Studies in the Pragmatics of Discourse 9783110826142, 9789027932495

Table of contents :
Preface
Acknowledgments
Chapter one Towards a pragmatics of discourse
1. Introduction
2. Generative-transformational grammar and text grammar
3. Text grammar and pragmatics
4. Some issues in the pragmatics of discourse
Chapter two Pragmatics, presuppositions and context grammars
1. The status of pragmatics
2. Context grammars
3. Pragmatic structures
4. Texts and contexts
5. Semantics and pragmatics
6. Presuppositions in text grammars and context grammars
Chapter three A note on the partial equivalence of text and context grammars
Chapter four Acceptability in context
1. The problem
2. The philosophy of ‘acceptability’
3. Relative grammaticalness
4. Pragmatic acceptability
5. Macro-structures
6. Discourse and cognition
Chapter five Issues in the pragmatics of discourse
1. Sentences and sequences
2. Further pragmatic constraints on sequences
3. Acts and speech acts
4. Focus, perspective and related notions
5. Final remarks
Chapter six Pragmatic Connectives
1. Speech act sequences
2. Semantic connectives
3. Pragmatic connectives
4. Conclusions
Chapter seven Sentence topic and discourse topic
1. The problem
2. Sentence topic
3. Discourse topic
4. Conclusions
Chapter eight Pragmatic macro-structures in discourse and cognition
1. Introduction
2. Semantic macro-structures
3. Sentence sequences and speech act sequences
4. Macro-acts
5. Pragmatic macro-structures
6. Pragmatics and cognition
Chapter nine Context and cognition Knowledge frames and speech act comprehension
1. The foundations of pragmatics
2. The cognitive nature of pragmatic conditions
3. Pragmatic comprehension
4. Frames and speech acts
5. Context analysis
6. Examples of context analysis
7. The ‘dynamics’ of context
8. Utterance analysis
9. Comprehension of speech act sequences and macro-speech acts
10. Psychological process assumptions
Chapter ten The pragmatics of literary communication
1. Pragmatics
2. Literary communication
3. The pragmatics of literature
4. Concluding remark
Chapter eleven The semantics and pragmatics of functional coherence
1. The problem
2. Pragmatic theory
3. Semantic relations in discourse
4. Functional coherence at the semantic level
5. Speech act sequences and functional pragmatic coherence
6. Conclusions
Chapter twelve Towards an empirical pragmatics Some social psychological conditions of speech acts
1. Introductory summary
2. The need for an empirical pragmatics
3. The cognitive model of speech acts
4. Social psychological factors
5. Concluding remark
Chapter thirteen Conclusion, open problems, further prospects and future research
1. General conclusions
2. Open problems
3. Further prospects and future research
Bibliography
Subject Index

Citation preview

Studies in the Pragmatics of Discourse

JANUA LINGUARUM Studia Memoriae Nicolai van Wijk Dedicata edenda curat

C. H. van Schooneveld Indiana University

Series Maior 101

Studies in the Pragmatics of Discourse Teun A. van Dijk

Mouton Publishers The Hague · Paris · New York

Library of Congress Cataloging in Publication Data Dijk, Teun Adrianus, 1943Studies in the pragmatics of discourse. (Janua linguarum. Series maior ; 101) Bibliography: p. Includes index. 1. Discourse analysis - Collected works. 2. Pragmatics - Collected works. 3. Speech acts (Linguistics) - Collected works. I. Title. II. Series. P302.D473 40Γ.41 81-11213 ISBN 90-279-3249-2 AACR2

ISBN: 90-279-3249-2 © 1981, Mouton Publishers, The Hague, The Netherlands Printed in The Netherlands

Preface

In this book are collected the papers I wrote between 1973 and 1979 about the pragmatics of discourse. On the one hand they were intended to extend the current theory of speech acts, emerging in linguistics around 1970 after its initial development in the philosophy of language, and on the other hand they represent an attempt to incorporate a pragmatic component in text grammar. There appeared to be many obvious links between textual structures and the structures of speech acts, speech act sequences and, in general, context structures. Some of these links, e.g. coherence properties of discourse, presupposition, topic and comment, connectives, macro-structures, etc. have been discussed in the papers in this book, showing that a theory of discourse cannot be adequate without a pragmatic component. Yet, the discussions have a tentative nature. No unified theory has been attempted, and many of the relationships between pragmatics and (text) grammar still remain obscure. One of the reasons to collect these papers was that several of them were published in relatively inaccessible books or journals, whereas other papers have not been published before. In order to mark the clear development of the papers, they have not been rewritten (except for minor changes, e.g. in the notes) which at some points resulted in inevitable overlap, but which nevertheless, leaves open the possibility of reading the papers individually and in any desired order. In order to allow the reader to skip the redundant parts of the text, these have been printed in the same size of type as the notes. I have added, however, both an introductory and a concluding chapter (both written in 1978), in which the links between the various chapters, the major

vi

Preface

problems, the development and backgrounds of the views presented, as well as open problems and further prospects are discussed, respectively. Especially, it was necessary to sketch briefly the place of text grammar with respect to the generative-transformational paradigm, and the role of pragmatics in linguistic theory in general and text theory and grammar in particular. The papers are presented in the order in which they were written. One of the tendencies has been to give increasing attention to the interdisciplinary aspects of (discourse) pragmatics, especially the relationships with the cognitive processing of semantic and pragmatic structures. At the same time, whereas in the first papers a pragmatic component was still discussed in a more or less general way, with only occasional specific relations to discourse structure, the later papers try to show the inherent pragmatic properties of discourse, and the links and analogies with local and global structures of speech act sequences. Although some of the papers provide introductory remarks about concepts from the theories of discourse and action which constitute the basis and background of the discussion, the reader is referred to my book Text and Context (London: Longmans, 1977) for general introductions to formal semantics, the theory of action, semantic structures of discourse (connection, connectives, coherence and macro-structures) and their relations to corresponding pragmatic structures. The studies in the present book are at the same time preliminaries and further elaborations of the pragmatic part of that monograph. I would like to thank the editors of the books and journals in which these papers originally appeared for their permission to reprint the papers in this book. Both acknowledgments and reference to the original places where the papers were published are made in the Acknowledgments. March 1980

TAvD University of Amsterdam Dept. of General Literary Studies Discourse Studies Section

Acknowledgments

Acknowledgment is made to the following editors and publishers of the respective papers: To Siegfried J. Schmidt and Fink Verlag for permission to reprint "Pragmatics, Presuppositions and Context Grammars", published in Pragmatik III Pragmatics, Munich, 1976, pp. 53-82. To Marvin Loflin and James Silverberg and to Mouton/De Gruyter, for permission to reprint "A Note on the Partial Equivalence of Text Grammars and Context Grammars", published in Discourse and Inference in Cognitive Anthropology, The Hague, 1978, pp. 135-144. To Sidney Greenbaum and to Mouton/De Gruyter for permission to reprint "Acceptability in Context", published in Acceptability in Language, The Hague, 1977, pp. 39-62. To Ladislav Matejka and the Journal of Slavic Philology (vol. 1) for permission to reprint "Sentence Topic and Discourse Topic", University of Michigan, Ann Arbor, 1977, pp. 49-61. To Marc de Mey and co-editors, and to Communication and Cognition for permission to reprint "Pragmatic Macro-structures in Discourse and Cognition", published in CC 1977, Papers of the International Workshop on "The Cognitive Point of View" (March, 1977), University of Ghent, Ghent, 1977, pp. 99-113. To Helmut Haberland, Jacob Mey and North Holland Publishers for permission to reprint "Context and Cognition. Macro-structures and Knowledge Frames in Speech Act Comprehension", published in the Journal of Pragmatics 1 (1977): 211-232, and for "Pragmatic Connectives", Journal of Pragmatics, 3(1979): 447-456. To Eduarde Forostieri, Gerald Guinness and Humberto Lopez Morales

viii

Acknowledgments Universitaria (UPR), for permission to reprint "The Pragmatics of Literary Communication". In On Text and Context. Papers of the Int. Conference on Literary Communication (Rio Piedras, April, 1977), Rio Piedras, 1980, pp. 3-16. To Leo Apostel and Asa Kasher for permission to reprint "Towards an Empirical Pragmatics" from Philosophia, special issue on speech acts (1981). To Sandro Ferrara and Julien Boyd for permission to reprint "The Semantics and Pragmatics of Functional Coherence in Discourse" from the special issue of Versus (1981) about 10 Years of Pragmatics.

Contents

Preface Acknowledgments Chapter one

Chapter two

v vii

Towards a pragmatics of discourse

1

1. Introduction 2. Generative-transformational grammar and text grammar 3. Text grammar and pragmatics 4. Some issues in the pragmatics of discourse

1

17

Pragmatics, presuppositions and context grammars

27

1. 2. 3. 4. 5. 6. Chapter three Chapter four

The status of pragmatics Context grammars Pragmatic structures Texts and contexts Semantics and pragmatics Presuppositions in text grammars and context grammars

3 5

27 29 33 41 45 51

A note on the partial equivalence of text and context grammars

59

Acceptability in context

71

1. The problem

71

Contents 2. 3. 4. 5. 6. Chapter five

Chapter six

Chapter seven

Chapter eight

The philosophy of 'acceptability' Relative grammaticalness Pragmatic acceptability Macro-structures Discourse and cognition

72 77 82 84 89

Issues in the pragmatics of discourse

97

1. Sentences and sequences 2. Further pragmatic constraints on sequences 3. Acts and speech acts 4. Focus, perspective and related notions 5. Final remarks

97 116 120 144 160

Pragmatic Connectives

163

1. 2. 3. 4.

163 165 166 174

Speech act sequences Semantic connectives Pragmatic connectives Conclusions

Sentence topic and discourse topic

177

1. 2. 3. 4.

177 179 186 192

The problem Sentence topic Discourse topic Conclusions

Pragmatic macro-structures in discourse and cognition 1. Introduction 2. Semantic macro-structures 3. Sentence sequences and speech act sequences 4. Macro-acts 5. Pragmatic macro-structures 6. Pragmatics and cognition

195 195 196 198 201 205 209

Contents Chapter nine

Context and cognition Knowledge frames and speech act comprehension 1. The foundations of pragmatics 2. The cognitive nature of pragmatic conditions 3. Pragmatic comprehension 4. Frames and speech acts 5. Context analysis 6. Examples of context analysis 7. The 'dynamics' of context 8. Utterance analysis 9. Comprehension of speech act sequences and macro-speech acts 10. Psychological process assumptions

Chapter ten

Chapter eleven

215 215

216 217 219 222 228 232 233 236 239

The pragmatics of literary communication

243

1. 2. 3. 4.

243 245 247 262

Pragmatics Literary communication The pragmatics of literature Concluding remark

The semantics and pragmatics of functional coherence 265 1. 2. 3. 4.

Chapter twelve

xi

The problem Pragmatic theory Semantic relations in discourse Functional coherence at the semantic level 5. Speech act sequences and functional pragmatic coherence 6. Conclusions

265 266 268

275 282

Towards an empirical pragmatics Some social psychological conditiions of speech acts

285

269

xii

Contents 1. 2. 3. 4. 5.

Chapter thirteen

Introductory summary The need for an empirical pragmatics The cognitive model of speech acts Social psychological factors Concluding remark

Conclusion, open problems, further prospects and future research

285 286 287 288 295 297

1. General conclusions 297 2. Open problems 299 3. Further prospects and future research 314 Bibliography Subject Index

319 327

CHAPTER ONE

Towards a Pragmatics of Discourse

1.

INTRODUCTION

1.1. The aim of this introductory chapter is to provide a general framework for and a discussion of the papers collected in this book. These papers have been written over a period of seven years, from 1973 to 1979, and therefore some critical reassessment of their aims, methods, theoretical proposals and terminology is necessary. Although there is a clear pattern in their development, no unified theory of pragmatics can be achieved at the moment. The papers are pieces of a puzzle representing the complex picture of the relations between language and action in communicative contexts. In the rapidly growing field of pragmatics they are specifically intended to study the pragmatic properties of discourse. In this respect they should establish the links between the theory and grammar of discourse on the one hand and the theory of speech acts on the other hand, thus extending both fields of research. 1.2. Our major claims are that a sound theory of discourse cannot be adequate without a pragmatic component and that, conversely, a serious pragmatics should account for the functions of utterances with underlying textual structures. This means, firstly, that for instance properties of local and global semantic coherence of sentence sequences should be complemented with coherence properties at the pragmatic level of speech act sequences. And secondly, it should be stressed that a pragmatic theory cannot be limited to an account of single speech acts, expressed by single sentences, but also must explicate the structure of speech act

2

Towards a Pragmatics of Discourse

sequences and global speech acts, realized by sequences of sentences of discourse and conversation. To assess these two claims is the crucial task of a pragmatics of discourse. 1.3. The papers in this book provide some evidence for these claims. As was said above, however, we are far from a unified theory linking a grammar of discourse with a theory of speech acts and verbal interaction. First of all, the papers study some properties of sentences and discourse, such as presuppositions, connectives and topic-comment, for which clearly also a pragmatic account is imperative. Secondly, the more general relations between linguistics, grammar and pragmatics are discussed. Thirdly, many of the papers try to (re-) formulate what the specific aims, problems, units and basic concepts of a linguistic pragmatics should be. It has become increasingly obvious that the foundation of such a linguistic pragmatics should be sought in a more general theory of action. Later it was realized that these foundations should also be extended to a more social theory of interaction, as well as to a cognitive theory of speech act processing and communication. The more recent papers tend towards this interdisciplinary orientation. Although we still believe that an independent linguistic pragmatics is possible and necessary, real understanding of its basic concepts does not seem accessible without philosophical and empirical, that is psychological and sociological, foundations. Much of current work on pragmatics in general, and the theory of speech acts in particular, taking place in a too narrow grammatical-linguistic framework, is therefore not only limited in scope, but lacks a fundamental explanatory character. Properties of action, cognition and social structure are often used in a more or less intuitive way, thus ignoring the analysis and empirical assessment of these primitive terms of linguisitic pragmatics that are given in other disciplines. 1.4. In the final chapter we will try to give a survey of the many problems which were not, or only briefly or inadequately, treated in the papers. Such a survey will at the same time imply programmatic prospects for future work on discourse and pragmatics.

Generative-transformational and Text Grammar 2.

3

GENERATIVE-TRANSFORMATIONAL GRAMMAR AND TEXT GRAMMAR

2.1. One of the characteristic properties of generative-transformational theory, which dominated linguistics in the sixties, is the central and nearly exclusive role of 'grammar' within a theory of language. The result of this emphasis on grammar was that properties of language and language use which were less fit to be accounted for within the more or less explicit but empirically narrow generative grammars received little systematic attention. In fact, this was even true for semantics, emerging in the second part of the decade, especially under the impetus of 'generative semantics'. That is, gradually the account of sentence meanings came to be identified with an explication of the 'logical form' of sentences. This tendency has led to the introduction of various kinds of logical systems into linguistic theory, first predicate logic, then modal logics and intensional logic (in the framework of Montague grammars). These extensions already went beyond the boundaries of classical generative grammars. The second deviation from the main paradigm, also occurring around 1970, was the rapid growth of sociolinguistics and pragmatics. Sociolinguistics, confronted with data and problems of actual language use, had to deal with all kinds of language variation, which conflicted with the Chomskyan assumptions about an 'ideal speech community'. Similarly, pragmatics appeared on the linguistic scene after Searle's (1969) book on speech acts, introducing notions such as 'speech act', 'felicity conditions' and 'context', which were difficult to integrate into syntactic or semantic generative grammars. Of course, the pervasiveness of the grammatical paradigm first led to attempts to account for social variation and speech act constraints within the grammar, e.g. with variable rules (Labov, 1972a, b) and hypersentences (Sadock, 1970, 1974), respectively. These attempts to save the role of grammar, however, were not successful, and it has become clear by now that in both cases independent, much more complex theories are needed to account for the phenomena of language use. 2.2. Actually, a similar development has taken place in the study of discourse. Also around 1970 it became clear that syntax and

4

Towards a Pragmatics of Discourse

especially semantics could not be limited to an account of sentences in isolation. In many cases, grammaticalness appeared to be relative. Thus, the interpretation of sentences and the treatment of PROforms, (in-) definite articles, presuppositions, topic and comment, etc. required interpretation of other, e.g. previous, sentences of the discourse. These and other arguments resulted in attempts to elaborate so-called 'text grammars'. But also in this case, many linguists preferred to try to account for such'phenomena within the classical sentence grammar format, thereby taking discourse as one 'complex sentence'. Text grammar itself, however, also kept very close to the basic principles of the paradigm: it was argued that a grammar should generate an infinite set of grammatical texts of a language, and that such a grammar should be considered as a more or less abstract account of the ability of native speakers to produce and understand any well-formed and interpretable text of their language. In that perspective, findings about the semantic relations between clauses in sentences were generalized to an analysis of similar relations between sentences of a text. This was primarily demonstrated on all kinds of coreferential phenomena, which linked original text grammars with developments in generative semantics. It was only at a later stage that it was realized that the fundamental problems of (co-) reference could not be handled in a grammatical meaning-semantics at all, and that a referencesemantics, as exemplified by logical semantics, should be added to the text grammatical framework. This was a first step away from the generative-transformational paradigm, as was mentioned above for the introduction of logic in sentence semantics. Yet, even this extension towards a logical grammar did not allow for a systematic study of discourse phenomena which seemed outside the scope of either intensional or extensional semantics, e.g. functional relations between sentences, the still unsolved problem of topic-comment articulation, the distribution of information in sequences in general, and certain conversation particles and connectives. Another fundamental problem for a sound theory of discourse was the treatment of global semantic structures, such as 'theme' or 'topic of discourse', which were explicated in terms of semantic

Text Grammar and Pragmatics

5

macro-structures, although at first we had only a very vague idea what these should look like and especially how they should be connected with semantic structures at the sentence and sequence level. In the framework of empirical psychological research about the role of such macro-structures in cognitive processing of discourse we then gradually discovered some of the rules relating local sentence and sequence meanings to global discourse meaning (van Dijk, 1975a, 1977a, 1980). But, this kind of'semantic transformation' linking micro-structures with macro-structures, which should also account for the necessary operations of complex semantic information organization and reduction during discourse processing, were clearly no longer compatible with the idea of a generative grammar taken in a narrow sense. Moreover, textual coherence, both at the micro- and at the macro-level, could not possibly be explained without the important role of 'world knowledge' applied by language users to establish connections between sentences and to construct macro-structures. This fact required further extension of the theory with a cognitive semantics, specifying among other things how 'frames'or 'scripts', i.e. organized conceptual structures of conventional world knowledge, play a role in the production and understanding of meaningful discourse.1 Thus, not only the boundaries of linguistic grammar but those of linguistics itself seemed to be transgressed, thereby making the theory of language and discourse an essentially interdisciplinary enterprise.

3.

TEXT GRAMMAR AND PRAGMATICS

3.1. It soon became obvious that these various developments in linguistics needed integration. Not only was it necessary to analyse textual coherence in terms of formal semantics and cognitive semantics, but the theory of speech acts also needed to be linked with text grammar. Although a first attempt in this direction was already made in Some Aspects of Text Grammars (van .Dijk, 1972a), there was no really specific connection between the two theories. The basic idea was only that a speech act is accomplished by an utterance in some context, and that such an utterance does

6

Towards a Pragmatics of Discourse

not necessarily consist of one single sentence. In other words, a pragmatically based text grammar should specify the conditions under which whole discourses, when uttered in some context, could be said to be appropriate with respect to that context. It was left open whether the discourse-utterance should be analysed as one speech act or as a sequence of speech acts, and not specified how the appropriateness of its component sentence-based speech acts is related to the appropriateness of the discourse-based speech act as a whole. 3.2. Typically, it was further assumed that if a grammar should handle speech acts there should be 'generative' rules of some kind, specifying the infinite set of appropriate speech acts of the language, thereby accounting for the 'communicative competence' of language users. In other words, the generative paradigm required not only recursive definition of all possible texts, but also of all possible contexts (and their relations). This is the ratio behind the notion of context grammar used in the first papers of this book. Clearly, the notion of grammar in this sense is rather far from the usual grammar of sentence surface structures, but there is no a priori reason to use the notion of grammar in this restricted sense. If it is acceptable to extend the syntactic grammar with a semantic component, abstractly accounting for the meaning and reference, i.e. the interpretation of generated sentences, we could use the same justification for the extension of the grammar with a proper pragmatic component. In that sense the grammar would have as an additional task to account abstractly for the functions an expression (with a certain meaning and reference) may have, viz. as certain speech acts. And, whereas in semantics we interpret sentences, expressing some proposition, relative to some possible world (or model structure), we would evaluate sentences, expressing some proposition, and realizing some speech act, relative to possible contexts. In other words, the grammar thus becomes a F O R M - M E A N I N G - A C T I O N rule system. Note that we do not combine fully heterogeneous theoretical components in this way. A grammar thus conceived is in fact an abstract theory of verbal utterances of a language. Of these utterances it specifies the abstract 'underlying' morphonological and syntactic structures,

Text Grammar and Pragmatics

1

the conventional meanings/interpretations associated with such utterances, and finally the conventional speech act associated with such utterances. Thus, not only the 'form' and the 'meaning' of the utterance should be specified, but also its 'functions', plus the way these functions are systematically related to both form and meaning.2 At this point, it seems unproductive to quarrel about terms. If there are practical reasons to limit the use of the term 'grammar' to grammars in the traditional sense, then there is no problem in talking about grammar being systematically related to a pragmatic theory. But in that case, we no longer are allowed to call a grammar an abstract, idealized theory of linguistic competence, because knowledge of language users about the pragmatic functions of their utterances is a general property of linguistic competence, if this term has any empirical content at all. One might argue that as soon as we talk about the functions and uses of language, we are definitely within the domain of what has been called 'theories of performance' in the paradigm. This argument, however, is not valid. First of all, even semantics, which role in the grammar is hardly disputed anymore, is an abstraction of the meanings and reference assigned to utterances by language users. In fact, we may say the same for all abstract grammatical structures: they are not 'inherent' properties of the phonetic sequence, but a cognitively assigned structure operating with categories and rules. Similarly for the pragmatic component: of course again we must make the necessary generalizations and abstractions with respect to the ad hoc properties of speech acts and contexts, but further the rules, categories, conditions, etc. assigning speech acts to uttered sentences/discourses, and determining under what conditions these are appropriate with respect to a context, are completely in Jine with the kind of things usually accomplished by a grammar. A further argument is that the different levels are systematically related: syntactic structures not only to semantic structures or interpretations, but also to pragmatic structures (e.g. indicative sentence form to assertive speech act types, etc.), and semantic structures to pragmatic structures (e.g. the prepositional content of speech acts, indexical expressions, performatives, particles, connectives, modalities, etc.). Whether we call it a

