book, we survey the accounts of CIs developed by Grice himself and ...... enforcement officials: book dealer Sebastiaan Hesselink and a book thief readily.
Imagination and Convention
FOR INDEXING
FOR INDEXING
Imagination and Convention Distinguishing Grammar and Inference in Language
Ernie Lepore and Matthew Stone
1
FOR INDEXING
3
Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © Ernie Lepore and Matthew Stone 2015 The moral rights of the authors have been asserted First Edition published in 2015 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 0000000000 ISBN 978–0–19–871718–8 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.
FOR INDEXING
Preface There is no hiding our agenda in this book. We are defenders of grammar, of meaning, and of common sense. We are exponents of the richness of human experience and the creativity of language. We believe in drawing useful and principled distinctions. But our book is not a traditional monograph. It is an opinionated survey. Whether you endorse our views or contest them, we hope our book helps you to articulate your own views about language more robustly and to draw more eclectically in your thinking on data and theories from across cognitive science. This book has been a long time coming. The ideas germinated in a seminar on metaphor at Rutgers in 2007, and were followed up in seminars on the semantics–pragmatics divide (with Gil Harman) in 2012, and on creativity and language (with Liz Camp) in 2014. We also taught preliminary versions of the book at the 2010 Summer School on Meaning at the Central European University in Budapest, the 2011 Beijing Conference on Language and Value, and the 2012 International Summer School on Cognitive Sciences and Semantics in Pumpuri, Latvia. We have also presented these ideas, in various stages of development, at Ankara University, Beijing Normal University, Boston College, Cambridge University, Concordia University, CUNY, Hampshire College, Harvard University, the Jagellonian University Krakow, Peking University, Shanghai University, the 2011 Siena–Rutgers Workshop in Cognitive Science, Swarthmore College, Texas Tech, UCLA, the University of Amsterdam, the University of Helsinki, the University of London, the University of Massachusetts, the University of Oslo, the University of Lisbon, the University of Kansas, the University of Michigan, the University of Minnesota, and at meetings of the Central and Pacific divisions of the APA. We gratefully acknowledge the unfailing generosity of the organizers and participants of these events, and the many other colleagues with whom we have discussed the work in progress, for offering their encouragement, suggestions and criticisms. Their contributions have helped us to strengthen and clarify our arguments substantially. We particularly thank the readers recruited by Oxford University Press for their detailed and instructive feedback on an earlier version of the book. Chapter 10 overlaps partly with our paper Against Metaphorical Meaning from Topoi 2010:29(2). The summary and projection of Part III overlaps partly with our essay “Figures of Speech” from The Philosopher’s Magazine, 1st Quarter
FOR INDEXING
vi
prefa ce
2012. Chapter 14 overlaps partly with our chapter “David Lewis on Convention” from Barry Loewer and Jonathan Schaffer’s A Companion to David Lewis, to be published by Wiley Blackwell. We would also like to acknowledge the financial support we have gotten for the research. Both Ernie and Matthew enjoy extra research time through their appointment in the Center for Cognitive Science at Rutgers. Matthew’s work was also made possible by a fellowship at Rutgers’s Center for Cultural Analysis in 2007–8, a sabbatical leave in 2009–10, and by NSF grants HSD-0624191 and IIS1017811. Matthew, finally, because it is his first book, completes his efforts with a special sense of gratitude for the patient enthusiasm of his friends and colleagues while the manuscript slowly came together, and particularly thanks Ernie Lepore and Doug DeCarlo for the accommodation that enabled him to put so much time and energy into the project.
FOR INDEXING
Contents 1. Overview
Part I.
1
The Landscape of Pragmatic Inference
Introduction to Part I
9 11
2. The Gricean Framework
15
3. The Linguistic Turn
40
4. The Psychological Turn
60
Summary of Part I and Projection
81
Part II.
85
The Interpretive Effects of Linguistic Rules
Introduction to Part II
87
5. The Scope of Linguistic Conventions
89
6. Speech Act Conventions: Indirection and Relevance
98
7. Presupposition and Anaphora: The Case of Tense and Aspect
115
8. Information Structure: Intonation and Scalars
128
Summary of Part II and Projection
147
Part III.
151
Varieties of Interpretive Reasoning
Introduction to Part III
153
9. The Scope of Interpretive Reasoning
155
10. Perspective Taking: Metaphor
162
11. Presenting Utterances: Sarcasm, Irony, and Humor
173
12. Leaving Things Open: Hinting
187
Summary of Part III and Projection
191
Part IV. Theorizing Semantics and Pragmatics
195
Introduction to Part IV
197
13. Interpretation and Intention Recognition
199
FOR INDEXING
viii
contents
14. Inquiry and the Formal Underpinnings of Communication
233
Conclusion
264
Bibliography
268
Index
281
FOR INDEXING
1 Overview
Should a speaker ask at a restaurant, “Can I have the French Toast?” you can safely infer she wants to place an order. Should a speaker describe the events of the last recession with, “Oil prices doubled and demand for consumer goods plunged,” you can reasonably take it that the plunge followed the doubling. Should you set out to describe a magic act with, “And then he put the red handkerchief on the side of the table,” and should your skeptical friend interject, “Well, it looked red,” you can confidently bet she disagrees with you about its color. What is it about each of these exchanges that permits an addressee to derive such a precise understanding of the speaker? And conversely, what lets the speaker anticipate and shape her interlocutors’ understanding so precisely? How much of this coordination derives from interlocutors’ specific knowledge of one another as people? How much exploits their knowledge of language itself? And what is the status of this understanding as the conversation goes forward? Have speakers signaled public meanings that contribute their shared inquiry? Or have they merely offered prompts to one another that suggest particular ways of imagining how things might be?
1.1 Semantics, Pragmatics, and Conversational Implicature A pervasive view answers these questions by establishing a division between two qualitatively different domains of knowledge and inference: a domain of semantics concerned with knowledge of language, and a domain of pragmatics concerned with explaining the behavior of rational agents. On this view, knowledge of language contributes only to settling what the speaker is saying. This enterprise is properly the domain of semantics, which describes the essentially arbitrary ways in which words and expressions are assigned semantic interpretations in context by the rules of language.
FOR INDEXING
2
over v iew
There’s more to a speech act, however, than just the content of an utterance as determined by context and the rules of language. Speakers can use utterances to ask or request, list or narrate, agree or doubt, as we saw in the interpretations with which we began. Speaker need to show what they are doing with the content they present. Conversely, addressees have to recognize what speakers are doing. These matters are properly the domain of pragmatics, which describes how interlocutors assess utterances against the aims and norms of ongoing collaborative activity. Pragmatic principles are fundamentally different from the principles of semantics. Most importantly, they lack the arbitrariness that characterizes linguistic rules. (The point is emphasized by Davis 1998.) To illustrate, consider the principle of accommodation in pragmatic inference, as proposed by Thomason (1990) and described further in Chapter 2. Thomason suggests that one of the things speakers can do with utterances is to prompt their audience to accommodate them, by acting to make sure that any preconditions of their communicative intentions are satisfied. Accommodation, for Thomason, is a particular instance of the general fact that we always assist with what’s needed for our collaborators’ intentions to succeed: We help each other out when we act together. Thus, we can explain pragmatic accommodation without any assumptions about the particular language interlocutors speak—or even about the language faculty itself. The paradigm case for this division of labor between semantics and pragmatics is the phenomenon of conversational implicature, as characterized by Grice (1975). In brief, a conversational implicature (CI) is a contribution associated with an utterance that the speaker does not explicitly say but that can be calculated from what the speaker does say on the assumption that her utterance is a reasonable way for her to advance the purposes of the conversation. As we shall see, the received view is that the characteristic interpretive judgments with which we opened the chapter are examples of CIs. The speaker’s request with Can I have the French Toast? is a CI. The temporal relationship connecting doubled and plunged is a CI. The sense of disagreement with it looked red—the suggestion that the handkerchief was not, in fact red—is also a CI. The theory of conversational implicature leads to characteristic answers to the questions we opened with. On the one hand, it emphasizes the role of open-ended explanatory inference and principles of rationality in deriving the contributions of utterances. In particular, it invites us to take a very narrow view of the conventions of meaning in language, incorporating only the broad lexical and compositional rules required to settle matters of reference, predication, quantification, and other truth-conditional effects. Part of the appeal of
FOR INDEXING
over v i ew
3
the theory of CIs, and its account of the semantics–pragmatics divide, is the streamlined, bare-bones model of semantics that researchers have appealed to in developing the theory. On the other hand, the theory promises a homogeneous account of speakers’ purposes in contributing information to conversation. Of course, the theory distinguishes between what a speaker says and what she does not say but merely implicates. However, the theory proposes that in either case the speaker will contribute information, update the context, and add to the common ground. This uniformity is also part of the appeal of the theory of CIs and its account of the semantics–pragmatics divide. Since Grice’s original statement of the theory of conversational implicatures in his 1967 William James Lectures at Harvard, Grice’s ideas have inspired a wide variety of modified and extended accounts. These frameworks turn out to conceptualize pragmatic principles and pragmatic reasoning in diverse ways, and as a result they make strikingly different claims both about what, in fact, pragmatic reasoning contributes to interpretation, and about how it does it. In Part I of the book, we survey the accounts of CIs developed by Grice himself and explored in subsequent research. We shall reach the perhaps surprising conclusion that, while Grice’s theory of CIs provides the foundation for most of modern pragmatics, the place and status of the semantics–pragmatic divide remains deeply contested.
1.2 The Diversity of Linguistic Rules and Interpretive Inference Our discussion in Part I will showcase influential arguments that suggest that the received understanding of semantics and pragmatics offers an elegant reconciliation of common-sense ideas about language and about action. Moreover, we shall also survey influential analyses that suggest that the received understanding offers simple and powerful pragmatic explanations of a wide range of interpretive effects. The apparent success of pragmatics represents a major factor in the appeal of the received view. We believe, however, that this success is merely apparent. In the rest of the book, we explain why we reject the received view, and advance an alternative view that, we think, better captures both the interpretive effects at play, and our common-sense intuitions about them. In Parts II and III of the book, we revisit the interpretive effects that researchers have sought to explain as CIs. We proceed in Part II by exploring ways that speakers can signal apparently pragmatic meaning through linguistic devices that allow them to explicitly mark what they intend their utterances to do. Griceans have been wrong to categorize these cases as CIs, we argue, because
FOR INDEXING
4
over v iew
these effects rely not on general interpretive principles but instead on essentially arbitrary links between forms and their interpretations. This is the “Convention” part of our title. We will look at three sets of phenomena in particular. The first concerns speech acts. Interlocutors often achieve effects such as polite indirection which serve their purposes in smoothing social interactions. But these effects, we will argue, are in fact triggered by the specific and varied rules of individual languages; there’s more than rationality involved. For example, it’s part of speakers’ knowledge of English, we will suggest, that Can I have the French Toast? is an indirect request, and therefore the request interpretation is not calculated and not a CI. We continue with presupposed and referential components of meaning, which serve to signal constraints on the context. This account eliminates other alleged CIs, like the interpretation that the plunge follows the doubling in Oil prices doubled and demand for consumer goods plunged. We suggest that the interpretive effects in question are due to the resolution of linguistically encoded referential meaning—temporal reference, in particular. Attending to these new dimensions of meaning gives a better account of the dependence of many aspects of interpretation on fine-grained linguistic choices. Moreover, it allows us to account for this interpretive variability without postulating ambiguity, one of the chief purported advantages of the theory of CIs. We end our tour with information structure. Language includes a range of devices that allow speakers to signal how their utterances connect to the ongoing discourse, including the variation in the accentuation and phrasing of English intonation. For example, the natural reading of Well, it looked red comes with intonation that conveys the speaker’s disagreement by invoking a contextually specified alternative and indicating the speaker’s uncertainty about it. In this case, of course, the alternative is the prior speaker’s commitment that the handkerchief was red. But if the speaker’s information structure encodes her skepticism explicitly, it cannot be a CI. Where Part II surveys the diversity of linguistic meanings in signaling interpretation, Part III surveys the varieties of interpretive reasoning that interlocutors can exploit. We turn to cases whose content is not resolved conventionally, but that instead require imaginative, productive inference on the part of the audience, and where speakers regularly anticipate and prompt such inference. The effects we consider cover a wide range of figurative and evocative language, including metaphor, sarcasm, irony, humor, and hinting. Many theorists have suggested that general interpretive principles fully explain these creative uses of language. However, we argue that such utterances are better characterized as invitations
FOR INDEXING
over v i ew
5
to the audience to follow a specific direction of thought in exploring the contributions of the utterance. This is the “Imagination” part of our title. The thinking involved is heterogeneous and diverse—it’s not just a circumscribed or uniform application of principles of rationality. And so, the insights interlocutors get by pursuing and appreciating this thinking also fall outside the scope of pragmatics as traditionally conceived. In particular, in Part III, our strategy is to confront diverse practices for using language indirectly, creatively or figuratively, and to describe, as faithfully as we can, the imaginative engagement that these practices require and the insights they offer to interlocutors. Metaphorical discourse, for example, offers an open-ended invitation to find insights from seeing one domain, analogically, as another. Sarcasm, meanwhile, is an open-ended invitation to explore some aspect of the ordinary meaning of a discourse as an inversion of an appropriate contribution to the conversation. Irony is an open-ended invitation to engage with an imagined speaker, who would sincerely use a discourse that has been exhibited by the actual speaker merely in pretense. Humorous discourse offers an open-ended invitation to appreciate and enjoy an ambiguity among contrasting perspectives with incompatible affective valences. Hinting is an invitation to explore the associations of otherwise incongruous details in a discourse for new insights into the communicative situation. By characterizing these practices more precisely, through a range of arguments and examples, we are able to substantiate our intuitive claim that these practices involve heterogeneous kinds of thinking, aimed at different goals and appealing to different principles. We can acknowledge that a speaker often invokes these practices collaboratively, with intentions we can understand. But, we suggest, our understanding of the speaker often emerges only retrospectively, from the insights that their utterances prompt. That’s why it’s a mistake to think of the process as a matter of collaborative reasoning about speakers’ intentions—or any other uniform strategy.
1.3 Rethinking Language and Communication Our survey of these diverse phenomena in Parts II and III leads us to a broad set of departures from the received picture. The rules of language are not exhausted by a bare-bones semantics; the rules of language can signal many other aspects of interpretation. Open-ended interpretive reasoning does not privilege rationality; utterances are wildly heterogeneous not only in the roles they play in shaping conversation but in the reasoning they recruit to do so. These claims represent a major departure from familiar views of the semantics–pragmatics divide.
