal relation is constructed interactively. Fritz produces an
utterance which is treated by a second speaker, Erik, as a consequence:
the cat does not appreciate the food because it is her first Christmas.
Therefore, you cannot expect her to react in a particular way. The second
part of Erik's utterance (die weifJ gar net dajJ mer do was kriegt- 'she
doesn't know that you get something at christmas-') provides another
potential reason for the cat's behavior. At the same time, another speaker
(probably Marie) comes in and completes another explanatory unit (die
weifJ gar net was los isch; 'she doesn't know what's going on;') for what
Fritz has said earlier (des hat er net emol aguckt 'she didn't even look at
it' ). Both explanatory units, Erik's and Marie's, contain reasons for the
cat's behavior, as well as consequences with respect to the first part of
Erik's utterance (des isch doch dere katz ihr erschtes weihnachte- 'it's
the cat's first christmas-'): hence, because it is the cat's first Christmas,
she does not know that you receive something on this occasion and she
does not know what is going on. The interactive nature of this production
of accounts shows that both speakers, as recipients, consider the action of
Causal relations in spoken discourse 103

blaming the cat for not appreciating the food as an action that calls for

4.6. Activity types as constraints on the interpretation of asyndetic


In my data, asyndetically linked accounts follow dispreferred second

actions and various other types of action which are organized as
adjacency pairs in that they make a second-verbal or non-verbal-action
sequentially necessary: first assessments, for example, call for second
assessments (Pomerantz 1984), requests call for responses in the form of
an acceptance or refusal, reproaches for denial or admission (Heritage
1984: 269; Levinson 1983: 308,336). The causal units in the latter cases
account for the production of the first actions. However, not all first parts
of actions that are organized as adjacency pairs require accounts. Rather,
accounting occurs, in particular, after activities which are sensitive to the
social relationship between the parties. Requests and reproaches, are, like
dispreferred seconds, face-threatening actions-requests, because they
impose some obligation on others, reproaches, because they imply
criticism of others. Although assessments are not face-threatening in the
same way, they are still sensitive to possible disagreement-especially as
'strong assessments'-in that they express a speaker's highly subjective
evaluative stance towards a particular fact or event. Accounting after first
assessments makes a subjective position more understandable and thus
more likely to lead to a preferred second assessment.
The analysis of the data shows that participants are sensitive to the
production of accounts in these particular environments. Therefore, I
would argue that these activity types constrain the interpretations that
arise when a causal unit is not explicitly marked as such. An utterance is
much more likely to be interpreted as an account, i. e. as an utterance
explaining and/or justifying a previous conversational move, if the
preceding utterance, by virtue of its sequential and social implications,
calls for accounting.
Concerning accounts after dispreferred seconds, Heritage (1988: 136)
argues that they help to avoid threatening the social relationship between
the participants. Accounting for the inability to produce an answer, for
example, leaves the responsibility for the dispreferred second with the
speaker. Not accounting for a dispreferred second or denying the right of
the first speaker to carry out the first action would constitute a threatening
act, endangering the participants' relationship. The same holds for the
104 Christine Gohl

production of accounts after reproaches: reproaches are, by their very

nature, disaffiliative acts threatening the recipient's face. Giving a reason
for such a disaffiliative move makes the act of reproaching more
understandable because it reveals the motives behind the reproach.
With respect to requests, which, of course, are much less face-
threatening than reproaches, it can be argued that a request always puts
some obligation on the recipient(s): he or she is asked to do something for
the speaker, for example, provide him/her with the information requested.
Again, providing a reason for requesting makes the act more
understandable because it gives the recipient some insight into the
speaker's motives.
In sum, the interpretation of an asyndetic construction depends on the
features described in sections 2 and 3 and on the sequential environment
of the construction in question, i. e. on the activity it can be construed as
relating to.

5. Concluding remarks

This study shows that speakers may give reasons without usmg a
conventionalized causal connective element.
Different sequential environments were investigated in which speakers
produce such asyndetically linked causal relations in order to give
accounts. These environments are constituted by different actions which,
for various reasons, call for accounting. One group of these activities-
dispreferred seconds, requests, and reproaches-is particularly sensitive
to the social relation between participants, either because these activities
are face-threatening or because they put some obligation on the recipient.
The second group, including first (strong) assessments, is also socially
risky. Accounts following strong assessments make the production of a
preferred second more likely because they support what is often a highly
subjective attitude on the part of the speaker and thus make it easier for
recipients and possible next speakers to understand the speaker's stance
and react to it in a preferred format.
Ford has shown in her studies (1994 and this volume) that accounts or
explanations and justifications often emerge in contexts involving some
form of contrast, which mayor may not be marked, for example, by
contrastive markers or negation particles. The analysis of the asyndetic
causal constructions in my data, however, shows that only about half of
the accounts occur in contexts involving contrasts. This means that
accounts have a much broader context of occurrence. In fact, much
Causal relations in spoken discourse 105

suggests that the production of accounts should be seen as motivated by

making reference to the kinds of actions which call for them. This is not
to deny that there may be a correlation between linguistic structure and
the kind of action that is carried out by an utterance-dispreferred
seconds, for example, often contain negation markers-but the
correlation is by no means obligatory. Consequently, analysis must go
beyond the linguistic realization of an utterance and ask what a speaker is
doing with a particular utterance.
Related to the issue of structure and action is the question of linguistic
practices. Which ones do speakers have at their disposal for the
production of reasons? What circumstances influence the choice of
linguistic means, especially the choice between explicit conjunctional
linkage and asyndetic linkage, when speakers produce accounts for
previous actions? First tentative analyses show that one of many possible
factors is whether speakers are giving reasons for their own actions or for
actions produced by other speakers: asyndetic linkage seems to be
preferred to conjunctional linkage in cases of interactively produced
constructions. Moreover, particular environments apparently allow only
asyndetic linkage. For example, when an account refers to a non-verbal
action, there is no previous utterance or implication to which a
conjunctional clause could be linked (cf example (6) SCISSORS).
These questions, which still need investigation, touch upon the general
issue of the linkage between (linguistic) practices and actions, discussed,
for example, in ScheglofI (1996, 1997). Further studies along the lines
suggested here could bring about a fuller understanding of the relation
between various linguistic practices and the action of 'giving reasons' in
spoken interaction.

Appendix: Transcription conventions

[] overlap
(.) micro-pause
(-), (--), (---) short, middle or long pauses of approx. 0.25-0.75 seconds, up to ca.
1 second
(2.0) estimated pause of more than 1 second
and=uh slurring within units
lengthening, according to its duration
uh,ah, etc. hesitation signals, so-called "filled pauses"
? intonation phrase-final: high rise
intonation phrase-final: mid-rise
intonation phrase-final: level pitch
106 Christine Gohl

intonation phrase-final: mid-fall

intonation phrase-final: low fall
/ break-off
« » interpretive comments over a stretch of speech
( ) unintelligible passage
(such) uncertain transcription
~ » onllssion of text
pitch step up
j, pitch step down
«t» forte, loud
«p» piano, soft
«all» allegro, fast
«cresc» crescendo, becoming louder


* I am grateful to Dagmar Barth and Susanne Gtinthner as well as to the

editors of this volume for their valuable comments on earlier versions.
1. For the connection between the term 'account' and the more generic
phenomenon of interactional accountability discussed in ethnomethodology
and conversation analysis, see Heritage (1988: 128, 138-141).
2. I view speaking as a form of social action (Heritage 1984: 135) and therefore
linguistic utterances as a means for carrying out particular actions. From this
premise it follows that coherence relations, e.g. causal relations, that hold
between utterances also hold between actions, i.e. an utterance referring to
some other utterance does so because of the action(s) carried out by that
particular utterance. This applies to utterances to the extent that they consist
of one or more turn-constructional units. For the notion of turn-
constructional unit (TCD), see Sacks, Schegloffand Jefferson (1974).
3. A first analysis of explicitly linked causal structures compared with asyndetic
causal structures shows that explicit linkage allows for more conversational
space inbetween accountable action and account. Exceptions are, moreover,
deviant cases (Heritage 1988: 133-135), in which an account is produced
with a delay, usually when a first speaker does not offer the account
immediately but only after a second speaker has elicited its production (see
excerpt (5) DECISION in section 4.1.).
4. The translations are meant to convey more the content of the utterances than
the stylistic subtleties of the Swabian dialect.
5. Chafe (1988: 1) defines the prototypical intonation unit as "single coherent
intonation contour characterized by one or more intonation peaks and a
cadence that is recognizable as either clause-final or sentence-final."
6. For this connective function see also Franck (1980: 253) and Lindner (1991:
174). Besides emphasizing their connective function, Ickier (1994: 377) also
stresses the interactive role of modal particles: they "anchor particular
utterances in larger contexts, which are usually characterized by their
Causal relations in spoken discourse 107

dialogic nature." For the interactive function of the particles doch, eben, and
ja in spoken discourse see also Liitten (1979).
7. Occasional remarks on the possible causal meaning of modal particles can be
found for doch (Feyrer 1998: 187-192), auch (Paul 1966), halt (Abraham
1983, Hinrichs 1983) eben andja (Abraham 1983).
8. This investigation is based on the analysis of approximately 7 hours of
spoken German comprising about 4 1/2 hours of informal conversational
data, i.e. family conversations, 1 1/2 hours of interactions in a medical
setting, and 3/4 of an hour of radio phone-ins. Most of the data was recorded
in the Swabian area of Southern Germany. I wish to thank Ute Lacher-
Laukenikat and project C4 of the SFB 471 for making their data available to
9. For the interpretation of asyndetic conditional structures, i.e. paratactic
conditionals, see Thumm (1999).
10. For the notion of 'dispreferred second' and a more detailed discussion of
accounts after dispreferred second actions, see section 4.2.
11. In fact, in excerpt (3), the utterance'.h .h Ooh I've no idea:.' also constitutes
an account: M accounts for her failure to provide J with the information
requested (Heritage 1984: 250). Compare also excerpt (6) in section 4.2.
12. For more details, including the marked-unmarked distinction, and a survey
of the most important issues concerning preference organization, see
Levinson (1983: 332-345). On the use of causal markers in English and
Japanese conversations in dispreferred environments, see Ford and Mori
13. For another example illustrating the production of a dispreferred second-
account-sequence after a suggestion, see excerpt (1) SMOKING. While in
excerpt (6) SCISSORS the declining component is not verbalized, in excerpt
(1) SMOKING the speaker formulates the declining of another speaker's
suggestion in a very explicit and highly marked way.
14. 'Pock-pock' is presumably the name for a kind of a fowl.
15. For a discussion of assessments and their sequential implications in spoken
German interaction, see Auer and Uhmann (1982). Some of their transcripts
show that second speakers may explicitly ask for the reason of an assessment
given by a first speaker if the latter does not account for the assessment him-
/herself (Auer and Uhmann 1982: 8-9).
16. The formulation der nie:st scho de ganze da:g ('he's been sneezing all day')
falls into the category of Extreme Case Formulations (Pomerantz 1986).
Such formulations are used by speakers when they justify complaining about
something or somebody to "legitimize a complaint and portray the
complainable situation as worthy of the complaint" (Pomerantz 1986: 228).


Abraham, Werner
1983 Starke Satzverkniipfer-Sinnsog und Partikelenquete. In: Harald
Weydt (ed.), Partikeln und lnteraktion, 25-39. Tiibingen: Niemeyer.
108 Christine Gohl

Atkinson, J. Maxwell and Paul Drew

1979 Order in Court. London: Macmillan
Auer, Peter and Susanne Uhmann
1982 Aspekte der konversationellen Organisation von Bewertungen.
Deutsche Sprache 10: 1-32.
Bergmann, Jorg R.
1991 Konversationsanalyse. In: U. Flick, E. von Kardoff, H. Keupp, L.
von Rosenstiel and S. Wolff (eds.), Handbuch Qualitative
Sozialforschung, 213-218. Mfinchen: Psychologie Verlags Union.
Blakemore, Diane
1987 Semantic Constraints on Relevance. Oxford: Blackwell.
Couper-Kuhlen, Elizabeth
1996 Intonation and clause combining in discourse: The case of because.
Pragmatics 6/3: 389-426.
Chafe, Wallace
1988 Linking intonation units in spoken English. In: John Haiman and
Sandra A. Thompson (eds.), Clause Combining in Grammar and
Discourse, 1-27. AmsterdamIPhiladelphia: Benjamins.
Feyrer, Cornelia
1998 Modalitiit im Kontrast: Ein Beitrag zur ubersetzungsorientierten
Modalpartikelforschung anhand des Deutschen und des
Franzosischen. Frankfurt am Main: Peter Lang.
Ford, Cecilia E.
1994 Dialogic aspects of talk and writing: because on the interactive
edited continuum. Text 14/4: 531-554.
Ford, Cecilia E. and Junko Mori
1994 Causal markers in Japanese and English conversations: A cross-
linguistic study of interactional grammar. Pragmatics 4/1: 31-61.
Franck, Dorothea
1980 Grammatik und Konversation. Monographien Linguistik und
Kommunikationswissenschaft 46. Konigstein, Ts.: Scriptor.
Gfinthner, Susanne
1993 "Kannst du auch fiber ANDERE Leute LAsTERN." Vorwiirfe als
Formen moralischer Kommunikation. Arbeitspapier Nr. 9.
MORAL-Projekt. Universitat Konstanz.
1994 Moral voices: Prosodic features of reproaches in "why"-formats.
Arbeitspapier Nr 11. MORAL-Projekt. Universitat Konstanz.
1996 From subordination to coordination? Verb-second position in
German causal and concessive constructions. Pragmatics 6/3: 323-
Heritage, John
1984 Garfinkel and Ethnomethodology. Cambridge: Polity Press.
Hartmann, Dietrich
1986 Semantik von Modalpartikeln im Deutschen. Deutsche Sprache 14:
Causal relations in spoken discourse 109

Hinrichs, U.
1983 Konnen Abtonungspartikel metakommunikativ funktionieren? In:
Harald Weydt (ed.), Partikeln und lnteraktion, 274-290. Tubingen:
IckIer, Theodor
1994 Zur Bedeutung der sogenannten 'Modalpartikeln'. Sprach-
wissenscho:ft 19: 374-404.
Levinson, Stephen
1983 Pragmatics. Cambridge: Cambridge University Press.
Lindner, Katrin
1991 'Wir sindja doch alte Bekannte' The use of Germanja und doch as
modal particles. In: Werner Abraham (ed.), Discourse Particles,
163-201. Amsterdam: Benjamins.
Lutten, Jutta
1979 Die Rolle der Partikeln doch, eben, und ja als Konsensus-
Konstitutiva in gesprochener Sprache. In: Harald Weydt (ed.), Die
Partikeln der deutschen Sprache. 30-38. Berlin: de Gruyter.
Pasch, Renate
1997 WElL mit Hauptsatz-Kuckucksei im DENN-Nest. Deutsche
Sprache 3: 252-271.
Paul, Hermann
[1966] Deutsches Worterbuch, 6.A. Tubingen: Niemeyer.
Pomerantz, Anita
1984 Agreeing and disagreeing with assessments: some features of
preferredldispreferred turn shapes. In: 1. Maxwell Atkinson and
John Heritage (eds.), Structures of Social Action, 57-101.
Cambridge: Cambridge University Press.
Pomerantz, Anita
1986 Extreme Case Formulations: A way of legitimizing claims. Human
Studies 9: 219-229.
Sacks, Harvey, Schegloff, Emanuel A., and Jefferson, Gail
1974 A simplest systematics for the organization of turn-taking for
conversation. Language 50/4: 696-735
Scheutz, Hannes
1998 Weil in spoken German. Presentation given at the conference on
'Cognitive and Discourse Perspectives on Cause, Condition,
Concession and Contrast'. Oktober 1998. UniversiUit Konstanz.
Schegloff, Emanuel A.
1996 Confirming allusions. American Journal of Sociology 102/1: 161-
1997 Practices and actions: Boundary cases of other-initiated repair.
Discourse Processes 23: 499-549.
Schleppegrell, Mary J.
1991 Paratactic BECAUSE. Journal ofPragmatics 16: 323-337.
110 Christine Gohl

Thumm, Markus
1999 How to do some things without 'if. Eine empirische Untersuchung
zu parataktischen Konditionalkonstruktionen im gesprochenen
Englisch. MA-thesis, Department of Linguistics, University of
Uhmann, Susanne
1998 Verbstellungsvariation in Weil-Satzen: Lexikalische
DifIerenzierung mit grammatischen Folgen. Zeitschrift .fur
Sprachwissenschaft 17/1: 92-139.
Constructions with if, since, and because:
Causality, epistemic stance, and clause order*

Barbara Dancygier and Eve Sweetser

English causal and conditional conjunctions show significant overlap both

in functions and in the ranges of grammatical constructions in which they
occur. These phenomena are related: one reason why different conjunctions
can express similar meanings is that the lexical semantics of one
conjunction may explicitly involve meaning which is not present in the
other conjunction's semantics but can be expressed by accompanying
grammatical constructions. Using Mental Spaces theory (Fauconnier
1985/1994, 1997), we analyze the semantics of English if, since, and
because, examining the interaction between their lexical semantics and the
semantics of their grammatical surroundings. Among the factors involved
are predictive function and causal relations (Dancygier 1993, 1998;
Dancygier and Sweetser 1996, 1997), epistemic stance (Fillmore 1990a,b),
and the relation of the mental space to the speech setting (Sweetser 1990,
Dancygier and Sweetser 1996, Dancygier 1998).

1. Introduction

In this paper we will analyze the uses of causal and conditional

constructions using the conjunctions if, since, and because. These
conjunctions specialize in distinct areas of meaning, but their areas of
use overlap significantly: a comprehensive analysis, therefore, should
explain the overlap as well as the distinctions between them. It has been
noted, for instance, that since could substitute for if in (1) without
significant change in interpretation, and that a causal relation underlies
the contingency described in (2):

(1) If (as you say) he is in town now, maybe we should phone him?
(2) If I take my pills regularly, the symptoms go away.

In (1), imagine that the hearer has said earlier in the conversation that
the relevant person is in town now. In the speaker's reference to this
previous statement, she could use since instead of if, with quite similar
112 Barbara Dancygier and Eve Sweetser

effect. (2) can naturally be interpreted to mean that taking pills regularly
makes the symptoms disappear-a causal interpretation related to
readings of because and since. There is thus a significant degree of
overlap among the uses of the conjunctions involved. But the potential
for such interpretations will often depend on factors other than the
conjunction itself; notice that when is a plausible substitute for if in (2)
but not in (1), suggesting that the lexical and constructional contexts are
important in determinining possible construals of the relations between
conjoined clauses.
In positing functional overlap between these conjunctions, we are by
no means suggesting that they are synonymous--or even that they have
specific senses which are synonymous. On the contrary, we will try to
specify the source of the similarities as well as to clearly delineate the
differences. Both similarities and differences are best accounted for not
by focusing only on the conjunctions themselves, but by describing the
range of constructions they participate in. The main goal of our analysis
is thus to distinguish aspects of constructional meaning which are
involved in various uses of each of the conjunctions, and to show how
particular formal parts of constructions combine with each other to
contribute to varying interpretations of the whole. 1
We will bring up three relevant parameters in trying to compare and
contrast the four kinds of constructions. First, we will consider the role
of causal models in interpretation. Causality is involved in the
constructions in a variety of ways and on various levels of
interpretation. Constructional aspects of meaning, and the cognitive
domains in which these constructions are interpreted, can be as
important as the choice of conjunction, in bringing causality into the
interpretation. The conjunctions will also be compared in terms of the
type of mental space set-up they engage in; this is often also specifically
signalled by the use of verb forms marking negative epistemic stance.
Finally, we will review the constructions with respect to their roles in
backgrounding or foregrounding elements of interpretation in different
clause order configurations.
Our analysis of meaning will be framed in terms of Mental Spaces
theol)' (Fauconnier 1985, 1997), an extremely general formal theory
which provides mechanisms for talking about the ways in which we
connect cognitive structures with each other. To some extent, traditional
treatments of conditional semantics in terms of Possible Worlds have
aimed at capturing similar insights. Somewhat in the way that an if-
clause can be said to set up a Possible World within which a then-clause
Constructions with if, since and because 113

is construed, we would argue that an ~clause sets up a Mental Space

which is the background for the construal of the then-clause. For
example, in If his computer gets repaired, then he'll finish the paper by
Friday, the ~clause sets up a space wherein the computer gets repaired,
and within that space, the speaker predicts that the paper will get
finished by Friday.
However, mental spaces are a more general mechanism than possible
worlds, referring not only to very partial cognitive "world" or
"situation" constructions as well as to more complete ones, but also to a
variety of non-world-like structures which can be connected and
mapped onto other cognitive structures. For example, the two mental
spaces consisting of our mental list of restaurant customers and our
mental list of their orders are mapped onto each other to allow
utterances such as The ham sandwich wants his check now, wherein the
name of the food item ordered can be used to refer to the customer who
placed that order (Fauconnier 1985[1994]). World-like structures are not
the only consistent, complex mental models which humans develop, and
which humans can interconnect with each other.
We have exemplified the use of if as a mental space builder.
Numerous linguistic devices can play such a role: in 1984, in the
picture, possibly, etc. All such expressions require that the content of
the clause or sentence be considered in a mental space other than the
base space-the reality space of the hearer and the speaker. In what
follows we will show how the conjunctions if, since, and because
participate in the construction of mental spaces and how the kind of
space being set up affects the way the conjunctions are used.

2. Mental spaces, causality, and levels of interpretation

We will start this section by outlining types of mental spaces which are
particularly relevant to our interpretation of the use of conjunctions. We
will then show how these types of mental space set-ups correspond to
levels of interpretation of constructions and how causal relations are
established at each of these levels.
114 Barbara Dancygier and Eve Sweetser

2.1. Mental spaces and cognitive domains

Causal and conditional constructions, and particularly if-conditionals,

are known to vary widely in function. It would be economical and
possibly even elegant to be able to attribute some of this functional
diversity to a few specific parameters of interpretation. In fact, mental
spaces theory opens up the possibility for such an elegant treatment, by
allowing us to talk about different kinds or classes of mental spaces. For
example, in a more or less "standard" conditional such as (3),the
speaker sets up a space of mental content (cf Sweetser 1990,1996;
Dancygier and Sweetser 1996)-a space which is about a possible state
of affairs in the world, namely the computer getting repaired.

(3) Ifhis computer gets repaired, he 'llfinish the paper by Friday.

Within this content-domain space (further filled out by the speaker's

and hearer's general knowledge about the individuals referred to), the
speaker predicts an added aspect of the content of this mental space-
the paper getting finished by Friday. As argued in Dancygier (1993,
1998), prediction in the content domain represents the central use of
conditional constructions.

"now (not Friday yet)"

Future Space = P Counter-Future Space =

(neutral stance) ",p
"his computer gets repaired" "his computer doesn't get repaired"

Figure 1. If his computer gets repaired, he'll finish the paper by Friday.
Constructions with if, since and because 115

As seen in Figure 1, a predictive conditional such as (4)

automatically also sets up an alternative space, where P does not hold
(no computer repair) and Q also does not hold (no finished paper by
Friday). Although this is not a logical property of conditionals, it is a
necessary property of the predictive use of conditionals, as we have
argued elsewhere (Dancygier and Sweetser 1996, 1997); (3) would not
be relevant as a prediction if the paper were just as likely to be finished
regardless of the computer repair.
By contrast, in an example like (4), the speaker does not predict
something about a good Thanksgiving on the basis of information about
seeing the addressee before Thursday.

(4) If! don't see you before Thursday, have a good Thanksgiving!2

Rather, the speaker sets up a discourse context, a speech-act space

wherein Thursday has arrived without her seeing the hearer, and then
utters a speech act which is to be taken as effective within that space.
This can't quite literally be true, since of course the well-wishing is
uttered in the current context and can't exactly be retracted; but the
conditional certainly modifies the context of interpretation from the
previously accepted one. In particular, the hearer cannot now reply with
but isn't it a bit early for that? As we might expect from the non-
predictive nature of (4), no alternative space is set up; that is, although
the speaker knows she may see the addressee before Thursday, she is
not providing for that separate contingency. Her good wishes are offered

P: S doesn't see H between

Monday and Thursday

Figure 2. If I don't see you before Thursday, have a good Thanksgiving.

116 Barbara Dancygier and Eve Sweetser

A conditional can thus set up, imagine, and negotiate possibilities in

either the world of linguistically described content, or in the world of
current speech-act context and performance, where the events and
participants are the Speaker, the Hearer, and the speech act with its
interpretive context. Speech act conditionals have been noticed as a
special and interesting case for some time (Vander Auwera 1986,
Sweetser 1990). Mental spaces theory provides a simple mechanism for
the description and analysis of the difference between them and content
conditionals: conditional constructions may build speech-act spaces as
well as content spaces.
What other sorts of spaces might speakers construct with
conditionals? Sweetser (1990) has argued for a distinction between
content conditionals and epistemic conditionals, which follow the
speaker's reasoning process in a conditional manner, as in (5):

(5) If he finished the paper by Friday, his computer must have gotten

In the described world of content, the paper's being fInished is not a

condition for the computer's being repaired, rather the reverse. But in
the epistemic world, where knowledge of effects limits possible
conclusions about causes, knowing when the paper was finished is a
condition which may be sufficient to warrant an inference about
computer repairs. In (5), an epistemic space is set up, wherein the
speaker believes or knows that the paper was finished, and it is in that
space that she draws the conclusion about the repairs. (Possible
paraphrase: 'If I know (or we know) that the paper was finished, then I
(we) must conclude that the computer was repaired. ')
A further point here is that the speaker of (5) is very possibly not
doing initial setup of a space wherein the paper is known to be finished.
We might say that she is evoking, rather than actively setting up, that
space, if she is borrowing structure set up by an interlocutor who has
just informed her that the paper was fInished by Friday.
Dancygier (1993, 1998) has set out a further variety of conditional
uses, as in (6), wherein metalinguistic negotiation is the subject of the
conditional relationship:

(6) Chris wants you to meet her fiance, if that's the right word for
Constructions with if, since and because 117

the speaker perhaps thinks that Chris is using fiance as a euphemism for
'live-in partner.' Crucially, the speaker is not here construed as
negotiating either the facts of the case (the content), or conclusions to be
drawn from them in the epistemic domain; nor is she conditionally
performing the speech act of stating that Chris wants the hearer to meet
a particular individual. She is apparently metalinguistically commenting
on, or presenting as merely conditional, her use of the label fiance to
refer to that individual.
In this case, we need a somewhat more complex mental space
structure, which we have referred to (Dancygier and Sweetser
1996,1997) as a metalinguistic space. A metalinguistic space
(generically diagrammed in Figure 3 below) consists of a speech-
context space, within which there is a space of content being discussed
and a space of linguistic forms available for referring to that content.
Metalinguistic conditionality concerns mappings between the content
space and the form space within a particular context. Here, the speaker
constructs a particular form-content mapping (uses fiance to refer to
Chris' partner), and then sets that in the context of a particular
metalinguistic space by calling the mapping into question explicitly (if
that's the right word). By implication, such a conditional usage at least
suggests that other metalinguistic spaces could be imagined, with
different form-content mappings; however, in (6) there is no explicit
attempt to evoke other specific mappings (e.g., partner).

Figure 3. Metalinguistic space

118 Barbara Dancygier and Eve Sweetser

Every speech act, of course, involves interlocutors and a context and

the interlocutors' ideas of each other's understandings of the content and
of the forms used to express it. So all of this mental space structure is
present and accessible in any speech situation. However, not all of this
structure is necessarily actively at stake in a given conditional use. (6)
demands that we bring all of these elements into our diagram overtly,
because in a metalinguistic conditional the content-label mappings
themselves are what are being conditionally represented and compared.
In (4), we needed to overtly represent speech-act context spaces
involving S and H to analyze a speech-act conditional, but the content-
label mappings were not part of the conditional relationship. Rather,
they were taken for granted as background to it. A diagram of (5) would
need to represent possible epistemic spaces of S, but would take the S/H
context and label spaces for granted as background.
Our claim, then, is that in all these cases, a conditional construction
involves setting up a mental space (in the case of if conditionals, this is
the job of the ~clause), and requesting construal of something (in if
conditionals, the then-clause or main clause) within that space. Much of
the diversity of interpretation can be attributed to the fact that the spaces
themselves can be quite diverse sorts of objects, related to the linguistic
form in a variety of ways. Crucial to this analysis is the idea that
domains related to the current speech interaction are generally
privileged with respect to mental space construction (cf Sweetser
1990): they are automatically "accessible" for interpretation, even
without explicit mention, in ways that other spaces are not. 3 This fact
has wide-ranging effects on interpretation, far beyond conditional
constructions. As noted in Sweetser (1990), broad classes of linguistic
forms show the same kind of possibility for multiple interpretations
depending on the level or domain accessed. Parallel to conditionals are
causal and adversative conjunctions (as well as many coordinate
conjunction usages, which are beyond the scope of the present paper):

(7) .Joe turned down the stereo because Sam was studying.
(Causal relation is between state of affairs described in P and
event described in Q, i.e. between contents of the clauses.)

(8) Sam is (must be) studying, because Joe turned down the stereo.
(Causal relation is between speaker's knowledge about content
ofP and speaker's conclusion about content ofQ.)
Constructions with if, since and because 119

(9) Could you turn down the stereo, because I'm trying to study.
(Causal relation is between the contextual state expressed in P
and the speech-act performance of the request in Q.)

(10) OK, Chris introduced me to her partner, since we're being

politically correct.
(Causal relation is between contextual situation P and a
particular form-content mapping used in Q.)

(11) Since you're a linguist, what's the Russian wordfor "blender"?

(Speech-act level: the addressee's professional status causes or
enables the current speech act of questioning. )

Examples such as these can be multiplied with ease in the domain of

conjunction. Some conjunctions seem to prefer certain domains: since in
English has preferentially a temporal sense in the content domain4
(Chris has been studying since he came home), but has readily
accessible causaVenablement senses in the other three domains (a
speech-act domain example is [11]). Ranging farther afield, Hom (1985)
has laid out in detail the differences between content and metalinguistic
uses of negation. And the well-known contrast between deontic and
epistemic modal uses is interpretable in terms of the content/epistemic
contrast here invoked. Thus Joe must be home by ten, because I say so
imposes a modality of compulsion in the content domain of Joe's
getting home. But Joe must be home by ten, because I always see his
light go on then involves instead a metaphoric 'compulsion' of the
speaker's reasoning processes: 'I am forced to conclude that he gets
home by ten, because 1 know that 1 always see his light go on then' (cf
Sweetser 1982, 1990, Talmy 1988).
What all these examples share is an implicit contribution of the
speech setting to the interpretation: some part of the linguistic form (for
example, because) need not be interpreted directly with respect to the
content expressed by surrounding forms, but can be interpreted as
meaning something about the speaker's mental processes or the speech
interaction, even though those mental processes or that speech
interaction have not been explicitly mentioned. So conditionals are not
alone in allowing speakers to implicitly access speech-act, epistemic and
metalinguistic structure as well as content, in building an interpretation.
As we shall see, however, each of the conjunctions participates in
mental space set-up in a somewhat different way.
120 Barbara Dancygier and Eve Sweetser

2.2. Conjunctions versus constructions: sources of causality in the

content domain

As we claimed above, causal relations in content mental spaces are

understood in terms of standard causal models, including an active
causation model wherein one event or state of affairs actively brings
about another. Content-level causality also includes enablement (cf
Sweetser [1990]). Causal because occurs in examples like He is tired
because he's been up all night, where staying up late is presented as the
cause of ensuing tiredness, and also in sentences such as He will come
and spend Christmas with us because I'm paying for his flight, where a
free ticket enables the person to spend Christmas with his family, rather
than causing him to do so. Such content-level causal models constitute a
major part of the actual interpretation of constructions containing
because and if.

2.2.1. Two ways of depicting causality: because and if

Causal conjunctions like because introduce causality or enablement into

the interpretation as part of their lexical semantics, although other
constructional features and the context help the hearer locate the causal
link in the appropriate cognitive domain. (As we will try to show in
section 4, intonation patterns and clause order in constructions with
because signal content and non-content interpretation of causality in a
fairly regular way.) In the cases of non-causal conjunctions, causal
interpretations may arise from other aspects of the context and
surrounding grammatical constructions: She insulted me and I got angry
is easily interpreted as meaning that the insult caused the anger, on the
basis of iconic interpretation of clause order, plus real-world knowledge
about likely results.
Although because lexically expresses causality, the direction of the
causal chain is marked by the conjunction's placement in a larger
construction: the because-clause presents the cause, the main clause
presents the result. Consequently, even though the conjunction seems to
be solely responsible for the causal meaning, the way in which causality
participates in the interpretation of the sentence is also dependent on the
subordination construction. It may initially seem trivial to focus on the
obvious fact that the because-clause is marked as representing the cause,
as opposed to the caused event, in the causal relation. But as Lightbown
Constructions with if, since and because 121

and Spada (1993) observe, children learning their first language may in
fact hypothesize that principles of iconicity override meanings
introduced at the lexical level and produce sentences like You took the
towels away because I can't dry my hands. In such cases, the strategy
used is to present events in their sequential iconic order (the towels
disappeared from the bathroom first, and consequently the child had
nothing to dl)' her hands with later on). The child needs to learn not only
the causal meaning of because, but also the way to use it
constructionally to mark causal connections correctly. 5
In many content-domain causal conjunction examples, only the
causal relation itself is asserted, while the two events or situations
between which the causal relation holds are presupposed background. In
the classic I'll do it because I want to, and not because you told me to,
the speaker does not assert either that she will do as she is told, or that
she wants to do so, or that she was told to do so: but she is necessarily
asserting that her actions are caused by her wishes and not by the
addressee's commands. .[fclauses generally set up a new mental space
distinct from the Base space of the speaker's assumed construal of
"reality", but because-clauses do not set up such a space. Main clauses
of content-domain because constructions decribe situations which may
already be known to hold in the interlocutors' current base space. 6 The
because-clause then offers a causal explanation of how that situation
came about, also not necessarily on the basis of brand-new information.
The causal relationship is at the core of the message.
Since is similar to because in being semantically causal, and in that
both are often said to presuppose the truth of their complements. These
differences interact with other parameters, such as information structure
and typical clause order. Since will be discussed in more detail in
sections 3 and 4, where we focus on epistemic stance and information
The conjunction if presents an interesting case. Unlike because, it is
not lexically causal. In the content domain, however, if-sentences
typically express causal or enablement relations among events. (12), for
example, would typically convey the message that the conditioner will
cause the hair to get softer.

(12) Ifyou use this conditioner, your hairwillfeel much softer.

Similarly, in example (3) above, it-is likely to be assumed that the repair
will enable the owner to start using the computer again.
122 Barbara Dancygier and Eve Sweetser

Dancygier (1993, 1998) has argued that causal meanings enter the
interpretation of content conditionals via conditionals' primary function,
prediction. 7 Prediction in conditionals is a type of reasoning which
consists in setting up a hypothetical (typically future) mental space and
attempting to predict its consequences based on knowledge of typical
cause-effect chains and general world-knowledge. In (12) above, the
speaker sets up a space in which the hearer applies the recommended
conditioner, and predicts in that space that the hearer's hair will become
softer as a result. Such causal interpretations tend to arise in conditionals
which set up content mental spaces, where causation-based predictions
can be made on the basis of assumptions of sequential correlations
among events or states of affairs and general knowledge. But the
interpretation of speech-act or epistemic conditionals also involves
setting up causal relations: the speaker of Since you're a linguist, what
do you think of Chomsky? constructs relationships of causality and
enablement as part of the background to the act of asking a question.
The most notable formal feature distinguishing English predictive
conditionals is the combination of verb forms: for future-reference
conditionals, a present-tense verb form is used in the If-clause, and the
modal will in the main clause (cf Fillmore 1990a,b). This verb form
pattern characterizes the whole sentence (not the two clauses in isolation
from each other) as representing a certain kind of reasoning.
Predictiveness (and, consequently, causality) is part of the interpretation
of certain If-constructions as wholes, rather than of the conjunction if or
any other lexical element of the sentence. Let us also note that such
constructions do not assert any of the clauses. The If-clause cannot
express an assertion because the conjunction sets up a hypothetical
mental space, different from the base space where assertions could be
made. The main clause describes the predicted result in that same
hypothetical space, so it is not asserted either. What a predictive
conditional asserts is the causal dependency and correlation between the
events or states of affairs described by its clauses.
To sum up, content-domain conditionals are interpreted causally as a
result of predictive uses which are explicitly marked by formal markers
such as verb forms. This is one of the basic differences between
sentences with because, and those with if. In the former, the conjunction
itself introduces causality; in the latter, constructional meaning is the
source of the causal interpretation. Though the constructions with
because and if are different in a number of respects (verb forms, clause
order, mental space set-up, to mention just a few), they can both convey
Constructions with if, since and because 123

information about the causal relations between the content of the

subordinate clause and that of the main clause. These differences and
similarities are related to other aspects of the constructions, such as their
epistemic stance and clause order, which we will consider in the
sections to follow.

2.3. Causality in other domains

As seen above, clauses marked by causal and conditional conjunctions

are often used to set up or evoke epistemic, speech act, or metalinguistic
mental spaces, in which relations are established among beliefs or
speech acts, not events or states of affairs. Does causality play any role
in such spaces?
Causal models are certainly present in the interpretation of non-
content domain constructions, but we will have to distinguish more than
one level of causal links to show how they contribute to the overall
interpretation. In the content spaces, causal links can be established
between events or states of affairs described. In the other domains,
similar links can be established between beliefs or between beliefs and
speech acts. In example (8) above (Sam is (must be) studying, because
Joe turned down the stereo), the causal relation is between the speaker's
knowledge about the content of P and the speaker's conclusion about the
content of Q. Similarly, in (11) (Since you're a linguist, what's the
Russian word for "blender"?, the belief expressed in the since clause
causes or enables the act of asking about Russian. We can interpret
metalinguistic uses similarly. For instance, in example (6) above (Chris
wants you to meet her fiance, if that's the right word for him) the
speaker uses the word fiance while also pointing out that the (possibly
mistaken) belief in fiance being the right word is all that enables her to
do so.
It is known that causal/enablement relations in non-content domains
often go in a direction opposite to the causal links in the content domain
(Sweetser 1990, Dancygier 1993, 1998). This is especially common in
the epistemic domain, where the speaker is often reasoning from effect
to probable cause. In (8), for example, we assume that Sam's need to
study normally causes Joe to tum his music down, so the fact of the
music being turned down enables one to infer that Sam is studying. In
fact, it is possible for epistemic sentences to be ambiguous, depending
on the underlying assumption about a content causal chain. For
124 Barbara Dancygier and Eve Sweetser

example, a sentence like If Tim went to the embassy, he has his visa
documents can have two epistemic interpretations. If the speaker
believes that Tim was going to obtain his visa documents at the
embassy, she may use the information about Tim's visit there to infer
that he now has his documents. If, on the other hand, the speaker
believes that acquiring the documents enabled Tim to go to the embassy,
she concludes that he had them prior to the visit. In either case,
knowledge of cause-effect links in the content domain is used in a
reasoning process carried out in the epistemic domain.
Causal models can thus participate in the construction of meaning on
various levels. This is due not only to our understanding of causality,
which allows us to locate it in various types of mental spaces, but to our
understanding of the relations among those spaces, and, more generally,
to our understanding of speech interaction. This understanding is
dependent on what we may see as a nested complex of mental spaces.
At the center of this complex is some representation of the content of
what is said: events or states of affairs which are referred to by the
speaker. The content space need not make any reference to the speaker's
reasoning processes or to the speech exchange itself (although these
may also be addressed as explicit COlltent: I am asking you why you need
an extension on your paper). However, reasoning processes relevant to
the speech interaction almost inevitably make reference to the content
space: the speaker may, for example, be reasoning specifically about the
structure of the content space. In He must be home; I see his coat, the
speaker's use of must refers to her understanding that her reasoning
processes are under compulsion, not that the described subject's actions
are compelled; but the reasoning processes about the act of returning are
necessarily based on the speaker's understanding of the content world,
where both the return and the coat's presence reside.
The structure of the speech exchange will necessarily involve both
the idea that some content is expressed, and the understanding that the
speaker has reasoning processes related to that content, plus the added
understanding that there are a speaker and a hearer interacting in the
current speech setting, with all the social interactional issues attendant
upon conversational exchange. As we have seen, each of these levels of
mental space structure may involve causal models appropriate to the
level. But the nesting effect explains how different causal models can
enter the interpretation at the same time. This includes the possibility of
exploiting a cause-effect chain wherein P causes Q, as the necessary
Constructions with if, since and because 125

background for an inferential chain wherein a belief that Q leads to a

conclusion that P.

3. Epistemic stance

As noted above, predictive conditionals with future reference typically

display a characteristic verb form pattern in English, with a present
tense verb in the protasis and the predictive modal verb will in the
apodosis. In fact, the same pattern is found in constructions with
temporal conjunctions like when: If/when we find the paper, we'll xerox
it right away. If and when have similar functions in this example, in that
both mark backgrounds to main-clause predictions.
The central difference between if and when is that when commits the
speaker to the reality of the space described in the when clause, even if
that reality will only take place in the future. If, on the other hand,
makes no such commitment. This contrast exemplifies the broader
phenomenon of epistemic stance (Fillmore 1990a,b), a parameter which
is by no means restricted to temporal and conditional clauses. When
identifies the speaker's beliefs with P as a description of the real state of
affairs (expresses positive epistemic stance); if, on the other hand,
expresses the speaker's lack of full positive stance with respect to the
content. The non-positive stance of if need not commit the speaker to a
negative or skeptical stance, but does indicate that she thereby distances
herself from full commitment to the contents of the ~clause. Other
aspects of a conditional construction may go further, and explicitly mark
the speaker's leaning towards non-belief in the reality of the described
situation. Sentences such as If we found the paper, we would xerox it
right away, If we had found the paper, then we would have xeroxed it
right away, and Even if we found/had found the paper, we wouldn't
xerox/wouldn't have xeroxed it right away express negative epistemic
stance towards the content of the conditional clause-and hence also
towards the content of the main clause, since that is situated in the
mental space defined by the conditional clause. The expression of
negative epistemic stance is formally marked by the use of verb forms
which in earlier work we have dubbed distanced, which pick up the
Present + will predictive pattern and backshift it, that is, contribute
added layers of past morphology. 8
We have seen examples of positive epistemic stance (with when) as
well as neutral and negative epistemic stance (with ij). We noted that
126 Barbara Dancygier and Eve Sweetser

these constructions use a special predictive pattern of verb forms, in

both non-distanced and distanced format, and that the distanced forms
correlate with negative epistemic stance.
Since and because, our two remaining conjunctions, express positive
epistemic stance~ the clauses in their scope are seen as representing
"factual" or presupposed information. This aspect of their use seems to
result from the mental space set-up they engage in; in contrast with
conditionals, they do not build new hypothetical spaces, but instead
evoke aspects of whatever space is currently the base space. They also
do not require any special pattern of verb forms. As one might expect,
they do not allow the use of distanced verb forms, which are appropriate
only in the cases of hypothetical mental spaces marked with negative
epistemic stance.
However, not all conditionals are predictive. We have shown above
that causality-asserting predictive reasoning (and, consequently, the use
of predictive and distanced verb forms) is necessarily characteristic only
of conditionals setting up their mental spaces in the content domain. It
has been noted (cf Dancygier 1993, 1998; Dancygier and Sweetser
1996, Sweetser 1996) that the use of if in other domains need not
involve any restrictions on the use of verb forms, and that such
conditionals need not necessarily bring up another alternative space9 .
Recall our earlier discussion of speech-act uses of since such as Since
you're a linguist, what do you think of Chomsky? A parallel speech-act
conditional could serve a similar function: imagine a context where the
interlocutor has stated that he is a linguist, and the speaker says If you're
a linguist, what do you think of Chomsky? Here the speaker does not
intend to compare the space where the addressee is a linguist with other
possible spaces, nor to predict the speech act, but only to situate the
actually performed speech act in a well-defined space. In such non-
predictive cases, ~clauses may rely on context in ways which are
reminiscent of the more positive epistemic stance of since-clauses.
Are we now calling into question our previous argument that ifhas a
non-positive (neutral) epistemic stance, while since has an overtly
positive stance? One might initially feel that (13) expresses a positive
stance regardless of whether since or if is used.

(13) If/Since he's (so) hungry (as you say he is), he'll want a second
Constructions with if, since and because 127

Sentences like (13) certainly suggest that the truth of the proposition
('he's hungry') has been asserted in earlier discourse. Rather than
setting up a novel space, they evoke a space already contextually
accessible. However, referring to a space available in previous discourse
does not necessarily mean that the speaker always aligns herself with its
content to the same degree. As argued by Dancygier (1998), ~f indicates
non-commitment to the clause it marks: but non-commitment covers a
wide range of possible attitudes, from strong disbelief to near-
commitment. Even in an example such as (13), where ~fand since seem
interchangeable, since sounds more completely positively aligned than
if, although in the presence of other strong indications of positive
attitude, the degree of non-commitment conveyed by if may be
So there are important differences between if and since. If maintains
a neutral stance, while since takes a positive one. Also, since, unlike ~f,
has explicitly temporal senses. However, in appropriate contexts, there
is predictable functional overlap between the two conjunctions-
predictable, in the sense that it falls out naturally from a mental space
analysis of the constructions and contexts in question. But in defining
the possible contexts of overlap, we need a clear understanding of the
factors which make some particular assertion or claim accessible to, or
assertable by, some particular participant in the speech interaction. What
licenses a speaker to treat a particular content as already accessible to
the hearer, or as not being subject to questioning by the hearer? We shall
make use of the standard constructs of speech act theory to define
contexts where if and since are naturally isofunctional. As will become
evident, the relevant factors have a general application to the analysis of
the relationship between space-builders and informational structure:
space-building background information has to be brought up in a way
appropriate to its status in the speech context.
When fills a genuinely temporal space-building function, as does
since when used to mean 'from some past time to the present.' Non-
temporal since, however, builds up causal relations; it states a causal
relationship between the space which it evokes (rather than constructs)
and the main clause. An ~clause may, as in (13), merely re-enact the
building of an already accessible space--or it may build a truly novel
space, as in many predictive constructions. A causal since-clause,
however, always evokes a space which is presupposed by the speaker,
as accessible already to the audience.
128 Barbara Dancygier and Eve Sweetser

Discourse accessibility of a mental space can arise in a variety of

ways. One of these ways is the speaker's right to access her
interlocutor's prior conversational record, quoting or summarizing or
making inferences from it.

(14) Since he's so inefficient, maybe you should make a formal

(The speaker accesses her interlocutor's prior claims about
inefficiency, as background to her suggestion.)

Since, with its positive epistemic stance, marks (14) as explicitly

accepting the claim about the inefficiency-although of course this
acceptance may be insincere or ironic. We could expect the speaker to
have used if here, if she were not fully convinced or simply wanted to
avoid taking sides. Another source of evocation is the immediate
physical context of the exchange. In (15), since is appropriate in two
situations: one where the speaker is expressing her own knowledge
about the rain, and one in which the hearer has reported the rain to the
speaker who has not observed it firsthand:

(15) Since it's raining, we can't go outside after all.

As (15) shows, contextually evident information should normally be

marked with since rather than if, unless for some reason doubt is being
thrown on its validity. For example, a speaker who is physically
examining the hearer's clothing to decide whether it is warm enough or
weatherproof enough could say:

(16) Since you're wearing those shoes, we'd better not go to the

But for a speaker considering the possibility of the hearer's changing

into sneakers before departure, ifwould be a valid choice.
Typically, then, the context and the speaker's evaluation of the
situation (and in particular, her epistemic stance towards it) dictate the
choice between if and since. Certain categories of information are by
their nature presupposed to be known to the speaker, or to the hearer,
while others are not. In particular, speaker-based preconditions for a
speech act are presumed to be accessible to the speaker, while hearer-
based ones are not. Hearer-based preconditions, on the other hand, are
Constructions with if, since and because 129

presumed to be uniquely accessible to the hearer; the speaker cannot

treat them as presupposed unless the hearer has already put them on the
conversational record.
Predictably, the non-presuppositional nature of if contrasts with the
presuppositional semantics of since in marking clauses expressing
speech-act preconditions. A speaker may hypothesize, but not
presuppose, the hearer's willingness to answer a question or comply
with a request or tolerate a potentially inappropriate remark:

(17) If/#Since it's not rude to ask, what made you get interested in
(since would be acceptable only if the hearer had already made it
plain that the question was acceptable.)

(18) If/#Since you don't mind, could you hold this stack ofbooks for a
(since is acceptable only if the hearer has indicated he doesn't

(# is here used to indicate that the sentence so marked is grammatical,

but would be unacceptable without special context; one would have to
imagine that the addressees of (17)-(18) had just said things like, No,
it's not rude to ask about my personal intellectual development, or qr
course I don't mind holding somethingfor you.)
On the other hand, speaker-based preconditions demand a positive
epistemic stance; it is normal for a speaker to presuppose her authority
on these subjects. She does not need to hypothesize, normally, about
whether or not she is speaking technically:

(19) Since/#IfI'm speaking technically, I'll call it a misdemeanor and

not a crime.
(The if form would be acceptable as a general condition, meaning
'Whenever I'm speaking technically, ... ' It cannot, unlike the since
form, refer to the current speech act.)

And if the speaker is making a request on the preconditional basis of her

own desires, she is presumed to know (and not hypothesize) her stated
reasons for those wishes:
130 Barbara Dancygier and Eve Sweetser

(20) Since/#If I'm stranded without a car, could you drop me at the
BART station?

If for some reason the speaker were not the most knowledgeable person
about her own access to a car, then if would seem more acceptable in
(20). (Imagine a situation where H, a service station employee, has just
told her that her husband picked up the car an hour ago, while she was
expecting to pick it up and drive home in it.)
Speaker and hearer are each presumed to be authorities on their own
states of mind and body, and not on each other's. Hence the speaker
should normally use since when evoking her own mental and physical
states, and ifwhen referring to the hearer's:

(21) If/#Since you're really tired, maybe we should stop work now.
(since is acceptable if the hearer has himself indicated his
tiredness. )

(22) Since/#IfI'm totally exhausted, maybe we should stop work now.