8

Towards a Pragmatics of Discourse

'grammar' or not, what is needed is an integrated theory of the abstract structures and functions of verbal utterances, and a pragmatics can no longer be ignored in such a theory. This integrative strategy in linguistic theory formation does not give carte blanche for the introduction of any other theory into the grammar. We still would prefer to keep grammar apart from cognitive and social theories, simply because a grammar has no strict and direct empirical basis. The cognitive and social theories will have to specify how language users actually communicate: how they listen to each other, understand what is said, transform and store information in an organized way into memory, loose access to this information (forgetting) under specific conditions, are able to (re-) produce some information, and how all this is in turn determined by the language user's knowledge and assumptions about the other language user, the social context, with its roles, rules, functions and institutions. And especially, how all this again varies over language users, over groups of language users, etc., e.g. as a function of individual knowledge, beliefs, interests, attitudes and wishes, personality factors, on the one hand, and social class or group, education, sex, power, etc. on the other hand. From this enumeration of some of the tasks and phenomena involved in the psychological and sociological accounts of speaking and communication we see that pragmatics is still highly abstract and general. It contributes to the abstract definition of an utterance, by specifying the function, viz. to be a certain act, of that utterance. And language users have general knowledge of these functions and of the contextual conditions determining appropriateness, and act according to this (implicit) knowledge. The rules involved are also general in the sense that they may hold for any utterance of the language used under normal conditions (i.e. sincerely, in social context, etc.). In this respect, for instance, the 'grammar' of narrative or argumentation does not belong to the grammar of natural language: not every utterance has a narrative structure. The same holds for the theory of rhetorical operations. In both cases, the structures generated by these theories are systematically mapped on the linguistic structures of utterances, but generally, and not only on linguistic

Text Grammar and Pragmatics

9

structures (narrative structures may be 'expressed' also by picture sequences) (cf. van Dijk, 1978a). From this discussion we conclude that pragmatics may be taken as a component part of a grammar taken in a wider, but still methodologically sound, sense, but that at the same time other properties of utterances and the use or processing of utterances should be accounted for in other theories. These would specify the possible and actual varieties of discourse types, language users, cognitions, and social situations, in relation to the relevant structures of the utterance as specified by the grammar. 3.3. Once a case is made for taking up pragmatics into the grammar, problems arise of which few have been solved at the moment. First of all, of course, the necessary specific categories, rules, constraints, etc. should be established for our new level of grammatical description. That is, is what is systematically involved in the conditions making utterances, taken as speech acts, appropriate? In other words: what kinds of properties of such contexts must be enumerated in the grammar? Clearly, knowledge of speakers/hearers is involved and hair colour is not, but for other contextual properties this is less obvious. Take for instance attitude, politeness, power or social and institutional relations in general. For the discrimination of some speech acts these properties of language-users seem relevant, but not for others. Thus, politeness, power or social status appear to be involved in distinguishing commands from requests, whereas to arrest or to sentence are speech acts restricted to language users in certain institutional functions. Important parts of the papers here collected repeatedly try to make the necessary inventary of the pragmatic component, a problem more or less neglected in speech act theory and linguistic pragmatics. That is, there was no precise formulation of the specific aims, categories or types of rule of a pragmatic theory or grammar component. And referring merely to 'speakers' or 'hearers' is obviously inadequate because speakers and hearers have so many properties, many of which have nothing to do with appropriate communication. Hence an attempt was made to take speakers and hearers as abstract language using systems, much in

10

Towards a Pragmatics of Discourse

the way current artificial intelligence would do in constructing simulating systems for language production and comprehension. Language users defined in this way would be characterized with a set of properties necessary for the grammatical, i.e. well-formed, meaningful and appropriate, production and understanding of utterances. Recall that they would certainly not be models of real language users: they would not have strategies for the efficient planning, production or understanding of utterances, they would not have limited working memories, affective problems or social misery. Therefore, it should be stressed, the grammar is not an empirically adequate theory of language and communication, but at most a component of such a theory. Thus, specifying a more ordess theoretical language user as the major feature of pragmatic context, required systematizing its relevant properties, viz. those involved in making utterances appropriate. We arrived at'the provisional assumption that both cognitive and social properties should be specified. Of the cognitive ones we seemed to have only knowledge, belief and want, viz. epistemic, doxastic and boulomaeic structures. The social features seemed less easy to isolate. Clearly, the social relation of 'dominance' (actualized in power, authority, etc.) was needed for many speech act conditions, but what about 'status', 'role', 'function', or similar categorical concepts of social structure? These and similar questions have not been answered in these papers, and-as will be spelled out in the last chapter in more detail—we here have a clear problem of future research. Moreover, it appeared that probably also the 'psychological' features should be further extended, perhaps not only at the cognitive level, but also at the affective level of feelings, attitudes, judgements or emotions, as we try to do, highly informally and provisionally, in Chapter 12. Thus, an evaluative component is present in many speech acts, such as accusations, reproaches, congratulations, thanking, etc. That is, we need conditions such as 1S finds it good/positive . .. '. The general framework for all these speculations about the necessary properties of language users and context in these papers has been the theory of action, as it has been developing in philosophy and philosophical logic. In this respect we have tried

Text Grammar and Pragmatics

11

to be more systematic than the philosophers of language, who of course have established this link. Obviously, notions such as appropriateness, happiness or felicity of speech acts were specific forms of the more general notion of success fulness of action. Thus, basic notions such as purpose, intention, doing, result, effect, etc. were needed as general properties of purposeful speech interaction, whereas knowledge, beliefs, wishes, preferences and social relations would function as defining specific kinds of action. In Chapter 4, we have tried to analyse the various properties of the basic notion of pragmatics, viz. appropriateness, parallel to well-formedness in syntax and truth or satisfaction in semantics. As was argued above, appropriateness in this sense is a theoretical notion of a theory of grammar, and distinct from the empirical notion of 'acceptability', related to actual understanding and functioning. Note, incidentally, that the theory of action itself is not part of pragmatics, but serves as a. foundation for a pragmatic theory, much in the same way as, for example, the theory of abstract automata served as a mathematical foundation for generative syntax (See for example Rehbein, 1977, for a detailed analysis of complex action used in this way). 3.4. Once the proper aim, categories and types of rule of pragmatics were established, the next major problem was how to link these components to the other components of the grammar. That is, it is a characteristic property of grammars that not only several levels are distinguished and analysed, but also that these levels are related by certain rules or constraints. Given certain expressions as specified by the syntax, the semantics will provide the interpretation of these expressions. Similarly, we may now argue, a pragmatics will also 'interpret' these semantically interpreted expressions, viz. as actions which are (in-) appropriate relative to a context. Although this seems a plausible approach, we know little about these kinds of links between pragmatics and the 'rest' of the grammar.3 We know that some syntactic and semantic aspects of sentences and discourse are somehow determined by pragmatic constraints, but we have no general meta-theory about the relationships between syntax and semantics on the one hand, and pragmatics on the other hand. Thus, whereas we speak about

12

Towards a Pragmatics of Discourse

'sentences' being well-formed, and perhaps even about sentences being true or false, we would hardly say that sentences are appropriate or not. We would at least need a function making sentence-utterances out of sentences, since appropriateness applies only to the use of language, i.e. to things which are done. This more general problem should be attacked in future research about which we will be more specific in the last chapter. 3.5. The same holds, mutatis mutandis, for the links, already discussed above, between pragmatics and the respective theories of cognition and social interaction (see also van Dijk, 1978b, 1980). We must know which concepts involved in successful communication are specifically relevant in pragmatics, and what the empirical basis of such concepts are. Thus, an empirical pragmatics would need a theory about how speech acts are actually understood as such, i.e. which contextual features are used by language users to assign some speech act to a given utterance. In other words: what cues of the context are necessary and sufficient such that some utterance may count as such a speech act for a given hearer? In Chapters 8 and 9 we have tried to spell out some of the cognitive conditions involved in the interpretation of utterances as speech acts. In part, these are identical to the cognitive properties of discourse comprehension and complex information processing in general. Similarly, chapter 12 specifies which further social psychological properties underly the acceptability of speech acts, thus establishing a link between cognition and social interaction. 3.6. The next, and major problem with which the present papers are concerned is the more specific link between pragmatics and text grammar. This means that, firstly, it should be specified which properties of discourse require additional account at the pragmatic level, and secondly, it should be investigated which properties of speech acts manifest themselves in discourse structures. Obviously, what was needed in such an approach was the introduction of the notion of a speech act sequence, to be related systematically to the sentence sequence of the discourse. Although this step, also within speech act theory or pragmatics, towards an analysis of speech act sequences is obvious, very little attention

Tex t Grammar and Pragmatics

13

has been given to the necessary conditions on appropriate speech act sequencing. In this respect, pragmatics stayed close to current sentence grammars, especially because it was tacitly assumed that one speech act is usually accomplished by one sentence, at least according to the majority of examples treated in the literature. Yet, even the analysis of the links between speech act and sentence structure has been superficial. It was not shown, for instance, whether one sentence could manifest also several speech acts, and if so, how this could seriously be explained; or whether some speech act is based on a complete sentence or only on some part of it, e.g. the comment, being the part expressing 'what is new' or 'focused upon'. Against this background, then, it became necessary to pay closer attention to the precise relationships between speech acts and simple, compound and complex sentences and sentence sequences, as we have tentatively initiated in Chapter 5. At the same time, the thinking which had led to text grammar was extended to the theory of pragmatics itself: usually language users do not accomplish single, isolated speech acts; they produce sequences of speech acts, often in dialogue or conversation; and these sequences, either of one speaker or of several speakers, most certainly will have to obey certain rules and constraints. The very elementary reason for this assumption lies in the theory of action. An action, i.e. an event brought about through intentional doings of persons, leads from some initial state to some intended final state (if it is successful). In speech act sequencing, thus, the conditions holding for the appropriate accomplishment of some following speech act must be at least consistent with the properties of the final state brought about by the preceding speech act. Or perhaps even more powerfully: some preceding act could be accomplished solely with the intention of bringing about the conditions necessary for a following speech act. We all know this intuitively from everyday talk: we make remarks about the weather, somebody's health or appearance, begin to make statements about our financial problems, and will only then make a request for getting a loan from an acquaintance. Clearly, in such cases we not only have general rules and constraints of pragmatic coherence as we may call it, but also cognitively based interactional strategies. These

14

Towards a Pragmatics of Discourse

strategies are necessary in maximizing the effectiveness of our way of talking, i.e. to ensure that our purposes are realized (get our loan), whereas the proper pragmatic rules determine which sequences are possible I appropriate. Although we have begun with some observations about the possible constraints on speech act sequencing, we certainly have not yet found the rules involved. And it is not yet clear even whether several kinds of rules and constraints should be distinguished. On the one hand, we have the general constraints, mentioned above, from a theory of action. They pertain to general properties of conditioning, enabling temporal and causal sequencing. Then we have the kind of rule determining speech act sequences in conversation (or other speech act 'exchanges'), such as answers required after questions, or thanks after offers or congratulations. These may be more or less conventional or even fixed. Thirdly, we have the rules which have more of ^.functional nature. That is, they assign a function to an utterance-speech act on the basis of its relation to surrounding speech acts. Thus, some speech act, mostly assertion, but also command, or request, may be functioning as an 'answer' to another speech act. Similarly, many speech acts may function as an 'explanation' of a previous speech act. And conversely, as a special type of conditioning, speech acts may be seen as 'preparation' for other speech acts, very often so in requests, for instance. We are unable to draw the full list of such functions of speech acts in sequences. The same holds, by the way, for the analysis of functional relations between sentences or propositions of texts. And in fact, the two levels of analysis are closely related. If we say, for instance, that one sentence functions as an 'explication' of another sentence, we may reconstruct this in semantic terms (specifying new information about some object, for instance), or in pragmatic terms (to add, to be more specific). This problem is further explored in Chapter 11. This and other reasons seemed to suggest that the analysis of speech act sequences could be profitably set up in analogy to the lines of research followed in the semantic account of textual structures. Thus, we were able to say of two speech acts that they are connected if one is a condition of the other—in the same way as propositions are connected if the denoted facts are for example

Tex t Grammar and Pragmatics

15

conditionally related (see van Dijk, 1977a, for detail). Similarly, other coherence conditions must be satisfied, such as speaker identity, or relations between speakers, and compatibility of the acts themselves. This consequent strategy of theory formation by analogy led to the very fruitful and now obvious assumption about the existence of pragmatic macro-structures, i.e. of global speech acts. Thus, we would have a sequence of individual speech acts at the 'micro-level', and in a coherent talk or conversation also one or more global speech acts organizing the individual speech acts. Again, we all know this from our everyday communicative behaviour. For instance, although a sequence of various speech acts may be performed, the whole may still function as a 'request' by A to B to do p, or as an 'invitation'. In this way, a lecture may function globally as a macro-assertion, and a letter to some government bureau as a macro-request. The two pieces of the puzzle seemed to fit so well that at the same time this hypothesis yielded a strong further argument in favour of semantic macro-structures. That is, each speech act needs a prepositional base, so also macro-speech acts would need such a global content. Clearly, this cannot be the propositions of subsequent speech acts, but must be the macro-proposition derived from them. To get macro-speech acts from speech acts, explicit mapping rules are needed. Again, these appeared to be very similar to those at the semantic level. In this case, the macro-rules do not operate on any kind of information, but more specifically on action information. So again, irrelevant speech acts (those not conditioning others) may be deleted (often greetings, for instance), sequences may simply be generalized (demand, requirement, request, etc. to 'ask' for instance), and finally a sequence of various speech acts may jointly entail another speech act, especially in institutional contexts, where there are fixed social frames for the successive accomplishment of speech acts, e.g. in court. Although we pay some attention to these rules, and their cognitive basis, in Chapter 8, they need further elaboration, and especially the formulation of further constraints. Some of this work has been done in our recent book on macro-structures (van Dijk,

16

Towards a Pragmatics of Discourse

1980). At the same time more empirical work is necessary on the pragmatic (macro-) structure of talk, so that we get to know more about the empirical constraints on speech act sequencing, functional relations, frame-like conventional structuring (as is known from openings and closings of conversation, for instance), and of course the social and cognitive bases for these kinds of structure. Note, finally, that pragmatic macro-structures not only account for the existence of global speech acts, but at the same time are a further condition on local pragmatic coherence. That is, if subsequent speech acts are in a conditional relation, this must be within the global framework ofa. pragmatic topic of conversation, so to speak, that is, with respect to the (global) act the speaker is 'now' accomplishing. In this sense, the sequence may really be goal-directed and thus organized by the global intention (plan) and purpose of the global speech act embracing the conversation as a whole. We see that the basic idea of 'macro-structures' taken as an information reducing and organizing principle plays a role anywhere in the theory of language: at the semantic level for handling complex structures of sequence meanings, at the pragmatic level in order to organize action structure, and at the cognitive level, linking both, to be able to handle complex information in general. The notion of macro-syntax, at least in the sense of macrostructures understood here, would fall outside linguistic grammar proper, and pertains to other kinds of categories, e.g. those of the theories of narrative and argumentation. The same holds for morphonological structures at the macro-level, to be accounted for in a theory of prosody and metrics. 3.7. It may be concluded from the few remarks made above that the marriage of text grammar and pragmatics was not only necessary but also theoretically fruitful. It has in fact led to the discovery of new rules, constraints, categories and structures in pragmatics, and on the other hand provided the possible solution of hitherto thorny problems in the grammar and theory of texts, e.g. coherence conditions, to which we will return briefly below. Finally, since now both the discourse and the action aspects of utterances can be accounted for, and mutually related, we have a

Issues in the Pragmatics of Discourse

17

much better theoretical starting point for the empirical investigation of language use in cognition and interaction.

4.

SOME ISSUES IN THE PRAGMATICS OF DISCOURSE

4.1. There are at least two major reasons for including grammar in an account of semantic discourse structures and pragmatic speech act structures. First, it should enhance the descriptive adequacy of the grammar in the analysis of natural language utterances viz. by the introduction of further generalizations, rules,categories, units and analytic levels, such that new phenomena can be studied and the grammar can become a more satisfactory component in an interdisciplinary theory of language. Secondly, however, such a development should at the same time provide a more adequate framework for the treatment of current issues, i.e., phenomena and problems which had been discussed in the actual paradigm, but clearly with only partial success. It has appeared often that problems of syntactic and semantic description required additional explication in terms of relations between sentences in discourse or in terms of properties of pragmatic context, e.g. speaker, hearer, and their knowledge or beliefs. In other words, both text grammar and pragmatic grammar can be called 'linguistic grammars' if they really contribute to a formulation and solution of specific linguistic problems. Of course, problems may turn out to be pseudo-problems, and the new kinds of grammar may have their proper theoretical and empirical aims giving rise to perhaps other kinds of problems, but the practical significance of new proposals very often is recognized only inasfar as well-known issues can be treated in a 'better', i.e. more explicit, more elegant or more consistent way. In this section, we will breifly enumerate some of these issues which have been given attention in the papers of this book. 4.2. A first issue, perhaps the most crucial one in the discussion of natural language semantics of the last decade, concerns the analysis of presuppositions. In the footsteps of the philosophy of language, linguistic grammar soon began the attempt to explain

18

Towards a Pragmatics of Discourse

the more or less intuitive insight that sentences may have presuppositions. In this sense, a presupposition was usually treated as a proposition which should be true in order for the presupposing sentence to be true or false, or even have meaning. Such presupposional phenomena were first observed for definite descriptions and restrictive relative clauses. Then, all kinds of other syntactic structures and lexical items (nouns, verbs, adverbs, particles, etc.) appeared to be involved as well, such as cleftsentence constructions, factive verbs such as to know and adverbs such as even. Since clearly propositions and truth values were involved, logical analysis was soon associated with these linguistic facts, the standard definition of presupposition in that framework being that p is a presupposition of S if both S and ~S entails p. In other words, a presupposition should not as such influence the truth value of a sentence, but should determine whether such a sentence can be assigned a truth value at all.4 Of course, there still remained a lot of problems, both formal and descriptive. For one thing, entailment could not be properly defined in classical logics, and hence the basic notion of the definition remained an insecure factor. Since a contradiction is entailed (i.e. necessarily implied) by any proposition, the definition would allow p & -p to be presupposed by any sentence; an undesirable by-product for natural language. Hence, the underlying conditional of entailment should guarantee that the 'contents' of the related propositions should be 'relevant' to each other in some way, thus defining 'relevant conditions', 'relevant strict implication' and 'relevant entailment':5 a formal task which has not yet been accomplished, and hardly applied in the theory of presupposition. Similarly, the descriptive analysis also met with problems, because the presuppositions of certain structures or words seemed to change with syntactic and semantic context. Recursive calculation of 'embedded' presuppositions, thus, became difficult.6 In early text grammar, and less Officially' elsewhere, it was observed that presuppositions of sentences very often appear as a part or an implication of preceding sentences of the same discourse. Thus, presupposition appeared to be a rather straightforward manifestation of semantic textual coherence (van Dijk, 1972a). For complex sentences, for instance, this would mean that the

Issues in the Pragmatics of Discourse

19

embedded clause could be derived from an underlying structure in which the presupposed proposition would precede the other proposition. Whatever the value of that kind of 'transformational' reduction to prepositional sequences, underlying both complex sentences and sequences of sentences, the basic idea was that both meaningfulness (intension) and truth value or satisfaction (extension) of a sentence are relative notions in natural language: a sentence may be true or false relative to the truth or satisfaction of preceding (or sometimes following) sentences in a discourse. In model-theoretic terms this would mean that a sentence is evaluated with respect to a model which is partly 'constructed' through the evaluation of preceding sentences in connected models. In other words, if some definite noun phrase presupposes the existence of an individual, this means that the individual has already been introduced contextually in the domain (universe of discourse) of the text-model. The same holds for Other presupposed propositions, denoting facts in possible worlds which have given access to the world(s) in which the present sentence should be evaluated. Of course, although discourse provided a natural framework for this contextual account of presuppositions, this picture is not complete. First of all, presuppositions also appear via a more indirect, even non-linguistic, road. Thus, previous sentences may not express the presupposed proposition, nor even strictly imply it. In that case, other information, typically frame-like world knowledge, is needed in order to supply the missing proposition. At first intuitively, later against the background of frame-theory in artificial intelligence, this introduction of connecting propositions appeared to be necessary anyway in the analysis of discourse coherence. Thus, the notion of implication became weaker in this sense, often merely involving plausible or probable implication. But even this extension of the theory of discourse was not sufficient. The contextual approach should clearly be generalized and extended to pragmatic or even other contexts, e.g. in order to account for sentences with presuppositions only determined by the context of speaking, viz. knowledge or expectations of speakers or hearers. Thus, definite articles simply are the grammatical 'indices' of the fact that the denoted referent of the noun

20

Towards a Pragmatics of Discourse

phrase has been identified, or is supposed to be known, by both speaker and hearer. This may be obviously the case through the utterance of previous sentences, or else by previous specific knowledge, general knowledge of the world, or perception of the actual context. At this point we are already halfway into a pragmatic account of presupposition. Contextual semantics though is still restricted to truth or satisfaction conditions, whether relative to other sentences/propositions, or relative to what language users know or believe. However, the acount of presupposition is clearly pragmatic as soon as the appropriateness of sentences, utterances or speech acts comes into play, formulated in terms of conditions involving the knowledge, beliefs, wants, expectations and evaluations of language users. Thus, pragmatically an utterance of the sentence Even F(a) involves the condition that the speaker did not expect that F(a) would be the case or would probably be the case, where expectations are based not only on general or specific world knowledge but also on all kinds of norms, scales and evaluations (For detail, see e.g. Ducrot, 1972, 1974, 1980). Both in Chapter 2 and in Chapter 3, these problems concerning the possible treatment of presupposition and the kinds of information processed in semantics and pragmatics have been dealt with only partially. Recognizing that interpretations may be relative to the semantic information expressed by or (weakly) implied by previous sentences or to the information of language users, it remained to be shown where generalization of these two kinds of 'contexts' is not possible. Since an expression such as the man may refer to any identified man in the language users knowledge, we at least need some device that accounts for the structural information of the text, e.g. by semantic identification (definite description), or 'the one we were just talking about', which requires a focusing or selection device in the actualization of knowledge; a step which brings us towards the boundaries where semantics, pragmatics and cognitive processing theory meet. In all these cases, however, presupposition has to do with formal or actual interpretations of sentences relative to verbal and/or pragmatic context. Thus, other kinds of presupposition should be formulated in