FOR INDEXING
6
over v iew
If we are right, then both the data and the explanatory apparatus of the Gricean program and its various successors require profound reinterpretation. Much more of interpretation than one might have expected is actually encoded in the rules of language. Conversely, the inferences speakers invite their audiences to draw from their utterances are not as pervasive as the theory of CI pretends, and do not form a natural class. Put most starkly: we have no use for a category of conversational implicatures, as traditionally and currently understood. If we are right, theorists across cognitive science must approach linguistic knowledge and interpretive reasoning in more nuanced ways than they often have. We cannot simply separate a domain of semantics, concerned with truth and reference, from a domain of pragmatics, concerned with the behavior of rational agents. When we study the nature of meaning, our critical work must target relevant linguistic knowledge flexibly and with an eye to the empirical diversity of linguistic rules. When we study the nature of interpretation, we must respect the heterogeneous and overlapping mechanisms that may contribute to the interpretive effects at issue. We conclude in Part IV with some key illustrations and applications of these ideas. Chapter 13 critiques the intentional framework Grice used to characterize interpretation. Grice saw intention recognition as constitutive of communication—for Grice, this was a way to naturalize linguistic knowledge. We think this strategy is incompatible with the results of Parts II and III. Once we acknowledge that linguistic meaning is broader in scope than Grice envisioned, and that interpretive reasoning is more diverse than he imagined, we are forced to abandon any reduction of meaning to communicative intention. This is a good thing, we suggest. Intention recognition has a special status in everything people do together, whether in performing linguistic utterances, using other signals, demonstrating how the world is, or accomplishing concrete tasks: Intention recognition is necessary for all collaboration. But, on our view, what makes linguistic collaboration analogous to those other kinds of collaboration is precisely that interlocutors’ contributions to conversation—just like their contributions to practical activity—carry things forward according to a dynamic that is antecedent to their intentions, and independent of them. It is a mistake, therefore, to think that “mind reading” inevitably intervenes in meaning making, or that the theory of CIs eliminates the need to describe linguistic conventions and imaginative mechanisms in detail. Meanwhile, in Chapter 14, we offer an illustration of the new philosophical positions that our view makes room for. Drawing on two separate but consonant insights, from Lewis (1969 and 1979), we suggest a way of thinking about the inquiry that interlocutors pursue in conversation as completely governed by
FOR INDEXING
over v i ew
7
linguistic rules. Lewis (1979) invites us to conceptualize certain contributions to conversation as updates to abstract scoreboard on which the propositions at issue in a conversation are made public and subjected to shared scrutiny. Now, such a scoreboard can evolve in whatever way agents agree on; agents must coordinate to align their understanding of the state of the proceedings. Thus, a scoreboard model naturally dovetails with Lewis’s (1969) understanding of a convention as a solution to a recurrent coordination problem, and suggests that inquiry itself is the natural scope of the conventional rules we use to communicate. Thus, Chapter 14 offers a new, broader philosophical conception of the domain of conventional meaning. This conception goes beyond matters of reference and truth to embrace a wide range of knowledge about linguistic action and conversational context. But it does not simply relabel implicatures as meaning. It specifically targets the way that some utterances transparently put propositions into play into a dialogue in particular ways by signaling them through linguistic conventions. And it specifically excludes the indefinite number of ways in which other linguistic practices less transparently allow us to get information across and shape others’ views of the world. In this sense, developing a linguistic model of inquiry is possible only when we abandon the received assumptions about the semantics–pragmatics divide that we criticize in this book: particularly the narrow range of semantic rules and the special place for collaborative rationality in reasoning about linguistic action. There are antecedents in the semantics and pragmatic literature for many of our critical points. Authors as diverse as Bach and Harnish (1979), Davis (1992, 1998), Horn (1984), Levinson (2000), Sperber and Wilson (1986), and Thomason (1990), among numerous others, have attacked the architecture of Gricean implicature. We will draw on some of the prior philosophical arguments that show that allegedly pragmatic content is in fact linguistically determined (Davis 1998), and we will also draw on prior linguistic theories that actually show how this determination might proceed (Asher and Lascarides 2003; Lascarides and Asher 1993; Pierrehumbert and Hirschberg 1990, among others). We will draw on prior philosophical arguments that distance the hermeneutics of figurative language from the recognition of conventional or even intended content (Black 1955; Davidson 1978, among others), and we will draw on specific interpretive theories that invoke distinctive principles to explain the significance of figurative or suggestive speech (Camp 2003, 2008, e.g.). We will reach conclusions that can readily be echoed in other frameworks, particularly formal approaches to discourse and dialogue, as epitomized by SDRT (Asher and Lascarides 2003) and Coherence Theory (Hobbs 1990; Kehler 2001). Indeed, our book
FOR INDEXING
8
over v iew
can be seen as an amplification and defense of the formal work of one of us on models of collaborative reasoning in conversation (Thomason et al. 2006)— the book can be taken as an explanation why their interpretive judgments and reasoning mechanisms turn out to be so different from those traditionally linked to cooperation elsewhere in the literature. We want to emphasize, however, that this book is first and foremost a philosophy book—its point is not to settle this or that resolution to a particular empirical issue, or to advocate for a specific theoretical framework of discourse interpretation. It is rather to sharpen intuitions and draw distinctions about language use. Ever since Grice (1957) first called attention to the problem, philosophers have faced a challenge in reconciling the complexities of actual language use with common-sense claims about communication, namely: (1) (2) (3)
Utterances are contributions to joint activity. What they communicate is public, and so can be shared. Utterances can nevertheless prompt us to idiosyncratic and open-ended understandings of the world and one another.
We explore in what follows how best to think of language, collaboration, and meaning, in light of the developments in philosophy and linguistics since Grice (1957), in order to frame an understanding on which claims (1–3) might actually turn out to be true.
FOR INDEXING
PA RT I
The Landscape of Pragmatic Inference
FOR INDEXING
FOR INDEXING
Introduction to Part I
We begin this book with a philosophically minded survey of theories of conversational implicature. In Chapter 2 we look at the seminal views of Grice himself and the researchers who have followed most closely in his footsteps: they all explain implicatures in terms of the norms of cooperative rationality. We focus in Chapter 3 on accounts that instead explain conversational implicatures by appeal to distinctively linguistic knowledge and principles. Finally, we focus in Chapter 4 on proposals to derive implicatures from psychological principles discovered through empirical study of the human mind. As we shall see, the shared terminology and common empirical coverage of these theories belies profound scientific and philosophical differences. Some researchers think linguistic knowledge goes further in spelling out the interpretations of utterances than Grice thought. Others think linguistic knowledge says even less about interpretation than Grice thought. Some think Grice’s explanatory principles should be maintained. Others think they should be refined to capture collaboration more precisely. Still others argue that new linguistic or psychological principles are needed to explain how audiences interpret utterances. These views may all present themselves as theories of conversational implicature, but they have very different consequences for the relationship between linguistics and psychology in general, and for the divide between semantics and pragmatics in particular. The views we survey in Part I cannot all be right. In fact, we will not endorse any of them. Nevertheless, here in Part I, we are reluctant to foreshadow too many of our objections. All the views we consider are important and influential in the field. For that reason alone, we think it’s important to approach them on their own terms. But we also suspect that many of our readers already sympathize with one or another of the perspectives we will consider. We hope to articulate, untendentiously, a map of extant theories that anyone can endorse.
FOR INDEXING
12
int ro duc t ion to p a r t i
All the same, however, our presentation will highlight assumptions of the various views that are often left tacit—assumptions that we think deserve much greater critical attention. We hope to challenge our readers by putting these assumptions front and center from the start. One key case underpins all the work we will consider in Part I, and yet we think it is particularly problematic. This is the claim, made by Grice already in (1957), that a speaker’s contributions to conversation are, first and foremost, a matter of her intentions.Neale (1992) offers a review of Grice’s philosophy that emphasizes the centrality of this principle. Grice’s view is not simply that communication requires the audience to recognize the speaker’s intended interpretation of her utterance. This is an obvious truism. Language is fraught with ambiguity, and it offers placeholders— she, he, it; this, that; here, there—that seem to fold the contents of consciousness directly into the meanings of utterances. If an audience is to understand the speaker, they must resolve ambiguities and recognize references in the way the speaker had in mind. Though Grice has certainly distinguished himself as a modern exponent of this common-sense view, he did not originate it. It seems to have been a commonplace at least since the British Enlightenment. Blackstone’s (1765) Commentaries on the Laws of England includes a famous invocation of this principle in (4), as well as specific suggestions about how it might be realized. (4)
The fairest and most rational method to interpret the will of the legislator, is by exploring his intentions at the time when the law was made, by signs the most natural and probable. And these signs are either the words, the context, the subject-matter, the effects and consequence, or the spirit and reason of the law. (Blackstone 1765: i, Introduction, §2)
Blackstone goes on to explain how jurists should adjudicate the meaning of the statute by a range of principles that provide evidence to reconstruct what lawmakers had in mind. We will assume something like this model of pragmatic disambiguation throughout the book. But Grice’s view, as we shall see, goes much further. Grice develops his ideas as part of an attempt to characterize what a speaker means with an utterance. Grice’s goal is to give a reductive analysis of linguistic knowledge, so his view of meaning is grounded directly in psychological notions. The heart of his proposal is that interpreting an utterance is nothing more than identifying the speaker’s communicative intention. Communicative intentions are special in that they must be overtly expressed and are fulfilled only by being recognized, but they are not special in involving particular kinds of actions,
FOR INDEXING
i n t ro du c t i on to p a r t i 13 having particular kinds of effects on the interlocutor, or even in relying specifically on linguistic knowledge. (Nothing in this book hinges on any analysis of speaker meaning, so we abstract away here from the various details of Grice’s view, and the intricacies of the critical literature that grew up around it, though we return to the link between meaning and intention in much more detail in Chapter 13.) Grice’s idea is thus that communicative intentions are prior to linguistic meanings. Linguistic meanings are just generalizations about the kinds of communicative intentions that speakers tend to have when they use certain expressions. In all cases, understanding a particular utterance is accomplished by inferring what the speaker means from whatever evidence the speaker provides. Facts about linguistic meaning are one source of evidence that can connect a speaker’s utterance to her meaning on a specific occasion. Context is another. The appeal of Grice’s approach is most evident in cases when speakers seem to act idiosyncratically and creatively in formulating communicative intentions. We’ll return throughout the book to the use of novel metaphors as a prototypical case, in Chapter 10 especially. Speakers’ meanings in these utterances go beyond what the semantic rules underwrite. Nevertheless, according to the Gricean tradition, these communicative intentions can be unproblematic. Speakers are manifesting their communicative intentions in the ordinary way, and hearers can and should recognize them just as they usually do. In fact, pragmatic theorists often cite the prevalence of conversational implicatures as evidence of the ubiquity of these creative kinds of meaning. Conversational implicatures thus suggest that the link between meaning and intention is a powerful explanatory principle. All the pragmatic theories we survey in Part I endorse it. In fact, however, Grice’s approach to communicative intentions already involves far-reaching implications for the relationship between semantics and pragmatics. We might want to resist these implications. For example, we may have to cut the Gricean link between interpretation and intention if we want to treat linguistic knowledge as providing conventional standards that speakers defer to. If speakers work to formulate their utterances in accord with external standards, and normally succeed in meeting those standards, then when hearers “recover an intended interpretation,” their job is just to apply the relevant standards to the speaker’s communicative act and infer an appropriate meaning. Explicit reasoning about what the speaker might have wanted or believed comes in, if at all, only when something goes wrong. Note that Blackstone’s commonsensical (4), and indeed all of his interpretive guidelines, are just as compatible with this perspective as as they are with Grice’s. In other words, this is a point where Grice’s specific philosophical innovations matter.
FOR INDEXING
14
int ro duc t ion to p a r t i
Conversely, cutting the link between interpretation and intention opens up room to characterize interpretive inference in new and open-ended ways. This freedom may be as important in language use as it is in other kinds of human activity. For example, consider how different chess is from language on the Gricean view. Normally, in chess, players intend the moves they make. But there are rules that say when and how each move is made. In the end, we identify the moves players make by consulting the rules, not by recognizing their intentions. Sometimes, in fact, we do recognize players’ intentions, perhaps as a side-effect of consulting the rules. But sometimes we don’t. Part of the fun of chess is how difficult it is to make sense of our opponents’ strategies and to predict how they intend the game to evolve; part of the fun is the unexpected insight we sometimes get from looking at a reconfigured chessboard—insight that prompts moves that neither player might otherwise have anticipated. In short, as diverse and far-reaching as these different theories of conversational implicature are, we see them as setting up a much larger debate about the nature of linguistic knowledge and communicative inference—one that will occupy us for the rest of the book.
FOR INDEXING
2 The Gricean Framework
We start with Grice (1975), and his observation that the point of offering an utterance to conversation sometimes seems far removed from its conventional meaning. The metaphor in (5) was a stock example already for Black (1955: 281). (5)
Richard is a lion.
The sentence means that Richard is a lion; it entails that Richard is not a fish. But normally, a speaker uttering this sentence isn’t likely to mean or imply either of these. More likely, the speaker’s point is to call attention to something about Richard that figures in our common-sense understanding of lions—perhaps Richard’s fierceness, leadership, or pride. This particular way of talking about metaphor is due to Searle (1979). Grice’s theory of conversational implicature (CI) offers an elegant and influential framework for making sense of speakers’ creative and flexible use of language in cases such as (5) and many others. Grice suggests that we should think of speakers’ points in terms of content they contribute to the conversation—information they want their audience to recognize and take up in further discussion. He calls such content speaker meaning. Sometimes, Grice suggests, speakers’ contributions can be recognized just from applying the rules of language. The resulting content specifies a contribution that the speaker evidently intends. Other times, however, as in (5), there is a looser relationship between linguistic content and speaker meaning. Then, Grice argues, to identify the contribution a speaker intends, interlocutors need to appeal to the principles of rational cooperation that govern conversation. In (5), for example, it wouldn’t make sense to claim that Richard is literally a lion. But lions are associated (metaphorically) with a range of other properties that it might be relevant to ascribe to Richard, and that we can assume the speaker intended us to think of.1 Grice’s category of CIs distills this insight. It has been profoundly influential. It helped to cement the modern perspective that ordinary language—warts 1 The interrelated significance of CI and speaker meaning is not always evident in Grice’s writing, but Neale (1992) reveals its importance as a consistent undercurrent in Grice’s thinking.
FOR INDEXING
16
the g r icean fr amewor k
and all—is rule-governed and amenable to analysis with formal tools, even as it sparked a diverse and exciting range of follow-up research across cognitive science. It can be justifiably viewed as a breakthrough in linguistics and philosophy. This chapter provides a tutorial about Grice’s ideas and an introduction to the thinking of later philosophers who have followed closely in his footsteps.