By this analysis, since should only be employed when the hearer

must concede the speaker's right to take a positive epistemic stance to
the content of the space-evoking clause. There remain a variety of
reasons to use if rather than since. First, of course, the material may
really be one towards which the speaker genuinely does not take a
positive stance (truly neutral or negative-stance hypothesizing), or about
which the speaker must defer to the hearer (as in hypothesizing about
the hearer's tiredness). But inheritance from the conversational record is
especially complex with regard to epistemic stance. On the one hand, if
the hearer has just said that it is raining, and there is no reason to doubt
his word, the simplest scenario may be to adopt it as given and mark it
with since. On the other hand, if (cf Akatsuka 1986; Schwenter 1997,
1998) marks a non-positive stance, which may mark genuinely reserved
judgement (If she was at· the party, I didn't see her), or may just
acknowledge that this claim is not the speaker's own. Quotative if can
thus be more polite than since, or less so, depending on whether the
issue is acceptance of the contents or taking credit for the claim.
How does since interact with the various parameters of interest to us?
Obviously, being positive in epistemic stance like when, it does not
cooccur with verb forms marking negative stance. Like if, it allows
interpretation of the relationship between the two clauses as existing in
Constructions with if, since and because 131

different domains. There are content, epistemic, and speech-act uses of

since; and it is easy to find metalinguistic ones as well (cf Sweetser
[1990; chap. 4]):

(23) Since my plane was delayed, I'm phoningfrom London.

(Content the plane's delay-a situation treated as already
familiar to the hearer-is what necessitates the phone call.)

(24) Since he's phoning from London, his plane must have been
(Epistemic: the presupposed knowledge of the phoning causes the
conclusion about the plane's delay.)

(25) Since you're so smart, when was George Washington born?

(Speech act the on-the-floor claim that the hearer is smart is what
provokes the query.)

(26) Since you're so picky, I'll just put another "trout" on the grill.
(Metalinguistic: The addressee has just corrected the speaker's
use of the term salmon to refer to the wrong species of fish. )

Sweetser comments that since may be preferentially used in the non-

content domains, precisely because of its presuppositional status; there
is a relationship between presupposed or given background-clause
material and non-content usage (cf. Dancygier 1998), as Nikiforidou
(1990) has also pointed out for Greek conditional forms. It is in the non-
content domains that it is often most useful to contextualize a main
clause, to make clear on what grounds or against what background
assumptions the speaker is performing the present speech act, drawing
the present conclusion, or using the present linguistic label.
Unlike when, which at the content level can refer to cotemporality at
any time relative to the speaker's present, since demands present or past
(realis) tenses in its content-level uses. It is not used to make content-
level predictions, and does not participate in the predictive use of tense;
backshifting does not occur in since-clauses, and in general predictive
uses of will in since-constructions can be interpreted as expressing
prediction at some other level than that of content. This may follow
from the fact that a conditional's business is to set up a (possibly
entirely unreal) space, as background to some further inferences, while
132 Barbara Dancygier and Eve Sweetser

that of a causal conjunction is to link situations, claims, and conclusions

causally to a given background.

4. Clause order, space set-up, and causal/conditional relations

Given the flexibility of clause order in conditional and causal

constructions, how do different orderings of the main and subordinate
clause interact with space building? This is a very general issue indeed:
not only if=-conditionals, but most of the broader class of conditionally
used constructions, involve conventional use of the order of the
expressions of antecedent and consequent. It seems intuitively natural
that the space-builder clause should precede the clause which elaborates
the space; and this is of course consistent with the observed tendency for
if=-clauses to precede their main clauses. 10
However, the other order is certainly also possible, although (as we
shall see) it is not possible for all conditionals-the clause containing
the apparently space-building (or more neutrally, space-delimiting)
material can follow the main clause. Doesn't this violate some basic
understanding of how space-structuring proceeds? The answer is that
perhaps it might, if the cognitive processing of mental spaces were done
in a vacuum; but in fact, context may already have done a great deal of a
speaker's building for her, in which case an explicit space-building
clause may only be needed to confirm or clarify an already-constructed
mental space.
Looking first at if=-conditionals, we may note that all three of the
following organizations ofP and Q are possible:

(27) ifP, Q If the home computer breaks down, I'll work at my


(28) Q, ifP I'll work at my office, if the home computer breaks


(29) Q ifP I'll work at my office if the home computer breaks down.

(27), diagrammed in figure 4, follows our intuitively basic strategy of

first building a space defined by P, then adding the assertion of Q to the
structure of that space. Note that the relevant space could still be in
Constructions with if, since and because 133

some sense pre-built by the discourse; (27) could respond to What 'II you
do ifyour home computer breaks down?

Set up conditional space defined by P, then assert Q within

that space. Space setup first, then contents.
There may be a background assumption that P is
already a cognitively accessible possibility, but Q is
not accessible in this way.
(Implicitly set up """'P, ~ space as well.)

Figure 4. If the home computer breaks down, I'll work at my office.

(28) instead asserts Q, without first specifying in what space Q applies.

Here it seems almost necessary that some background space be already
established, to make sure that the hearer will be adding Q to the relevant
space. The speaker can make sure that has been achieved by adding a
disambiguating if P of her own, following Q, but this would be an odd
way to introduce a genuinely new space. Note that in both (27) and (28)
we have standard predictive conditional structure, with the implicit set-
up of a """'P, ......,Q space alongside the P, Q space; the difference lies not in
the relations expressed but in the order of their construction and its
relation to the informational context of the discourse.
The contrast between (28) and (29) exemplifies a more general
contrast between conjuncts juxtaposed with and without comma
intonation. (29) would be an answer to Would you ever work at your
office? or When will you work at your office?, rather than to What will
134 Barbara Dancygier and Eve Sweetser

you do if the home computer breaks down? Its major assertion is the
identity of the conditional relation under which the working in the office
would happen; the main clause content is already "on the floor." In
mental space terms, (29) begins by expressing a prediction Q, but in an
intonation pattern which makes it clear that Q is not being
independently asserted, and is therefore unlikely to belong to the base
space. The focal stress (marking new information), and the final
intonation fall (marking the end of an assertion), happen only once we
are identifying the space P within which Q is to be placed. Q is "old
information," while its space-setting is not.
So (27)-(29) correspond to different sequences of instructions for
building what are rather similar final mental space constructs. (27) is the
"canonical" sequence of setting up a space and then making a prediction
within it. (28) might be said to involve at least a potential pre-built P-
space, to which the utterance adds Q, subsequently confirming with an
~clause that yes, P was the intended space for elaboration by Q. (29)
could be seen as involving a presupposed Q, and asserting the
alternative space structure which links that Q with P rather than with ,-..;P.
These differing sequences of instructions are useful in different
contexts, just the way physical instructions to construct the same object
might usefully differ depending on which components of the object
were already pre-assembled or ready to the addressee's hand.
The three space set-up options presented in (27)-(29) differ in
foregrounding or backgrounding causal relations. (27), being a typical
predictive conditional, asserts the causal relation (rather than any of the
clauses) in the way discussed above. (28) is different in that it initially
asserts Q, and thus pushes the nature of the relation between P and Q
into the background. In fact, as Dancygier (1998) claims, this clause
order in conditionals is perhaps most common among speech act and
metalinguistic conditionals, which often perform a speech act in the
current space, and use the ~clause to negotiate the background structure
of that space. In (29), the stress on identifying the space in which the
"old" prediction is valid gives additional focus to the causal relation on
which the prediction is based. Dancygier (1998) argues that this clause
order and intonation pattern is available only to predictive, causality-
asserting conditionals, while all the other types of space-settings are
excluded. (29) thus represents a clause pattern which constructionally
requires causality to be asserted. 11
Similar clause order options are available for because-constructions,
but the preferences are distinctly different. While ~clauses typically
Constructions with if, since and because 135

precede their main clauses, clauses with because typically follow theirs.
This regularity (observed by Ford 1993) is influenced by the fact that
because-constructions do not set up new spaces, but establish causal
relations in the base space. It is thus common for these constructions to
first mention the fact requiring explanation, and then establish the cause
of the state of affairs in question. Consequently, the main contrast
among because-sentences is between two intonation contours available
in the Q because P pattern.
Sweetser (1990, 82ff) discusses parallel examples involving causal
conjunction, as examples of Chafe's (1984) generalization about
"bound" (commaless) and "unbound" (comma) intonation: comma
intonation demands an interpretation involving assertion of both the
main clause and the causal connection between the clauses, while
commaless intonation is readily given an interpretation which
presupposes the assertion (in this case, prediction) embodied in the main
clause, and newly asserts only the causal connection between the
clauses. Thus (30) is likely to be understood as asserting that the reason
for Anna's love is Victor's resemblance to her first love, but
presupposing that Anna loves Victor; (31), on the other hand, asserts
that Anna loves Victor, as well as the reason why.

(30) Anna loves Victor because he reminds her o..f herfirst love.

(31) Anna loves Victor, because he reminds her ofherfirst love.

Sweetser comments that the bound interpretation is unlikely to occur

in conjunction with a speech-act or epistemic reading of the causal
relationship between the clauses. Thus, (31) but not (30) is readily
susceptible to non-content causal readings-for example, to conveying
the message that the inference about Anna's love for Victor is caused by
knowledge about a resemblance to her earlier sweetheart, rather than the
content-level message that the love is caused by the resemblance.
Sweetser's explanation for this contrast is that it would be exceptional to
presuppose the main clause of a speech-act or epistemic conditional,
since this would involve presupposing the very speech act being
performed, or the new conclusion being drawn.
The mental-space structures of (30) and (31) above are similar to the
conditional examples (28) and (29). The main difference between the
causals and conditionals is the preference for PQ clause order in
conditionals and QP clause order in causals, which appears to be a
136 Barbara Dancygier and Eve Sweetser

consequence of the differing space-building functions of the two

conjunctions. Space-building background has a natural reason to
precede the content of a newly-built space, while a base-space state of
affairs in need of explanation may well reasonably be described first,
before giving the explanation for why or how it came to be that way.
In contrast with because and when, since seems awkward in the
commaless Q conjunction P construction: #1'1/ marry you since you're
so sweet, or #He invited me since I live next door are strange, though it
seems normal to say Since I live next door, he invited me or I'll marry
you, since you're so sweet. Since does not seem to allow foregrounding
or focus on the assertion of the causal relation, perhaps does not even
assert it 12 . In this, since contrasts with a wide range of temporal
conjunctions (before and after behave like when) as well as with
because and if. Compare the following examples of conjunctions in
constructions which employ stress or syntactic focalizing structures to
mark the focus of contrast:

(32) I love you because you're sweet, not because you're rich.

(33) We'll eat when Sue arrives, not now.

(34) They had coffee after lunch, not before dinner.

(35) #He invited me since I live next door, not since I'm his boss.

(36) It's because you're sweet that I love you, not because you're rich.

(37) It's when Sue arrives that we'll eat, not now (/not when Joe

(38) It's after lunch that they had coffee, not before dinner.

(39) #It's since I live next door that he invited me, not since I'm his

Various causal and adversative conjunctions fall on one side or the other
of this contrast: though and while behave like since, and phrases like in
spite of the fact that or despite the fact that behave more like because.
Interestingly, since in its temporal sense does allow assertion and focus
of the temporal relation:
Constructions with if, since and because 137

(40) It's since the pollen came out that I've been sniffling, not since I
moved to the new apartment.

(41) I've been sniffling since the pollen came out, not since I moved to
the new apartment.

Of the two causal conjunctions we have looked at only one, because,

readily allows focus on the causal relationship itself Interestingly,
another conjunction which can foreground the causal link in this way is
if, in its predictive use.
A look at clause order and intonation options thus reveals further
similarities and distinctions between the constructions under scrutiny.
Both if and because constructions allow the use of sentence-final
adverbial clause placement and commaless intonation to unambiguously
assert the causal relation between the contents of the clauses, while the
"comma" intonation usually renders the sentences ambiguous between
domains. On the other hand, since does not seem to allow foregrounding
or focus on the assertion of the causal relation. These facts are added
evidence of the varying degrees to which asserted, presupposed, and
construed causality are integral parts of building these constructional

5. Conclusion

The functions of causal and conditional constructions can be compared

along a variety of parameters. We considered both the contribution of
conjunctions as lexical items, and the constructions in which these
conjunctions appear. Particular conjunctions participate in mental space
configurations in differing ways: some conjunctions set up their own
spaces, some evoke spaces set up in preceding discourse, some merely
elaborate the base space of the ongoing interaction. Further, important
differences arise with respect to the cognitive domains in which the
spaces are set up, but significant aspects of the overall interpretation
result from the spaces being nested one within each other.
Such parameters of interpretation correlate significantly with formal
features of constructions, such as the choice of verb forms, clause order,
or intonation. Some of these formal features may be claimed to invite
certain types of mental space set-up or contextual use, regardless of the
choice of conjunction. One good example of such a feature is clause
138 Barbara Dancygier and Eve Sweetser

order and accompanying intonation, since we observe recurring patterns

of interpretation in constructions which use the same clause order and
intonation patterns.
It is often difficult to isolate the study of a small group of linguistic
forms. Temporal and adversative conjunctions, which were not
originally meant to appear in our data, had to be mentioned along with if
or since; otherwise, important facts about the behavior of our primarily
chosen conjunctions would have to be missed. As we have also claimed
elsewhere (Sweetser 1990; Dancygier 1993, 1998) coordinate
conjunctions and even conjunctionless and paratactic constructions 13
also share some constructional characteristics and aspects of
interpretation with if, because, and since.
What is needed is an analysis which uses parameters of
constructional meaning (verb forms, clause order, intonation, use of
mental space builders) to outline the range of constructions which
participate in the construal of related meanings (causality, sequentiality,
conditionality), and explores the similarities and differences between the
constructions with respect to these parameters. We hope to have shown
that such an analysis can reveal generalizations and aspects of
constructional meaning which accounts of individual conjunctions can
describe only partially.


* For Ekkehard Konig, on his sixtieth birthday.

1. It is impossible to cite all the works which have helped us to develop an
understanding of the semantics of grammatical constructions. Among salient
ones are Fillmore, Kay and O'Connor (1988), Fillmore and Kay (1999),
Langacker (1987, 1991a, 1991b), and Goldberg (1995). Sweetser (1999) deals
generally with the relation between lexical meaning and constructional
2. We are grateful to the late Suzanne Fleischman for this example as well as for
characteristically insightful comments on it.
3. By "automatically accessible," we mean that speakers do not need any special
shared experience, knowledge or context besides the fact of the speech
exchange, in order to have implicit access to these spaces.
4. Though see section 3 for a full discussion of uses of since in different
5. Traugott (1982) gives an illuminating discussion of the differences between
earlier stages of English, wherein both clauses in a causal relation were
marked, and the modern English situation.
Constructions with if, since and because 139

6. Or, of course, in whatever space is currently being added to; the crucial fact is
that no new space need be set up when because is used.
7. For discussion of the relationship of then to predictive function, see
Dancygier and Sweetser (1997), Schiffrin (1992), and Iatridou (1991, 1994).
8. Conditionals with distanced form have been discussed extensively as
"counterfactual" or "unreal". The discussion of this issue exceeds the limits of
the present paper. The reader is referred to Sweetser 1990, Dancygier 1993,
1998, Dancygier and Sweetser (1996, 1997). For reasons why distanced
conditionals should not be treated as counterfactual, see Comrie (1986); for
discussion of the crosslinguistic connection between past tense forms and
distanced forms, see James (1982) and Fleischman (1989).
9. Cf. Dancygier and Sweetser (1996, 1997), for discussion of predictive and
non-predictive epistemic and metalinguistic conditionals, and for analysis of
reasons why content conditionals are necessarily predictive and speech-act
conditionals necessarily non-predictive (see also Sweetser (1990), Dancygier
10. Cf Haiman 1980, 1986; also Ford 1993 on because vs. if.
11. All the patterns just discussed are also found among even if constructions,
with the same priorities and restrictions involved.
12. As Elizabeth Couper-Kuhlen has pointed out to us, this incompatibility with
non-comma intonation brings us back to Sweetser's (1990) original concern
about the preference of since for non-content uses, and raises doubts as to
whether examples like (23) should really be treated alongside other content
13. For discussion of parallels with paratactic conjunction, see Lakoff (1971),
Grice (1978), and Sweetser (1990).

Akatsuka, Noriko
1986 Conditionals are context-bound. In: Traugott et al. (eds.), 333-52.
Chafe, Wallace L.
1984 How people use adverbial clauses. BLS 10: 437-449.
Comrie, Bernard
1986 Conditionals: a typology. In: Traugott et al. (eds.), 77-99.
Dancygier, Barbara
1998 Conditionals and prediction: time, knowledge, and causation in
conditional constructions. Cambridge: Cambridge University Press.
1992 Two metatextual operators: negation and conditionality in English
and Polish. BLS 18: 61-75.
1993 Interpreting conditionals: time, knowledge and causation. Journal of
Pragmatics 19: 403-434.
Dancygier, Barbara and Eve Sweetser
1996 Conditionals, distancing and alternative spaces. In: Adele Goldberg
(ed.). Conceptual structure, discourse and language, 83-98. Stanford
CA: CSLI Publications.
1997 Then in conditionals. Cognitive Linguistics 8/2: 1-28.
140 Barbara Dancygier and Eve Sweetser

Davis, W.
1983 Weak and strong conditionals. Pacific Philosophical Quarterly 64:
Fauconnier, Gilles
1985 Mental spaces. Cambridge MA: MIT Press [Reprinted 1994,
Cambridge: Cambridge University Press].
1996 Analogical counterfactuals. In: Fauconnier and Sweetser (eds.), 57-
1997 Cognitive mappings for language and thought. Cambridge:
Cambridge University Press.
Fauconnier, Gilles and Eve Sweetser (eds.)
1996 Spaces, worlds and grammars. Chicago: University of Chicago
Fillmore, Charles J.
1990a The contribution of linguistics to language understanding. In: Aura
Bocaz (ed.). Proceedings of the first symposium on cognition,
language, and culture, 109-128. Universidad de Chile.
1990b Epistemic stance and grammatical form in English conditional
sentences. CLS 26: 137-162.
Fillmore, Charles J., Paul Kay, and Mary Catherine O'Connor
1988 Regularity and idiomaticity in grammatical constructions: the case of
let alone. Language 63/3: 501-538.
Fillmore, Charles and Paul Kay
1999 Grammatical constructions and linguistic generalizations: the what's
X doing Y? construction. Language 75/1: 1-33.
Fleischman, Suzanne
1989 Temporal distance: a basic linguistic metaphor. Studies in Language
13/1: 1-50.
Ford, Cecilia E.
1993 Grammar in interaction: Adverbial clauses in American English
conversations. Cambridge: Cambridge University Press.
Goldberg, Adele
1995 Constructions. A Construction Grammar approach to argument
structure. Chicago: University of Chicago Press.
Grice, H. P.
1978 Further notes on logic and conversation. In: Peter Cole (ed.), Syntax
and Semantics, Vol. 9: Pragmatics, 113-127. NY: Academic Press.
Haiman, John
1980 The iconicity of grammar: isomorphism and motivation. Language
56: 515-40.
1986 Constraints on form and meaning of the protasis. In: Traugott et al.
(eds.), 215-227.
Hom, Larry
1984 Toward a new taxonomy for pragmatic inference: Q-based and R-
based implicature. In: Deborah SchifIrin (ed.), Meaning, form and
use in context: linguistic applications (GURT 1984), 11-42.
Washington, D.C.: Georgetown University Press.
Constructions with if, since and because 141

1985 Metalinguistic negation and pragmatic ambiguity. Language 61:

Iatridou, Sabine
1991 If then, then what? NELS 22: 211-225.
1994 On the contribution of conditional then. Natural Language
Semantics 2: 171-199.
James, Deborah
1982 Past tense and the hypothetical: a crosslinguistic study. Studies in
Language 6: 375-403.
Konig, Ekkehard
1986 Conditionals, concessive conditionals, and concessives: Areas of
contrast, overlap and neutralization. In: Traugott et ale (eds.), 229-
Lakoff, Robin
1971 Its, and's, and but's about conjunction. In: Charles 1. Fillmore and
D. Terence Langendoen (eds.), Studies in Linguistic Semantics, 114-
149. New York: Holt, Rinehart and Winston.
Langacker, Ronald W.
1987 Foundations ofcognitive grammar, vol. 1: Theoretical prerequisites.
Stanford CA: Stanford University Press.
1991a Foundations ofcognitive grammar, vol. 2: Descriptive applications.
Stanford CA: Stanford University Press.
1991b Concept, image, symbol: the cognitive basis of grammar. Berlin:
Mouton de Gruyter.
Lightbown, P.M. and N. Spada
1993 How languages are learned. Oxford: Oxford University Press.
Nikiforidou, Kiki
1990 Conditional and concessive clauses in Modern Greek: A syntactic
and semantic description. Ph.D. dissertation, University of California
at Berkeley.
Schiffrin, Deborah
1992 Anaphoric then: aspectual, textual, and epistemic meaning.
Linguistics 30: 753-792.
Schwenter, Scott
1997 The pragmatics of conditional marking: Implicature, scalarity, and
exclusivity. Ph.D. dissertation, Stanford University.
1998 From hypothetical to factual and beyond: refutational si- clauses in
Spanish conversation. In: Jean-Pierre Koenig (ed.), Discourse and
cognition: Bridging the gap, 423-436. Stanford CA: CSLI
Sweetser, Eve
1982 Root and epistemic modals: causality in two worlds. BLS 8: 484-507.
1990 From etymology to pragmatics. Cambridge: Cambridge University
1996 Conditionals and mental spaces. In: Fauconnier and Sweetser (eds.),
142 Barbara Dancygier and Eve Sweetser

1999 Compositionality and blending: semantic composition in a

cognitively realistic framework. In: Theo Janssen and Gisela
Redeker (eds.), Cognitive Linguistics: Foundations, scope and
methodology, 129-162. Berlin: Mouton de Gmyter.
Talmy, Leonard
1988 Force dynamics in language and cognition. Cognitive Science 2: 49-
Traugott, Elizabeth Closs
1982 From propositional to textual and expressive meanings. In: Winfred
Lehmann and Yakov Malkiel (eds.), Perspectives on Historical
Linguistics, 245-271. Amsterdam: Benjamins.
Traugott, Elizabeth Closs, Alice ter Meulen, Judy Snitzer Reilly and Charles A.
Ferguson (eds)
1986 On conditionals. Cambridge: Cambridge University Press.
Van der Auwera, Johan
1986 Conditionals and speech acts. In: Traugott et al. (eds.), 197-214.
On affirmative and negative complex conditional

Estrella Montolio

This paper describes the characteristic features of clauses introduced by

complex conditional connectives (CCCs). Specifically, I analyze the
grammatical (§2), semantic (§3, §5, and §6) and pragmatic (§7) properties
of these clauses which distinguish them from ij=-conditionals. CCCs are
organized into two subsets that stand in complementary distribution: (i)
affirmative CCCs and (ii) negative CCCs. These subsets show essentially
the same grammatical behavior but differ in terms of their semantics (§ 1
and §4) and their pragmatics (§7). The data are drawn principally from
Spanish, but the results of the analysis are, in general, equally applicable to
English and French.

1. Preliminaries

Though it is true that the literature on conditionals is quite extensive,

conditional structures other than !fclauses have traditionally attracted
little attention from linguists. This is most definitely the case with what
might be called complex conditional connectives (Dostie 1987; Leard
1987; Montolio 1991b, 1999; Visconti 1994), which have received
anything but extensive treatment. Perhaps this has to do with the general
impression that complex conditional connectives (henceforth, CCCs)
are not only equivalent to one another but are also completely parallel to
!fclauses. 1 The goal of this paper is to refute such an idea and
demonstrate how mistaken it actually is. Contrary to the view of
equivalence, I shall show that CCCs cannot appear in the majority of the
contexts in which it is possible to use !fclauses. As we shall see, this is
because CCCs express a much more specific and restricted relationship
between the protasis and the apodosis than does if. I concentrate mostly
on Spanish data, but I believe that, for the most part, my observations
are equally valid for both Romance and Germanic Languages. 2
The connectives that we shall include in the category of CCCs
correspond to two complementary subsets (cf Montolio 1990, 1991b,
144 Estrella Montolio

1999). The first group, which we shall call the affirmative group (or
ACCCs), is made up of constructions such as:

(1) a. a condici6n de que (= French: acondition que)

'provided that', 'on the condition that', 'so long as'
b. con tal de que (= French: pourvu que)
'provided that', 'so long as'
c. siempre que 3 (= French: pourvu que)
'provided that', 'so long as'

Expressions such as the following are included in the second, negative

group (or NCCCs).

(2) a. a menos que (= French: amoins que)

b. a no ser que (= French: amoins que)
c. salvo que (= French: saufsi)
'except if
d. excepto que (= French: saufsi)
'except if

Spanish conditional connectives excepto si ('except if) and salvo si

('except if) are semantically analogous to the other connectives
included in this group (2), but, as if-compounds, they can combine with
either the subjunctive or the indicative. Given the fact that the rest of the
complex connectives to be analyzed here all trigger the subjunctive
mood, 1 have preferred not to pursue their description any further here.
Because these two subsets stand in complementary distribution, any
member from the ACCC group can appear in the same contexts as all
the others of this group but in none of the contexts required by NCCCs.
The complementary distribution of these subsets is due to a difference in
semantic content, which explains the pragmatic inadequacy of these
connectives when an ACCC is replaced by an NCCC (or vice versa)
within the same context. This can be seen in examples (3b) and (4b)
below. (I use the #sign to indicate that the utterance under consideration
is grammatically correct but pragmatically inadequate).
Complex conditional connectives 145

(3) a. Ire con vosotros al cine a condici6n de que/con tal de

que/siempre que me acompaneis luego a casa en coche.
'I'll go with you guys to the movies provided that/so long as
you take me home afterwards by car. '
b. # Ire con vosotros al cine a no ser que/a menos que me
acompaneis luego a casa.
'# I'll go with you guys to the movies unless you take me
home afterwards by car. '

(4) a. No ire con vosotros al cine a no ser que/a menos que me

acompaneis luego a casa en coche.
'I won't go with you guys to the movies unless you take me
home afterwards by car. '
b. # No ire con vosotros al cine a condici6n de que/con tal de
que/siempre que me acompaneis luego a casa en coche.
'# I won't go with you guys to the movies provided that/so
long as you take me home afterwards by car. '

The utterances in question become acceptable again if their polarity

is changed (see example (4a». Hence, we can say that the
complementary distribution of these two subsets (ACCCs and NCCCs)
stems from their complementary polarities. Notice that it is not a matter
of whether utterances with NCCCs should necessarily be negative (as
shown by a simple example such as Manana ire a hacer jogging a no
ser que llueva 'Tomorrow I'll go jogging unless it rains'). Rather, the
point is that its potential commutation by an ACCC always entails a
change in polarity (Manana ire a hacer jogging a condicion de que no
llueva 'Tomorrow I'll go jogging provided that it doesn't rain').
As we shall see below, the reason for the strangeness of the (b)
sentences is not, strictly speaking, a matter of ungrammaticality, i.e., a
problem in the syntax of the constructions. Instead, it probably has more
to do with a pragmatic inadequacy between the informational content of
the two clauses and the processing instructions conveyed by the
connectives in question. Despite these distributional constraints, it is
easily shown that si ('if) and its corresponding negative counterpart si
no ('if ... not') can appear in the ACCC and NCCC contexts,
146 Estrella Montolio

(5) a. Ire con vosotros al cine si me acompafiais luego a casa en

'I'll go with you to the movies if you take me home afterwards
by car.'
b. No ire con vosotros al cine si no me acompafiais luego a casa
en coche.
'I won't go with you to the movies if you won't take me home
afterwards by car.'

The distribution of the data leads me to believe that si 'if is the

prototypical and superordinate member of the category of conditional
structures. The connective si is semantically unmarked as compared to
CCCs (as well as other possible conditional constructions), which
express a more specific and restrictive type of condition. The import of
this observation is that CCCs are less abstract and more specific in
meaning than if and its negative counterpart if ... not. That is to say, if
and if ... not constitute the superordinate terms of the relationship, in
which CCCs are the subordinate terms.

2. The order and the value of p in CCCs

Although research on clause order in conditional structures has not

specifically examined CCC constructions, the data they provide seem to
show that clauses introduced by CCCs, unlike if-clauses, usually occupy
the second position in conditional constructions (for English, see
Traugott, 1997: 160 and Dancygier, 1998: 173; for French, cfr. Dostie,
1987: 177-178; see Montolio 1991b: 134-135, for Spanish data).
Therefore, in a high percentage of the cases, CCCs invert the normal or
unmarked order of conditional expressions, which of course tends
prototypically to be [if p, q]. CCC-ciause order, then, has to be related to
the schema [q ifp].4
Ramsey (1985: 405) observes that while preposed protases tend to
have discourse functions, postposed protases "look more like a
parenthetical comment for the main clause". The corpus-based studies
by Ford and Thompson (Ford and Thompson 1986, Ford 1993, 1997)
have corroborated this observation, namely that when the protasis is
preposed it either conveys information that is in some sense already
known or creates the framework for the following discourse. However,
Complex conditional connectives 147

when !fclauses are postposed, they function more like afterthoughts,

evaluations, justifications.
Prosodic features also play a role. When the postposed clause is
preceded by a pause [q, ifp], on occasion at least, the !fclause seems to
convey the sense of less probability. The expectation that the contents in
p will not be fulfilled is often taken advantage of in advertising slogans:

(6) a. Si puede mirela. 0 cambie, si puede. (Catchphrase for the

Spanish TV series Serie Rosa 'Pink Series') (EI Peri6dico de
Catalunya, March 21, 1991).
'If you can, watch it. Or change [the channel], if you can.'
b. Resistete, si puedes. (Advertising slogan for the Renault 19)
(EI Pais, September 15, 1992).
'Resist [the temptation], if you can.'

In each case the effect of postposition in conjunction with the

preceding pause clearly conveys an inferential interpretation that the
fulfillment of the modal puede(s) 'can' is improbable. In fact, in the
television commercial corresponding to the same Renault advertising
campaign as that in (6b) even a priest succumbs to the temptation. The
interpretation that 'p will not be easily fulfilled' is used as sort of a
challenge, a strategy to incite the consumer's interest.
I would like to suggest that the CCC-schema conveys an analogous
sense of improbability or exceptionality. The characteristic semantics of
postposed protases headed by CCCs represents an a-posteriori
specification that rectifies 5 the content of the main clause, by stipulating
the conditions under which q will or will not be carried out. In these
constructions the hearer is instructed that p entails a certain type of
correction of what is affirmed in q. As pointed out by Dancygier (1985,
1998) for unless constructions, we can propose that the relation between
q and p in CCC structures is not one of cause (as is prototypically the
case of conditionals), but rather one of circumstance. That is, p
constitutes a specification of the circumstances under which q can be
maintained. This can be seen in the following examples:

(7) a. A mi me parece que las especulaciones son muy poco

pertinentes, a no ser que seamos capaces de imaginarnos a
nosotros mismos convertidos en un anciano de setenta y siete
afios [ ..}. (El Europeo, October 1988, pp. 59-60).
148 Estrella Montolio

'It seems to me that the speculations are rather inappropriate,

unless we are able to imagine ourselves turned into a 77-year-
old elderly person [... ].
b. La respuesta a tan dificil pregunta es sf, es posible broncearse,
siempre que la exposicion solar se realice de manera
inteligente. (Marie-Claire, February, 1993).
'The answer to such a difficult question is yes, it is possible to
get a tan, provided that exposure to the sun is done in an
intelligent manner.'

The specification implied by p is so strong that, orally, it is often

introduced by a long pause or, graphically, ~y a comma or even a
period. In many cases, CCC-ciauses constitute phonologically and
graphically independent utterances. The schema [ q. CCC p] is quite
frequent in Spanish. It can be observed in the following example:

(8) Sf, sf, vamos. Estoy convencido de que en mi vida futura la

filosofia sera mi hobby. Ahora, no mi profesion, desde luego. A
no ser que cambie mucho, 0 me vea en una situacion... economica
muy ahogada y me tenga que dedicar a dar clases. (Esgueva and
Cantarero (1981: 24).
'Yes, yes, well.... I'm convinced that in my life to come
philosophy will be my hobby. But not my job, that's for sure.
Unless my life changes a lot, or I find myself in a very tight
financial. .. situation and I have to earn a living teaching. '

As opposed to prototypical If-clauses, which are preposed, the order

[q, CCC p] is not iconic with the sequence of events in the 'real world'.
Instead, the order appears to be iconic with the mental processing of the
instructions to be given to the hearer. That is, CCC-ciauses tend not be
iconic with event structure, but rather with argumentational patterns, in
that their cognitive processing seems to mirror the sequence used in
assertion-restriction or the specification thereof

3. The biconditional value of CCCs

The hypothesis put forth here is the following: whereas the construction
if p designates a very broad and generalized condition, roughly
paraphrasable as 'in the event that p,' CCCs express highly specific
Complex conditional connectives 149

conditions. As noted above, CCCs establish a much more precise and

restricted relationship between p and q than if does. It is true that, as it
has been seen, !fclauses can convey a similar "restricted" relation
betwen p and q in certain grammatical circumstances (for instance, in
cases in which the subordinate clause is postposed and preceded by a
pause). The difference lies in the fact that if only suggests such a strong
contingency in the appropiate context, whereas CCCs make the
"restricted" relation explicit, which is consistent with the greater
transparency of CCCs.
The relationship between p and q in a CCC-structure is so strong
that, as we shall see, it cannot be canceled. In fact, the exceptionality of
the condition expressed by CCCs is such that it cannot be later
neutralized without leading to a semantically and pragmatically
unacceptable utterance (cf. Dostie 1987: 190).
Ducrot (1973: 267) and Fauconnier (1984: 147) have both pointed
out that there exist mechanisms that nullify the biconditional reading of
!fclauses, which is the normal interpretation in ordinary language
(known as conditional perfection). Fauconnier proposes that de toute
far;on ('anyway') can be employed to nullify the 'invited' inference of a
conditional reading (ifp, q , if not p, not q) (Geis and Zwicky 1971).
The neutralizing character of de toute far;on is confirmed by
"l'apparition simultanee possible de mais ou cependant" (Fauconnier
1984: 147).

(9) a. Si tu viens,} 'irai, mais si tu ne viens pas,} 'irai de toute fa~on.

b. Si til vas, ire, pero si no vas, yo ire de todos modos. 6
c. Ifyou go, I'll go, but if you don't, I'll go anyway.

However, as can be seen in (10), such expressions cannot be used

with CCC-ciauses.

(10) a. *J'irai a condition que tu viennes, mais si tu ne viennes pas,

} 'irai de toute fa~on.
b. *Ire a condici6n de que vayas, pero si til no vas, ire de todos
c. *1'11 go provided that you go, but if you don't, I'll go anyway.

These examples corroborate the hypothesis that the condition

established by CCC-ciauses is in no way defeasible. This behavior is
similar to that of the connective if and only if. A condicion de que
150 Estrella Montolio

'provided that' and a menos que/a no ser que 'unless' convey unique or
exceptional circumstances that will allow (or block) the event expressed
in the main clause to occur. This explains why such CCC-constructions
cannot be iterated in coordinate structures:

(11) a. *,Sa/dre contigo a cenar a condicion de que no hab/emos de

trabajo y a condici6n de que me invites.
'I'll go out to dinner with you so long as we don't speak about
work and so long as you pick up the tab. '
b. *Me ire de esta empresa a no ser que me asciendan y a no ser
que me aumenten e/ sue/do.
'I'll leave this company unless they promote me and unless
they give me a raise. '

Notice that both constructions are perfectly viable with si:

(12) a. Sa/dre contigo a cenar si no hablamos de trabajo y si me

'I'll go out to dinner with you if we don't speak about work
and if you pick up the tab. '
b. Me ire de esta empresa si no me ascienden y si no me
aumentan e/ sue/do.
'I'll leave this company if they don't promote me and if they
don't give me a raise. '

The reason for the unacceptability of the iterated CCCs lies in the fact
that it is simply not possible to present the only circumstance under
which something will or will not happen and then add another
exceptional condition. In short, CCCs disallow the presence of other
exceptional conditions.

4. ACCCs and NCCCs

Both groups (ACCC and NCCC) express exceptional conditions. The

meaning of a condici6n de que 'provided that' and other ACCCs in
general is restrictive: they establish the only condition under which q
can be realized. That is, ACCCs have approximately the same value as
[q only ifp]. ACCCs are therefore formulations of a favorable exception
for something to happen. NCCCs, on the other hand, indicate the only
Complex conditional connectives 151

situation by which q will not be realized. 7 That is, NCCCs convey the
instruction [q only if not pl. Hence, NCCCs convey something to the
effect of "in any situation other than p." In other words, NCCCs express
the exceptional circumstance by which the proposition expressed by the
main clause will not be fulfilled. Both types of CCCs express tff
relations. Both types of CCCs present a modal and argumentative value,
since they instruct the hearer whether a given circumstance is favorable
or disfavorable to the proposition in the main clause.
The perspective that the speaker adopts in the face of the possibility
of q 's realization is therefore quite different: while ACCCs signal that q
can only be realized if p, NCCCs express that q will not be realized only
in the case of p. This is what determines the affirmative character of the
first group and the negative character of the second. ACCCs convey the
necessary basis for something to occur; NCCCs express the
circumstance under which something will not be realized. Dancygier
(1985: 68) proposes quite insightfully that what unless really denies is q,
not p. Her proposal seems to be clearer than the one advanced by Geis
(1973): except if p, q. To observe a case in which the main clause
displays negative polarity (for instance, example (4)), we can
extrapolate from Dancygier's observation as follows in (13b):

(13) a. q unlessp q; [(not q) ifp] (1985: 65)

(b. not q unlessp notq; [ (q) ifp])

The negative character of NCCCs is the reason why a no ser que

'unless' and like connectives appear to be generally the equivalent of si
no 'if ... not' -clauses. In fact, however, there are quite a few differences
between the· two types of constructions (cf Geis 1973; Dancygier 1985;
von Fintel 1991; Dancygier 1998: 169-171). Consider, for instance, the
odd combination of NCCCs such as unless with negative-polarity
clauses which are perfectly acceptable with if (Cf *Unless .John cares a
bit for Mary, he shouldn't marry her vs. If John doesn't care a bit for
Mary, he shouldn't marry her (Geis 1973:283)).
In my opinion, regardless of whether they are equivalent to if or if ...
not clauses, the fundamental value of NCCCs is to specify in p the
circumstances under which q will not be fulfilled. The point is that the
circumstances present in p are construed as improbable or contrary to
what one might expect. The contrast between the two following
examples adds credibility to the hypothesis advanced here (cf Smith
1983: 21 for similar examples):
152 Estrella Montolio

(14) a. Los cocodrilos no podrian sobrevivir si no tuvieran dientes

'Crocodiles couldn't survive if they didn't have extremely
sharp teeth. '
b. # Los cocodrilos no podrian sobrevivir a menos que tuvieran
dientes afiladisimos.
'Crocodiles couldn't survive unless they had extremely sharp
teeth. '

(14a) sounds perfectly normal but (14b) is surprising from a pragmatic

point of view. In (14a), the negation of the counterfactual structure
presupposes that crocodiles actually do have extremely sharp teeth. On
the other hand, (14b) is decidedly odd because the clause introduced by
the NCCC presupposes that the condition in p is exceptional. That is,
the condition is contrary to what would normally be expected. Since
NCCCs convey an exceptional obstacle for q, in an eternal-truth context
such as (14b) they are obviously inappropriate: the clause suggests that
crocodiles do not have extremely sharp teeth, and this conflicts whith
our encyclopedic knowledge.
This nuance of less expectation conveyed by unless-constructions
explains why they are often inappropriate for counterfactuals. As is well
known, counterfactual clauses are presupposed to be false:

(15) Yo Ie habria dado ese medicamento a la nina, si no 10 hubiera

desaconsejado el medico explicitamente.
'I would have given the medicine to the girl, if the doctor hadn't
advised against it explicitly. '

The utterance in (15) conveys a factual interpretation in which the

content of p is negated. The doctor is understood as indeed having
advised against giving the girl the medicine. Clauses headed by a menos
que 'unless' and a no ser que 'unless' convey information which is
necessarily understood as hypothetical, precisely because they carry a
sense of high improbability-and improbability is necessarily
hypothetical. Thus, a version of (15) with a menos que or a no ser que
indicates that the hypotheticality (i.e., nonfactuality) of p is real, since in
reality we do not know if the doctor did or did not authorize the
Complex conditional connectives 153

(16) Yo Ie habria dado ese medicamento a la nina, a menos que/a no

ser que 10 hubiera desaconsejado el medico explicitamente.
'I would have given the medicine to the girl, unless the doctor had
advised against it explicitly. '

Precisely because they can only receive a hypothetical interpretation

and are primarily future-oriented (see Traugott 1997: 158 for the case of
unless), a menos que/a no ser que-clauses are incompatible with
counterfactuals that presuppose the creation of an 'impossible world.'
This is also shown by the behavior of the following examples.

(17) a. Si Colon no hubiera descubierto America, los indios habrian

sido mas felices.
'If Columbus hadn't discovered America, the Indians would
have been happier. '
b. *A no ser que Colon hubiera descubierto America, los indios
habrian sido mas felices.
*' Unless Columbus had discovered America, the Indians
would have been happier. '

As we have seen, a no ser que presents q as well as the possibility of not

q. Nevertheless, counterfactuals only allow for one interpretation: q, if
the clause is in negative form, or not q, if there are no overt markers of
negation. Hence, (17b) is odd because we know that Columbus did
actually discover America.

5. CCCs and contextually given conditions

As we have just seen, the greater specificity of CCCs-in contrast to the

neutral, unmarked character of si 'if-makes them inappropriate with
counterfactual structures. This same reason, I contend, stands behind
their inability to appear in constructions in which the content of p is
contextually asserted. Take for example sentence (18):

(18) A: Ya he acabado el trabajo.

'I've just finished the job.'
154 Estrella Montolio

a. B: Pues si ya has aeabado e/ trabajo, sa/gamos a eenar por

'Well, if you've just finished the job, let's go out for
dinner. '
b. 'B: *Pues a condici6n de que ya hayas aeabado e/ trabajo,
sa/gamos a eenar por ahi.
*'Provided that/so long as you've finished the job, let's go
out to dinner. '

In other cases, the assertion hidden in the protasis is less explicit, but it
is information that is present in the context of the speech event (the term
context should be understood in the broad sense of Sperber and Wilson
19962) . The thematic or given nature of such rrotases is demonstrated
by the fact that the protasis cannot be postposed:

(19) a. Si hay una pasion franeesa que no //eva camino de

desapareeer, esa es /a pasion por /a po/itiea. [E/ Pais, June 21,
'If there is one French passion that is not on its way to
disappearing, it is their passion for politics. '
b. *Es /a pasion por /a po/itiea, si hay una pasion franeesa que
no //eva camino de desapareeer.
'It's is a passion for politics, if there is one French passion that
is not on its way to disappearing. '
c. *A condici6n de que haya una pasion franeesa que no //eva
camino de desapareeer, esa es /a pasion por /a po/itiea.
'Provided that there is one French passion that is not on its
way to disappearing, it is their passion for politics. '

(20) a. Si no te /0 ha dicho es porque te quiere.

'Ifhe hasn't told you it's because he loves you.'
b. *Es porque te quiere, si no te /0 ha dicho.
'It's because he loves you, ifhe hasn't told you.'
c. *A no ser que te /0 haya dicho es porque te quiere.
*' Unless he's told you, it's because he loves you.'
Elsewhere (Montolio 1999) I have called such conditional structures
identificatives (19) and exp/ieatives (20), respectively. Unlike
prototypical (i.e., hypothetical) conditionals, the If-clauses here are far
from left-open (in the sense of Lehman 1974: 234; that is, the
Complex conditional connectives 155

proposition in the antecedent of a conditional sentence is not asserted).

On the contrary, these uses of si-clauses seem to be what I propose to
call left-given. This is shown by the fact that these clauses cannot be
postposed, as is demonstrated by the (b) examples, nor can they be used
with CCC-structures, as in the (c) examples.
In Spanish, as well as in French and English, the apodoses of if-
constructions can be-indeed frequently are-yes/no-questions «21a))
as well as wh-questions «22a) and (23a)).

(21) a. Sifuese obispo, ~condenaria el aborto?

'If you were a bishop, would you condemn abortion?'
b. *A condici6n de quefuese obispo, ~condenaria el aborto?
*'Provided that you were a bishop, would you condemn

(22) a. Si a Gonzalez Ie gusta tanto participar en la campana de un

referendum, ,",por que no organiza uno en su pais? dice
Philippe Seguin. [El Pais, June 21, 1992]
'If Gonzalez likes participating in the campaign for a
referendum so much, why doesn't he organize one in his
country? says Philippe Seguin. '
b. *A condici6n de que a Gonzalez Ie guste tanto participar en la
campana de un referendum, ,",por que no organiza uno en su
pais? dice Philippe Seguin.
*' Provided that Gonzalez likes participating in the campaign
for a referendum so much, why doesn't he organize one in his
country? says Philippe Seguin. '

(23) a. Si no estas preparada para tener un hijo, ,",por que demonios

has querido quedarte embarazada.
'If you're not ready to have a child, why the hell did you want
to get pregnant?'
b. *A menos que estes preparada para tener un hijo, ,",por que
demonios has querido quedarte embarazada.
*' Unless you're ready to have a child, why the hell did you
want to get pregnant?'

In view of examples such as these, it seems feasible to state that ACCCs

are difficult to combine with yes/no-questions, as is the case with
example (21b). Such a combination gives rise to utterances that are
156 Estrella Montolio

clearly very strange. Furthermore, CCCs are, in general, difficult to

interpret with wh-interrogatives (e.g., (22b) and (23b». Actually, this
difference in behavior with regard to the two different basic types of
interrogatives does not come as much of a surprise. This is so because in
wh-interrogatives the tt:protasis functions as a framework for a latent
assertion (e.g.,(22a) and (23», whose superficial hypothetical form lays
the groundwork for and justifies the question that follows it. That is, the
use of a hypothetical form for contextually asserted information
constitutes a mechanism that attempts to safeguard the relevance of the
subsequent question (cf Haegeman 1984).
CCCs are also incompatible with so-called indicative
counterfactuals, since they also presuppose some kind of previous

(24) a. Mira, si esa chica es guapa, yo soy Claudia Schiffer.

'Look, if that girl is pretty, I'm Claudia Schiffer.'
b. *A condici6n de que esa chica sea guapa, yo soy Claudia
*'Provided that that girl is pretty, I'm Claudia Schiffer.'
(25) a. Si tit no eres rico, yo entonces soy un mendigo.
'Ifyou're not rich, then I'm a beggar. '
b. *A no ser que tit seas rico, yo soy un mendigo.
*' Unless you're rich, I'm a beggar.'

All of these kinds of conditionals fall under the heading of

nonpredictive conditionals (according to Sweetser 1990; Dancygier
1993, 1999). Note that, to a greater or lesser extent, they all constitute
"echoes" from some previous discourse. Furthermore, they can be
characterized by their lack of strict iconicity. The sequence of clauses is
not iconic with the sequence of events, though in fact there does seem to
be an iconic correspondence between the linear order of the clauses and
the sequence of steps taken in argumentation: Given x: X allows me to
assert ... (cft. Sweetser 1990: 128). That is, speakers use p to show the
hearer how they arrive at the assumptions in q. As we can see, CCC-
structures cannot appear in any context of epistemic conditionals, where
the tt:clause introduces some type of assumption present in the context
of the speech situation (Sweetser 1990: 128). That is, CCC-ciauses
cannot appear in any context whose content is given, assumed, or
clearly presupposed. The information that CCC-ciauses convey is
Complex conditional connectives 157

therefore necessarily rhematic. This is consistent with post-main-clause

order, because focus tends cross-linguistically to occur in sentence-final
position (Comrie 1986: 86).
The insertion of a CCC, especially NCCC, seems slightly more
acceptable--especially when they appear preposed-with speech-act
and metalinguistic conditionals than with any of the former types.
Nevertheless, such uses do not, by any means, constitute prototypical
contexts for CCCs. Let us consider Austin's celebrated example:

(26) a. There are biscuits on the sideboard ifyou want them.

b. #There are biscuits on the sideboard provided that/so long as
you want them.

Clearly there is something peculiar about (26b). Consider the examples

in (27):

(27) a. En espanol americano a eso Ie llaman 'durazno', SI no

confundo la palabra.
'In Latin-American Spanish that's what they call a 'durazno'
(peach), if I'm not confusing the word.'
b. ?En espanol americana a eso Ie Ilaman 'durazno', a no ser que
confunda la palabra. 9
'In Latin-American Spanish that's what they call a 'durazno'
(peach), unless I'm confusing the word.'

The ability of NNNC clauses to be inserted in these nonpredictive,

nonprototypical schemas can be explained by the parenthetical nature
that both !fclauses and CCC-ciauses acquire in these types of
conditionals: the independence of the propositional content of CCC-
clauses with regard to the main clause also corresponds to ·the
nonembedded character of the subordinate clauses in speech-act and
metalinguistic conditionals.

6. CCCs and the discourse deictic then

Another feature that characterizes CCC-constructions as noncentral

members of the family of conditional structures is their inability to
combine with the phoric then: 10
158 Estrella Montolio

(28) a. Si hace sol, entonces iremos a la playa.

'/fit's sunny, then we'll go to the beach.'
b. *A condici6n de que/con tal de que/ siempre que haga sol,
entonces iremos a la playa.
'Provided that/so long as it's sunny, then we'll go to the

(29) a. Si no llueve, entonces iremos a la playa.

'/fit doesn't rain, then we'll go to the beach.'
*A no ser que/ a menos que/ salvo que/ excepto que llueva,
entonces iremos a la playa.
*' Unless/except ifit rains, then we'll go to the beach.'
The presence of entonces ('then') in conditional schemas is a matter that
remains to be thoroughly studied in Spanish. Nonetheless, as pointed
out by Haiman (1978: 576), in contradistinction to the traditionally
studied schema if p, then q, the phoric then does not constitute by any
means an obligatory element in ordinary-language conditionals. In fact,
certain types, such as speech-act conditionals or metalinguistic
conditionals, reject it.

(30) a. ?? Si no me equivoco, entonces ese es tu primo.