Issues in the Pragmatics of Discourse

21

different terms. If we say, for instance, that in a promise it is 'presupposed' that the speaker has the intention to do p, we should rather say that ΙχΌΟχρ is an appropriateness condition for the speech act, and not a presupposition of the sentence uttered in order to accomplish that speech act. The same holds for other, cognitive or social, conditions for the acceptability of an utterance in communicative interaction. We may conclude from this brief sketch that a text grammar with a pragmatic component is a satisfactory framework for the theoretical account of presuppositions, because it provides an explication of the notion of 'contextual information processing' involved in all presuppositional phenomena. The empirical basis of this framework should be supplied by a cognitive theory of complex semantic information processing, in which it is specified how new input information is linked with stored information derived from previous sentence comprehension and from actual context knowledge or general world knowledge. 4.3. Another fundamental problem in current linguistics for which such a framework appears necessary is that of the topiccomment articulation of sentences. In fact, this problem is very closely connected with that of presupposition, and we may even assume that topic-comment structures and presupposition are manifestations of the same properties of information distribution in discourse and conversation. In both cases, the intuitive notion of Old' or 'given' information is involved, such that topical function is assigned to those phrases of the sentence—or components of the underlying propositions—which are at the same time presupposed. (See e.g. Clark & Hariland, 1977). In the third place, the topic of a sentence may be associated with the more or less cognitive notions of 'attention' or 'actualization', involving concepts being processed in semantic working memory. In that case, the topic has the function to indicate the concepts which were actualized in working memory due to close previous information. This is also the reason why topic-comment tests so often make use of a question-answer format. That part of the answer which has been already actualized by the question may serve as the topic. In other words, the topic-comment

22

Towards a Pragmatics of Discourse

distinctions according to this account do not simply represent the 'given' and the 'new', but rather what of the old is being selected in memory to be connected with incoming information. Whereas the notions distinguished above are relevant at the level of the sentence, we already have it that under each interpretation the notion of topic involves relationships with other, given, old or actualized information. In many cases, as with presupposition, this information comes from previous sentences in a sequence. We will show in Chapter 7 that in that case a fourth distinction is necessary. Thus, if intuitively the notion of topic is associated with 'what the sentence is about', it very often is confused with notions such as 'topic of a sequence', 'topic of discourse', or 'topic of conversation'. Many of our intuitions derive sentence topics rather from what 'has been talked about' in the last few sentences of the discourse. But, sequential topic and sentential topic need not coincide. A sequential.topic is a property of linear coherence and involves repeated co-reference to discourse referents, and thus may be part of the macro-structure of the sequence. A sentential topic is strictly local, and only determines the way a sentence is connected with the established informational structure. Finally, then, and via the notion of 'sequential topic', we have global topics, viz. topics of discourse or conversation. In that case, however, we are no longer at the local or linear level of semantic information organization or processing, but at a more global level. A global topic, then, is any concept contained in or entailed by the macro-structure proposition of a sequence or a whole discourse. Thus, given the macro-structure proposition 'Peter is travelling to Paris' derived from a story in which Peter is travelling to Paris, we may take 'Peter' or 'travelling' or 'Paris' as a possible global topic. If the global topic is 'Peter', then the linear coherence rules will often also have 'Peter' as a sequential topic, or even as a sentence topic. If 'travel' would be a topic, the individual expressions may refer to many, even unrelated, persons in the subsequent sentences. Very often, however, our intuitive notion of a topic of conversation is not limited to such isolated concepts. We would not say, for instance, that 'Peter' or 'travel' is the topic of the story, but 'Peter is travelling to Paris'. In other words, the topic of

Issues in the Pragmatics of Discourse

23

discourse or conversation is a full proposition, and made explicit in terms of macro-structures. We see again that for the phenomenon of topic and comment, and for a full analysis of the notion and its different implications, we need an account in terms of relative interpretations, sequences of sentences, macro-structures, and the pragmatic structures of the context—again with an additional cognitive theory as empirical base. From our observations in Chapter 7, for instance, it follows that the identification of the topic at the sentence level may depend on the topicality of expressions in previous sentences, on the sequential topic and on (part of) the global 'topic', i.e. the macro-structure of the discourse. All this, does not mean that further problems, e.g. those related to the notion of perspective, are not present in this issue, and our Chapter 7 is far from a complete account of the major problems involved. 4.4. The category of connectives, to which certain sentence adverbials, conjunctions and certain particles belong, has been much neglected in both generative sentence grammars and in text grammar. Connectives express relations between propositions, both in complex and compound sentences, and in sequences of sentences. An adequate theory of natural connectives involves both a semantic and a pragmatic account. At the first level they represent all kinds of connecting relations between 'facts' (the values of propositions), and at the second level they represent more the relations between speech acts. Thus, connectives may function as semantic or pragmatic connectives. This rather complex issue is treated in Chapters 5 and 6. It is related to the much broader problem about the relations between clauses in complex/compound sentences, between sentences in a discourse, and the ways in which these relations are determined by speech acts and speech act sequences. It appears, for instance, that certain sequences of sentences cannot simply be reduced to a complex sentence, which is one of the arguments for not considering textual structures as being a simple variation of compound/complex sentence structures. Sentence boundaries, thus, are not arbitrary, but are determined by both semantic and pragmatic factors —and of course by psychological (memory span) and social deter-

24

Towards a Pragmatics of Discourse

minants. Changing a topic of discourse, introducing new discourse referents, changing possible worlds, perspective, etc. usually are sufficient semantic grounds for beginning a new sentence. Similarly, for a change in illocutionary force, e.g. when two different speech acts follow each other. More in general, this problem is an aspect of the relations between pragmatics and semantics (or the rest of the grammar). One of the thorny questions in this framework, for instance, is whether a sentence has some kind of pragmatic correlate, in the sense that by uttering a sentence only one, but possibly composite, speech act may be performed. This is an interesting but very strong hypothesis. Apparent counter-examples can be explained by arguing that one speech act may be based on a composite proposition, expressed by one sentence, or, if different speech acts are involved, that there is possibly a hierarchical pragmatic structure, with a major speech act and 'embedded' speech acts. Although our discussion certainly does not settle this problem, it is at least brought up as one of the crucial issues which must be considered if pragmatics is to be incorporated into the grammar. In the last chapter we will come back to these more general problems about the relations between pragmatics and grammar. The question is whether, apart from their clear pragmatic function when relating speech acts, all connectives can be treated exclusively in purely semantic, intensional and/or extensional, terms. In Text and Context (van Dijk, 1977a) where a systematic study of the relations between the semantics and pragmatics of discourse is given, this issue of natural connectives is given detailed attention. Although an attempt has been made to give much of the meaning of natural connectives in terms of semantics, it seems that at least certain connectives, such as but and because, also involve pragmatic aspects, related to expectations of language users (Ducrot, 1980). Given the discussion about discourse structure above, this may appear to be a more general property of certain natural language expressions, such as quantifiers, definite articles, pronouns, adverbs, particles,7 modalities,8 aspect, etc. In all these cases, the knowledge, expectations or beliefs of language users seem to be involved, about the world, the context, the (other) language user, etc. The important conclusion arising

Notes

25

from this hypothesis, which we try to back up with the example of connectives, is that a pragmatic component is a descriptively necessary part of the grammar, without which the 'meaning' or function of many categories cannot properly be accounted for. NOTES 1.

2.

3.

4.

This development towards a 'cognitive semantics' was initiated, again around 1970, in cognitive psychology and artificial intelligence, in order to account for memory structures and their roles in higher cognitive processes such as language processing, problem solving, action and perception. Somewhat later notions such as 'frame' appeared, especially in artificial intelligence, as part of the general problem of knowledge representation, needed in order to account for the cognitive tasks mentioned above. In that context too interest emerged rapidly for problems of discourse understanding. For a survey, see van Dijk, 1975a, 1977a, van Dijk & Kintsch 1977, Kintsch & van Dijk, 1978, and the references given in chapters 4, 8 of this book. It is surprising that in the many studies of pragmatics, mentioned in the following chapters, no such view has been defended. Either speech act theory was considered independently from the grammar, or certain ad hoc 'pragmatic' modifications (e.g. hypersentences, implicatures) were integrated into syntax or semantics. We by no means would like to claim that the integration of a pragmatic component as proposed in the various chapters, and summarized and systematized here, is the only possible and adequate way of linking grammaticalness and appropriateness, sentences/discourse and action. Yet, we think the proposal is simple and consequent, and not at all incompatible with a serious theory of grammar. See however Katz (1977), Cole, ed. (1978), Franck (1980). It is sometimes argued that pragmatics should not be merely a component of the grammar, but the very basis of the grammar, thus leading to a 'pragmatic theory' of grammar. Although we would readily agree that the functions of language have crucial importance, also in a theory, we would not say in any strict sense that pragmatics has a fundamentally different status, even if pragmatic rules, categories or conditions operate on interpreted syntactic structures, e.g. sentences with a meaning (and reference), uttered in some context. Similarly for the relations between syntax and semantics: even if semantic structures determine syntactic structures, we would not say that the semantics is basic or primary, e.g. because expressions only function in order to express meaning. Of course, in all this there are a number of meta-theoretical intricacies, but in general we would prefer to say that a level of pragmatics, just because it is a component of the grammar, has constraints on both semantics and syntax (and on phonology for that matter), as is usual for the links between grammatical levels in general. Which does not, by the way, conflict with the possibility of a certain 'independence' of structures of each level: it would be hard to prove, for instance, that all syntactic expressions, categories, structures and rules are pragmatically relevant. See the references given in the next chapter. There have been many studies about presupposition since that time. See e.g. Kempson (1975), Wilson (1975), Choon & Dinneen (1979) and Gazdar (1979). For a more extensive analysis of

26

5. 6. 7. 8.

Towards a Pragmatics of Discourse entailment, as the defining notion of semantic presuppositions, we now also have Anderson & Belnap (1975). For these notions, cf. van Dijk, 1977a, b, and references given there. See e.g. Karttunen (1973a, b, 1974), and references given there. See Franck (1980). See Groenendijk & Stokhof (1975).

CHAPTER TWO

Pragmatics, Presuppositions and Context Grammars*

SUMMARIZING INTRODUCTION

The aim of this chapter is to present a general discussion of the linguistic status of a pragmatic theory. It will be argued that much recent work on 'speech acts' takes the basic pragmatic categories for granted, without explicit introduction and definition within a consistent theoretical framework. A sketch for such a framework will be presented below. We will assume that a generative grammar should include a pragmatic component. A grammar which has this property will be called a Context Grammar. Within such a framework we will briefly discuss the relations between semantics and pragmatics, especially with respect to different conceptions of presupposition.

1.

THE STATUS OF PRAGMATICS

1.1. In much classic and recent work on pragmatics, the status of a pragmatic theory is not clear. This is also one of the conclusions of the Jerusalem conference (Bar-Hillel, 1972) on pragmatics of natural language, and the confusion in the field has not been eliminated since. Of course, this situation is to be expected when we recognize how many disciplines are directly or indirectly engaged in the study of 'language in context', as we intuitively may characterized the domain of pragmatics. In the writings of Morris (1938, 1947) pragmatics seems to have a rather different character from its syntactic and semantic co-components in a semiotic theory. Especially the psychological

*Shortened version of a paper presented at the symposium "Zur Grundlegung einer expliziten Pragmatik", Bielefeld (W.-Germany), January 19-21, 1973.

28

Pragmatics, Presuppositions and Context Grammars

and sociological aspects of the study of language use(rs) are retained in his conception of pragmatics. Hardly any remarks are made on the abstract categories and structures to be specified by a pragmatic theory.1 The same holds for much recent work on pragmatics in psycho- and sociolinguistics, anthropology and ethnolo-

gy.2

In the philosophical writings on ordinary language, usually taken as the source of the recent interest for pragmatics, some useful intuitive distinctions were made, such as performativeconstative, illocutionary-perlocutionary, etc., but these have not been made explicit so that they remained vague and often overlapping (Austin, 1962). The linguistic work on pragmatics was either close to sociology and anthropology or tried to integrate some global pragmatic notions into the current framework of syntactically or semantically based generative grammars, e.g. by introducing the device of 'hyper-sentences' (Ross, 1970; Sadock, 1970; see also Lakoff, 1974 and Katz, 1977).

1.2. It cannot be denied that many interesting observations and proposals have been made on some aspects of pragmatics in these philosophical and linguistic writings. Especially in recent work on the happiness conditions for speech acts (e.g., Searle, 1969), on pragmatic meaning (Grice, 1967; Schiffer, 1972) and on some basic notions and empirical foundations (e.g. Wunderlich, 1976b), there is a fruitful discussion of some central problems. In many respects, however, these attempts are often of a heuristic nature, the construction of a systematic set of hypotheses on pragmatic structures of natural language is still a desideratum, and no insight has been gained into the relations between pragmatic structures on the one hand and semantic and syntactic structures on the other hand. It seems clear that before we are able to discuss the actual use of pragmatic rules, we should first formulate them and have some idea about their form and methodological status. On the other hand attempts are being made in philosophy, philosophical logic and modal logic to capture some of the properties of context (Montague, 1974; Kamp, 1976), communication (Harrah, 1963, 1972) and language users (Apostel, 1972), such as

Context Grammars

29

'action', 'intention', 'knowledge and belief, 'message', 'pragmatic indices', etc.3 We will in general follow this more abstract line of research because it at least promises to give an explicit basis for further discussion. Nevertheless, it must be kept in mind that in this initial stage of research empirical observations of actual verbal interaction and of specific grammatical phenomena related to pragmatic structures are very important for the construction of the abstract theoretical framework which must account for them. In fact, much of the formal approaches just mentioned formulate hypotheses wMch are not specific for natural language communication but have a more general application. 1.3. Thus, it is necessary, firstly, to investigate briefly which types of grammar may reasonably be allowed by a theory of natural language, and which links should be established between such grammars and a sound theory of pragmatics. Secondly, in order to be able to provide a serious basis for a 'pragmatic grammar', we must try to give an independent specification of the main categories of pragmatics. Finally, it must be shown how pragmatic structures thus specified are to be related to other structures characterized by the grammar.

2.

CONTEXT GRAMMARS

2.1. The aims and scope of generative grammars cannot be determined a priori. Their descriptive and explanatory goals are set by a developing theory of grammar, which is part of a general theory of natural language. We will not here discuss the form of current grammars. It may roughly be said that generative grammars are restricted to a recursive characterization of some abstract structures 'underlying' idealized utterances of given natural languages. More in particular, they provide structural descriptions of all and only the sentences of that language, at several levels, viz. morphonological, syntactic and semantic.4 More recently it has been claimed that grammars should also provide structural descriptions of well-formed se-

30

Pragmatics, Presuppositions and Context Grammars

quences of sentences, viz. of texts, of natural language (van Dijk, 1972a, Petofi & Rieser (eds), 1973). Although it is not yet exactly known which additional categories, rules and constraints grammars should have in order to meet this last requirement, we will provisionally assume here that any adequate grammar should at least specify the abstract linguistic structures of sentences and texts. We will further assume that the rule system thus constructed has an idealized empirical correlate in the cognitive system of language users, i.e. somehow provides a model of their linguistic competence (Chomsky, 1965). 2.2. Following some recent intuitive suggestions,5 we will further assume that language users have the rule-governed ability to 'adequately' use utterances of their languages in all possible communicative situations, and that this ability is part of their linguistic competence. On the 'model of competence' view of grammar, this hypothesis implies that an empirically adequate grammar must formulate the rules, categories and other constraints abstractly reconstructing these abilities. However, even when we assume that such a hypothesis can be consistently formulated and that some empirical warrants can be provided to sustain it, it is not yet clear which abstract objects a grammar thus extended should describe. If indeed it should represent the ability to 'use' certain utterances adequately in a given situation, it may be expected to characterize all possible situations and match these with all possible utterances. It is obvious that such a strong claim can be seriously advocated only when we put heavy restrictions on the notion 'possible situation'. That is to say, we must at least have some idea which situations or properties of situations are linguistically relevant and, for that matter, which of these properties should be accounted for in the grammar. 2.3. In order to come closer to a formulation of a serious answer to these questions, we will introduce the abstract notion of context. Intuitively, a context is the linguistically relevant set of characteristics of a communicative situation, the latter being the

Context Grammars

31

state of affairs in which communicative events 'in' natural language take place. Thus, a context must have exactly those properties which are sufficient and necessary for the formulation of the conditions and rules for the adequate use of utterances. Specifically, a context may be characterized simply as a set of components in 'happiness conditions' for utterances in natural language.6 We will say that a given utterance is happy or appropriate with respect to a given context. Conversely, a context may be said to be satisfactory, or not, for a given utterance. These notions are all scalar: an utterance is more or less happy, a context more or less satisfactory. It is clear that the notions intuitively introduced here need explication in a theoretical framework. The notion of context seems rather static when it is merely used to refer to a state of affairs. We therefore additionally introduce the term communicative event, which we provisionally take as a specific change relation over contexts. A communicative event is said to be successful if a given context changes into a specific new context. A context grammar is required to describe the structure of all appropriate contexts, relative to well-formed utterances, and to specify under which conditions a change from context into context is successful. Since happiness conditions do not merely depend on the structure of appropriate contexts, but also on the properties of utterances, context grammars include a component for sentences and text characterization. Rather generally speaking, indeed, we may say that conditions for morphonological, syntactic and semantic well-formedness are a subset of the set of happiness conditions. Below, however, we will restrict the notion of happiness condition to the structure of pragmatic contexts. The status of logical or referential (truth-) conditions will be discussed separately. 2.4. Furthermore, context grammars must specify how given pragmatic structures, i.e. contexts, impose constraints upon the structure of utterances. More in particular, assuming that the abstract structures underlying utterances are characterized by semantically based text grammars, a context grammar must formulate the rules mapping pragmatic structures into semantic struc-

32

Pragmatics, Presuppositions and Context Grammars

tures, i.e. contexts into texts. Let us try to make these very general requirements on context grammars more precise by systematically constructing the notion of context. 2.5. It should be stressed that the usual competence-performance distinction, although not wholly unproblematic from a psychological point of view, is respected here. Utterances which are formally speaking happy with respect to their contexts, may not be actually acceptable in concrete communicative situations, and conversely. Of course this raises the more general problem of the empirical basis of a grammar which we cannot discuss here. It must be kept in mind, however, that serious theory construction cannot be based on ad hoc data and on features of contexts which are only psychologically and sociologically relevant. In principle we take the view that native speakers have intuitive insight in to the pragmatic rule-system of their language, so that we may test the hypothetical conditions and rules of the context grammar on actual judgements about the appropriateness of contexts and the happiness of utterances. We are also aware of the fact that pragmatic rules, perhaps even more than syntactic, morphonological and semantic rules, may be different depending on class and group conventions in verbal and social interaction 'within' the 'same' language group. However, these differences should perhaps rather be characterized by specific constraints upon utterance structures than by specific context structures, which might very well be general if not universal: the underlying structures of questions, insults, etc., but not their manifestation in the utterance, seem to be independent of dialect or even language differences. Of course this hypothesis needs empirical investigation.22 2.6. In the following chapters we will only use the notion of appropriateness for utterances interpreted as speech acts. This notion is relative, viz. with respect to a context. Hence, the inverse relation, viz. the notion used to indicate that the structure of the context is such that some utterance taken as some speech act is appropriate in it, will need another term, e.g. 'pragmatic satisfaction'.

Pragmatic Structures 3.

33

PRAGMATIC STRUCTURES

3.1. One of the basic intuitive properties of verbal interaction is undoubtedly the establishment of a special kind of social interaction in which a 'speaker', by way of producing an 'utterance' tries to 'change the mind and/or the actions' of a 'hearer'. Let us try to reconstruct this property in a more or less systematic way. 3.2. Although notions such as 'communicative event' or 'communicative interaction' should be defined in a more general theory of communication, let us briefly recall their major properties, in order to be able to specify which specifically linguistic and grammatical aspects should be distinguished. An event may most simply be characterized as an ordered η-tuple of state of affairs, , where the ordering is defined by a change-relation over states. This relation is a linear function of time (intervals or moments). Simple events are defined for n=2, complex events for n >2, where s^ in each case is the 'initial state', sn the 'final state', and s{ (/ prepositional attitudes) are based on propositions, it is also possible, and usual, to have more than one such act accomplished by the utterance of a single (complex) sentence. Thus the utterance of a sentence like Although I have a head-ache, I'll help you to solve your problem is both an assertion and a promise (CONTRA 6).