2.1 Grice on Conversational Implicatures The idea that our conversational purposes shape our understanding of utterances in conversation is a natural one. Take the exchange in (6). (6)
a. A : I’m out of gas. b. B : There’s a gas station around the corner. (A variant of Grice 1989: 32 Ex 1)
B’s utterance in (6) is naturally understood as a suggestion for A to get gas at a particular station. In proposing that the station is a good place to get gas, B seems to implicitly vouch that the station can be expected to be open, stocked, and staffed. Why don’t we just understand B’s utterance as the factual description that it seems to be on the surface? One obvious factor is our awareness of what these two interlocutors are doing. A has a problem and B is trying to help. It seems right that this background understanding of the interlocutors’ relationship, situation, and goals leads us to the interpretation we find. Perhaps, then, Grice suggests, we can fully explain the use B makes of his utterance in (6) by appealing to the straightforward conventional meaning, and additional non-conventional principles that describe interlocutors’ rational behavior in contributing to conversation. Grice offers a couple of defenses for drawing the line he does between conventional meaning and conversational interpretation. One is the feature he calls cancelability. Suppose B follows up the utterance in (6b) by immediately adding, “but I don’t think it’s open.” B’s contribution remains coherent. We perhaps still understand B as offering a proposal to A about a possible course of action, but B explicitly disavows any endorsement that the plan will work. In Grice’s words, the suggestion otherwise present in (6) that A can get gas at this place has been canceled. For Grice, this difference in interpretation is evidence for his hypothesis that rational principles rather than conventional meanings are at work. We naturally understand B as having different purposes in mentioning a gas station that he thinks is probably closed than he would have in mentioning one he thinks
FOR INDEXING
th e g r i ce a n f r a m ewor k 17 is probably open. This difference seems to explain why it is that we take the alternative utterances in contrasting ways. Importantly, because the interpretive contrast is due to rationality rather than to conventional meaning, we do not have to postulate new meanings for B’s words in the alternative utterances. In both cases, we can appeal to a simple linguistic meaning describing the location of a gas station. We do not expect conventional meaning to be cancelable in this way. Consider (7). (7)
a. John is a bachelor. b. But he is married.
It’s a consequence of the meaning of bachelor that John is not married. So to continue (7a) with (7b) creates a contradiction. To reconstruct a coherent overall interpretation, one is forced to assume that the speaker does not want us to understand bachelor or married in the normal way. Perhaps John’s slovenly housekeeping makes him a metaphorical bachelor; perhaps his obsessive passion for work counts as a metaphorical marriage. Either way, though, the conventional meaning remains active in trying to make sense of the speaker’s utterance. It is not canceled. A second reason to explain interpretive effects indirectly, through reasoning about speakers’ conversational purposes, is that it often allows us to think about the rules of language in simpler and more perspicuous ways. Here’s an example of this effect that Grice made famous. A number of theorists, notably Strawson (1950), have thought that the word and is ambiguous between temporal, causal, and listing readings. They base their conclusions on intuitions about what typical utterances of sentences like (8–10) contribute to a conversation. (8) (9) (10)
Lee got out of bed and got dressed. Chris ate a bad oyster and got sick. Kim was tall and Sandy was short.
These sentences clearly differ in the relationships we understand between the propositions they conjoin. Grice suggests, however, that this interpretive difference reflects our different understanding of the speaker’s purposes, rather than a different disambiguation of her linguistic meaning. That is, we recognize that the speaker in (8) is narrating a story, and thus describing successive events in temporal order. We recognize that the speaker in (9) is presenting an explanation for the unfolding of causally related events. We recognize that the speaker in (10) is describing a contrast that held at a particular moment in time. If this is right, there’s only one meaning for the English and on display in (8–10), with
FOR INDEXING
18
the g r icean fr amewor k
the familiar meaning of logical conjunction. The other interpretive effects, as Grice conceptualizes them, reflect our reasoning about the rational choices of the speaker. We return to this empirical claim in more detail in Chapter 7. But Grice’s view is clearly much more compelling than Strawson’s. Grice’s plan, then, is to develop an intuitive and appealing view: our conversational purposes and the rational organization of conversation work in consistent ways to shape our meanings and further our understanding of one another. The philosophical work comes in saying more about how this happens. To this explanatory end, surprisingly, Grice begins by offering but a single general “principle of conversation,” the Cooperative Principle, on which to base such reasoning: (11)
Cooperative Principle (CP): Speakers are to “make [their] conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which [they] are engaged.” (Grice 1989, 26)
Grice’s idea is that the CP is not just true; it’s mutually known among interlocutors. That means that the CP is operative in interlocutors’ reasoning whenever they try to take one another’s perspective into account. It therefore lets them agree on content that’s inferentially derivable from one another’s rationality. Grice sees the CP as fundamental. But, of course, the CP is somewhat programmatic. To help theorists spell out how the CP actually leads to an understanding of a contribution to a conversation that includes its CIs, Grice distinguishes among categories of more specific considerations a speaker must take into account in order to follow the CP, including considerations of truthfulness, informativeness, relevance, and clarity. These are his famous Maxims defined as below (Grice 1989: 26–7): (12) (13)
(14) (15)
Quality: Do not say what you believe to be false; do not say that for which you lack adequate evidence. Quantity: Make your contribution as informative as is required (for the current purposes of the exchange); do not make your contribution more informative than is required. Relation: Be relevant. Manner: Be perspicuous: avoid obscurity of expression; avoid ambiguity; be brief (avoid unnecessary prolixity); be orderly.
Quality (12) says that cooperative speakers must contribute information that other speakers can rely on to be accurate. Quantity (13) says that cooperative
FOR INDEXING
th e g r i ce a n f r a m ewor k 19 speakers must not give too much information, or too little, given the goals of the conversation. Relation (14) says that cooperative speakers must provide relevant information. And Manner (15) says that cooperative speakers must provide information in a format that is appropriate to the audience and to the circumstances of the exchange. Grice notes, by the way, that the CP might entail additional maxims that speakers follow. We return to this suggestion in section 2.2. Grice’s general concept of speaker cooperation can be characterized in terms of obedience to these maxims. Grice suggests that these maxims are “standardly (though not invariably) observed by participants in a talk exchange” (Grice 1989, 41). More precisely, it is common knowledge, Grice claims, that the CP and the four maxims provide standards for purposeful communication. Grice defends this idea by appeal to a particular view of the nature of conversation: “The use of language is one among a range of forms of rational activity” (Grice 1989, 341). So, in response to “a fundamental question about the Cooperative Principle and its attendant maxims, namely, what the basis is for the assumption which we seem to make, that talkers will in general . . . proceed in the manner that these principles prescribe” (Grice 1989, 28), he answers, “I would like to be able to think of the standard type of conversational practice . . . as something that it is reasonable for us to follow,” setting out the claims in (16–18). (16)
Certain goals are fundamental to conversation.
That is, conversations are not disconnected remarks. Rather, in conversation, interlocutors commit at least to share their perspectives, and in fact, prototypically engage in a process of inquiry intended to result in agreement on the answer to a question of shared practical interest. (17)
These goals can be accomplished only in conversational exchanges conducted in accordance with CP.
At each stage, the purpose or the direction of the conversation limits which discourse moves are appropriate. Discourse moves that violate the CP, or the particular maxims which describe how it applies, cannot contribute to achieving the purposes of the conversation. (18)
Therefore, all rational speakers must behave in accord with CP.
This is why the CP provides a fundamental constraint on utterance interpretation that speakers and their audiences can rely on. As part of his effort to ground his account in foundational considerations involving rationality, Grice demonstrates that each of the maxims has an
FOR INDEXING
20
the g r icean fr am ewor k
analogue in the non-linguistic domain, in other types of goal-oriented interactions. In any collaboration, we make commitments to working towards a mutually satisfactory outcome. These commitments constrain which actions we can rationally perform as part of our cooperative project, much as the maxims constrain what utterances we can rationally perform in a conversation. We present examples from Grice (1989: 28) along with our take on the operative maxims of practical collaboration: (19)
(20)
(21)
(22)
Quality. Collaborators should choose actions that further their joint project and that do not undermine it. “If I need sugar as an ingredient in the cake you are assisting me to make, I do not expect you to hand me salt.” Quantity. Collaborators should act to do precisely their full part in their joint project. “If you are assisting me to mend a car . . . if at a particular juncture I need four screws, I expect you to hand me four, rather than two or six.” Relation. Collaborators should respond to the immediate needs of the task. “If I am mixing the ingredients for a cake, I do not expect to be handed a good book, or even an oven cloth (though this might be an appropriate contribution at a later stage).” Manner. Collaborators should act efficiently and recognizably. “I expect a partner to make it clear what contribution he is making, and to execute his performance with reasonable dispatch.”
Grice understands the four conversational maxims as specializations of these more general maxims to conversation, and indeed, in Chapter 13, we shall see that subsequent research on collaboration has substantiated Grice’s views about the generality of such considerations. Grice posits that speakers can exploit the shared assumption that they follow the CP and its attendant maxims to add additional layers of meaning to their utterances. More precisely, Grice suggests that speakers often use utterances in ways that are compatible with the CP and the conversational maxims only under certain assumptions. These assumptions become CIs—that is, they are content that the speaker contributes indirectly to a conversation by using an utterance. Schematically, CP and the maxims function as follows: (23)
By saying p, the speaker conversationally implicates that q just in case she is presumed to be following the maxims or at least the Cooperative Principle; and the supposition that she is aware or thinks that q is required in order to maintain that she is following these maxims; and she
FOR INDEXING
th e g r i ce a n f r a m ewor k 21 thinks (and would expect her audience to think that she thinks) that it is within the competence of her audience to work out or grasp intuitively, that the supposition that q is required (Grice 1989, 30–1).2 So when A says (6a) I’m out of gas and B replies with, (6b) There’s a gas station around the corner, B thereby conversationally implicates that A can get gas there; for, otherwise, B would be violating the Maxim of Relation. If B doesn’t think that A can get gas at the station, then there is no reason to think that its existence offers A a solution to the problem at hand. Since B is presumably trying to be cooperative, it follows that his utterance makes sense only on the assumption that he believes the gas station is open. So it’s a CI by (23). Grice’s framework requires that CIs be calculable, or “capable of being worked out”; more precisely, he writes: (24)
the presence of a conversational implicature must be capable of being worked out; for even it if can in fact be intuitively grasped, unless the intuition is replaceable by an argument, the implicature (if present at all) will not count as a conversational implicature; it will be a conventional implicature. To work out that a particular conversational implicature is present, the hearer will rely on the following data; (1) the conventional meaning of the words used together with the identity of any references that may be involved; (2) the cooperative principle and its maxims; (3) the context, linguistic or otherwise, of the utterance; (4) other items of background knowledge and (5) the fact (or supposed fact) that all relevant items falling under the previous headings are available to both participants and both participants know or assume this to be the case. A general pattern for the working out of a conversational implicature might be given as follows: “He has said that p, there is no reason to suppose that
2 Grice’s definition awkwardly straddles the perspective of the speaker and the audience, a distinction that has proven to be important. Grice normally locates what the speaker means by an utterance in terms of intention. When we ask whether a speaker means something in particular by an utterance, what matters is what the speaker is aiming to do, and not whether it succeeds. We could describe CI the same way. That is, a speaker implicates something if the speaker has a suitable intention, envisioning that her audience reasons to an intended proposition from their presumption of her rationality. Communication, by contrast, requires success. A speaker communicates only if her audience gets the message. Many have argued that it is important to have a notion of implicature that requires either actual audience uptake, or an objectively reasonable expectation of audience uptake (Davis 1998; Saul 2002). Grice’s definition stops just short of this requirement, but in fact, in ordinary English discourse, it seems that while a speaker can mean something without getting the idea across, a speaker cannot imply something without getting the idea across. From this perspective, a CI necessarily involves an inference drawn by the audience—though as Bach (2005) emphasizes the reverse is certainly not the case. We believe that the fundamental notion must be the speaker’s presumptions about her audience, and we write accordingly. Since we do not consider misunderstanding of CIs or other problematic cases, we expect that this choice will not prove distracting to readers who are used to different ways of thinking (Hirschberg 1991, 17–19).
FOR INDEXING
22
the g r icean fr amewor k he is not observing the maxims, or at least the cooperative principle; he could not be doing this unless he thought that q; he knows (and knows that I know that he knows) that I can see that the supposition that he thinks that q is required; he has done nothing to stop me thinking that q; and so he has implicated that q.” (Grice 1975, 50)3
We interpret these passages rather more stringently than other commentators sometimes do, but that’s because we take so seriously the bridge between semantics and pragmatics that they seem to establish. First, in interpreting (24), it’s important to recognize that Grice uses this notion of calculability as a diagnostic to distinguish conversational implicatures from elements of interpretation that arise by convention. To play this role, the contextual and background information described in (24: 3–4) cannot make reference to further linguistic conventions, but should simply describe the interpretive effects of interlocutors’ factual knowledge about the discourse, the conversational situation, and the common-sense world. This means that (24) makes essential appeal to rationality. Cooperation does real work in underpinning inferences characterized by (24). Second, we understand (24) as characterizing a very strong kind of argument: arguments that an assumption on the part of the speaker is, as (23) puts it, required. We will see many examples throughout the book of probabilistic and uncertain reasoning about speakers. Such reasoning is no doubt important and commonplace. But, by sticking to the strong reading that (23) and (24) seem to have, we get a theory where CIs appeal to shared standards and therefore contribute public meanings in a transparently recognizable way. That lets (23) link utterances to content in a way that’s continuous with semantics as traditionally understood in philosophy. It’s in this combination of rational inference and public content that we locate the radical power of Grice’s ideas.4 3 See, also, his “Presupposition and Conversational Implicature,” where Grice emphasizes that a crucial test of the presence of a CI is that “one has to produce an account of how it could have arisen and why it is there” (Grice 1981, 187). 4 A number of critics have complained that Grice never endorsed a determinacy constraint on CIs. They sometimes cite the following passage from Grice in support of their challenge:
Since to calculate a [CI] is to calculate what has to be supposed in order to preserve the supposition that the [CP] is being observed, and since there may be various possible specific explanations, a list of which may be open, the [CI] in such cases will be a disjunction of such specific explanations; and if the list of these is open, the [CI] will have just the kind of indeterminacy that many actual [CIs] do in fact seem to possess. (Grice 1989: 39–40) Other pragmatic approaches definitely do embrace this indeterminacy, but our arguments will ultimately suggest that this move does not help the defender of general pragmatic principles. At the same time, this move comes with substantial costs, and we think the Gricean has much to gain from attempting to avoid them. Hence our reading of (23) and (24).