??'/fI' m not mistaken, then that guy is your cousin. '
b. ?? Si se Ie puede llamar asi, entonces es una estructura
??' /f you can call it that, then it's a conditional structure. '

Of course, speech-act and metalinguistic conditionals, by their very

nature, do not express a grammatical relationship where p entails q. By
contrast, entonces 'then' explicitly expresses that q follows from p.
Therefore, the semantics of these constructions clashes with that of then
and thus produces an unacceptable utterance.
In those constructions that do accept its presence, the insertion of
then is neither pragmatically, nor semantically identical to its omission:
it is one thing to imply that q follows from p, as in prototypical
conditionals using the schema [if p, q], and it is another to mark this
explicitly, as in [ifp, then q]. In any event, it should be noted that those
if-constructions that do allow the presence of then do not allow it when
the normal order of the clauses is altered, which happens to be the
schema normally used by CCCs:
Complex conditional connectives 159

(31) *Entonces ire a la playa, si llueve.

'Then I'll go to the beach, ifit rains.'

Nevertheless, the examples in (28) and (29) show that CCCs are
incompatible with entonces 'then' even when they appear in a preposed
position. Such a fact invites the question of just what the function of
then in a conditional is. There seems to be agreement that one of its
main functions is to act as an anaphoric emphatic correlate to the
protasis, i.e., something like a resumptive pronoun (Dancygier 1998;
Dancygier and Sweetser 1997; latridou 1992; Montolio 1991b, 1999).
One might say that the use of then constitutes an attempt on the part of
the speaker to insure that the interlocutor retains the hypothetical
framework on which the following clause or utterance is based. This is
clearly the case of conditionals in which various protases are linked to
one apodosis:

(32) Todos creemos tener el mismo derecho: si uno se puede hacer

millonario en dos dias, si no es el resultado de una vida de
esfuerzo, si se consigue con solo jugar a la Bolsa, con disponer de
informacion privilegiada, entonces todo el mundo cree que
tiene derecho a ganar mas de 10 que gana. [EI Pais, May 17,
'We all believe that we have the same right: if one can become a
millionaire in two days, if it's not the result of a life of hard work,
if it can be obtained by playing the stock market, by having
privileged information, then everybody believes that they have the
right to earn more than they do.'

As opposed to prototypical ~clauses, CCC-ciauses do not function

as a hypothetical framework for the assertion of q. Rather, as we have
seen, they restrict or specify what is expressed in q. They are not,
however, the basis for asserting the content of the main clause. These
facts account for why CCCs cannot collocate with then, since the
cohesive function of this element is to serve as an elliptical framework
for the apodosis. However, when the CCC-structure is included in a
larger framework, entonces 'then' is indeed acceptable.

(33) A: Papa, iftle Ilevaras al cine?

'Dad, will you take me to the movies?'
160 Estrella Montolio

B: J.~i,
a condici6n de que acabes los deberes. Entonces te llevare
al cine.
'Provided that/so long as you finish your homework. Then I'll
take you to the movies. '

As can be seen, in (33) the insertion of entonces 'then' sounds perfectly

normal because its presence is not linked solely to the information in the
CCC-ciause. If this were the case, it would have given rise to an
unacceptable utterance:

(34) *A condici6n de que acabes los deberes, entonces te llevare al

*'Provided that/so long as you fmish your homework, then I'll
take you to the movies. '

Instead, in (33) B's answer, entonces ('then') is linked to a larger

discourse context, to the whole previous sentence ("yes, I will take you
to the movies, provided that you fmish your homework. Then --'if this
is the case' --, I'll take you").

7. The pragmatics of CCCs

In various papers, Fillenbaum has analyzed the relationship that exists

between a series of conditional constructions and certain speech acts
(see, specially, Fillenbaum 1986). He points out that the semantic
content of unless makes it appropriate to appear in deterrents, while at
the same time making it unnatural-sounding in inducements. This
semantico-pragmatic relationship between unless-constructions and the
illocutionary value of threats!! and prohibitions has been documented
diachronically for English (Traugott 1997) as well as for Spanish. Bartol
Hernandez (1992: 89) briefly comments that, historically, in the
majority of the cases, they are prohibitions (or negative orders,
expressed with can, should, or the subjunctive) whose compliance
carries an exception, a restriction, presented as a necessary condition.
(35) is an example given by this author:

(35) E si por aventura el Rey fuere de tan gran piedat que 10 quiere
dexar vivir, non 10 pueda facer a menos que no Ie saque los ojos.
(Fuero Real, 350)
Complex conditional connectives 161

'And if by chance the King were so kindhearted as to want to let

him live, he cannot do so unless his eyes are tom out. (Regional
Code, 350)'

Fillenbaum has shown that in a high percentage of cases provided that

appears in promises or inducements (e.g., Fillenbaum 1976: 148). The
fact is that it is very difficult to invert these roles; that is to say, the
formulation of a promise with unless or that of a threat with provided
that is highly infrequent (being generally inadequate). This is true even
if we change the polarity of the clauses, in accordance with the
complementary distribution of the CCCs that we have analyzed. As can
be seen, (36b) is no longer a deterrent (as (36a» but a promise not to do
something undesirable. In contrast, (36d) seems a threat more than an
inducement (in contradistinction to (36c»:

(36) a. Te dejare todo el verano sin vacaciones, a no ser que mejores

estas notas.
'I'll take away your vacation, unless you improve these
grades. '
b. # No te dejare todo el verano sin vacaciones, a condici6n de
que mejores estas notas.
#'1 won't take away your vacation, provided that you improve
these grades. '
c. Iremos de viaje a Disneylandia, a condici6n de que mejores
estas notas.
'We'll go to Disneyland, provided that you improve these
d. # No iremos de viaje a Disneylandia, a no ser que mejores
estas notas.
#'We won't go to Disneyland, unless you improve these

The favoring character of positive stimuli presented by provided that

is corroborated by the fact that it can paraphrase-with the right
modifications in the clauses-any paratactic conditional construction
with the conjunction and, but only in the event that it expresses an
inducement stimulus, such as:

(37) a. Dame un besito y te dare un caramelo.

'Give me a little kiss and I'll give you a piece of candy.'
162 Estrella Montolio

b. Te dare un earamelo, a condici6n de que me des un besito.

'I'll give you a piece of candy, provided that you give me a
kiss. '

However, when these coordinate structures are the equivalent of a threat

the paraphrase is impossible. Compare the following sentences:

(38) a. Toea a mi hermano y te parto la eara.

'Touch my brother and I'll bash your face in. '
b. #Te parto la eara, a condici6n de que toques a mi hermano.
#'1'11 bash in your face, provided that you touch my brother. '

The illocutionary specialization of these connectives is derived from

their semantics. The lessening of expectation or high improbability
conveyed by a menos que ('unless') conveniently explains its
inadequacy as a stimulus: it does not seem pragmatically profitable to
try to act upon one's interlocutor under the pretext that something
probably WILL NOT happen.
This same reason explains the adequacy of unless for prohibitions
and threats. The coercive power of a threat lies in the fact that the
negative event for the hearer is presented as a general assertion, which
only has a distant possibility of being revised: X, unless Y. The hearer
must therefore either comply with the proposition in the unless-clause or
suffer the consequences. 12
To exemplify this, we can recreate a cliche-ridden dialogue from the
film-noir genre between a gangster type and "the other guy's girl." In
the improbable event that the gangster were to use a CCC-construction
instead of the more predictable paratactic structure, he could choose
between formulating his threat with provided that or unless.
Nevertheless, one of them would seem much more persuasive than the

(39) a. No matare a tu novio, a condici6n de que te fugues eonmigo.

'I won't kill your boyfriend, provided that you run away with
b. Matare a tu novio, a no ser que te fugues eonmigo.
'I'll kill your boyfriend, unless you run away with me.'

The greater force of (39b) lies precisely in that the threatening action
(i.e., kill the boyfriend) is expressed in an affirmative factual manner.
Complex conditional connectives 163

This seems to be more effective than posing the matter in the converse
terms, as in (39a), where the threatening action is negated and the
restriction on this negated threat appears later.
The same pragmatic explanation can be applied to resolve the greater
or lesser inadequacy of utterances (36b) and (36d). As can be seen,
exchanging connectives causes the speech acts to be exchanged too.
Thus, (3 6b) seems to be a much weaker version of a threat than that
formulated in (36a), though it also could be interpreted as an
unconvincing stimulus. Insofar as the clear stimulating promise in (36c)
is concerned, it is converted directly into a threat as soon as its
connective is changed (36d).
Besides the illocutionary value generally equated with CCC-
structures, we can also pose the question of what their pragmatic
behavior is as compared to that of if-clauses. Also worth considering is
the question of whether there exist differences between the two subsets
of CCCs from such a point of view. An if-clause can appear in any
context that a CCC can, and in fact, speakers prefer to use the unmarked
connector if even when the insertion of a CCC is possible. The very
subtle differences that these constructions convey with regard to if (and
if ... not) are so faint that most speakers probably do not consider them
to be very operative. Furthermore, in the Romance languages, CCC-
clauses trigger the subjunctive mood, which according to the literature
on the matter is in the decline. This seems to be why CCC-ciauses are
more characteristic of formal, written registers than of spontaneous, oral
CCC-structures are, in effect, noticeably less frequent than if-clauses
in oral and spontaneous registers. In this sense, we can see the
correspondence between ontogenesis and linguistic universals, because
such constructions are acquired much later than if-structures. Although
most authorities on the matter of the acquisition of conditionals only
deal with if-clauses, Wing and Kofsky (1981) point out that just as
common sense would lead us to believe, CCC-structures are among the
latest to be acquired. Given their formal complexity and the specificity
of the condition they introduce, CCC-structures surface in child
language at a later stage than if-clauses and other types of synthetic
conditional structures, such as paratactic structures (McCabe et al. 1983,
Wing and Kofsky 1981). In the case of NCCCs, this late appearance
seems to be even more justified. NCCCs imply a greater cognitive-
processing complexity in that to the uncertainty of the truth value they
add a negative relationship between the two clauses.
164 Estrella Montolio

My claim is that no matter what the speech act may be, ACCCs and
NCCCs are pragmatically very different from one another. When
speakers choose to use an unless-clause, they are often trying to avoid
committing themselves completely: q is said first, and afterwards, as a
kind of afterthought, the only situation in which their utterance will not
be true is signalled (P). This claim is consistent with the historical
corpus-based analysis of Traugott (1997), who concludes that in general
unless-clauses express 'possible events that will block promises, threats,
expectations, and even probable recurrence. '

(40) Creo que [el condicionamiento social y economico] impiden que

en la practica se pueda decir: yo me caso cuando quiera. A no ser
que la ... no se, tambien la situacion economica familiar del novio
y de la novia, su plan, claro (Esgueva and Cantarero 1981: 34).
'I think that [social and economic conditioning factors] keep one
from saying in practice: I'm going to get married whenever I
want. Unless the ... I don't know, also the boy's and the girl's
family financial situation, their plans, of course. '

Thus, in most cases, the use of unless can be interpreted as aface-saving

repair (cf Goffman 1967). In such cases the speaker tries to take
precautions against an interpretation of his utterance in terms of overall
truth, which could jeopardize the speakers "face" in the future. In this
sense, the use of unless is much like metadiscursive if=-clauses of the
type if I'm not mistaken, if that's the right word, if I remember
correctly, etc., and has to do with the strategy of guarding oneself from
possible criticisms from one's interlocutor.
On the other hand, ACCCs are more problematic from an interactive
perspective: once established, provided that cannot be mitigated. Such a
specific meaning explains-at least in part-its low frequency: speakers
prefer to propose a condition with an if=-clause, rather than impose a
condition with provided that. In fact, according to the data I have
worked with, speakers often use an ACCC-ciause when they are talking
about a third party. In effect, ACCCs can be too imposing, and therefore
are used typically when there exists a certain asymmetry ("power") in
the social status of the interlocutors. Thus, a tenured professor can easily
indicate to his graduate student:
Complex conditional connectives 165

(41) De acuerdo, Ie dirigire la tesis, a condici6n de que trabaje sobre

el subjuntivo.
'Agreed, I'll supervise your thesis, provided that you work on the
subjunctive. '

On the other hand, this utterance would not be very likely the other
way around; that is, it would be precarious if the graduate student
declared something like the following to his superior:

(42) Muy bien, dare sus clases a condici6n de que usted me ayude a
preparar la bibliografia.
'Okay, I'll teach your classes provided that you help me with the
bibliography. '

Such an utterance, without a doubt, would seem less risky formulated as

an ~clause:

(43) Muy bien, dare sus clases si usted me ayuda a preparar la

'Okay, I'll teach your classes if you help me with the
bibliography. '

The rule of linguistic politeness which indicates that indirectness is

more courteous than direct expressions also explains why the ACCC-
structures can seem not very polite: with these structures, there is no
interpretation other than the imposition of a necessary condition. Such
constructions are learned; furthermore, they can also seem impolite if
they are used in direct address.

10. Summary

To summarize, I shall try to list the principal traits of CCC-structures

and place them in relation to other structures belonging to the family of
conditional structures:

(i) Non-predictive conditionals are non-prototypical.

a. CCCs cannot appear in non-predictive contexts (i.e., where, in
most cases, [ifp] equals [givenp]);
b. CCCs are related to predictive and future-oriented contexts.
166 Estrella Montolio

(ii) Hypotheticals and counterfactuals are non-prototypical

(Dancygier 1993: 428).
a. CCCs rarely appear in counterfactual contexts;
b. CCCs are basically related to central, predictive conditionals
(i. e., prototypical conditionals).

(iii) In prototypical, predictive conditionals, CCCs are non-central

a. whereas, in prototypical If-conditionals, clauses are related
through causal relationships, in CCC structures, clauses are
related through circumstantial relationships;
b. whereas the prototypical order of central If-conditionals if p, q
makes them iconic with the sequence of events, the
prototypical order of CCC-conditionals q, CCC P makes them
iconic with the order involved in processing assertion-
c. whereas the function of If-clauses is, prototypically, to
establish a state of affairs to be taken as the basis for
considering the consequent, the function of CCC-ciauses is to
restrict or specify the contents of the apodosis (hence, their
incompatibility with then).


* I would like to thank Maria Rosa Vila and Joseph Hilferty for the many hours
that they spent with me discussing many of the points expanded upon here, as
well as the editors of this volume, Bernd Kortmann and Elizabeth Couper-
Kuhlen, for their exhaustive observations on this chapter. I would also like to
thank lbon Sarasola for providing me with the Basque example in endnote 4. I
would further like to mention the help received from Joseph Hilferty in
making an intelligible English version. It goes without saying, however, that
all errors are of my own responsibility.
1. The lack of quality grammatical descriptions regarding CCCs is reflected in
the fact that most Spanish textbooks and reference grammars for foreign
learners give a rather jumbled treatment of these constructions. Furthermore,
they tend to present them as if they were mutually commutable with one
another as well as with ij=-clauses in general. This has the unfortunate side
effect, as I have seen in my own classes, of prompting foreign students to
produce unacceptable utterances.
2. It may be possible that the conclusions reached here are also largely
applicable to languages such as Italian and German. Given my unfamiliarity
with these languages, however, I leave it to others to confirm or disprove this
Complex conditional connectives 167

possibility. It should also be noted, however, that in Spanish--as in other

Romance languages--clauses containing CCCs are generally constructed
with the subjunctive mood. The inflectional contrast between the present and
the imperfect subjunctives allows speakers of the Spanish language to
differentiate the values that conditional constructions with if express through
the indicative/subjunctive opposition.
3. The Spanish ACCC siempre que (literally: 'always that') maintains a
temporal meaning in many of the utterances in which it appears. An
interesting question is which formal elements must be involved so as to
obtain a conditional reading; in this sense, it seems that the presence or
absence of a pause between the relevant clauses is decisive. Conditional
readings tend to be marked by a pause, whereas a lack of a pause seems to
signal a temporal interpretation. Compare the following examples:
(i) a. Nos traera un regalo siempre que vue/va de viaje. (temporal)
'She'll bring us a present every time she returns from a trip.'
b. Nos traera un regalo, siempre que vuelva de viaje. (conditional).
'She'll bring us a present, provided that she returns from a trip.'
4. The relation between CCC-ciauses and postposed ~clauses appears to be
corroborated by the fact that languages such as Basque, which do not possess
specific unitary connectives to convey such conditional nuances, often
express them by postposing the ~clause (as in the Basque ba):
(i) Hondartzarajoango ara eguzkia egiten
badu bederen
make-IMP COND-AUX(CONJ/TR/3SG) at least
'We'll go to the beach if it's at least sunny out.'
5. Something similar is suggested by Leard (1987: 164) upon isolating a
particular group of conditional expressions which he calls restrictives. Dostie
(1987: 191-192) also associates this rectifying mechanism with CCCs, though
she only considers those structures that I have included in the ACCC group.
6. Of course, such utterances would sound more natural if an optative
concessive conditional were used:
(i) Vayas 0 no vayas,
yo ire.
'I'm going whether you go or not.'
7. NCCCs are, in a manner of speaking, related to concessive-conditional
connectives. This applies, specifically, to incluso si ('even if), since
concessive-conditionals express the idea of 'contrary to expectation'. On the
other hand, a no ser que ('unless') and incluso si ('even if) have, in a certain
sense, complementary meanings: a no ser que conveys the only circumstance
that will be an impediment for the fulfillment of q; in contrast, incluso si
entails that p will never be an obstacle for q to be realized.
8. In Spanish there is another type of conditional structure labeled a contrastive
conditional, which is characterized by the thematic nature of the protasis. It
does not appear to have a clear counterpart in English:
168 Estrella Montolio

(i) Si Chatwin eligi6 vlaJar caminando, Paul Theroux redescubri6 las

glorias y zozobras del ferrocarril (El Europeo, May 1993).
*lfChatwin chose to travel by foot, Paul Theroux rediscovered the glory
and angst of the train.
However, just as in the case of other conditionals whose protasis expresses a
contextually given condition, this conditional cannot appear with a CCC, as is
demonstrated by the grammatical behavior of the sentences in (ii).
(ii) *A condici6n de que Chatwin eligiera viajar caminando, Paul Theroux
redescubrio las glorias y zozobras del ferro carri I.
*Provided that Chatwin chose to travel by foot, Paul Theroux
rediscovered the glory and angst of the train.
For more information on Spanish contrastive conditionals, see Montolio
9. In fact, speech-act conditionals are a habitual context for a natural unless-
clause anteposition.Thus, for instance, the example at hand would seem more
plausible with a preposed CCC: A no ser que confunda la palabra, en espanol
americana a eso Ie llaman udurazno" 'Unless I'm confusing the word, Latin-
American Spanish they call this [i.e., a peach] "durazno'"
10. In Spanish there exist two particles that correspond to this value of then:
entonces (which, like then, coincidentally also has a temporal meaning) and
pues (which historically meant 'after'). In fact, the connective pues
'thus'seems actually to be much more frequent than entonces 'then' in spoken
Spanish conditionals and seems to correspond to a marker of sequentiality
(Dancygier, 1998: 178 et passim), that is, a particle that introduces
conclusions. This matter still needs to be examined further, but a first
approximation can be found in Montolio (1991a, 1999).
11. It is worthwhile pointing out that in a conversational context, especially
where direct reference is being made to the interlocutor, conditional threats in
Spanish frequently take the form of a special construction: como + VPRES-
SUBJUNCT. This construction seems to be quite similar to the Japanese -tewa
(Akatsuka, 1997) and the disjunctive paratactic conditional, as in: Cena 0 no
veras la television ('Eat your dinner or you can't watch TV'). For more on
such threatening conditionals in Spanish, see Montolio (1999).
12. To this effect, Brown and Levinson distinguish concrete types of acts that
threaten negative face and that are formulated in the following way: Threats,
warnings, dares (S[peaker] indicates that he (or someone or something) will
instigate sanctions against H[earer] unless he does A) (Brown and Levinson
1987: 65-66; the italics are mine).


Akatsuka, Noriko
1997 Negative conditionality, subjectification, and conditional reasoning.
In: Angeliki Athanasiadou and Rene Dirven (eds.), On Conditionals
Again, 323-354. Amsterdam: Benjamins.
Complex conditional connectives 169

Austin, John L.
1956 1ft and cans. Proceedings ofthe British Academy: 109-132.
Bartol Hernandez, Jose Antonio
1992 Algunas conjunciones y locuciones condicionales en la Edad Media.
In: Jose Antonio Bartol Hernandez, Juan Felipe Garcia and Javier de
Santiago (eds.), Estudios jilol6gicos en homenaje a Bustos Tovar,
vol. 1,85-110. Salamanca: Ediciones Universidad de Salamanca.
Brown, Penelope and Stephen C. Levinson
1987 Politeness. Cambridge: Cambridge University Press.
Comrie, Bernard
1986 Conditionals: A typology. In: Elizabeth Traugott, Alice Ter Meulen,
Judy Snitzer and Charles Ferguson (eds.), On conditionals, 77-99.
Cambridge: Cambridge University Press.
Contreras, Lidia
1963 Las oraciones condicionales. Boletin de Filologia de la Universidad
de Chile XII: 273-290.
Dancygier, Barbara
1985 If, unless, and their Polish equivalents. Papers in Contrastive
Linguistics XX: 65-72.
1990 Conditionals: Sequence of events and sequence of clauses. In: Jaeck
Fisiak (ed.), Further insights into contrastive analysis, 357-373.
Amsterdam: Benjamins.
1993 Interpreting conditionals: Time, knowledge, and causation. Journal
ofPragmatics 19: 403-434.
1998 Conditionals and prediction. Time, knowledge and causation in
conditionals structures. Cambridge: Cambridge University Press.
Dancygier, Barbara and Eve Sweetser
1997 Then in conditional constructions. Cognitive Linguistics 8(2): 1-28.
Dostie, Gaetane
1987 Etude semantique de quatre connecteurs conditionnels: a condition
que, pourvu que, en autant que et d'abord que. Le jran9ais moderne
LV/3-4: 174-203.
Ducrot, Oswald
1973 La preuve et Ie dire. Paris: Marne.
Esgueva, Manuel and Margarita Cantarero (eds.)
1981 El habla culta de la ciudad de Madrid. Madrid: Consejo Superior de
Investigaciones Cientificas.
Fauconnier, Gilles
1984 Espaces mentaux. Paris: Minuit.
Fillenbaum, Samuel
1986 The use of conditionals in inducements and deterrents. In: Elizabeth
Traugott, Alice Ter Meulen, Judy Snitzer and Charles Ferguson
(eds.), On conditionals, 179-195. Cambridge: Cambridge University
von Fintel, Kai
1991 Exceptive conditionals: the meaning of unless. North Eastern
Linguistics Society 22: 135-148.
170 Estrella Montolio

Ford, Cecilia E. and Sandra A. Thompson

1986 Conditionals in discourse: A text-based study from English. In:
Elizabeth Traugott, Alice Ter Meulen, Judy Snitzer and Charles
Ferguson (eds.), On conditionals, 353- 372. Cambridge: Cambridge
University Press.
Ford, Cecilia E
1993 Grammar in interaction. Adverbial clauses in American English
conversations. Cambridge: Cambridge University Press.
Geis, Michael and Arnold M. Zwicky
1971 On invited inferences. Linguistic Inquiry 2: 61-66.
Geis, Michael
1973 If and unless. In: Braj B. Kachru, Robert B. Lees, Yakov Malkiel,
Angelina Pientrangeli and Sol Saporta (eds.), Issues in Linguistics.
Papers in honor of Henry and Renee Kahane, 231-253. Urbana:
University of Illinois Press.
Goffman, Erving
1967 Interaction ritual: Essays on face to face behavior. New York:
Garden City.
Haegeman, Liliane
1984 Pragmatic conditionals in English. Folia Linguistica XVIII/3-4: 485-
Iatridou, Sabine
1991 If Then, then what? North Eastern Linguistics Society 22: 211-225.
Leard, Jean-Marcel
1987 La syntaxe et la classification des conditionnelles et des concessives.
Le Fran~ais Moderne LV (3/4): 158-173.
Lehmann, Christian
1974 Prinzipien flir "Universal 14". In: Hansjakob Seiler (ed.), Linguistic
Workshop 11, 69-97. Munich: Wilhelm Fink Verlag.
McCabe, Anne E., Susan Evely, Rona Abmmovitch, Carl M. Corter and Debra 1.
1983 Conditional statements in young children's spontaneous speech.
Joumal ofChild Language 10: 253- 258.
Montolio, Estrella
1990 Expresi6n de la condicionalidad en espano!. Ph.D. dissertation.
Barcelona: Universitat de Barcelona.
1991a Asi, pues entonces, 10 mejor sera que pienses bien 10 de casarte.
Acerca de los procondicionantes en espana!. Foro Hispanico 2: 43-
1991b On conditional structures with complex connectors. In: Johan van
der Auwera (ed.), Eurotyp working papers: Adverbialia, volume 1:
1999 Capitulo 57: Las construcciones condicionales. In: Ignacio Bosque
and Violeta Demonte (eds.), Gramatica descriptiva de la lengua
espanola, Volume III: 3643-3737. Madrid: Espasa Calpe.
Complex conditional connectives 171

Ramsey, Violeta
1987 The functional distribution of preposed and postposed if- and when-
clauses in written discourse. In: Russell S. Tomlin (ed.), Coherence
and Grounding in Discourse, 383-408. Amsterdam: Benjamins.
Smith, Neil V.
1983 On interpreting conditionals. Australian Journal ofLinguistics 3: 1-
Sperber, Dan and Deidre Wilson
19962 Relevance. Communication and cognition. Oxford: Blackwell.
Sweetser, Eve E.
1990 From etymology to pragmatics: Metaphorical and cultural aspects
ofsemantic structure. Cambridge: Cambridge University Press.
Traugott, Elizabeth Closs
1997 Unless and but conditionals: a historical perspective. In: Rene
Dirven and Angeliki Athanasiadou (eds.), On conditionals again,
145-167. Cambridge. Cambridge University Press.
Visconti, Jacqueline
1994 Entre hypotheses et conditions. Sur la semantique des connecteurs
conditionnels complexes. Cahiers de linguistique fran~aise 15: 77-
Wing, C.S. and Ellin Kofsky
1981 Children's comprehension of pragmatic concepts expressed in
because, although, if, and unless. Journal ofChild Language 8: 347-
Pre- and post-positioning of wenn-clauses
in spoken and written German *

Peter Auer

This paper is concerned with the distinction between pre- and postposi-
tioned (initial and final) wenn-clauses in German, and with the distinction
between written and spoken language. A simple cross-tabulation of the two
features [spoken/written] and [pre-/postpositioned] shows that initial wenn-
clauses are preferred in spoken German, but final wenn-clauses are pre-
ferred in written German. An attempt will be made to explain these find-

1. Introduction

The fmdings and discussions in this paper are corpus-based. They are
partly quantitative, partly qualitative. With respect to both dimensions,
the claim is that a full understanding of the syntax of (particularly) spo-
ken language eludes the possibilities of a purely introspective method-
ology. Of course, no (quantitative or qualitative) corpus-based investi-
gation can do without a strong reliance on the analyst's knowledge ('in-
tuition') about the language being researched; in fact, fmding valid gen-
eralizations always involves Gedankenexperimente, playing with struc-
tural changes in and recontextualizations of the 'examples' found to be
used by the informants. On the other hand, not even the empirical start-
ing point of the present investigation (i.e., the (differing) preferences of
spoken and written language for post- and prepositioning) is available to
a purely introspective approach, since it is of a quantitative kind. In or-
der to reach an explanation of these findings, this quantitative analysis
has to be complemented by an in-depth analysis of individual cases of
usage. Such an analysis will pay attention (a) to the in-time emergence
of syntactic patterns, including the details of their delivery such as hesi-
tations, reformulations, break-offs, etc., and (b) to the interactional as-
pects of this emergence, including hearer feedback (or lack of it) and
sequential placement. In this respect, spoken language research can
profit in important ways from conversation analysis.
174 Peter Auer

2. German wenn-clauses and English ~clauses: a brief syn-

tactic and semantic overview

This, of course, is not the place for a full contrastive analysis of the two
constructions. In order to facilitate the reader's access to the examples
to be discussed below, and in order to link up the present investigation
with previous ones on tf-clauses in spoken (and written) Engiish, a short
sketch of some important similarities and differences may, however, be
in order here.

2.1. Syntax

The syntax of English suggests a (misleading) parallel between pre- and

postpositioned adverbial clauses (including conditionals) since they can
usually be exchanged without structural changes in either the main or
the dependent clause. German syntax, on the other hand, treats the two
positions quite differently. While post-positioned adverbial clauses al-
ways occupy the so-called post-field (Nachfeld), which is not obliga-
tory, and are thus tagged on to an already complete syntactic pattern,
pre-positioned subordinated clauses may be (and in written, normative
language usually are) more tightly integrated into the syntactic structure
of the following main clause: they occupy the so-called front field (Vor-
feld) of the sentence, i.e., the uniquely available and obligatory position
before the finite verb. Moving adverbial clauses from one to the other
position therefore involves structural changes in the main clause:

(1) a. wenn sie==n JOB haben wollen, () mUssen sie==n bisschen da

aufn PUNKT kommen. (original utterance) 1
'if you want to have ajob, you need to get down to the point'
b. sie miissen n bisschen auf==n PUNKT kommen wenn sie==n jOb
haben wollen. (fabricated)
'you need to get down to the point if you want to have a job'

Positioning the wenn-clause in the post-field (lb) instead of the front-

field (la) implies that another constituent will fill this position (in the
present case, it is the subject pronoun Sie). The dominant syntactic pat-
tern in which pre-positioned adverbial clauses occur in written German
may therefore be called "integrative", while the English treatment is
"non-integrative" (cf Konig and van der Auwera 1988: 103-109 for this
Positioning ofwenn-clauses in German 175

terminology and some further remarks). In spoken German, however,

the fully integrated placement of the pre-positioned adverbial clause in
the front-field is only one possibility. Alternatively, pre-positioned
wenn-clauses may be followed by a resumptive particle (a local-
temporal adverbial such as dann or one of its regional equivalents, e.g.
na, denn, no, etc.; cf (2b»;2 or they may even be used in a non-
integrative way, rather like in English (cf (2c»:

(2) a. wenn sie==n JOB haben wollen, () mUssen sie==n bisschen da

aufn PUNKT kommen. (original utterance)
'if you want to have ajob, you need to get down to the point'
b. wenn sie==n JOB haben wollen, () dann mOssen sie==n bisschen
da aufn PUNKT kommen (fabricated)
(same meaning)
c. wenn sie==n WIRKlich n JOB hiitten haben wollen; () sie hiitten
dann SCHON n==bisschen aufn PUNKT kommen miissen
'if you had really wanted a job, (then) you would have needed to
get down to the point'

For (2c), special conditions of use hold, and it may therefore be consid-
ered to be "marked" (cf Konig and van der Auwera 1988, Kopcke and
Panther 1985, Giinthner 1999, and below).

2.2. Semantics

The semantics of German wenn-clauses3 is not strictly equivalent to

English conditional tf-clauses either. The cognate of English if, German
ob (> Germanic *eba), has lost its original (OHG/MHG) function of in-
troducing conditional clauses, although remnants of this usage may still
be found in (etymologically) composite concessive conjunctions such as
Modem German ob+wohl and ob+gleich and in concessive conditionals
of the type ob X oder nicht...('whether X or not'). Filling the gap, the
temporal conjunction wenn (or rather, its predecessors, MHG
swenne/swanne), a cognate of English when, has taken over most of its
functions. As a consequence, the semantics of wenn-introduced clauses
oscillates between a temporal and conditional reading in the indicative
mood. (For this reason, German wenn-clauses, other than English tf-
clauses, cannot be called conditional clauses. 4 ) Other conjunctions are
176 Peter Auer

available for a nonambiguous encoding of a conditional or a temporal

Disambiguation of wenn is possible on the basis of contextual infor-
mation (i.e., the semantics of the remainder of the clause and/or its con-
versational context) in some, but certainly not all cases. Even in the fol-
lowing examples (which are among the clearest in my data) the para-
phrases are not entirely beyond dispute. However, they do represent the
prevalent readings ofwenn-clauses:

(3) (temporal: consecutive)

dann MELD ich mich morgen bei ihnen? (-) wenn «=SOBALD,
SOWIE)) ich bei AUror angerufen hab,
'so I'll be in touch with you tomorrow as soon as I have given
"Auror" a ring'

(4) (temporal: simultaneous)

TRIFFST du den () wenn «=WAHREND, sOLANGE)) du in PEking
'will you see him while you are in Peking?'

(5) (temporal: iterative)

ich sprEch UNdeutlich, (1) LISpel auch n=b' etwas, (-) und (-) ich
sprEch dann after zu SCHNELL. (-) wenn «=JEDESMAL WENN)) ich
beGEIStert bin, oder (-) eh im element bin; (-) dann sprech ich zu
'I speak inarticulately, I also lisp a little, and then I often talk too
fast. whenever I am enthusiastic about something, or ehm get
carried away; (-) then I talk too fast,'

(6) (conditional: hypothetical)

er will sie jetzt wieder HElraten, und die haben so ne FRIST ehm in
den islamischen liindern dass innerhalb von nem halben JAHR
oder so, muss die frau dann wieder zum MANN zuriick wenn
«=FDRDEN FALL DASS, FALLS)) er sie DOCHwiederwil1 h.
'he now wants to marry her again and they have kind of a deadline
ehm in the Islamic countries that within half a year or so the wife
has to return to her husband in case he wants her back again'
Positioning ofwenn-clauses in German 177

(7) (conditional: factual)

«radio phone-in, psychotherapeutic consulting; the caller has
complained about having no-one to tum to with his marriage
problems; the therapist recapitulates and formulates her advice»
denn LETZTlich .h wenn ((=DA)) sie in ihrer verwAndtsch~ft
niemand HAbn mit dems REden kOnnen, .h iih is=danns BESte, (-)
sie wiirden zu am Eheberater gehn?
'for in the end, since you have nobody among your relatives who
you could talk to, ehm it's best then to tum to a marriage counselor'

Clearly, the temporal readings of wenn are not covered by English if,
but by when instead.
Some verbal and prosodic features of wenn-clauses may facilitate or
even enforce one or the other reading: (a) the temporal, non-iterative
reading is not available in sentences referring to past events; here, the
temporal conjunction als takes over (while English allows when); (b)
focussing adverbials such as stressed nur ('only') in the main clause
strongly suggest a conditional reading of the (following) wenn-clause;
(c) the particle schon (no English equivalent) in the wenn-clause sug-
gests a factual-conditional reading; (d) immer wenn ('always when')
instead of a simple wenn as a conjunction enforces a habitual-temporal/
contingent reading; (e) selbst wenn (' even if) and wenn ... iiberhaupt
('if ... at all') instead of a simple wenn enforce a conditional reading; (f)
subjunctive (Konjunktiv II) in the wenn-clause enforces a hypothetical-
conditional reading; (g) stressed wenn-conjunctions suggest a condi-
tional instead of a temporal reading.
Two special uses of wenn-clauses need to be mentioned. The first is
the expression of concessivity through the combination of wenn and
auch (wenn + auch or auch + wenn), roughly similar to Engl. even
if/even though: 5

(8) a. auch wenn sie KEInen job haben wollen, () miissen sie=n
bisschen da aufn PUNKT kommen. (fabricated)
'even if you don't want ajob, you need to get down to the point'
b. wenn sie auch keinen JOB haben wollen, () sie miissen n
bisschen da aufn PUNKT kommen. (fabricated)
'even though you don't want a job, you need to get down to the
c. obWOIa sie KEInen job haben wollen, () miissen sie=n
bisschen da aufn PUNKT kommen. (fabricated)
178 Peter Auer

'although you don't want to have ajob, you need to get down to
the point'

The auch wenn-construction (8a) differs from obwohl-concessives

(Eng!. although, (8c» in that the truth of the proposition it expresses can
but need not be taken for granted ("neutral epistemic stance"; cf Fill-
more 1990, Couper-Kuhlen, 1999): whereas the proposition 'you don't
want a job' is not asserted in (8a)/auch wenn, it is in (8c )/obwohl. Auch-
wenn-clauses therefore differ from if=-conditionals and resemble true
(obwohl-) concessives in that the presupposed generic statement is nega-
tive (for the above example: 'someone who does not want a job does not
have to get down to the point'). At the same time, they differ from true
concessives and are similar to true conditionals in that the truth of the
antecendent may but need not be asserted. Note that, differently from
auch wenn pre-positioned wenn auch-clauses (8b) co-occur with non-
integrative word order in the consequent, and always receive a factual
Finally, it should be noted that German wenn-clauses occur some-
times as obligatory constituents of the verb. 6 (English often uses non-
finite forms such as participle or infinitive clauses for this purpose, al-
though if=-clauses are also possible.)

(9) das EINfachste, da ham sie RECHT, das war fiir uns, wenn sie
mal==n MOnat (-) im teleFONmarketing ARbeiten wiirden.
'the simplest solution for us, and here you are right, would be if
you could work in our direct marketing sector for a month'

Syntactically speaking, wenn-clauses of this kind can be replaced by

dass-(complement) clauses (das Einfachste ware, dass sie mal im Tele-
fonmarketing arbeiten). Semantically speaking, various differences re-
sult from the choice between dass- and wenn-complements; most of
them pertain to the presumed status of the information in the comple-
ment clause (cf Eisenberg 31994: 365f for some further discussion).

3. German wenn-clauses from a quantitative perspective

The observations in this section are based on a collection of 500 wenn-

clauses taken from a corpus of spontaneous, direct conversations. All
instances of wenn were considered for analysis, apart from obvious syn-
Positioning ofwenn-clauses in German 179

tactic break-offs in the wenn-clause, some non-reconstructable utter-

ances, and the comparative uses of als/wie wenn (see note 5). In Figure
1, the total of n=500 tokens is broken down according to the position of
the adverbial clause relative to the main clause: pre-positioning, post-
positioning, parenthetical positioning within the clause, 8 independent
use of the wenn-clause as a turn-constructional unit of its own, and a
residual category of ambiguous cases (e.g. apo-koinu constructions, see
below). There can be no doubt that the front position is preferred in
spoken German.







pre- paren- post-
single others
positioned thetical positioned
IJ percentage 56,00 4,00 32,60 5,40 2,00

Figure 1. Position of German wenn-clauses relative to the main clause~ n=500.

The results agree with Ford and Thompson's fmdings on If-clauses in

English conversations, according to which initials outnumber finals by a
ratio of 4:1 (n=316, initial=81%, fmal=19%; Ford and Thompson 1986:
362), with Ford's findings based on a smaller collection (n=52, 50% of
which where preposed, 35% postpositioned, and 15% single; cf Ford
1993: 24), and with more general claims about a universal preference
for pre-positioning of antecedents in conditional constructions (Green-
berg 1963). Note, however, that the preference found in the English data
for pre-positioning of conditional clauses does not extend to temporal
(e.g., when-) clauses; rather, Ford (1993: 24) found these to follow their
main clauses by a ratio of 2:1. Given the ambiguity of German wenn-
clauses (in the indicative mood) between a conditional and a temporal
reading, it may be asked if the preference for initial placement of wenn-
clauses holds for both. To answer this question, those instances ofwenn-
clauses have been singled out (n=203) which have either a clear tempo-
180 Peter Auer

ral or a clear conditional reading (based on the substitution tests and cri-
teria discussed in section 2.3).9
Of the n=203 disambiguated wenn-clauses, 24% have temporal, the
remainder conditional meaning. Exactly half of the wenn-clauses with
temporal meaning are prepositioned and postpositioned respectively.
(Of the 76% unambiguously conditional examples, about two thirds are
pre-positioned (55% of the total), one third is postpositioned (21 %).)
There is, then, a clear difference between conditional and temporal uses:
only for the former does the preference for pre-positioning hold. Since
the majority of German wenn-clauses are semantically ambiguous be-
tween a temporal and a conditional reading, this fmding also suggests
that, taken as a whole, they behave syntactically like (English) condi-
tional rather than temporal clauses.
Figure 2 shows the percentage of integrative, resumptive and non-
integrative constructions among the pre-positioned wenn-clauses in the
sample (n=280).







integrative resumptive non- others
25,72 47,83 18,48 9,42

Figure 2. Integration (%) of pre-positioned wenn-clauses into the subsequent main

clause (n=280)

The relatively large residual category ("others") covers wenn-clauses

plus subsequent main clauses within larger hypotactical constructions
(see below example (26)-(28)). Again, the results are very clear: re-
sumptive constructions are preferred to fully integrated and totally non-
integrated constructions. The canonical, integrative construction of
standard written German only plays a secondary role in spoken German.
Positioning ofwenn-clauses in German 181

Some comments on non-integrative wenn-clauses in German are

necessary at this point. Pre-positioned wenn-clauses occurring in the
pre-front field of a sentence are basically of two types (cf Auer 1996).
We find instances which cannot be positioned in the front field (i.e., in-
tegrated into the main clause); in other words, the only available pattern
for them is non-integrative syntax. This is sometimes for syntactic rea-
sons; in particular, yes/no-questions and imperatives, which are verb-
initial syntagms in German, do not have a front field, and in w-
questions, the w-question word is usually said to occupy the front-
field. 10 In these contexts, adverbial clauses either need to be post-
positioned (despite the general preference for pre-positioning), or to be
non-integrative. Of the 45 questions/imperatives in the sample, 16 have
pre-positioned wenn-clauses, i.e., non-integrative word order (cf (10»,
while 29 have post-positioning; this means that the normal preference is
reversed in this syntactic environment.

(10) ich mein ich muss ihnen (-) ganz SCHNELL und GANZ: vehement
sagen wenns IRgendwie gEht (-) fahrns InN
'I mean I have to tell you without hesitating and very vehemently:
if you can make it at all, go there!'

However, there are also semantic reasons why certain wenn-clauses

have to occur in the pre-front instead of the front field. This is the case
for "speech-act related" 11 wenn-clauses which do not conjoin two
propositions on the content level; often, they are used in order to miti-
gate subsequent face-threatening acts (such as, in the following exam-
ple, an interruption). The apodosis is asserted independently of the pro-
tasis, and this semantic independence corresponds with obligatory syn-
tactic non-integration:

(11) (Gob interview»

wenn ich (-) grad WElter ausfiihren darf; (0.5) Sie wissenja in de:
in der A Utoinduschdrie .h herrschen SEHR grofJe k'
konkurRENZ, markt
'if I may continue elaborating on that; (0.5) you know that in
the car industry there is a lot of competition «etc.»'

In such cases, the marked position of the wenn-clause in the pre-front

field helps to contextualize a marked (non-referential) semantic inter-
182 Peter Auer

But there are also contexts in which non-integrative syntax is fre-

quent although not obligatory. For instance, there is a tendency for non-
integrative clause-combining to occur in concessive wenn nicht-

(12) wenn auch die theoRIE; (-) eh (-) so IRgendwo mal gehOrt wurde
im KOPF? (-) eh das UMsetzen das ist ja das entSCHEldende,
'even though the theory (-) ehm (-) may have been heard
somewhere in one's head (-) the decisive thing is putting it into

Another frequent function of non-integrated wenn-clauses is topicaliza-

tion; in this case, the wenn-clause is typically followed by an anaphoric
pronoun back-referencing the proposition expressed in the wenn-clause
as a whole, or an element contained in it. In the following example, the
wenn-clause introduces a new discourse referent or topic; it is in many
ways equivalent to other topicalization constructions (such as a cleft
construction: was Ihre Fragen angeht, die konnen Sie jetzt stellen), with
the additional implication that the speaker is not certain about the rele-
vance of the new discourse referent for the co-participant.

(13) also wenn sie FRAgen ham zwischendufch, eh DIE konnen Sie
ruhig STELlen?
'well if you have any questions in between, ehm you can ask
THEM of course. '

A similar topicalization (not of a single referent, but of a whole proposi-

tion) is involved in the following example:

(14) un wenn ich mein Eltem anrufh wUrde, ==DS wiirde AUCH nix
'and if I called my parents, that wouldn't be any use either. '

Here, the wenn-clause could even be entirely replaced by an infinitival

construction (meine Eltern anzurujen), since potentiality is already ex-
pressed by the conditional verb form wiirde... bringen and redundantly
coded by wenn. Finally, non-integrated wenn-clauses often express em-
phasis and lend an emotional meaning to the utterance: 12
Positioning ofwenn-clauses in German 183

(15) wenn WIRKlich==n ganzen tag das telefon klingelt, und acht
STUN' (-) man IS hinterher' < <acc>man WEISS was man>
getan hat. () geb ich ehrlich ZU
'if the phone really rings all day, and eight hOD' (-) afterwards
you are' you know what you have done. i have to admit that. '

In (15), the speaker describes her working-day in a call-centre and

wants to underline that dealing with callers is a tiring job~ one of the
strategies used to convey this meaning is the non-integration of the pro-
tasis into the apodosis.

4. Some reasons for pre- and post-positioning

What are the advantages of pre-positioning wenn-clauses? This question

seems less difficult to answer than the opposite one of why a certain
number of these clauses - roughly a third in our data - are post-
positioned. We will deal with each question in turn.

4.1. The advantages ofpre-positioning

To start with, it should be noted that the preference for pre-positioned

wenn-clauses is not just a quantitative fmding but is reflected in speak-
ers' changes in the design of an emerging syntactic pattern 'in mid-
stream'. Particularly striking are cases such as Ex. (5), in which a post-
positioned wenn-clause is retrospectively turned into a pre-positioned
one via what might be called an apo-koinu construction:
The K01VOV here, of course, is wenn ich beGEIStert bin, oder (-) eh
im element bin. It seems that the speaker, having completed the three-
part list of his verbal handicaps, wants to qualify the last item retrospec-
tively. He could have done this by simply adding the wenn-clause in the
post-field but recycles this last component instead, with the wenn-clause
inserted before it. The wenn-clause here is both final and initial. in-
stances in which a clause is broken off and a wenn-clause is inserted be-
fore it is re-started (as in (16)) are also evidence for the interactional
relevance of pre- vs. postpositioning.

(16) ichfahr (-) wenn (-) wenns iiberHAUPT geht

dennfahr ich NA:CH(er) erscht in Urlaub,
184 Peter Auer

'I'll have if if it works out at all

then I'll only have my holidays afterwards'

So why this additional effort? There seems to be some kind of cognitive

'naturalness' in the way in which conditionals create the ground - or, in
more recent but equally metaphorical parlance, set up a "mental space"
(Fauconnier 1985) - in which some hypothetical or factual proposition
is located. 13 For cognitive reasons, it is the grounding which (iconically)
precedes the focal proposition, and not the other way round. Ford, for
instance, suggests that "the prevalence of initially placed if-clauses may
reflect the general tendency to signal «... » that the interpretation of the
coming clause will be, in some general way, limited by the contents of
the if-clause" (1993: 15). Further evidence for the 'naturalness' of this
position can be derived from the affinity of conditional clauses and
topic-introducing devices (topics precede comments), for which some
evidence has been given in the preceding section (see Haiman 1978,
Ford and Thompson 1986 for an in-depth treatment of this line of argu-
mentation), and from the affinity of conditional and causal clauses
(where causes iconically precede their effects). The advantages of this
discourse function seem to outweigh the cognitive costs linked to the
deployment of a syntactic pattern which projects considerably into time.
It may not have been sufficiently taken into account in previous re-
search on clause positioning, however, that this projection in time has an
interactional side as well: 14 speakers who open up far-reaching syntactic
gestalts claim the tum for at least the time which is necessary to bring
them to a well-formed conclusion. In other words, producing a wenn-
clause gives the speaker the right and obligation to go on talking; it
functions as a tum-holding device until the formulation of the conse-
quent is completed. There are numerous cases in the data in which
highly complex turns emerge in this way, since the speaker uses the
space between a gestalt-opening wenn-clause and a terminating main
clause for detailing the "mental space" opened up by the first compo-
nent. Two elaborate examples (as they seem to be typical for institu-
tional talk) are (17) and (18):

(17) (Gob interview; applicant B is talking about his previous

employment in a West German consultant company which,
however, withdrew from East Germany, despite the fact that it
had highly experienced consultants»
Positioning ofwenn-c/auses in German 185

B: zum beispiel einen herren, () KELler?

(-) eh der () is: () FONFundzwanzig jahre
unterNEHmensberater? () der hat=n STAM!vfklientel in uh es ah
I: mhm,
B: und DER war naturlich, (-) ein FACHmann. (-) aber er KOMMT,
in die neuen BUNdesliinder? () «acc>er war ja nu> (-) eh
hatte es ja gar nicht mehr NDtig gehabt; =da () so [VIEL] (-) zu
I: [mhm,]
B: aber (-) er IS in die neuen BUNDdesliinder gekommen, (-)
um auch etwas zu be WEgen. (-) aber wenn er dann nur auf der
STRASse () sitzt, (-) und DANN (-) den () kliENten () mit nach
schweRIN nehmen muss um=n FDRderantrag zu stellen; (-) dann
wieder zur BANK, (-) und die BANK sagt () wir brauchen
erst=ne ZUstimmung von dem FDRderinstitut,
I: =<p>mhm,
B: vor[her () ko]nnen wir nicht die geSAMTfinanzierung,
I: [(h) ]
<p> wie mit KDpenick. ja
B: und () fund er da]nn NUR aufder STRAsse ist;
I: [(h)]
B: (-) dann SAGT er das LOHNT sichfiir mich nicht. (-)
dann bleib ich LIEber () in nordrhein westFAlen.

'B: for instance a Mr Keller

ehm who has been a consultant for 25 years
he has his regular clients in the U.S. and Canada
I: mhm,
B: and he was a specialist of course. (-) but he is coming
to the New States (.) he certainly had (-) ehm he had no need to do
that any more; to travel so [much there
I: [mhm,
B: but (-) he did come to the New States, (-)
in order to get something moving. (-) but if he is on the road all
the time (-) and then (-) he has to take his client with him to
Schwerin in order to hand in the proposal for the subsidies; (-)
and then back to the bank, (-) and the bank says (.) first we need
the subsidizing body's consent
186 Peter Auer

I: mhm,
B: before [that we cannot (do) the total financing
I: [(h)
like with Kopenick 15
B: and (.) and he is just on the road then he says this isn't worth it
for me.
then I'd rather stay in North Rhine-Westphalia «a West German

This passage is embedded into a larger report the applicant gives of his
participation in a West German consultant agency in the New States,
which however closed down its East German office, making him redun-
dant. The interviewer does not seem to know the company and ques-
tions its importance on the market. The applicant counters by stating
that although small, the company had very professional consultants. At
the same time, he has to deal with the interviewer's innuendo that the
company withdrew from the East German market because it was not
working successfully. In this context, the case of "Mr. Keller" is men-
tioned, an experienced consultant who was disappointed by the
kafkaesque way in which state and bank authorities made it hard for
new enterprises to get subsidies, and returned to the Old States.
After he has been portrayed as a successful consultant who came to
East Germany mainly for idealistic reasons, "Mr. Keller's" dissatisfac-
tion with the situation is described in a complex tum construction which
starts out with a wenn-clause (wenn er nun aufder Straj3e sitzt... ). In the
given context, the interpretation is not hypothetical but refers to a (fac-
tual) state of affairs ('since he was always on the road... '), which is es-
tablished as the ground from which some conclusion can be drawn. Be-
fore this conclusion is reached, however, the speaker elaborates at con-
siderable length on the unfortunate situation in which "Mr. Keller" and
his clients found themselves; in four clauses each introduced by (und)
dann, the various fruitless journeys between the financing bank and the
state authorities in Schwerin are described. Towards the end of this
elaboration (securely produced by the speaker within the realm of his
own tum, since a syntactic projection - that of the when-clause - still
remains to be taken care of), the interviewer produces some recipiency
tokens which, although not claiming the tum (cf their reduced loudness,
indicating non-competitiveness), nevertheless acknowledge the
speaker's point: two laughter particles and one comment (wie mit
Kopenick) display understanding. Only after this feedback does the
Positioning ofwenn-clauses in German 187

speaker close the syntactic gestalt with two resumptive dann-clauses.