However, although as we see a function from speech acts to sentences is not one-one, we might introduce the notion of a macro-speech act, for reasons parallel to the introduction of the notion of macro-structure in text grammar. 12 That is, it may be the case that an utterance as a whole, although 'including', locally, several (micro-) speech acts, say as a type of sub-programs, has the function of one global illocutionary act. In the example above, this would be a promise. A long request may contain several assertions, e.g. as motivations for the request, but nevertheless as a whole have the function of a request. It might well be the case that the assumed macro-structures of a discourse thus receive additional confirmation from the pragmatic level. A macro-speech act would in that case be a function from a given context structure to the set of macro-propositions of an uttered discourse. If this assumption can be made explicit we

66

Partial Equivalence of Text and Context Grammars

would here have a decisive argument' against the reduction hypothesis, because, even when sentence grammars would claim to be able, in principle, to generate ordered η-tuples of sentences, they would (at least in their present form) not be able to generate macro-structures. Since certain pragmatic phenomena require such macro-structures as their scope, a context grammar must be (include) a text grammar. (CONTRA 7). 9. The last argument is, as far as we are concerned, of very remarkable interest for the explanation of a series of major problems in linguistic and cognitive theory (see also Chapter 8). It is a well-known fact in cognitive psychology that all complex behaviour is programmed on the basis of plans.13 These plans have a global character and direct the well-ordered and functional execution of mental and bodily acts towards a certain goal. The same must hold for the accomplishment of illocutionary acts. The central, global aim of such an act is to cause a change in the internal knowledge and preference sets of the hearer (leading eventually to the formation of intentions to act in a certain way). In order to attain that goal, a speaker may have a complex illocutionary strategy at the 'micro-level'. Such lower level acts may then become functional for higher level acts. Such assumptions, although briefly and vaguely formulated here, find confirmation in psycholinguistic research. It is known that we are normally unable to memorize a longer utterance in all syntactic or even semantic detail. We mentally summarize and abstract from the set of propositions—depending on our knowledge, preferences, and interests of a given moment—by forming macro-structures. For pragmatic theory such cognitive evidence implies that the speaker of a longer utterance cannot possibly intend (at least appropriately, and with some positive effect) that the hearer 'knows' all propositions manifested in an uttered discourse. At least, this 'knowledge' is merely intended to have a 'short term' character, and is intended to construct a more global point or theme of the utterance. This requires in the theory a weighting procedure with respect to the 'importance' of certain propositions. The usual pragmatic formulas like 'Speaker intends hearer to

Partial Equivalence of Text and Context Grammars

67

know/believe . . . ' would thus become more complex when macro-structures and macro-acts are to be accounted for. The very existence of performative verbs like resume or 'performative adverbs' like in brief or briefly is an indication for the actual relevance of macro-structures in actions and discourses. It has been shown above that for a serious treatment of the 'unity' of very complex illocutionary acts, we must assume that utterances are construed at the syntactico-semantic levels as discourses, not as sentences. Conversely, taking utterances as uttered texts guarantees the process-character of the communicative event. 10. The few and informal counter-arguments formulated above seem to point to a rejection of the Reduction Hypothesis. If the addition of a pragmatic component to the grammar is necessary, we must assume that such a grammar provides structural descriptions of texts (including, of course, sentences, i.e. assigning structural descriptions of sentences relative to those for other sentences). In other words: to make a sentence grammar context-sensitive is descriptively insufficient because the structure of previous sentences cannot be given in pragmatic terms; the 'previous' structural descriptions must be connected in one structural description for the whole discourse. Of course, the names 'sentence grammar' and 'text grammar' are merely labels: we just have grammars. When grammars of the current generative transformational type, based on explicit natural logics accounting for the abstract underlying structures interpreted in adequate modeltheoretic ways, are able to characterize discourses, there is no need for specific 'text grammars', because they simply would be text grammars in the sense intended. Very little is known at present about the precise relations between pragmatic and other grammatical structures. The very general discussion given above is meant as part of the methodological framework in which these relations may fruitfully be studied and as an informal refutation of the wide-spread conception of discourse/text as merely a pragmatic or even a performance entity.

68

Partial Equivalence of Text and Context Grammars

NOTES 1.

2. 3. 4.

5. 6.

The concept of 'relativity' in linguistic theory appears in several recent studies, especially in connection with treatments of presupposition. See Lakoff (1971), Karttunen (1973a: 185; 1973b, and other writings; his 1973b comments on Thomason's paper (1973) who, in the current logical discussion on contextual constraints on truth value assignment, discusses the boundaries between semantic and pragmatic presupposition. See also Kasher (1973). For numerous references to work on text grammars, see e.g. van Dijk (1972a, 1977a), van Dijk & Petofi (eds.) (1977), Petöfi & Rieser (eds.) (1973), Dressier & Schmidt (1973) and Dressler (ed.) (1977). An interesting example of such a formal semantic treatment of textual relations is provided by Ballmer (1972). For the notion of 'context grammar', see Chapter 1, as well as the references to other pragmatic work given there, which cannot be accounted for in this short note. For a serious formal treatment of 'pragmatic' aspects of sentences, viz. in the form of a model theoretic semantics for explicit performatives, see Aqvist (1972). Roughly the view of pragmatic theory and its tasks, in this chapter, is close to the work of Kummer (e.g. in Kummer, 1975). The notion of 'context' used in this paper-which is not here de fined-differs from other uses of the term. In the first place it should not be identified with the performance notion of 'situation', which is more general and less abstract. Many linguists simply count (preceding parts of) discourse as belonging to the ('verbal') context. In recent discussions this notion of context is often more narrowly restricted to a set of sentences, being true or presumed to be true, at the moment of utterance. In that case such a set would represent an epistemically accessible possible world obtaining at the moment of utterance. Without being explicit, we could view a context 'statically' or 'dynamically', viz. as an (initial) state description or as a process description. In both cases this description would pertain to the states of the speech participants (speaker, hearer)-internally (wants, preferences, intentions, knowledge, belief) and externally (topological, actional)-and of the utterance itself. Hence a context is an ordered subset of possible worlds (situations), selected by partial characterizations of some 'speech' individuals and events, viz. those which are relevant for the syntactic and semantic structure of the uttered discourse. This point has also been made in a paper by Isard (1975). See the references in note 1. From our discussion it follows that a presupposition set, as assumed by Karttunen and others, must either somehow be structured (ordered) or part of what it performs theoretically must be accounted for in a semantics of discourse. The idea of a 'presupposition (or data) set/base' has been used by other authors as well, e.g. Bartsch (1972). Stalnaker (1973), also using a similar construction, localizes it completely at the pragmatic level, viz. as a set of speaker-assumptions. Although the different treatments are converging on some very important properties of natural language none of the mentioned articles gives explicit elaboration of such adopted concepts as 'presupposition set' or 'context'. A possibility remaining undiscussed in this paper is a sort of Converse Reduction

Notes

7. 8. 9. 10. 11.

12. 13.

69

Hypothesis, assuming that presuppositions as well as the whole context in the form of a maximally consistent (?) Henkin set describing the context, may be accounted for in a text grammar, thus reducing all pragmatic determinants to those, syntactic and semantic, of expressions and their interpretations of the language itself. Although such an hypothesis is interesting and although we do not have at present enough insight into the implications of such a proposal, based on the philosophy of 'expressibility' (defended e.g. by Searle, 1969), it may be assumed that such a procedure is reductive in the wrong sense, because it would not meet the basic tasks of pragmatic theory, which is not to define truth for sentences but appropriateness for utterances. See Langacker (1969) for these constraints and the discussion in van Dijk (1972a, Chap. 2) where it is argued that backward pronominalization is transformationally derivative and requires ordinary precedence relations. See Cohen (1969) for the view that 'referring' as an act should be treated as a part of our linguistic competence. For current work in logical semantics of natural language referred to here, see Davidson & Harman (eds.) (1972) and Hintikka, Moravcsik & Suppes (eds.) (1973), Keenan (ed.) (1975). For a discussion of these logical aspects of identity and reference in texts, see (among other papers) Kummer (1971a, b) van Dijk (1973a) and Ballmer (1972). An attempt to give some semi-formal semantic conditions for connectives has been made in van Dijk (1977a, b), in which numerous references can be found to the current logical work on connectives and on the problem of 'connection' and relevance between formulas in general, especially for conditionals of different strength (counterfactuals, entailments, etc.). In this work too, the notion of 'context' or 'circumstances' in logical theory becomes increasingly important. See especially Aqvist (1973) and Lewis (1973). For the notion of 'macro-structure* see the informal treatment in van Dijk (1972a: Chap. 3; 1977a, Chap. 5; 1977c, 1980). The importance of the cognitive concept of plan-see Miller, Galanter & Pribram (1960)-in linguistic theory has been stressed early by Miller & Chomsky (1963). A fortiori such a concept is important for the explanation of our ability to produce and interpret such still more complex units as discourses. At the level of linguistic actions, we are thus able to relate macro-structures of texts with macro-speech acts of contexts. See Nowakowska (1973) and the discussion in Kummer (1975). See van Dijk, (1977c, 1980) for a general, cognitive treatment of macro-structures in discourse comprehension, and the relations to notions such as 'schema', 'frame' and 'plan'. See also the recent book by Schank & Abelson (1977) studying these notions from an artificial intelligence point of view. See the discussion in Chapter 9, for detail.

CHAPTER FOUR

Acceptability in Context*

1.

THE PROBLEM

1.1. One of the crucial problems of current grammatical theories is their empirical foundations. What do grammars actually account for? Linguistic intuitions of native speakers, systematic language use, a set of socioculturally determined conventional norms or something else? It need hardly be recalled here that Chomsky's Saussurean distinction between 'competence' and 'performance' has met with growing criticism, both in psycholinguistics (cf. Bever 1970) and in sociolinguistics (Labov 1970). On the one hand, it has been stressed that linguistic intuitions are only a limited part of our linguistic abilities and that cognitive strategies play an important role in our verbal performances. On the other hand, a social basis in the form of a 'homogeneous speech community' has turned out to be a too gross simplification, and the usual conception of grammatical 'rule' is presumably too idealized. Such and similar objections, based on serious empirical research, has blurred the handy methodological distinction between theoretical properties of utterances, viz. their grammaticalness, and their 'real' properties, determining their factual acceptability. 1.2. It is the aim of this chapter to shed some light on this problem from the point of view of recent work in text or discourse grammars and in pragmatics and their empirical, especially cognitive, basis. In both research directions it has been shown that grammaticalness of sentences and acceptability of utterances, This paper was written in 1974.

72

Acceptability in Context

respectively, should be accounted for relative to the structure of verbal and non-verbal context. Nevertheless, it should be made clear whether the attempts toward the elaboration of some fragments of a text grammar can be empirically warranted (at least from a formal or theoretical point of view). Some familiar questions, concepts and criteria are expected in such an inquiry, e.g., with respect to the 'psychological reality' of discourses or discourse rules and categories or with respect to the existence of linguistic intuitions about the well-formedness or coherence of discourses or of a clear distinction between a text and a semi-text or non-text of a language.1 In the perspective of current work in psychology and psycholinguistics such a 'traditional' approach to the problem of the empirical basis of text or context (pragmatic) grammars must be made with care. For example, we have no a priori grounds for deciding whether some phenomenon should be accounted for in terms of a grammatical 'rule' or in terms of a cognitive 'process'. That deficiency seems to weaken the usual criticism of text grammar, viz. that sentences belong to competence/grammar and discourses to performance/cognitive or social theory, since such a distinction is no longer clear even for sentences. Thus, we witness a progressive merging of grammars and cognitive models, expecially in recent proposals in Artificial Intelligence.2 1.3. Our main point, thus, is to show that, even if a more sophisticated version of the distinction between grammaticalness and acceptability is maintained, such a distinction should be made explicit in the perspective of systematic (con-)textual analysis. In other words, if a grammar is the central, theoretical core of a theory of performance, it should contain a set of discourse rules and a pragmatic component in order to establish the required connection with models of cognitive strategies and social conventions.

2.

THE PHILOSOPHY OF 'ACCEPTABILITY'

2.1. Since the notion of 'grammaticalness' has been given ex-

The Philosophy of 'Acceptability'

73

tensive attention, it seems useful to begin our discussion with a brief specification of some central features of the different notions of 'acceptability' in social, psychological and linguistic (meta-) theories. We want to know which explicit meaning can be assigned to such terms as 'accept', 'acceptance', 'acceptability' and their cognates. 2.2. A first problem in such a preliminary philosophical account is whether '(to) accept' should be described as an act. According to recent work in action logic,3 an act is a bodily event, a 'doing'— or the absence of such a doing—caused by a conscious organism, a 'person', able to control his own doings, intending to perform such a doing and usually with the purpose to thereby cause other events to happen, to not happen or (not) to continue. This awkward sentence is an abbreviated informal definition of the results of much philosophical and logical work and neglects specific intricacies. Now, under this definition, at least one reading of the term 'accept' can be described as an act. When I accept a present or an invitation I usually perform a series of doings (e.g., moving my hand, nodding or saying something). These doings are intentional because I have to decide whether I shall accept something or refuse or reject it. The purpose of such an act may be the event/state of my becoming/being 'happy' with the thing I accepted, i.e., the realization of the purpose may be consistent with my wishes and/or those of somebody else or be consistent with some social convention. The accepted 'thing' may be an object that is concrete and transferable (i.e., in possession), an act or an interaction, with which the desired state or event can be brought about. We use the verb to accept, however, in this reading only in order to denote an act which is not 'situationally evident', so to speak. If say, somebody gives or sends us a book, it is natural to 'take' or 'receive' it. In an 'accepting' situation, however, there exists a serious possibility that the thing offered may be refused or rejected, as we saw above. In such situations we decide whether to accept or to refuse, and hence we have reasons or grounds for such acts. These reasons are based on our evaluation of the offered object with respect to the chance that its acceptance realizes or continues a desired state of affairs. Hence, the object must satisfy

74

Acceptability in Context

a number of specific properties judged satisfactory in that perspective. One of the additional conditions is finally that the one who accepts, the 'acceptor', not only has the freedom to decide whether to accept or not but also has the recognized ability or right to judge whether the desired properties are satisfied. This places the acceptor, perhaps only momentarily, in a dominating position with respect to the Offerer'. The logical form of the predicate 'to accept', thus, is at least a four place relation:

(1) X accepts Υ from Ζ because of V or

(2) X accepts Y, with properties W, from Z. A formulation closer to our normal usage of the term would be: «

*

(3) X accepts Υ as U from Z. e.g., in such sentences as 'He accepted that book as a present from John', or The faculty has accepted that book as a doctoral dissertation', where the objects must satisfy' certain criteria. In the formulate (l)-(3) X and Z are person variables, Υ an object variable and V, W and U are property (or intentional object) variables. Other arguments, e.g., for time, place or circumstances, may be added: we may accept something in some possible world which we reject in another because our desires or their possible chances of realization may be different. 2.3. The above analysis, in philosophical jargon, of the act of accepting seems to bear also, at least partially, on our usual understanding of 'acceptability' in linguistics. Substituting the variables in (3) for the corresponding constants, we would get something like: (4) A (native) speaker-hearer accepts an utterance from another speaker (e.g., a linguist) as a sentence of his language. This is roughly the full form as it is usually understood, but a couple of difficulties arise. It is true that a native speaker,

The Philosophy of 'Acceptability'

75

according to our definitions given above, must be assigned the abilities to judge whether the object satisfies certain properties and, if not, to reject or refuse the object. The properties in question are both grammatical properties and 'cognitive' properties like structural complexity, length, etc. However, it is well-known that the language user may well not be able to accept the utterance 'as a sentence' because he may not have the notion or concept of a sentence at all and certainly not the theoretical concept of a sentence. Thus, he may accept the utterance simply 'as a (good, normal) utterance of his language'. Acceptance, in such cases, may be implicit (tacit) or explicit. The explicit case is rare and only occurs in those situations where evaluation of utterances is required (in tests and in teaching the language). Acceptance (or rejection) in that case is the intentional act, based on motivated decisions, where the motivation may be intuitive but where the decision itself must be 'known'. As Bever (1970) has pointed out, such an act requires specific (linguistic) abilities. The implicit case of acceptance is the 'normal' or natural one and occurs in the course of a conversation. It may be asked whether 'accepting' in such a case is an act or whether accepting is involved at all. When hearing an utterance of his language during a conversation, a language user does not seem to do something which we may qualify as the act of acceptance. To be sure, he does listen or read, which are acts, because they are controllable doings: he may decide not to listen or to read. But, the act of acceptance is not simply performed by accomplishing the acts of listening or reading, since these acts are also conditions for rejection. Next, phonemes/morphemes will be identified and the syntactic structure analyzed with a set of rules and strategies together with a semantic 'parsing' that puts interpreted words or phrases in the appropriate 'logical' categories of a meaning representation, etc. This is (very, very) roughly what happens, and perhaps the language user also does these things, but they occur to his mind rather than his causing them intentionally to happen. They are not even 'mental acts' (whatever that may mean) but series of mental events, i.e., processes. If the process takes a normal course, i.e., if the syntactic structure corresponds to the rules present and

76

Acceptability in Context

the semantic interpretation makes sense, the utterance (as such) is automatically accepted, much in the same way as a computer may accept a sentence if it has the appropriate programs to process it. The mental processes going on are thus normally not thought about and may even be inaccessible. We only know that something is wrong when the imput utterance does not satisfy the different rules, strategies and categories we have available (our 'tacit' language knowledge). In that case there are grounds to reject the sentence or part of it, a rejection which may be expressed or not. The expressions of rejection are often conventionalized and need not be treated here. Thus, whereas rejection is an act according to the definition, acceptance of a linguistic utterance is either simply a cover term for the complex mental processing or it is a 'negative' act like omissions or forbearances. However, as acts these doings must be intentional, which is hardly the case in sentence processing. So, we'll say that acceptance is an automatized mental act (or series of mental acts) because in certain circumstances it is controlled consciously, thought about and carried out. In this respect the act is like that of a decision (or even is a decision, whether the input utterance satisfies the rules and structures in the processors). Hence, the sole external evidence that a mental act of acceptance is carried out in a course of conversation is the absence of explicit rejections. The acceptance of an utterance, moreover, is not merely based on syntactic and semantic rules/strategies but also on pragmatic rules, conditions and structures. In that case the utterance is accepted not (only) as an object but as a speech act. Instead of talking of an utterance-object (a token of an utterance type or a sentence type), we may analyze the speech act in the usual way and perhaps speak of morphophonological, syntactic and semantic (propositional, referential) 'acts', with the same proviso for 'acts' as made above. At these levels, the different acts may be said to be appropriate or not, depending on whether in most similar circumstances most language users would accept them, which is part of the definition of a convention (Lewis 1968). This pragmatic condition reminds us of the fact that the simple phrase 'acceptability of an utterance' (worse: Of a sentence') not only can be intended as implying an 'act' by hearers/readers but should

Relative Grammaticalness

77

also imply some act of the speaker having produced the utterance. 'Producibility' seems too awkward, but we would need a term indicating the fact that an utterance also satisfies the rules and structures (and the intentions) of the speaker. Although we may ideally require that the producible is also acceptable, this may in fact often not be the case. We shall now attempt to demonstrate that only relative or contextual grammaticalness reflects systematic processes of acceptance.

3.

RELATIVE GRAMMATICALNESS

3.1. There are well-known cases of sentences which are grammatical but not acceptable. Let us here analyze some examples of the converse case: 'sentences' which are acceptable but not grammatical in the strict sense: (5) A: B: (6) A: B: (7) A: B:

Did you hit him? No. He me. With what has the postman been murdered? John thinks with a knife. Sorry, I couldn't make it in time. Obviously.

As such, the utterances of speaker B can hardly be called grammatical by a grammar accounting for isolated sentences, although they are perfectly acceptable in the whole conversation. Hence cognitive strategies should be called on to help explain such cases of systematic (i.e., not ad hoc} acceptability. Strategies operating on the typically incomplete B-inputs alone, however, can not be of much help. In (5) there may be ambiguity between direct and indirect object me, in (6) the first interpretation would be that with a knife is an instrumental linked with the immediately preceding predicate think and in (7) the adverb may be sentence- or predicate-modifying. Hence, any serious cognitive model can account for the B-utterances only in relation to the Α-utterances. Structural and substantial information from the

78

Acceptability in Context

preceding utterance is thus necessary for the interpretation, and hence the acceptability, of the following utterances of the conversation: in (5) he subsitutes for you (referring to B) and me substitutes for him, where the verb/predicate remains identical and is deleted in surface structure, whereas in (6) and (7) a whole proposition is deleted for which one additional category (instrumental, adverb) is provided. It should be stressed that whatever cognitive processes are involved, they must at least partially be based on (all-or-none) rules: e.g., He I, in (5) would be both ungrammatical and unacceptable. So it seems to make sense to speak of grammatical and ungrammatical incomplete sentences and that requires treatment of dialogue discourses like (5)-(7) in the grammar. The particular rules, e.g., of substitution and deletion, involved in these examples may well have proper cognitive correlates (avoiding repetition, perceiving/marking contrastive information, etc.). 3.2. The problem, then, is: what would such a grammar look like? First of all, it must be stressed that the phenomena discussed above regularly appear within the sentence: (8) I hit him, and he me. (9) Peter thought that the postman was murdered with an umbrella, but John thinks with a knife.

So, in any case, a grammar must capture the generalization that certain rules apply similarly in complex sentences and in sentence sequences. Given that a classical sentence grammar (syntactically, semantically or categorically based) can account for (8) and (9), the argument would be that the same rules (transformations) would account for the discourse phenomenon. This view is the standard criticism against text grammars, and it is correct-at least, as far as these cases are concerned. Of course, some adjustments must be made (for example, transformations would have to apply across 'independent' S-nodes, in which case, what sort of a node would the topmost be?) but they would (perhaps) be marginal. If this is correct, and if the grammar captures the generalization, it would no longer be a sentence grammar but a (weak) discourse

Relative Grammaticalness

79

grammar because it characterizes structures of discourses. More correctly, the difference would disappear; we would just have grammar, and that is the way it should be. Thus, an adequate grammar accounts for the structure of (complex) sentences and of certain structures of sentence sequences (discourses) which are based on the same rules determining the structure of complex sentences. In classical terms this might imply, for example, that in such cases the deep structures of the sentence/sequence are identical but that their surface structures are different. One problem immediately arises here: why are these surface structures different; are they just structural variants, i.e., having the same meaning and pragmatic function? Other questions arise at the same time: if the different structures are variants and if they are based on the same rules, may we freely make sentences out of sequences and sequences out of complex sentences? If so, sentence grammars and discourse grammars coincide. Of course, that assumes that classical sentence grammars provide all the semantic constraints determining complex sentences (which they don't).4 3.3. Although it may be demonstrated that important differences (especially pragmatic) hold in general between complex sentence expressions and sequences, we shall take some clear cases where a sequence is not easily reduced to a complex sentence with the same meaning: (10) (a) (b) (11) (a) (b) (c) (12) (a) (b)

John! Can you hear me? Shall I help you? *John, can you hear me and shall I help you? Can you tell me the time? I have no watch. *Can you tell me the time and I have no watch. Did you tell him the time because he had no watch. I promise to be there in time. Will you also be there? *I promise to be there in time and will you also be there? (13) (a) Yesterday I had a funny dream. I was president and. . . (b) * Yesterday I had a funny dream and I was president. .. (14) (a) Peter is drunk. He always is when he visits Amsterdam.

80

Acceptability in Context (15) (a) Perhaps Harry is ill. He was not at the meeting. (b) *Perhaps Harry is ill because he was not at the meeting.