FOR INDEXING
th e g r i ce a n f r a m ewor k 23 We can illustrate these ideas by returning to the three cases presented in Chapter 1. In (25), the CI is that the speaker wants the hearer to give her French Toast: (25)
Can I have the French Toast?
This is a calculation from the maxims of Quality and Relation. The Maxim of Quality shows that the ordinary question meaning of her utterance is not a sufficient contribution to the conversation on its own; there’s little doubt that it would be possible for the speaker to have French Toast. However, as Searle (1975) observes, there is a natural associative link that suggests that a speaker who asks about a possibility may go on to request that it be brought about. The possibility is a precondition to the request, in the sense that the addressee can and will carry out the request only if it is possible for him to do so. Meanwhile, the Maxim of Relation reminds us that having food for each customer is a goal which the conversationalists here are likely to share. This confirms that the speaker must have intended the addressee to identify the request suggested by this associative link: in this way, the addressee derives that the speaker has not only asked about a possibility but has also indirectly made a request. In (26), we calculate the CI that the plunge followed the doubling by appeal to the Maxim of Manner. (26)
Oil prices doubled and demand for consumer goods plunged.
That maxim enjoins the speaker to be orderly, which includes describing events in the order that they occur. We readily recognize that this principle applies to establish a temporal relationship in the speaker’s discourse. Finally, in (27), we calculate the speaker’s CI that the handkerchief might not be red by appeal to the Maxim of Quantity. (27)
Well, it looked red.
The prior speaker has already raised the question of the color of the handkerchief and offered the apparently stronger claim that the handkerchief actually was red. That means that the speaker of (27) should have assented to the stronger claim if she could have. Thus, by not doing so, she implicates doubt about that claim. The derivations at play in (25–27) seem to be quite powerful and general. Though the CP and the framework of CIs establishes a link between the interpretations of specific utterances and the purposes and strategies of specific interlocutors and specific conversational settings, we can often calculate these CIs in a broad and schematic way. This is an important reason why linguists and philosophers, beginning with Grice, have focused on theories of generalized CIs, which seem to be present across a wide range of utterance types and
FOR INDEXING
24
the g r icean fr am ewor k
across a wide range of contexts. For example, (27) is an example of a diverse class of scalar implicatures which seem to follow simply from the speaker’s use of a weaker sentence where a stronger alternative is relevant and available to the speaker. Scalar implicatures seem to apply across a wide range of linguistic items that are ordered by logical relationships, including quantifiers, connectives, modals, and the like. See Chapter 8 for further discussion. As a final case study, let us return to the example of metaphor with which we opened this chapter. Here we discover an important further dimension of Grice’s theory. Applying the CP, for Grice, is not always just a matter of filling out or amplifying the ordinary literal meaning expressed by an utterance, as we find prototypically in inferences driven by the Maxim of Relation as in (6), the Maxim of Manner as in (26), or the Maxim of Quantity as in (27). Instead, applying the CP can sometimes require the audience to derive a replacement for the literal meaning of an apparently deviant utterance. The way Grice explains this is that maxims can be flouted. Remember that the job of interpretation, on Grice’s theory, is to reconstruct a speaker’s communicative intention. The rules of language give evidence of speakers’ intentions, because these rules are ultimately generalizations about what speakers tend to do with their utterances. But these rules admit exceptions, and so the audience needs to consider them in light of all the relevant information—including, of course, the constraints of the CP and the maxims. Normally, even when the ordinary meaning of an utterance creates a seemingly blatant violation of some maxim, the audience nevertheless assumes that the speaker is working to respect the various demands of the CP. The audience must reconstruct the alternative, cooperative purpose that the speaker must have had. Grice sees this as a matter of considering the preponderance of the evidence—using broader mechanisms of intention recognition to develop some new, creative understanding of the speaker that reconciles the CP with what the speaker actually said. Let’s consider how this theory applies to (5), Richard is a lion. The utterance flouts the Maxims of Quality and Relation. The literal meaning obviously abandons the standards of truth normally required by the CP: Richard is not a lion but rather a human being. Equally problematic is the irrelevance of the literal meaning: interlocutors’ interests in answering questions about Richard simply aren’t advanced in the usual case by addressing whether or not he’s some kind of great cat. Clearly, the addressee can reject the hypothesis that the speaker’s intended contribution to conversation involves the literal meaning. Instead, more open-ended intention recognition is required. In this case, we have suggested, following Searle (1979), that the speaker refers to lions here
FOR INDEXING
th e g r i ce a n f r a m ewor k 25 as a way of prompting his audience to bring to mind associated properties, perhaps fierceness, leadership, or pride. Since the speaker obviously expected the hearer to think of those properties while rejecting literal lionhood as a possible interpretation, we should think of the metaphorical meaning of the utterance as a conversational implicature to the effect that Richard has these associated properties. Importantly, even in these cases of flouting the maxims, all the ingredients— the CP, the maxims, the linguistic rules—remain common knowledge. Therefore, speakers can confidently use utterances that flout the maxims to convey their meanings indirectly and creatively, and addressees can readily recognize such indirection. The naturalness of mechanisms that repair apparently problematic utterances and recover sensible contributions to conversation is one of the most distinctive elements of the Gricean program.
2.2 Gricean Pragmatics Meets Formal Semantics Grice’s theory of CIs frees the content a speaker contributes to conversation from the conventional rules that link form to meaning in language. As we shall see in the next chapters, a wide range of researchers have accepted this key insight, while departing from Grice in his account of pragmatic principles and conversational inference. Even if we accept the Gricean picture wholeheartedly, however, we can hope to be more precise about the goals of conversations and the rational principles by which agents pursue those goals. In this section, we review the work of Stalnaker (1973, 1979) and Thomason (1990) as paradigmatic examples. Both outline more precise classes of CIs—including, notably, new cases of flouting, where interlocutors rely on new ways of revising linguistic meaning to speaker meanings that respect the CP. And both do so by combining mathematical models of the meanings of utterances from formal semantics with more precise philosophical conceptions of the organization of conversation. Together, their work epitomizes the ability of the Gricean framework to offer sophisticated defenses of philosophical views and linguistic theories.
2.2.1 Possible-Worlds Semantics and the Implicatures of Assertion Stalnaker invokes Grice’s theory of CIs to defend a bare-bones account of propositional content in semantics. Stalnaker’s view of propositions involves two main ideas with broad theoretical appeal. The first is that content is to be understood purely in terms of information about the world. When you learn that a proposition is true, you learn that the world is a certain way—this is the information that the proposition carries. Stalnaker’s claim is that this information
FOR INDEXING
26
the g r icean fr amewor k
completely specifies the proposition and its content. For Stalnaker, there can be no difference among propositions that carry the same information in the sense that they say the world is a certain way and would be true in exactly the same circumstances. Mathematically, this idea is realized in Stalnaker’s abstract identification of a proposition and a set of possible worlds. To make things concrete, consider (28). (28)
Aristotle taught Alexander.
The information in (28) simply distinguishes worlds where Aristotle taught Alexander from worlds where he did not. Stalnaker’s understanding of propositions as abstractions of information contrasts with a more traditional understanding of propositions, going back to Russell, as structured abstract objects. Structured propositions organize realworld objects, properties, and relations into complexes that reify claims about the world. For a Russellian, for example, the proposition expressed by (28) corresponds to a tuple specifying the entities Aristotle, the teaching relation, Alexander, and the past (or, if we follow the account of tense in Chapter 7, a contextually specified past interval). This object embodies the claim that this entity stands in this relation to that entity at that time. See Salmon (1986); Soames (1987); King (2007). The second ingredient of Stalnaker’s view is his acceptance of direct reference as an organizing principle in semantics. Direct reference is the doctrine that—except in certain exceptional cases—the meaning of a word or other atomic expressions in natural language is simply the individual, property, or relation that the expression refers to. For example, in (28), the meaning of Aristotle is just the individual; same for Alexander. We approach taught as the realization of two morphemes: teach plus the past tense: teach picks out the teaching relation; the past tense picks out the past—or, on the assumption that the past refers to a contextually specified interval, contributes just the specified interval. The alternative to direct reference is Frege’s doctrine of senses. A sense is an abstract object that determines reference but that embodies a particular way the mind can grasp an idea. (It’s common to make Frege’s idea of senses more concrete by modeling them as a kind of covert description, following Russell.) For Frege, synonymous terms can have the same reference but different senses. Famously, for example, both Hesperus and Phosphorus refer to the planet Venus, but Hesperus has a sense that identifies Venus as the evening star and Phosphorus has a sense that identifies Venus as the morning star.
FOR INDEXING
th e g r i ce a n f r a m ewor k 27 However, Kripke (1972) offered arguments against the doctrine of senses for names and natural kind terms that many philosophers regarded as decisive. For Kripke, thinking of the meaning of such atomic expressions as being descriptive is conceptually problematic. Moreover, it seems that the way we use language is not sufficient to determine useful senses for expressions over and above their referents. Anyway, when we consider the interpretations of utterances, the direct reference view does a better job. Again, that has not prevented philosophers from continuing to defend alternatives to the direct reference view. But Stalnaker is again in good company. Stalnaker’s two principles have an unfortunate consequence however: Stalnaker’s propositions are very coarse-grained. One way the difficulty surfaces is that the theory assigns the same content to linguistic expressions that intuitively seem very different. Take (29) and (30). (29)
Hesperus is Phosphorus.
(30)
Hesperus is Hesperus.
According to direct reference, both Hesperus and Phosphorus contribute references to the planet Venus to the truth-conditions of these sentences. It follows that both (29) and (30) express the obvious necessary proposition that the planet Venus is self-identical. By contrast, we naturally understand (29) as the substantive report of an interesting discovery. Another problematic consequence is that there are intuitively salient aspects of meaning that the theory prevents us from capturing. Consider (31) and (32). (31)
Chris knows that Sandy has left.
(32)
Chris doesn’t know that Sandy has left.
When a speaker uses (31) or (32), our typical sense is that the speaker takes it for granted that Sandy has left. The question at issue is whether Chris knows the fact or not. It’s natural to expect that this effect is part of what the verb know means. For example, if we adopt a framework for sentence meaning like Discourse Representation Theory (Kamp and Reyle 1993), where meanings are thought of as structured, we can capture the effect by treating the verb know as a presupposition trigger: whenever know is used, its meaning enriches the interpretation of the utterance with the constraint that the speaker takes the sentential complement to be true. We’ll return to this conception of presupposition in Chapter 7. Stalnaker’s understanding of content has no room for such dimensions of meaning, however. For Stalnaker, A knows that P just in case A
FOR INDEXING
28
the g r icean fr amewor k
stands in a particular relation to the proposition P. Such a proposition, like any other, will simply indicate what the world must be like for A to be so related to P. Stalnaker’s appeal to Grice provides an elegant and powerful tool to defend his coarse-grained account of propositions against such apparent problems. Stalnaker begins by revisiting Grice’s understanding of communication in light of the precise understanding of content Stalnaker has developed. For example, consider the case of interlocutors who are attempting to agree on the answer to a question. Stalnaker invites us to characterize the initial information state of the interlocutors as a set of worlds which are open possibilities as far as their mutual knowledge allows. This is called the context set. Insofar as the interlocutors are open to many alternative answers to the question at issue, the context set contains alternative worlds where the question is answered in different ways. As the interlocutors exchange information, however, their contributions eliminate incompatible possibilities from the context set. Eventually, they narrow the options down to a context set where the question at hand finds a single determinate answer. Stalnaker uses this model of inquiry to give more precise characterizations of the constraints of rational interaction than Grice’s Cooperative Principle and maxims. Here’s a simple case that Stalnaker considers in detail. Cooperative speakers should use terms whose meanings their interlocutors know. If interlocutors don’t know what proposition a speaker is offering to the conversation, they cannot register the information she has provided and so cannot track the state of the conversation. Stalnaker cashes this idea out in terms of his mathematical model of inquiry. Suppose interlocutors do have shared knowledge of the meaning of a sentence. Then there can be no uncertainty across the context set in how the relevant rules of language apply to the sentence, and thus no variation across the context set in the proposition the sentence expresses. Cooperative speakers should strive for this. Formally, this condition is captured by Stalnaker’s Principle 3, given in (33): (33)
The same proposition is expressed relative to each possible world in the context set. (Stalnaker 1999, 88)
In other words, in rational inquiry, interlocutors should always know what proposition the speaker is asserting. That way, interlocutors can tell, for each world in the context set, whether that world is compatible with the information advanced to the conversation thus far. They can continue to agree on the issues that are settled and the issues that are open. In the spirit of Stalnaker’s fundamentally rational motivation for this principle, we will call it the Maxim of Clarity.