Their content is higWy predictable, given the fact that it has been men-
tioned before that the company closed down its East German branch. It
seems, then, that what the speaker wanted to convey by this complex
turn is not so much this consequent but rather the details of the situation
which led to it. The relevant information of this complex construction is
what is produced BETWEEN the initial wenn-clause and the final dann-
clauses. The speaker employs the projecting force of the first in order to
claim conversational space for himself: and makes use of this space as
long as he needs it to 'convince' the recipient of his point (as evidenced
by the recipient's responses). The 'orderly' conclusion of the turn is
produced as soon as this purpose is reached.
The following extract similarly shows how pre-positioned wenn-
clauses can be used to claim conversational space:

(18) (bulimia therapy)

M: aso ich hab ma mit einer zuSA!vfMgewohnt, ==
==und .h die hab ich EH nich so leidn kOnn un sie mich AUCH
und dann hab ich IMmer so.h (0.5)
und (-) DIE: is schon wesentlich DICker als ich; ==
==und dann hab ich ECHT immer gedacht (0.5)
ich hab so alles des (-) AUF se projeziert
und wenn se viel geGESsn hat, ==
==die hat sich .h SAHne n ganzn becher SAHne mit
Apfelschnittchen drin gegessn. ==
==und das warfiir mich ECHT der ABscheu. ==
<<fast>n hab ich gedacht> .h des is ja wohl (1.0) des is
FURCHTbar (1.0)
wie KAMmer denn sowas ESsn un auch noch mit gUtm ge WISsn.

'M: you see I once lived with a girl

and I couldn't really stand her and neither could she me
and then I always
and she really was a lot bigger than I was
and believe me I always thought
I projected everything on her
and when/if she ate a lot,
she put cream a whole cup of cream she ate with slices of apple in
188 Peter Auer

and to me that was really disgusting.

then I thought .h isn't that (1.0) that is really appalling
how can you eat anything like that and without even feeling
guilty. '

Once more, a speaker is involved in telling a story which in this case is

supposed to show how she projected her own feelings of guilt for eating
too much onto her flatmate. And once more, a wenn-clause is the first
component of a syntactically cohesive turn construction which spans six
intonation units. The speaker does not go into gestalt closure (apodosis)
after the wenn-clause, but rather parenthetically includes information
detailing the claim that the roommate 'ate a lot', and how she herself
reacted to that emotionally. Only then does a (dan)n-clause follow
which ties back to the initial part of the turn, where a story concerning
'projection' (ich hab so alles des aufse projeziert) was announced.
There is only one legitimate way for a recipient to share (or rather,
intrude into) the conversational space which a wenn-projection creates
for the current speaker: by becoming a co-speaker herself, i.e, by col-
laboratively producing the gestalt-closing apodosis matching the already
produced protasis (cf. Lerner 1991). Both inserted material between pro-
tasis and apodosis and collaborative constructions pivoting around this
transition suggest that there is some interactional work going on, and
that, at least in a substantial subgroup of examples, the construction is
not planned and executed as one whole, but rather develops in (at least)
two steps.

4.2. Why post-positioning at all?

If pre-positioned wenn-clauses are both cognitively more 'natural' and

interactionally more advantageous than post-positioned ones, why do
the latter occur at all? Two reasons have already been mentioned in sec-
tion 3: wenn-clauses may be used for expressing the temporal circum-
stances of an event, and since temporal adverbial clauses do not follow
the preference for pre-positioning, wenn-c1auses of this semantic type
need not do so either. I Secondly, it was shown that the absence of a
front-field in questions and other verb-initial syntagms makes their post-
positioning more likely. 17 There are, however, other important reasons.
First of all, it may be asked if there are any further syntactic envi-
ronments in which post-positioning is preferred or even necessary.
Positioning ofwenn-clauses in German 189

There is indeed another construction in which the front-field is not

available: that in which the wenn-clause plus subsequent clause are
themselves embedded into a larger construction. The various types of
embedding show different patterns with respect to the possibility of pre-
positioning. As in Ford and Thompson's English data (1986: 359), final
positioning is preferred "when a conditional clause occurs within a
nominalization, an infinitive, or a relative clause". Take, for instance,
the following case of a relative clause:

(19) (therapy session)

TM: s==ESsn isch wie? ein Teddybiir. ==
TW: ==ja,
TM: den' den sie: (-) .h mit sich RUMtragn. (2.0)
damit SIE < <p>nich allEin sein miissn. >
un dem==mer (-) sich RANzieht, ja? (-) (-) wenns HARTwird;
(3.0) an dem=mer sich FESCHThiilt, (2.0) wem==mer EINsam
isch, (1.0) nd der ii: berall MIT muss.

'TM: eating is like a teddy bear.

TW: yeah
TM: who who you carry around with you.
so that you don't have to be alone.
and whom one holds close, right? (-) (-) when life becomes hard;
whom one clings to, (2.0) when one is lonely,
and who has to come along all the time. '

Both wenn-clauses in this extract are part of a relative clause introduced

by an oblique relative pronoun, i.e., their matrix clause is itself subordi-
nated, and therefore has verb-final syntax (cf the placement of the finite
verbs ranzieht and festhiilt). Here, the wenn-clause cannot be placed in
front of the relative clause (*und wenn's hart wird den man sich ran-
zieht); pre-positioning would require a superordinate main clause in-
stead of the relative clause (und wenn's hart wird, zieht man sich den
ran).18 The same applies to dependent clauses introduced by wie 'as',
obwohl 'although', weil 'because', etc. which likewise do not allow ini-
tial wenn-clauses.
However, subordination by the most frequent complementizer dass
('that') shows a different pattern. Here, we frequently encounter initial
placement of the pre-positioned wenn-clause be..fore the complementizer
190 Peter Auer

(20) MEIN interesse is natiirlich

WENN ich da: .h schon als POSTdoc auf==m ZEITvertrag bin;
dass ich wahrend de dieser ZEIT dann; (-) auch==n paar
ergebnisse MITnehme
'my interest of course is
if I am there as a post-doc on a temporary contract
that I can take at least some results with me during this time'

(21) und DESwegen war es natiirlich; (-) fir uns WUNschenswert; ()

<<scanning>WENN wir uns einigen KONNten, >
dass sie so friih wie MOGlich <dim>natiirlich anfangen.
'and therefore of course it would be desirable for us
if we could come to an agreement
that you start as soon as possible'

The additional stress on wenn in these examples may give us a clue to

the origin of this construction; arguably, it underlines the semantic link
between antecendent and consequent. Fronting the wenn-clause to a po-
sition before the dass-complementizer may be another way of focussing
on the semantic link established by wenn. 19 Note in passing that the
fronting of the wenn-clause renders its scope ambiguous both in (20)
and (21): it mayor may not include the initial phrases mein Interesse ist
natiirlichl ...ware es natiirlichfir uns wiinschenswert (i.e.: 'of course, if
I am only there as a post-doc on a temporary contract, then my interest
is to take at least some results with me' and 'if we could come to an
agreement it would of course be desirable for us that you start as soon as
possible' respectively).20
In addition to these syntactic constraints, there are semantic-syntactic
reasons for post-positioning wenn-clauses. In particular, wenn-clauses in
complement function are usually post-positioned (cf (9) above). As a
rule, the main clause contains an evaluative two-place predicate, with
the wenn-clause expressing the proposition which is evaluated.21 The
opposite serialization is not unacceptable; nevertheless, it is very rare.
The dominant pattern obviously parallels that of dass-introduced com-
plement clauses which can, but rarely do, precede the main clause as
well. Complements make up ca. 25% of all the post-positioned wenn-
clauses in the spoken materials investigated.
Finally, and most importantly, post-positioning of wenn-clauses is
linked to the pragmatic status of the proposition they express, and to the
interactional possibilities this position opens up both for the speaker and
Positioning ofwenn-clauses in German 191

the hearer. As outlined in section 1, final subordinated clauses in Ger-

man are added onto an already complete syntactic structure. They are
therefore a straightforward means for expanding a syntactic gestalt, and
thereby the tum-at-talk. This is particularly obvious in cases where syn-
tactically complete syntagms preceding the wenn-clause are marked as
terminal by intonation, e.g. by a pitch fall to the speaker's base line (full
stop in the transcription); the wenn-clause then appears as an after-
thought, or epexegesis (cf Auer 1991):

(22) «hypothetical talk about a situation in which two people are in

conflict over where to put the cup for the coffee; B is asked to
B: ich wiird [einfach] die () die tasse kaffee NEHmen,
II: [<p>h:m,]
B: und eh (-) WEGste//en. (-) .hja? ()
eh==s da stunden/ang streiteREIen gibt,
wiird ich sagen, a/[so: ] jetzt is:-
II: [hm,}
II: da hiitt==ich iirger [mit IHnen.
B: [sch/uss aus ENde?
II: [wenn SIE mir dann auch noch den KAFfee (wegschlieften.)]
B: [he he he he he he

'B: I would [simply take the (.) the cup of coffee

II: [mhm
B: and ehm (-) put it away. (-) you see?
before they start quarreling for hours,
I would say right [now it is
II: [hm,]
II: then I would have trouble [with you.
B: [over and out
II: [if you (shut away) my coffee
B: [he he he he he he'

At a point where B has already suggested to simply 'taking away' the

disputed coffee cup, but is in the middle of a syntactic construction
elaborating on this proposal (eh's da stundenlang Streiterei gibt wiirde
ich also sagen: Schluss, aus, Ende) II intervenes during an intra-tum
192 Peter Auer

hesitation pause to refute this solution: 'if you did that, there would be
trouble between the two of us' (i.e. between the mediator, B, and one of
the two people quarreling, i.e. himself). The utterance is linked to B's
proposal by the initial anaphoric da; it is semantically and syntactically
complete, and being marked by a final fall, it certainly is a candidate for
a complete tum. However, B does not pick up this refutation, but con-
tinues with the production of the unfinished syntagm in another piece of
simultaneous talk. Sequential structure and temporal development are
now out of phase: a response has been produced to an utterance which is
still in need of being completed, and is only completed after the re-
sponse. In this context, II's following wenn-clause, syntactically ex-
panding an already complete turn/syntagm, can be seen as a skillful way
of re-aligning sequentiality and timing: it re-instantiates II's refutation
of B' s proposal without repeating it, by retrospectively transforming a
simple construction into a hypotactical one with a post-positioned ad-
verbial clause. Semantically, this expansion adds nothing new: it just
restates what B herself has said before.
The possibility of such an expansion is not only available to the
speaker but also to the recipient, of course, who may become a co-
speaker and co-producer of the emerging syntactic pattern by adding a
wenn-clause himself/herself:

L: .h dann: eh () wir' der Hundwird auchjetz zunehmend ruhiger;
S: mHM (-) des GUT so; (-)
L: JAja des==also wird langsam (a)==richtiger HUND;
S: aHA
L: hm, (-)
S: wenn==er (nicht mehr) abhaut, (-)
hat (name) des ANgebot jetz fiir den zaun? ==

'L: then ehm beco the dog is becoming more and more calm now;
S: this is how it should be;
L: yeah slowly he's tuming into a real dog.
S: I see
L: hm,
S: if he doesn't escape (any more),
did NN get the offer for the fence in the meantime?'
Positioning ofwenn-clauses in German 193

So it is not only the transition between a wenn-clause and its subse-

quent main clause which is sensitive to turn-taking, but also the inverse
transition between a (main) clause and its subsequent wenn-clause. But
obviously, there is an important difference: while in the first case an
open syntactic projection is in play, in the second case the first speaker
has already come to an orderly completion of the sentence/turn.
Post-positioned wenn-clauses thus offer the possibility not only of
expanding a turn, but also of expanding a sentence by adding a post-
field constituent. At least example (22)22 also points to an important
pragmatic feature of such expansions: its low information value. Indeed,
this applies to a very large number of post-positioned wenn-clauses. Of-
ten it is the whole previous text which functions to build up the 'mental
space' that is necessary to come to the conclusion expressed in the main
clause, while the post-positioned wenn-clause only summarizes this pre-
ceding text, sometimes slightly changing the focus. In (24), the
introductory adverbial insofern explicitly establishes this resultative link
between pre-text and conclusion, while the post-positioned wenn-clause
just repeats what is known from the previous conversation anyway (the
wenn-clause is factual here):

(24) «after a long discussion of the applicant's career aspirations in the

bank, and an equally long description of the branch of the bank in
Stralsund and its sophisticated private customer service, which
seems to match these wishes»
(( ..)) das HAM wir alles in stralsUnd, also inSOfern, (-) eh ware
das==ne ideAle (-) STELle, (-) wenn sie (-) praktiZIEren wollen im
verTRlEBSbereich. im KUNDdennahen bereich.
'we've got all that in Stralsund, so in that regard, this would be
an ideal position if you want to be a trainee in the sales
department. in client-oriented business.'

As in other, similar cases of low-relevance wenn-clauses in final posi-

tion, the front-field is used here for a connecting (anaphorical) adver-
bial, which is preferentially placed in sentence-initial position, where its
indexical meaning is most easily processed. Since only one constituent
may be placed in the front-field, this position is not available for the
wenn-clause any longer.
Since post-positioned wenn-clauses are often of low pragmatic rele-
vance, upgrading their informational value requires special means; a
standard technique for doing so is the use of focussing particles such as
194 Peter Auer

(stressed) auch, nur or dann, or a combination of these. In this case, it is

the focussing particle which projects syntactically: it requires a constitu-
ent to follow which is in its scope. Therefore, wenn-clauses such as the
following cannot be treated in the same way as post-field wenn-clauses
in general: they do not expand an already complete syntactic pattern but
rather close a gestalt projected by the particle.

(25) «about wearing glasses))

ich zieh==se nur DENN ouf wEnn==i==se wIrklich (-) Effektiv
'I only put them on then when I really and positively need them'

Ex negativo, the necessity of using such focussing particles in order to

upgrade the following wenn-clause to rhematic status is evidence for the
(sub)thematic status which wenn-clauses usually have in the post-field.

5. Pre- and post-positioning of wenn-clauses in written Ger-


In the last sections, it has been shown that wenn-clauses are preferen-
tially pre-positioned with respect to their main clauses in spoken Ger-
man, and that this serialisation has a number of cognitive and interac-
tional advantages. It has also been shown that the more marked struc-
ture, i.e. post-positioned wenn-clauses, which does occur in about a
third of all instances, has its own specific contexts of usage. These are
partly due to (a) syntactic constraints on pre-positioning in superordi-
nate clauses without a pre-field or in which two complementizers occur
in adjacent position; partly to (b) tum-taking (afterthought position);
partly to (c) semantic-syntactic reasons (wenn-clauses in complement
function are postpositioned); and partly to (d) pragmatic reasons (post-
positioned dependent clauses are thematic or subthematic, unless focus-
sing particles indicate the contrary).
In written German, wenn-clauses are generally less frequent than in
spoken discourse, a finding which contradicts the frequent claim that
spoken language avoids syntactically complex constructions: the fre-
quencies of wenn-clauses per 100 words in the corpus of spoken lan-
guage used above is 0.54, but in a corpus of written language, taken
from the newspapers DIE ZEIT (politics section) and Frankfurter All-
gemeine Zeitung (culture section), it is 0.33. 23 To put it differently,
Positioning ofwenn-clauses in German 195

every 186th word is wenn in our spoken corpus on an average, but only
every 300th word in our written texts. Once more, this fmding is in line
with comparative work on written and spoken English (Ford and
Thompson 1986, 354: 0.72 vs. 0.46; similarly: Beaman 1984 and Biber
1986), but also with previous work on German (Leska 1965: 450).
There are of course also qualitative differences between the wenn-
clauses used in the two corpora; in particular, certain rather idiomatic
patterns (constructions) seem to be more or less exclusively used either
in spoken or written language. For instance, the reduced wenn-(dann)-
constructions of spoken German (cf section 2.2. above) do not occur in
the newspaper corpus, while, on the other hand, the topicalizing causa-
tive construction wenn p dann (deswegen), weil q ('if p, then that is be-
cause of q'), as in (26), seems to be used exclusively in writing.

(26) (DIE ZEIT Nr. 8 17.02.1995)

Wenn wir das Leugnen von Auschwitz, anders als das Leugnen
der kopernikanischen Wende, unter Strafe stellen, dann deswegen,
weil es uns nicht nur hypothetisch angst macht.
'If we punish the denial of Auschwitz, and not the denial of the
Copernican revolution, then that is because it does not make us
feel afraid only hypothetically. '

Also, and contradicting received wisdom according to which written

language is more logical and more explicit, we fmd instances of wenn as
a conjunction in the newspaper texts with semantics which are exceed-
ingly vague, as for instance in (27):

(27) (F.A.Z 19.06.1993, S. 27 / Nr. 139)

Aber wenn zum stets und instiindig angestrebten "Weltniveau" der
DDR die Stellvertreterschaft des iiberragenden, fortschrittlichen,
darin sagar selbst "biirgerlichen" Kulturerbes gehorte, dann
iiufJert sich nun das AufschliefJen zur Weltoffenheit seltsam
'But if the representation of an outstanding and progressive
cultural heritage, one which includes even the "bourgeois",
belonged to the ever and urgently sought after "international
standard" of the GDR, then growing into cosmopolitan open-
mindedness expresses itself rather timidly nowadays. '
196 Peter Auer

Here, the wenn-dann construction seems to vaguely express something

between adversativity and concessivity.
However, these differences only affect a relatively small number of
examples and are not directly linked to the positioning of the wenn-
clause. The important question for the present discussion is rather
whether the preference for pre-positioning of wenn-clauses is also to be
found in written German (as it is in written English, cf Ford and
Thompson 198624). Fig. (3) shows that this is not the case:








a percentage 51,60 1,60 1,00

Figure 3. Percentage of post- and pre-positioned wenn-clauses in written Ger-

man (n=626)

In the written materials, post-positioned wenn-clauses are almost one

and a half times more frequent than pre-positioned ones, while the op-
posite ratio is found in the spoken material. 25 As would be expected, the
number of integrative pre-positioned wenn-clauses is higher than in
conversational language (65.56%), and both the number of non-
integrative (9.28%) and resumptive (26.16%)26 structures is markedly
reduced. The preference for post-positioning is only slightly less pro-
nounced in unambiguously conditional wenn-clauses than in unambigu-
ously temporal ones (65 % vs. 71%) (n=301).
What could be the reason for this reversal of preferences for post-
and pre-positioning in written German compared to spoken language?
Three factors seem to be primarily responsible for it. First, the number
of (almost exclusively final) wenn-clauses with a preceding focus parti-
cle in the main clause is about eight times as high in the written as in the
spoken material (40 vs. 5 occurrences). An example is:
Positioning ofwenn-clauses in German 197

(28) (DIE ZEIT Nr. 7 10.02.1995)

Danach durfen Frauen ungewollte Schwangerschaften in den
ersten drei Monaten nur beenden, wenn sie sich vorher haben
beraten lassen: in einer Beratungsstelle und vom abtreibenden
'Accordingly, women may only terminate an unwanted pregnancy
during the first three months if they have undergone counselling:
in an advice centre and also by the physician who does the
abortion. '

Secondly, although embedded wenn-clause plus main clause construc-

tions are not more frequent in the written than in the spoken material, all
52 wenn-clauses of this type are post-positioned, while a majority of
them (34 of 54, all of which are embedded into dass-constructions) are
pre-positioned in the spoken materials. In other words, fronting of
wenn-clauses before the embedded matrix clause such as in (20) or (21)
does not occur in the newspaper texts.
Finally, one of the important reasons outlined above for frequent pre-
positioning in interactional language use is simply not applicable to
writing: this is the need for the speaker to claim conversational space for
the production of a larger tum, through projecting syntactically beyond
the current clause. Instead, another factor becomes relevant: in writing,
the wenn-clause may become so complex that processing it would be-
come difficult even in reading if it was pre-positioned with respect to its
main clause; cf for instance:

(29) (DIE ZEIT Nr. 03 13.01.1995)

Und man denkt an Talleyrands Feststellung: "Hochverrat ist eine
Frage des Datums", wenn man sich daran erinnert, daft Hans
Modrow in einer Phase als Reformer und Hoffnungstriiger galt,
aber in der niichsten fiir schuldig erachtet wurde, weil er
mitverantwortlich warfiir das DDR-System.
'And one thinks of Talleyrand's statement: "High treason is a
question of the date" when one recalls that Hans Modrow was
regarded as a reformer and as a source of hope in one phase, but
that he was found guilty in another, because he shared
responsibility in the GDR political system. '
198 Peter Auer

Neither of these reasons for post-positioning in written German can ex-

plain the difference between the English and the German results, of
course. Why should these same reasons not lead to a preference for
post-positioning in written English as well? If one was looking for a
structural explanation, one would probably try to find an answer based
on the most prominent difference between English if- and German
wenn-clauses, i.e. the semantic ambiguity of the latter. For instance, it
might be argued that since wenn can often be interpreted either condi-
tionally or temporally, German newspaper journalists try to disambigu-
ate their sentences by using other, strictly conditional conjunctions in-
stead, such as falls, im FaIle dass, fiir den Fall dass, sofern or soweit.
However, this hypothesis receives little empirical support in my data:
not only are these conjunctions very rare in the newspaper texts (a total
of 31 tokens!), they also fail to show a positional distribution different
from that of wenn (6 initial vs. 17 fmal tokens, with 8 parentheticals).27
Another possibility to express conditionality in German which is not
available in present-day English (apart from peripheral cases) is inver-
sion (as in: kommst du zu spat, bestraft dich das Leben := wenn du zu
spat kommst, dann bestraft dich das Leben 'if you are late, you will be
punished by life'). This possibility is almost never used in spoken Ger-
man because of its bookish and high-register connotations but its occur-
rence cannot be excluded in rather conservative newspapers such as DIE
ZEIT and F.A.Z.; and since the distribution of pre- and post-positioned
conditional clauses with inversion is unknown we cannot exclude a bal-
ancing effect, for instance due to a preference for pre-positioning in this
case. Since this syntactic pattern can only be quantified in syntactically
labelled corpora, there is no possibility to test this hypothesis in a
straightforward way in our materials. However, preliminary analyses of
some texts suggests that inversion hardly occurs in newspapers. Alterna-
tively, one might look for a non-structural explanation which would lo-
cate the reason for diverging English and German patterns on the textual
level, possibly in the stylistic preferences of English and German text
composition. The matter clearly awaits further investigation.

6. Conclusion

In this paper, I have looked at the placement of German wenn-clauses in

spoken and written texts. Various explanations for the general quantita-
tive results - i. e., that spoken German prefers pre-positioning, written
Positioning ofwenn-clauses in German 199

German post-positioning - have been presented and discussed on the

basis of individual conversational contexts in which wenn-clauses occur.
The general conclusion of this study is that the supposed parallel be-
tween 'left' and 'right' in syntax (suggested by parlances such as 'left
extraposition' vs. 'right extraposition', or 'left-adjoined' vs. 'right-
adjoined') is fundamentally mistaken when applied to spoken syntax; in
speaking, to be sure, there is no 'left' and 'right', but only 'earlier' and
'later'. At least for an approach to syntax which takes the in-time ('on-
line') emergence of (particularly) oral language units seriously, what is
dealt with first and what is taken care of later cannot be seen as a deci-
sion between two logical equivalents (as between 'right' and 'left').
Rather, it involves one of the most basic and far-reaching decisions a
speaker can make, with all kinds of cognitive, interactional and struc-
tural repercussions. 28

Appendix: Regularisation ofTranscriptions ofWord-Count(example)

original transcription: regularised transcription:

M .hh ich will UMgehend den (-) nachs- Ich will umgehend den nachsten An-
ten (-) ANrufer wieder einen Horer rufer, wieder einen Horer, begrii.6en:
begrii.6en guten Abend? Guten Abend.
A guten Abend, Guten Abend.
B guten Abend? Guten Abend.
A .hhh ja; ALso hh das proBLEM ah Ja, also das Problem aller steht auch
ALler steht ah auch ah so fib vor MIR so vor mir jetzt irgendwo. Wie und
jetzt irgendwo-wie wie und wo AN- wo anfangen am besten? Es ist eine
fangen am besten? etwas auBergewohnliche Problema-
B rnhm, tik, die vielleicht nicht so ganz haufig
A = ahrn: (--) es is eine etwas AUSser- in ihrer Sendung erscheint, obgleich
gewohnliche Proble ProbleMAtik, die das eigentlich eine total menschliche
vielleicht nicht nicht so ganz ah: : Angelegenheit ist. Ich lebe seit, urn
HAUfig in ihrer sendung erscheint, gleich mal auf Fakten zu kornrnen,
.hh obgleich des eigentlich ah::: ne ich lebe seit circa fiinfzehn Jahren
total MENSCHliche angelegenheit is- mit einern Mann zusarnrnen.
ich lebe: seit =um gleich mal auf
FAKten mal zu KOMM: ich lebe seit
hh fib circa FONFzehn JAHRN mit
einern MANM zusArnrnen? [.hh ]
B [ja] Ja?
A und fib auch ah recht GU:T eigentlich Und auch recht gut, eigentlich. Na-
=natiirlich mit den iiblichen ALLtags- tiirlich mit den iiblichen Alltags-
schwierigkeiten die iiberall existieren schwierigkeiten, die iiberall existie-
200 Peter Auer

original transcription: regularised transcription:

auch in HEterobeziehungen, ren, auch in Heterobeziehungen.

B ja, la.
A u:nd fih das=(dies) ist nicht unsere Und das ist nicht unsere Problematik,
probleMAtik eigentlich, wir kommen eigentlich. Wir kommen mit unserer
mit diesem mit unserer verANlagung Veranlagung sehr gut zurecht, also
sehr gut zuRECHT? also=[das]= ist das ist alles o.k. soweit.
alles o.k. soweit,
B [ja:] Ja.


* The corpus research on which this paper is based was supported in many and
substantial ways by Benjamin Stoltenburg. Thanks to Jessica Wallace for cor-
recting the worst blunders in my English, and to Susanne Giinthner as well as
to the editors of this volume for their comments on a previous version.
1. Transcription of the spoken extracts follows GAT-conventions (cf. Selting et
al. 1998); capital letters indicate stress positions. English translations are sim-
plified, particularly with respect to prosody and hesitation phenomena. In case
of conflict, less idiomatic versions have been chosen in order to give a better
impression of German syntactic structure.
2. The wenn-clause itself should be seen as adjoined to the resumptive particle,
i.e., as a co-constituent of the front-field; cf. Eisenberg (31994: 364f).
3. For a detailed discussion, see Metschkowa-Atanassowa 1983 and Zifonun et
al. 1997: 2280-2293.
4. A certain kind of ambiguity between a temporal and a conditional reading can
also be observed in the German question word wann (usually asking for tem-
poral information, 'at which time') which, when followed by a verb in the
subjunctive mood, often takes on a conditional meaning (wann wiirdest du
kommen = 'under which conditions would you come' or 'at which time would
you come'). Thanks to Bernd Kortmann for drawing my attention to this par-
5. For an analysis of these "concessive conditionals", cf. Konig 1985.
6. Among these wenn-clauses in the role of obligatory constituents, we may also
count comparisons using wie wenn and als wenn, as in: du kOmmsch dir
vielleicht vor wie wenn dir deine wErte verLORN gangn sin. ('maybe it seems
to you as ifyour values had been lost '). This usage of wenn will not be taken
into account in the following discussion, nor has it been included in the quan-
titative analysis.
A note in passing: some grammarians believe that wenn-clauses in comple-
ment function are obligatorily marked by a resumptive es (e.g. Eisenberg
31994: 365); this is not supported by my data, however.
7. Some 400/0 of the corpus are job interviews, mainly collected among north and
east German speakers, some 30% are therapeutic conversations, both in face-
to-face and in radio phone-in contexts, and the remaining 30% represent pri-
Positioning ofwenn-clauses in German 201

vate everyday conversations, partly on the telephone. In the latter two types of
data, southern German speakers prevail.
8. Embeddings of wenn-clauses into complex hypotactic constructions were not
counted as parenthetical.
9. The count excludes, in addition to all polyvalent cases, all factual conditionals
(Le. those expressing a positive epistemic stance), which are always non-
temporal, and all concessives, as well as wenn-clauses used as complements,
but includes counterfactual conditionals and reduced wenn-dann routines.
10. *Warum [wenn Du Kirschen magst] pfliickst Du Dir keine vom Baum? 'Why
[if you like cherries] don't you pick any from the tree?' therefore has to be
understood as parenthetical. Resumption by dann is also excluded here, Le.,
non-integration is the only option.
11. The term is used in a broader sense here than in Sweetser 1990. Details on this
construction may be found in Gfinthner 1999. Note that Sweetser's "epistemic
conditionals", although not "content conditionals", do not allow pre-front field
placement in German (* Wenn er sich jeden Tag voillaufen lasst, sie hat ihn
verlassen. 'If he gets drunk everyday, she has left him. ').
12. Cf. Konig and van der Auwera 1988: 128 ("assertive emphasis on a conse-
quent of a concessive allows non-integration"), Kopcke and Panther 1989:
700 ("high degree of ego involvement") and Gfinthner 1999 for details.
13. Cf., among others, Ford and Thompson 1986: 370; Ford 1993; Dancygier and
Sweetser 1996.
14. But see Ford 1993: 56.
15. Speaker I is most likely alluding to Zuckmayer's play (and a famous German
movie) Der Hauptmann von Kopenick, in which the Prussian state and army
authorities are caricatured.
16. This of course, leaves the question open why temporal adverbial clauses
should behave differently from conditional ones-a question which requires
an investigation of its own.
17. For a similar remark on English, cf. Ford and Thompson 1986: 369.
In some cases, however-though not in (26) with its oblique relative pro-
noun-the wenn-clause can follow the relative pronoun (das Essen ist wie ein
Teddybtir, der, wenn es hart ist, immer bei Ihnen ist, und der, wenn man ein-
sam ist, zum Festhalten da ist). But here we are dealing with parenthetical
placement in the middle field of the sentence; this is exceedingly rare in spo-
ken German.
19. Of course, wenn is not always stressed in fronted wenn-clauses. Cf. the fol-
lowing example:
ich (.) hab (-) FUNF jahre lang an der schule franzOsisch geHABTh, mir
jEhlts eigentlich an (.) PRAxis, .h aber: (-) ich bin aberZEUGT davon, =wenn
ich: eh eh OFters mal die geLEgenheit hatte zum beispiel in FRANKreich, eh
mich aufzuhaltn, .hh dass des: (-) eh SICherlich Ausbauftihig is.
'I had French at school for five years, actually I'm lacking practice, but I'm
convinced if I on occasion had the chance to spend some time for instance in
France, that I could work on it. '
20. The tendency to place the wenn-clause early in dependent constructions is also
evidenced by the fact that parenthetical placement immediately after dass is
202 Peter Auer

frequent (cf. Note 18). Often, a second, resumptive dass is added at the begin-
ning of the consequent:
kAnnst du ihm vielleicht (-) A USrichten dass ich ANgerufn habe?
und dass wenn er mit dem trelitz geSPROChen hat uber meine priifungk,
dass er sich dann irgenwie=mal=GANZ kurz bei mir mElden soli?
'could you perhaps tell him that I called?
and that if he has talked to Trelitz about my exams,
that he should give me a quick ring some time?'
21. The same was found in English conversations by Ford and Thompson 1986:
22. In other-speaker produced post-positioned wenn-clauses this does not always
hold, since second speakers may choose this way of intimately linking their
speech to a preceding syntactic pattern but nonetheless produce unexpected
and even contradictory information under this 'disguise'.
23. More exactly, the corpus included the F.A.Z. Feuilleton-Glossen from Jan 8,
1993 (Ausgabe Nr. 6) to Dec 31, 1993 (Nr. 304) and DIE ZEIT politics sec-
tion of Dec 30, 1994 (No.1) to Feb 17, 1995 (Nr. 8). Transcriptions of con-
versational speech were regularized in order to make a comparative computer-
based word-count possible; see the appendix for an example.
24. The written corpus used by Ford and Thompson consisted of philosophical
essays, a professional text for automobile mechanics, and a personal narrative
account (1986: 355). The preference for pre-positioning held for all these
25. A separate count for the two newspapers shows that the quantitative results
are identical.
26. The resumptive particle is dann in 46 cases and so in 16 cases. Dann and so
are not freely interchangeable; in particular, so can be used with concessives
(introduced by auch/selbst wenn), while dann cannot. Cf.: Auch wenn das Ab-
geordnetenhaus erst im Herbst, voraussichtlich am 22. Oktober, gewtihlt
wird, so hat mit der Urabstimmung schon der Wahlkampf begonnen. ('Even
though parliament will not be elected until autumn, probably on October 22,
SO the electorial campaign has already begun with the strike ballot. ')
27. Of the 15 tokens in the spoken material, the three positions were about equally
28. For a similar argument, cf. Thompson 1985.


Auer, Peter
1991 Das Ende deutscher Satze-Rechtsexpansionen im deutschen Ein-
fachsatz. Zeitschriftfur germanistische Linguistik 1991: 139-157.
1996 The pre-front field in spoken German and its relevance as a gram-
maticalization position. Pragmatics (Special issue, eds. Johannes
Wagner and Cecilia Ford), Vol. 6, No.3: 295-322.
Positioning ofwenn-clauses in German 203

Biber, Douglas
1986 Spoken and written textual dimensions in English: Resolving the
contradictory findings. Language 62: 384-416.
Couper-Kuhlen, Elizabeth
1999 Varieties of conditionals and their emergence in discourse. In: Aditi
Lahiri, Alexander Patschosvky and Christoph Schwarze (eds.), Is-
sues in interdisciplinary research on the lexicon = Working Paper
No. 99, University of Konstanz, Fachgruppe Sprachwissenschaft, 89-
Dancygier, Barbara and Eve Sweetser
1996 Conditionals, distancing, and alternative spaces. In: Ann E. Goldberg
(ed.), Conceptual structure - Discourse and language, 83-98. Stan-
ford: CSLI Publications
Eisenberg, Peter
1994 GrundrifJ der deutschen Grammatik. Stuttgart: Metzler.
Fauconnier, Gilles
1985 Mental spaces. Cambridge: Cambridge University Press.
Fillmore, Charles J.
1990 Episternic stance and grammatical form in English conditional sen-
tences. In: Chicago Linguistic Society 26, University of Chicago,
Ford, Cecilia E.
1993 Grammar in interaction. Adverbial clauses in American English
conversations. Cambridge: Cambridge University Press.
1997 Speaking conditionally: Some contexts for if-clauses in conversation.
In: Angeliki Athanasiadou (ed.), On conditionals again, 387-413.
Amsterdam: Benjamins,
Ford, Cecilia E. and Sandra A.Thompson
1986 Conditions in discourse: a text-based study from English. In: Eliza-
beth Traugott et al. (eds.), On conditionals, 353-372. Cambridge:
Cambridge University Press.
Greenberg, Joseph H.
1963 Some universals of grammar with particular reference to the order of
meaningful elements. In: Joseph H. Greenberg (ed.), Universals of
language, 73 -113. Cambridge, Mass. : MIT-Press.
Gtinthner, Susanne
1999 Wenn-Satze im Vor-Vorfeld: ihre Formen und Funktionen in der
gesprochenen Sprache. Deutsche Sprache 27(3): 209-235.
Haiman, John
1978 Conditionals are topics. Language 54: 564-89.
Konig, Ekkehard
1985 Where do concessives come from? On the development of conces-
sive connectives. In: Jacek Fisiak (ed.), Historical semantics, 263-
282. Berlin: de Gruyter.
204 Peter Auer

Konig, Ekkehard and Johan van der Auwera

1988 Clause integration in German and Dutch conditionals, concessive
conditionals, and concessives. In: John Haiman and Sandra A.
Thompson (eds.), Clause combining in grammar and discourse, 101-
133. Amsterdam: Benjamins.
Kopcke, Klaus-Michael and Klaus-Uwe Panther
1989 On correlations between word order and pragmatic function of con-
ditional sentences in German. Journal ofPragmatics 13(5): 685-711.
Lerner, Gene
1991 On the syntax of sentences-in-progress. Language in Society 20(3):
Leska, Christel
1965 Vergleichende Untersuchungen zur Syntax gesprochener und ge-
schriebener deutscher Gegenwartssprache. Beitrage zur Geschichte
der Deutschen Sprache und Literatur 87: 427-464.
Metschkowa-Atanassowa, Sdrawa
1983 Temporale and konditionale "wenn "-Satze. Dusseldorf: Schwann.
Selting, Margret, Peter Auer, Birgit Barden, Jorg Bergmann, Elizabeth Couper-
Kuhlen, Susanne Gunthner, Christoph Meier, Uta Quasthoff, Peter
Schlobinski, Susanne Uhmann
1998 Gesprachsanalytisches Transkriptionssystem. Linguistische Berichte
34(173): 91-122.
Sweetser, Eve E.
1990 From etymology to pragmatics. Metaphorical and cultural aspects of
semantic structure. Cambridge: Cambridge University Press.
Thompson, Sandra
1985 Grammar and written discourse: initial vs. final purpose clauses in
English. Text 5: 55-84.
Zifonun, Gisela, Ludger Hoffmann and Bruno Strecker
1997 Grammatik der deutschen Sprache. Vol. III. Berlin: de Gruyter.
Counterfactual reasoning and desirability*

Noriko McCawley Akatsuka and Susan Strauss

This paper calls into question the popular view of counterfactual thinking
under the influence of formal logic, which considers the utterance of a
counterfactual sentence as an instance of complex and intricate reasoning
skills. Using naturally occurring data from English, Japanese, and Korean,
we will demonstrate that there exists a pattern of prototypical
counterfactual reasoning which appears to be natural and spontaneous to
every human being. Speakers/writers express a particular stance of
desirability versus undesirability toward a particular event, based on their
subjective evaluation of reality. Counterfactual conditionals are invoked as
a necessary step in this line of thinking.

1. Introduction l

This chapter represents a continuation of our inquiry into the most typical
usage of counterfactual conditionals in everyday life across language and
culture (Akatsuka 1997, 1999). We use the word "typical" here in the
sense of "natural and spontaneous," and thus underscore our claim that
human beings appeal to counterfactual reasoning in dealing with the
many aspects of everyday life that we find ourselves facing.
In the tradition of mathematics, philosophy and formal linguistics, the
conditional sentence has been regarded as the epitome of Man's rational
capacity, the height of Man's ability to reason logically. In the same vein,
it has long been argued that underlying the counterfactual conditional is a
similar, and perhaps even more logically complex type of reasoning,
engendering such analytical sentences as in (1), from Fauconnier (1985:

A: If Napoleon had been the son ofAlexander, he would have won the
battle of Waterloo.
B: But he would have died long before that.
A: Well, suppose he lived a very long life, without ever ageing, or that
Alexander was resurrected in Corsica in the eighteenth century.
206 Noriko McCawley Akatsuka and Susan Strauss

Obviously, this string of utterances for Fauconnier represents a typical

instance of counterfactual conditionality. The utterance is clearly an
invented dialogue in which both participants realize the gross
counterfactuality (and absurdity) of the proposition 'if p'. And while the
type of reasoning displayed here could indeed be considered "complex,"
the view of counterfactual conditionality that we adopt actually has little
to do with this type of exercise in logic, truth values, and absurdity.
Based on the discourse data that we collected in American English,
Korean, and Japanese just following the 1994 Los Angeles earthquake
(hereafter, the earthquake data), we will show how and to what degree a
qualitative, cross-linguistic analysis can illuminate the role of the
speaker's evaluative stance of desirability. More specifically, we will
demonstrate the importance of the dichotomy of desirable/undesirable in
understanding the inherent ·relationship between counterfactual thinking
and everyday human existence.
The outline of the paper is as follows. In section 2, we will introduce
Fauconnier's (1985) "mental spaces" hypothesis as one current view on
counterfactuals. We will show first that while Fauconnier considers his
framework as "cognitive," it is actually inherently non-cognitive. We will
systematically demonstrate why this is so and why the framework fails to
account for usages of counterfactuals in everyday situations, by
introducing and then explicating a number of his invented examples in the
light of Akatsuka's desirability framework. In section 3, we will examine
spontaneous instances of counterfactual conditionals in the earthquake
data in the three languages (English, Korean, and Japanese) and will
contrast these with Fauconnier's constructed examples. We will also
present a discussion of the theoretical implications of this study. In
section 4, we will demonstrate that the present study sheds critical light
on the treatment of counterfactuals in Ford and Thompson (1986) and
Labov (1972). Section 5 is the conclusion.

2. Fauconnier (1985) and counterfactuality

2.1. Fauconnier (1985) and the tradition offormallogic

A close and careful examination of Fauconnier's (1985) analysis of

counterfactuality reveals that his approach is actually inseparable from the
tradition of formal logic. That is, the major concern of previous scholars
in philosophy and logic has centered on the logical problem of truth
conditions. Within that discipline, the primary aim was to specify under
Counterfactual reasoning and desirability 207

what strict conditions a particular counterfactual statement can be viewed

as a case of 'true' or 'valid' reasoning (cf Jackson 1991; Fauconnier
1985; McCawley 1981). Example (2) excerpted from Fauconnier (1985:
109) encapsulates his approach to the phenomenon of counterfactuality.
As is clearly stated, rather than grappling with issues of logic and truth
conditions, the focus of analysis shifts to the process by which we
establish "counterfactual spaces"-an issue which Fauconnier labels a
"cognitive semantic question."

(2) Linguistically, we do not directly tackle the logical problem of truth

conditions for counterfactuals, but rather the cognitive semantic question
of how counterfactual spaces are set up and structured...Counterfactuality
is a case of forced incompatibility between spaces (emphasis added); a
space Ml is incompatible with another space M2 if some relation
explicitly specified in Ml is not satisfied for the corresponding elements
in M2 (1985:109).

Thus, he solves the truth conditional problem of logic by simply

proposing a hypothetical world. What is crucial here is that he
characterizes counterfactuality as a case of "forced incompatibility
between spaces." He does not, however, specify any other conditions
relating to the spaces M1 and M2, so that it is irrelevant which of the two
is the real world, R, and which is a hypothetical world, H. Further, in
spite of his labeling the problem as a "cognitive semantic" one, we
actually find no evidence of a true theoretical attempt to illuminate the
notion of counterfactuality from a cognitive semantic point of view. We
discuss this in detail in the next section.

2.2. "Space-builders"

In order to understand Fauconnier's view of counterfactuality, it is

perhaps best to first examine his theoretical construct "space-builders."
According to this framework, "space-builders" include various linguistic
elements such as conditionals (if ), verbs like wish, and negatives
such as not and prevent. The particular semantic properties of these
elements, according to Fauconnier, are supposed to be capable of setting
up a counterfactual mental space which functions in such a way that
'some relation,' which does not hold in the parent space, may be satisfied
in the newly created counterfactual space. In this light, consider the
utterances listed in (3a)-(3d), which appear in the original as examples
(3), (4), (5) and (6), respectively (1985: 109-110).
208 Noriko McCawley Akatsuka and Susan Strauss

(3) a. If Lucky had won, I would be rich. I would have moved to Tahiti.
b. I wish Lucky had won. I would be rich.
c. Fortunately, the fire did not cross the highway. My house would
have been destroyed.
d. Luckily, the fire was prevented from crossing the highway. My
house would have been destroyed.

According to this framework, it is simply the linguistic elements not in

italics which are assumed to create the so-called counterfactual space-
that is, the conditional element if in (3 a), the verb wish in (3b), the
negation not in (3c), and the verb prevent in (3d).
However, there is one crucial observation missing here: what these
examples have in common is the fact that they are all subjective
statements. Utterances such as (3a) through (3d) can by no means be
construed as objective descriptions, whether invented by the researcher or
culled from spontaneous discourse data. In other words, the speaker of
(3a) and (3b) sounds unhappy or disappointed because what s/he was
hoping to happen did not actually happen. Conversely, the speaker of (3c)
and (3d) sounds happy and relieved because his/her home was not
destroyed by a particular fire which did not cross a particular highway.
In this light, Fauconnier's framework of mental spaces takes no
account whatsoever of an individual speaker's mental attitude, and thus
pays no special attention to the function of the adverbs fortunately and
luckily, which figure overtly in his own constructed examples (3 c) and
(3d). The function of these adverbials also accounts for analytic gaps in
examples (3a) and (3b). We will explicate this function in the next section
from the perspective of the concept of desirability.

2.3. Desirability and conditionals

We will begin this discussion by examining Fauconnier's analysis of (3c)

and (3d). According to Fauconnier, the lexical items not and prevent
represent the strongest type of lexical space-builder by virtue of the fact
that they deal in the domain of negation, and hence counterfactuality,
since semantically they are capable of reversing and/or negating some
expression of reality (i.e., factuality). Accordingly, lexical items such as
these are much stronger than the if and wish types, since the latter do not
always build counterfactual space. Fauconnier states:
Counterfactual reasoning and desirability 209

(4) Clearly, counterfactuality may be lexically imposed, as it is by strong

negatives such as not and prevent in (5) [i.e., our (3c)] and (6) [i.e.,
our (3d)] (1985: Ill).

What is interesting here is the focused concentration on negation

markers as lexical agents for the construction of this so-called
counterfactual space. However, close examination of these sample
sentences reveals the force of other semantic factors beyond these
negators which playa significant role, if not a more significant role, in the
expression of counterfactuality.
To illustrate, let us begin by omitting the attitudinal adverbs
fortunately and luckily from Fauconnier's examples in (3c) and (3d). The
sentences so altered appear as (5a) and (5b) below:

(5) a. ??Thefire did not cross the highway. My house would have been
b. ?? The fire was prevented from crossing the highway. My house
would have been destroyed.

While the change consists simply of a seemingly minor lexical deletion,

the occurrence of the counterfactual in the second sentence is rendered
unnatural in both examples. The negation markers nonetheless are still
intact-a fact which now casts serious doubt on the claim that
counterfactuality is 'lexically imposed' by such elements as negatives.
We propose instead that the phenomenon of counterfactuality is
inherently related to the critical role of the speaker's stance of desirability.
And while we do not deny the importance of negation in this
phenomenon, we do not recognize negation as the single factor involved.
That is, negation comes into play in conjunction with, and only in
conjunction with, an expression of the speaker's stance of desirable vs.
This becomes much clearer when we compare Fauconnier's examples
with their Japanese counterparts. The English expression,
'Luckily/Fortunately + p (proposition) happened' corresponds most
closely to the Japanese sentence pattern, p-te yokatta. What this grammar
pattern expresses is precisely the speaker's attitude, desirable, towards
the past event/state of affairs described by the proposition, p. Observe
210 Noriko McCawley Akatsuka and Susan Strauss

(6) p-te, yokatta.

[desirable} was lucky
'(I, it, etc.) was lucky that p.'

Now, let's compare Fauconnier's original example from (3c), repeated

below as (7), with the Japanese translation of the same utterance in (8).

(7) Fortunately, the fire didn't cross the highway. My house would
have been destroyed.

(8) Kaji ga dooro no mukoo ni moeutsura-nakute, yokatta.

Fire SM. highway over burn-cross-not fortunate
(Moshi) moeutsutte-itara, uchi ga yakete itadaroo.
IRREALIS burn-cross-if my house SM had burnt down-MDL

'I was lucky that the fire did not cross the highway. If it had, my
house would have been destroyed. '

The essential difference between the two examples is that the conditional
antecedent appears in the Japanese example, (8). That is, unlike English,
Japanese syntax does not normally tolerate monoclausal conditionals. And
once again, we see the overt representation of the speaker's attitude,
desirable, towards the past event/state of affairs described by the
proposition, p. In this case, p is synonymous with 'the fire did not cross
the highway' (=fact).
To illustrate in English, we repeat examples (Sa) and (5b) below.

(5) a. ??The fire did not cross the highway. My house would have
been destroyed
b. ??Thefire was preventedfrom crossing the highway. My house
would have been destroyed.

Note that when we supply an overt conditional antecedent in both

examples, the awkwardness disappears. Observe (Sa') and (5b'):

(5) a' The fire did not cross the highway. If it had, my house would
have been destroyed.

(5) b' The fire was prevented from crossing the highway. If it hadn't
been, my house would have been destroyed.
Counterfactual reasoning and desirability 211

Now, (Sa') and (Sb') both contain the explicit conditional antecedent
'if not P,' and it is here that we can begin to understand what the speaker
is actually doing by invoking counterfactuality: she is presenting the
reason why she considers it desirable that p, i.e., 'the fire did not cross
the highway' (=fact), happened. The counterfactual situation, i.e., 'not p'
or 'the fire's crossing the highway,' would have led to the loss of her
house. We informally represent this line of reasoning, in (9):

(9) It was desirable that p (=fact) happened. If 'not p' (=counter to

fact) had happened, it would have led to undesirable consequences
(='not q' ).

Contrary to Fauconnier's claim, then, it is not the lexical items not and
prevent that build the so-called counterfactual space. Rather, we maintain
that it is the speaker who invokes counterfactual thinking as an integral
step in the line of reasoning, as delineated in (9). Thus, the use of
negation figures strongly in the expression of p vs. 'if not P,' as well as q
vs. 'not q,' or in other words, in the speaker's expression of stance of
desirable vs. undesirable, and not in the construction itself of the so-
called 'counterfactual space. '
Let us now tum to Fauconnier's analysis of (3a) and (3b), repeated
below as (lOa) and (lOb) for convenience.

(lOa) IfLucky had won, I would be rich. I would have moved to Tahiti.