In all these cases the (b)-sentences.are ungrammatical, awkward or mean something different from the (a)-sentences. Apparently, utterances manifesting different speech acts (like request and statement in (11)), having different (meta-)levels of communication as in (10), sentences under the scope of some 'worldcreating' noun (predicate) as in (13), generalizations (14) and motivations/conclusions can not simply be expressed in one sentence. Although we cannot, in this brief chapter, give a serious analysis of the examples, we may note that the ungrammaticalness of the (b)-sentences is essentially for pragmatic reasons. Thus, in (11) the meaning of the second sentence is not directly connected with the meaning of the first sentence. Rather the proposition underlying the second sentence is a condition for the appropriateness of the speech act of a request or question, i.e., a question is appropriate only if I do not yet know the answer or have no means to supply the answer myself (by looking at my watch). More generally, we might suppose that a sequence of sentences is a one-many mapping from speech acts: one speech act may be accomplished by the utterance of one or more sentences, but several speech acts cannot be accomplished by uttering one sentence except perhaps in cases of indirect speech acts (see Searle (1973) and Franck (1974)). The theoretical problems are intricate here. An alternative proposal may be that sentences and speech acts are related by a one-one mapping: each sentence-utterance accomplishes one speech act. Yet, at least at one level of description (see below), it might be useful to let a whole sequence-utterance be one speech act, e.g., a statement. Thus the sequence (16) I am cold. Could you please close the window? manifests a statement and a request, but as a whole seems to function as a request, where the statement specifies one of the conditions, viz. the motivation, of the request. If this observation is correct, mappings from pragmatic structures may have dis-

Relative Grammaticalness

81

courses or fragments of discourses as their scope and not only sentences. Notice also that a one-sentence version of (16), viz. (17) I am cold, so could you please close the window? has very peculiar properties. The connecting so, here, does not have semantic character, i.e., relating facts or propositions by causal, logical or conceptual implication, but has pragmatic character: it relates the fact expressed in the first sentence with the action (viz. the request) of appropriately uttering the second sentence, not with the 'content' of that second sentence, at least not directly. This is perhaps a reason why the written version of (17) is not well-formed because so should introduce a new, independent sentence. Our intuitions about sentence boundaries seem to be very unreliable indeed as an empirical base for a (sentence) grammar. 3.4. The conclusions which may be drawn from the preceding (admittedly still highly informal) remarks seem to be the following: (a) a grammar accounting for isolated sentences underpredicts their (un-)acceptability—as utterances—in a discourse; (b) acceptability of sentences in a discourse is not only based on cognitive interpretative processes but on rules which may also hold for complex sentences; (c) theoretical notions like 'sequence', 'discourse' or 'text' are necessary because not every discourse can be reduced to a complex sentence (nor the converse, cf. //-clauses); (d) whatever the semantic equivalence of sentences with sequences, their surface differences are based on underlying pragmatic differences. Finally, it may be noticed without further examples that sentences which as such are both grammatical and acceptable may become unacceptable in the discourse, e.g., because of presupposition violations. That is, a sentence can be interpreted only, and hence accepted, relative to the set of interpretations of previous sentences (a set which may be empty, see below). In formal semantic terms, not to be spelled out here, this means that a sentence can be interpreted only in those model structures related with the model structures of the previous sentences. That is, instead of sentence models we have discourse or text models

82

Acceptability in Context

(Ballmer 1972). Thus, the general conclusion is that acceptability is in principle better 'modelled' by a grammar accounting for the structure of sequences.

4.

PRAGMATIC ACCEPTABILITY

4.1. Sentences are not only interpreted/accepted with respect to previous sentences (indeed, the previous sentence set may be empty as we saw) but also to pragmatic context. Neither the precise structure of the pragmatic context nor the rules and constraints relating it with the structure of the utterance can be discussed here.5 We shall assume a pragmatic context to be defined primarily in terms of sets of propositions and rules characterizing the internal structure of speaker and hearer: their knowledge, beliefs, wishes, etc. Semantic models are in this way 'contextualized' with respect to speaker and hearer and some other properties of the (fragment of the) possible world in which they are communicating. A first generalization which comes to mind is to reduce all 'textual' rules to 'contextual' rules. That is, we let constraints from preceding sentences be equivalent to constraints from 'previous' propositions in the epistemic sets of speaker and hearer, changing linearly with the production of subsequent sentences. In Chapter 3, we have given some arguments against such a reduction of a discourse grammar to a pragmatic sentence grammar: the pragmatic component must be added to a proper discourse grammar. Above, for example, we hinted already at the fact that pragmatic mappings may have whole discourses as their scope. Although it cannot be denied that sentences change the pragmatic context (e.g., the knowledge of the hearer), the constraints in an uttered discourse cannot be fully explained on the basis of an (ordered?) set of propositions alone, e.g., a presupposition base. On the other hand, such a pragmatic base is necessary, e.g., for the interpretation of pronouns that have no antecedents and are being used deictically. Similarly, previous sentences as such are not enough to provide the necessary information for the inter-

Pragmatic Acceptability

83

pretation of following sentences; entailments, meaning postulates, etc., based on previous sentences must be supplied by the set of pragmatic presuppositions and corresponding inference rules. What is also needed is the preceding verbal (surface) structure of previous sentences. Assuming that a passive sentence has an abstract underlying proposition with a structure like its active counterpart, we would not be able to explain why the following discourse is ungrammatical: (18) A: Was he hit by you? B: *No. He me. Other arguments against the reduction hypothesis relate to the use of connectives, adverbs, predicates, etc. (thus, consequently, conclude), presupposing previous sentences/utterances, not merely propositions (which might have entered the epistemic set of the hearer in a different way). Hence, a discourse may have its proper underlying theoretical unit (a 'text') even when, in performance, production and perception are controlled by linear cognitive processes moving from sentence/clause to sentence/clause. 4.2. Utterances, as we saw, are acceptable if their underlying discourses satisfy the rules of relative grammaticalness and interpretability. At the same time an utterance is acceptable in a conversation only if it is a speech act which is also appropriate relative to other (speech) acts of the conversation or interaction. The various appropriateness conditions are those given in recent philosophical and linguistic work and need not be specified here for the different speech acts. Thus it seems difficult to have one discourse-utterance manifesting both a request and an order because the preparatory conditions for these acts are inconsistent. Similarly, a request for an object can not precede or co-occur with the act of taking the object by force. Thus, just as a sentence can not be said to be grammatical/acceptable in isolation, so a speech act can not be said to be acceptable in isolation: two speech acts which as such are both appropriate may be incohernet or inconsistent or may have conflicting contextual pre-conditions. The same holds'for the function of speech acts in interaction, though

84

Acceptability in Context

this is not an object for linguistics proper but for sociolinguistics or sociology. 4.3. An additional argument must be given for the hypothesis that a discourse apparently manifesting more than one speech act has nevertheless, as a whole, the function of one speech act. Consider the following dialogue: (19) A: May I borrow your car? I'll bring it back at five. B: O.K. The affirmative answer of speaker B may be given only after a question-request and is usually pointless (or optional) after a promise. Therefore, it seems as if A's discourse should be interpreted as a request or as a request act where the promise act is 'auxiliary' or subordinated to the request. If the latter observation is correct, it would seem sensible to assign a hierarchical structure to a speech act sequence of a conversation, much in the same way as our other_ actions, though linearly (or concurrently) ordered in time. We light our pipe as an auxiliary to the act of smoking and strike a match in order to light the pipe, etc. Hence the acceptability of a word, phrase, sentence, discourse and their respective utterances and speech acts themselves is relative not only to linearly preceding elements but to a hierarchical structure. This means that discourses, at the pragmatic level, too, cannot possibly be satisfactorily described in terms of a sentence grammar merely generating sequences of which only the sentences have hierarchical structure.

5.

MACRO-STRUCTURES

5.1. Returning to the structure of the discourse, we have one result, viz. that grammaticalness of sentences is linearly relative to preceding sentences or propositions in text and pragmatic context, and one hypothesis, viz. that sentences are hierarchically relative to the semantic representation of the discourse, as it is syntactically expressed in complex sentences. Consider, e.g.,

Macro-structures

85

the following discourse: (20) Mary has met a bright guy from Harvard. He is red-haired and wears horn-rimmed spectacles. Here, the second sentence is semantically modifying the object of the first sentence. Similarly, further specification may be given of the particular meeting of the pair. In this perspective, a third hypothesis may be formulated: underlying the linear and hierarchical structure of the sequence of sentences, there seems to be evidence for the presence of a more global level of semantic representation, which has been called the macro-structure of the discourse. A macro-structure is a theoretical construct, consisting of a hierarchical structure of propositions. Predicates or propositions from this macro-structure abstractly represent sets of sentences or propositions at the micro-level (sequence) level of the discourse. Thus a structure like (PAST) (MEET) [Mary, (CLEVER)(MAN)] may under-lie, intuitively, a long description of the meeting and of the man Mary met. Such macrostructures may be directly expressed in the discourse, e.g., in the first sentence of (20), where they may announce or resume the global meaning of the whole discourse. This is a first piece of linguistic (semantic) evidence. Secondly, a sentence in a discourse may have the necessary syntactic and semantic relations with previous sentences, but this constraints is not sufficient to define one globally coherent discourse. That is, each sentence might provide a further associative proliferation of any of the concepts in the preceding sentences. However, it is an empirical fact that in general such discourses are not acceptable. The linear expansion must be structured under macro-constraints. That is, each sentence must be functionally dependent on at least one macro-category, e.g., AGENT or ACTION, or PLACE, etc. Thus, each semantic representation of a sentence in a discourse is conceptually associated with a macro-concept. Again, this is a property of the discourse itself and not (merely) a fact of cognitive processing. Thirdly, (types of) discourses are acceptable only if their macro-structure satisfies a number of further constraints. Both

86

Acceptability in Context

theoretical and empirical research on narrative discourse, for example, have resulted in the establishment of a macro-syntax, for which terms such as 'introduction', 'complication', 'resolution' and 'conclusion' have been used.6 Similar categories were known already in ancient rhetoric. 5.2. Other arguments in favor of a macro-structural component in a discourse grammar may be provided. However, there are a number of difficulties which must not be overlooked and which require particular care with the formulation of such an hypothesis: 1. 2.

3.

Although perhaps empirical (cognitive) evidence for macrostructures can be assessed—see below—the arguments in favour of an account in the grammar are weak. Whatever the grammatical evidence for such structures, a serious grammar for macro-structures requires explicit categories, rules of derivation and interpretation, which have not yet been provided. Even if a formal language for their description would be provided we would not yet have the rules mapping macrostructures into sentential representations of the sentences of the discourse—and hence a relation with surface 'structures—which is essential for any grammar.

These and similar critical arguments against a macro-structure hypothesis have been mentioned in the literature,7 and they are justified. If a solution to these problems can be provided at all, it would at least take the following arguments into account: 1. 2. 3.

No strict distinction can be made between a justification for grammatical rules and categories, on the one hand, and cognitive processes and categories, on the other hand. A formal language for macro-structures has the same format as any adequate language accounting for 'meaning'structures. If a solution for argument (3) above can be found this would provide sufficient ground for having a macrostructure theory in the grammar.

Macro-structures 4.

87

Besides purely formal problems, the hypothesis requires much more empirical (descriptive) warrants: hardly any systematic and explicit discourse descriptions have been given in linguistic research.

Another problem upon which the hypothesis depends is the precise formulation of the aims and tasks of a (linguistic) semantics. If such a semantics would have to assign 'meaning-structures' to utterance-types, then there are no a priori reasons why such meanings would not also have the 'holistic' nature postulated in a macro-structure hypothesis. Along the usual Frege-Tarski line, both linguistic and formal semantics are required to construct an expression on the interpretation of its parts. It is not so easy to decide whether our hypothesis is inconsistent with this requirement: a macro-representation of a discourse is also based on the interpretation of its parts, viz. its sentences. Typically 'semantic' categories like Agent, Object, Event or Action, etc., do not necessarily dominate 'words' or concepts, but surely also propositions, as is already the case in classical transformational grammars where S-nodes are recursive. Similarly, those categories may dominate sets of propositions. Intuitively, as we mentioned earlier, the Agent of the discourse may be identified by a set of propositions, and the same holds for an Event. The formal problem is, however, that we do not simply need a set of sentences to identify, e.g., the Agent, but parts of the sentences ofthat set. That is, we must 'extract' from each sentence precisely the information relevant for the identification and construction of some macro-category. This may all be very plausible, e.g., in the perspective of cognitive processes, but it is pretty vague. What we need are the rules to do the operations described. 5.3. What is the moral of these general remarks for the grammaticalness-acceptability discussion? Clearly, what we want is that the grammar predicts that certain utterances are unacceptable because of a violation of macro-constraints. Let us give a concrete example to illustrate this point. Consider the following discourses:

88

Acceptability in Context (21) A: Did you hear about the bank robbery? B: No, what happened? A: Yesterday morning I was at our bank round the corner. Suddenly, one of the clients took a gun out of his pocket. He shot a couple of times in the air and then aimed at the cashier. He said that he wanted all the cash she had in her desk. She was very frightened and gave it to him. Then he ran away. The police have not yet caught him. (22) Suddenly, one of the clients took a gun out of his pocket. She was very frightened and gave it to him. The police have not yet caught him. Yesterday morning I was at our bank round the corner ( . . . ) (23) Yesterday morning I was at our bank round the corner. The bank is a terrible building. The buildings in this part of town are horrible. But I like living here. The town has no industry. My brother works in a factory. His boss is a terrible guy. He was born in New York in 1909. At that time you could still live in the city (...)

Intuitively, we find (21) an acceptable conversation (apart from some stylistic aspects and elements of spoken language, neglected here) and A's narrative acceptable. In (22), however, the order of the sentences of that narrative has been changed, which clearly makes the discourse not only ungrammatical at the semantic level, i.e., incoherent, but also unacceptable: the linear referential relations are mixed up. Discourse (23) does satisfy the constraints for linear coherence: each sentence has a semantic link with a previous sentence. Still, we would hardly consider it acceptable: there is no 'point', 'line' or 'theme' in it, and we would probably qualify it as pathological. In other words, there is no macromodel in which in each sentence can be interpreted: the relations between the (micro-)models in which the sentences are interpreted exist but are fully arbitrary. Discourse (23) would at most be a collection of the topics talked about during an informal conversation. In case the grammar would be able to formulate the macro-constraints, it would be able to predict the differences in

Discourse and Cognition

89

acceptability of (21) and (23). Notice, also, that macro-structures may have direct 'linguistic' relevance: in (21) A's question is about a bank robbery, and the narrative he gives in his reply is about the robbery. But the term 'robbery' itself does not appear in the narrative and yet B knows that the proposition of A's question and the meaning of the narrative have the same referent, viz. some event which may be characterized as (YESTERDAY) (ROB) [SOMEBODY, (OUR) (BANK)] . Hence, macro-structures may be directly expressed in surface structure. A macro-semantics should specify the specific relations between such a sentence and the global and specific meaning of the discourse. That such relations exist may also be concluded from the fact that a sentence like 'The bank was not robbed' would be inconsistent with the narrative discourse, even if it is not strictly inconsistent with any sentence of it. The inconsistency, then, can only be explained between the sentence and the underlying macroproposition of the discourse. In other terms, the whole ordered sequence of sentences of the discourse entails The bank was robbed'. Even if a macro-structural component in the grammar is superfluous, the semantics should provide the rules specifying why this sort of entailment may be true or false. Similarly, at the empirical level, the grammar should account for the fact that native speakers can make such conclusions, identify the equivalence relation in a conversation like (21), and detect inconsistencies between sentences and whole discourses. These are only a few of our semantic abilities. The conclusion from these arguments and examples seems to be that the phenomena involved are directly relevant to a theory of grammar and that therefore it seems warranted to look for a solution of the puzzles mentioned in section 5.2.

6.

DISCOURSE AND COGNITION

6.1. Since only some of the formal problems involved in pragmatically based discourse grammars can be solved at the moment, it is important to investigate the empirical foundations of such a grammar. More specifically, we want to know which cognitive

90

Acceptability in Context

structures and processes are involved in the production, interpretation, storage, retrieval, recall, recognition, etc., of such complex verbal utterances as complex sentences and discourses. Are these cognitive processes fundamentally different from those operating on simple sentences? Is it possible to make a distinction between perceptual, interpretative and productive strategies on the one hand and rule-based operations on the other hand? 6.2. Independently from work on discourse in linguistics, such and similar questions have recently attracted the attention of psycholinguists and cognitive psychologists. Since the early explorations by Bartlett (1932) on memorization of stories, a number of interesting experiments have been carried out on discourse material. Because no explicit grammar or cognitive model for the processing of such material was available, most of the experiments were based on intuitive analytic categories and procedures. Some of the major findings and assumptions of this work will be briefly given here. It is clear that such results can be considered as arguments in favour of a text grammar, in particular of the macro-structure hypothesis, only if it is agreed that an adequate grammar must provide the 'closest possible fit' to the systematic phenomena of verbal behavior. 6.3. It is well-known that early experimental psychology of language behavior very often made use of nonsense material to test learning, recall, etc. Together with a general reaction against behavioristic approaches to natural language, mainly initiated by Chomsky and Miller around 1960, the nonsense and word-list material was gradually replaced by sentences. In this research one of the major goals was to find psychological evidence for particular grammatical rules (transformations), e.g., by testing reaction times. The hypotheses involved there, however, turned out to be overly optimistic: there is no direct relation between particular rules and regularities in test behavior. Bever (1970), in his article referred to earlier, demonstrated that much of our verbal behavior is based on cognitive strategies which need not run parallel to the rules and their operations as provided in the grammar. At the same time the interest in learning and recall shifted to

Discourse and Cognition

91

semantics. It was pointed out that surface structure of sentences— or syntactic structure in general—is not learned and only fragmentarily recalled or recognized: information processing is based on underlying semantic structures. The memory models involved, hence, became 'semantic' or 'conceptual', constructed out of propositions (with their case structure represented in a tree graph, for example) or networks (for discourse representation, see, e.g., Kintsch 1974). 6.4. It is against this background .that recent research on discourse must be viewed. Even if discourse material was used, this was often not to test learning or recall for complex material as such or for the rules or constraints underlying it, but to get insight about 'contextual cues' in learning sentences. In other discourse experiments the tradition goes back to Bartlett's work. Having a story reproduced several times (often a very long time apart) after presentation, he found that the exact words, phrases and sentences were not recalled in the later trials, and that in the long run only some Outline' of the story, perhaps together with some striking detail, was recalled. New detail may reappear if consistent with, or inferable from, that outline. The main hypothesis, then, is that recall is essentially constructive, i.e., in perception and storage a basic structure, a schema, is formed upon which detail of the original input or new (added) detail can later be recalled. The core of our processing of complex structures like discourses is obviously such a schema. It was further found that elements in the discourse which could not easily be understood were transformed (by reduction or further explication) to more explicit or better-known structures (rationalization). Finally, it was found that the inference and rationalization procedures, the construction of schemata, and the recall or addition of detail were strongly dependent on personal interests, attitudes and feelings, and on social conventions. Similar conclusions were drawn from experiments with serial reproduction of discourses by various subjects after each other. Paul (1959) replicated some of Bartlett's experiments in a more sophisticated way and arrived at similar conclusions. In general, discourses which are relatively 'explicit' are better recalled than

92

Acceptability in Context

their Obscure' counterparts, where coherence must be established by the subject through inferential steps not presented in the text itself. However, this facilitation is more marked for unfamiliar material: we have no difficulty in supplying inferences in discourses about topics we know rather well. In repeated serial reproduction the more familiar and the more explicated texts remain coherent, even when 'skeletonized', which may not be the case for relatively difficult, obscure or unfamiliar material. One of the explanations is that schemata can be constructed more easily when related to schemata already present in memory. Other differences in task behavior are based on differences in cognitive style of the subjects: some subjects look for the global meaning (forgetting detail, and hence importing new detail in recall) whereas others are interested in specific details, being less able to construct a more generally coherent pattern. But, as was already demonstrated by Cofer (1941), recall of ('logical') ideas is always better than verbatim recall. Further detail in this direction was supplied by Gomulicki (1956), who showed that the recall of elements of a passage is directly proportional to 'its contribution to the total meaning of the passage'. Apparently, during interpretation, special mnemonic processes operate such that relatively 'important' concepts are abstracted from the material. Thus, in a narrative, Agent and Action are more important than descriptive elements. More generally, our recall for events and actions is better than that for states, scenes or object properties. Such phenomena may be explained—Gomulicki did not provide such explanations—by the fact that events and actions are better Organized', i.e., temporally and causally ordered, than, e.g., scenes. Similar results about the abstraction of important concepts from discourse material were obtained by Lee (1965), who showed in particular that the organization of paragraphs and the presence of summaries, conclusions and title facilitate this abstraction process. The importance of titles in the interpretation of discourses was also demonstrated by Dooling and Lachman (1971) on material which was intentionally vague and ambiguous. Earlier, Pompi and Lachman (1967) and Lachman and Dooling (1968) had postulated the construction of 'surrogate structures', which

Discourse and Cognition

93

need not directly depend on the words of the passage themselves, but which are a combination of theme, image, schema, abstract and summary. During encoding and reproduction such abstract categories serve as 'cores' around which individual concepts/words, phrases and sentences are interpreted or produced. The assumption that reading and (re-)production of discourse are not only based on properties of isolated sentences has been demonstrated also in several contributions to the volume edited by Carroll and Freedle (1972), e.g., by Crothers, who identified the 'theme' with higher order nodes, and by Frederiksen, who showed that problem solving is based on 'superordinate (semantic) processing'. One of the more concrete results in the investigation of the relations between grammar rules and memory processes was Slobin's (1968) finding that recall of passive sentences in discourse —and their storage as passives—depends on the degree of focus or importance of the (logical) subject of such sentences in the discourse: if that subject is unimportant or not specified at all, the sentences may be stored as (truncated) passives of which the subject is the 'logical subject' of the passage. This result is one of the examples where original coding hypotheses (complexity of grammatical rules entails complexity in storage and retrieval) had to be modified. Bower (1974), finally, made a distinction between microstructure and macro-structure of a discourse and showed that interference of interpolated details does not improve recall of macro-structures but causes confusion in the recall of microstructural elements. Macro-structures in this account are not welldefined and seem to be constructed of'major categories' representing key events in the discourse. 6.5. Although the survey of some selected papers on the cognitive processing of discourse given in the previous section is very fragmentary and oversimplified, the main results seem to be consistent with our hypotheses.8 However, a great number of problems have not yet had sufficient theoretical and empirical attention. For our discussion it remains to be demonstrated that the processes of abstraction (cf. Bransford and Franks 1972) and the formation of schemata or macro-structures are not merely cognitive strategies

94

Acceptability in Context

for the organization, storage and retrieval of complex information, but that such strategies presuppose the existence of grammatical rules. The inference processes involved seem to be based not only on inductive world knowledge but also on our knowledge of conceptual meaning structures, meaning rules and postulates of natural language. The structure of a macro-structure is thus determined by general semantic rules, identical with those for propositional structures: in interpreting a discourse we select the elements which may be the Agent, the Action(s), the Object or the Circumstances of the global meaning. Strategies and rules, just as for sentences, may not run parallel: the first Agent introduced in a story will as a hypothesis be taken as the discourse-Agent, but this strategy may be falsified by further information. Similarly, a conclusion or result of an argument or narrative may be given initially in the discourse, but the rules of macro-syntax/semantics will nevertheless assign the correct interpretation to such a discourse. Reordering of global segments of a discourse seems to follow the same rules as those for sentences. The interesting advantage of this hypothesis of the parallelism of structures and rules at the micro- and macro-levels is that we need only one set of rules and strategies to process both. The big problem remaining, however, is the formulation of the grammatical rules (and the cognitive processes underlying them) of macro-interpretation, taking concepts, concept clusters and propositions at the sequence level to concepts, concept clusters and propositions at the macrolevel by abstraction, reduction ('semantic pruning') and generalization. Although our—still not sufficiently explicit—hypothesis seems to have some 'holistic' properties characterizing much work in Gestalt theory, the underlying methodology is rather 'constructivistic', that is 'global' structures are derived from 'local' structures by explicit macro-rules (see e.g., van Dijk 1977a, c, and Reference given there to other work). Macro-structures are not mystical 'emergent' properties of a discourse but are constructed on the basis of semantic properties of words and sentences. If a formal model of this construction can be provided we would have an explicit warrant for our main assumptions in this paper, viz. that the factual acceptability of sentences depends on their function in

Notes

95

the discourse as a whole and that the acceptability of discourses is based on the presence of a macro-structure defining its 'unity' and on the pragmatic function in the conversation. At this point the hypotheses put forward have their direct bearing on social psychology and sociology, since they specify the cognitive foundations of our representation of reality and the organization of interaction. NOTES 1.