FOR INDEXING
th e g r i ce a n f r a m ewor k 29 The Maxim of Clarity provides the basis for Stalnaker’s explanations of the intuitive meanings of (29) and (30). Perhaps surprisingly, the Maxim of Clarity predicts that (29) Hesperus is Phosphorus is pragmatically anomalous in just those circumstances where it seems to make an informative contribution. Let’s focus on a particular case to make the discussion concrete. Suppose that the hearer knows that the word Phosphorus names the planet Venus, but is unaware of what planet the word Hesperus refers to. This is exactly the kind of situation in which we’d expect (29) to carry useful information. Stalnaker’s technical system models the example as follows: because of interlocutors’ ignorance about the external circumstances that determine the reference of the terms in their language, there are worlds in the context set where the word Hesperus refers to something other than Venus. Remember, meaning is just reference on the direct reference theory. So in these cases the speaker has used an expression whose meaning the hearer does not know. These cases also leave the hearer uncertain about which proposition Hesperus is Phosphorus expresses. In the simple situation we’re describing, we can capture the hearer’s uncertainty as follows. Perhaps the word Hesperus names the planet Venus, in which case Hesperus is Phosphorus expresses a necessarily true proposition. But perhaps the word Hesperus names the planet Mars, in which case Hesperus is Phosphorus expresses a necessarily false one. Thus, because the hearer does not know what the word Hesperus refers to, the hearer cannot identify a definite proposition that the speaker is contributing by uttering Hesperus is Phosphorus. Put another way, a speaker who utters (29) is flouting the Maxim of Clarity. She is using an utterance that obviously violates the cooperative requirement not to use expressions whose meanings her audience does not know. By Gricean reasoning, she must intend to contribute something other than the literal meaning of what she says. Stalnaker proposes that interlocutors resolve the incongruity by a Gricean calculation. Again, we sketch the calculation in what follows, assuming interlocutors agree on the reference of Phosphorus but are uncertain about the reference of Hesperus. The speaker has used an utterance whose literal meaning her interlocutor cannot identify, since he is uncertain about what the word Hesperus refers to. The speaker must have some ulterior motive. The speaker obviously means to call attention to the problematic status of the audience’s knowledge, since her utterance has done exactly this. But since the speaker is cooperative, we expect that she is not merely announcing the problem but actually resolving it. In other words, she must intend to offer a contribution that eliminates the worlds in the context set where the reference of Hesperus differs from its actual reference. And it’s straightforward to determine from Hesperus is Phosphorus just what
FOR INDEXING
30
the g r icean fr am ewor k
worlds are the ones that should stay: those where the circumstances surrounding reference and meaning of the interlocutors’ language are as they actually are, and the word Hesperus is a name for the planet Venus. Those where the linguistic facts are different and the word Hesperus names something else should go. Note that there is no uncertainty about what the speaker is doing on Stalnaker’s story. This is in keeping with our understanding of Grice’s (23) and (24). There is no way that a speaker can cooperatively use a tautology for its literal meaning. With this meaning off the table, Stalnaker’s logic tells us exactly what proposition she contributes instead. In addition to the Maxim of Clarity, Stalnaker motivates additional constraints on rational inquiry. His suggestions here are somewhat sketchy, but we don’t see any philosophical danger in amplifying on them somewhat to bring out their close relationship to Grice. The key idea is that it’s a bad strategy to attempt to settle a complex issue directly. Doing so can cost effort. There’s no reason to resolve a difficult question first, for example, when another matter we could readily settle beforehand might well make that question moot. At the same time, interlocutors must inevitably address the basic underlying issues in order anyway. They must clarify the questions involved, resolve disagreements, and reach a reliable shared understanding of how things are. Thus, when interlocutors do ultimately aim to settle a complex issue, they should raise constituent questions one by one, and do so in the order they expect to best advance their interests. Call this need to choose efficient strategies in conversation the Maxim of Progress. For Stalnaker, to ask whether somebody knows some proposition is naturally understood as a complex question. It implicitly involves settling whether the proposition is true, whether the subject has the relevant belief, whether the belief is appropriately justified—the whole nine yards. By the Maxim of Progress, it’s not rational to address this question from scratch, out of the blue. As Stalnaker puts it: (34)
Suppose a speaker were to assert that x knows that P in a context where the truth of P is in doubt or dispute. He would be saying in one breath something that could be challenged in two different ways. He would be leaving unclear whether his main point was to make a claim about the truth of P, or to make a claim about the epistemic situation of x (the knower), and thus leaving unclear what direction he intended or expected the conversation to take. Thus, given what “x knows that P” means, and given that people normally want to communicate in an orderly way, and normally have some purpose in mind, it would be
FOR INDEXING
th e g r i ce a n f r a m ewor k 31 unreasonable to assert that x knows that P in such a context. (Stalnaker 1999, 55) Since speakers are normally cooperative, when a speaker uses (31) or (32) in conversation, it’s because an appropriate range of subsidiary questions have already been answered in such a way that makes it possible to address in a simple and expeditious way whether the complex knowledge claim is true or false. We have every reason to assume that this prediction is satisfied in the hypothetical context which we imagine as readers of (31) or (32). What answers must the conversation have already produced in this case? Amplifying Stalnaker’s remarks, it would be a simple matter to settle the truth of the propositional complement. And if inquiry revealed that the complement is not true, we could dispense with further discussion. Thus we can derive the suggestion, common to (31) or (32), that Sandy has in fact left, as a side-effect of understanding the speaker’s communicative intention in light of the Cooperative Principle and the maxims. And again, the CI we derive is a precise proposition that piggybacks in a constrained way on the structure of the speaker’s utterance and the correlative assumptions we must make about the speaker’s choices in using it. Stalnaker’s work provides an object lesson in the robustness of Grice’s views and the sophistication with which they can support further philosophical investigations. In particular, as we sharpen our understanding of content, context, and inquiry, we naturally refine our understanding of what’s required to use meanings cooperatively in conversation. These refinements let us account for intuitive uses of utterances in clear new ways.
2.2.2 Presupposition Triggers and the Implicatures of Accommodation We can refine our understanding of cooperative conversation not only by refining our understanding of meaning and communication, but also by refining our understanding of rationality and cooperation itself. This is the approach of Thomason (1990). As we have seen, Grice expected to find close analogies between conversation and other kinds of collaboration. But he explored these analogies only in a loose, intuitive way. By contrast, subsequent research, particularly in artificial intelligence (AI), has sought to develop more precise models of collaboration, both to explain patterns of human collaboration in detail and to build computer systems with richer collaborative skills. Thomason uses these general models of collaboration to shed light on the relationship between linguistic meaning and utterance interpretation.
FOR INDEXING
32
the g r icean fr amewor k
Thomason, like Stalnaker, starts from an independently motivated theoretical challenge. Thomason’s challenge is the problem of informative presuppositions. The problem is perhaps best illustrated through the case of definite referring expressions, like my sister in Stalnaker’s example (35). (35)
I have to take my sister to the airport. (Stalnaker 2002: 709)
You might have thought the speaker could only utter (35) cooperatively if she and her audience were already mutually aware that she had a sister. In fact, given the rational constraints on inquiry advanced by Stalnaker within the Gricean tradition, speakers should never presuppose something unless it’s part of the common ground in the conversation. Otherwise, the speaker is making an assumption that might not apply; there are worlds in the context set where the presupposition is false. It’s unclear what import the speaker’s claim has for such worlds: is she ruling them out, or making a claim that does not apply to them? In fact, however, utterances such as (35) are used effortlessly all of the time without any special action to establish their presuppositions. The presuppositions are taken up unproblematically as true. The problem is to explain why cooperative conversation allows speakers to do this.5 In an influential paper, Lewis (1979) proposes to account for such examples as an instance of a general rule of accommodation for presupposition in conversation. (36)
If at time t something is said that requires presupposition P to be acceptable, and if P is not presupposed just before t, then—ceteris paribus and within certain limits—presupposition P comes into existence at t. (Lewis 1979: 340)
For Lewis, the generalization in (36) reports the empirical discovery of a linguistic rule—a fact either about particular natural languages or a universal but contingent constraint on all possible natural languages. So understood, the rule is just a description of the phenomenon. In effect, one way to read it is simply as contradicting the claim that presuppositions must be part of the common ground. This is a direct challenge to Stalnaker’s suggestion that presupposition is ultimately a pragmatic phenomenon involving speaker intentions and rationality. 5 Note that this problem is independent of how the speaker’s presupposition actually arises. On theories like Strawson’s (1950), a use of an expression like my sister has a linguistic meaning that triggers the presupposition that the speaker has a sister. Alternatively, we can follow Russell’s (1905) account of descriptions, but adopt Kripke’s (1978) Gricean suggestion that speakers simply use such expressions referentially in order to refer to a particular person. In either case, the speaker seems to be committed to more than the common ground warrants. So, we need to explain why the speaker’s presupposition in (35) is unproblematic.
FOR INDEXING
th e g r i ce a n f r a m ewor k 33 Thomason offers his refinements of Grice as a way to defuse the puzzles associated with Lewis’s account of accommodation. Thomason reconstructs the rule of accommodation not as a rule of language but as a general principle of collaboration. Moreover, Thomason argues that this collaborative principle is straightforwardly compatible with rational constraints on presupposition. It simply shows—like Grice’s mechanism of flouting the maxims—that social norms can always be exploited, as well as followed. Thomason starts from a comprehensive model of collaboration in which recognizing others’ intentions plays a fundamental role; we will discuss this model further in Chapter 13. But here’s one of its key principles: whenever somebody acts in a collaboration, her collaborators aim to recognize her intention, and she herself aims to make her intention recognizable. Accommodation, Thomason argues, is another general feature of the model. Accommodation in this general sense describes the actions people take to eliminate obstacles to the recognized intentions of their collaborators. In other words, accommodation is intervening to help. Because collaborators work to keep their intentions public, they are normally in a position to accommodate one another, and it would be both uncooperative and irrational for them to fail to do so. In conversation, accommodation includes making fixes to the state of the conversation in order to rectify potential problems in others’ recognized communicative intentions. This is precisely the rule of accommodation, as Lewis describes it. Accommodation thus provides a mechanism for explaining particular cases of Gricean CIs, via the following reasoning. The speaker presents an utterance U . The audience’s intention recognition delivers a particular plan or intention P as an explanation of U , which is associated with what might at first appear to be the most plausible interpretation of U . However, the audience finds that plan P does not, on the face of it, respect the norms of rationality: The plan only works if a certain condition C is part of the context set, but C is not yet established in the conversation. The audience does not, however conclude that the speaker is being irrational. The audience concludes that the speaker is flouting the normal rules of conversation. The speaker is counting on the audience to accommodate. In particular, it will be obvious to everyone that condition C is necessary in order to continue the conversation along the way that the speaker envisions. It will be obvious that the audience can just accept C, adding it to the context set, and allow the speaker and the audience proceed as though it had always been there. And it will be obvious to all that this indirection—this ulterior plan P ′ , to use P to trigger accommodation and get across condition C—is what the speaker really intended.
FOR INDEXING
34
the g r icean fr amewor k
Let us return to (35), I have to pick up my sister at the airport, to illustrate the idea. Initially, the audience does not know the speaker has a sister, but it’s easy enough to see that the speaker is committed to having one. So the audience just adjusts things as needed, and assumes the speaker does have a sister. Thereafter the conversation moves forward in a perfectly normal way. The speaker’s sister might not have started out as part of the context set, but she winds up there because the audience accommodates the speaker. Afterwards, the speaker and the audience can all take for granted that the speaker has a sister. This reasoning is strikingly similar to the kind of reasoning Grice articulates in his definition of CI. The speaker gets across that she has a sister not by asserting it, but just by making it clear that she is presuming she has a sister, that she expects you to recognize this assumption, and that she expects you to take it up. Unlike Grice’s theory, however, there are no potentially conflicting maxims; there are no appeals to indefinite extra assumptions. You can, in Thomason’s theory, actually work out where this information comes from, based on Thomason’s model of cooperation, his model of context, and linguistic meaning. Indeed, the specificity of Thomason’s explanation in these cases delivers a precise implicated proposition that must be accommodated. The result is that the speaker gets some assumption across in just the way that Grice characterized CI, but without having to appeal to the particular logic that Grice had in mind for calculating CIs. The cases where this sort of explanation is most compelling are those that bridge relevance and presupposition and the coherence of discourse;6 we can explain them within Thomason’s framework much more precisely than we could within Grice’s framework. For example, imagine the following discourse where A utters (37a), and B replies with (37b): (37)
a. Did anyone in the class pass? b. Chris passed.
B’s utterance of (37b) only makes sense if Chris is in the class. How can we calculate this based on a model of collaboration? Thomason suggests that such implicatures are “set in motion by the discourse expectation that the normal response to a question should help to answer the question” (Thomason 1990: 355). We have seen that in any collaboration there are tasks under way that collaborators expect to complete before moving on. Moreover, as part of making their intentions recognizable, collaborators must flag their intentions explicitly when they are not doing what’s expected of them. B’s utterance of (37b) isn’t flagged as exceptional; it doesn’t override the default, and so it must 6 Thomason’s example is responding to the question Where’s Harry? with the answer He’s sick.
FOR INDEXING
th e g r i ce a n f r a m ewor k 35 be an answer to the question posed by (37a), and in fact, if we assume Chris is in the class, B’s utterance of (37b) is a straightforward exact answer. There’s good reason to expect that a plan recognition module or operation will naturally deliver a plan in which B takes for granted that Chris is in the class, and has formulated an utterance based on that information with a goal of answering the A’s question. But the audience might not have known that Chris is in the class. That’s not a problem; they discover that the assumption is necessary, and so, they accommodate it. Accommodation, then, is a way of calculating a kind of CI. Something related happens in the relevance example (6) There’s a gas station around the corner. Here you are to imagine that the intention recognition operation is naturally going to deliver an interpretation where B wants to help A by providing a solution to A’s problem. The context is collaborative, and we have reason to expect, other things being equal, that the interlocutors will take up one another’s domain goals, that is, their goals for accomplishing tasks in the world, in order to smooth their interactions with one another and get things done together. So when A’s utterance raises the question of where A can get some gas, we expect B to help out. If B were going to depart from what’s expected of him, then he would have to show A that he was doing so, and he doesn’t. In fact, all A needs to assume is that the gas station around the corner is open, and that B is aware of this fact; and then, A immediately sees how B is providing a plan that does resolve A’s problem. So A concludes that the station must be open. A doesn’t know this ahead of time, but can effortlessly accommodate it. Here, again, accommodation is a way of calculating the CI that the gas station is open without appealing to maxims, but, rather, by appealing to the mechanisms of intention recognition and collaboration, as Thomason articulates them. In short, Thomason’s suggestion, then, is that in order to pursue the Gricean program further, and in order to explore the range of inferences that can be explained through the Cooperative Principle, we need to think about collaboration, intention recognition, and general rationality much more precisely. One suggestion is that computational approaches, inspired by work on the philosophy of agency, provide the tools to meet this goal. We will return to philosophical and computational approaches to intention and agency in detail in Chapter 13. Precisely for the methodological reasons Thomason emphasizes, we shall see that capturing Thomason’s insights about accommodation involves very different commitments from endorsing a general Gricean theory of CIs.