(lOb) I wish Lucky had won. I would be rich.

According to Fauconnier's framework, it is the "space-builders" if and

wish that are responsible for the appearance of the counterfactuality in
these examples. Now, compare the same examples, with the first
sentence of each altered to express the fact p that Lucky (a racehorse) did
not win a particular competition, and the second sentence altered to
include an overt conditional antecedent expressing 'if not p,' as noted in
(lOa)' and (lOb)':

(lOa)' Unfortunately, Lucky did not win. If he had won, I would be rich.
IfI were rich, I would have moved to Tahiti.
(lOb)' Unfortunately, Lucky did not win. I wish he had won. If he had
won, I would be rich.
212 Noriko McCawley Akatsuka and Susan Strauss

We claim that an adequate theory of counterfactuality must explain on

a principled basis that the workings of counterfactual reasoning in
Fauconnier's examples (lOa) and (lOb) are exactly the same as those in
(lOa)' and (lOb)'. According to our analysis, the line of reasoning
involved here will be illustrated informally as in (11).

(11) It was undesirable that p (=fact) happened. If 'not p' (=counter-to-

fact) had happened, it would have led to desirable consequences
('not q").

It will be obvious that (9) and (11) represent the identical pattern of
reasoning, with the sole difference being the speaker's desirability
assessment taking opposite values. The line of reasoning delineated in (9)
and (11) will be referred to henceforth in this study as "prototypical
counterfactual reasoning."
We have demonstrated that Fauconnier's framework is an i1mpersonal
one, not unlike those proposed by scholars from the tradition of formal
logic and mathematics; the view is necessarily limited to an incomplete
observation of the distribution of counterfactuals. Crucially, the
framework fails to account for all four examples (i.e., (3a)-(3d» as
squarely representative of prototypical counterfactual reasoning.
Unwittingly, too, even example (1), based on an invented dialogue
positing absurd conditions ('if Napoleon had been Alexander's son'),
expresses a strong stance of desirability ('he would have won the Battle
of Waterloo'). And what Frenchman would find this anything but
We now tum our discussion away from constructed examples and
examine authentic discourse excerpted from the earthquake data in
English, Japanese, and Korean, where "prototypical" counterfactuals
emerge spontaneously.

3. A qualitative analysis of the earthquake data3

On January 17, 1994, at approximately 4:30 a.m., a strong earthquake hit

Los Angeles, significantly startling its residents. January 17 happened to
be a Monday, and that Monday happened also to be a national holiday
(Martin Luther King's birthday). Because of this, many people had
planned not to go to work that morning. Commuter traffic was
exceptionally light, and many people were still asleep when the
Counterfactual reasoning and desirability 213

earthquake hit. Because of these and other factors, the casualties were
rather light, given the magnitude of the quake .
Approximately three weeks following the incident, researchers from
UCLA collected 'earthquake stories' from dyads of native speakers of
English, Japanese, and Korean. Subjects were predominantly UCLA
graduate students, though some undergraduates also participated. In all
cases, we attempted to ensure that the dyad participants did not know
each other previously, though here too some of the Japanese data reveals
that a handful of subjects did know their interlocutor to some extent. For
the most part, however, subject pairs were intended to be strangers.
The only instructions given to the subjects were to tell each other
about the experiences they had had during the earthquake, and that the
data recording session would last for approximately 20-25 minutes. Other
than these criteria, no explicit instructions were given. The interactions
were both audio and video taped and were transcribed using the
conventions of Conversation Analysis (Atkinson and Heritage 1984).
What is remarkable is that counterfactual conditionals emerged in all
three languages, and in a number of cases we find significant overlap in
terms of the types of issues captured with counterfactuals (e. g., 'if it had
happened during rush hour,' 'if it had happened in a place other than
LA,' 'if it had happened before the recent earthquake retrofitting of some
of the campus buildings,' etc.). None of the conversational dyads was
aware of what the other dyads had said in other recording sessions.
Further, because the study was designed solely to elicit spontaneous talk
in the respective languages, no mention of counterfactuals was ever made
at the inception. It was not until we transcribed and analyzed the data that
we discovered this parallelism.
In the discussion that follows, we limit ourselves to one representative
sample from each language. In all three examples, the subjects
systematically express their reactions to the relatively light damage and
small number of injuries and deaths, in the light of the strength of the
earthquake itself
Interestingly, the abstract line of reasoning of each case is quite similar
to what we have noticed in Fauconnier's examples (3c) and (3d), even
though Fauconnier's examples are monologic and constructed, while our
data derive from spontaneous, natural dialogue. Instead of the initial
occurrence of such attitudinal adverbs as fortunately and luckily as we
witnessed in the Fauconnier examples, we observe here the speakers
exchanging and building on the expression "we were lucky", in all three
languages. We will call this activity 'the opening.' Just following this
sequence emerges the counterfactual utterance, 'if not p' had happened,
214 Noriko McCawley Akatsuka and Susan Strauss

'not q,' would have happened in all three languages. Henceforth, we will
adopt the formula 'if not p, then not q' to represent counterfactual
Let us now examine each excerpt, starting with English.

3.1. English: "Actually it's good that this happened now"

The speakers, T and H, were just talking about the fact that friends living
in the dorms said that the building seemed to sway at the time of the
earthquake; T then adds that some of the UCLA dormitories had recently
undergone seismic retrofitting: "actually, it's good that this happened now
because..." In other words, "we were lucky," hence the opening
Observe, too, that after this opening T immediately appeals to
counterfactuals. What is of particular importance here is T's initial clause
"I jus' kept thinking," which precedes the counterfactual construction,
underscoring just how naturally and just how spontaneously we appeal to
counterfactual reasoning.

(12) English:
«participants: T=Female, H=Male))

(a) Opening

T1 Actually it's good that this happened now because (then)

these dorms were just respir- retrofitted () fer- () y'know ()
earthquakes () two to three years ago.

(b) 'if not p, not q'

==> T2: I jus' kept thinking if this earthquake had been in

==> the middle ofthe afternoon on a day that wasn't a
==> holiday how ba: :d this would have been, like the San
==> Francisco quake. 1- That was at five 0' clock in the ==>
afternoon on a Friday "hh or something like that It ==>
was just. y'know I think it-I think there would have ==>
been so many more lives lost and this would have been
==> a rea: : I tragedy.
Counterfactual reasoning and desirability 215

In the stretch of talk in section (b) 'if not p, not q' we witness the
gradual emergence of a complex set of thoughts, all of which express
some counter-to-fact notion with respect to some aspect of the reality of
this earthquake. That is, first T proposes a time and a day different from
the actual time and actual day of the incident p and then indicates that the
consequences q would have been much worse (or undesirable), a line of
reasoning we encountered earlier in example (9), repeated below as (13):

(13) It was desirable that p (=fact) happened. If 'not p' (=counter-to-

fact) had happened, it would have led to undesirable consequences
(='not q').

Crucially, the speaker attitude underlying this utterance surfaces in the

linguistic manifestation of how T expresses the degree of potential
undesirability of the consequences ('not q'), had the situation played out
under the, counterfactual conditions she just established (i.e., 'different
time, different day'): her utterance is structured with an exclamatory
degree of intensity marker how, coupled with an ostensible sound stretch
in the negative adjective ba:d (I jus' kept thinking how ba::d this
would've been).
But she doesri't stop there. We actually witness her imagination
emerge linguistically, as she now upgrades the potential counterfactual
consequences by not only changing the time of the occurrence, but now
the place, i.e., San Francisco, where in 1989 a comparable earthquake
occurred during the height of rush hour traffic. And once again, the
pattern of reasoning is identical to that in (13).
The significance of this upgrade is twofold: first, it establishes a
recognizable frame of reference for this new level of negative
consequences by recalling an actual incident where those very same
conditions held (i.e., different time of day, different day of the week, and
now a different place); it is a well-known fact that many more people
suffered and died in the San Francisco quake. Second, it underscores all
the more the crucial relationship between desirability stance and
counterfactual thinking by again linguistically matching the upgraded
scenario with an equally upgraded utterance: there would have been so
many more lives lost and this would have been a rea:: I tragedy, clearly
more semantically intense than the initial 'how bad this would've been,
and notice too the audible sound stretch on the intensifier in a rea:: I
216 Noriko McCawley Akatsuka and Susan Strauss

3.2. Co-construction in Korean and Japanese

We have just observed an example from the English database where a

single speaker imagines a scenario different in specific respects from the
actual one, in order to emphasize her current stance of desirability toward
the actual incident.
In the Korean and Japanese examples that follow, we note a similarity
in this type of complex thinking; only here, the imagination belongs to
both parties as they collaboratively participate in picking out specific
aspects of the actual incident which could have been different, and in
jointly constructing the potential undesirable circumstances.

3.2.1. Korean co-construction

First we present the Korean extracts to illustrate the co-construction of

counterfactual reasoning. Example (14) below is framed much in the
same way as its English counterpart in (12). That is, it begins with an
assessment in (a) concerning the fortunate fact that the earthquake
happened very early in the morning, followed by the utterance of a
counterfactual scenario in (b) where the participants underscore just how
lucky they were. There, we will observe that the counterfactual
conditional 'if not p, not q' is collaboratively constructed by the two
H opens this sequence with the Korean expression 'be lucky' tahayng
ita, and also notably ends the sentence with the same expression. He
continues his turn with mannyak-ey, 'if foreshadowing that some
conditional utterance will most likely follow. And T provides an
agreement token, kurehcho 'that's right. '

(14) Korean:
«participants: H=Male, T=Male))

(a) Opening

==> HI: tahayng-i-n kes-un cihca saypyek-ey cicin

lucky-COP-ATTR thing-TM dawn at earthquake
tn-ssunikka tahayng-i-cyo
occur-CONN lucky-COP-TAG
'The lucky thing is that the earthquake happened at
dawn, that was lucky, huh. '
Counterfactual reasoning and desirability 217

H2: mannyak-ey..
if (lit. in a one in ten thousand chance)

==> TI: kureh-cho, ku..

be like-TAG FLLR
'That's true.'

Directly following this assessment sequence, T then builds upon H's

foreshadowed conditional, by introducing a counterfactual element
concerning the time that the earthquake occurred: 'traffic' manha-ssu-I -
ttaykena, 'if it had been at a time when there was a lot of traffic,' in other
words 'if not p.'
H produces the consequent clause 'not q' koyngcanghi manhun
salam-tul-i ama, 'my, a lot of people would probably have died,' which
itself is an emotionally charged statement, underscored all the more by
the exclamatory marker ama '(oh) my. '

(b) 'if not p, not q'

==> T2: ce 'traffic' manha-ssu-I -ttaykena xxx ... freeway'

FLLR traffic be a lot-PST-ATTR time freeway

'ifit had been at a time when there was a lot of traffic

or...the freeway."

==> H3: koyngcanghi manhun salam-tul-i ama

rather many-ATTR people-PL EXCL
cwuke-ss- keyss-cho

'my, a lot of people would probably have died.'

The Korean example represents a very neat case where one speaker
produces the conditional antecedent, 'if not p' clause and the second
speaker the consequent, 'not q.' And the interactional dynamic which
engenders such a collaborative construction is an interesting one.
Even more revealing, however, is the Japanese example, where we
find the most elaborate instance of collaboratively produced
counterfactual reasoning of the three excerpts under investigation.
218 Noriko McCawley Akatsuka and Susan Strauss

3.2.2. Japanese co-construction

In the stretch of talk that follows, we observe an opening interactional

pattern much like that in example (14), where one participant, in this
case, H, expresses the assessment that it was "lucky" the earthquake
happened so early in the morning. This is responded to in the very next
line by A with an expression indicating a stance of agreement, "I thought
about that, too."
What is interesting here is that in both the Korean and the Japanese
examples, the participant who utters the assessment does so in such a
way that co-participation is invited from the outset through the use of
interactional particles. In the case of the Korean utterance, the speaker
ends the assessment with -ci, functioning at times like a TAG-like
particle. Similarly, in the Japanese example below, we find the use of nee,
which among other functions, also serves as a TAG-like marker. And, in
each case, their interlocutors do provide both an appropriate and relevant
response, i. e., utterances providing agreement.

(15) Japanese:
«participants: H=Male, A=Female))

(a) Opening

==> HI : demo nee asa de yokatta desu yo nee

but PRT morning in lucky-PST COP PRT PRT
gozen yo)i han de?
a.m. four 0' clock halfpast at

'But, see, we were lucky it happened in the morning.

at four thirty, right?'

==> AI: atashi mo omotta-n desu yo.

1 too think-PST-PLN-NOML COP PRT
'I thought about that, too.'

In the lengthy collaboratively built sequence that follows, H expands

his own initial utterance and begins to weave a "complex" mesh of
counterfactual thought with his interlocutor, first by altering the time of
the earthquake, from the early morning hours (4:00 a.m.) to the current
time, i.e., the approximate time of recording (4:00 p.m.), and then by
altering the place in conjunction with the changed time ('if it had
Counterfactual reasoning and desirability 219

happened in the library at a time when students were in it'). And finally,
the relevance of an altered time emerges in the talk, referring to the
freeway overpass which collapsed as a result of the quake, beginning at
line A3, and resumed again from H7 through the end of the excerpt.

(b) 'ifnotp,notq'

==> H2: moshi nee kore ga nee imagoro=

ifPRT this 8M PRT now around
'If (it had happened) around now [i.e., about 4 p.m.],'

==> A2: =kore ga nee, yuugata no:

this 8M PRT evening PRT
'(if it had happened) in the evening,'

H3: ima toka ne

now or something PRT
'like now or something. '

==> A3: rasshuji no go-ji toka ne ha ha ha ha

rush hour GEN five o'clock or something PRT
'at rush hour like five 0' clock ha ha ha ha ha'
H4: soo SOD SOD yuugata no
right right right evening GEN
rasshuji no goji toka
rush hour GEN five 0' clock or something

'Right, right, right at rush hour, at like five o'clock

in the evening. '

A4: ha haha haha ha

'ha haha haha ha'

==> H5: anD toshokan nanka de mada gakusei nanka

FLLR library or something LOC still students something
ir<[irujikan dattara:
be at time COP-COND

'Well, like at the library, if it had happened at a time when

students were still there. '
220 Noriko McCawley Akatsuka and Susan Strauss

A5: [iru san yoji toka:

be at three four 0' clock or something
'like three or four o'clock or,'

H6: moo atama kara saa hon ga ochitekite:

EXCL head from PRT book SM fall-come-GER
'well, the books would have fallen on their heads and,'

A6: aa zaa - sugokatta desu nee.

'Oh, like 'CRASH'-it would've been terrible.'

H7: haiuei furiiuei nanka mo zenbu nee

highway freeway something too all PRT
'the highways, freeways, all of them,'

A7: zenbu
'all of them'

==> H8: sundanshichattara moo

'If they had (all) been cut'

==> A8: kanbotsushichatte:

'they would have collapsed and'

==> H9: shisha nanka...juubai gurai

dead people or something ten times about
n n natta deshoo nee?

'The death toll would've been ten times higher, right?'

(c) Closing

==> A9: nn atashi mo omotta-n su asa de yokatta-n daroo naa tte

I too think-PST-PRT moming-INST good-PST-PRT-MDL-PRT
Counterfactual reasoning and desirability 221

'So I thought too that we were lucky it happened in the


HIO: nnn

AIO: minna neteru toki d~tta kara

everyone sleep-te PROG time COP-PST-PLN because
'Everybody was sleeping at that time, so'

HII: honto hukoo chuu no saiwai desu yo.

really disaster in GEN blessing COP PRT
'Right, it was the only blessing within this disaster. '

All: nn=

HI2: =nn nn

In the first line of this sequence, H2, foreshadows his conditional

thought with the word moshi 'if,' and then sets the counterfactual hour to
be 'around now' imagoro. In response, A builds on H's utterance by
repeating it almost verbatim, with the exception of one change: rather
than recycling imagoro, a temporal deictic term whose meaning can only
be recovered from the situated context, she substitutes the time reference
with yuugata, thus slightly upgrading the level of specificity of the
counterfactual time in question. And she substitutes and upgrades the
time reference yet once more in A3, by specifying even more detail in this
counterfactual switch: rashuji no go-ji toka '(if it were) at rush hour, like
5:00 p.m. '-a piece of information which will become supremely
relevant later in the talk.
From H2 through A4, the speakers have created a counterfactual
scenario in which the earthquake could have occurred at a time other than
at the actual time that it did. At H5, they carry this one step further by
now zooming in on a specific location which could have suffered serious
effects, had that quake indeed happened at the altered hour. What is
crucial here, as well, is the fact that this specific location is one which
both purportedly know very well-the library, ('well [suppose this
happened] like at the library, if it had happened at a time when students
were still there'). Note too, that H is the one who proffers this new
222 Noriko McCawley Akatsuka and Susan Strauss

location (H5), and in mid-utterance, as he struggles through the verb iru

('to be-at a location'), A actually begins her tum, by supplying the verb
iru as well as the temporal specification of 3 or 4 0' clock.
And their collaborative imagining continues, as they jointly create a
scene, complete with vivid visuals, supplied by H (moo atama kara saa
hon ga ochitekite 'books would have fallen on their heads'), and
onomatopoetic expressions (aa zaa 'crash!'), supplied by A.
Observe that this very description represents the first instance of a
consequent clause in this entire sequence, and that this consequent clause,
just like its antecedent, is collaboratively produced by both participants,
ending with the assessment of sugokatta desu ne 'it would have been
terrible,' at A6.
At H7, H continues to build on the counterfactual scenario to include
all freeways and highways, thereby creating a much more intense and
dangerous situation than the actual one, zenbunee sundanshichattaramoo
'if all freeways and highways had been cut (like the one that actually was
cut'). A in tum concurs with the first half of the undesirable consequent
kanbotsushichatte: 'they would have collapsed, and', which H completes
by providing the actual upshot of this entire conversational sequence and
line of reasoning, shisha nanka juubai gurai n natta deshoo nee, 'the
death toll would have been ten times higher, right?'
Crucially, as soon as the above conclusion is jointly reached, the
speakers go back to where they left the conversation in the opening. That
is, they again congratulate themselves by saying asa de yokatta 'we were
lucky it was in the morning. '

3.3. Summary

To sum up, in the excerpts from all three languages, we find the
consistent parallel whereby speakers invoke counterfactual thinking to
express the pattern of reasoning originally introduced as (9) and (13), and
repeated below as (16):

(16) It was desirable that p (=fact) happened. If 'not p' (=counter-to-

fact) had happened, it would have led to undesirable consequences
(='not q').

Every instance of counterfactual conditional is invoked to highlight the

speaker's appreciative attitude towards the actual incident. That is, in
each language, we find that speakers have created a domain of
Counterfactual reasoning and desirability 223

imagination in which they alter one or more specific aspects of a given

reality (either on their own, as in the English example, or in collaboration
with their interlocutor, as in the Korean and Japanese examples). In all
three cases, we have witnessed how speakers shifted coordinates of time,
coordinates of place, or both, and then viewed the situations from a
counterfactual perspective which enabled them all to appreciate just how
lucky they were that the damages caused by such a large earthquake were
so mmor.

3.4. Discussion

Why are there such significant overlaps in the contents of counterfactual

conditionals uttered by the American, Korean, and Japanese speakers
when they recall the occurrence of the earthquake from the point of view
of the common experience of the community? In the discussion that
follows, we suggest that this seemingly simple question is the key which
will shed light on the cognitively dynamic aspects of counterfactual

3.4.1. The predictability of 'not p'

In response to the question above, we propose that the counterfactual

utterances by the American, Korean, and Japanese speakers overlap in
the way they do for the reason that the contents of 'not p' are highly
predictable. The speakers in all three languages, in producing
counterfactual utterances, reflect their shared common sense about the
earthquake. In other words, irrespective of language and culture,
residents of metropolitan cities like Los Angeles share the view that if an
earthquake were to occur early in the morning while most residents are
still asleep, casualties would be relatively small, even if freeways and
large public buildings were to suffer heavy structural damage, precisely
because few people would have been present in those areas.
Thinking this way, we realize that the processes of imagination we
have observed in the cross-linguistic data are not really "complex" mental
activities. This is the reason why the two Japanese speakers can enjoy the
lengthy collaborative construction of counterfactual reasoning, as
illustrated by the laughter tokens ha ha ha ha ha (lines A3 and A4) and
their enthusiastic endorsement of the other party's comments soo SOD SOD
'right right right' (H4). Note that they are not co-reasoning to create new
224 Noriko McCawley Akatsuka and Susan Strauss

ideas. There is absolutely no exchange of new ideas going on here.

Instead, in these extended stretches of talk the speakers are jointly
repeating what they have been repeating to themselves and to others and
what they have heard repeated on so many occasions since the earthquake
first hit five weeks earlier. And by repeating what they know and what
basically everybody else already knows, they are congratulating
themselves by conveying the message, "We are really lucky. It could've
been so much worse."

3.4.2. The speaker's belief: "if 'not p' actually happens, then terrible
consequences will result"

In sharp contrast to Fauconnier's 'absurd' conditions in example (1) ('if

Napoleon had been the son of Alexander'), all instances in the earthquake
data of 'not p' of "if 'not p' had happened" actually represent a variety of
situations which could have indeed happened and could have resulted in
much greater casualties. Why is this the case? We suggest that when the
speaker utters "if 'not p' had happened," this utterance reflects the
speaker's belief that if 'not p' does indeed happen, then terrible
consequences will result. In other words, each instance of counterfactual
reasoning in the earthquake data is endorsed by the same speaker's belief
that "if 'not p' actually happens, then 'not q' (=terrible consequences)
will result." What we observe here is the co-existence of conditional
judgment and counterfactual judgment in the consciousness of the
speaker-which explains why the earthquake data contains so many
instances of expressions of great relief

3.4.3. The 'causal link' between p and q5

In order to explain the co-existence of conditional reasoning and

counterfactual reasoning in the consciousness of the speaker that is
observable in the earthquake data, we will need a dynamic theory of
conditionals which postulates that the connection between p and q is an
integral part of the grammatical meaning of natural language conditionals.
Logicians have long noted that normally there is some kind of connection
between p and q. It was partly due to the difficulty in pinpointing the
exact nature of this connection in the framework of formal logic that they
generally concluded, regardless of their personal stand on the analysis of
~f, that this connection should be treated as a problem of pragmatics
Counterfactual reasoning and desirability 225

rather than grammar (e.g., Quine 1950, Grice 1975, Stalnaker 1975). The
position of these logicians has been more or less inherited by many
linguists (e.g., Geis and Zwicky 1972, Haiman 1978, Karttunen and
peters 1979, McCawley 1981). It is important to note that Fauconnier
(1985: 113-115) explicitly rejects the idea that there is an inherent
connection between p and q of both indicative conditionals and
counterfactual conditionals. Fauconnier even takes the position that the
counterfactuality of q in counterfactual conditionals is implicature and
therefore cancelable. Basically, then, Fauconnier (1985) treats natural
language conditionals very much like mathematical conditionals. Note
that in the case of mathematical conditionals, there is no connection
between p and q of 'p ::> q'; they are independent propositions. Further,
there is no sequential relationship between the two, either. However,
McCawley (1981) observed that in all English conditionals, p is
temporally and/or causally or epistemologically prior to q. We now come
to the conclusion that it is far from being an accident that Fauconnier's
mental space theory has failed to account for the prototypical
counterfactual reasoning phenomenon in our discussion in Section 2.

4. More on aspects ofprototypical counterfactual reasoning

4.1. Ford and Thompson (1986): 'contrast'

It will be illuminating here to examine Ford and Thompson's (1986)

analysis of conditionals in the light of the present discussion. Ford and
Thompson (1986) is a pioneering study of discourse functions of the if-
clause as a link between the preceding context and the following
discourse. Based on their quantitative analyses, they have identified four
basic discourse functions shared by their written and spoken data, one of
which is to present a contrast with the preceding context. This view is
shown by the formula in (17) below. What is relevant to our study is that
(18) is provided as a clear case illustrating that such a contrast can be
indicated by means of a counterfactual. (18) is an excerpt from a personal
narrative about the behavior of a chimpanzee by the name of Nim
Chimpsky, trained to use American Sign Language. (17) and (18) appear
in the original as examples (6) and (8), respectively (1986: 359).

(17) X. (But) if not X, then Y.

226 Noriko McCawley Akatsuka and Susan Strauss

(18) Nim's aggression increased mainly because of the necessity of

introducing more and more teachers into his life... 1f it had been
possible for him to have grown up with a small and stable group of
caretakers, he would have experienced fewer separations from his
trusted caretakers and had far fewer opportunities to test his dominance
through aggression. (Terrace 1979: 145).

According to our analysis, what underlies (18) is the same line of

reasoning as delineated in (19) below. This reasoning is identical to that
expressed earlier in (16), with the exception that the speaker's
desirability stance has taken the opposite value. (19) was previously
introduced as (11), and together with (16) these expressions represent
instances of 'prototypical counterfactual reasoning.'

(19) It was undesirable that p (=fact) happened. If 'not p' (=counter-to-

fact) had happened, it would have led to desirable consequences
('not q").

The writer of (18) has a warm and sympathetic understanding of the

poor chimp's aggressive behaviors resulting from his unfortunate
upbringing. This has been made unmistakably clear by his appeal to
counterfactual reasoning. We agree with Ford and Thompson that
'contrast' is indeed a useful notion in understanding the discourse
function of the counterfactual conditional; however, we propose that what
is contrasted in (18), is not 'if not X' and its preceding context X, but
rather the undesirable reality 'caused by X' and the desirable situation Y
which could have taken place.

4.1.1. Clancy, Akatsuka, Strauss (1997)

Strong empirical support of our re-analysis of Ford and Thompson's

formula comes from our recent cross-linguistic research project (Clancy,
Akatsuka, and Strauss 1997). Based on a total of 84 hours of spontaneous
discourse data, we examined how American, Japanese, and Korean
parents use conditionals when talking to children less than three years of
age. Since the subjects were so young, these interactions were
particularly rich in such adult-produced speech acts as commands,
prohibitions, and granting permission. We discovered that
overwhelmingly the most frequent type of conditional uttered by the
caregivers was what we called "predictive" conditionals, which predict
Counterfactual reasoning and desirability 227

that if one does an undesirable thing, then undesirable consequences will

result (American English 63%, Japanese 74%, Korean 80%).
Observe the following example where an American mother is talking
to her two-year-old child who wants Santa Claus to bring her two dolls
for Christmas:

(20) Mother: You have to be a good girl. If you are not a good girl,
Santa Claus won't bring you anything.

Throughout the data, the caregivers often gave warnings, criticisms, and
advice to their children, constituting a particular type of persuasive
speech act. Typically, this pattern of persuasion takes the following form:
An initial utterance expressing "do X" or "don't do X" is followed by the
provision of a reason, couched in a conditional statement. In other words,
the message takes the general shape of "if children don't listen to what
adults say, undesirable consequences will result." What is being
contrasted here are desirable consequences versus undesirable
consequences. We can informally represent this line of reasoning involved
here as (21):

(21) It is desirable that p will happen. If 'not p' happens, it will lead to
undesirable consequences.

There is an obvious parallelism between prototypical counterfactual

reasoning and the pattern of reasoning delineated in (21). From this, it
will be clear to the reader that the speaker's stance of desirability plays a
critical role in understanding the many usages of conditional utterances in
our daily lives, not just the role it plays in prototypical counterfactual
reasoning alone.

4.2. The discourse domain: The case ofa book review

Because the prototypical counterfactual serves to underscore the

speaker's stance of desirability or undesirability toward a particular
consequence, the discourse domain in which this kind of reasoning
manifests itself can be quite broad. In order to illustrate this point, let us
look at a one-page book review which appeared in the "Dividing Line"
column of a recent issue of Time magazine (October 4, 1999: 90). The
book is entitled Gone Boy and was written by a father whose 15 year old
son was killed by a fellow student in a college campus shooting incident.
228 Noriko McCawley Akatsuka and Susan Strauss

The reviewer, Jack E. White, is a long-time friend of the author~ the

review is directly related to White's stance on gun control.
The column heading reads "Elegy for a Gone Boy," in large, bolded
black letters, accompanied by the subtitle, "A father's searches for
meaning in his son's murder," which appears in much smaller font. It is
clear that the function of the heading and its subtitle is not simply to
provide an efficiently abridged abstract of the contents of the book-it is
an unmistakable indicator of the reviewer's mental attitude toward an
undesirable reality.
The review begins as a personal narrative, which appears in (22)


AT Swathmore College. Greg was 15-the same age his elder son
Galen had reached in 1992 when he was slaughtered in an act of
senseless violence. Galen was in his second year at Simon's Rock
College in Great Barrington, Mass., when a fellow student named
Wayne Lo went berserk and shot up the campus with a cheap imported
rifle, killing Galen and a teacher and wounding four others.

White devotes a sizable amount of space to the body of the article,

which provides an account of how Galen's grieving father spent seven
long years in the investigation into the cause of his son's murder. He goes
on to report that the result is "a tapestry of shared pain and guilt in which
everyone, including the killer is in some sense a victim." The review ends
with the concluding remarks reproduced in (23) below. Notice that this
utterance is a textbook case of prototypical counterfactual reasoning, "if
'not p' had happened, it would have led to desirable consequences."

(23) None of this would have happened if Wayne Lo, at the age of 18,
had not been able to walk into a gun store, flash his driver's license
and $129 and walk out with a deadly weapon. Or if he had not been
able to have 200 bullets sent to him at Simon's Rock College by a
mail-order arms company. To my friend, Greg, there is a
straightforward conclusion to be drawn from the mystery of Galen's
death. "We've just got too many guns in this country. We've got to get
rid of them". Anyone who reads Gone Boy will find it hard to disagree.

By appealing to counterfactual reasoning, the reviewer is providing the

reader with detailed and specific new information about just how easily
Wayne Lo was able to buy a gun, have the ammunition sent to him at
school, and actually carry out this fatal shooting. The bottom line here is
that there are indeed too many guns in America.
Counterfactual reasoning and desirability 229

Thus, counterfactuals do not just describe unrealized possibilities; they

can and do eloquently report what actually happens in the real world.

4.3. Labov (1972): 'comparators'

This section will serve as a preview of our next project, which involves a
re-examination of Labov's (1972) seminal study of narratives in the light
of our prototypical counterfactual reasoning hypothesis. Specifically, we
are interested in Labov's treatment of counterfactuals in relation to his
key notions of "evaluation" and "comparators". Simply stated, Labov's
"evaluation" is a syntactic unit with which the narrator shows why a
particular event is a reportable experience; "comparators" are
grammatical devices that the narrator utilizes to evaluate the narrative.
According to the original framework, neither notion is related to the
speaker's evaluative stance of desirability. Consider the following excerpt
(1972: 387) (emphasis added), where Labov cites one of the most
dramatic danger-of-death stories to illustrate the notion of what
constitutes a comparator:

(24) One of the most dramatic danger-of-death stories was told by a retired
postman on the Lower East Side: his brother had stabbed him in the
head with a knife. He concludes:
And the doctor just says, "Just about this much more", he says,
"and you'd've been dead".
Comparators then include negatives, futures, modals (emphasis added),
quasimodals, questions, imperatives, or-clauses, superlatives, and
comparatives, more or less in increasing order of syntactic complexity.

As stated here, Labov (1972) treats counterfactual utterances as

evaluating clauses because they utilize "modals".
We are impressed by Labov's perceptive observations. For example,
he continues the above discussion by saying that "A comparator moves
away from the line of narrative events to consider unrealized possibilities
and compare them with the events that did occur." Unfortunately,
however, the question of "what precisely is it that the speaker is
evaluating with those devices called comparators?" remains unaddressed.
Further, as we have just pointed out in the previous section,
counterfactuals do not just consider unrealized possibilities. Skillful
speakers/writers such as the narrator of this story know that they can
effectively report what actually happened by appealing to counterfactual
reasoning, and in so doing, they convey the message "I was really lucky."
230 Noriko McCawley Akatsuka and Susan Strauss

5. Concluding remarks

Throughout this study, we have attempted to demonstrate that there exists

a pattern of "prototypical counterfactual reasoning," which appears to be
natural and spontaneous for every human being. We represent this pattern
informally as in (25) and (26).

(25) It is desirable that p happened. If 'not p' had happened, it would

have led to undesirable consequences (='not q').

(26) It is undesirable that p happened. If 'not p' had happened, it would

have led to desirable consequences (='not q').

These two represent the identical pattern of logic.

We have argued that prototypical counterfactual reasoning in the
earthquake data calls for a cognitively dynamic theory of natural language
conditionals which can account for the fact that there is a necessary,
simultaneous co-existence of conditional and counterfactual conditional
reasoning in the consciousness of the speaker. That is, each occurrence of
"if 'not p' had happened" in the cross-linguistic data entails the same
speaker's belief that if 'not p' actually happens, then 'not q' (=terrible
consequences) will result. This analysis explains why the earthquake data
is permeated by expressions of such a great sense of relief In this line of
thinking, we realize that the seemingly "complex" imagining processes
that we observed in the cross-linguistic data are not really "complex"
mental activities because 'not p' is highly predictable; it is a reflection of
the shared common sense of the residents of Los Angeles about the
earthquake and its damages, irrespective of language and culture.
We have further argued that prototypical counterfactual reasoning
calls for a theory of natural language which postulates that the antecedent
and consequent of both indicative and counterfactual conditionals are not
only sequentially ordered but they are 'causally' related to each other.
What emerges from the study of prototypical counterfactual reasoning,
then, is the understanding of the fundamental differences between the
grammar of the natural language conditional and that of the mathematical
conditional. Recall that in the case of the mathematical conditional, the
antecedent and consequent are independent propositions. Moreover, there
is no ordering relationship between them. Our study suggests strongly
that the popular view of counterfactual thinking as "complex" is a by-
product of theoretical frameworks which consider that the natural
Counter/actual reasoning and desirability 231

language conditional has a similar structure to that of the mathematical

In conclusion, we hope to have demonstrated instead that
counterfactual reasoning as it occurs typically in everyday situations is
inextricably related to an expression of the speaker's stance toward the
desirability or undesirability of a particular outcome.




* We would like to dedicate this study to the memory of James D. McCawley.

If Jim had not written McCawley (1981) we would never have learned the
joy of discovering the logic of natural languages.
The authors have contributed equally to the production of this chapter and
are both equally responsible for its contents.
1. The authors are grateful to the two volume editors, Bernd Kortmann and
Elizabeth Couper-Kuhlen, and to Sandy Thompson for their insightful
comments and constructive criticisms, which greatly enhanced the quality of
this paper.
232 Noriko McCawley Akatsuka and Susan Strauss

2. See Akatsuka (1985) for a full discussion of the epistemic scale where this
relationship is discussed in cognitive terms.
3. The Japanese data were collected and transcribed by Shoichi Iwasaki and his
research assistants, Eri Yoshida and Yumiko Kawanishi. Data collection and
transcription for the Japanese data appearing here were partially supported
by the University of California's Pacific Rim Study Grant (1992-94). The
English and Korean used in this study data were collected and transcribed
independently from the Iwasaki data.
4. The magnitude of the Northridge earthquake was 6.6. In all, 51 people lost
their lives and over 5,500 people were injured. For the sake of comparison,
the earthquake in Kobe, Japan, which occurred exactly one year later at
approximately the same time of the morning, measured 7.2 on the Richter
scale~ the number of fatalities in the Kobe earthquake exceeded 5,000.
5. See Akatsuka (1986) for an earlier discussion of this point based on
introspective data


Akatsuka, Noriko
1985 Conditionals and epistemic scale. Language 61: 625-639.
1986 Conditionals are discourse-bound. In: Elizabeth Closs Traugott,
Alice ter Meulen, Judy Snitzer Reilly, Charles A. Ferguson (eds.),
On conditionals, 333-352. Cambridge: Cambridge University Press.
1991 Dracula conditional and discourse. In: C. Georgopoulos and R.
Ishihara (eds.), Interdisciplinary approaches to language: Essays in
honor of S.-Y. Kuroda, 25-37. Dordrecht/BostonILondon: Kluwer
Academic Publishers.
1997 On the co-construction of counterfactual reasoning. Journal of
Pragmatics 28: 781-794.
1999 Towards a theory of desirability in conditional reasoning. In: Akio
Kamio and Ken-ichi Takami (eds.), Function and Structure, 195-
213. AmsterdamlPhiladelphia: Benjamins.
Atkinson, J. Maxwell and John Heritage (eds.)
1984 Structures of social action: Studies in conversation analysis.
Cambridge: Cambridge University Press.
Clancy, Patricia M., Noriko Akatsuka and Susan Strauss
1997 Deontic modality and conditionality in adult-child discourse: a
cross-linguistic study. In: A. Kamio (ed.), Directions in functional
linguistics, 19-57. Amsterdam: Benjarnins.
Fauconnier, Giles
1985 Mental spaces. Cambridge, MA: MIT Press.
Counterfactual reasoning and desirability 233

Ford, Cecilia E. and Sandra A. Thompson

1986 Conditionals in discourse: A text-based study from English. In:
Elizabeth Closs Traugott, Alice ter Meulen, Judy Snitzer Reilly,
Charles A. Ferguson (eds.), On conditionals, 353-372. Cambridge:
Cambridge University Press
Geis, Michael L. and Arnold M. Zwicky
1971 On invited inferences. Linguistic Inquiry 2: 561-566.
Grice, Paul H.
1975 Logic and conversation. In: Peter Cole and Jerrold L. Morgan
(eds.), Syntax and semantics 3: Speech acts, 41-58. New York:
Academic Press.
Haiman, John
1978 Conditionals are topics. Language 54: 564-589.
Jackson, Frank (ed.)
1991 Conditionals: Oxford readings in philosophy. New York: Oxford
University Press.
Karttunen, Lauri and Stanley Peters
1979 Conversational implicatures. In: C.-K. Oh and D.A. Dinneen (eds.),
Syntax and semantics 11: Presupposition, 1-56. New York:
Academic Press.
Labov, William
1972 The transformation of experience in narrative syntax. In: William
Labov (ed.), Language in The Inner City: studies in the Black
English vernacular, 354-396. Philadelphia: University of
Pennsylvania Press.
McCawley, James D.
1981 Everything that linguists have always wanted to know about logic
**(but were ashamed to ask). Chicago: University of Chicago Press.
Quine, Willard Van Orman
1950 Methods oflogic. New York: Henry Holt and Co.
Sacks, Harvey, Emanuel Schegloff and Gail Jefferson
1974 A simplest systematics for the organization of turn-taking for
conversation. Language 50: 696-735.
Stalnaker, Robert
1968 A theory of conditionals. Reprinted in: William L. Harper, Robert
Stalnaker and Glenn Pearce (eds.) (1981), 1ft: conditionals, belief,
decision, chance, and time, 41-56. Dordrecht/Boston: Reidel/
Terrace, Herbert
1979 Nim: A chimpanzee who learned sign language. New York:
Washington Square Press.
White, Jack E.
1999 "Elegy for a Gone Boy" Book review in Time magazine October 4,
234 Noriko McCawley Akatsuka and Susan Strauss

1994 Los Angeles earthquake data:
English data: Susan Strauss, 1994
Japanese data: Shoichi Iwasaki, 1994
Korean data: Susan Strauss and Yumiko Kawanishi
Adversative connectors on distinct levels of
discourse: A re-examination of Eve Sweetser's
three-level approach *

Ewald Lang

While basically acknowledging the discemability of three levels on which a

sentence is interpreted (content, epistemic and speech act level), the paper
tackles Sweetser's claim that the correct interpretation of interclausal
connections depends not on form, but on a pragmatically motivated choice.
I argue for a division of labour between grammar and pragmatics proper by
(i) showing that there are more level-specific structural differences at issue;
(ii) proving the notion "epistemic" to be feasible only if statements,
assumptions, and inferences are kept distinct; (iii) providing an account for
the unavailability of content-level interpretation for adversative

1. Introduction

In Chapter 4 of her exciting book From Etymology to Pragmatics (1990),

Eve Sweetser offers a fresh view of the notorious diversity of
interpretations we observe with causal and adversative connectors. She
rejects the view that connectors like because, although or and and but are
lexically polysemous in the classical sense and suggests instead that the
extremely varied uses of these connectors should be explained in terms of
pragmatic ambiguity. Drawing on Sweetser (1990:76-78) the main tenets
of her approach can be summarized as in (I)-(ill):

(I) One meaning, several uses

In polysemy, a morpheme has several related semantic values; in

pragmatic ambiguity <... >, a single semantics is pragmatically
applied in different ways according to pragmatic context.
236 Ewald Lang

(II) Three levels of interclausal connection ("conjunction"l)

... conjunction may be interpreted as applying in one of (at least)

three domains [where] the choice of a "correct" interpretation
depends not on form, but on a pragmatically motivated choice
between viewing the conjoined clauses as representing content
units, logical entities, or speech acts. (1990:78 ) [emphasis mine

The domains of interpretation are illustrated by clear-cut examples

such as (la-c) below, all of which have in common that the second clause
asserts a proposition in the form of a statement.

(1) a. John came back because he loved her. [content or fact

b. John loved her, because he came back. [epistemic level]
c. What are you doing tonight,
because there's a good movie on. [speech act level]

The levels are distinguished depending on what the clauses are taken
to represent and hence on what sort of entities the causal relation is taken
to obtain. Thus, in (la) it is two content units between which a factual
causal relation is asserted to obtain. More exactly: the content units are
two propositions whose couching in declarative sentences enables them
to render assertable statements. The linking pattern for a causal relation
on the content level thus amounts to STATEI\ffiNT because STATEI\ffiNT.
In (1 b), the first clause is not asserted as a statement but only as an
assumption which is inferred from what is stated as fact in the second
clause. The causal connector marks the second clause as sufficient
evidence justifying the assumption rendered by the first clause. The
linking pattern on the epistemic level is ASSUMPTION, because
STATEMENT (EVIDENCE). The hypothetical status of what is asserted by
the first conjunct can be made explicit by embedding it under modals -
cf (1 b-i, ii) or by rephrasing it as an inference - cf (1 b-iii):

(1) b-i. John must have loved her, because he came back.
b-ii. John undoubtedly loved her, because he came back.
b-iii. From the fact that he came back,
I conclude that John loved her.
Adversative connectors on distinct levels ofdiscourse 237

In (Ic), the first clause serves the purpose of a speech act of asking
whose performance is being justified by the statement rendered by the
second clause. The linking pattern thus is SPEECH ACT, because
STATE:MENT (JUSTIFICATION). The because-clause does not form part
of the question but is linked instead to an understood assertion which, if
explicitly inserted, may read like this:

(1) C-l. What are you doing tonight? I am asking that

because there's a good movie on.

In the course of presenting a variety of cases that are less clear-cut

than the ones in (la-c) above, Sweetser does of course mention
disambiguating features that may (but need not) be present in the
grammatical structure of the conjunction.
For example, she hints at the role of intonation: commaless intonation
is linked with the content reading in (Ia) but unavailable to the other
readings in (lb) or (lc); the latter require a so-called "comma intonation"
such that the clause-final intonation drop marks the first clause as an
independent assertion. Sweetser also discusses the role of iconic ordering
of clauses in different domains (see her Ch. 4.2; more on this in § 3
The overall tendency of her approach, however, is based on the
following claim :

(ill) Pervasiveness of pragmatic ambiguity

Given sufficient context, we can almost always force either a

content-conjunction reading or an epistemic-conjunction reading
on any pair of clauses conjoined by because; it is just harder to
find reasonable contexts for some readings than for others. <...>
[in addition to because] therefore, since, so, although, and
despite < ... > all show such multiple usage. (1990:77-8)

2. Objections and amendments

While 1 fundamentally agree with claim (I) One meaning, several uses (I
consider this to be the only promising way of dealing with connectors!),
and while 1 am basically sympathetic to the notion of "pragmatic
ambiguity", 1 see some serious difficulties in accepting the claims in (II)
and (ill).
238 Ewald Lang

The notion of "pragmatic ambiguity" is in need of refmement and

elaboration. To achieve this aim Sweetser's approach will be relativized

(N) a. challenging the ambiguity claim advocated in (IT) and (ill);

b. inspecting more closely the features based on which the
supposed levels are distinguished;
c. checking the consistency and exhaustiveness of the three-level

Pragmatics is not considered to be the sort of stop-gap device which

Sweetser purports it to be; rather, what Sweetser calls "a pragmatically
motivated choice of interpretation" (see (IT) above) will be shown to draw
heavily on structural information available from the conjoined clauses as
well as the connectors occurring in the conjunctions at issue. Specifically,
I will defend (and provide evidence for) the following counter-claims,
each of which has its bearing on the three points of criticism in (N). For
ease of reference, these claims will be prefixed by "C-". The rest of this
section will concentrate on (C-l), the others will be discussed in § 3.

(C-l) There are more level-specific structural differences in the way

in which clauses are conjoined than Sweetser's claims (I) - (ill)
would admit.

(C-2) The lexical items, i.e. the connectors that are used to connect
clauses on the three levels of interpretation, display more level-
specific distributional restrictions than the sample of cases
discussed by Sweetser would suggest.

(C-3) The three-level distinction proposed by Sweetser must be

supplemented by another level-provisionally called "textual
progression" or "discourse perspective"-which is not located
within the Sweetserian hierarchy of levels but cuts across them.

(C-l) emphasizes the role of structural information in determining the

interpretation of a given conjunction. Its import can easily be
demonstrated by taking up Sweetser's introductory set of examples. A
closer look at (la-c) immediately reveals that there is a whole range of
structural dimensions with respect to which the clauses that are linked
together by connectors differ. Spelling out (C-l) in detail means checking
the clauses of what Sweetser calls a "conjunction" for a possible display
Adversative connectors on distinct levels ofdiscourse 239

of some crucial structural features, e.g. those listed in (CI-I) to (CI-5)


(CI-I) Are the clauses of the same or of different clause type? (in
morphosyntactic terms of being declarative, interrogative, or
imperative sentences)

It is obvious that a content-level interpretation in the sense of

Sweetser's paragon example (la) is confined to clauses which share
declarative mood and thus may render STATEMENTS. Taking this as a
necessary condition for content-level interpretation implies that any
conjunction whose clauses differ in type is excluded from being
interpreted on that level and is thus confined to being interpreted on
epistemic and/or speech act level. 2
Sharing declarative mood is a necessary, though not a sufficient,
condition for a conjunction of clauses to be interpretable on the content
level. Consider again (lb) and a version of (1 c) in which the SPEECH ACT
of ASKING is not grammatically overt but concealed in a declarative:

(1) b. John loved her, because he came back.

c'. I am asking you what you are doing tonight,
because there's a good movie on.

This is the point where further criterial features enter the picture. One of
them, of course, is comma intonation. Due to its interaction with other
features the indicative role of comma intonation is more far-reaching than
Sweetser assumes. Below the level of sharing declarative mood there are
important indicators of clause-internal differences which interact with
comma intonation. Thus, in isolation, the first clause of (1 b) may render a
STATEMENT as does the first clause in (la). Both behave differently,
though, if we add an adjunct like presumably or a parenthetical like I
guess, which mark the hypotheticality or assumption status of the
assertion being made:

(1) a'. Presumably John came back because he loved her.

John, I guess, came back because he loved her.
b'. Presumably John loved her, because he came back.
John, I guess, loved her, because he came back.

While in (la') hypotheticality markers like presumably take scope over

both clauses, their scope in (1 b') is confined to the first clause. Thus the
240 Ewald Lang

function of comma intonation to mark the first clause as an independent

assertion is confirmed by its effect of marking scope boundaries. Similar
observations can be made for negation and other operators. Despite
looking alike on the surface, the first clauses in (la, Ia') and (Ib, Ib')
differ in scope determination as well as in intonation, which undoubtedly
both form part of core grammar, and hence provide distinct cues for the
interpretation to be assigned to the conjunction as a whole.
Thus despite the fact that the first clauses in (1a) and (1 b) are alike in
sharing declarative mood, they also differ. The first clause in (Ia) asserts
a proposition in the form of a factual statement thus allowing for a
content-level interpretation of the conjunction. The first clause in (1 b),
however, asserts a proposition in the form of an ASSUMPTION as
evidenced by comma intonation and characteristic scope properties. The
assumption status of the first clause is crucial to epistemic-Ievel
interpretation of the conjunction.
The next dimension in support of level-specific structure formation as
claimed by (C-l) is mentioned by Sweetser in discussing "iconic
ordering" in coordinate structures. However, the point at issue can be
extended to causal connectors and exploited in a more systematic way.

(C 1-2) Restrictions on the order of clauses with subordinators

(Ia) on the content level allows both orders-John came back

because he loved her and Because he loved her John came back, though
there is a difference that distinguishes backgrounded vs. focussed
material on the level of discourse-related information structure. In any
case, preposing the subordinate clause does not prevent a content-level
reading as shown by (2a). Coherence is maintained independently of the
order of clauses. However, as shown in (2b-c'), Sweetser's examples
(Ib) and (Ic) as well as their German equivalents do not allow of
preposed because/weil-clauses if they are to keep their respective
readings on the epistemic or speech act level (marked * and ??,

(2) a. Weil er sie liebte, kam John zuriick. [content]

'Because he loved her John came back. '
b. Weil er zuriickkam, liebte John sie. *[epistemic]
'Because he came back, John loved her. '
Adversative connectors on distinct levels ofdiscourse 241

b'. Weil er zuriickkam,

LIEBTE John sie vermutlich / denke ich. [epistemic]
'Because he came back,
John presumably / I guess LOVed her. '
c. Because there's a good movie on,
what are you doing tonight? ??[speech act]
c'. Because there's a good movie on, I'd like to ask you:
what are you doing tonight? [speech act]

Thus, unlike (lb), a conjunction like (2b) cannot be interpreted as

justifying the conclusion "he loved her" from the fact that "he came
back". What is lacking here is the sort of coherence that is rendered by
the original ordering. To ensure coherence in such cases requires
additional lexical or prosodic focus indicators, as is illustrated in (2b').
Similarly, while clause type difference and intonation break may serve as
compensatory cues, (2c) is still incoherent. An amended version of it
should read like (2c').
Sweetser recognizes the role of conjunct ordering in discussing
conjunctions with coordinators like and, or and but by stating that
pragmatically "the primary, independent conjunct precedes the secondary,
dependent conjunct" (1990:98). Yet, she simply ignores the ordering
effects in conjunctions with subordinators. The facts in (2b, 2c) above
show that (Cl-2) has to be taken as a condition on level-specific structure
formation, which in tum has serious repercussions on Sweetser's
ambiguity claim in (II)-(ill) above.
First, it is a grammatical category feature of subordinators like
because, since, etc. to allow for clause preposing while keeping their
interpretation constant-cf (2a). If this built-in property fails, as with
(2b) for example, this has to be acknowledged as a disambiguating
grammatical fact, too. It is on this basis that pragmatic interpretation
starts to work, contrary to the ambiguity claim in (II).
Second, unlike subordinators, coordinators like and, or, but (though
notfor or German denn) are grammatically unrestricted as to the order of
conjuncts and hence do not provide category-based grammatical
information that is taken up by pragmatic interpretation. Among other
things, it is the sequential ordering of the conjuncts that is submitted to
pragmatic interpretation (along the lines discussed by Sweetser).
Third, in view of this, we should expect coordinate conjunctions to be
less fertile, hence more ambiguous or vague, in providing grammatically
determined information on which, following Sweetser, the "pragmatic
choice of the' correct' interpretation" may draw.
242 Ewald Lang

In what follows it will be shown that this assumption is by no means

justified. Due to space restrictions the further pursuit of (C-I) to (C-3)
will be confined to coordinate conjunctions whose clauses are linked by
adversative connectors. Moreover, we will employ German data3 and
present arguments which draw on results from current work being carried
out in the connector project at the Institut fur deutsche Sprache
Mannheim (for details see HdK [=Pasch et al. (forthcoming)]).