2.

3.

4. 5. 6. 7. 8.

For references to work in text grammar, see van Dijk (1972a, 1977a), Dressier & Schmidt (1973), Dressier, ed. (1977) and van Dijk & Petöfi, eds. (1977). For systematic descriptions of English discourse structures, see Werlich (1974) and Halliday & Hasan (1977). See e.g., Schank & Colby, eds. (1973), Schank & Abelson (1977), Bobrow & Collins, eds. (1975), and Norman & Rumelhart, eds. (1975) for work on the boundaries of linguistics, cognitive psychology and artificial intelligence, a domain now commonly called 'cognitive science'. See the work on action logic by von Wright (1967) and further elaboration by Brennenstuhl (1974). See also Nowakowska (1973), and Kummer (1975), and van Dijk (1977a) for further introduction and references to the theory analogic of action. See Rehbein (1977) for complex action analysis. We think of several directions of research in philosophy, linguistics, and the social sciences as mentioned in Chapter Two. For conditions on connection, see van Dijk (1974, 1977a, b), and references given there. For an action theoretical explication of this assumption, see van Djjk (1975c, and 1976a, b), where also further references on narrative can be found. A collection of reviews of our earlier work in text grammar has been collected by the Projektgruppe Textlinguistik (1974). For further detail on experimental work in this area, see van Dgk (1975a). See also more recent work, e.g., Kintsch & van Dijk (1975, 1978), and van Dijk & Kintsch (1978), in which many references to recent experimental work on discourse comprehension are also made, e.g., Just & Carpenter, eds. (1977), Meyer (1975), Thorndyke (1975). For an artificial intelligence approach to discourse (or story) comprehension, see Charniak (1972) and Schank & Abelson (1977).

CHAPTER FIVE

Issues in the Pragmatics of Discourse*

1.

SENTENCES AND SEQUENCES

1.1. One of the shortcomings of most approaches to establish a grammar of discourse is that a number of basic problems have been taken for granted which should have been solved before we could seriously speak of a (specific) 'text grammar'. One of these problems concerns the relations between (single, compound, complex) sentences on the one hand, and sequences of sentences on the other hand. There is an intuitive sense in which we take a discourse as a sequence of sentences. However, under what conditions a certain amount of information is expressed in one— possibly compound/complex—sentence or in several sentences has not been systematically studied. In part this problem could be ranged under what often, all too conveniently, has been called a 'theory of performance'. That is, a number of these constraints pertain to a theory of cognitive information processing and to stylistics or sociolinguistics. On the other hand, a number of systematic conditions are more properly 'linguistic'. More in particular, it will be argued below that they should be formulated in terms of a pragmatic theory of language. 1.2. The problem at issue can be demonstrated on some simple examples:

""Unpublished paper, written August 1975. Some parts of it have been summarized in vanDijk(1977a).

98

Issues in the Pragmatics of Discourse (1) John is ill. He won't come tonight. (2) Because John is ill, he won't come tonight. (3) John won't come tonight because he is ill.

In a rather intuitive sense, it may be said that these discourses express roughly the same 'conceptual content'. In other words, they are, at least from a certain point of view, semantically equivalent. Yet, discourse (1) expresses this semantic information in two sentences and discourses (2) and (3) contain only one complex sentence. Taking a functional view of language, we may then ask why this is the case: why do we use one sentence in one context and several sentences in another context in order to convey the same (semantic) information? Assuming for a moment the semantic equivalence of these discourses, it seems that we have to look at the pragmatic level in order to find grounds for a systematic differentiation as the one exemplified in (l)-(3). The answer to this query is by no means simple and can be given only in a theoretical framework in which a clear relationship is established with other problems: the distribution of information in a discourse, the relations between topic, focus and comment, the mappings of speech act sequences onto sentences and sentence sequences on the one hand and onto propositions and proposition sequences on the other hand, the relations between presupposition and assertion, and so on. For example, we may express a certain amount of information in one sentence, but these sentences, as in (2) and (3) may be structurally different in the sense that one of the clauses is subordinated to the other. Similarly, we may add a fourth example in which the two clauses are coordinated in one sentence: (4) John is ill, so he won't come tonight. which raises the problem in which respect coordination of clauses in one sentence is equivalent to 'coordination' of sentences in a sequence, which in turn touches upon the more general problem of the empirical basis for the theoretical unit of the sentence.

Sentences and Sequences

99

1.3. Although the relations between (1) and (4) should be made clear first, the differences between sentences and sequences appear more clearly in a comparison between (1) and (2)-(3). Let us take a context in the form of a previous piece of discourse, e.g. (5) ( . . . ) John has the flu ( . . . ) In such a context, (1) cannot be appropriately uttered as a following sequence, whereas (2) can be appropriate. The obvious reason is that (5) implies 'John is ill', a proposition which is asserted in the first sentence of (1), whereas the conditions for appropriate assertion require that we may only assert what is not yet known to the hearer. Sentence (2) may be appropriately uttered after (5) because 'John is ill' is not asserted but presupposed—at least in one reading—, occurring in typical subordinate, initial position. The same holds for (1) and (3) with respect to the proposition 'John won't come tonight'. In this case the main clause appears in initial position and the subordinate clause in final position accompanied by special intonation and stress on ///. Neglecting for a moment the differences between (2) and (3), it seems that one of the differences between sentences and sequences is that in a sentence a proposition may be uttered without being asserted if another proposition is asserted by uttering the sentence, whereas the utterance of that proposition in one single sentence usually implies the assertion of that proposition. The same argument more in general holds for any context such that hearer B knows that 'John is ill': in such a context (1) is inappropriate (for B) whereas (2) may be appropriate. In other words: (2) can be uttered in a context where 'John is ill' is presupposed whereas (1) can not. The hierarchical structure of a sentence, in which one of the propositions/clauses is subordinated, enables the speaker to assert a proposition in relation to a proposition which at that point of the conversation is already (assumed to be) known. Much more subtle, however, are the differences between (1) and (4), because in coordinate constructions presupposition cannot simply be expressed by clausal subordination. Taking a variant

100

Issues in the Pragmatics of Discourse

(Γ) John is ill. So he won't come tonight. and comparing it with (4) at first sight seems to yield pragmatic equivalence if we are unable to establish a set of contexts in which (Γ) is appropriate and (4) is unappropriate, and conversely. Notice, first of all, that if there is a systematic difference between (Γ) and (4) this should not only be based on arbitrary differences in graphical surface structure, but also on phonological differences, e.g. a longer pause after ///, a possible pause after so, and different sentence-terminating intonation in (1'). In other languages there are also syntactic differences, e.g. in German and Dutch: (6) (7) (8) (9) (10) (11) (12) (13)

Jan Jan Jan *Jan Jan Jan ?Jan ?Jan

was ziek. Dus, hij k warn niet. war krank. Also, er kam nicht. was ziek, dus hij k warn niet. war krank, also er kam nicht. was ziek, dus kwam hij niet. war krank, also kam er nicht. was ziek. Dus, kwam hij niet. war krank. Also, kam er nicht.

Subject-Verb ordering is possible in both compound sentences and sequences, but Verb-Subject ordering seems acceptable only in compound sentences. Verb-Subject ordering is typical in these languages after initial sentence adverbs. Cf.: (14) Jan was ziek. Daarom kwam hij niet. (15) Jan war krank. Deswegen kam er nicht. There also seem to be differences here between Dutch and German: (8) being acceptable in Dutch, but (9) not in German which requires V-S ordering after clausal also. Similarly, (12) and (13) seem acceptable to some speakers at least when there is no pause (comma) after the adverb. Although the differences between these examples are not very clear cut and although our intuitions are weak for such cases, we may try to attack the problem with some of the properties which seem less obscure.

Sen fences and Sequences

101

If we take (1) or (Γ) with heavy stress on he, the use of these sequences implies that others are present in the context of utterance. That is, we do not (merely) provide a description of John's actions but rather a description of the present situation. Whereas the first sentence of the sequence is clearly about John, the second somehow changes the perspective of the discourse, by drawing consequences of a certain state of affairs for the present context. In the non-contrastive case, with he unstressed, we also make reference to the expectations of hearers which are (or will be) present in tonight's situation. It seems that in such cases, and especially in the contrastive variant, the two sentences are less acceptable when combined in one compound sentence. In (4) the focus is rather on the relation between John's illness and his absence, viz. on the cause or reason expressed by so. We might say that here this reason is described rather from the point of view of John, whereas in (1) and (Γ) a conclusion is drawn by the speaker. Similarly, in the Dutch and German examples, where the past tense has been used ('John was ill, so he didn't come' vs 'John was ill. So he didn't come'), separate sentences in a sequence seem more appropriate if the focus of narration is on some meeting, and a compound sentence will be used if the focus remains on John's condition and the consequences for his actions. This difference is clearer in examples such as: (16) John felt ill, so he went to bed. Here the point of view is clearly John's, i.e. a sequence of events about him is related, and not about some situation which is worth telling about also independently of his presence in it. Since perspective differences can also be expressed by the verb (e.g. 'come' vs. 'go'), we would expect a corresponding difference between: (17) John felt ill, so he didn't go. and (18) John felt ill. So he didn't come, which would make (19) less acceptable: (19) ?John felt ill, so he didn't come.

102

Issues in the Pragmatics of Discourse

However, (19) does not seem very unacceptable, and the same holds for the Dutch and German examples (6)-(l 1). Moreover, we still should account for the possible differences between (8) and (10) and between (9) and (11), i.e. between the Subject-Verb and the Verb-Subject versions. In order to avoid differences of perspective and some other specific properties of the examples of (6)-(l 1), let us take another example: (20) (21)

De lucifers zijn nat, dus branden zij niet. De lucifers zijn nat, dus zij branden niet. [The matches are wet, so they don't burn]

Note that in the first example dus may be preceded by en ('and'), but not in the second example. Hence in (21) dus is a conjunction and in (20) it is rather a sentential adverb, as is the case in (14) and (15) and in English: (22)

The matches are wet. Therefore, they don't burn.

In (20), as well as in (14), (15) and (22), the focus seems to be on the cause or the causal connection (or the reason). In that case, the second clause or sentence may sometimes express a proposition which is presupposed or a likely consequence of the context. By stressing therefore we imply that the hearer believes that there is some other cause for the event denoted by the second sentence than the one assumed by the hearer. In (21) however, the focus seems to be on the consequence of the causal relation. Here, the consequence cannot be presupposed. This emphasis on the consequence is even more prominent in the following Dutch sentence: (23)

De lucifers zijn nat, dus die zullen wel niet branden. [The matches are wet, so those probably won't burn]

Whereas (21) is rather unacceptable in a situation where obviously the matches do not burn, (23) is perfectly OK, because something is predicted about a future event (which cannot be presupposed). It is not clear whether there is a device in English to express

Sentences and Sequences

103

differences like those in (20) and (21). Perhaps (20) corresponds with a specific stress on burn and (21) with heavier stress on they. Furthermore, the differences between this English sentence, viz. (23)

The matches are wet, so they don't burn

and (22) should further be made explicit for the case in which therefore does not have specific stress. Whereas inter-clausal so is primarily a causal connective, denoting a relation between facts, therefore and sentence initial so are what we may call inferential adverbs: they indicate that the sentence they introduce may be inferred (in a non-logical way, of course) from the premises expressed by the previous sentence(s). A similar distinction may be established between entailment as a connective and the process or act of deductive inference in which the conclusion is also entailed by the set of premises. Apparently, the sentence-sequence distinction is closely connected with the distinction and ordering of speech acts. A distinction between argument- or premise-giving assertion and concluding-assertion lies at hand. Connectives of the sequence forming type can of course also be used to make inferences about possible or probable causes or reasons of events, e.g. (24)

The matches don't burn. So they are [must be] wet.

However, it is not obvious whether the following sentences are acceptable: (25) (26)

The matches don't burn, so they are wet. The matches don't burn. Therefore they are wet.

Clearly, in the latter case we may not have a stressed therefore referring back to a mentioned cause or reason, because the preceding sentence does not express a cause or reason. Only an inferential therefore, followed by a pause, having a specific fallingrising intonation, and followed by stress on wet, would be acceptable here as a variant of inferential so. In such a case a comma after so and therefore is usual. However, is (25) different

104

Issues in the Pragmatics of Discourse

from (24)? If it is, it would mean that the speaker refers to two causally related facts, of which the cause is under focus. More in particular, it might be the case that (25) is used in a situation in which the first fact is discovered first (as a consequence) and that the second fact was discovered (or inferred) later. In that case we would not only have the possibility to map linear and causal orderings in events in a discourse, but also the ordering of our perception, discovery, etc. of these events. Note also that in subordinate (presupposed) clauses the use of connectives is more restricted: (27) (28)

?The matches are wet, because they do not burn. ?Because the matches do not burn, they are wet.

More natural would be the use of because in this inferential sense in a dialogue: (29) A: Why are the matches wet? [or: How do you know that the matches are wet?] B: Because they do not burn. Here, indeed, because does not refer to the connection between the two causally related facts, but indicates the relation between my discovery and knowledge of the respective facts. Similarly, a typical dialogue may serve to show the differences between various meanings of Dutch dus ('so'): (30)

A: Jan heeft geen zin [John doesn't feel in the mood] (i) B: (En) Dus komt hij niet? (ii) B: Dus hij komt niet? t So he doesn>t come? 1

In B's first reply-question, which may be preceded by en ('and'), B refers to the reason of a fact he knows (that John doesn't come) and questions the reason as being sufficient. In the second question B simply accept a certain fact (John's not feeling in the mood) but he wants to make sure whether this has the consequence mentioned by him. The first dus is a sentence adverb

Sentences and Sequences

105

denoting a cause or reason relation, whereas the second dus, followed by Subject-Verb order, is again an inferential pragmatic connective, relating premises and conclusions. This 'so' more directly relates a certain fact, or knowledge of that fact, to the present conversation or to interests of speaker and hearer. 1.4. Similar remarks, although provisional and still far from explicit, may be made about other types of connection and their corresponding connectives. For conjunctions we have the usual differences between interclausal and, sentence initial And, interclausal';' and intersentential period: (31) (32) (33) (34)

We were in We were in We were in We were in

New York. We visited uncle Tobias. New York. And (,) we visited uncle Tobias. New York; we visited uncle Tobias. New York and (we) visited uncle Tobias.

The classical semantics for conjunctive and is of course that both conjuncts be true in order for the compound sentence to be true. If we could speak of truth of sequences, the same would hold for (31), (32) and (33). As such, the connective does not seem to indicate a further relation between the conjuncts, but natural language connectives require that the propositions connected be 'relevant' to each other.1 This notion of relevance or connectedness should be defined at the object level, viz. in terms of the relations between the facts denoted by the propositions. In general this connectedness between facts can be defined as a relation of conditionality, varying in strictness for the different types of connectives (probability, physical, psychological, logical necessity): proposition A is connected with B, if the fact V(A) denoted by A (i.e. the value of A), is a condition for V(ß), i.e. if B is true or satisfied in some (most, all) possible worlds 'selected by' A. Further details of this formal semantics of intensional connectives are omitted here. It may be asked, however, in which respect conjunctions also can be defined in these terms. For a sentence like (34) we might say, that at least under one reading (implying that uncle Tobias was in New York during our visit) the

106

Issues in the Pragmatics of Discourse

first conjunct determines the possible world in which the second conjunct is true: i.e. the proposition 'We visited uncle Tobias' is true only in those worlds which are selected by the proposition 'We were in New York'. In other words: our presence in New York and our visit of uncle Tobias are in a weak sense conditionally connected. This semantic relation holds also for the other sentences and sequences, whether the relation is expressed by and or by period or ';'. Under another reading the two facts may as such have no relation at all with each other, e.g. when uncle Tobias lives in Boston. In that case, we would merely have the classical (extensional) truth conditions (both propositions true). Yet, one of the principles of natural language conversation is that in normal cases the sentences of one discourse be somehow relevant to each other.2 If the conjuncts of (31)-(34) do not have this relation directly, we should assume that they are indirectly related. Indeed, in that case (31X34) seem appropriate only with respect to a set of other propositions entailed by previous text or context, e.g. after a question like: (32)

What did you do this summer?

Here the connection is established for each conjunct relative to the proposition 'We did something this summer', defining the possible world (time and activity) in which they can be satisfied. If we would prefer to maintain a 'conditional' reading for this and (which often is commutative), we would have to read it as 'in the same situation'. A more sober approach would however stick to the 'minimal' properties of classical conjunction, thereby attributing the conditional aspects to the order of the clauses or sentences and to their semantic content.3 In that case, 'We visited uncle Tobias' would e.g. be short for 'We visited uncle Tobias in New York', where 'in New York' as local adverb can be deleted under identity. The same would hold for temporal indicators. The second (indirectly conditional) reading would become necessary as soon as local differentiation excludes direct relatedness of facts: (35) We were in New York and we visited uncle Tobias in Boston.

Sentences and Sequences

107

This sentence seems acceptable only, however, in e.g. a narrative about our activities of a certain period. Sentence (35) would be derived from the underlying sequence: (36) This summer we were in New York. This summer we visited uncle Tobias in Boston. The surface structure and then either expresses indirect condition or expresses a sequence or listing of facts. We will assume, however that listing in principles is expressed by asyndetic sequences, whereas and, like all other connectives in natural language is directly or indirectly conditional. We might say that if and in (35) does not express indirect condition, its 'listy' meaning is close to the and relating noun phrases, relating members of a group or set. There is perhaps another argument why a 'conditional' reading of natural language and has advantages above a mere 'listing' reading. Whereas all other connectives have coordinative and subordinative forms, and. at first sight does not seem to have a subordinative correlate. However, the following sentence seems to have the same meaning, apart from presuppositional differences: (37) While we were in New York, we visited uncle Tobias. The adverb here seems to have the same relation to and as because has to .so in: (38) (39)

We were in New York, so we visited uncle Tobias. Because we were in New York, we visited uncle Tobias.

In both cases the connective expresses a condition: the while/and a very weak condition (compatibility), the so/because a stronger condition (likeliness). Now, what are the further differences between (31)-(34), given this semantic base? The use of and in the compound sentence (34) expresses, as we assumed above, the direct or indirect relatedness of the facts denoted by the conjuncts. The foucs is in that case directly on the sequence of actions reported. Sequence

108

Issues in the Pragmatics of Discourse

(31) would be inappropriate in such a case,, but would rather be used in a situation in which the speaker provides a further specification or an explanation after assertion of the first sentence, e.g. after a question like "Where were you yesterday?", and thus preemptying a possible further question like "Why?" or "What did you do there?". The sequential structure is also natural in those cases where a change of scene or perspective is expressed: (40) We safely arrived in New York. The next day we visited uncle Tobias. Here the perspective is changed from the travel to and arrival in New York on one day, and the activities in New York on another day. Still less clear are the systematic conditions for the use of sentence initial And, usually followed by a pause (expressed by a comma). One of the more or less 'rhetorical' uses is when the speaker wants to create a certain suspense in the hearer, thus announcing an assertion which is somehow unexpected. Another condition is based on addition, which has a stronger variant in the connective moreover. So, just as in (34) initial And expresses relatedness of facts, but in a different pragmatic context: it focuses attention on the second fact, either by addition or by asserting an unexpected proposition. It is not clear in which respect the semi-colon is equivalent with the period, and whether a possible difference corresponds to a phonological difference. At least syntactically the semi-colon requires an independent clause (whereas a comma either separates non-independent clauses, where e.g. certain deletions are possible, or separates clauses in an enumeration). We will assume provisionally that a semi-colon is used in those cases, e.g. change of perspective or focus, where a proposition must be independently asserted, but where at the same time a connection between the facts must be asserted. The linguistic conventions for the use of semi-colons, however, do not seem well established, and perhaps limited to certain writing styles. 1.5. Let us finally illustrate our provisional assumptions with

Sentences and Sequences

109

respect to sentence vs. sequence differences on contrastive and concessive connections (we omit a separate discussion on disjunction, which seems to have similar properties in this respect as conjunctions; cf. Chapter 8): (41) We slept too late, but we were still in time. (42) Although we slept too late, we were still in time. (43) We slept too late. Yet, we were still in time. (44) We slept too late. We were still in time though. These are only some of the possible forms of expressing contrast or concession. Just as we had initial So and And, we may here have initial But, also followed by a pause (c.q. a comma). The distribution of forms is the same as for the other connectives (causals, conjunction): coordinate and subordinate inter-clausal and intersentential, respectively. The subordinate clauses, as usual, may be used to express a presupposed proposition (or a proposition with 'lower' focus; see below). As we remarked above, the use of one sentence, by its possible hierarchical structure, enables this immediate expression of presupposed information, but we will not pay particular attention to the differences between coordinated and subordinated constructions within the sentence. The semantics for these connectives is roughly the following: p but q is true if both p and q are true and -q is a probable consequence of p. The notion of 'probable consequence' will not further be discussed here: we will simply define it as follows: 'q is a probable consequence ofp' or 'p probabilizes q'' = d e f # is true in most p-worlds. In a concessive relation we thus express the fact that two facts are the case, but that the second fact is exceptional in circumstances defined by the first fact. In our example: sleeping too late usually implies that we do not come in time: (45)

We slept too late, so we were not in time.