FOR INDEXING
36
the g r icean fr amewor k
2.3 Grice and his Impact Grice’s proposals stick close to our intuitions, but their implications are surprisingly far-reaching. In particular, as Stalnaker’s and Thomason’s efforts show, Grice’s framework reveals deep interconnections that relate the philosophy of language to metaphysics and the philosophy of mind—even as it supports and rewards close engagement with the formal proposals of linguists and computer scientists. As Neale (1992: 509) sums it up, Grice’s work “lies at the center of research on the semantics–pragmatics distinction and shapes much discussion of the relationship between language and mind.” Grice is most famous for his general interpretive principles—the Maxims of Quality, Quantity, Relation, and Manner. Nevertheless, it is wrong to think of the maxims as the key feature of Grice’s approach. The maxims are heuristic, and up for grabs. Stalnaker and Thomason don’t hesitate to revise them. Rather, it is the features that are conserved in Stalnaker and Thomason that capture the underlying perspective of Gricean theory. We think three claims give the essentials of the view. First is the idea that meaning is intimately intertwined with interlocutors’ intentions in contributing to conversation, so that there is a special place at the interface of semantics and pragmatics for general principles of rationality—principles that describe rational considerations that make one utterance a better move than another to achieve interlocutors’ purposes in a particular context. These views refer key ingredients of the content of utterances in conversation to these principles, rather than to the rules of language. Second is the idea that contributions to conversation have an analogous status, whether they follow from principles of rationality or the conventions of meaning. Part of this, we have emphasized, is the view that pragmatic reasoning can determine public, shared meanings on a par with the conventional meanings of sentences. The Gricean uses conversational implicature, not literal meaning, to derive the courage attributed to Richard in an utterance of (5) Richard is a lion, the identification expressed in an utterance of (29) Hesperus is Phosphorus, and the existence of a sister in an utterance of (35) I have to take my sister to the airport. But in all these cases the conversational implicature is something that the speaker contributes to the conversation. In fact, in such cases, the literal meaning can be incidental to the speaker’s overriding communicative intention. Last is the idea that the principles of rationality distill normative considerations. They are justified just by what makes sense. But interlocutors do not follow them blindly. They are open to exploitation. We can readily flout the maxims and accommodate one another—and this social indirection explains speakers’
FOR INDEXING
th e g r i ce a n f r a m ewor k 37 theoretically surprising but nevertheless frequent use of utterances such as (5), (29), or (35) with apparently deviant literal meanings. Researchers continue to pursue a wide range of work that shares these assumptions, and develops them further in important ways. One direction attempts to strengthen the formal foundations of Gricean reasoning. Researchers in AI have appealed to computational models of planning and action to characterize the inferential mechanisms underwriting CIs. The close connection between AI techniques and philosophical models of agency enables these models to realize Gricean principles in fairly transparent ways. However, since they give a consistent and complete specification of interlocutors’ decision-making in conversation, the models are much more precise and fine-grained than Grice’s provisional account in terms of the CP and the maxims. Early work along this line includes Cohen and Perrault (1979); Allen and Perrault (1980) and Perrault and Allen (1980), but subsequent work by Litman and Allen (1990a), and Grosz and Sidner (1986, 1990) was particularly influential in linking AI models of collaboration to empirical accounts of conversation; Stone (2004a) offers a more recent overview. An alternative approach is to use economic models of strategic interaction among agents, particularly game theory. In game theory, we can model complex decisions where ideal choices would take into account agents’ uncertainty about the world, agents’ preferences for different possible outcomes, and agents’ dependence on one another. A complete game-theoretic formalization lets us associate an observed message from a rational agent with an indirect meaning: the message must have been the best choice for its sender. This characterization of implicated meaning is as precise as the specification it starts from. Game-theoretic pragmatics reconstructs Gricean implicature in terms that emphasize the information of decision-makers and abstract away from reasoning mechanisms. Nevertheless, in key respects, the account matches communicative decisions as characterized by Grice. For example, in the right games, we get an indirect meaning that reinterprets a statement as a suggestion (Parikh 1992); in other games, we get an indirect meaning for partial answers that reinterprets them as exhaustive (van Rooij and de Jager 2012). Clark (2011) offers an overview of the issues involved and a survey of recent research. A complementary research direction aims to develop a broader and more systematic perspective on CIs by adducing relevant generalizations about people’s contexts, utterances, and goals in conversation. For example, Roberts (2012) maps out the rational dynamics of inquiry in more detail, characterizing moves that make progress in conversation and using the characterization to describe the implicatures of a diverse range of linguistic forms. Similarly, Simons (2001,
FOR INDEXING
38
the g r icean fr amewor k
2003) uses a more detailed characterization of the rational principles involved in communication to give a more precise explanation of how a speakers’ presuppositions might be calculated from the informational content of their contributions. And Geurts (2010) argues that many apparent puzzles in Gricean theories of implicature can be defused with more accurate statements of the choices speakers must make in selecting the right utterance to meet their communicative goals from among the possibilities provided by their language. Space prevents us from offering a detailed treatment of such work. As we emphasized from the outset, our goal is philosophical. These different theories are developed as part of an interdisciplinary effort to understand the role of meaning in conversation. To the extent that they offer useful empirical insights and descriptive frameworks, we can endorse them—given an appropriate understanding of the true status of their theoretical constructs. On the other hand, insofar as they share Grice’s broad assumptions, these approaches are subsumed under our more general philosophical critiques. However, we shall also see, in the rest of Part I of the book, that there are many schools of pragmatic theory that claim Grice as an inspiration but do not follow Grice in these three key claims. For example, in Chapter 3, we will consider a range of frameworks that treat pragmatic principles as aspects of speakers’ conventional understanding of language rather than as consequences of general rationality. These theories thus explain the bread-and-butter CIs of (25), (26), and (27) as fundamentally linguistic judgments. In Chapter 4, meanwhile, we will consider a range of frameworks with a different conception of the general rational inference that underlies implicature—a conception that appeals to speakers’ actual psychology and their actual use of utterances. Here implicatures are explained empirically, in terms of psychological mechanisms for conversational reasoning, not from speakers’ intentional decisions to respect or depart from idealized norms. We shall see that psychological frameworks cannot appeal to the kind of creative flouting of the CP that Gricean theorists invoke to explain utterances such as (5), (29), or (35). In this sense, the status of the Gricean framework is more comprehensively contested than researchers’ universal affirmation of their debts to Grice might suggest. We will assess Grice’s three principles systematically in Part II and Part III of the book, in the context of the full range of theoretical alternatives. For now, let us close by briefly foreshadowing where exactly we will agree and disagree with Grice. We will reject Grice’s central claim that contributions have the same status in conversation no matter how they are contributed. In particular, we reject flouting and exploitation as general social mechanisms that suffice to derive specific contents for otherwise deviant utterances. An apparent violation
FOR INDEXING
th e g r i ce a n f r a m ewor k 39 of the CP can provide a reliable cue that the literal meaning is not intended, but it cannot explain the actual force of the speaker’s utterance. To explain that, we will suggest, we need to develop specific accounts of the diverse range of linguistic practices through which speakers invite interlocutors to engage in specific creative ways with the content the speaker has offered. Any conclusions the audience thereby discovers are implicit and tentative suggestions, rather than transparent, public contributions. A wide range of alleged CIs remain, of course. Generally, in these cases, we agree with Grice that the speakers normally would succeed in contributing implicated content with such utterances. In fact, in many examples of alleged CIs in the literature, such as those in (25), (26), and (27), we are struck by how precise speakers seem to be in the “extra” meaning they are getting across with their utterances, and how essential the uptake of this “extra” meaning is to their audience’s understanding. However, we will deny that that such inferences can in fact be calculated from what the speaker says by inference from CP or its four maxims, that is, from facts pertaining to the nature of human interaction. Instead, we will emphasize the crucial role of conventional rules of language in delivering these alleged CIs.
FOR INDEXING
3 The Linguistic Turn
As we have seen in Chapter 2, Grice’s pioneering work made a case for a pervasive gap between semantics and pragmatics. Much of our understanding of utterances in context, the view suggests, arises from the interaction between conventional meaning, as specified by the rules of language, and the application of general interpretive principles. In particular, Grice and his close followers supposed that conversation is governed by norms of collaboration, and argued that the contributions speakers make to conversation with their utterances are derived in large part from these rational constraints. As we saw with the work of Stalnaker (1973, 1979) and Thomason (1990), it’s natural to develop Grice’s ideas by refining our understanding of collaboration. Another natural direction, however, is to attempt to redescribe interpretive effects directly as a function of bare-bones meanings and further principles that shape interpretation. On this approach, language itself gives ample evidence for an interpretive architecture that distinguishes between semantics and pragmatics. Moreover, language itself can show us what interpretive principles we need in pragmatics and how those principles work to supplement and transform semantic meanings. In this chapter, we focus on one influential thread of research within this broadly Gricean tradition, epitomized by the work of Horn (1984, 1989) and Levinson (2000). Both researchers wind up developing a linguistic reinterpretation of Gricean theory and its distinction between conventional semantics and inferential pragmatics. In particular, for Horn (1984, 1989), the domain of application of Gricean principles has to be suitably restricted, by linguistic rules, in order to give an accurate and explanatory account of the distribution of CIs. For Levinson (2000), meanwhile, general interpretive principles shape utterance meanings in a much more general way than merely rational considerations would dictate. That means the interpretive principles themselves must figure in the language faculty. Horn (1984, 1989) and Levinson (2000) both back
FOR INDEXING
t h e l i n g u i s t i c tu r n 41 up their ideas with empirical arguments for viewing many CIs as linguistic in origin. Thus, both argue that the study of CIs reveals previously unrecognized or unexplained rules of language as well as constraints on what the possible rules of language are. Although Horn and Levinson offer a new account of how many CIs arise, their analyses and derivations are broadly similar in form to Grice’s. They also agree with Grice that rationality is essential for understanding why CIs exist in language. For Horn and Levinson, pragmatic principles determine one possible solution to the problems inherent in systems of communication. Pragmatics therefore affords rationalistic or functional explanations at the level of language evolution or language change, even though individual speakers are not compelled to respect the principles purely on the grounds of rationality or the Cooperative Principle. These affinities with Grice explain why these approaches are generally called neo-Gricean. A key contribution of the neo-Gricean program, in adapting Grice in these ways, has been to broaden the range of phenomena that naturally seem to group together under the heading of CIs. Reviewing the neo-Gricean arguments and analyses in detail will give us an instructive opportunity to explore the diversity of apparent pragmatic effects—and the challenge of accounting for them all through general principles of rationality. We begin with Horn’s approach in section 3.1, and continue with Levinson’s in section 3.2. Section 3.3 takes stock, by outlining the linguistic and philosophical impact of these and other neo-Gricean proposals.
3.1 The Linguistics of Principles in Conflict: Horn Horn takes Grice to have established a fundamental link between utterance interpretation and a theory of efficient communication. CIs arise because interlocutors can assume that utterances capture just what’s essential for their communicative purposes. For Grice, this efficiency is a matter of individual speakers’ rational choices about what they will say in contributing to a cooperative conversation. For Horn, by contrast, this efficiency is a consequence of the rules of language, which themselves strike a balance between the sometimes conflicting interests of the speaker and of the hearer. Horn motivates his rule-governed, linguistic conception of efficient communication by abstracting Grice’s maxims through two competing principles—the Q principle and the R principle: (38) (39)
Q: Say as much as you can (given R). R: Say no more than you must (given Q). (Horn 1984: 13 Ex 1a and 1b)
FOR INDEXING
42
the linguist ic tu r n
These two principles highlight the somewhat different standards of communicative success that apply to speakers and their audiences in conversation. A hearer needs information; for an utterance to be of use to him it has to provide this information. The Q principle, then, shows what makes for a good utterance for a hearer. (40)
The hearer wants the speaker to say as much as she can.
On the other hand, it costs the speaker effort to plan and perform her utterances; the speaker wants to get away with saying as little as possible. This leads to the R principle that captures the speaker’s interests: (41)
The speaker wants to say no more than what she has to.
For Horn, any communicative exchange is a balancing act between these two opposing principles. Efficient communication is a manifestation of the interlocutors’ shared observance of the one particular balance that’s codified in the rules of their common language. Horn’s Q and R principles, like Grice’s Cooperative Principle, promise to explain richly contextualized speaker meanings while postulating only a simple, logical semantics for the utterances speakers use. Moreover, Horn’s principles must be universal, just like Grice’s, since the operation of the language faculty and the pressures that shape speakers’ and hearers’ behavior are the same across all human communities. Horn’s discussion largely presupposes this broadly Gricean role for pragmatics in simplifying semantics and explaining prevalent patterns of interpretation. Indeed, Horn’s Q and R principles afford many of the same kinds of explanations as Grice’s Maxims. The Q principle corresponds to the Maxim of Quantity. Its effects are typified by so-called scalar implicatures (as introduced in Chapter 2 in connection with example 27 and explored further in Chapter 8). As an illustration, contrast the interpretations of (42) and (43). (42) (43)
You ate some of the cookies. (Horn 1984: 14 Ex 2b) You ate all of the cookies.
When a speaker uses (42), she suggests that her addressee did not eat all of the cookies. The idea is that in saying he ate some of the cookies, the speaker has given as much information as she has about the situation; she’s said as much as she can. So she must not have been able to make stronger statements, for example because she knows them to be false. The alternative (43) is such a statement, so, by the Q principle, since the speaker didn’t say it, she must not have been able
FOR INDEXING
t h e l i n g u i s t i c tu r n 43 to (for example, because it’s not true). That’s a sketch of an explanation of scalar implicatures in terms of Horn’s Q principle. The R principle, meanwhile, is Horn’s way of getting at the Maxims of Relation, Manner, and Quality. For Horn, this is typified by pragmatic strengthening, as in the relevance implicature of (6), repeated here: (6)
a.
A : I’m out of gas.
b.
B : There’s a gas station around the corner.
For Grice and Thomason, the CP lets us recognize that (6b) is intended as a solution to A’s problem. That intention underwrites the CI that the gas station is open: If the gas station is not open, A cannot get gas there. The explanation works out slightly differently in Horn’s framework. The key idea is that what the speaker must say in such circumstances is just what the hearer needs to figure out what she has in mind. In this case, just by the speaker’s mentioning the gas station, you clue in to the possibility that she wants you to go there, expects it will be open, and expects that you’ll be able to purchase gas there. Since this is all the speaker has to say to get this idea across, by the R principle, this is all the speaker should say. By reinterpreting (6b) in light of the R principle, you realize that you can take the utterance as being stronger than what the speaker literally says. Horn’s reinterpretation of Grice is designed to make sense of the idea of ambiguity in pragmatics. Grice’s CP applies without qualification; every rational speaker should respect it. However, Horn observes that interpretive principles seem to be operative in different ways in different cases. Horn’s framework is designed to allow for such variation. What’s crucial about Horn’s Q and R principles is that they are opposed. They make inverse predictions. The Q principle enriches interpretation to exclude something the speaker didn’t say but could have; the R principle enriches interpretation to include something the speaker didn’t say but could have. There are sometimes cases where we might imagine applying either Q or R to enrich interpretation, yielding opposite results. Because of this conflict, you cannot tell how to enrich the interpretation of an utterance purely on rational grounds. Instead you need linguistic rules. The possible ambiguity in the application of Q and R is substantiated by minimal pairs in which different principles are operative in creating CIs. For an example, in English, contrast (44) and (45): (44)
John could have solved the problem.