3. Adversative connectors

3.1. How ambiguous are aberlbut-conjunctions?

From the set of about 30 lexical items that may be subsumed under the
label "adversative connector" in German we will select only those needed
to illustrate (C-I) to (C-3).
Let us first turn to coordination involving the connector aber, which
(like but in English) is commonly considered as the standard adversative
connector. Compared with connectors such as trotzdem ('nevertheless'),
hingegen ('whereas'), or dennoch ('nonetheless', 'all the same'), aber is
much less restricted regarding distribution and hence to a much higher
degree prone to "pragmatic ambiguity". Nevertheless, the supposed
ambiguity of aber-conjunctions is drastically reduced by cues which
emerge from structural features of the coordinate clauses in the domain of
sentence structure. In addition to CI-I and CI-2, the relevant features

(CI-3) Matching or independently varying syntactic clause structure

(CI-4) Parallel or independently varying information structure

(C 1-5) (Un)availability of clause-internal deletions like "gapping",

"right node raising", etc. (Lang 1991, see also th. 4 below).

Let us now look at some data that are representative of the range of cases
to be taken into account.

(3) a. Dein Vortrag war sehr lang, aber das soli keine Kritik sein.
'Your talk was rather lengthy, but this is not a criticism. '
Adversative connectors on distinct levels ofdiscourse 243

b. Wir sind mit dem Manuskript in Verzug,

aber welcher Autor halt schon Termine ein?
'We are late with the manuscript,
but what author keeps the dead-lines?'
c. Willst du heute etwa ins Kino-aber ich komme nicht mit.
'Do you want to go to the movies today-but I'm not coming

All examples in (3) are easily recognizable as cases to be interpreted on

the speech act level. Just recall the relevant indicators mentioned in (C I-
I), namely "clause type difference" - cf (3b, 3c), "intonational break" -
cf (3a), and what was said about "independent clause-internal structures"
in CI-3 to CI-4 (i.e. lack of parallelism as to constituency and topic-focus
articulation). Just as in (lc) above, it is the conjunct-internal differences
which prevent the examples in (3) from being interpretable on the content
level and qualify them for speech act level interpretation. Generally
speaking, the reason is this: aber/but marks what is rendered by the
second clause as being in contrast with inferences that might be drawn
from the speech act that is performed by uttering the first clause. In short,
what counts are the conjunct-internal differences in terms of C1-1 to C1-
4. If such differences are missing, more effort is needed to determine the
respective levels of interpretation. Consider the following examples,
which Sweetser tends to locate at the epistemic level.

(4) a. Hans hat Unmengen von Zigarren gehortet,

aber er raucht gar nicht.
'Hans has hoarded an obscene number of cigars,
but he doesn't even smoke'
b. John keeps six boxes ofpancake mix on hand,
but he never eats pancakes. (cf Sweetser (1990: 100»

Sweetser comments: "But presents two conjuncts which clash with each
other in some way ... At the epistemic level, the available premises may
clash with an apparently necessary conclusion" (1990: 100). The first part
of the quote is unproblematic; but what exactly is "epistemic" about the
entities connected by aber/but? (4a, b) instantiate the well-known type of
"denial of expectation" interpretation of but-conjunctions (Lakoff
1971:133). In both cases, the clauses are mutually independent, hence
semantically compatible STATEMENTS. Given this, (4a, b) qualify as
coordinations of statements at the content level with the additional
requirement that the interpretation involves inferring an ASSUMPTION
244 Ewald Lang

which is in contrast with the two statements being presented side by side.
This analysis reduces the "epistemic" aspect of (4a, b) to the inferences to
be drawn and thereby does justice to the structurally determined level-
distinction. The same point was made in Foolen (1992:83).

3.2. Problems with interpreting but/aber at the content level

The awkwardness of Sweetser's use of the notion "epistemic" is

furthermore revealed by the stand she takes on cases of what since Lakoff
(1971:133) is commonly known as "semantic opposition" but. Sweetser
(1990:103) claims: "The but in (56) [= (5a) below] does indeed indicate
contrast: an epistemic contrast between two semantically opposed
propositions" (italics mine - EL). From this we may conclude that (5b)
lacks such an epistemic contrast, other things being equal. Again, (5b)
with and is a clear-cut case of content-level conjunction.

(5) a. John is rich but Bill is poor. [= Sweetser's (56), 1990:103 ]

b. John is rich and Bill is poor.
c. John is rich, but Bill is rich, too.
d. *John is rich but Bill is rich.

What we are left with is the problem of where to locate (5a): it does
indeed indicate a different contrast from (5b). I doubt, however, whether
this contrast can be accounted for in terms of the "epistemic level"
without rendering this notion completely void. I would contend that the
alleged "epistemic" difference between (5a) and (5b) is that the
interpretation of (5a) requires some additional inferencing which the
interpretation of (5b) does not. Thus, (5a) and (5b), like (5c) when
compared with (5d), clearly differ as to the conditions they impose on the
context in order to get interpreted properly.
According to an observation first made by Bellert (1972), the
interpretation of but/aber-conjunctions involves at least two semantic
contrasts to be read off or inferred from the conjuncts. (5d) contains the
overt contrast John-Bill but fails to provide cues to reconstruct the
second contrast needed, i. e. an INFERRED ASSUMPTION which is in
opposition to the STATEMENT Bill is rich. This pragmatic insufficiency
has its grammatical counterpart: (5d) can be considered deviant because
it does not allow a contrastive stress pattern - just try and you·will see'
Moreover, contrary to most work done in the field, we claim that the
"semantic opposition" (more exactly, the occurrence of antonymous
Adversative connectors on distinct levels ofdiscourse 245

predicates in logically compatible STATEMENTS) in (Sa) is not a crucial

feature of this type of conjunction (as shown by (5c) which lacks
opposition); and that this opposition is by no means the carrier of the
contrast indicated by aber/but. Instead, the contrasting entity has to be
inferred from outside the coordination. However, the opposition in (5a) as
well as its absence in (5c) serve as structural clues to reconstructing it.
We will return to this in 3.3 below.
Now, if (5a) cannot be located on the epistemic level, why not
consider it a content-level conjunction - very much like (5b)? Sweetser
(1990:103) rejects this idea by stating: "I have not been able to unearth
any indubitable content-conjunction examples with but" and by offering
"as a plausible explanation for the use of but in only two domains"
[epistemic level and speech act level - EL] the following consideration:
"what does it mean to say that A and B "clash" or "contrast" in the real
world? How can discordance or contrast exist outside of the speaker's
mental concept of harmony or non-contrast?"
Sure enough, the contrast induced by aber/but is not something to be
looked for in the real world but something to be established by the
interlocutors in assessing what, by way of coordinate conjoining, is
presented as co-existing in the world. But this does not prevent aber/but-
conjunctions from being interpretable at the content level, that is, as
conjoining two STATEl\1ENTS which lay the ground for additional
inferencing. Sweetser's problem with interpreting but at the content level
is caused by confusing two notions which, though not unrelated, have to
be kept distinct: inference (whether a proposition is asserted or inferred)
and epistemic status (whether a proposition forms the content of a factual
The alternative approach suggested here tries to avoid this confusion.
It draws on the fact that the interpretation of an adversative conjunction
involves relating the conjoined clauses to an ASSUMPTION which-if not
available in explicit terms (see (V) below)-has to be inferred from
outside the conjunction. Based on these considerations it is posited that

(V) (a) adversative (and probably also concessive) connectors inher-

ently contain pointers to previous information available from
the context, and due to this
(b) adversative conjunctions necessarily involve some back-
tracking that may well go beyond the domain of sentence
structure and operates on (what may be called) the level of
"textual progression" or "discourse perspective".
246 Ewald Lang

Following (C-3) it is this level which provides the appropriate means to

interpret adversative conjunctions, and therefore we consider it a
necessaIy supplement to Sweetser's approach. Note that the level of
"discourse perspective" must not be understood as an additional level
within, or on top of, the Sweetserian hierarchy, but rather as embracing
the three levels discussed so far as grammatically determined patterns of
discourse formation. Recall that C1-1 to Cl-5 apply to sentence structure
and thus clearly contribute to discourse formation and comprehension.

3.3. Adversative conjunctions and the division of labour between

grammar and pragmatics

This section will provide an outline of the way in which sentence and
discourse level interact in determining the interpretation of adversative
conjunctions. To begin with, note that the semantic contribution of
aber/but in combining two clauses into a conjunction is twofold. In a
nutshell, it may be conceived of as:

(VI) (a) conjoining semantically compatible and non-inclusive

propositions that-depending on the number of structural
parallels between the conjuncts-may be bundled up or
enumerated as instances of a "common integrator" (Lang
1984, 1991, in print; see also fn. 4 below);
(b) indicating that the assertion rendered by the second clause
is in contrast to an ASSUMPTION that either may be read
off, or must be inferred from, previous information
(Braul3e 1998).

(VIa) states what aber/but share with und/and; therefore (VIa) is the
basis for (Sa) and (Sb) having identical truth conditions. This is a feature
of aber/but that probably nobody will deny. (VIb) is what separates
aber/but from undland, and hence (VIb) is what causes (Sa) and (Sb) to
have distinct conditions of use as regards semantic coherence. As such,
(VIb) is perhaps not as uncontroversial as (VIa), so it may be worth
commenting on.
What has been stated in (VIa + b) should not be understood as
amounting to "but = and + pragmatics". This would defmitely miss the
target. The conditions of use of aber/but and other adversative connectors
are not simply a matter of "pragmatically motivated interpretation" (as
Sweetser suggests); rather they are rooted in the lexical semantics as well
Adversative connectors on distinct levels ofdiscourse 247

as in the syntactic category features of these connectors. On the other

hand, I do not deny the role which contextually induced pragmatic
interpretation inevitably has to play in specifying the "correct" reading
which a conjunction of clauses eventually gets assigned. I differ from
Sweetser, however, by claiming that in order to account for (what she
calls) "pragmatic ambiguity", the division of labour between grammar
and pragmatics has to be redefined. Regarding aber/but as characterized
in (VIb), this division of labour may be determined as follows:
That part of (VIb) which reads "indicating that the assertion rendered
by the second clause is in contrast to an ASSUMPTION" has to be spelled
out primarily in terms of grammatical structure formation (including C1-1
to CI-5), while the other part, which reads "an ASSUMPTION that either
may be read off, or must be inferred from, previous information", is
where pragmatics proper enters the picture.
It seems reasonable to consider the interpretation of an adversative
conjunction as involving a search device which is determined by a
specific source-target relation. Triggered by the lexical indication that a
contrast must be established, the device takes the assertion rendered by
the second clause as the source from which to look for an appropriate
ASSUMPTION that meets the condition of contrast, i.e. the target. Though
the search device is to some extent guided by grammatical structure, as
claimed by (Cl-l) to (CI-5), it is inevitably accompanied and
supplemented by pragmatically based reasoning. So the part to be played
by pragmatics proper is to provide supplementary means to find out what
the target, i.e. the INFERRED ASSUMPTION, is.
Note that by dividing up (VIb) between grammar and pragmatics as
suggested afore, it is not an accident that the notion "ASSUMPTION"
figures in both, and rightly so. The INFERRED ASSUMPTION forms the
bridge between grammar and pragmatics in interpreting adversative
conjunctions. Almost all studies of adversative connectors have attempted
to account for it in one way or another (for a survey see Rudolph 1996).
What is new is the way to tackle it as proposed in (VIb).
Source and target are both propositional but differ as to their epistemic
status. That the target (i.e. the proposition which the second clause is put
into contrast with) has to be an ASSUMPTION, not a factual STATEMENT,
can be seen if, for example, (4a) is extended by a clause that renders the
target in explicit terms, cf the bold-faced additions in (6):
248 Ewald Lang

(6) a. Hans hat Unmengen von Zigarren gehortet,

aber er raucht gar nicht. [= 4a]
'Hans has hoarded an obscene number of cigars,
but he doesn't (even) smoke'
b. Hans hat Unmengen von Zigarren gehortet,
*folglich raucht er, aber er raucht nicht.
' ... hence he smokes, but he doesn't smoke'
c. Hans hat Unmengen von Zigarren gehortet,
also konnte man vermuten, daB er raucht,
aber er raucht nicht.
, ... so one could assume that he smokes,
but he doesn't smoke'

If the supposed target of contrast is expressed in the form of a

FACTUAL STATEMENT, the result is a contradiction-cf (6b), but if it
appears with overt markers of ASSUMPTION status as in (6c), the inter-
pretation of the entire construction is consistent. This is what follows
from the grammar part of (VI), which requires the target of contrast to be
Let us now turn to the pragmatic aspects of (VI). If the clauses
conjoined by aber/but both represent STATEMENTS (as is the case with
the majority of adversative conjunctions), the ASSUMPTION forming the
target of contrast has to be inferred. Here is where pragmatics comes
into play by providing the means to infer the target assumption from
previous information. Reconsidering the conjunctions in (4) and (5) above
will then amount to the following analysis.
In (4a/6a), the inference is obviously drawn by resorting to world
knowledge about smokers such that the STATEMENT rendered by the
first clause (Hans has hoarded an obscene number ofcigars) is taken as
a premise which by abductive reasoning makes it plausible to assume
"Hans smokes" and thus to yield a suitable target assumption. By the
way, the fact that this inference is not based on deductive but on
abductive reasoning provides additional evidence of the ASSUMPTION
status of the target.
In (Sa-d), repeated as (7a-d) below, the inferential steps leading to the
target are less clearly set out. (This is what caused the trouble for
Sweetser's approach, cf § 3.2 above.) How come?
The clauses conjoined by aber/but in (5/7) are clear-cut STATE-
MENTS, so the target of contrast has to be inferred from outside the
conjunction. Moreover, the clauses conjoined are absolutely parallel as to
their internal structure. So they cannot but be taken as instances of a
Adversative connectors on distinct levels ofdiscourse 249

"common integrator".4 The common integrator, which has to be

identified within the conjunction (see fu. 4 below) or else has to be
inferred from outside the conjunction, then serves as the location frame
within which the contrast is to be placed. Thus, in comparison to (4a16a),
the conjunctions in (5/7) require additional cues from the discourse level
in order to be assigned an appropriate interpretation.
The simplest way to illustrate these context conditions consists in
showing what sort of questions the respective conjunctions in (7) might
serve to answer. Suitable questions (added in 11...11 ) thus reveal the
distinct effects that are caused by the structural differences of the
conjoined clauses.

(7) a. John is rich but Bill is poor.

liAs for John and Bill, are they both rich? II
b. John is rich and Bill is poor.
II As for John and Bill, what about their income? II
c. John is rich, but Bill is rich, too.
II As for John and Bill, they aren't both rich, are they? II
d. *John is rich but Bill is rich.
II??? II

The structural differences are clearly brought out by the fact that none
of (7a-d) is replaceable with any of the others as an appropriate answer to
the respective question-context given in II. ..11. Assessing (517) once more,
we may say: in isolation and in written form these conjunctions are not
really ambiguous but are underspecified regarding the cues they need to
get interpreted.
Finally, we will look at those grammatical features of adversative
connectors that form the basis of the division of labour between grammar
and pragmatics just outlined. The most relevant point to note is this:
indicating a contrast in the way suggested in (VIb) is a semantically built-
in feature of aber/but which correlates with likewise built-in syntactic
properties of these connectors. In contrast to and, or, etc., adversative
coordinators like aber/but display the following features:
First, aberlbut allows for binary connection only. This is the basis for
marking the assertion following aberlbut as the source from which to
look for an appropriate CONTRASTING ASSUMPTION, that is, the target.
Second, aber Ibut is intrinsically asymmetric in marking what has
been called a "change of perspective" (Spooren 1989): for example, (4a)
or (7a) with the reversed order of conjuncts would not render the same
250 Ewald Lang

information on the level of discourse perspective, because source and

target are reversed.
Third, unlike und/and, which allows for scope ambiguities, aber/but is
a marker of scope boundaries: negation or other scope-taking operators
cannot take scope over aber/but, cf:

(8) Hans ist nicht dumm undfaul [ambiguous as to scope of negation]

'Hans is not stupid and lazy'

(9) Hans ist nicht dumm, aberfaul [scope of negation unambiguous]

'Hans is not stupid but he is lazy'S

With these features in mind let us now examine a couple of aber/but

conjunctions which display subtle but telling differences in the
information structure of the conjuncts.
To begin with, note that in written form (10) below is underspecified
as to its information structure and hence to the way it is linked with
previous discourse. However, when (10) occurs as a continuation of (11)
or (12), which (along the lines of (5/7) above) form distinct discourse
contexts which (10) has to match, we realize that (10) assumes
correspondingly distinct patterns of information structure shown in (11)
and (12), respectively. (In the following examples, SMALL CAPS indicate
main stress, I and \ contrastive stress, labeled square brackets topics and
foci, respectively.)

(10) Hans ist krank, aber seine Frau geht arbeiten.

'Hans is off sick but his wife is at work'

(11) II Wie geht's den beiden? II 'How are the two doing?'

a. Tl[/Hans] F[ist KRANK], aber T2[seine Frau\] F[geht ARBEITEN]

'Hans is off sick but his wife is at work'

b. T[[Hans] und [seine Frau]] sind F[[IKRANK] bzw. [auf ARBEIT\]]

'Hans and his wife are off sick and at work, respectively'

(12) II Was regt dich so auf?11 'What's upsetting you?'

a. F[[Hans] ist KRANK], aber F[[seine Frau] geht ARBEITEN]

'Hans is off sick but his wife is at work'
Adversative connectors on distinct levels ofdiscourse 251

b. F[[Hans] ist KRANK], dennoch F[[seine Frau] geht ARBEITEN]

'Hans is off sick, nevertheless, his wife is at work'

c. F[Obwohl [Hans] KRANK ist], F[geht [seine Frau] ARBEITEN]

'Although Hans is off sick, his wife is at work'

Let us take a closer look at the way in which information structure at

sentence level and at discourse level interact. The contexts adduced in
(11) and (12) each form a discourse topic whose internal structure
determines the way in which it is elaborated by the sentences following it.
The question in (11) marks the NP den beiden 'the two' as a (con-
tinued) topic and specifies the VP as question focus, thus determining
focussed VPs as answers. In view of this, (I1a) shows a topic-focus
articulation which elaborates that of the discourse topic in a strictly
parallel way: the topic den beiden is resumed by the topical subjects
Hans and seine Frau, respectively; the VP question focus is instantiated
by the VPs ist krank and geht arbeiten, respectively. This mode of
parallel elaboration is evidenced also by two symmetry features: (i) the
clauses in (lIa) could be reordered without changing their links to the
discourse topic; (ii) (IIa) can be regrouped into a respectively
construction like (lIb). Hence, (11a,b) instantiate what might be called
"strictly parallel elaboration of a discourse topic".
In (12), where was ('What') refers to a whole proposition, the
question is such that it requires all-focus sentences as answers. Given
this, (10) is assigned the information structure in (I2a), which determines
the conjuncts as all-focus sentences at the discourse level. The
informational status of Hans and seine Frau is open: syntactically they
are sentence topics, but it depends on further discourse factors whether or
not they resume a discourse topic as they do in (I1a,b). Now, (12a) is a
predetermined elaboration of the discourse topic Was regt dich so auf?
('What's upsetting you?'), but there is no need for the elaboration to
spread over a conjunction of clauses-Hans ist krank ('Hans is off sick')
would suffice as an answer to (12) alone. In case the elaboration involves
a conjunction of clauses, these are not bound to the sort of parallelism we
observe with (11a,b). The relation between the all-focus clauses in (12a)
is not dependent on the discourse topic. Hence, (I2a) is what might be
called a "non-symmetric elaboration of a discourse topic. " This is
evidenced by (12b,c), which answer the question by couching the same
asserted propositions in concessive constructions. Concessive
constructions6 , no matter whether coordinate like (I2b) or subordinate
252 Ewald Lang

like (12c), are inherently asymmetric regarding the relation between the
clauses they link.
To summarize: due to their distinct information structure, (12a-c) are
not replaceable with (1Ia,b) while keeping the respective discourse
contexts constant, and vice versa.
Furthermore, the examples in (11) and (12) reveal an important
difference in the way interclausal aber/but relate to the discourse. In the
non-symmetric cases, the target from which the ASSUMPTION that
contrasts with the STATE1vffiNT in the second conjunct ('His wife is at
work') has to be derived is supplied by the first conjunct. So, in (12a-c) it
is the assertion of John's being off sick that triggers the relevant
inference: sick people need care, and spouses make good candidates for
this job.
In (11a,b), which form symmetric continuations of the given discourse
topic, the target that licenses the use of aber/but is not as easy to
reconstruct. Structurally, the question Wie geht's den beiden? ('How are
the two doing?') does not provide an appropriate target per se; all it does
is to license a list of and-conjoined items, cf (11 b). This is where
pragmatics enters the picture. In interpreting (1Ia), we face the following
situation: if searching the wider context for a suitable target fails, we are
left with no alternative but to take the first conjunct Hans ist krank
('Hans is off sick') and generate from it the implicature that the news
about his wife will be of similar quality.
Given that, as claimed in (VIb), aber/but has the built-in feature of
triggering an inference, the notion of a but of "semantic opposition"
(introduced by Lakoff (1971» seems to be misconceived: semantic
opposition per se never suffices to license the use of aber/but.

4. Concluding remarks

Sweetser's three-level approach has proved to be valuable as a heuristic

means in sorting the various interpretations that conjunctions can be
assigned. However, in order to serve as a means of explanation, it needs
to be modified in some respects.
Regarding the level of grammatically determined sentence structure,
Sweetser's level-distinction can be more precisely reconstructed in terms
of criteria like Cl-l to CI-5, among which Cl-l (sameness or difference
of clause type) is the most crucial one. Given this, we might reconstruct
what should be kept of Sweetser's three-level distinction of conjunction
interpretation in the following way:
Adversative connectors on distinct levels ofdiscourse 253

(VII) a. Content-level interpretation is reserved for conjunctions of

declarative clauses that render STATEMENTS (i.e. entities
to be related to the world via truth-conditions); see (la, 3a,
b. Epistemic-Ievel interpretation is confined to conjunctions of
declarative clauses, of which at least one is marked as
rendering an ASSUMPTION; see (lb',lb-i, 1b-ii, 1b-iii,
c. Speech act level-interpretation is available for conjunctions
of non-declaratives (with reservations about or-con-
junctions like (2» as well as for conjunctions of clauses
differing as to type but with one clause being marked as
rendering a SPEECH ACT; see (lc, 1c', 1c-i, 2c, 2c', 3a-c).

In addition, Sweetser's approach has to be modified in a way that

would enable it to account for contextual requirements imposed on the
interpretation of adversative conjunctions. The level that is assumed in
(C-3) to do this is the all-embracing level of "discourse perspective". This
is the level where (a) the information structure of sentences or clauses
interacts with that of the discourse containing them-as exemplified by
the cases in (7) and (10)-(12); (b) where the peculiar syntax and
semantics of adversative connectors can be accounted for in a promising
way; (c) where the division of labour between grammar and pragmatics
can be defined - cf (VI a+b).
Pragmatics comes in when grammar does not provide sufficient cues
to get an adversative conjunction properly interpreted.


* This paper is dedicated to Ekkehard Konig on the occasion of his 60 th

I wish to thank the editors for useful comments on an earlier draft of this
paper and Chris Wilder for improving my English.
1. Note that Sweetser uses the term "conjunction" for the syntactic operation of
clause-linking without distinguishing coordination and subordination. In
order to avoid confusion with the word class "conjunctions" I will call the
latter "connectors" (subdivided into "coordinators" and "subordinators") and
the constituents they link "conjuncts"~ otherwise I will follow Sweetser's
2. It may be noted in passing that sameness of clause type with non-declaratives
raises some problems concerning the conjoining of clauses on the speech act
254 Ewald Lang

level. I will take for granted that pragmatically (i) is asymmetric, while (ii) is
symmetric regarding the alternatives presented:
(i) Give me liberty or give me death! [= Sweetser's (46), p. 98]
(ii) Give me a salami sandwich or a hotdog!
What needs to be clarified, however, is this: is it justified to consider (ii) as
an or-{;onjunction of two speech acts? Speech acts normally resist being
presented as alternatives.
Another open question is whether we are to take (ii) as a conjunction of two
speech acts or rather as a single speech act of request which comprises an
alternative of how the request might be complied with?
Again, these questions show that a closer examination of the internal
structure of the conjuncts seems to be unavoidable.
3. The advantage of illustrating the issue at hand with German data consists in
the fact that aber is an unambiguous adversative connector. It does not share
the range of additional uses and readings that are covered by but, as in
German these are rendered by separate lexical items like sondern, auf3er etc.
Analysing but-constructions in comparison with aber-constructions thus will
help us to pick out only those having adversative readings.
4. Without going into details (see Lang 1984, 1991, in print), let me just point
out that the derivation of the Common Integrator (CI) of a given conjunction
(a) is sensitive to surface structure parallels shared by the conjoined clauses,
and (b) can best be illustrated by means of a (multiple) question to which the
given conjunction would serve as an appropriate answer.
So while (i)-(iv) share the propositional formula "someone transfers
something to someone" as constituent part of their respective Cis, the
specific make-up of the latter rests on the information structure of the
clauses, specifically on the (non-)coreference of the items instantiating the
argument places. (Non-)coreference is indicated by (non-) coindexing.
Coreference requires destressing: if e.g. the beast is coindexed with the dog
the former serves as an anaphoric epithet for the latter.
(i) Grandpa will bequeath the dogk to his grandchildreni and
Grandma (will)(bequeath) the beastv*k to us j/*i
CI: WHO will bequeath WHAT TO WHOM?
(ii) Grandpa will bequeath the dogk to his grandchildreni and/but
Grandma will sell the beastk/*l to us i/*j
CI: WHO will transfer the dogk to USi IN WHAT WAY?
(iii) Grandpa and Grandma will bequeath the dogkand the beastv*k
to their grandchildreni and to us j/*i, respectively
CI: WHO will bequeath WHAT TO WHOM?
(iv) Grandpa will bequeath, and Grandma will sell, the dog to us
CI: WHO will transfer the dogkto USi IN WHAT WAY?
Note that the internal structure of the conjuncts in (i) allows for regrouping
the parallel constituents into a respectively construction like (iii) which leads
to the same CI as (i) but is, for instance, unavailable for (ii). On the other
hand, (ii) can also be rendered by a right node raising construction like (iv)
which leads to the same CI as (ii) but is unavailable for (i). So (iii) and (iv)
may serve as overt syntactic diagnostics for information structure and CI
Adversative connectors on distinct levels ofdiscourse 255

differences that seem to be concealed when conjunctions like (i) or (ii) are
presented in written form only.
5. (8) and (9) point to an important difference between German aber and
English but regarding scope determination under negation. Since German
adversative coordinators for contrast (aber) vs. correction (sondern) are
lexically distinct, their scopal behaviour is built-in and need not be indicated
by other means-ef. (i, ii). English but-constructions, however, have to
render the distinction of contrast vs. correction under negation by distinct
conditions on deleting non-focussed material in the second clause. Thus, but-
constructions like (iii) expressing contrast disallow, whereas but-
constructions like (iv) expressing correction require, the deletion of repeated
material in the second conjunct (for details see Lang (1991).
(i) Hans ist nicht dumm, aber (er ist) faul [contrast]
(ii) Hans ist nicht dumm, sondem (er ist) faul [correction]
(iii) Hans is not stupid but *(he is) lazy [contrast]
(iv) Hans is not stupid but (*he is) lazy [correction]
The same point can be made to explain the difference in the following cases:
(v) John ist reich, aber Bill (ist) arm.
(vii) John is rich but Bill *(is) poor.
(vi) John ist reich, aber Bill nicht lauch.
(viii) John is rich but Bill *(is) not Itoo.
Despite the differences shown, aber and adversative but both mark scope
boundaries in preventing propositional operators from taking scope over
6. The relationship between aber-conjunctions and concessive constructions
cannot be tackled here, but see the contributions of Barth and of Crevels (in
this volume).
Another point that deserves mentioning is the fact that German aber (unlike
English but) is not confined to interclausal positions, but can also float in the
second clause (this is why in HdK [=Pasch et a1. (forthcoming)] aber has
been classified as a "connective particle"). At first glance, this distributional
flexibility seems to support the view that aber is most prone to "pragmatic
ambiguity". It can be shown, however, that being allowed to float within the
second clause is what enables aber to syntactically mark the information
structure relation of the second clause with respect to the discourse topic as a
non-symmetric one, similar to (12a-c) above:
(12) d. F[[Hans] ist KRANK], F[[seine Frau] geht aber ARBEITEN]
'Hans is off sick, his wife is, nevertheless, at work'
The prosodic details of contrastive vs. concessive clause linking are being
investigated in a current DFG project of mine which forms the source of the
data discussed here.
256 Ewald Lang


Bellert, Irena
1972 On certain properties of the English Connectives and and but. In:
Senta Plotz (ed.), Transformationelle Analyse, 327-356. Frankfurt
am Main: Athenaum.
BrauBe, Ursula
1998 Was ist Adversativitat? aber oder una? Deutsche Sprache 26:138-
Foolen, Ad
1992 Review of Sweetser 1990. Lingua 88:76-86.
HdK [= Pasch, Renate, Ursula Brausse, and Eva Breindl]
(to appear) Handbuch der deutschen Konnektoren. 2 vols. Ms. Mannheim:
Institut flir deutsche Sprache.
Kortmann, Bernd
1997 Adverbial subordination. A typology and history of adverbial sub-
ordinators based on European languages. (EALT 18.) Berlin/New
York: Mouton de Gruyter.
Lakoff, Robin
1971 If's, and's and but's about conjunction. In: Charles J. Fillmore and
D. T. Langendoen (eds.), Studies in linguistic semantics, 114-149.
New York: Holt, Rinehart & Winston.
Lang, Ewald
1984 The semantics of coordination. (SLCS 9.) ArnsterdamlPhila-
delphia: Benjarnins.
1991 Koordinierende Konjunktionen. In: Semantik/Semantics, Art. 26,
597-623. (HSK 6.) BerlinlNew York: Mouton de Gruyter.
(in print) Die Wortart "Konjunktion". In: Lexicology/Lexikologie, Art. 88,
(HSK 17.) Berlin/New York: Mouton de Gruyter.
Raible, Wolfgang
1992 Junktion. Eine Dimension der Sprache und ihre
Realisierungsformen zwischen Aggregation und Integration.
Heidelberg: Winter.
Rudolph, Elisabeth
1996 Contrast: adversative and concessive relations and their
expressions in English, German, Spanish, Portuguese on sentence
and text level. (Research in text theory 23.) BerlinlNew York:
Mouton de Gruyter.
Spooren, Willibrordus
1989 Some Aspects of the form and interpretation ofglobal constrastive
coherence relations. PhD Dissertation, University of Nijmegen.
Sweetser, EveE.
1990 From etymology to pragmatics. Metaphorical and cultural aspects
ofsemantic structure. Cambridge: Cambridge University Press.
Viewpoints and polysemy: Linking adversative
and causal meanings of discourse markers

Scott A. Schwenter

This article examines the use of a connective discourse marker in Spanish

(si), which can be used to convey both adversative and causal meaning, de-
pending on the discourse context. The semantic/pragmatic link between the
two uses of the connective, which are considered distinct, but overlapping,
polysemies, is mediated by viewpoint. The refutational meaning of the ad-
versative polysemy is paralleled by the high-scalar meaning of the causal
polysemy. This parallelism results from the transfer of dialogical features
from the adversative use, typically found in dialogues, to the causal use,
typically found in monologues. The study provides a general framework for
analyzing the role of viewpoint in cases of polysemy.

1. Introduction

The representation of more than one viewpoint in a single lexeme or con-

struction can be illustrated by a multitude of forms in English or any other
language. A prototypical example of how multiple viewpoints or "voices"
get encoded into linguistic structure can be seen clearly by looking at one
of the "e" relations forming the theme of this book: concession. In an
English concessive sentence like the following,

(1) Even though it rained, they played the baseball game.

what is strictly encompassed by the speaker's point of view is clear: the

rain did not affect whether the baseball game was played or not. But be-
yond this, another point of view is evoked which cannot necessarily be
ascribed to any particular individual, namely the expectation (based on
knowledge of "how the world works") that the rain would have been suf-
ficient cause for the baseball game not to be played. Though not explicitly
stated in (1), this expectation, and the point of view from which it arises,
is taken into account by the speaker, and grammatically indexed by the
concessive ('even though') construction.
In this paper, I intend to show that taking multiple viewpoints into ac-
count is a useful practice for establishing connections between distinct
258 Scott A. Schwenter

polysemies of individuallexemes. I will focus in particular on adversative

(contrastive) and causal polysemies of the Spanish discourse marker si,
which derives from the conditional marker si 'if (and is unrelated to to
the sf of affirmation meaning 'yes'; indeed, as will be shown below, its
meaning is often quite close to 'no'). Although I will be interested pri-
marily in the synchronic connections exhibited by these polysemies, such
an analysis is not necessarily restricted to synchrony. In fact, this kind of
analysis has been advanced for some time now with regard to the rise and
structure of conditional constructions, which are claimed to present in
condensed form (i.e. in a complex sentence) a discursive sequence con-
sisting of a (yes-no) question (speaker A), an implied affirmative answer
(speaker B), and a stated consequence following from B's answer
(speaker A) (Haiman 1978; Jespersen 1940).
Although the title of this paper makes explicit the connections to two
of the four "C" relations that make up the theme of this volume-contrast
(of which I take adversativity to be a subtype; cf Schwenter 1999a: Ch. 4
and below) and causality-the full story of what I am about to present
actually brings another of the "C" relations into the picture, namely condi-
tionality. I will focus most of my attention on discourse marker (DM)
uses of Spanish si, which, in its most canonical use, is a conditional
marker meaning 'if, marking "non-assertion" of the protasis, as in the
constructed example (2a). With other functions, however, it is found in
assertions with declarative sentence form as an adversative DM (2b), akin
to certain uses of 'but', or as a "justificatory" DM or connective (2c-d),
akin to certain uses of 'because' (the latter two often, but not always, with
exclamatory intonation):

(2a) Si traen el pastel, me 10 como.

'If they bring the cake, I'll eat it.'

(2b) A: jComete un poco de pastel!

'Eat some cake!'
~ B: jSi ya he comido!
'But I've already eaten!'

(2c) Date prisa. jSi se va el autobus!

'Hurry up. (Because) the bus is leaving!'

(2d) Juan no viene a la fiesta. Si su madre no 10 deja.

'Juan won't come to the party. (Because) his mother won't let
Viewpoints and polysemy 259

In the constructed examples (2b-d) si does not contribute to the pro-

positional content of the utterance-indeed, it could be removed without
affecting this content in the least-but rather signals how the utterance in
which it appears relates to the preceding utterance (cf Fraser 1988). In
this function, then, si is like many other DMs, which primarily express
procedural as opposed to conceptual meaning (Blakemore 1996) as well
as a "metatextual" comment on the text being constructed (Traugott
1999). In (neo-)Gricean terms, si in each of these examples conveys a
conventional implicature which is non-truth-conditional but also uncan-
celable; the specific semantic content of this conventional implicature, as
well as its pragmatic effects, are what I intend to flesh out in the discus-
sion below. Throughout the paper, I will take a neo-Gricean perspective
on implicature, and in particular on scalar implicature (cf Hom 1989,
among others).
The remainder of this paper is structured as follows. In section 2 I
provide an overview of the analytical framework used throughout. Sec-
tion 3 analyzes uses of si as an adversative DM. Section 4 analyzes uses
of si as a causal DM in two different contexts, exhortative and epistemic,
and summarizes the polysemies of DM si. Section 5 offers concluding
remarks, including ways of extending such an analysis to some English
forms. The data to be analyzed in the paper come principally from a 40-
hour corpus of colloquial conversational Spanish collected by the author
in Alicante, Spain, and indicated by [ALC]. The sources of other natu-
rally-occurring examples will be given in the text, and constructed exam-
ples will also be used when necessary.

2. Analytical framework

Let me begin by stating explicitly my position regarding the concepts of

contrast and adversativity. Contrast is not a strictly linguistic notion, but
rather a general cognitive ability linked to structures of expectation. I de-
fine contrast as "[t]he perception of difference, at some level and by any
means possible, between two entities comparable on some other dimen-
sion" (Schwenter 1999a: 126). Adversativity, on the other hand, is a
strictly linguistic notion. It involves "[c]ontrast between different points of
view as these are constructed in language use (Schwenter 1999a: 127,
emphasis in original). Contrast can be manifested indirectly in language,
for example in a sentence like John is short and he's a good basketball
player. Here, the conjunction does not explicitly signal contrast, but hear-
ers familiar with the link between height and basketball-playing ability
260 Scott A. Schwenter

will perceive it nevertheless. Adversativity differs in that it typically pro-

vides explicit marking of clashes between different viewpoints, as in John
is short but he's a good basketball player, where the conjunction but
forces an interpretation in which the speaker asserts (supposed) incom-
patibility between shortness and basketball-playing ability.
Now, although there are a number of ways to conceptualize the inter-
play between viewpoints in language use and linguistic structure (for
some examples, see Ducrot 1984, 1996), in this paper I will utilize a
rather basic schema for analysis. Following Roulet (1984), I distinguish
between monologues (one physical speaker) and dialogues (two physical
speakers), as well as between monological (one "viewpoint") and dialogi-
cal (two "viewpoints") discourses. These two sets of distinctions can be
merged to yield four different combinations of speaker/viewpoint struc-
ture, as displayed in Table 1:

Table 1. Structure of discourse by speakers and "viewpoints"

No. of "Viewpoints"
No. of Speakers One Two
(Monolo~ical ) (Dialo~ical)
One (Monologue) Monologue/ Monologue/
Monological Dialo£ical
Two (Dialogue) Dialogue/ Dialogue/
Monological Dialogical

The distinction between monological and dialogical viewpoints in this

paper is closely tied to the argumentative function of utterances in their
discourse context, and in particular to whether the viewpoints are con-
strued as argumentatively parallel, i.e. oriented towards the same conclu-
sion, or as argumentatively opposing, i.e. oriented towards different con-
clusions. To illustrate what is meant by this, consider again the English
concessive sentence in (1) above. Here it is possible to isolate both the
stated viewpoint that the game was played despite the rain, and the under-
lying expectation that rain would cause the game to be cancelled. Thus, in
this case, a single speaker uttering (1) is simultaneously advancing two
viewpoints which are in some sense incompatible. Therefore, in accor-
dance with the schema in Table 1, a concessive sentence is a monologue,
yet expresses a dialogical viewpoint.
As the discussion below will make clear, the discreteness of the four
combinations of speakers and viewpoints illustrated in Table 1 is some-
what illusory. Instead, a gradient perspective on viewpoints appears nec-
essary, since the vestiges of indisputably dialogical interactions remain
Viewpoints and polysemy 261

palpable in meanings/functions that otherwise appear monological in

viewpoint. It is possible to determine, for instance, different degrees of
"addressee orientation" or, from another perspective, the degree to which
the meaning/function of a lexeme is based strictly in the perspective of the
speaker. This is not meant to imply, however, that the schema provided in
Table 1 is misguided. Indeed, it would have been much more difficult to
recognize the gradient nature of viewpoints without this starting point for
the analysis.

3. Si as an adversative DM

The adversative polysemy of si is typically found in dialogues, and is

always dialogical-it signals a confrontation of "incompatible" view-
points. The utterance introduced by si presents a "retort" (Montolio 1999)
to another speaker/viewpoint, and has a "refutational" (Schwenter 1998)
pragmatic value, indicating the exclusive relevance of the speaker view-
point. As a retort, the speaker's viewpoint is presented in the context of
another, competing viewpoint (often the immediately preceding utterance
of an interlocutor), and via inference excludes the competing viewpoint
from further consideration in the discourse.
To illustrate this process in action, consider the exchange in (3), taken
from the Valencia corpus of colloquial conversational Spanish (Briz [ed.]
1995). Here, a couple is engaged in a rather heated argument, and appear
to be on the verge of "breaking up" (§ = "latching" between two speak-
ers; / = .5 second pause):

(3) A: § mira yo te quiero// y cre- y creo que /0 SABES/// pero

NO/no puedo DEMOSTRARTELOJ 0 sea no no puedo dedi-
carte todo /0 que tit necesitas
'Look, I love youl/ and I thi- and I think that you KNOW it/II
but NO/ I can't DEMONSTRATE IT TO YOU, I mean I can't
give you everything that you need. '
~ B: pero si yo no te pido que me /0 demuestres
'but SI I'm not asking you to demonstrate it to me.'
[Briz (ed.): 1995: 83]

B's response specifically picks up on A's claim no puedo demostrarte/o

and contravenes an assumption inferrable from A's utterance: (A thinks
that) B wants A to demonstrate his love for B. In other words, what B is
refuting here is not any part of A's utterance per se, but rather his (prag-
262 Scott A. Schwenter

matic) presupposition about what B wants him to do (viz. demonstrate his

love for B).
The adversative nature of these si-marked utterances is often rein-
forced by the adversative DM pero, as in (3). However, such reinforce-
ment is not necessary for the adversative meaning to be conveyed, and si
is often used on its own to signal refutation of a competing viewpoint. 1 In
(4) si is used by M to introduce a refutation of (part of) the propositional
content derivable from A's utterance:

(4) [A and M want advice for their next vacation]

A: S~ .ha viajado por toda Europa.
'S has traveled all over Europe. '
~ M: iSi no ha salido de Espana!
'SI she hasn't been out of Spain!'

Here, A's position that S has traveled extensively is directly denied by M

via negation. Since A's utterance also conversationally implicates that S
would be a good person to ask for advice, M's retort functions, via impli-
cature, to refute the validity of this implicit line of reasoning.
Semantic contradiction as in (4) is not a necessary feature of si-
marked utterances. The refutation involved may be based in considera-
tions of relevance (in the Gricean sense), as in the next example, where
the rhetorical argumentation introduced by si is presented as the only one
relevant to the discourse situation.

(5) [R is watching A drink orange juice]

R: A, no te bebas el zumo tan rapido, que te va a doler la
'A, don't drink the juice so quickly, because your stomach
is going to hurt. '
~ A: Si tengo sed.
'SI I'm thirsty.'
R: Bueno, pues 10 que te de la gana.
'Fine, then (do) whatever you want.'
(ALC 7b.OI15.338)

The argumentative sequence of the first two turns in (5) can be summa-
rized as in (5'), where the parenthesized segments are conclusions left
implicit, but easily inferrable from the exchange.
Viewpoints and polysemy 263

(5') R: beber rapido ~ dolor de barriga (:. no beber rapido)

'drink quickly' ~ 'stomach ache' ( :. 'don't drink quickly')
A: tener sed ~ (heber rapido)
'be thirsty' ~ ('drink quickly')

It is important to notice that the utterer (=A) of the si-marked turn does
not dispute the "truth" of the proposition ("your stomach is going to
hurt"), but she does dispute its relevance and the relevance of the associ-
ated argumentative sequence within the discourse situation. Her response
to R signals the exclusive relevance of the "be thirsty" argument and, as a
consequence, justifies the conclusion "drink quickly".
Refutation with si may also be directed towards conversational impli-
catures that go beyond the literal meaning of the proposition, like the one
generated by A's utterance in (6):

(6) [B (age 6) looks strangely at a glass of a purple beverage]

B: i,Que es eso?
'What is that?'
A: Es el que te gusta a ti.
'It's the one (flavor) that you like.'
~ M: Si a ella Ie gustan todos los sabores.
'SI she likes all of the flavors. '

The cleft construction in A's utterance leads to an upper-bounded scalar

implicature from "it's the flavor that you like" to "it's the only flavor that
you like" (cf Atlas and Levinson 1981). M's si-marked utterance contra-
venes the upper bound implicated by A's utterance.
Finally, si can also head an utterance employed to object to more
strictly metalinguistic aspects of utterances, such as the pronunciation of a

(7) A: Manana tengo que dar clase, sobre la GENEOLOGiA.

'Tomorrow I have to teach a class, about geneology.'
~ R: Si es GENEALOGiA.
'SI it's genealogy.'

R's objection in (7) is to the appropriateness of A's pronunciation of the

word genealogia 'geneology', not to any propositional content derivable
264 Scott A. Schwenter

from A's utterance. The correction of A's pronunciation by R implicates

that A's rendering of the word was inappropriate.
As the preceding examples illustrate, si-marked utterances can intro-
duce a wide range of opposition types between two (or more) viewpoints.
This should not be surprising, since refutation as a discourse-pragmatic
strategy is not limited to denying and correcting the "truth" of prior
propositions. Like negation, it is also free to target "oppositions which are
created in the pragmatic aspects of utterance interpretation" (Ladusaw
1996: 323; cf Hom 1989), e.g. rhetorical argumentation, conversational
implicatures, and judgments of metalinguistic appropriateness.

3.1. "Hybrid" uses

The bulk of the examples of si presented in 3.1 above had an overwhelm-

ingly adversative flavor, since in each case two speakers were in "opposi-
tion" to each other. There are examples of other interactions which on the
surface appear to express a similar kind of adversativity, but on further
inspection reveal themselves to be subtly-yet crucially-different from
most of those seen above. The one exception among the preceding exam-
ples is (6), which is a "metapragmatic" (Schwenter 1999a) example ob-
jecting to a scalar implicature, but not to the propositional content or to
the argumentative orientation of the prior utterance. Examples like (6) are
therefore distinct because their interpretation is not so obviously constru-
able in strictly adversative terms, i. e. strictly as a case of opposition be-
tween (two) competing viewpoints. Another example of this type is (8):

(8) M: Creo que te han tintado MAL el pelo.

'I think they dyed your hair POORLY. '
~ B: iSi me 10 han tintado FATAL!
'SI they dyed it HORRIBLY!'

In (8), M makes a hedged assertion about the dye job done on B's hair,
and characterizes it as done mal 'poorly'. B' s response disputes this
characterization of the dye job; but instead of characterizing the dye job
positively, B strengthens the negative orientation already made explicit by
the adverb in M's assertion, stating that the dye job was donefatal 'hor-
ribly, lit. fatally,.2 Such an example expresses some degree of adversativ-
ity, since there is a clear correction of one speaker's evaluation by another
speaker's evaluation. At the same time, because the two speakers actually
Viewpoints and polysemy 265

AGREE on the overall implicated point-the dye job was not a good
one-the rhetorical purpose of each utterance in the exchange is parallel.
Here there is no ambiguity between a dialogical, adversative interpreta-
tion and a monological interpretation with parallel arguments. Rather,
both components of meaning combine to produce the overall interpreta-
tion of the example.
The adversativity expressed by a particular si-marked utterance can be
diminished even more when a "correction" of (some aspect of) an inter-
locutor's utterance cannot be identified:

(9) [Looking at a bicycle for a 4 year-old boy]

S: Mira, se puede bajar el asiento tambien.
'Look, you can lower the seat also.'
~ A: Si incluso se puede bajar mas.
,SI you can even lower it more. '

In the situational context of this example, S and A are inspecting the bi-
cycle, and S' s utterance is accompanied by the physical act of lowering
the bicycle seat a certain amount. A's utterance is in response to the (non-
locutionary) implication that the seat cannot be lowered beyond the point
to which S has lowered it. This is where the adversativity of the example
lies. Nevertheless, the contributions of both speakers are argumentatively
oriented in the same direction, each constituting positive assessments of
the bicycle in question, and implicating a conclusion like "we should buy
this bicycle" (which, in fact, they did). The co-oriented nature of the ut-
terances is made clear by the appearance of incluso 'even' in A's tum,
which marks the expressed proposition as a parallel, but stronger, argu-
ment for the aforementioned conclusion (which A took to be implicated
by what S said and did) (Kay 1990; Konig 1991).
The two preceding examples, along with (6) above, illustrate that,
even in dialogues, the degree of opposition marked by si is not fixed, but
sensitive to contextual factors, lexical relationships, etc. The "hybrid" na-
ture of each example is due primarily to the parallel argumentative orien-
tation of the asserted propositions: in (8) both the mal proposition and the
fatal proposition are directed towards a negative evaluation of B's hair,
while in (9) the propositions derivable from both utterances are being
used to evaluate the bicycle positively. The greater adversativity ex-
pressed in (8) as opposed to (9) results from the explicit correction of a
weaker predicate with a stronger one.
266 Scott A. Schwenter

4. Si as a "causal" DM

4.1. "Exhortative" contexts

DM uses of si where it functions to signal "connection" between two ut-

terances that form part of a complex discursive act in monologue, such as
(2c) and (2d) above, can be divided into two classes. The first is com-
prised of a paratactic sequence of two utterances, normally separated by
period intonation. Specifically, this class consists of an utterance express-
ing a command or exhortation which is "motivated" by a following utter-
ance prefaced by si. The second class consists of a similar two-utterance
sequence linking a speaker's subjective assessment with an argument
serving to justify that assessment. The distinction between the two classes
corresponds roughly to the now well-known division between "speech-
act" and "epistemic" causality (Sweetser 1990). Hence from a pragmatic
perspective they are distinct and, following Sweetser, I consider them to
be pragmatically ambiguous. Both differ from adversative DM si in that
they cannot be prefaced by pero 'but'. However, they are also similar in
th~t both are typically expressed in paratactic structures. This is the norm
for "speech-act" and "epistemic" causals, which assert the content of
their clauses. Both types are moreover opposed to "content" ("real-
world" or "direct") causals, which presuppose the content of their clauses
and tend to display features of subordination (Couper-Kuhlen 1996, inter
alia). In this section I examine the first class of uses, where "exhortation"
is the central pragmatic act being performed.
To the extent that any uses of si like those in (2c) and (2d) above have
been considered in the literature, the explanation for them has been as-
similated to dialogue-based, adversative uses like those in §3 above. Con-
sider for instance the following naturally-occurring example, from a re-
cent article analyzing the functions of "independent" si:

(10) [To the disgusted husband whose favorite soccer team has just lost
three goals to none]
No te preocupes, jhombrel, si eljUtbol no es importante.
'Don't worry about it, man! SI football [=soccer] isn't important.'
(Porroche 1998: 236)

In her analysis of this example, Porroche points out that the target of the
husband-directed utterance is not an actual utterance produced by the
husband, but rather "a belief that ... forms part of the mental universe of
the speaker, but which, at times, can also be attributed to the interlocutor"
Viewpoints and polysemy 267

(Porroche 1998: 236~ my translation). The second part of this explanation

is, of course, implicitly recognizing the multiple viewpoints discernible in
these examples: the wife believes that her husband thinks soccer is (very)
important, and attempts to provide a correction to his viewpoint. How-
ever, at the same time, the segment introduced by si can also be under-
stood in a causal sense, since it presents a piece of motivation for the
negative imperative No te preocupes 'Don't worry', and, moreover, goes
on to provide a reason not to worry.
The causality involved in examples like (10) above or (12) and (13)
below thus concerns the relationship between a command/exhortation and
motivating material which provides "just cause" for what the command
requests. In this "exhortative" use, there exists a clear overlap between si
and discourse connective uses of que (lit. 'that') which perform a similar
function (cf Garcia 1996: 49-52)-indeed, que could have been used in
place of si in (10) above. Consider the constructed (near) minimal pair in
(11 ):3

(IIa) Come, que no te queda casi nada.