We see that in the interpretation of but a normal causal (or reason) connective is 'presupposed'. The probable consequence relation involved may be of various types: physical probability but also

110

Issues in the Pragmatics of Discourse

psychological probability (expectation): (46) It was January, but very warm. (47) We went to New York, but did not visit uncle Tobias. There is another type of but with different truth conditions, viz. not based on probable consequence but on probable presupposition or precondition: (48)

We want to go to the movies, but we have no money.

Here, having money is the normal precondition for going to the movies, and the but expresses the fact that this precondition of the fact expressed in the first conjunct does not exist. Finally, there is a weak but with similar truth conditions, but instead of expected q we do not have ~q, but r. This is the simple contrastive but, also denoting counter-expectation, but not based on a causal or rational probable consequence or precondition: (49)

He wanted to catch a pike, but he caught a shoe.

From these truth definitions we already see that but has clear pragmatic aspects, viz. the condition that the consequent is unexpected, which may be based on unexpectedness with respect to the speaker, with respect to the hearer, or with respect to both. Since, however, this unexpectedness need not be related to the particular speech context, but has a more general character (knowledge of the world) we may still speak of a proper semantic characterization. What, then, are the differences between (41) through (44)? First of all, some syntactic differences should be noticed. Concessive although may only be used for unexpected-consequence but in subordinate position, not for unfulfilled-condition or for contrastive relations: (50) *Although we want to go to the movies, we have no money. (51) * Although he wanted to catch a pike, he caught a shoe.

Sentences and Sequences

111

The same holds for the use of yet in a following sentence: (52) *We want to go to the movies. Yet, we have no money. (53) We want to go to the movies. But, we have no money. (54) *He wanted to catch a pike. Yet he caught a shoe. We see that initial But is grammatical in such a context, where yet is not. Apparently, the semantic equivalence of but and yet only holds for the unexpected-consequence case. Another characteristic feature of contrastives and concessives is that they are not usually expressed by simple juxtaposition (i.e. by pause/period): (55) *We slept too late. We were still in time. The omission of but seems more acceptable in the contrastive uses of but, although in that case a special intonation is necessary on the following sentence, e.g. on the contrastive predicates. But may be followed by the concessive adverbs yet or nevertheless, e.g.: (56)

We slept too late, but nevertheless we were in time.

At first sight this seems merely to strengthen the unexpected consequence. However, there are cases where we may have but, and not yet or nevertheless, and conversely, and not only in the simple contrastives as was mentioned above. Compare the following sentences: (57) ?He cannot fish, but he caught a pike. (58) He cannot fish, and yet he caught a pike. (59) ?The glass was very thick, but it broke. (60) The glass was very thick, and yet it broke. Conversely, the use of yet or nevertheless is less acceptable in cases like: (61)

The glass is thick, but it is very fragile.

112 (62)

Issues in the Pragmatics of Discourse The glass is thick, and yet it is very fragile.

Notice also that the negation in sentences like (59) makes them more acceptable: (63)

The glass was very fragile, but it didn't break.

What are the underlying conditions which determine these (often very slight) differences? First of all, it seems that the use of yet and nevertheless is somehow linked with affirmative (positive) assertion: whereas they seem required in-(59) they are less so in (63). Apparently, if the underlying implication has a negated consequence, the occurrence of a (positive) fact can be thus expressed by yet or nevertheless. Secondly, the use of yet and nevertheless is more acceptable in sequences of narrated events (past tense) than in sequences stating a more general state of affairs. When I just speak about the quality of glass, I use but to indicate an apparent contradiction between properties, but when I tell a story about certain events which were unexpected under the conditions specified in the antecedent, I will use but or but nevertheless. In the first case we rather have a contrast between Objective' properties, whereas in the second case, the contrast is 'subjective' (unexpectedness of an event). Third, yet or nevertheless will have a still more pragmatic use in argumentation. In the following dialogue, the use of simple but in B's reply does not seem to be acceptable: (64) A: This glass is really very thin. B: And yet (but nevertheless) it is unbreakable. But what are the differences between but (or but nevertheless) in compound sentences like (41) and Yet, Nevertheless in a following sentence of a sequence? Above we have noticed that in certain readings yet (nevertheless) is not admitted or at least less acceptable. In those cases, the use of complex sentences would be preferred in order to express a contrast or an unexpected consequence. A first difference is that in compound sentences, the antecedent may sometimes have a more or less presupposed

Sentences and Sequences

113

character or topic function, such that the main assertion is focused on the consequent. Of course, in this type of connective, based on unexpected consequences, the focus is normally on the consequence anyway. In compound sentences two propositions are asserted denoting an irregular course of events. In a sequence an assertion is made which in the hearer creates expectations about the probable consequences. The speaker, therefore, will continue with an assertion 'correcting' these expectations in the hearer. In other words, as was the case for the other connectives: the interclausal connective and the compound sentence focus on the course of events and the relations between the events, the sequence rather expresses the sequence of speech acts (e.g. assertions of various types) about these (courses of) events. Following sentences, then, may give specifications, draw conclusions, confirm or disconfirm expectations. The sentence-sequence distinction, indeed, seems to have an important pragmatic function, where the sequence of sentences is closely related to the sequence of speech acts. Below we will have to mention other functions of this distinction, and make the functions assumed above more systematic and explicit. 1.6. Before we discuss another set of constraints determining the expression of a sequence of propositions in complex sentences or in sequences of sentences, let us resume some of the observations made above. 1.

At a certain level of description it may be assumed that complex/compound sentences and sequences are equivalent; in other words, there is a level of representation from which both may be derived, viz. the level of semantic representation.

2.

Differences between subordinate and coordinate sentence structures are based on, among others, principles of semantic information distribution, e.g. on presupposition and assertion.

3.

The use of sentences instead of sequences (or conversely)

114

Issues in the Pragmatics of Discourse is based on a number of constraints, e.g. on the implications of the structural properties of complex sentences as mentioned in 2.

4.

Connectives in natural language (grammatical conjunctions, adverbs, other particles) are to be interpreted as conditional relations between the facts denoted by the connected propositions.

5.

Connectives usually have three types of surface forms, viz. subordinate interclausal, coordinate interclausal and intersentential. They do not only differ lexically, but there are also graphical and phonological (stress, intonation, pause, etc.) differences related with them. The very existence of specific intersentential connectives (and their corresponding phonological, graphical and syntactical differences) seems to suggest that sequences cannot fully be described in terms of compound/complex sentence structure at at least one level of description.

6.

It will be assumed that the differences between sentential and sequential expression of semantic representation are essentially pragmatic. The pragmatic constraints operating here are the following: (i)

the distribution of presupposition and assertion (see 2. and 3. above);

(ii) ordering and type of speech acts: in a compound sentence an assertion is made of two propositions and of the connection between them—directly relating to facts and relations between facts; in a sequence the assertion of the connection and the second proposition may be made relative to the assertion made by uttering the previous sentence, e.g. as a conclusion, correction, specification, addition, etc. Here not only the facts and their relations are represented, but (also) the way this relation is interpreted by the speaker,

Sentences and Sequences

115

and presented and evaluated in the context; (iii) focus and perspective: by expressing a proposition in a separately asserted sentence, the proposition is assigned special focus. In most cases this focus is assigned to the consequent expressing consequences or results of facts previously referred to. If previously mentioned causes/reasons are focused upon, special expressions (stress and intonation, e.g. stressed sentential therefore) are used. This distinction seems parallel to the topic-comment and presuppositionassertion distinction in the sentence itself. Here however, preceding sentences are 'presupposed' in the form of premises with respect to which specific 'following' (e.g. concluding) speech acts can be made. The change to a new sentence, and a new assertion, enables the speaker to change the perspective of the discourse, e.g. from the points of view of the events or persons referred to, to his own view or comment about these, for example in order to pre-empty possible questions or comments of the hearer; (iv) change of scene and topic: this condition could still be called 'semantic' because it merely reflects properties of semantic representation and reference; by starting a new sentence in a sequence, a speaker may change the possible world (e.g. place or time) or the persons or objects involved in a course of events referred to; (v) introduction of new referents: a new sentence may also be needed in those cases where a new referent must be introduced which has no direct relation with referents (things or properties) referred to in the previous sentence; (vi) change of speaker in dialogue and conversation: this nearly trivial condition, however, is important in

116

Issues in the Pragmatics of Discourse order to understand sequencing of speech acts, e.g. in argumentation, question answering, etc., also in monologue sequencing. Clearly, previous discourse of the previous speaker may have the same presupposition and premise functions as previous discourse of the same speaker; notably, it defines discourse referents, topics of conversation, etc. with respect to which focus, perspective and speech act types of the (next) speaker are to be determined. These conditions determining the differentiation of sentences and sequences are neither complete nor theoretically explicit. It is not yet clear how speech act sequences relate to sentence sequences, and how notions such as 'focus', 'perspective' and 'topic of conversation', should be defined, and how they are related to notions such as 'presupposition' or 'topic/comment'.

2.

FURTHER PRAGMATIC CONSTRAINTS ON SEQUENCES

2.1. In the previous section we have given examples of propositional information which could be expressed either in compound/ complex sentences and/or in sequences, and we have concluded that sequential structure is mainly determined by pragmatic constraints, e.g. those relating to sequencing of speech acts. There are a number of cases, however, where a sequence cannot possibly be expressed in one sentence. Although the examples given earlier were all of the assertive type, we saw that one ground for expressing information in a sequence was the introduction of various kinds of assertion, viz. explication, explanation, conclusion, etc. Similarly, sequence structure is necessary in order to express sequences of various speech acts: (65) (66) (67)

It is cold in here. Could you please shut the window? *Could you please shut the window because it is cold in here. I wouldn't go to Italy at the moment. The weather is horrible there.

Further Pragmatic Constraints on Sequences (68) (69) (70)

117

Shut up! I am busy! *Because I am busy, shut up! What is the time? I have no watch.

We see that in these discourses we have expressed sequences of assertions, questions, commands, advices, etc. In most cases these sequences cannot simply be expressed in one sentence and connected with the (semantic) connectives discussed above. One of the characteristic features of the examples in (65) through (70) is that in all cases one of the speech acts is an assertion. More in particular, it appears that in sequences of speech acts of this kind the assertion in general provides a condition for the other speech act to be appropriate. In (65) the assertion in the first sentence specifies the motivation for the request expressed by the second sentence. One of the conditions for requests is that the hearer knows whether the speaker has a plausible 'ground' or 'interest' for his request. The reasoning relating the two speech acts is fairly complex, but will not be made explicit here: it is cold—»I feel cold—»I want not to be cold — ( . . . ) etc. In (65) we may use sentence initial So. However, as we saw earlier, this connective does not relate the facts referred to (cold - your shutting the window), but indicates the relation between the motivation and the request. We therefore call it a pragmatic connective: it does not connect propositions but speech acts or conditions and consequences of speech acts (see Chapter 9). A similar 'ground' is specified by the assertion expressed by the second sentence of (67): an advice is appropriate only if it is clear from the context or the text what the causes and reasons are why the hearer should engage in an alternative course of action as the one intended. The same remarks can be made about the command in (68) and the question in (70): as for requests, commands (at least of the 'non-authority' type) require grounding, whereas questions are appropriate only if it is clear or made clear that the speaker cannot provide himself an answer to his question.

118

Issues in the Pragmatics of Discourse

From these few examples, we may conclude that sequences may serve as expressions for speech act sequences or for parts of one speech act, e.g. the conditions and the act itself. Note, that in the examples given we strictly speaking have one speech act, or at least a complex speech act in which one speech act dominates, viz. a request, an advice, a command and a question, respectively. What is at issue is not the cold, the weather in Italy, or my not having a watch, but an action to be accomplished by the hearer. 2.2. We have assumed above, provisionally, that a sequence of different speech acts must be expressed in a sequence of different sentences. There are, however, examples where different speech acts seem to be expressed in one, compound or complex sentence: (71) (72) (73) (74)

I am in a hurry, but tell me your story. Although I'll be in a hurry, I'll come and see you. Please shut the door or turn the heater on. Because John might become angry, you better don't stop him.

A certain number of the remarks made above about speech acts and the expression of some condition for their appropriateness also hold for these examples. The sentence form, then, may be needed for additional reasons, e.g. the expression of presupposed conditions for speech acts, as is the case in (72). Furthermore, a sentence may be used to express two speech acts of the same type, e.g. two requests or commands, as in (73). It may be asked in such cases, however, whether the disjunction is a disjunction of speech acts (whatever that may mean) or a disjunction of the prepositional content of one speech act. Note that the following disjunction seems unacceptable: (75)

Please shut the door or please turn the heater on.

The use of please apparently has one request as its scope. Finally, although it may be said that the first clause of (74) expresses an assertion which is the ground for the appropriate advice expressed by the second clause, there is at the same time a connection

Further Pragmatic Constraints on Sequences

\ 19

between the facts (viz. actions) referred to: the first action may be a sufficient reason for the second action, as is the case in: (76)

Because John became angry, I didn't stop him.

The problem is why (74) is acceptable, whereas (69) is not. Since (74) becomes less acceptable when we change the second clause into a command, and because (69) becomes more acceptable when we change the second clause into an advice, the difference may lie in the fact that we may only connect speech acts in one sentence when the grounds given in the assertion for the other speech act to be appropriate are at the same time conditions for the action of the hearer. Now, in a command there is no condition pertaining to conditions of actions which are commanded, only reasons why I give my command. In advices the conditions for successful action of the hearer are at the same time criteria for the appropriateness of the advice when the speaker has epistemic access to such conditions and the hearer-agent has not, and/or if the speaker considers these conditions to be sufficient for the action to which his advice pertains. This seems to account for the fact that compound commands, requests and questions are possible if their conditions are specified, e.g. in conditional commands, etc.: (77)

Give me the book, if you can reach it!

Here, reaching the book is not a condition for an appropriate command, but a condition for the execution of the action commanded. In other words: the command is to be satisfied only in a world defined by the conditional clause. The problem whether the assertion (or other speech act) expressed by the utterance of compound or complex sentences is one (compound or complex) action or a sequence of actions will be discussed in a broader action theoretical framework below.

120 3.

Issues in the Pragmatics of Discourse ACTS AND SPEECH ACTS

3.1. From the discussion above it has been concluded that speech act sequencing is one major constraint on the structure of sentence sequences. At the same time, however, a certain number of basic notions remained unclear, and additional problems have been created. These can only be adequately treated in a more general framework of a theory of acts and speech acts. Resuming a long philosophical and logical discussion, we will briefly define actions as binary change operations on possible worlds brought about through an intentional and controlled change of the bodily state of a conscious person. This bodily change will be called a doing. Doings are extensional objects. Actions, however, are interpretations of doings, i.e. intensional objects. The same doing may be assigned various actions according to the intended change of the world. We will neglect all 'negative' actions, i.e. omissions, forbearances, lettings, etc. Actions, thus, have three phases: the mental conditions (wishes, desires, decisions, purposes, intentions—of which the internal structure is still pretty obscure, and irrelevant for our discussion), the actual doing, and the consequences of the doing (purposed or not): state, event or process. Important for our discussion is the fact that we may further speak of compound and complex actions, as well as of sequences of actions. Most actions we perform are compound/complex, although intended and perceived/interpreted as one action (e.g. opening a door, writing a letter). As extensional objects, doings are linearly ordered in time, and as such form a sequence. In what respect, then, could we speak of actions as being compound or complex, and how do we distinguish such composite actions from sequences of actions, and why is such a distinction useful? The criterion for one (simple or composite) action may be sought at different levels. We may use the unit of intention, or the unit of interpretation (by an observer), which is the basis for intersubjective and conventional interaction. Thus, smoking a pipe, as an action, consists of other actions, e.g. lighting a match. Lighting a match, however, is an action which may also be part of other actions, like making a fire. The identification of an action (within

Acts and Speech Acts

121

a sequence) is thus based on the possible function of that action in another sequence. Now, although lighting a match is based on a sequence of doings (like moving my arm, scratching the match, etc.), we still consider it as a simple action, because its constituents do not, as conventionally 'recognized' actions, take part in other actions. Of course, we may intentionally and consciously move our arm, even in an 'as-if-match-striking' way, without actually striking a match, or as part of another action. Yet, this action is no longer conventionally identified because it is not associated with a necessary purpose. In the narrow sense, moving my arm is, by definition an action if I only intend to move my arm and accomplish the doing in some way. In the broader sense, however, it is not an action (but, say, 'merely' an intended doing) because it lacks a purpose: I do not satisfy a wish or need, nor try to change the world accordingly. Whereas in striking a match the purpose is to light the match which has as purposed consequence the lighting of my pipe, moving my arm has no other purpose as its own accomplishment. Although there are a great number of theoretical difficulties involved we will call such intentional and identifiable doings: basic actions. To use a metaphor which seems inevitable anyway in this context: a basic action is like a phoneme: we may utter and identify it (as an ideal type) but it has no meaning; but, together with other basic actions (phonemes), it forms a simple action (a word) which as such as a conventional meaning. Notice, that the identification and naming of actions depends not only on cognitive structures (perception, planning, etc.) but also on cultural and contextual structures: what for a strange observer is merely a doing may for others be a significant action. The same holds for words or morphemes. Now, striking or lighting a match may be an identifiable and conventional simple action, but such actions are normally part of composite actions; in our example: I light a match in order to light my pipe, in order to smoke my pipe. That is, although I intend to light a match this intention is part of the higher order intention to smoke a pipe, viz. as part of the linear execution of the composite action. Smoking a pipe, in this case, is a (composite) unit action because it is independent and not an action which is again part of a higher order action: I

122

Issues in the Pragmatics of Discourse

may smoke a pipe, just in order to satisfy my needs. Striking/ lighting a match, however, conventionally has another status. It is typically an auxiliary action, e.g. embedded in a broader purpose (lighting a pipe, a stove, a fire, etc.). An auxiliary action is an action of which the result is intended only as a sufficient and/or necessary condition for the successful accomplishment of another action. The smoking of the pipe itself, however, also consists of a series of actions, beginning with lighting the pipe, drawing smoke, etc. Unlike lighting a match (which could be auxiliary in other composite actions) these actions are essential parts of smoking a pipe; without them I would not smoke a pipe at all, or the action of smoking a pipe would not be successful. Now, we will call an action composite if it consists of several simple actions as defined above. An action is complex it it contains an auxiliary action which is a condition for the (or a) main action, and an action is compound if it consists of a sequence of actions which together define one action and hence are essential components of such an action. The actions in a compound action may also be conditions for following actions but are necessarily so. Unlike auxiliary actions they are not substitutable by other actions. Thus, mixing concrete is an auxiliary action for one of the essential actions of the action of building a house. It is, however, not a necessary action, because I may build a house of wood, and I may be mixing concrete in order to build a floor or a street. Building a wall, then, is already building the house because it is (roughly) speaking an essential component of building a house. The next question is how we distinguish between composite (and especially compound) actions on the one hand, and sequences of actions on the other hand. A composite action, as we saw, functions conventionally as a unit, it is intended and executed as such, and perceived, interpreted and named as such. In our everyday (inter-)action, however, these simple or composite actions are usually followed by states or events which call for further actions, either by the same agent or by other agents (reactions). When I come home (a t ) I may kiss my wife (a 2 ), drink a beer (a 3 ) and smoke a pipe (a4). Each of these actions is one (simple or composite) action, but together they form a temporal sequence, which may be either new or more or less habitual. Still, none of

Acts and Speech Acts

123

the actions is an essential part of (a necessary condition for) the whole sequence: I may not kiss my wife, or kiss my wife elsewhere, and smoke a pipe but not take a beer first, etc. As a sequence it is a unit with respect to possible activity in a certain period, and a description of the sequence may be given as an answer to the question: "What did you do before dinner?" Theoretically, the sequence has no definite length, and its description may coincide with a full biography or historiography of a person or group of persons. Yet, we do often distinguish such sequences—at several levels—: we talk about a particular war or revolution as highly complex action sequences of groups or nations, or of the attack of a bridge by a group of persons during that war or revolution, also as a sequence of actions, although in the latter case, given the unique purpose (destruction of the bridge) we may again speak of one complex action. We see that in such cases the identification of composite units vs. the identification of sequences depends on the point of view of the observer or describer, and on the purposes of the observation or the description. In many cases of fairly complex actions we do not notice or describe those actions of agents which are irrelevant for the 'main action', e.g. taking a beer of somebody who is co-agent in destroying a bridge in a war. An action may be called relevant if it is a sufficient or necessary condition for a (component, essential action of a) main action. In our example, then, we may call the destruction of the bridge one complex action, given a clearly identifiable purpose with respect to a change in a state of affairs of which the component actions are conditionally related with each other. The war in question would be a sequence of actions, identifiable as one sequence, i.e. with a beginning and an end and with an identifiable purpose of the fighting parties involved, but composed of actions by different agents and groups which as such need not be conditionally related. That is, the destruction of a bridge and taking some town may both be sufficient conditions for successfully conducting a war, but these complex actions as such need not be related conditionally. In other cases, e.g. in the action of governing a (capitalist) country, there is no clear final purpose beyond a permanent, process-like, maintaining or ameliorating a certain political and socio-economic