(45)
John was able to solve the problem. (Horn 1984: 15 Ex 3a and 3b)
FOR INDEXING
44
the linguist ic tur n
When someone says (44), that suggests John did not solve the problem. It suggests he had the opportunity, but passed up on it. This is a kind of scalar implicature that is explicable in terms of the Q principle, according to which you say as much as you can. If all you say is that he could have solved the problem and you do not say he did solve it, this suggests he must not have. If you knew he had solved it, you should have said so, if you were saying as much as you could. By contrast, if you say (45), then, normally, in English, you suggest he did solve it. And you can explain this in terms of the R principle, by enriching. Normally, if you want to do something, and you are able to do it, then you do do it. Your saying he was able to do it thus gives a reason to expect that he would do it. If you are saying only what you have to say, you can therefore say simply that he was able to do it; your hearer, appealing to R, takes you to mean that he did do it. Of course, it may be wrong to attribute the interpretive difference between (44) and (45) to general principles. But Horn’s explanations in terms of Q and R do seem to reflect the goal-directed organization of conversation, and the pressures that speakers and hearers face in communicating efficiently, and thus seem similar in spirit to Grice’s explanations. Indeed, Horn’s discussion leads us to appreciate that the maxims of cooperative conversation conflict, even on Grice’s understanding. (See also Thomason 1990.) Nevertheless, if different principles apply to different examples, we cannot follow Grice in thinking that CIs are calculated from principles of rationality that always apply. We need to think about pragmatic principles in a very different way. Horn’s proposal is that general principles like Q and R constrain pragmatic interpretation, but leave certain details open that must be specified on a case-bycase basis. The key is the mutual reference of Horn’s principles. The Q principle isn’t just, “Say as much as you can”—it’s say as much as you can given R. The R principle isn’t just “Say no more than you must”—it’s say no more than you must given Q. We have, in effect, a set of simultaneous equations that narrow the CIs that might arise, but do not determine them fully. The balance between Q and R, the exact set of solutions that a language chooses to satisfy the two constraints, depends on further linguistic rules. CIs are thus a necessary and universal feature of utterance interpretation across languages. Nevertheless, to predict the interpretation of a specific utterance, we need to know how a specific language resolves the potential conflict between Q and R. Perhaps the most compelling evidence Horn offers for linking pragmatic interpretation to linguistic rules is a case study from the fieldwork of the linguist Deborah Tannen (1975). Here two different people draw different CIs from the very same utterance, contingent upon which principle each thinks applicable. The example involves a wife and husband talking about whether to attend a party.
FOR INDEXING
t h e l i n g u i s t i c tu r n 45 They wind up differing in their understanding of the conversation, because they differ about which CIs to draw from their contributions. (46)
a. Wife: b. Husband:
Bob’s having a party; do you want to go? OK.
c. Wife [Later]: Are you sure you want to go? d. Husband: Ex 1–6)
OK, let’s not go. I’m tired anyway. (Tannen 1975: 206
The next morning they are both unhappy with the other’s choices. The wife charges, “We didn’t go to the party because you didn’t want to!” The husband replies, “I wanted to go; it’s you who didn’t.” How did they get to these different interpretations of their exchange? What went wrong, as Tannen describes it, is that the wife was asking (46c) Are you sure you want to go? to find out whether the husband really wanted to go, whereas the husband took her to be doing something entirely different: he took her to be hinting that she really didn’t want to go. There is, then, a sort of clash of maxims when the wife asks (46c). On the one hand, she might really want to know what her husband wants; on this interpretation she refrains from offering any information about her feelings about the party. The Q principle underwrites this interpretation. If the wife is reluctant to go to the party, she can say so. She has not said so, so we should not suspect that she is. On the other hand, she might be saying more than what she is literally asking; she might be hinting that she doesn’t want to go. That’s the conclusion reached by someone following the R principle. Given the question, it’s obvious that the wife has some doubts about going to the party, so by R she needn’t say explicitly that she’s reluctant to go; we infer it. In short, it’s as if the wife takes the utterance to be governed by the Q principle, while the husband takes it to be governed by the R principle. Tannen argues that in America, men and women systematically assign different interpretations to many kinds of indirect utterances; it’s not a matter of confusion and idiosyncrasy, but rather represents a systematic difference in the approaches men and women bring to the communicative enterprise. It’s as if they are speaking different languages governed by different rules of conversation. Horn adopts this suggestion explicitly: the interpretive strategy the husband is following and the one the wife is following amount to different sets of rules for drawing CIs. The difference between the two sets of rules is not the Q and R principles; they are always available. Instead, the difference is their relative strength: that is to say, when and to what extent do interlocutors take the Q
FOR INDEXING
46
the linguist ic tu r n
principle to be operative and when and to what extent do they take the potentially conflicting R principle to be operative instead. This aspect of Horn’s pragmatic architecture represents a fundamental departure from Grice’s philosophy, but it exhibits a structure that’s common in linguistics. Throughout grammar, linguists expect to find evidence of universal principles; in this case, those universals are the Q and R principles. At the same time, linguists expect to find variation across languages because there can be different resolutions of the conflicts among universal tendencies. For example, different principles can be instantiated to different parameters across languages. In this case, the parameters of variation are the resolutions of preferences for Q and R, as we see in the two interpretations of (46c). We might worry, following Davis (1998), that the variability in interpretation Horn identifies in (44) and (45), and (46c), shows the activity of arbitrary conventions for implicated speaker meaning, rather than the action of general interpretive principles. This is a telling criticism that we revisit in Chapter 6. Horn’s appeal to communicative pressures can’t help resolve the difficulty: we can explain specific conventions in terms of pressures just as well as we can explain a more general interpretive preference that way. Horn attempts to meet the challenge by defending the theoretical parsimony of Q and R. He argues that invoking Q and R results in an economical statement both of linguistic semantics and of the derivation of implicatures. More precisely, Horn suggests that many pragmatic inferences can be controlled just by postulating a primitive grammatical distinction between marked and unmarked alternative expressions. Unmarked expressions are the basic ones; they are not derived; they have common distributions—and they define the cases where R interpretations dominate. Marked expressions are derived, specialized, or unusual in the language—and they define the cases where Q interpretations dominate. Horn calls the opposition the division of pragmatic labor, and introduces the diagram in (47) to explain the specialized meanings that marked and unmarked expressions take on through the differential application of Q and R principles.
(47)
The typical cases The possibilities at issue
FOR INDEXING
t h e l i n g u i s t i c tu r n 47 The rectangle in (47) visualizes some aspect of logical space. It maps out all the possibilities that might arise within some delimited semantic field or domain of discourse. The inner circle, meanwhile, designates a set of possibilities that speakers consider normal, stereotypical, or the default case. In the absence of more specific overriding information, you’d expect that a point in logical space would turn out to fall among the stereotypical cases associated with the inner circle. We can now see how differential application of Q and R principles to marked and unmarked terms can lead to complementary pragmatic enrichments. Consider an unmarked expression U whose semantic reference embraces the full rectangle of logical space in (47). The R principle applies. If someone uses U , you can take her to be talking about the usual case: the operative principle R reminds us that speakers are lazy and are not going to spell out explicitly that they are talking about the usual case, since you’d assume it applied anyway. By contrast, consider a marked expression M whose semantic reference also embraces the full rectangle of logical space in (47). Now the Q principle applies. When a speaker uses M, she has chosen not to use the unmarked term, U . But if she were in the typical case, she would have used the unmarked term. So, by the Q principle, she must be in the unusual situation where using U would have led to inaccurate implicatures. Synonymy in semantics thus induces differentiation in pragmatics. We wind up with two terms with distinct pragmatic meanings: the unmarked term U refers to the usual situation; the marked term M to the unusual one. This is a difference in meaning in the language, since the contrast between U and M is ultimately traceable to the conventional fact that U is unmarked and M is marked.1 Nevertheless, we can simplify our statement of the facts about meaning and explain them better if we specify the meanings of U and M in two steps: first, the introduction of simple bare-bones meanings; and second, the enrichment of these meanings by principles Q and R. Perhaps surprisingly, Horn does not directly apply the division of pragmatic labor to the examples he uses to motivate a linguistic account of CIs.2 But the exact treatment of these cases is not crucial to Horn’s argument in introducing the division of pragmatic labor. He is not committed to explaining all CIs through it. Horn’s key claim is just that the division of pragmatic labor establishes a 1 See our discussion of (50a–51) for the conventionality of markedness. 2 The contrast in (46c) is not relevant because the different interpretations inhabit different idiolects.
We might speculate that the division of pragmatic labor makes ‘was able’ unmarked in (45), where we assume with R that you do what you are able to; while ‘could have’ is marked in (44), where we assume with Q that you have not done what you could have.
FOR INDEXING
48
the linguist ic tu r n
fundamental place for Q and R in stating the rules of language. And Horn does illustrate the power of this approach through a diverse array of examples. One notable case involves the difference between certain lexical causative expressions in English and their periphrastic counterparts. Horn uses (48) and (49) to illustrate the contrast. (48) (49)
Lee stopped the car. Lee got the car to stop. (Horn 1984: 27 Ex 29a and 29b)
Intuitively, (48) and (49) seem to describe very different situations. In (48), we understand that Lee stopped the car in the ordinary way, by pressing on the brakes and allowing the car to slow to halt. In (49), however, we do not imagine things unfolding in the usual way. Something must have gone wrong: perhaps the brakes didn’t work properly, and Lee had to press on them really hard or bring the car under control by some extraordinary measure—plowing up onto the curb maybe. What makes this difference in interpretation particularly striking is the close relationship in the meanings of the two sentences. Fundamentally, both (48) and (49) seem simply to say that Lee arranged things so that the car, which had been moving, was no longer moving. But this is exactly the kind of situation we can explain with the division of pragmatic labor. Suppose that (48) is an unmarked way of expressing this fundamental meaning, while (49) is a marked one. Then we should understand (48), via R, with an enriched meaning that suggests that Lee stopped the car in the normal way. Conversely, we should understand (49), via Q, as describing unusual situations, where the default understanding of (48) would not apply. Concretely, if all Lee did were to press on the brakes and bring the car to a halt in the normal way, the speaker wouldn’t have used (49). Thus, by appealing to the division of pragmatic labor, we don’t have to say explicitly that (49) has a special meaning that excludes that Lee stopped the car. The relationship between these two forms—one is unmarked, the other marked—settles this issue for us. The case of causative expressions is particularly instructive because not all lexical and periphrastic causatives in English seem to share a distinction in markedness of this kind. For example, Horn suggests that neither (50) nor (51) is marked with respect to the other: (50) (51)
That sort of behavior really angers me. That sort of behavior really makes me angry. (Horn 1984: 28 Ex 31a and 32a)
As speakers of English, we don’t find an immediate difference between (50) and (51). We wouldn’t predict this synonymy if (50) was unmarked and (51) was
FOR INDEXING
t h e l i n g u i s t i c tu r n 49 marked in a way analogous to (48) and (49). We’d expect that only (50) could describe an ordinary case where the speaker has been wronged. With (51), we’d expect the emotions to work themselves out in some exceptional way— we might expect (51) to be particularly appropriate, for example, in paradoxical cases where another’s assistance unexpectedly provokes displeasure. Obviously, no such contrast exists. We see then that the division of pragmatic labor does not eliminate the need for rules of language. We still need rules that tell us which forms are unmarked alternatives and which are marked ones. In effect, these rules establish that certain forms in the language compete with one another, or as Horn says sometimes, form a scale. Once we have these linguistic stipulations, however, the division of pragmatic labor allows us to predict their semantic and pragmatic consequences. In many respects, this theory echoes a longstanding functionalist tradition in linguistics, going back to de Saussure (1916), who saw linguistic elements as organized in systems of related elements whose meanings stand in opposition to one another. Linguists such as Halliday (1985) have suggested that we must fundamentally rethink grammar in order to capture the effect of such oppositions. By contrast, Horn argues that we can capture such insights without modeling grammar itself in functional terms—but, to do so, we must appeal to general interpretive principles such as Q and R. Critical discussion of Horn has emphasized his commonality with Grice. Horn accepts the category of CI, and indeed, has found a large number of apparent novel cases of CI, and offered explanations in terms of general pragmatic principles. Nevertheless, we draw a very different message from Horn’s work. Horn recasts pragmatics from being a part of rational principles we use on the fly to interpret utterances, as in Grice, to a set of background principles we use to hypothesize and explain the rules of language in a constrained way. Because Horn repositions his interpretive principles as linguistic constructs, he does not offer an approach in which CIs are calculated purely on the basis of rationality. Horn’s theory of CIs therefore cannot carry the philosophical weight that Grice hoped for. We explore the significance of this departure in section 3.3, but first, we consider another approach to interpretive principles as linguistic constructs: Levinson’s.
3.2 Linguistic Principles for Disambiguation and Enrichment: Levinson Like Horn, Levinson argues that the principles that explain CIs are linguistic in nature. And, like Horn, Levinson argues that this linguistic approach makes the
FOR INDEXING
50
the linguist ic tu r n
theory of CIs broader in its scope and more straightforward in its predictions. Levinson, in particular, focuses on the possibility of using linguistic principles of interpretation to disambiguate the meanings of linguistic expressions and enrich their interpretations. This further departure from the Gricean picture adds to the flexibility of the theory and so allows Levinson to describe a wide range of new effects as CIs. Levinson develops his theoretical picture in three works that we draw on closely here: Atlas and Levinson’s “On It-Clefts” (1981), his Pragmatics (1983); and, finally, the major published statement of his views in Presumptive Meanings (2000). Levinson’s framework overlaps closely with Horn’s in its data and explanations—indeed, the two researchers see themselves as having developed a mutually reinforcing body of work. Thus, it will be possible for us to explore the key innovations of Levinson’s approach after just a brief review of his account of interpretive principles as linguistic rules. Like Horn, Levinson starts by streamlining the maxims. Levinson’s three interpretive principles cover much the same ground as Horn’s two. Levinson invokes a quantity principle (52) that is an analogue of Horn’s Q: (52)
Q (Quantity) Principle: Say as much as you can (“Do not provide a statement that is informationally weaker than your knowledge of the world allows . . . ” Levinson 2000: 76).
Levinson’s Q, like Horn’s, enriches the meaning of an utterance by excluding certain relevant possibilities. In place of Horn’s R principle, Levinson offers two separate principles that enrich the interpretation of linguistic expressions by relating them to a default case. Levinson’s informative principle (53) allows interlocutors to take unmarked utterances to describe the usual case. (53)
I (Informative) Principle: Let the hearer fill in expected details (“Produce the minimal linguistic information sufficient to achieve your communicational ends” Levinson 2000: 114).
Meanwhile, Levinson’s manner principle (54) rules out the normal case when the speaker uses marked expressions: (54)
M (Manner) Principle: Use unusual language precisely when things are unusual (“Indicate an abnormal, non-stereotypical situation by using marked expressions that contrast with those you would use to describe the corresponding normal, stereotypical situations” Levinson 2000: 136).