'Eat, QUE there's hardly anything left.'

(11 b) Come. Si no te queda casi nada.

'Eat. SI there's hardly anything left.'

Either of these utterances could be uttered by a parent to a child who

is still sitting at the table with food in front of her, long after the rest of the
family has finished eating. Both markers preface an asserted proposition
that functions pragmatically as an argument in favor of carrying out the
action demanded by the imperative come 'eat', and both could be sup-
pressed, i. e. the causal relationship between the imperative come and the
declarative sentence that follows is inferrable without any intervening
connective element. Pragmatically, the difference between the two utter-
ances is that the second one with si is more "emphatic"~ it would be in-
terpreted as a stronger (more persuasive) attempt on the parent's part to
get the child to finish the food than the first utterance with que, which
presents the "motivation for eating" in a more neutral way, pointing out to
the child the fact that she has little food remaining, and possibly softening
the command to eat (cf Pons 1998). Put somewhat differently, the sec-
ond utterance seems to assume a priori stronger opposition from the child
towards eating-it strives to take into account the child's perspective and
presents a corrective measure to that perspective. In short, it displays dia-
logical features. Thus, though the propositional content of the two exam-
268 Scott A. Schwenter

pIes in (11) is equivalent, the version with si offers stronger motivation

for eating than the version with que.
Unlike most of the dialogue-based examples shown above, example
(11 b) would most likely not be interpreted as a reply to an explicit utter-
ance by an interlocutor. As a result, further reinforcement of si by the ad-
versative DMpero, a possibility in all of the adversative examples in §3,
would be odd in (11 b), unless the child had previously stated or implied
that there was an overabundance of food left on her plate (i. e. as justifica-
tion for not eating more). The (im)possibility of prefacing si-marked ut-
terances with pero thus constitutes a handy diagnostic for distinguishing
adversative uses (usually dialogues) from those with a more causal sense,
like the monologue example in (11 b).
A naturally-occurring example where eating was precisely at issue can
be seen in (12), where A (the mother) produces two utterances which
relate back to the command to eat in the first utterance:

(12) [Mother to her 18-month-old son, who is sitting at the dinner ta-
A: jVenga hombre! jCome!
'Come on man! Eat!'
A: te quedan dos bocaditos.
'there are two bites left. '
~ A: Si ya no te queda casi nada.
'SI there's hardly anything left. '

Here, both of A's final two utterances can be seen as trying to per-
suade the child to eat, or more accurately, to finish eating. Both of these
utterances can thus be considered "pragmatically subordinate" to the ini-
tial tum where the command is uttered; the si-marked utterance being the
"stronger" of the two. Indeed, it is interesting to note that if the proposi-
tional contents of A's latter two utterances are switched, the sequence
will become pragmatically anomalous, as shown in (12'):

(12') A: iVenga hombre! iCome!

A: Ya no te queda casi nada.
~ A: ?? Si te quedan dos bocaditos.

The reason for this is that the sequence introduced by si must be inter-
pretable not only as motivation for carrying out the action demanded by
the imperative, but also as the "strongest" piece of motivation among
Viewpoints and polysemy 269

those available. Here, the negative sentence explicitly minimizes the

quantity of food left, especially through the use of casi nada 'almost noth-
ing', thereby providing the strongest argument among those put forward
in favor of eating. By contrast, the sentence te quedan dos bocaditos
'there are two bites left' presents the quantity in positive terms, leaving
strictly to conversational inference the pragmatic connection between it
and the imperative come. What this type of example illustrates, then, is
that si in this use encodes a component of scalar meaning, in effect
imposing a ranking on propositions that are being used as motivation for
the action requested.
The dialogical nature of the preceding examples, though not linguisti-
cally explicit, does appear to be easily inferrable from the discourse con-
text. In each case, one can assume a reluctance or "opposition" on the
part of the addressee towards carrying out an action (e.g. eating) or, as in
(10), an emotional opposition to "not worrying" about the losing soccer
team. Moreover, the si-marked utterances are plainly addressee-directed,
since they attempt to motivate or persuade the addressee to carry out the
action specified in the command. But one might ask whether it is possible
to generalize this analysis to all "exhortative" examples. An example
which, at first glance, might not seem analyzable in this way is (13):

(13) B: jcono! aqui /e da e/ so/// jmecagiien /a put-!§

'Damn! the sun is shining here// son of a bi-!'
D: § i,ahora te enteras?/
((estam-)) una mierda ahi
'you just figured that out? «we-)) a piece of shit there'
~ C: mira! a/ Toni/ si /e sobra un pedazo dee- dee- de
hierba por ahi/ co/gando
'Look at Toni/51 he's got a piece of plant there/ hanging'
D: [dos dos]
'two two'
A: [i,a quien?J
C: en e/ bocadi//o//parece que //evas unn§
'in the sandwich// it looks like you've got a'
A: § jcono! es- son ajos tiernos
'damn! it's- they're green garlic'
[Briz (ed.) 1995: 68]

Here there is no expressed opposition to the command 'look at Toni' ut-

tered by C. However, this utterance represents an obvious and abrupt
270 Scott A. Schwenter

change in topic, as is evidenced by the exchange between B and D at the

beginning of the example. C's goal is to capture the other participants'
attention and redirect it towards Toni. It appears, then, that C's si-marked
utterance-presenting an argument in favor of looking at Toni-can also
be interpreted as attempting to overcome opposition, specifically as a way
of persuading the others to heed his command by refocusing their atten-
tion on Toni. Given the discussion above of si and que (which could re-
place si in [13]), C's choice of si to provide linkage between the com-
mand and the assertive motivation that follows is not arbitrary: the scalar
strength of si lends the propositional content a sense of "noteworthiness"
that que would not necessarily convey, and thereby makes a stronger re-
quest for the addressees' attention.
Because the illocutionary force of the two utterances linked by si in the
above examples differs (a command and an assertion related to that
command), it is tempting to assimilate them fully to the "speech-act" ex-
amples of Sweetser (1990). However there exists an important differ-
ence. Sweetser's speech-act examples consist of sentences in which it is
the act itself that is motivated or "enabled" by the causal clause, as in
What are you doing tonight? Because there's a good movie at the thea-
ter. In this sentence, the causal clause provides the motivation for per-
forming the speech act of asking the question-in other words, it provides
a "context of relevance" for the question. The "exhortative" examples
with si differ: they do not provide motivation for the speech act itself, but
rather for the action specified in the command (e.g. looking, eating). In-
stead of being based in considerations of speech-act relevance, their pur-
pose is fundamentally persuasive in nature.

4.2. Epistemic contexts

Like the exhortative examples seen in 4.1, in this section we will once
again consider discursive sequences consisting of two utterances with the
structure [VI + si V2]. Once again, there is a paratactic relationship be-
tween the two utterances, and in this relationship the si-marked utterance
V2 remains "pragmatically subordinate" to VI. Specifically, VI repre-
sents an inferential conclusion that the speaker presents as "warranted" or
"justified" by the content of an assertion in V2 (cf Schiffrin 1987: 202).
Because they refer to speaker-internal cognitive positions, these examples
are labeled "epistemic"; indeed, they appear to be good examples of cau-
sality in the epistemic domain, to use Sweetser's (1990) terminology.
Viewpoints and polysemy 271

Two representative examples of si used as a DM in epistemic contexts

can be seen in (14) and (15):

(14) [talking about a friend who has cellulitis]

Q: Se que todas las mujeres la tenemos.
'I know that all women have it. '
Q: Pero ella, es que es una cosa por de mas.
'But her, it's a thing that is too much. '
~ Q: Si encima Ie han dicho que caminara un poco para
'SI they even told her to walk a little to reduce it. '
A: Pero no.
'But she doesn't. '

(15) [A has been telling C how she likes the actor Mel Gibson]
A: Y a L tambien Ie gusta mucho.
,And L likes him a lot too. '
~ A: Si incluso Ie grabe yo una pelicula de el.
'SI I even recorded one of his movies for her.'
A: i,C6mo se Ilamaba?
'What was it called?'
C: i,EI hombre sin rostro?
'The Man without a Face?'
A: Si, esa es.
'Yes, that's it.'

In (14) Q evaluates her friend's cellulitis negatively as "a thing that is

too much". This evaluation is then justified by the si-marked utterance
that follows. Presumably the underlying reasoning involved is this: given
that doctors have told the friend to walk to reduce her cellulitis, it is rea-
sonable to conclude that the cellulitis is "too much". The force of this ar-
gument gains strength in the context of Q's initial assertion that "all
women have it", but despite that fact not all women require the interven-
tion of doctors. In (15) A's statement that her friend L likes Mel Gibson a
lot is justified by A's following assertion that she recorded one of Gib-
son's movies for L. Both examples show si in combination with scalar
additive particles (encima and incluso) which, in these contexts, can both
be translated as 'even'. The pragmatic force of such particles is to mark a
proposition as a strong-if not the strongest-argument for a contextual
272 Scott A. Schwenter

conclusion (Kay 1990; Konig 1991; Schwenter 1999c), and to imply that
other, weaker, arguments for that same conclusion also exist. The combi-
nation of si + scalar particle thus signals very strong speaker commitment
towards an inferential conclusion, which, as in (14), is often a controver-
sial and/or critical assessment. Notice that, unlike the exhortative exam-
ples seen above, epistemic examples like (14) and (15) do not contain
explicitly addressee-directed utterances. To the extent that an opposing
viewpoint can be identified, or even inferred, it is construed in opposition
to the conclusion, not to any aspect of the justificatory utterance intro-
duced by si. As a result, it is in the speaker's best conversational interest
to support the conclusion to the greatest degree possible.
It is interesting to note that most of the examples of si introducing an
assertion that justifies (some aspect of) speaker viewpoint, as in (14) and
(15) above, are functionally quite similar to the epistemic use of the
causal connective porque 'because'. Indeed, porque could replace si in
(14) and (15) without affecting the propositional content in the least, with
the proviso that there would typically be a shorter intonation break
(comma intonation) between the connected segments. The most salient
semantic/pragmatic difference between the two markers is that speakers
intuit a tighter "fit" between the content of the two segments with si so
that, for instance, (14) would be more likely used in a situation in which
the speaker felt that there already was or possibly would be opposition to
her opinion ("her cellulitis is too much") stated in the first segment. The
use of si instead of porque thus tends to reflect emphatic epistemic com-
mitment to the truth of the conclusion. While it is possible that porque
could be used in such a context, the relevant point regarding the epistemic
porque construction is that it has a wider domain of application than, and
is pragmatically unmarked with respect to (cf Lambrecht 1994: 16-17),
the epistemic si construction.
Further evidence for this particular analysis comes from the two con-
nectives' distinct co-occurrence possibilities with epistemic expressions
like me parece 'It seems to me' or creo 'I think.' These epistemic ex-
pressions function to attenuate speaker commitment to the truth of the
propositional content of an assertion. While such expressions seem com-
pletely natural with porque, they are by contrast pragmatically bizarre
with si, as the following examples illustrate (one might imagine these ut-
tered by a speaker A in a context in which another interlocutor B assumes
that Juan is coming to their party that same night):
Viewpoints and polysemy 273

(16a) (Me parece que/Creo que) Juan esta en(ermo, porque 10

he visto en el medico.
'(It seems to me thatlI think that) Juan is sick, because I
saw him at the doctor's office. '

(16b) (*Me parece que/*Creo que) Juan esta enfermo. Si 10 he

visto en el medico.
'(It seems to me thatII think that) Juan is sick. SI I saw
him at the doctor's office. '

Likewise, the well-known epistemic use of Spanish synthetic future

morphology to convey probability or inferred certainty (like English
That'll be John calling upon hearing the phone ring) is wholly compatible
withporque, but not with si:

(17a) Juan estara enfermo, porque 10 he visto en el medico.

'Juan must be (lit. will be) sick, because I saw him at the doctor's
office. '

(17b) *Juan estara enfermo. Si 10 he visto en el medico.

'Juan must be (lit. will be) sick. SI I saw him at the doctor's of-
fice. '

The principal explanation behind these patterns of pragmatic accept-

ability appears to reside in the strong degree of epistemic commitment
that si expresses, not only to the asserted segment that it introduces
(whose content is usually not under dispute), but also, and principally, to
the justificatory relationship between that segment and another segment
that expresses a potentially more controversial speaker position. In this
use, si is "modally harmonic" (cf Bybee, Perkins, and Pagliuca 1994),5
i. e. in epistemic concord, with non-attenuated forms like the indicative
mood, but is clearly not harmonic with markers of lessened epistemic
commitment like those seen in (16) and (17). By contrast, porque carries
fewer restrictions with respect to strength of speaker epistemic commit-
ment it can appear felicitously in both epistemically attenuated and non-
attenuated environments.
As a causal DM in epistemic contexts, then, si expresses justification
of an emphatic sort, as shown by its scalar contrast with porque. The co-
occurrence of si with scalar additive particles as in (14) and (15) above is
therefore not surprising, as these too are informationally emphatic forms.
274 Scott A. Schwenter

Note however that it would be misleading to ascribe an additive meaning

to si, as is made clear by (18):

(18) R: "Que vas a hacer con la cortina?

'What are you going to do with the curtain?'
A: Pues arreglarla.
'Well fix it.'
A: ;,No ves que esta rota?
'Don't you see that it's ripped?'
[A takes the curtain and shows the ripped part to R]
A: Sera muyfacil.
'It'll be very easy. '
~ A: Si total, nada mas es cuestion de coserle esta parte de aqui.
'SI after all, it's only a question of sewing up this part here. '

Here A's inferential conclusion that fixing the curtain "will be very
easy" is warranted by the si-marked assertion that follows. The DM total,
which is most plausibly translated in (18) as 'after all', indicates that the
following proposition represents the only argument relevant to the conclu-
sion. And nada mas 'only, lit. nothing more' further minimizes the de-
scription of the effort involved in fixing the curtain. A's argument for how
easy fixing the curtain will be, then, constitutes the only argument for
coming to this conclusion, i.e. si appears in a non-additive context. All of
these markers thus work in tandem to justify the conclusion, which "fol-
lows" from the assertion in A's fmal tum.
The exhortative examples in the preceding section included a strategy
of lexical minimization (e.g. through negation) in a si-marked utterance as
motivation for carrying out the action specified in an imperative-mood
command. But as the preceding example shows, such minimization is not
limited to exhortative examples-epistemic examples display it also. An-
other epistemic example that uses minimization in support of a previously
stated conclusion is (19). Here, Y is attempting to justify her opinion,
stated at the outset, that the man she's talking about doesn't pay attention
to anyone:

(19) [Y describing a man that she knows]

Y: Es que el, es una persona que no Ie hace caso a nadie.
'it's that him, he's a person who doesn't pay attention to
Viewpoints and polysemy 275

~ Y: Si una vez fuimos a su casa y se qued6 ahi sentado sin decir

nada ... ini se movi6!
'SI once we went to his house and he just sat there without
saying anything ... he didn't even move"

The negative words in Y's second utterance mark an absolute endpoint:

nada 'nothing' and ni 'not even'. They are used in service ofY's conclu-
sion that the man doesn't pay attention to anyone, and run counter to ex-
pectations that Y holds regarding another person's behavior when in the
presence of visitors in that person's home. In addition, the clause ini se
movi6! 'he didn't even move" licenses an inference about other things
the man refuses to do when in the company of others, further sanctioning
the conclusion "he doesn't pay attention to anyone".
The connection between all the examples in this section is that in each
case it is strongly implicated that (i) the si-marked justification offered for
the conclusion is the strongest available, and (ii) the conclusion simply
follows, as a matter of course, from this justification.

4.3. Summary

Implicit in the organization of this paper has been the position that the
adversative DM si and the causal DM si are distinct polysemies of the
same lexeme. This is supported on discourse-structural grounds: the ad-
versative DM presents, within a dialogue, a dialogical "retort" directed at
an interlocutor's utterance, while the causal DM introduces an assertion
which provides motivation/justification for a preceding utterance by the
same speaker. By contrast, the exhortative and epistemic uses of the
causal DM are not distinct polysemies but rather offer a case of prag-
matic ambiguity (Hom 1989; Sweetser 1990): what si is expressing in
each use is the same semantic content, the interpretation of which differs
due to the pragmatic domain of application. The two polysemies also dif-
fer with respect to collocational properties: perhaps most importantly,
only the adversative DM can be prefaced, perhaps in every case, by the
adversative DM pero 'but', which reinforces the contrast expressed by si
(Montolio 1999). Utterance-initial pero is not found with either sub-type
of causal DM si.
Despite these distinctions, however, it should also be obvious from the
foregoing discussion that the boundaries between each use of si as a DM
are not clear-cut, and that the strong dialogical nature of the adversative
276 Scott A. Schwenter

examples endures, albeit to a lesser extent, in the causal examples, where

it is now partially interpretable as an aspect of scalar meaning. Indeed,
one can arrange the uses presented above in relative fashion with respect
to the degree to which they are dialogical or monological. This is shown
in the table below:

Table 2. Uses ofDM si, by viewpoint

Dialogical Monological
Adversative +++
Exhortative ++ +
Epistemic + ++

This display helps illustrate what has been stressed above: exhortative
and epistemic uses of si do not lack dialogical features completely. The
importance of these features is somewhat diminished in these uses, but
paradigmatic contrasts with other forms like porque still indicate a dia-
logical residue: a more salient feature now is high scalar commitment to a
conclusion, as the examples in 4.2 illustrate. The same cannot be said for
adversative uses of si, which always appear to be exclusively dialogical
(even in monologue examples, see Montolio 1999). The greater dialogical
nature of exhortative uses of DM si, as compared with the epistemic uses
in Table 2, results from the addressee-directed characteristics of the for-
mer-most importantly, the assumption of strong opposition on the part
of the addressee-which are exhibited to a lesser extent by the epistemic

5. Conclusion

An obvious implication of the analysis presented in this paper is that there

is an underlying affinity between (some types of) adversativity and (some
types of) causality. This affinity seems commonsensical, especially if one
considers a pragmatic function/strategy like "justification" to be one that
speakers employ when they sense that their positions may potentially be
in dispute. But it also shows that certain pragmatic relationships, for in-
stance, refutation, which hold between (strictly) dialogical viewpoints,
may have as their (relatively more) monological counterparts other kinds
of pragmatic relationship, like justification. Thus, while it is not the case
that all types of adversative expression have a counterpart in the domain
Viewpoints and polysemy 277

of causality, the meaning conveyed by Spanish si in the first conceptual

domain does have an analogue in the second.
The analysis also shows that the distinction between dialogical and
monological viewpoints is not absolute, but rather gradient. The meaning
associated with adversative DM si, especially its function as a marker of
often vehement opposition between viewpoints, is carried over into the
strong scalar meaning of causal uses of si, which, I have argued, conserve
aspects of the dialogical features characteristic of adversative uses. A
strong type of adversativity, namely refutation, is mirrored by the high-
scalar argumentative strength expressed by causal si, which, indeed,
strives to present "irrefutable" justification for speaker-based epistemic
conclusions. Put more generally, then, dialogical "footprints" can be
found in lexemes and constructions which otherwise appear similar to
their more strictly monological counterparts. The gradience revealed here
between types of viewpoint parallels the gradience among the conceptual
domains which are under investigation in this volume (cf Konig 1986;
Kortmann 1997).
Though still under-utilized as an analytical device, the framework em-
ployed in this paper can and has been extended to the study of other lex-
emes. For instance, Traugott (1997: 6) argues th~t the historically distinct
uses of English after all are linked by "dialogic use, whether in dialog or
monolog". In addition, adversative and additive uses of English infact, as
in (20) and (21), respectively, are also amenable to this type of analysis:

(20) They think he's crazy, (but) in fact he's quite sane.

(21) That movie was bad, in fact it was terrible.

These uses, which arise at different times diachronically (Schwenter and

Traugott 2000), can be distinguished synchronically by the dialogical
viewpoint in the adversative example and the monological viewpoint in
the additive example (Schwenter 1999a). Many other differences be-
tween the two types of in fact can also be related to this difference in
viewpoint: the greater bleaching of the nounfact in (21), the sense of "re-
formulation" in the same example, and the epistemic scale in (20) versus
the "rhetorical" or "argumentative" scale in (21).
Finally, even lexemes whose different uses do not appear to constitute
separate polysemies convey nuances of meaning which are sensitive to
the number of viewpoints involved. This is the case, for instance, of epis-
temic sentence adverbs like certainly in dialogues like these:
278 Scott A. Schwenter

(22a) A: John isn't smart.

B: He certainly is!

(22b) A: John sure is smart.

B: He certainly is!

In both examples the main semantic contribution made by certainly is to

signal and reinforce speaker B' s full epistemic commitment to the propo-
sition "John is smart". Because of this semantic parallelism, it does not
seem plausible to consider the two uses to be different polysemies. None-
theless, as a result of the distinct argumentative orientation of B' s utter-
ance with respect to A's in each case-anti-oriented in (22a) vs. co-
oriented in (22b)-it is feasible to identify certainly as a dialogical prag-
matic marker of disagreement in (22a) but as a monological marker of
agreement in (22b).


1. See Schwenter (1999b) and Montolio (1999) for discussion of the pero si
2. The scalar relationship between the two adverbs in (8) is easily demon-
strated. Consider the results when they are inserted into a syntactic frame of
the type "X but not Y", where X represents the "weaker" adverb, y the
"stronger" one:
(ia) Te 10 han tintado mal, pero no fatal.
'They dyed it badly, but not horribly.'
(ib) *Te 10 han tintado fatal, pero no mal.
'They dyed it horribly, but not badly.'
The infelicity of (ib) results from the fact that any speaker of Spanish would
order these adverbs as <fatal, mal> along a semantic ("Horn") scale, such
thatfatal semantically entails mal, but not vice-versa.
3. The intonational break between the imperative and the utterance segment
justifying it tends to be longer when si introduces the latter. This is reflected
by the comma intonation in (11a) as opposed to the period in (11b).
4. Note too that the translation equivalent of because in Spanish, porque, could
not replace si in the examples seen above. In fact, upon translating examples
(10) through (13) into English I am unable to replace si with any English
connective form.
5. My use of "modally harmonic" with respect to discourse markers extends the
term beyond that of the scholars cited, who limit it to cases involving a mo-
dal verb and other devices like adverbs.
6. The differences between this connective use of si and epistemic porque are
very similar to those separating discourse marker uses of English after all
Viewpoints and polysemy 279

and epistemic because (Traugott 1997). Nevertheless, after all and si in

these uses are not always possible translations of each other.
7. In Schwenter (1999b), I have pointed out the extreme similarity between si
and the Japanese connective datte, which can be translated as either 'but' or
'because', depending on its context of use (Mori 1994), and which "by itself
indicates disagreement" (Ford and Mori 1994: 56). Unfortunately, these au-
thors do not tackle the issue of whether the separate uses of datte constitute
distinct polysemies of a single lexeme.
8. Another potential difference not mentioned in this paper is the intonation
pattern of each use. According to Montolio (1999), utterances introduced by
adversative DM si are characterized by a rise-fall pattern with strong stress
on the final tonic syllable. In my data from Alicante, this pattern is found
mainly with adversative examples in which si is reinforced by pero, and
therefore could be interpreted as an iconic reflection of the speaker's
strengthened adversative stance. But whatever the case may be for the adver-
sative DM, it appears that causal OM si does not occur with this unique in-
tonation pattern, though at times increased loudness (as in an exclamatory
utterance) can be found.


Atlas, Jay David, and Stephen C. Levinson

1981 It-clefts, informativeness, and logical form: radical pragmatics (re-
vised standard version). In: Peter Cole (ed.), Radical pragmatics, 1-
61. New York: Academic Press.
Blakemore, Diane
1996 Are apposition markers discourse markers? Journal of Linguistics
32: 325-47.
Briz, Antonio (ed.)
1995 La conversacion coloquial (Materiales para su estudio). (Cuadernos
de Filologia 16.) Valencia: Universitat de Valencia.
Bybee, Joan, Revere Perkins, and William Pagliuca
1994 The evolution ofgrammar. Chicago: University of Chicago Press.
Couper-Kuhlen, Elizabeth
1996 Intonation and clause combining in discourse: the case of because.
Pragmatics 6: 389-426.
Ducrot, Oswald
1984 Le dire et Ie dit. Paris: Minuit.
1996 Slovenian lectures/Conferences Slovenes: Argumentative seman-
tics/Semantique argumentative. Ljubljana: ISH.
Ford, Cecilia, and Junko Mori
1994 Causal markers in Japanese and English conversations: A cross-
linguistic study of interactional grammar. Pragmatics 4: 31-61.
Fraser, Bruce
1988 Types of English discourse markers. Acta Linguistica Hungarica
38: 19-33.
280 Scott A. Schwenter

Garcia, Serafina
1996 Las expresiones causales y finales. Madrid: ArcolLibros.
Haiman, John
1978 Conditionals are topics. Language 54: 564-89.
Horn, Laurence R.
1989 A natural history ofnegation. Chicago: University of Chicago Press.
Jespersen, Otto
1940 A modern English grammar on historical principles, Volume 5:
Syntax. London: George Allen and Unwin.
Kay, Paul
1990 Even. Linguistics and Philosophy 13: 59-111.
Konig, Ekkehard
1986 Conditionals, concessive conditionals and concessives: Areas of
contrast, overlap and neutralization. In: Elizabeth Closs Traugott,
Alice ter Meulen, Judy Snitzer Reilly, and Charles A. Ferguson
(eds.), On conditionals, 229-46. Cambridge: Cambridge University
1991 The meaning offocus particles: A comparative perspective. Lon-
don: Routledge.
Kortmann, Bernd
1997 Adverbial subordination: A typology and history of adverbial sub-
ordinators based on European languages. Berlin: Mouton de
Ladusaw, William A.
1996 Negation and polarity items. In: Shalom Lappin (ed.), The hand-
book ofcontemporary semantic theory, 321-41. Oxford: Blackwell.
Lambrecht, Knud
1994 Information structure and sentence form. Cambridge: Cambridge
University Press.
Montolio, Estrella
1999 iSi nunca he dicho que estuviera enamorada de ell Sobre construc-
ciones independientes introducidas por si con valor replicativo. Ora-
lia 2: 37-69.
Mori, Junko
1994 Functions of the connective datte in Japanese conversation. In: No-
riko Akatsuka (ed.), Japanese/Korean Linguistics, Vol. 4, 147-63.
Stanford: CSLI Publications.
Pons Borderia, Salvador
1998 Conexi6n y conectores: estudio de su relaci6n en el registro infor-
mal de la lengua. (Cuadernos de Filologia 27.) Valencia: Universi-
tat de Valencia.
Porroche Ballesteros, Margarita
1998 Sobre algunos usos de que, si yes que como marcadores discursivos.
In: Ma Antonia Martin Zorraquino and Estrella Montolio Duran
(eds.), Los marcadores del discurso: Teoria y analisis, 229-42. Ma-
drid: ArcolLibros.
Viewpoints and polysemy 281

Roulet, Eddy
1984 Speech acts, discourse structure, and pragmatic connectives. Jour-
nal ofPragmatics 8: 31-47.
Schiffrin, Deborah
1987 Discourse markers. Cambridge: Cambridge University Press.
Schwenter, Scott A.
1998 From hypothetical to factual and beyond: refutational si-clauses in
Spanish conversation. In: Jean-Pierre Koenig (ed.), Discourse and
cognition: Bridging the gap, 423-35. Stanford: CSLI Publications.
1999a The pragmatics of conditional marking: Implicature, scalarity, and
exclusivity. (Outstanding Dissertations in Linguistics.) New York:
1999b Discourse markers and the PAISN distinction. Ms., The Ohio State
1999c Two types of scalar particles: evidence from Spanish. Advances in
Hispanic linguistics. In: Javier Gutierrez-Rexach and Fernando
Martinez-Gil (eds.), 546-61. Somerville, MA: Cascadilla Press.
Schwenter, Scott A. and Elizabeth Closs Traugott
2000 Invoking scalarity: the development of in fact. Journal ofHistorical
Pragmatics 1: 7-25.
Sweetser, Eve E.
1990 From etymology to pragmatics. Cambridge: Cambridge University
Traugott, Elizabeth Closs
1997 The discourse connective after all: A historical pragmatic account.
Paper presented at the 16th International Congress of Linguists,
Paris, July.
1999 The rhetoric of counter-expectation in semantic change: A study in
subjectification. In: Andreas Blank and Peter Koch (eds.), Histori-
cal semantics and cognition, 177-196. Berlin: Mouton de Gruyter.
The treatment of contrasts in interaction *

Cecilia E. Ford

Relying on data from natural conversations and emphasizing the on-going

analyses afforded by speakers as they respond to one another, this study ex-
plores the functions of contrasts in interaction. Considering contrast as
broadly including neutral contrast, concession, and antithesis, and recogniz-
ing that contrasts can be manifested as disagreements across interlocutors, a
recurrent rhetorical pattern is found wherein contrasts are followed by ex-
planations or solutions. Furthermore, cases of contrast deployed without ex-
planation or solution are found to be followed by talk in which interlocutors
treat the contrast as requiring further elaboration. Unelaborated contrasts are
also found to be used in a variety of ways related to activity and speaker
role. The methodology is an adaptation of conversation analysis, with the
talk of interactants themselves-specifically talk following contrasts-used
as a primary resource for interpreting the functions of the rhetorical patterns
under scrutiny.

1. Introduction

Documenting the recurrent resources and practices of language use is a

primary goal for researchers in both discourse-functional linguistics (for
recent reviews see Cumming and Ono 1997; Tomlin et al. 1997; Hopper
1998, among others) and conversation analysis (see Atkinson and Heri-
tage 1984; ten Have and Psathas 1990; Whalen 1992, among others).
Understanding linguistic structures and how such structures are com-
bined in use, is fundamental to the study of language development, lan-
guage change, and language as social action. As part of a functional re-
search program focusing on language in use, the present study examines
the realization of a rhetorical relation which is recurrently deployed in
spoken interaction: the relation of contrast. I explore how speakers-
participants in interaction-treat contrasts through their talk directly
following the textual relation, with attention to how that subsequent talk
both builds on and interprets the functions of the contrasts themselves.
My attention will thus be on the functions of contrasts as they are elabo-
rated on and responded to by participants in face-to-face interaction.
This method of inquiry is an adaptation of the conversation analytic
284 Cecilia E. Ford

practice of relying on the talk of interactants as a primary source for the

analysis of functions of talk-in-interaction (see Petrakyla 1997).
For the purposes of this study, a contrast consists of two text spans
which present an opposition. This type of textual relation subsumes the
definitions of contrast, antithesis, and concession in Rhetorical Structure
Theory (RST) (Mann and Thompson 1988). Included in my collection
are cases where two parts of a text, generally (but not exclusively)
clauses, are presented as entities, actions, or propositions which could

(a) comprehended as the same in many respects

(b) comprehended as differing in a few respects
(c) compared with respect to one or more of these differences
(Mann and Thompson 1988)1

A feature of RST which makes it especially useful for the broad study of
rhetorical relations is that it allows for relations to be either explicitly
marked or inferable based on the content of text spans in question; thus
no explicit linguistic marking is required for the interpretation of two
spans as being in an oppositional relation. Furthermore, RST places no
fixed limit on the size of text spans to be related, though in practice the
lowest unit has often been the clause.
Example (1) offers an initial sense of the type of pattern I will be ex-
amining. The quote is from a magazine interview, and the speaker is a
student who organized a protest at her high school. She is describing the
effort it took:

(1) (From Ford 1994)

<" lIt sounds eas~,1

!but i t wasn' ~

'cause pep assembly's a real big deal,
especially at homecomdng; the whole school

The first part of the contrast ("It sounds easy") explicitly refers to an
assessment that the interviewer (or the magazine readers) could poten-
The treatment ofcontrasts in interaction 285

tially make;2 the second part of the contrast ("but it wasn't") stands in
opposition to the first. What follows the contrast is an explanation of the
opposition: the speaker explains that the setting for the protest was a
particularly important and consequential one. Here then, through her
subsequent talk, the speaker shows her understanding of the contrast as
presenting a problem needing resolution, and the resolution she provides
is an explanation supporting one pole of the contrast. At one level of
rhetorical analysis, we see that the incompatibility between the parts of
the contrast is addressed and resolved through the explanation.
At another level, when we take these rhetorical moves in their con-
text we see a contrast plus explanation doing particular persuasive work.
A contrast, here, functions as a format for presenting two perspectives,
of which the speaker identifies strongly with only one. Through this rhe-
torical combination, the speaker efficiently acknowledges a potential
incompatibility but builds support for her own perspective as she adds
an explanation for one pole of the contrast.
In this paper, I examine the functions of contrasts in interaction, with
a specific focus on the relationship between a contrast and the talk
which follows. While the rhetorical relation of contrast is a format for
the expression of diverse actions, a pattern is evident: across contexts,
contrasts are treated as needing resolution in subsequent talk; that is,
they are regular building blocks for larger actions involving explana-
tions or solutions for problematic states of affairs. In the data I have ex-
amined for this study, there is indeed a pattern of explanation and solu-
tion relations following contrasts; however, as we will see, it is also evi-
dent that the larger context of activity and the role of a speaker in an
activity are mediating factors affecting how and whether further resolv-
ing talk is treated as relevant after a contrast.
The remainder of the paper will be organized as follows: In section 2,
I describe my data sources; section 3 offers some examples of contrasts
in spoken English. Section 4 sketches the lexical and grammatical en-
coding devices associated with contrasts in the present data. In 5 and 6,
bearing in mind that the social and sequential context of any piece of
talk is central to the formulation and interpretation of what that talk is
doing, I discuss the rhetorical combinations found in the data:
section 7, I examine cases where contrasts are not followed by explana-
tions or solutions. I discuss relationships between activity type, speaker
role, and the location of the contrasting action in an interactional se-
quence in relation to the absence of further talk in these cases. Section 8
286 Cecilia E. Ford

concludes the paper with a general review and discussion of the find-

2. Data sources

Two different videotaped conversations provide the primary data for

this study; both involve spontaneous talk-in-interaction. One is in an
institutional setting and includes a science teacher and a class of high
school students. The other source is an evening gathering of five female
friends in a home setting.
The science class, in an urban U.S. public high school, was video-
taped twice a week during the course of a semester (Spring 1997) as part
of a broader longitudinal study of language socialization among diverse
students. 3 I have concentrated on the interaction in one 50-minute pe-
riod. The class hour involves pre-class interactions before the official
bell rings, a teacher-fronted segment (including many student contribu-
tions), and a laboratory activity during which one group of four female
students was videotaped. 4 The recording comes from the second semes-
ter of a two-semester course, so the students and teacher are quite famil-
iar with one another.
The second source is a 20-minute segment from a three-hour video-
taped gathering, involving five women between the ages of 30 and 46
(taped in December 1995). The women are familiar with each other to
varying degrees: Terry (T) and Pam (P) are domestic partners, Abby (A)
is a friend of theirs, and Rachel (R) is more of an acquaintance of T and
P. R was invited by a friend ofT and P's (Stacy), who is out at the store
during this segment of talk.

3. Contrasts in conversation

In spoken language, contrasts have been previously documented in a

variety of uses. A contrast can be produced as part of a more planned
public address, as part of a longer turn in informal interaction, and
across speakers in the expression of disagreement.
Contrastive combinations are treated as significant units of talk in
public speaking events. Audiences orient to contrasts by responding to
such pairings with applause or other forms of group action (Atkinson
The treatment ofcontrasts in interaction 287

(2) simp~i£ied from Atkinson (1984) Xs app~ause; =

the contrasting units are bracketed on the ~e£t5

Heath: ~owo the la:bour:: (0.4) prime mdnister and

his collea:ques are boasting in this election
campaign (0.7) that they have brought
infla:tion down from the disah:strous level

r twenty six percent (1.4)
But we are: entitled to inquire
(0 • 4)
Who put i t up to twenty six percent=
Audience: hehh[eh
Audience: [x-xxXXXXXXXXXXXXXX ...

Contrasts are produced by single speakers either as explicitly oppos-

ing propositions, or as negations of presupposed but not explicitly stated
information, as in (3), from the data:

(3) A teacher is handing out papers £or an assignment

T: This thing that you're getting now, (.) the maze?

?: ( )
T: Okay?
( .)
=>T: You don't have to turn this in.
( .)
T: Listening?
( .)
T: I'm gonna give you this 'cause you're gonna write
on it,

To understand this example, one needs to know that it is the custom in

this class that written assignments which the teacher distributes be
turned in once they are finished. In this instance, the teacher explicitly
tells the students that this sheet with a maze on it is an exception. At the
arrow, the teacher produces a clause which references, through nega-
tion, a corresponding affirmative expectation; that is, in stating that the
students "don't have to turn this in", the teacher indexes a shared under-
288 Cecilia E. Ford

standing in this class that assignment sheets in general are to be returned

to the instructor.
Contrast may also occur within the matrix of an interactional se-
quence, as in example (4). During a laboratory activity involving can-
dles, the lights in the classroom have been turned off, but at this point
most of the students have fmished the task:

(4) ("I'monna" = 'I am going to')


- -
::) T: You prefer them aw- I prefer them on, I can see:.

These examples illustrate the occurrence of contrasts in spoken lan-

guage. In the case of audience response, in particular, prior research
supports an interpretation of these complex units as real and relevant to
the participants themselves.

4. Forms of contrast

For the current study, I collected 50 instances of what could broadly be

interpreted as utterances and utterance sequences expressing contrasts,
with reference to the Rhetorical Structure Theory definitions of contrast,
antithesis, and concession, as discussed in section 1. In addition to in-
stances where both poles of a contrast were expressed, I also collected
cases where a negative clause implicitly indexed a corresponding af-
firmative, as in example (3) above. Cases of negation can be understood
as responding to or constructing a context in which the corresponding
affirmative is relevant or assumed (Giv6n 1978; Tannen 1979; Fair-
clough 1989).
The collection also includes disagreements across speakers as a form
of contrast, as in (4):
The treatment ofcontrasts in interaction 289

( 4) (repea ted)



?: ( (O.~) ) 7-
T: You prefer them aw- I prefer them~, I can see:.

The second explanation is embedded in the first: as can be seen in (4),

the linguistic markers of contrast in this collection are in line with what
has been described in the work of Halliday and Hasan (1976), Martin
(1993), and Rudolph (1996), and, for discourse in science classes,
Lemke (1993). These markers include a variety of lexical and gram-
matical strategies and devices for expressing dissimilarity, including
antonymy (in adjectives, nouns, and prepositions), negation, compara-
tives, and conjunctions. Adverbs associated with contrasts were actu-
ally, really, and only. Temporal oppositions such as then versus still
were also represented. Finally, prosodic highlighting (contrastive stress,
pitch and rhythmic parallelism) was a frequent feature in contrasts.
The defining criterion for contrasts in this study was that incompati-
bility be expressed, whether in content, through linguistic marking, or
by both. The collection contains cases of contrasting clauses, phrases,
and words; and contrasts may be produced by single speakers or across
speakers, with negation and disagreement included.
In sections 5 and 6 I provide analyses of the two types of rhetorical
combination that form the majority of cases in my data: CONTRAST +
verse with respect to the functions of the contrasts, i. e., the presentation
of the incompatibility or opposition itself, serves different interactional
ends. However, in common across cases, we fmd participants treating
contrasts as presenting problems in need of explanation or solution.

5. Contrast + Explanation

In my data base, a pattern was evident in which contrasts were followed

by explanations. The textual relation I am calling explanation corre-
sponds to what, in an early form of Rhetorical Structure Theory, was
categorized as REASON relation. As defined by Fox (1987), this textual
290 Cecilia E. Ford

relation involves one span "which makes a statement about something"

and another span "which provides the reason for that statement (either
the reason for making it or the reason for it being so)" (1987:82). In my
data the statement for which a reason is provided involves a contrast.
The rhetorical cluster of CONTRAST + EXPLANATION is produced
in a variety of different interactional sequences, within a single
speaker's talk and across speakers.
An initial example comes from a teacher-dominated portion of the
science class. In (5) the science teacher is involved in a lecture, which
includes prompts for student responses. In this excerpt, the teacher pro-
duces two contrasts and two explanations.

T: As I move away, (.) does my height get
CONTRAST (1) ~ T: =No.
?: 0No. 0
EXPLANA- (1):::> T: I look smaller,because I'm farther away,


First, the teacher answers his own question, "does my height get differ-
ent?" with the contrasting "No." Next, he explains this negation by con-
ceding that while there is an illusion of getting small, it is only because
of the added distance. And within that explanation, he introduces yet
another contrast, the contrast between his original denial of difference
and his subsequent concession that there is an appearance of difference.
This contrast is highlighted by the stress on look (as opposed to actually
being different). As with the first contrast, the second is followed by the
explanation, "because I'm farther away."
The contrasts in (5) are vehicles through which the science teacher
highlights expectations and assumptions, thus leading the students
through a discovery process. The first contrast draws attention to the
fact that there is an incompatibility between the image and the object
being reflected. This highlighted contrast then leads into the explana-
tion. Within that explanation is the more subtle contrast between ap-
pearance and reality, indexed through the contrastive stress on look.
And again, the opposition highlighted through the contrast leads into an
explanation component.
The treatment ofcontrasts in interaction 291

The conversation recorded in a home setting also contains instances

of the CONTRAST + EXPLANATION cluster. The exchange in (6)
takes place as A, R and T have been playing in the board game Piction-
ary. R remarks on the striking difference between where T's team was
located just a short while earlier and where they are now-far ahead on
the gameboard relative to R's team. As an explanation, T suggests that it
was the participation of Stacy on T's team which allowed them to ad-
vance so far ahead:

(6) R: 0: :kay.
R: W'll yeah. Look at that.
R: You were back here'n
CONTRAST:::) [we were way over there.
A: [Yau: :p?
A: Yup?
T: [huh huh huh huh huh
A: [eh [ heh heh heh
R: [uh hah hah hah hah hah hah hah hah=
A: =I say we keep her

To understand T's tum as an explanation requires a bit of background in

North American popular culture. T is using a currently common joking
formula borrowed from the children's television show, "Mr. Rogers."
Mr. Rogers has an excessively gentle style of teaching, which contextu-
alizes the addressee as very young and just learning new concepts. Mr.
Rogers introduces something new and then prompts the children in the
viewing audience to try saying the new word: "Can you say (.) __?"
Imitating the same stylized intonation, including the pause before the
new word, comedians as well as regular folks in North America have
taken to using this question format as a way of stating something obvi-
ous, thus treating the situation as one in which the answer or "new
word" should have already been clear. The question is now a joking
formula for stating the obvious; the formulation imitates a nursery
school teacher's request for choral repetition of a new vocabulary item.
T's tum, "Can you say (.) Stacy?", is presented in an exaggerated,
mock-didactic intonation and at a raised amplitude. After Stacy joined T
and A's team, they moved ahead on the gameboard; thus, "Stacy" is the
explanation for the discrepancy between where the team was and where
292 Cecilia E. Ford

they are now. That A and R understand the joke is evidenced by their
laughter. A's "I say we just keep her" also demonstrates that the refer-
ence to Stacy's being on the team is interpreted as an explanation for the
team's success.
In (6), the contrast is presented by one speaker and the explanation
by another. By providing an explanation, T treats R's statement of the
contrast as unfinished in itself and calling for elaboration. Thus, in terms
of its treatment by the conversation participants, we can analyze this
contrast as expressing an opposition that is remarkable and puzzling, an
observation that seems to draw further talk providing a resolution, even
if it is done in a joking format.
Example (4), repeated once again, is from the setting of the science
class, but the activity is quite different from the one in (5), in which the
teacher was in more firm control of the direction of the talk. Here the
students are working in small groups on laboratory tasks. There is much
more student talk in this portion of the class; and even as the teacher
announces an action he's about to take, at least one student feels free to

(4) (repea ted)

?: [
CONTRAST~ T: You prefer them aw- I prefer them on,
EXPLANATION~ I can see:.

The student's turn presents an unelaborated contrast; in stating her pref-

erence, she offers no explanation or justification. However, when the
teacher partially repeats K' s contrasting turn, we see another case of the
CONTRAST + EXPLANATION cluster: the teacher adds his own con-
trast, "I prefer them on", and he follows this with the explanation, "I can
see." In this case, the contrast is formed across speakers as a disagree-
ment, and thus the explanation comes as the account portion of what
conversation analysts understand as a dispreferred response.
As reviewed by Levinson (1983:333-335), conversation analytic re-
search has shown that turns which are in some sense disagreements (in
CA terms "dispreferred responses") regularly contain components ex-
The treatment ofcontrasts in interaction 293

plaining or otherwise accounting for the disagreement. Thus, from a

conversation analytic perspective, the contrast in (4) would be a case of
a dispreferred response with a characteristic account portion:

(4) (Teacher's tuxn)

T: You prefer them aw- I prefer them on, I can see:.

11 11

Accounts, as Levinson (1983) defines them, are "carefully formulated

explanations for why the (dispreferred) act is being done" (334). The
rhetorical relation of explanation is associated with dispreferred tum
formats in cross-tum contrasts.
In a study of adverbial clauses in interaction (Ford 1993), I noted that
because regularly introduced account portions of disaffiliative re-
sponses. This can be seen in the following example taken from a phone
conversation. The topic is R's new job:

(7) simp~i£ied £rom Ford (1993:115)

A: Did you get your (.) first pay check from it?
( .)
A: [At least?
=>R: [NO: I won't get that for a couple of weeks yet.=
A: =Oh,
( .)
A: [W'l
=>R: ['Cause i t takes a long time.