124

Issues in the Pragmatics of Discourse

state of affairs. It seems warranted, though, to consider actions and action sequences as having a hierarchical structure. What at one level of purpose, interpretation or description is a sequence of actions, may at another level be a more global conditional action. In that case we will speak of macro-actions: they consist of sequences of actions on a level 1. and function as one action at level li+l· A sequence of actions will be called globally coherent if they can be assigned to a macro-action, i.e. if they have an overall-plan. A sequence is linearly coherent if each action is a necessary or sufficient condition for at least one following action in the sequence. 3.2. Although this discussion of the structure of actions, composite actions and sequences of actions is not yet fully explicit and leaves open a number of further problems, we are now in a position to discuss speech acts and sequences of speech acts. According to definition, a speech act is an act if there is a conscious person who by intending and executing a certain doing has the intention of somehow bringing about some change (or preventing such a change). A speaker seems to satisfy these requirements: when we speak intentionally we are normally conscious, execute a doing (writing, moving our speech organs), and thereby directly change the physical environment (air, paper) and indirectly the perceptory organs of some reader or hearer. I may thus speak to myself, without wanting anyone to hear me, or if it is my intention that someone hears, then may be I do not wish them to understand, or even if I do wish someone to both hear and understand what I say, then it may be my intention that he should not assign any further meaning to my words, e.g. in terms of desired mental or overt action or reaction on his, the hearer's, part. In other words, speech acts are fairly complex actions, having a hierarchical structure. The doing involved is not more than a certain movement of mouth or hand resulting in a certain linear, transient or permanent change of physical state. Only under a number of abstracting and conventional conditions are we able to interpret this doing as a phonological or graphical act, viz. as speaking or writing. Following our earlier definition, this act is itself mostly compound. We may pronounce intentionally one

Acts and Speech Acts

125

(token of a) phoneme, but when we are speaking this intention is fully automatized with respect to the intention to pronounce a sequence of phonemes having a certain function and/or meaning (a morpheme). As such, a morphematic act has a conventional function; by accomplishing it we at the same time may accomplish a semantic (lexical) act. The metaphor used above now works the other way round: we accomplish morphematic acts and thereby lexical acts only with a broader purpose in mind, viz. the accomplishment of a syntactic-semantic or propositional act. This act is in turn the basis for a referential act which is itself composite because individual lexico-syntactic acts may themself be the basis for a referential act. Hence, the utterance of phonemes are basic actions, the utterance of morphemes simple actions, the utterance of a sentence composite actions. These composite utterance actions are the basis for corresponding meaning (or intensional) actions which are the basis for composite referring actions. I may accomplish a meaning action without accomplishing a referring action, e.g. when I just utter a meaningful sentence of a natural language without the intention to refer to things, properties or facts (truth values in some possible world). We may take actions of this type as functions: taking phonemes/morphemes as arguments and having lexical meanings as values (interpretations), or more correctly as composite functions because these meanings are again arguments for referring-act functions. We have here characterized in somewhat more detail what has traditionally be called a locutionary and a propositional act. The purpose of these higher order and compound actions is that the hearer understands what I mean and (thereby) knows what I am referring to. These actions are successful if this understanding of the hearer (i.e. a specific change in his knowledge set) is identical with the final state as purposed and if this final state is a consequent of the locutionary action of the speaker. The propositional act, however, may itself again be composite, by expressing it as a composite sentence, and thereby not only referring to facts but also to a certain relation (e.g. of cause and consequence) between these facts. Propositional and referring acts, finally, are again normally arguments for other conventional act functions, viz. illocutionary

126

Issues in the Pragmatics of Discourse

acts. By uttering a sentence, thereby accomplishing a prepositional act, and thereby accomplishing a referential act, I do not only have the intention that the hearer should understand what I say, but also why I say it, i.e. I want my speech to have some specific function: by understanding what I say-I want the hearer to change his beliefs about the world (to inform him), or I want the hearer to operate a change in the world in accordance with what I say (to request, order him) or to expect future events or actions to occur (to promise or warn him). This is all well-known and need no further discussion, but the point is that such illocutionary actions are indeed actions as defined: changes of the world—viz. in the various knowledge, beliefs, expectation, wish systems of the hearer—brought about by an interpretation (of an interpretation o f . . . ) of some doing, viz. a locutionary action. More important still is the fact that an illocutionary act has values only for propositional (meaning and reference) acts as arguments, where the basis for the propositional acts may either be spoken written or another conventional act. The point of illocutionary acts, and hence of speaking, is to contribute to the successful interaction of the participants. That is, I have certain wishes with respect to the state of the world and the role of the possible actions of the hearer to bring about that state. In order for the- hearer, as for agents in general, to accomplish an action successfully, he must have sufficient information about the initial state of the world, not only about the actual state of Objective' affairs, but also about his own abilities, his own wishes, and above all about the wishes of others, the obligations and norms constraining the set of his possible actions at a given moment. If I want the hearer to accomplish his action successfully, I must provide him all information which he does not have, both about the initial conditions of his action and about type and desired consequence of his action. The basic illocutionary act, therefore, is that of simple information transmission, e.g. by uttering "I am hungry". Only if my hearer's knowledge set has this information, it makes sense for him to engage in some action, after a rather complex process of inductive inference in which much other information from memory and contextual perception is needed (see Chapter 9). That is, my hearer may spontaneously give me a

Acts and Speech Acts

127

piece of bread, or he may interpret my information as an indirect request and give me a piece of bread. I may as a speaker leave no room for interpretation, and add to the first utterance: "Please, give me a piece of bread", thereby accomplishing a request, i.e. specifying a wanted action of the hearer. 3.3. Until now, everything is relatively straightforward even if a number of important action theoretical problems have not been treated adequately. For our discussion it becomes crucial in what respect we should speak of compound and complex speech acts and of sequences of speech acts (where a speech act is understood as an illocutionary act accomplished by an act of speaking or writing) and how these relate with the structure of sentences and sentence sequences. First of all, it should be recalled that the minimal basis for an illocutionary act is a propositional act/referring act. Our knowledge of the world, and hence that of our hearer, comes in propositional chunks, referring to the facts defining the properties of each possible world. I may refer to a pipe, by expressing and uttering the phrase that pipe, but do not thereby alone express a proposition nor refer to a fact. Hence, if, according to definition, an illocutionary act must result in a condition for possible action of the hearer, its mapping in the discourse is at least propositional. But, is that mapping one-one or may it also be one-many? Do we accomplish one illocutionary act by expressing/uttering one proposition, or may we accomplish one illocutionary act by expressing/uttering several propositions? An answer to this question at least requires a sound theory of propositions. Propositions, and hence propositional acts, may themselves be compound or complex. A compound proposition, in classical logical theory, consists of several propositions connected by one or more binary connectives. Semantically, a compound proposition, just like any proposition, refers to (is interpreted as) a fact, viz. the existence of a relation between several facts. More in particular, it is expressed that one proposition (itself again single or compound) is a condition for another proposition. In order to refer to an actual relation between facts, I must refer to the terms of that relation (or operation), viz. to the component

128

Issues in the Pragmatics of Discourse

facts, just as in accomplishing the compound action of building a house I must accomplish a certain number of component actions in order to be successful. A proposition may also be complex when one of the propositions is used to form the main proposition, e.g. when a relation is established between a fact and another particular individual. In that case one of the propositions is one of the arguments of an η-place predicate. Thus (78)

I am poor, but I do not care,

expresses a compound proposition, whereas (79)

My poverty does not bother me.

expresses a complex proposition because the proposition Ί am poor' is the first argument of the predicate 'to bother', whereas in (78) the fact that I am poor is a condition determining another fact (my not caring). Now, in both cases we may say that by uttering these sentences (in appropriate circumstances) I accomplish one assertion, i.e. the basic illocutionary act of information transmission. Yet, for (78) at least it may be maintained that I accomplish two or even three assertions. That is, I want my hearer to know that I am poor, that I do not care, and that I do not care inspite of my poverty. The most elegant way to solve this theoretical puzzle is to adopt again compound actions, viz. a compound assertion, consisting of three assertions. This is reasonable, because I may only assert that my poverty does not bother me, if the hearer already has the information that I am in fact poor, that is if he knows the referent of the argument of the relational expression. If not, a component action is needed first in order to provide this information. Hence by uttering (78) I accomplish a compound assertion. What about (79)? In the same circumstances as for (78), the speaker may already have the information that I am poor. In that case no assertive illocutionary act is needed, only an appropriate act of (contextually determined) reference to an identified individual, viz. to the fact that I am poor. So, from one point of view, (79) when uttered is one single assertion. But, in what

Acts and Speech Acts

129

respect is (79) illocutionarily different from the utterance of (80)

Although I am poor, it does not bother me.

Again we have two facts and a (concessive) relation between those facts, expressed by a connective. The first clause is subordinated and presupposed. That is, the speaker knows or assumes that this information is already available in the knowledge set of the hearer. Hence, no specific assertion is necessary to change this set in that respect. The difference with (79), first of all, is semantic: not only reference is made to my poverty and to my state of mind (bothering) but also to the specific relation between these two facts, viz. the fact that I do not bother in spite of my poverty, implying that the speaker believes that in normal cases poverty can be a sufficient reason for bothering, an implication which is not necessarily present in (79)—although there the negation has presuppositional aspects, viz. with respect to other's beliefs about my bothering about my poverty. Hence, by uttering (80) I make one composite assertion consisting of two assertions. The uttering of (80) would be a complex assertion if we could say that by uttering thes first clause in subordinate position we thereby accomplish the illocutionary act of presupposition. Although a full discussion of the various aspects of the notion of presupposition cannot be given here, we should at least examine in which respect presupposition could be viewed as an act. 3.4. The conditions to be satisfied by presuppositions in order to be qualified as actions are that there is an agent accomplishing a doing with the purpose of somehow changing a state or course of affairs. If we exclude mental acts for the moment, we can perform a presuppositional act only by expressing a proposition byiuttering a sentence. In a propositional act we assign properties or relations to (η-tuples of) individuals, and, by uttering a sentence we want that the hearer—by an act of interpretation of the perceived sentence—accomplishes the 'same' act. Similarly, by accomplishing a propositional act, and uttering a sentence expressing the proposition, we may intend to refer to particular individuals, properties and relations in some possible world from the point of

130

Issues in the Pragmatics of Discourse

view of the context of utterance. We have assumed that by thus expressing a proposition we refer to facts, and an act of reference has the further purpose that the hearer, by accomplishing the corresponding propositional act, also accomplishes the corresponding referential act. Since we thus change the state of the hearer, propositional and referential acts, just like locutionary acts, could be called 'illocutionary' in the wide sense. Yet, they are all what we could call 'linguistic' acts: we only want that the hearer hears and understands what we say and what we are talking about, but do not yet want to change him in a way such that the conditions for his future actions are modified. In other words: he does not yet know what we want with our utterance. In this perspective, a presupposition should be viewed as part of a referential act, viz. as that part of the act of reference in which we refer to facts, e.g. in (80) to the fact that I am poor. The act of reference, as we saw above, is based on knowledge of the world: we can refer only to those individuals whose existence (in a particular world) is known to us. In communicative contexts, however, it normally and typically occurs that some individuals, and especially individual facts, are known to the speaker, but not to the hearer. Hence we need a device such that the hearer is able to change his knowledge set with respect to the existence of such individuals. Now, by accomplishing the act of reference to individuals unknown to the hearer we may at the same time accomplish an illocutionary act of knowledge transmission, viz. we perform an assertion. This reference to unknown individuals is usually also indicated in the structure of the sentence. The information which is to be asserted may be given in the verb phrase, final position, or in initial position with specific stress; we use different articles to refer to 'known' objects and to 'unknown' objects, or to objects the identification of which the speaker judges irrelevant in the knowledge set of the hearer. A presupposition, thus, is an act of reference to a fact which is (assumed to be) known to the hearer, but not an illocutionary act in the strict sense because I do not, by this referential act, want to change the knowledge set of the hearer (beyond a change of knowledge pertaining to the fact that I am speaking, meaning something and talking about something). A presupposition is 'pragmatic' in the

Acts and Speech Acts

131

sense, however, that it is a referential act in which assumptions are made about the available knowledge in the hearer. Reference to 'new' information, and hence assertion, is usually made with respect to Old' information: known individuals are assigned unknown properties or relations, or conversely. Although it may be relevant to distinguish in a sentence between topic and comment in this respect, it does not seem possible to distinguish between a presuppositional act and an assertive act in the utterance of a sentence. First of all, a presupposition is a referential (semantic) act, whereas an assertion is an illocutionary act. Second, an assertion can be made only on the basis of the complete referential act, accomplished by the utterance of a whole sentence. More in particular, an assertion takes the whole (simple or composite) propositional and referential acts as argument, including reference to old and new individuals (things or facts). We may of course introduce a new term for those referential acts (or parts of a sentence) which pertain to facts unknown to the hearer, e.g. the act of introduction, thereby assigning a binary structure to each sentence based act of reference. Just as the act of presupposition is related to definite description (a presupposition may be called a definite description of a fact), the act of introduction is related to indefinite descriptions with which individual things are introduced. Note that introductions are propositionally based. That is, in a sentence like: (81) Peter met a girl. there are introduced two units of propositional information (at least), viz. that Peter met somebody (introduction of a relation, or more precisely, introduction of the fact that some relation exists) and that that somebody is a girl (and that the meeting took place in the past). Hence, there is a difference between the topiccomment distinction (at least in one of its interpretations) and the presupposition-introduction distinction, although they are closely related. Topics are the linguistic (syntactic) realizations of presupposed information, comments those of introduced information, i.e. the noun phrase Peter and the verb phrase met a girl in

132

Issues in the Pragmatics of Discourse

(81) respectively.4 The act of introduction is usually compound: several units of information may be conveyed by it at the same time. If necessary, we might also distinguish between basic and simple propositional and referential acts in order to relate simple acts with simple syntactic units (clauses) and units of conventional information transmission. In that case the 'logical' atomic propositions like GIRL (a) and MEET (a, b), or MEET (/) and PAST (f [a, b]), etc. would as basic propositions together form the single proposition 'Peter met a girl' underlying (81). Single propositions and propositional actions of this sort are the elements of composite sentences and composite propositional, referential and illocutionary acts. 3.5. From this discussion about the action theoretical status of presupposition we have concluded that presuppositions are not illocutionary acts in the strict sense. Unlike assertions, they are acts of semantic reference. Hence we cannot say that the utterance of a sentence like (80) is a complex illocutionary action of which the presupposition would be an auxiliary illocutionary action. Just like all referential/prepositional acts they are the basis for an illocutionary act. Although we may say that the introduction, i.e. the reference to new individual objects and facts, is the specific base of the assertion because it indicates the propositions with which the hearer is expected to enlarge his epistemic set, the assertion nevertheless must have the whole sentence as its 'scope' because it must also be indicated with which existing information about facts the new information should be connected. It seems to follow that the subordinate-coordinate distinction does not run parallel with a distinction between 'subordinate' (auxiliary) and 'coordinate' (component) illocutionary acts, but that it is a structural indication for two types of referential acts. In other words, it is the sentence boundary which is necessary for a distinction of different illocutionary acts, even if one illocutionary act may be compound and be performed by the utterance of one sentence. As we saw in earlier examples, sequences of illocutionary acts require sequences of sentences. It is at this level also that we may speak of auxiliary illocutionary acts. Thus, by uttering the

Acts and Speech Acts

133

following sequence: (82)

I have no money. Can you lend me ten dollar?

I first make an assertion and then a request. The assertion, as was spelled out above, has the basic purpose to provide the hearer with the information which may become relevant for his possible further actions. This action is referred to (as wanted) in the following sentence. This request could also be made by only uttering the second sentence. Hence the assertion has auxiliary function, providing a reason for making the request and thus enhancing the successfulness of the request. The assertion being a condition for the request in this case, we could say that the utterance of the whole sequence is a request. This does not mean that each sequence must have one global speech act of which the component speech acts are merely conditions, components or consequences. The following sequence: (83)

This summer I am going to Italy. I'll send you a postcard.

is an expression of a sequence of an assertion and a promise, because the assertion is not (only) made to create a condition for successful promising. The relations involved are essentially semantic: the act of sending a post-card is locally conditioned by the presence in Italy, because it follows from this sequence that the post-card will be sent from Italy. More intricate is the relation between speech act sequences and discourse in such sentences as (84)

I'll send you a post-card this summer, because I am going to Italy.

Is this a promise, an assertion, or both (or is the sentence pragmatically ambiguous in this respect)? Under one reading, it is an assertion of a compound proposition, i.e. asserting a conditional relation between two facts, of which the first proposition may be presupposed and the second, expressed in final position, intro-

134

Issues in the Pragmatics of Discourse

duced. The assertion of conditions of events in this case has the character of an explanation. We see that a proposition may have illocutionary function only under specific conditions, e.g. whether it is part of the introduction part of the sentence or not. The first clause has the typical semantic structure of a promise (reference to future acts of the speaker) but is no promise here. Under another reading, in which the first clause does not express a presupposed proposition, the utterance of the sentence, or at least of its first clause, is a promise. The subordinate clause in final position, however, is not presupposed, but an independent assertion made about the reason for sending the postcard. So, do we here have two different speech acts manifested in one sentence anyway? Intuitions, again, do seem little clear on this point. As far as we can judge the second reading could not be acceptably expressed in a sentence like (84), and would at least require a coordinate construction like: (85)

I'll send you a post-card this summer, for I am going to Italy.

In the first reading of (84)—assertion that the fact that I'll send a postcard is conditioned by my stay in Italy—there is no longer pause between the clauses. In a reading where a promise is made, but where because is used, we would have at least a longer pause before because, which in that case becomes equivalent with sentence initial for, indicating explanatory assertion. As soon as the promise-reading is dominant the assertion looses its independence, and becomes auxiliary to the promise. It provides not only a specification under which circumstances I send you a post-card, but at the same time provides grounds for the credibility of the promise. These following auxiliary speech acts (mostly assertions), however, require at least coordinative construction. Specifically, we have the feeling that as soon as the causal/ reason relation expressed by for (or because) is stressed, the first clause looses more or less its promising character and becomes a simple announcement about future actions (i.e. no longer has commitments as in promises). Using sentence initial for or

Acts and Speech Acts

135

because (i.e. making two sentences out of (85)) is a way to reduce the causal relations between the facts as such, and to stress the relation between the speech acts. Some additional evidence for this interpretation of the differences between (84), (85) and the sequence structure, comes from Dutch. Here the future is usually expressed also with a present tense. The direct causal relation between future actions would in that case be expressed as follows: (86)

Ik stuur je van de zomer een kaart, omdat ik naar Italie ga.

Promises, however, may also be accompanied by the auxiliary zal ('shall', 'will'): "Ik zal je een kaart sturen!". The same holds for menaces: "Ik zal je wel krijgen!" [I'll get you!]. The use of zal in (86), however, seems unacceptable, but normal in: (87)

Ik zal je van de zomer een kaart sturen. Want ik ga naar Italie.

Another, even more typical, example is the following sentence, traditionally/conventionally uttered for children: (88)

Because you have been so sweet, I'll give you a rabbit.

This is clearly a promise, and not primarily an assertion or praise, which would have been expressed as follows: (89)

You have been very sweet. Therefore, I'll give you a rabbit.

In the first example a relation is established between the behaviour of the hearer and the future action of the speaker (who could, indeed, later say: "I gave you a rabbit, because you were so sweet"). The presupposed subordinate clause specifies the conditions for the action mentioned in the introduction part of the sentence, viz. in the second clause. Although the utterance of the whole sentence is a promise, the purposed change in the hearer, consisting in the establishment of expectations based on obliga-

136

Issues in the Pragmatics of Discourse

tions of the speaker, is however based on the introduction part of the sentence. In (89), however, first an assertion or praise is made. The positive evaluation of other's behaviour in such cases typically leads to obligations, whereas obligations in turn may be grounds for actions of promise. This promise is executed by uttering the next sentence, and the grounds for this promise, entailed by the performance of an act of praise, are then indicated by sentence initial therefore. 3.6. These examples are closely related to conditional illocutionary acts, like conditional promises: (90) If I go to Italy, I'll send you a post-card. (91) If you are sweet, I'll give you a rabbit. Such examples, especially their truth conditions, are well-known from discussions in philosophical logic.5 The term 'conditional promise' may, however, be misleading. It does not seem correct to say that by uttering such conditional sentences, I make a promise under such conditions. The promise is made allright but the 'domain of its validity' is specified. That is, I refer to my future action in a possible world which is selected by the proposition expressed in the //-clause. In the first case this condition is a condition for the ability, i.e. for the possibility of the action referred to, where the obligations are self-imposed or more in general conventional (if you go on your holidays, you send your friends or relatives a post-card). In the second case, the condition has direct reference to the creation of such obligations by the behaviour of the hearer. Although the //-clauses are in subordinate position, they do not express presupposed propositions, because the information is not known to the hearer, nor to the speaker for that matter. Conditionals typically refer to facts in non-actual possible worlds to which the speaker does not have epistemic access at the moment of utterance. If he had, we would have the because examples as in the previous section. It follows that the relation between the facts as such does not differ (in all cases it is a relation of 'sufficient reason'), but only the possible world in which this relation

Acts and Speech Acts

137

exists. Strictly speaking, then, i f . . . (then) is not itself a connective at all, but an operator, whereas the under-lying connection is 'reason' or 'cause'. These and other properties of conditionals and counterfactuals will not further be discussed here.6 More important for our discussion is the observation that subordinate clauses in initial position need not be presupposed. Hence, the whole complex sentence expresses an introduction, although it might be said that in //-conditionals the //-clause expresses not a presupposition but a supposition. In both cases, the introduction is given with respect to the condition given in the first part of the composite sentence. With presuppositions this condition is known to be true, with suppositions it is assumed to be true (in some possible world), or assumed to be false in the actual world as is the case for counterfactuals.7 Another characteristic of //-conditionals relevant to our discussion, and an indication for their specific operator-like status, is the fact that if-then sentences do not easily split up into sequences. In English and other languages there is no specific sentence initial form of // (indeed, // is sentence initial) comparable with for, therefore, so, yet, etc., although we may sometimes have sequences like: (92)

I'll go to Italy this summer. If I have enough money.

Here sentence // requires heavy stress, and is often preceded by phrases like, at least, that is, etc. Such independent //-sentences have the function of 'correction' or 'limitation' with respect to previous assertions. Without the operator // the first sentence of (92) refers to a fact in a future actual world, and hence has the illocutionary force of an assertive announcement. Without operators (modal adverbs like maybe, probably, etc.) sentences are neutrally interpreted as referring to facts in the actual world, or to the world established by predicates (want, wish, dream, etc.) and propositions expressed in previous sentences. Following ifsentences may be used to correct or limit this (unconditional) interpretation. Just as the other sentence initial connectives it thus rather links illocutionary acts than facts: it corrects a previous sentence.

138

Issues in the Pragmatics of Discourse

3.7. Although the previous sections have yielded somewhat more insight into the issue sentence vs. sequence and related problems, the data and the observations are still too weak for explicit and systematic hypotheses. Nevertheless we may resume some of the main principles governing the structure of discourse and conversation in this respect. We will take the following six basic forms as our point of reference (using causals as example): (93)1. II.