A's extension, "at least?", displays the pursuit of agreement from R. At

the first arrow, R instead produces a negative and disaffiliative re-
sponse, which is followed, at the second arrow, by a because-prefaced
account for this disagreement.
In a subsequent study (Ford 1994), following up on the observation
that because introduced accounts for disaffiliative responses, I looked at
the use of because in different types of talk and writing. I found that if
disaffiliation was understood as a form of contrast, then the pattern of
294 Cecilia E. Ford

because-prefaced units introducing accounts could be generalized across

text types.
While my findings in that study supported the observation that be-
cause regularly functioned to introduce accounts for contrasts, I did not
look for a broader pattern beyond units connected with that specific
causal conjunction. In other words, the 1994 study involved a collection
of cases based on linguistic form, the connector because: in a data base
of instances of because, I found that a pattern in their immediately pre-
ceding textual environments regularly involved contrasts. 6 In the present
study, however, I fmd evidence of a more general pattern whereby the
contrasts and explanations come in sequenced combinations regardless
of the use of any particular formal connector.
This pattern of explanation following contrast was common in the
more freely interactional portions of the science class. In (8), just before
the class officially begins, student J has asked for change for his fifty
dollar bill. His intent is to buy some candy from S. S now offers him
change in the form of a check she has received from someone else in
payment for the candy she is selling (a second-party check):

s: I have a check.
S: Eight fifty.
CONT~T=> J: Nah. I won't take- I don't take second-
party checks.
S: eh huh huh huh huh
EXPLANATION=> J: I don't got no way of telechecking 'em,

J's refusal (contrast) is followed by an explanation in which he cites

his inability to verify whether the check is good ("telechecking" is a
process used by stores to get clearance for cashing checks). The tum
format includes normative features of disagreeing responses: the declin-
ing "Nah" and the account portion (what 1 am calling an explanation), "I
don't take second party checks... 1 don't got no way of telechecking
'em." Note also that there are nested contrasts and explanations here, as
in (5), above. J's "Nah" is followed by the explanation that he does not
take second party checks, and that negation is further followed by the
explanation that he has no way of electronically verifying the validity of
the checks.
As a final example of contrast plus explanation in my data, (9), be-
low, is an instance which supports the claim that explanations after con-
The treatment ofcontrasts in interaction 295

trasts are relevant not only from a discourse analyst's perspective, but
also, and in a significant manner, to the interactants themselves. In (9) T
presents an assertion which contrasts with a prior assumption (line 8),
but she does not immediately provide an explanation that might resolve
the asserted incompatibility. Note that both A and R treat the absence of
an explanation as problematic and deserving of further attention.
A and R have been remarking on the fact that there are many pictures
of people and sheep on the walls of T and P's dining room, where they
are seated. At lines 4-7, R begins to refer to the people in the pictures as
relatives, "grandma" and "grandpa":

1 R: I see there's a lo:t of pictures of sheep.
2 A: Oh yeah. there's more in the other [r(h)oo:m
3 t(h)oo.
4 R: [There
5 there's sheep there, Right by grandma.
6 A: ha ha ha
7 R: Gramma, (.) an' (.) an' (.) and grandpa,and the
8 the sheep and [baby sheep.
9 =>T: [We don't know these people.
10 (. )
11 T: t>or< these people.
12 R: y~ dotn't?
13 ( .)
14 T: No.
15 ( .)
16 A: Do you know the one [just below that one
17 ?: [just
18 (0.5)
19 A: the one in the mid[dle?
20 T: [No.
21 T: The little cryin' girl?
22 A: No?
23=>R: [You just like the >pictures.<
24=>T: [We just like- silver: frames and
25 pictures,
26 R: And what about the sheep.=Do you know the
27 sheep?
296 Cecilia E. Ford

28 A: Eh(h) [heh heh heh

29 R: [Ub heh heh heh
30 T: These have been in our family [for
31 gener(h) at(h) ions
32 R: [Ha ha ha ha ha

At line 9, T produces a tum which contrasts with R's representation (4-

7) of the people as relatives. Instead of agreeing with R's statements, T
asserts that she and her partner P "don't know these people." It is after
this contrasting assertion that T fails to provide an explanation. What
happens next supports the existence of an expectation, on the parts of
these participants, for an explanation to be offered as a resolution to the
incompatibility expressed through the contrast. They treat the contrast
here as a remarkable announcement needing elaboration.
At 12, with marked pitch height, R expresses surprise, with "You
do l'n' t?", a form of repetition which regularly functions as a prompt for
further elaboration on the part of the prior speaker. But at 14, T, rather
than providing an explanation now that her contribution has been treated
as noteworthy, simply reasserts her negation. From 16 through 22, A
tries to salvage the assumption of T and P's familiarity with the subjects
of the pictures: she draws attention to persons in other pictures, asking
whether T and P are at least familiar with them. However, T continues
to respond with unelaborated, unexplained denial (line 20).
It is not until lines 23 and 24 that the sought after explanations are
offered, and at that point they are produced by R and T, in overlap. R
offers her guess at a possible account, "You just like the pictures," and
simultaneously, T gives her explanation, "We just like- silver: frames
and pictures." Only after these explanations does the talk move beyond
the problematic contrast. What follows is a joking question from R
about the sheep; and T joins in the joking by pretending that the sheep
have been in her family "for generations."
In the clear orientations of the interactants in (9) to something pro-
blematic and unresolved after T's contrasting assertion, we find further
evidence for a shared expectation that explanation should follow at least
certain types of contrasts.
In addition to the cases in my data, I have found other instances of
CONTRAST + EXPLANATION in published studies of conversation.
Notably, cases reported for English in the volume Analysing Causal
Conversation (Eggins and Slade 1997) and in a study of tum-
constructional units in German by Margret Selting (1998). While space
The treatment ofcontrasts in interaction 297

limitations preclude the inclusion of those cases here, and while the au-
thors do not draw specific attention to a CONTRAST +
EXPLANATION rhetorical cluster in their data, I take its occurrence in
their examples as suggestive of a more general pattern, one that is likely
not exclusive to my data nor to English conversations.
In sum, in the present data, we find a rhetorical pattern involving
contrasts, in a variety of functions, regularly followed by explanations.
As shown here, contrasts may be used pedagogically to highlight puz-
zles or problems; they may be the vehicles for observing remarkable
states of affairs; and, through interactional sequences, they may involve
disagreements between speakers. Significantly, explanation is a treated
by interactants as normatively relevant subsequent to contrasts in a vari-
ety of functions. Whether produced by single speakers or across speak-
ers, and whether prefaced by because or not, speakers in these data
regularly treat explanations as relevant after contrasts.
In addition to the pattern of contrasts followed by explanations, an-
other set of cases from my data involve contrasts followed by solutions.
Examining these cases provides further perspective on sequential con-
texts in which elaboration is added after a contrast. These cases also
support the observation that contrasts are used to present incompatibili-
ties or problems which participants treat as needing resolution.

6. Contrasts + Solutions

A further pattern was evident in the data whereby contrasts were fol-
lowed either by solutions or by displays that solutions, if not actually
stated, were understood as relevant or expected. In Rhetorical Structure
Theory, a relation of SOLUTIONHOOD requires that one span present
a problem and that the "situation presented in [the other text span be] a
(partial) solution to the problem" (Mann, Matthiessen, and Thompson,
1992:72). While Mann et al. do not point to the rhetorical relation of
CONTRAST as a pattern used in expressing the problem in the solu-
tionhood schema, they do note that the "scope of a problem includes...
conditions that carry negative values, either expressly or culturally, in-
cluding...frustrations" (1992:72). This resonates with Longacre's (1983)
notion of frustration in relations between predications. Frustration in a
text relation involves an expectancy which is countered. Contrasts ap-
pear to be common textual expressions of situations that go against ex-
298 Cecilia E. Ford

pectations. The expression of solutions after contrasts evidences partici-

pants' interpretation of contrasts in just this way.
Solutions also have commonalities with explanations in such rhetori-
cal contexts. Both types of additions treat contrasts as problematic and
needing further talk in the service of resolution. In the case of explana-
tions, resolution is in the form of support or justification. With solu-
tions, the contrast is resolved through the expression· of a potential rem-
In (10), R offers a solution to the problem A has just expressed
through a contrast. A and P have been Christmas shopping at a store that
carries bird feeders. P is now on the phone in the another room, and A is
telling R about their shopping dilemmas:

(10) "she" =P
A: She's trying to decide if sh[e wants to SPEND=
R: [She has a lot of
A: =thirty five dollars on her dad. for Christmas.
R: Oh.
A: And she really doesn't.
=> (0.4)
A: But- (.) the gift that she thought of was thirty
five dollars, and she can't find i t anyplace
R: Mm-mrmn.
A: so,
=> R: Has she tried Chickadee Depot.

There are actually two contrasts in this example: (1) spending $35 or
not, (2) not wanting to spend $35 (for which finding the gift more
cheaply would be a solution) but being unable to find the item anywhere
else for less. Note the pause after the first contrast, an indication that R
is treating A's tum as not yet complete. Given the pattern I am observ-
ing, this would be a relevant place for an expression offering resolution,
either an explanation or a solution. A normative pattern whereby con-
trasts are elaborated in such ways could explain R's lack of uptake fol-
lowing this first contrast and during the pause at the first arrow; she may
be awaiting an explanation, for example.
The treatment ofcontrasts in interaction 299

The second arrow points to R's response to the second contrast. This
contrast is between P's not wanting to spend $35 but not being able to
find the birdfeeder elsewhere for less. By asking whether P has tried
"Chickadee Depot," a local bird store, R is proposing a potential solu-
tion. In suggesting this solution (in the form of a question), R displays
her interpretation of the contrast as expressing a problem, for which a
search for solution is relevant.
In the science class, not surprisingly, there was a great deal of talk
focusing on problems and solutions. In the following case, the teacher
treats a solution as relevant in the context of a contrast, but he suggests
that the students solve the problem on their own. This strategy seems
particularly appropriate to one of the pedagogical goals in this discourse
context: to teach the students to formulate and solve physics problems.

T: If I stand thi-Oand this is the mirror
th- the screeno here, (0.3) I'm at a
«pointing to wall in front of him»
certain height, right? Look in.
~ If I go back farther, am I smaller?
CONTRASTS I'm still the same height, right?
( .)
J: Yeah,=

ORIENTATION T: =Try (.) If you care to. (.) try this

TO SOLUTION uh (.) problem, ( .) but doing i t wi th
uh scale drawing.

The teacher's contrasts here are formulated as questions, and although

he does not provide a solution, he does explicitly name what he has pre-
sented a "problem," and he suggests that the students may want to solve
So in addition to being treated as presenting incompatible states of
affairs needing explanation, as described in section 5, contrasts can also
be treated as presenting problems, after which solutions are relevant.
Both solutions and explanations are elaborations which contextualize
preceding talk as unfinished, problematic, and needing resolution. The
present data provide us, then, with a basis for positing two kinds of rhe-
torical clusters and for using these clusters as a basis for interpreting the
work that contrasts do in talk-in-interaction.
300 Cecilia E. Ford

7. Contrasts without explanations or solutions

In a major portion of the cases I have analyzed, speakers either produce

or treat as problematically absent the patterns described in sections 5
and 6; that is, contrasts are either followed by explanations or solutions,
or speakers treat the lack of such elaboration as notable, through delays
or through further talk in pursuit of resolution. However, the collection
also includes instances of contrasts after which explanations and solu-
tions are not produced nor are they explicitly treated as relevant. In a
small subset of these instances, the contrasts appear to function perfectly
well alone, that is, they are in no way treated as unresolved in their con-
texts. The functions of contrasts which are treated in this way is worthy
of further investigation. In the majority of these non-explanation, non-
solution cases, however, there are good sequential and functional rea-
sons for a lack of orientation to resolution.
One type of contrast which was left unelaborated involves connec-
tives used as markers of discourse organization, without any literal con-
trast. Contrastive marking works as a format for text cohesion without a
concomitant expression of a problem or other content in need of resolu-
tion. Thus, while most of the contrasts in the collection involved some
contrastive marking and some real incompatibility at the content level,
several cases involve conjunctions serving text organizational functions
but not encoding any content level incompatibility. In such cases,
contrastive marking structures chunks of discourse, rather than focusing
on contrasts in meaning. In (12), the science teacher is involved in the
activity of discussing itemized tasks with which the students have
already been working. He refers briefly to one item and then moves to
another, using contrastive markers as a way of shifting the addressees'
attention, without drawing any problematic incompatibility into focus.
In introducing the next item in his list, he uses contrastive expressions
and marked stress: "Now this one, (.) but this one." In addition to the
use of but, we can see a use of now which Schiffrin (1987:234-40)
describes as expressing "a speaker's progression through the discourse
time of a comparison."

(12) T's first and second uses of this refer to

different tasks on which the students are working
T: After today this one will be easy. Now this
one, (.) but this one >I don't want to take
an hour on it.<
The treatment ofcontrasts in interaction 301

After using contrastive devices to introduce the shift "in discourse time"
to a next item, the teacher does not go on to produce any kind of expla-
nation or solution. The contrast functions only at a discourse organiza-
tional level, and subsequent talk does not attend to any content-level
More interesting than the discourse structuring uses of contrastive
marking are cases in which sequential context, and the related factor of
speaker role in a local interactional activity or setting, are associated
with unelaborated contrast. Contrasts which come at the close of story-
telling sequences, for example, appear to function more as concluding
assessment than as focused expressions of problems which might call
for further elaboration and resolution. (13) is such a case. T has been
telling a story about her partner P's nephew; R and A respond with turns
which deliver displays of appreciation, action types which are typically
produced in the sequential contexts of story endings. In this case, the
displays are done through negative statements:

T: He was cute,>you know, sending the thank you
card before Ch(h)r(h)istm(h)as(h) ,<
hh[uh huh huh huh
R: [Yeah (h) huh huh hah
R: Stra[tegic.
T: [( )
T: We saw right through him.
=> R: [He is no: t a dumb kid.
A: rYe [h
T: [eh heh huh huh huh huh huh.
(): .hhh
=> R: [No sense in sending i t in August.
=> A: [Well at least he didn't send his list to
T: =That's right
A: °Yeah. °

While there are contrasts implicit in these negations, they are not treated
as invoking any need for explanation or solution. This seems related to
their coming as contributions to an assessment sequence at the close of a
story telling. As such, rather than presenting focused topics of talk on
~02 Cecilia E. Ford

their own, they are specifically produced as positive and agreeing re-
sponses to the upshot of T's story, a story about a clever and somewhat
manipulative child (note T's "We saw right through him"). In the sense
that they are affiliative with the teller's displayed stance and with the
story's gist, contrasts in such contexts are probably less likely to be
treated as needing resolution or even any special attention or elaboration
in themselves. That is, they may well express incompatibility at some
level (incompatibility with the expectation that children are guileless,
and that thank you notes should be sent soon after the action for which
they express gratitude); however, these turns are fully aligned interac-
tionally with the stance displayed in the immediately preceding talk. 7
In another kind of unelaborated contrast, the lack of orientation to
resolution seems to serve as an enactment or indexing of authority. In
the classroom context, there are explicit rules of conduct and there is a
clear authority figure. There are instances in which the teacher makes a
negative command or when a student makes a command which is based
on institutional rules, but no explanation or solution follows. While
there may be incompatibility expressed, the call for elaboration is art-
fully ignored in a manner which claims or underscores institutionally
supported rules and authority.
In one case, a student says to her classmates, "No talking." As this is
a rule during specific classroom activities (though one that is often vio-
lated), it is not surprising that the student offers no further explana-
tion-the explanation being treated as self-evident from shared knowl-
edge. And when the teacher at one point says, "Don't complain", in fail-
ing to produce a subsequent explanation, he may well be enacting his
role as authority. These cases suggest there may be institutional contexts
and specific activities (issuing directives based in or with reference to
institutional authority) in which a lack of explanation or solution is in
fact a resource for aligning with or reinforcing authority. Thus, explana-
tions may be relevantly absent in certain types of turns produced by per-
sons who are either referring to institutional authority or enacting it. It
must be noted, however, that this pattern is by no means absolute; recall
example (4), in which the teacher offers an explanation for his prefer-
ence for having the lights on, "I can see."
Related to the enactment of authority are cases of correction, them-
selves expressions of special access to relevant knowledge. (14) is a
case of third tum repair, involving a speaker repairing a misunderstand-
ing of her tum after the recipient has produced a responsive tum (Sche-
gloff 1992):
The treatment ofcontrasts in interaction 303

R: Ohm (.) who drew the last time. me or her.
T: I did.
( .)
T: very poorly.
=> R: No.=Me or her.

In the game they are playing (Pictionary) the drawing of pictures alter-
nates between teams and between teammates. Interpreting R's question
is complicated by the fact that both teammates and teams alternate in
this way. T understands R's first tum to be asking who among all the
players drew the last time. It turns out that R does not want to know
which individual among all the players was the last to take a tum; R's
question pertains to her team only: Was it she or her teammate who
drew last? The overlapped "me or her" in the first line specifies this ver-
sion of the question, but T is already responding at that point.
In the contrasting tum, R offers a repeat of her own previous tum,
and in so doing, she speaks from authority-having exclusive access to
what she actually meant in the first place. An explicit explanation in this
context would likely seem irrelevant and redundant (e.g., "No. Me or
her, because that's what I meant").
Finally, as a corollary to the treatment of contrasts as problems, there
are cases where contrasts express complaints or troubles-tellings (Jeffer-
son and Lee 1992). It can be a delicate matter whether or not the recipi-
ent of a complaint should offer a solution ("advice giving") or should
take the role of a sympathetic listener ("troubles recipient"). In (15), A
is describing an ongoing dispute in her extended family. The contrast
involves the families' living remarkably close to each other but still ar-
guing about where to spend their time:

A: ... she an'- her husband, (.) her parents
and her husband's parents, live within
~ six blocks of each other here in town?
CONTRAST'iR: Oh (.) hm hmmm
A: and ih- (.) where they spend most of their
time is like this major bone of
>contention about whether they spend time
at ~ parents' house< or his parents'
304 Cecilia E. Ford

( .)
A: Six blo[cks-
R: [Six blocks away?
A: °Yeah. °
A: °It's insane.o

The incompatibility lies in the contrast between the proximity of the

dwellings and the existence of tension about where time is spent. In this
instance, the contrast is indeed treated as problematic, but this treatment
is not made evident though any following explanation or a solution.
Note that after presenting the contrast, A repeats "Six blocks," a key
element of the contrast. With the repeat, A prompts R for a response
which would align with the remarkableness of the proximity of the
families relative to the intensity of their issue regarding where to spend
time. R produces a surprised uptake, "Six blocks away?", in overlap
with A's prompt. After a pause, A goes on to express her stance even
more directly by assessing the situation negatively with "It's insane."
A's prompt, R's surprised uptake, and A's negative assessment all attest
to the fact that the contrast is interpreted as presenting a problem. Yet
neither participant produces a move toward a resolution.
This case suggests that contrasts interpreted as being in the service of
troubles telling may be less likely to be followed explanations or solu-
tions than are contrasts in other interactional activities. It is precisely the
lack of explanation or solution that makes the state of affairs count as
"trouble." Relevant to this analysis is the work of Jefferson and Lee
(1992), who illustrate similarities between advice-giving and troubles-
telling sequences. In (15), by taking the stance of a surprised recipient,
R embodies the role of "troubles recipient" rather than "advice giver."
Similarly, A's subsequent negative assessment treats the contrast as a
problem, but in this case as a problem in a "troubles-telling sequence"
rather than as an occasion for searching out remedies; A does not con-
textualize herself as one in search of a solution. While R is not strongly
affiliating as a sympathetic recipient here, at the same time, she does not
offer any solution or explanation.
It may be the case, then, that when contrasts are used in the service
of expressing troubles or complaints, the absence of an orientation to the
relevance of a solution or explanation reinforces the complaining force
of the contrast. The problematic incompatibility leads to further nega-
tive assessment rather than moves toward resolution.
The treatment ofcontrasts in interaction 305

The examples and discussion in this section underscore the fact that
contrasts are by no means always followed by explanations or solutions,
or even by clear orientations to the need for resolution. The contexts and
functions documented here provide a wider view of the uses of contrasts
in interaction, an area worthy of further study. Contrasts can be used in
text organizational functions, exploiting the contrastive format to
achieve a shift in focus without placing the contrast itself in focus. Such
subordination of the role of the contrast relative to a broader context of
activity is also seen in story closings. In these contexts, interactants may
use negative statements as presentation formats for otherwise agreeing
or affiliating turns without implicating any need to focus on the con-
trasts themselves as topics for further development.
Contrasts can also express commands and perform repairs. The
speakers of such turns act in the role of authority, act with reference to
shared institutional rules, or display their special access to knowledge
precisely through not explaining their contrastive moves. That is, the
authoritative stance can be seen as both associated with and enacted
through the absence of explanation. Finally, contrasts may be used in
the service of troubles telling. The lack of a move toward resolution is
one of the interactional features that distinguishes troubles-telling from
advice-giving sequences. Instead of proposing explanations or solutions,
interactants produce assessments which underscore treatment of the
contrast as indeed expressing trouble not easily remedied.

8. Conclusion

In this study, I have examined some of the functions of contrasts in in-

teraction, using the talk that follows contrasts as a special source for in-
terpretation. I have found significant orderliness associated with uses of
contrast, with explanations and solutions either coming immediately
after contrasts or oriented to as being expected and relevant in such con-
texts. In the present data, I have also observed that contrasts without
elaborations are artfully used for a variety of functional ends. In some of
these instances-commands from authority, other correction, and trou-
bles telling-particular actions seem to achieve their functional force at
least in part precisely through the absence of moves toward explanation
or solution. Thus, contrasts may still be associated with explanations
and solutions, but a speaker can index the authority from which s/he is
speaking by strategically producing an unelaborated contrast, or she
306 Cecilia E. Ford

may display an interpretation of a problem statement as a complaint by

offering a show of sympathy rather than a move toward remediation.
Finding evidence that CONTRAST + EXPLANATION and
CONTRAST + SOLUTION constitute schemas-recurrent and expect-
able rhetorical clusters-builds a foundation for further research into the
contexts for and encoding of such rhetorical constellations. Logical next
steps in this research agenda will be several: A first will be the study of
linguistic coding options for these paired actions, and the circumstances
in which different coding options are used. When and to what ends are
contrasts and explanations explicitly marked with contrastive or causal
connectives? Looking at languages other than English, how does this
action sequence fare cross-linguistically? Thinking diachronically, one
could explore whether these recurrent rhetorical combinations leave a
mark on linguistic practices over time (Hopper and Traugott 1993). At a
more sociolinguistic level, in reference to the cases of troubles telling
and speaking with institutional authority, it would be revealing to know
how interactional roles and relationships are invoked and performed
through contrasts as well as through other linguistic and pragmatic
means. Perhaps demanding explanations or solutions serves as a vehicle
for challenging the roles invoked in these contexts (e.g., troubles recipi-
ent, commanded/controlled subject).
In responding to an earlier version of this paper, Susanne Giinthner,
Harrie Mazeland, and Elizabeth Couper-Kuhlen each noted that con-
trasts are not actions in the same sense that explanations or proposals for
remedies (solutions) are. In rethinking this research, I have taken this
point most seriously; it suggests that while contrasts, solutions, and ex-
planations are all relations which language users can be shown to em-
ploy in constructing turns at talk, these categories of relation are in-
vented by analysts, and there is good reason to question the degree to
which these categories reflect a comparable type of rhetorical connec-
tion (see also Ford 1986 for a related critique of RST). In the current
version of this paper, I have taken pains to consider what is made evi-
dent through the talk of the participants themselves. What I hope to have
demonstrated is that, in treating contrasts as needing solutions and ex-
planations, the participants in these interactions themselves offer us a
basis for the functional analysis of these contrasts.
The treatment ofcontrasts in interaction 307


* I thank Elizabeth Couper-KuWen, Donna Dallos, Christine GoW, Susanne

Giinthner, Bernd Kortmann, Harrie Mazeland, and Sandra Thompson for help
in my thinking for this research and for feedback on various drafts of the pa-
per. I take full responsibility for my use of their feedback.
1. Mann and Thompson's definition of "concession" involves an acknowledge-
ment of "potential or apparent incompatibility" between two units of text be-
ing compared (1988). Given that this relation has the expression of contrast
within it, I included such cases in my collection (see the concession, "I look
smaller" in (5), below).
2. In another investigation, I am looking at the initial parts of contrasts. In this
instance, the use of the perception verb, sound, expresses an evidential dis-
tance between what is apparent and what is real, projecting that what is real
may be expressed next. The second part of the contrast makes good on this
3. These data come from a longitl;ldinal, ethnographic study headed by Jane
Zuengler at the University of Wisconsin-Madison.
4. The laboratory interaction is analyzed in Ford (1999).
5. Transcription symbols
Symbol Interpretation
(.) A short, untimed pause
(0.3) A timed pause
hhh Audible breath
thi- Hyphen indicates a sound cut off
Latching, rush into next turn or segment
[ Brackets indicate the onset of overlap
she Prominent stress
SHE Higher volume than surrounding talk
°sheo Lower volume than surrounding talk
she: Sound stretch
Low falling intonation
? High rising intonation
Intermediate intonation contours:
level, slight rise, slight fall
>talk< At a faster tempo than surrounding talk
6. While contrasts formed the most prevalent textual environment for subse-
quent use of because, that study also pointed to "strong evaluations" as regu-
larly followed by because-prefaced continuations (Ford 1994:548).
7. I know of no studies looking specifically at the use of negative statements in
affiliative assessments such as these.
308 Cecilia E. Ford

Atkinson, J. Maxwell
1984 Public speaking and audience responses: some techniques for invit-
ing applause. In: 1. Maxwell Atkinson and 1. Heritage (eds.), Struc-
tures of Social Action, 370-410. Cambridge: Cambridge University
Atkinson,1. Maxwell and John Heritage (eds.)
1984 Structures ofSocia/Action. Cambridge: Cambridge University Press.
Cumming, Susanna and Tsuyoshi Ono
1997 Discourse and grammar. In: T. A. van Dijk (ed.), Discourse as Struc-
ture and Process, 112-137. London: Sage.
Eggins, Suzanne and Diana Slade
1997 Analysing Casual Conversation. London: Cassell.
Fairclough, Norman
1989 Language and Power. New York: Longman.
Ford, Cecilia E.
1986. Overlapping relations in text structure. In: Scott DeLancey and Rus-
sell S. Tomlin (eds.), Proceedings ofthe Second Annual Meeting of
the Pacific Linguistics Conference, 107-123. University of Oregon:
Department of Linguistics.
1993 Grammar in Interaction: Adverbia/ Clauses in American English
Conversations. Cambridge: Cambridge University Press.
1994 Dialogic aspects of talk and writing: because on the interactive-
edited continuum. TEXT 14(4): 531-554.
1999 Collaborative construction of task activity: Coordinating multiple
resources in a high school physics lab. Research on Language in So-
cia/Interaction 32 (4): 369-408.
Ford, Cecilia E., Barbara A. Fox, and Sandra A. Thompson
1996 Practices in the construction of turns: the "TCU" revisited. Pragmat-
ics 6(3): 427-454.
Fox, Barbara A.
1987 Discourse Structure and Anaphora: Written and Conversational
English. Cambridge: Cambridge University Press.
Given, Talmy
1978 Negation in language: Pragmatics, function, ontology. In: Peter Cole
(ed.), Syntax and Semantics, Vol. 9,69-112. New York: Academic
Goodwin, Charles
1979 The interactive construction of a sentence in natural conversation. In:
George Psathas (ed.), Everyday Language: Studies in Ethnomethod-
ology, 97-121. New York: Irvington.
1981 Conversational Organization: Interaction between Speakers and
Hearers. New York: Academic Press.
1995 Co-constructing meaning in conversations with an aphasic man. Re-
search on Language and Socia/Interaction 28(3): 233-260.
The treatment ofcontrasts in interaction 309

Goodwin, Marjorie Harness

1990 He-Said-She-Said: Talk as Social Organization among Black Chil-
dren. Bloomington, Indiana: Indiana University Press.
Halliday, M. A. K., and Ruqaiya Hasan.
1976 Cohesion in English. London: Longman.
Have, Paul ten, and George Psathas (eds.)
1990 Studies in the Social Organization of Talk and Embodied Activities.
Washington, D.C.: University Press of America.
Hopper, Paul J.
1998 Emergent grammar. In: Michael Tomasello (ed.), The New Psychol-
ogy of Language: Cognitive and Functional Approaches to Lan-
guage Structure, 155-176. Mahwah, New Jersey: Erlbaum.
Hopper, Paul J., and Elizabeth C. Traugott
1993 Grammaticalization. Cambridge: Cambridge University Press.
Jefferson, Gail
1973 A case of precision timing in ordinary conversation: overlapped tag-
positioned address terms in closing sequences. Semiotica 9(1): 47-
1978 Sequential aspects of story telling in conversation. In: J. Schenkein
(ed.), Studies in the Organization of Conversational Interaction,
219-248. New York: Academic Press.
Jefferson, Gail and John R. E. Lee
1992 The rejection of advice: managing the problematic convergence of a
"troubles-telling" and a "service encounter" In: Paul Drew and John
Heritage (eds.), Talk at Work: Interaction in Institutional Settings,
521-548. Cambridge: Cambridge University Press.
Lemke, Jay L.
1990 Taking Science: Language, Learning, and Values. Norwood, NJ:
Lerner, Gene H.
1991 On the syntax of sentences-in-progress. Language in society 20(3):
1996 On the "semi-permeable" character of grammatical units in conver-
sation: conditional entry into the tum space of another speaker. In:
Elinor Ochs, Emanuel A. Schegloff, and Sandra A. Thompson (eds.),
Interaction and grammar, 238-77. Cambridge: Cambridge Univer-
sity Press.
Lerner, Gene H., and Tomoyo Takagi
1999 On the place of linguistic resources in the organization of talk-in-
interaction: A co-investigation of English and Japanese grammatical
practices. Journal ofPragmatics 31(1): 49-75.
Longacre, Robert E.
1983 The Grammar ofDiscourse. New York: Plenum.
Levinson, Stephen C.
1983 Pragmatics. Cambridge, UK: Cambridge University Press.
310 Cecilia E. Ford

Mann, William C., and Sandra A. Thompson

1988 Rhetorical structure theory: toward a functional theory of text or-
ganization. TEXT 8(3): 243-81.
Mann, William C., Christian M.I.M. Matthiessen, and Sandra A. Thompson
1992 Rhetorical structure theory and text analysis. In: W.C. Mann and S.
A. Thompson (eds.) Discourse Description: Diverse Linguistic
Analyses ofa Fund-raising Text, 39-78. Amsterdam: Benjamins.
Ochs, Elinor, Emanuel A. Schegloff, and Sandra A. Thompson (eds.)
1996 Interaction and Grammar. Cambridge: Cambridge University Press.
Orestrom, Bengt
1983 Turn-taking in English Conversation. (Vol. 66). Lund: CWK
PerakyHi, Anssi
1997 Reliability and validity in research based on tapes and transcripts. In:
David Silverman (ed.), Qualitative Research: Theory, Method and
Practice, 201-220. London: Sage.
Rudolph, Elizabeth
1996 Contrast: Adversative and Concessive Relations and their Expres-
sions in English, German, Spanish, Portuguese on Sentence and Text
Level. New York: Gruyter.
Sacks, Harvey
1987 On the preferences for agreement and contiguity in sequences in
conversation. In: Graham Button and John R. E. Lee (eds.), Talk and
Social Organization, 54-69. Clevedon, England: Multilingual Mat-
Sacks, Harvey, Emanuel A. Schegloff and Gail Jefferson
1974 A simplest systematics for the organization of turn-taking for con-
versation. Language 50(4): 696-735.
Schegloff, Emanuel A.
1980 Preliminaries to preliminaries: can I ask you a question? Sociological
Inquiry 50: 104-52.
1988 Discourse as an interactional achievement II: an exercise in conver-
sation analysis. In: Deborah Tannen (ed.), Linguistics in Context:
Connecting Observation and Understanding, 135-158. Norwood,
NJ: Ablex.
1992 Repair after next turn: the last structurally provided defense of inter-
subjectivity in conversation. American Journal of Sociology 97(5):
1996 Turn organization: one intersection of grammar and interaction. In:
Elinor Ochs, Emanuel A. Schegloff, and Sandra A. Thompson (eds.),
Interaction and Grammar, 52-134. Cambridge: Cambridge Univer-
sity Press.
Schiffrin, Deborah
1987 Discourse Markers. Cambridge: Cambridge University Press.
The treatment ofcontrasts in interaction 311

Selting, Margret
1998 TCUs and TRPs: The Construction of Units in Conversational Talk.
InLiSt NO.4. (Interaction and Linguistic Structures.) Sekretariat LS
Angl. Sprachwissenschaft. Konstanz, Germany: Universitat Kon-
Tannen, Deborah
1979 What's in a frame? Surface evidence for underlying expectations. In:
Roy A. Freedle (ed.), New Directions in Discourse Processing, 137-
181. Norwood, New Jersey: Ablex.
Tomlin, Russell S., Linda Forrest, Ming Ming Put, and Myung Hee Kim
1997 Discourse semantics. In: T. A. van Dijk (ed.), Discourse as Structure
and Process, 63-111. London: Sage.
Whalen, Jack
1992 Conversation analysis. In: E. F. Borgatta and M. L. Borgatta (eds.),
Encyclopedia ofSociology, 303-310. New York: Macmillan.
Concessives on different semantic levels:
A typological perspective *

Mily Crevels

It will be argued (a) that concessive clauses are expressed on four different
semantic levels: the content level, the epistemic level, the illocutionary or
speech-act level and the textual level, and (b) that there exists a systematic
cross-linguistic correlation between the semantic level at which adverbial
clauses apply, on the one hand, and the way they are expressed, on the
other. An illustration of this correlation is the fact that languages may use
different conjunctions to express adverbial relations if these apply at
different levels. The discussion of three hypotheses stands at the centre of
this paper (section 3). These hypotheses have been formulated to investigate
whether, depending on the semantic level, concessive clauses show
distinctive formal properties and have been tested against data from a wide
range of languages.

1. Concessive relations

I will start by discussing the major properties of conceSSIve

constructions, drawing heavily on Konig (1986, 1988 and 1994).
Someone who produces an utterance of the general format under (1), is
committed to the truth of both propositions (cf Konig 1988: 146).

(1) Although p, q

An example is given in (2):

(2) Although it is raining, I am going outfor a walk.

Moreover, a connection is implied between the propositions of the two

related clauses in question: the speaker asserts these two propositions
against the background assumption that the two types of situations
which p and q describe are generally incompatible. This implication has
the status of a presupposition rather than of an entailment (cf Konig
1986: 233). For an example such as (3a) the concessive presupposition
may be expressed as in (3b):
314 Mily Crevels

(3) a. Even though he had not eaten for days, he looked strong and
b. {f one does not eat for days, one normally does not look strong
and healthy.

Konig (1988: 147) formulates the relevant presupposition as follows:

(4) If p', then normally --q'

Concessive clauses occur in all positions where adverbial clauses are

permitted in a language. In many languages they may either precede or
follow the main clause. However, as Konig (1994: 679) has pointed out,
concessive clauses differ from other types of adverbial clauses in a
number of ways: (i) in contrast to most other types of adverbial clauses,
there do not seem to be many languages with a concessive interrogative
pronoun analogous to English when (Time), why (Reason), how
(Manner), etc., (ii) concessives cannot be the focus of a focusing adjunct
(focus particle) like only, even, just, especially (in contrast to a causal
construction as in (5a», (iii) concessives cannot occur as focus in a cleft
sentence (5b) and (iv) concessives cannot be the focus of a negation or a
polar interrogative (5c).

(5) a. i. Only because it was raining ...

ii *Only although it was raining ...
b. i. It was because it was raining that ...
ii *It was although it was raining that ...
c. i. Was he harassed because he was a boxer?
ii ?Was he harassed although he was a boxer?

As Konig claims, all of these divergent properties seem to be

manifestations of a single syntactic/semantic l constraint on the use of
concessive clauses: they cannot be focused. This constraint is generally
considered to indicate that the relevant clauses are less tightly integrated
into a main clause than other types of adverbial clauses.
Concessives on different semantic levels 315

2. Distinct levels of linking

2.1. Sweetser's domain theory and layers in Functional Grammar

Sweetser (1990) clearly demonstrates the contrast between root,

epistemic and 'speech-act' uses of modal verbs and multiple uses of
conjunctions, such as conditional if. Linking between three different
types of entities, i.e. (i) real or hypothetical situations, (ii) aspects of
knowledge or (iii) speech acts, therefore takes place at the content, the
epistemic or the illocutionary level. In Crevels (1994, 1998) I have
shown that these distinctions may also be made for concessive
constructions in Spanish, and that Sweetser's multiple semantic
domains correspond, respectively, to the predicational, the propositional
and the illocutionary layers within the layered structure of the clause in
Functional Grammar (cf Dik et al. 1990). The idea of this structure is
that layers of lower complexity are fully contained within layers of
higher complexity. Thus, every speech act contains a proposition, and
every proposition contains a predication. In addition, a fourth layer, the
so-called text layer (Crevels 1994, 1998, 2000; Hengeveld 1997), is
required. Concessive clauses pertaining to this layer have to do with the
organization of the discourse and, therefore, apply to text units which
may contain more than one sentence.
In Functional Grammar, six layers are claimed to be relevant for the
analysis of clause structure in natural languages (Hengeveld 1989, 1992,
1997; Crevels 1994, 1998, 2000). As Hengeveld (1996: 119) points out,
a serious problem lies in providing evidence for the validity of each of
these layers in main clauses. An important way to solve this problem is
to study the properties of subordinate constructions. Since subordinate
constructions may be classified according to the highest layer they
contain, and since each of the layers present in the hierarchical clause
model may be turned into a subordinate construction, the study of
different types of subordinate constructions will lead to a better
understanding of the differences between the layers.
A strong argument in favour of the existence of the different
semantic levels is the fact that languages may use different conjunctions
to express the same kind of adverbial relation. French parce que
'because' is used specifically for content conjunction, while puisque is
the correct causal conjunction at the epistemic or illocutionary level (cf
Sweetser 1990: 82, 156).
316 Mily Crevels

(6) a. II va I 'epouser parce qu'ill 'adore.

'He's going to marry her because he adores her.'

b. (Mais si,) il va I 'epouser, puisqu'ill 'adore.

'(But of course,) he's going to marry her, since he adores her.'

Likewise, Latin quoniam, English since and Spanish ya que show a

strong tendency towards an epistemic or speech-act reading, rather than
toward a content-conjunction reading (Quirk et al. 1985; Bolkestein
1991; Crevels 1994, 1998).

2.2. Entity types

Extending the analysis proposed by Lyons (1977: 442-447), Hengeveld

(1989, 1992, 1998) classifies the semantic types of adverbial clauses on
the basis of the entity types they designate. Linguistic units may refer to
entities of six different types. In view of the text layer, Hengeveld's
typology of entities may, therefore, be extended by adding fifth order
entities, which consist of text units that can be evaluated in terms of
their thematic continuity (cf Table 1).

Table 1. Entity types

Entity type Description Evaluation

Zero order Property or Relation Applicability
First order Individual Existence
Second order State of affairs Reality
Third order Propositional content Truth
Fourth order Speech act Felicity
Fifth order Text unit Thematic Continuity

2.3. Integration

Zero and first order entities cannot be realized by concessive adverbial

clauses, but the other four types are indeed relevant. Examples are given
in (7):
Concessives on different semantic levels 317

(7) a. Although it's raining, we're goingfor a walk.

b. He's not at home, although his car is parked in front ~f the
c. Even though I am calling a bit late, what are your plans for
this evening?
d. I speak and write Serbian, Albanian, Turkish and Dutch, but I
cannot express my true feelings in any other language than
~omani. .A lthouF now that I come to think of it, I have done
It many tImes ...

The difference between (7a) and (7b) corresponds to the distinction

Lyons (1977) makes between second and third order entities, i.e.
between states of affairs and propositional contents. The concessive
construction in (7a) is an example of what Sweetser calls a real world or
content relationship: although in the real world the rain indeed forms an
obstacle, it nevertheless cannot prevent our going for a walk. In (7b) the
concessive construction concerns an epistemic relationship: the
concessive connective combines two items of knowledge, a premise and
a conflicting conclusion. Thus, in this particular example, the speaker,
even though s/he knows that the car of the person in question is parked
in front of his house, reaches the conflictive conclusion that this person
is not at home. In (7c) we find an example of an illocutionary
relationship; in this case the concessive clause forms an obstacle to the
speech act expressed in the main clause, a possible paraphrase being: 'I
know that 1 should have phoned you sooner, so 1 normally wouldn't
phone you this late to ask you what you are doing this evening'.
Concessives at this speech-act level are often closely related to the
conversational maxims of Grice (1975); thus, the concessive clause in
(7c) appeals to Grice's principle of informativeness, which in some
cases allows the hearer to read into an utterance more information than
it actually contains. In (7d), fmally, an example of a textual relationship
is given. In this case the concessive clause Although, now that I come to
think of it, I have done it many times ... stretches over a whole series of
preceding utterances, signalling an unexpected turn in the discourse
context, a possible paraphrase being 'I come to the conclusion that 1
have expressed my true feelings many times in Serbian, Albanian,
Turkish and Dutch, so 1 normally wouldn't first say that I speak and
write Serbian, Albanian, Turkish and Dutch, but that I cannot express
my true feelings in any other language than Romani' .
318 Mily Crevels

2.4. A four-level approach to concessives

In the content domain a concessive connection indicates that the event

or the state of affairs described in the concessive clause forms an
obstacle for the event or the state of affairs described in the main clause,
but does not impede its realization.

(8) She's just given birth to a beautiful baby girl although she's fifty-

In the epistemic domain concessive connection expresses the idea that

the speaker, in spite of being convinced of the content of the concessive
clause, still reaches the opposite conclusion contained in the main
clause. In other words, in the epistemic domain concessive conjunction
will mark the impediment of a belief or a conclusion. Example (9) does
not express any factual conflict, but a conflict between the conclusion
and the potential counterargument expressed in the concessive clause.

(9) He left his wife and children, although he loved them very much.

In the speech-act domain the content of the concessive clause does not
form an obstacle for the realization of the event or the state of affairs
described in the main clause, but raises obstacles for the realization of
the speech act expressed by the speaker in the main clause.

(10) Although it's none ofmy business, your behaviour is a disgrace.

The speaker's recognition of the fact that s/he has no business

commenting on the interlocutor's behaviour is in no way incompatible
with the speaker's belief that the interlocutor's behaviour actually is a
disgrace. A possible paraphrase of (10) might be 'I know that 1 have no
business commenting on your behaviour, so 1 normally wouldn't do so',
or 'I know that 1 have no business commenting on your behaviour, so 1
normally wouldn't perform the speech act of commenting on it'.
Let us consider, finally, concessive conjunction at the text level,
where the modification is based on an unforeseen turn in the discourse

(11) My favourite poster is, I think, a French one for Nesquik, which
shows a sophisticated-looking small boy leaning nonchalantly
Concessives on different semantic levels 319

against something and saying that thanks to Nesquik he went hack

on to milk. He really looks a nice child. Though, there are some
Adchildren that one would feel quite ashamed to have around the
house ... (Greenbaum 1969: 68)3

In contrast to illocutionary concessives, textual concessives do not

modify the main clause of a concessive construction, but generally a
whole preceding text unit which may be composed of various sentences.
lliocutionary concessives are always part of a single construction, and,
therefore, specify or modify a single preceding or following speech act;
textual concessives, on the other hand, usually modify a whole series of
preceding propositions, rather than just single utterances and often seem
to be functioning as an afterthought.
Although in written language one is confronted with the bias of
punctuation conventions, textual concessive clauses do not always
manifest themselves in the form of a simple sentence. Moreover, they
typically exhibit main clause word order in those languages, like Dutch
and German, where main and subordinate clauses are distinguished on
the basis of word order (Konig and van der Auwera 1988, Giinthner
1993). In (12a) we have an example of a concessive with main clause
word order, whereas (12b) shows the typical V-2 order:

Dutch (Indo-European)
(12) a. Ze komt vast en zeker; hoewel, hi} haar kan je maar nooit
weten ...
'She will certainly come; though, one can never be sure with
her ... '
b. Ze komt vast en zeker, hoewel je hi} haar maar nooit weten kan

'She will certainly come, although one can never be sure with
her ... '

This difference in word order between main and subordinate clauses

may well be explained by so-called iconicity principles, or, in other
words, principles which govern natural form-function correlations (cf.
Givan 1984: 30-31).
Different subtypes of concessive clauses manifest different degrees
of subordination to and integration into a main clause. As mentioned
above, concessive clauses cannot be focused--even in their standard
use--and tend to take wide scope over any operator in the main clause.
320 Mily Crevels

Therefore, it is assumed that they are less tightly connected to a main

clause than conditional, temporal or causal clauses. The higher the
semantic level they pertain to, the looser they seem to be integrated into
a main clause and the more paratactic-like the construction becomes.

3. Hypotheses

On the basis of the discussion in section 2.4 it is now possible to

investigate whether the expression of concessive clauses at the four
semantic levels shows distinctive formal aspects. To do this I have
constructed a small questionnaire containing nine concessive
constructions in English at the four relevant levels with two content, two
epistemic, three speech-act and two textual concessives, respectively.
With the help of native speakers and specialized linguists I have
collected data on a sample of 30 extant languages. 4 The consultants
were requested to express the concessive constructions in their own
language and to use a concessive linker where possible. The data of the
sample languages will now be analyzed on the basis of three specific
hypotheses which may be subjected to further typological investigation.
The basis for each of these hypotheses is constituted by the following
hierarchy of semantic levels, which is based on the concessive relations
given in (7):

(13) Content level> Epistemic level> lliocutionary level > Text level

The content level, which is the semantically least complex level, is

followed by the semantically more complex epistemic level, which in
tum is followed by the illocutionary or speech-act level, and, finally, the
text level. In the concrete application of this hierarchy we will see that
in each case there is a correlation between the semantic level of the
concessive construction and its formal expression. Thus, by testing this
hierarchy it will be possible to establish the validity of each of the four
categories which belong to it. Of the three hypotheses that are to be
tested, the first two have a bearing on the way concessive constructions
are marked, while the third one has to do with the nature of concessive
Concessives on different semantic levels 321

3.1. The marking ofconcessive constructions

3. 1. 1. Syndetic versus asyndetic constructions

The first hypothesis to be tested ties in with other observations and

hypotheses in the domain of clause-linking in functional typology (cf
van der Auwera (ed.) 1998; Devriendt et al. (eds) 1996; Haiman and
Thompson 1988). If we may assume that a concessive clause becomes
less and less integrated into its main clause and more and more
paratactic-like the higher its semantic level is, we may assume at the
same time that languages are more likely to use asyndetic strategies to
express a concessive construction if this construction belongs to a
higher semantic level. Thus the first hypothesis, represented in (14),
may be formulated as follows:

The probability that a concessive construction will be expressed by
asyndetic means increases as the construction pertains to a higher
semantic level.

(14) Content > Epistemic > Illocutionary > Text

Syndetic > > > Asyndetic

Table 2 provides a schematic representation of the clause combining

strategies used to express the various concessive constructions: '-'
stands for syndetic and '+' for asyndetic concessive linking. For
instance, in Kiwai-a complex Papuan language spoken in south-
western Papua New Guinea, Western Province, Fly River
Delta--syndetic and asyndetic constructions may occur at the speech-
act and text level. This implies that my consultants in some cases use
syndetic means to express an elicited illocutionary concessive
construction, and on other occasions resort to asyndetic means.
The results of the testing of Hypothesis 1 are interpreted as evidence
for three cut-off points within this hierarchy, since, as can be seen from
the bottom end of Table 2, in Lakxota (South Dakota, USA) asyndetic
linking may take place at the epistemic, the speech-act and the text
level, while Pima (Arizona, USA) and Kiwai show signs of asyndetic
marking at the speech-act and text level, and in Hungarian, Boboda
(Burkina Faso), Kannada, Burmese, Turkish, Uzbek and Mongolian
asyndetic concessive constructions only occur at the text level. The
322 Mily Crevels

almost extinct Australian Aboriginal language Duunggidjawu from

south-eastern coastal Queensland provides a good example of a
language which uses juxtaposition of clauses as a clause-combining
strategy for all concessive constructions. Examples (15) and (16) show
that although in Duunggidjawu the word order is very free and varying,
as is the case in many Australian languages, it is rigid for the expression
of concessive clauses, which are always preposed.

Duunggidjawu (Australian; Stephen Wurm p.c.)

(15) dyinnang jO:ri wakka galang yingi,
foot his NEG good be:INCPAST

guyumgu biye dyingi.

camp.to returning he. went.there
'He went back to the camp although he had broken his foot.,5

(16) gurangur jo: ri minna na barang jO: ri

spear his this and boomrang his

ganna yingi, jO: wakka guyum-gu.

there are he NEG camp-in
'He's not in the camp, although his spear and his boomerang are
there. '

Table 2. Schematic representation of asyndetic constructions

Language Content Epistemic Illocutionary Text

Bahasa Indonesia
Romani (Erli)
Concessives on different semantic levels 323

Language Content Epistemic Illocutionary Text

Hungarian -/+
Boboda -/+
Kannada -/+
Burmese -/+
Turkish -/+
Uzbek -/+
Mongolian +
Pima -/+ -/+
Kiwai -/+ -/+
Lakxota -/+ -/+ -/+
Duungidjawu + + + +

In Lakxota, which shows a cut-off point between the content and the
epistemic level, syndetic means are used at the content level, but an
asyndetic strategy is possible at the higher levels. Consider examples
(17) and (18) with a Lakxota content concessive and an epistemic
concessive, respectively:

Lakxota (Amerind; Violet Catches p.c.)

(17) Ishtagxungxa eyash, ishtamanza un shni.
blind although glasses wear NEG
'He doesn't wear glasses although he sees very little.'

(18) Txawicu na cinca tewichaxila nus,

wife and off-spring love supposedly

'He left his wife and children, although he loved them very much.'

As mentioned above, Kiwai shows a cut-off point between the epistemic

and the speech-act level, since syndetic means are used at the content
and the epistemic level, but an asyndetic strategy is possible at the
324 Mily Crevels

speech-act and the text level. Consider the epistemic concessive in (19)
and the speech-act one in (20):

Kiwai (Indo-Pacific; Stephen Wurm p.c.)

(19) Nanie nou kara moto tatan ai-r-otoi,
although his car house beside FAcT-itsBJ-stand

nou pai moto-wa.

he NEG house-in
'He's not at home, although his car IS parked in front of the

(20) Mo pai ubi ai-r-erea ro k-emarogo-gido,

I NEG wish fact-sBJ-be yOU:SG vN-scolding-ENWH

ro uba ai-r-erea.
yOU:SG bad FACT-sBJ-be
'Although I should be minding my own business, your behaviour
is a disgrace. '

Another cut-off point is the one between the illocutionary and the text
level. In Turkish, for example, syndesis is the only option on the
illocutionary level (21), while at the text level asyndetic linking is also
possible (22).

Turkish (Altaic; Hiiseyin Demirel p.c.)

(21) Beni ilgilendir-me-me-si-ne ragmen
me concem-NEG-VN-3SG:POSS-DAT although

senin hareket-ler-in {fok kOtii.

your behaviour-PL-2sG:Poss very bad
'Although I should be minding my own business, your behaviour
is a disgrace. '

(22) Ben ingilizce konuruyor ve yazl-yor-um

I English speak-PROG and read-PROG-ISG

ama his-ler-im-i tiirkr;e dl~lnda

but feeling-PL-ISG:poss-ACC Turkish except
Concessives on different semantic levels 325

ba~ka bir dil-de anlat-am 1-yor-um

other one language-Loc express-IMPOT-PROO-ISO

As11nda dii~n-iince birf(ok deja

actually think-OER several time

yap-t 19 -1m -1 hat1r11-yor-um.

do- VN -ISO:POSS- ACC remember-PROO-Iso
'I speak English, and I write it, but I cannot express my true
feelings in any other language than Turkish. Although, now that I
come to think of it, I have done it many times ... '

At the text level it is sometimes difficult to establish whether a

construction contains a concessive linker or not. Thus I have classified
(22) as an instance of asyndetic linking because as11nda does not
convey any concessive connotation and, moreover, because the
concessive nature of the sentence does not change when asl1nda is left

3.1.2. Clause marking

If it is true that concessive clauses get more and more paratactic-like as

they pertain to a higher semantic level, the following is to be expected:
in eliciting a subordinate construction of the general format 'although p,
q', the possibility that the informant will mark the q (apodosis) instead
of the p (protasis) increases as the construction pertains to a higher
semantic level. This allows us to formulate the following hypothesis
based on hierarchy (13):

The probability that in an elicited subordinate concessive construction
the q will be marked instead of the p increases as the construction
pertains to a higher semantic level. 6

(23) Content > Epistemic > lliocutionary7

p > > q

Table 3 shows a schematic representation of the clause-marking

strategies used to express the various concessive constructions. In
326 Mily Crevels

Mongolian, for instance, both p and q marking is possible at the speech-

act level (P/q). This means that my informants in some cases choose to
mark the p in order to express an elicited illocutionary concessive
construction, and in other cases the q (apodosis). Furthermore, pq stands
for correlative marking in which both the concessive subordinate clause
and its subsequent main clause are marked. IRR, finally, stands for the
absence of an overt concessive linker.
Again there is evidence for two cut-o