Levinson’s M and I principles together create an analogue of the division of pragmatic labor. You normally say things in unmarked ways to describe the
FOR INDEXING
t h e l i n g u i s t i c tu r n 51 normal situation. Conversely, you tend to describe anything unusual explicitly— which you can achieve just by saying things in an abnormal way. Levinson’s interpretive principles, like Horn’s, provide grounds for enriching a meaning in alternative, incompatible ways. For example, suppose the conversation concerns the ordinariness of some aspect of a situation and the speaker uses an unmarked utterance with a more general meaning to describe things. We could potentially assume the ordinary case does not hold, via Q, or we could assume that the ordinary case does hold, via M and I. Of course, for any particular utterance and context, the rules of language will specify a particular resolution for such potential conflicts. Levinson (2000) offers an extensive catalogue of such interactions, along with their resolutions. While Horn and Levinson agree on the linguistic nature of interpretive principles, Levinson sees them as much more tightly integrated into the rest of grammar. Recall that Grice’s definition of CI assumes that the speaker’s utterance is unproblematically associated with a semantically determined proposition p, what the speaker says. For Grice, this contribution gets enriched or revised through the calculation of CIs. By contrast, according to Levinson, an interpreter can apply Gricean reasoning even before he knows how the rules of grammar apply to an utterance or what the semantics of the utterance is. For Levinson, pragmatic principles guide disambiguation. Levinson challenges the Gricean view by attempting to describe the resolution of indexicality and reference in rational terms. For example, imagine that today is Wednesday, and consider the contrast between utterances of (55) and (56). (55)
The meeting is tomorrow.
(56)
The meeting is Thursday. (Levinson 2000: 178 Ex 10b)
(55) is the normal way that you’d say that the meeting will take place on the Thursday immediately following. You wouldn’t use (56) to express this meaning. Assuming the speaker knows that today is Wednesday—which some addressees, hearing (56), might start to doubt—(56) presumably means that the meeting will occur eight days later. It will be a week from tomorrow, not tomorrow. Now, when speakers use words like Thursday to name a day within a specific week, they respect a complex array of constraints that narrow down the referential possibilities (?).3 We see some of this complexity in the fact that the 3 Speakers can also use words like Thursday predicatively, as in Tomorrow is Thursday. In this case, we suppose, the speaker uses tomorrow to fix the day and Thursday to characterize how the day falls on the calendar. We needn’t expect accounts of speakers’ referential intentions to extend to such utterances.
FOR INDEXING
52
the linguist ic tur n
word Thursday refers to six days earlier on a Wednesday when a speaker utters (57) rather than (56). (57)
The meeting was Thursday.
The sentence frame in (57) indicates that the referent of Thursday must be located in the past, whereas the sentence frame of (56) indicates that the referent of Thursday must be located in the future. Given the richness of this inferential process, we might handle the contrast of (55) and (56) by stipulation: when we interpret (56) we simply exclude the day immediately following the utterance as a possible referent for Thursday. However, it’s unclear why we should have such a constraint. Intuitively, we can respect the future reference of (56) just by considering the week beginning now. Levinson suggests that pragmatic principles are an important part of the explanation. His idea is that Thursday and tomorrow in these utterances are alternatives, where Thursday is the marked item and tomorrow is the unmarked one. Thus, tomorrow is the default word you would use if you are in a position to apply it. If, as a speaker, you are describing something that is going to happen the next day, you’ll use (55). Hence the day after the day of utterance is excluded as a possible interpretation of Thursday, and we must therefore resolve the reference of Thursday within the week beginning the day after the utterance. Thursday and tomorrow always refer to different days, because you make the best choice when you decide which word to use. Levinson’s idea intuitively parallels the resolution of reference for definite descriptions, a process that Neale (1990), for example, characterizes in terms of Gricean reasoning. For Neale, expressions like the man have an underlying quantificational meaning; nevertheless, following Kripke (1978), speakers can use these expressions to refer to specific individuals. An audience can calculate the speaker’s reference from the choice of description the speaker uses and the assumption that the speaker was speaking cooperatively. In many cases, this calculation involves a division of pragmatic labor intuitively similar to what Levinson highlights with (55) and (56). Discourse (58)4 illustrates the pattern.
Perhaps an alternative meaning for Thursday is at play, which contributes a property to semantic interpretation. But even if the predicative interpretation arises pragmatically—e.g. through the reasoning proposed by Stalnaker and reviewed in Ch. 2.2.1, where the speaker flouts the Maxim of Clarity by using a name whose reference his audience does not know, and thereby triggers a pragmatic reinterpretation and repair—Thursday does not carry speaker reference. 4 Retrieved from on June 1, 2012.
FOR INDEXING
t h e l i n g u i s t i c tu r n 53 (58)
a. In the spring of 1994, Dutch dealer Sebastiaan Hesselink received a phone call from a New York man who had a number of rare books and letters to sell. . . . b. Hesselink contacted Interpol, the FBI and local Dutch authorities, and they set up a sting for four o’clock the next day. c. On time, the man arrived in Utrecht’s central square with his bag of loot.
At issue is the interpretation of the man in (58c). This discourse most centrally concerns two particular men, in addition to the various unnamed law enforcement officials: book dealer Sebastiaan Hesselink and a book thief readily identifiable from the details of the story as one Daniel Spiegelman. In (58c), the man obviously refers to Spiegelman rather than to Hesselink, when we might have expected Hesselink to be more prominent at this stage in the arc of the narrative. But we have a ready Gricean explanation for this reference in the division of pragmatic labor. If the speaker had wanted to refer to Hesselink, the speaker would have used the name Hesselink or the pronoun he. Since the speaker used the man instead, the speaker must not have intended to refer to Hesselink; he must therefore have intended to refer to Spiegelman.5 In short, names and descriptions are alternatives for pragmatic purposes. On ordinary Gricean theory, this explanation of the reference of the man in (58c) is only available because we can assign a meaning for the man that is separate from and prior to its pragmatic reference to Spiegelman. Because we have this meaning, we can identify an abstract proposition involving it that counts as what the speaker says in (58c). We can then enrich and refine the semantic contribution to arrive at a pragmatic interpretation, along Gricean lines. We could treat (56) the same way, if we analyzed Thursday as some kind of concealed definite description, perhaps the fifth day in the week in question. However, Levinson argues that it would be surprising to assign that kind of meaning to Thursday—the word seems more like an indexical. But indexicals just contribute their reference to the meaning of utterances, so Levinson argues that what the speaker says in an utterance (56) already resolves the reference. Thus there’s no way to resolve the reference by Gricean reasoning analogous to that of (58c). 5 Note, by the way, that the symmetric inference is not available. The speaker has never used the name Spiegelman so we can assume it’s unfamiliar to his audience. Thus, the speaker could not have said Spiegelman even if the speaker wanted to refer to Spiegelman.
FOR INDEXING
54
the linguist ic tur n
Levinson proposes a revision of Grice’s theory that does make it possible to give a uniform pragmatic description of reference determination for names, indexicals, and descriptions. According to Levinson, rather than applying interpretive principles just to what is said, hearers can use general interpretive principles to refine and elaborate meanings at any stage of the interpretive process. This assumption leads to precisely the explanation of (56) we considered earlier. Levinson’s reframing of Gricean analysis seems quite natural from a linguistic and psychological view, especially given the parallels between (56) and (58c), but it undercuts Grice’s philosophical program. It suggests that any attempt to parse pragmatic effects into a level of rationality entirely separate from the rules of language must fail. Indeed, there are deep philosophical problems in the relationship between disambiguation and Gricean pragmatics, which we will return to in Part IV of the book. Levinson’s influential analysis of embedded implicatures gives a different motivation for integrating defeasible interpretive principles with the rules of language. As with (56) and (58c), Levinson starts from an interpretive parallel which, prima facie, calls for a uniform theoretical explanation. Start with (59). (59)
The chair sometimes comes to department meetings.
A natural understanding of (59) is that the chair comes to some department meetings but also misses a sizeable fraction of them. This more precise interpretation is plausibly explained, like (42), as a case of scalar implicature. The idea is that the adverb sometimes has a purely existential meaning. But we expect the speaker of (59) to have a good idea of how many meetings the chair comes to and to describe that quantity as strongly and specifically as possible. Since the speaker has not said that the chair comes to all the meetings, we can assume the stronger statement is probably false. The problem comes when we try to extend this explanation to other examples whose interpretations seem parallel. (60)
If the chair sometimes comes to department meetings, that is not enough. (Levinson 2000: 205 Ex 24c)
Most English speakers have the judgment that (60) is about the possibility that the chair comes to some department meetings but also misses many: the speaker suggests the chair should come to all the meetings. This makes sense if we see pragmatic principles as operating in tandem with other rules for calculating meaning in language, as Levinson has argued. To interpret (60), we need to identify the hypothetical situations under discussion. In keeping with the division
FOR INDEXING
t h e l i n g u i s t i c tu r n 55 of pragmatic labor, we assume that the speaker has described those situations in the best possible way, and refine the interpretation of the conditional accordingly. The simplicity of this explanation belies the depth of its departure from the Gricean picture. To appreciate the issues, note that conditionals like (60) are about what’s possible, not what’s true. Unlike the speaker of (59), who seems to be reporting on knowledge of the chair, a speaker of (60) seems to know only that the chair might sometimes come to meetings. Lots of reasons for uncertainty suggest themselves, but surely one plausible reason is that the speaker thinks that the chair might come to no meetings, or half, or all. Why, then, shouldn’t we take the speaker of (60) to be describing what all the situations have in common where the chair comes to at least some of the meetings? To make the reading more vivid, imagine a university where any deliberation must be inclusive but absolutism is permitted. Then once the chair comes to any meetings, she has obligations to consult the faculty by email in addition; she cannot meet her obligations even by attending all the meetings. Her only alternative would have been to avoid the meetings altogether and make the decision on her own. Since such general interpretations are coherent, there seems to be no reason why it would be irrational or uncooperative for a speaker to make them, and so no reason to favor a weaker interpretation purely on the Cooperative Principle. Nevertheless, (60) does seem to be about situations where the chair misses many meetings. To explain this pragmatically, we must think about the pragmatic principles that inform interpretation in a new way. According to Levinson, we take sometimes, across the board, as a marked description of how often a chair might come to meetings. We infer that the speaker must have intended to comment on a specific set of hypothetical situations because we observe the marked description. The parallel with (59) underscores the appeal of Levinson’s approach. Levinson, in short, appropriates Grice’s architecture of general principles that provide defeasible enrichments to interpretation, but reconceptualizes this architecture as an essential dimension of the rules for deriving meaning in language. Levinson’s interpretive principles cannot be separated from the rest of semantics. The interpretive principles shape how we resolve references and other semantic ambiguities; they shape how we understand a complex utterance as a function of the meanings of its parts. This represents a radical theoretical departure from a Gricean semantics–pragmatics interface. In fact, Levinson argues that the key difference between semantics and pragmatics is simply that pragmatic contributions to meaning are defeasible and so can be canceled, for example, unlike semantic contributions.
FOR INDEXING
56
the linguist ic tur n
Levinson’s abandonment of the received distinction between semantics and pragmatics is in many ways an extension of the more thoroughgoing rejection of Gricean ideas that Levinson shares with Horn. Interpretive principles, for Horn and Levinson, are linguistic, and cannot be grounded in rationality or psychology. Their linguistic status is reflected in their sensitivity to linguistic constructs like competitors and scales; and in the language-specific weights and preferences that resolve the inherent conflicts among them. Ultimately, of course, their linguistic status is reflected in the rich language-specific patterns of interpretive variation within and across languages, which their theories make sense of.
3.3 Scales and the Influence of Pragmatic Rules The theoretical perspectives advocated by Horn and Levinson are widespread in linguistics. Particularly influential has been their idea that the rules of meaning in language are to be specified in terms of constraints on the applicability of general default principles. A wide range of grammatical resources can be productively characterized this way, and the paradigm case of how to do it is Horn’s and Levinson’s architecture of marked and unmarked alternatives, scalar relationships, and the division of pragmatic labor. The subfield of linguistic pragmatics explores and systematizes these effects. The interpretation of pronouns and related forms across languages offers a case study that’s suggestive of the breadth and depth of this research. We can start with a case that both Horn and Levinson consider: the contrastive interpretation of English argument pronouns, for example him, and their reflexive counterparts, for example himself. The contrast is illustrated in (61) and (62). (61) is a way of saying what ordinarily happens when John looks in the mirror. (61)
John saw himself.
By contrast, when a speaker uses (62), it normally describes a situation where John sees someone else. (62)
John saw him.
It can’t describe the ordinary case where John looks in the mirror.6 The difference is naturally characterized as a syntactic constraint specifying the possible structural relationships between different classes of pronouns and their antecedents (Chomsky 1981). A reflexive must be linked with an antecedent within 6 Cases of mistaken identity aside. More precisely, him must be interpretively independent of John; him must be someone we cannot identify as John given the current information and point of view of the discourse.
FOR INDEXING
t h e l i n g u i s t i c tu r n 57 a specified local domain; in English, to a first approximation, the antecedent for a reflexive must be a more prominent argument of the same predicate. This is Principle A of Chomsky’s (1981) Binding Theory. By contrast, it seems, an ordinary pronoun cannot be linked with an antecedent within this same domain. This is Chomsky’s (1981) Principle B. Levinson’s pragmatic architecture, in which interpretive principles inform the resolution of ambiguity, allows us to explain the relationship of Principle B to Principle A. In the relevant cases, the forms himself and him are alternatives. Grammar therefore requires the speaker to choose the term that best fits her referential intention. Given that Principle A applies to himself, it’s a consequence of English grammar that (61) can only describe the situation where John is looking in the mirror. Because it specifically carries this meaning, it’s the best choice for a speaker who’s describing this kind of situation. Conversely, a speaker will only use the alternative (62) if she couldn’t use (61). In other words, him in (62) must be someone other than John. As usual in Horn and Levinson’s linguistic pragmatics, this is offered as a streamlined account of the conventional interpretations of the different pronouns—not an explanation in the Gricean spirit that obviates the need for the relevant conventions. In particular, we still need a linguistic statement of the contexts in which reflexives and ordinary pronouns count as alternatives, and this is a difficult problem at the interface of syntax and semantics. Some environments, such as the spatial prepositional phrase argument in (63), seem to allow pronouns to take close and prominent antecedents, where reflexives are also natural. (63) (64)
He pulled his cloak around him. He pulled his cloak around himself. 7
Conversely, other environments—whether because of long-range syntactic dependencies or the active argument structure of certain kinds of nominals— seem to permit reflexives despite the absence of any overt local antecedent. The use of herself to refer to Garcia in (65) is such a case. (65)
Among the loans Garcia said she wrote were two for herself totaling $177500.8
Again, such environments do not always bring an interpretive difference between reflexives and ordinary pronouns. Our judgment is that replacing herself in (65) 7 (63) and (64) are cliches of fantasy fan fiction, and occur tens of thousands of times each on the web, in a ratio of roughly 4:1. 8 From the LA Times, March 1, 2001: .
FOR INDEXING
58
the linguist ic tur n
with her would still result in a felicitous reference to Garcia. Thus, it’s wrong to offer a purely rational account that says that reflexives preempt ordinary pronouns because reflexives are invariably more specific, a point Davis (2014) emphasizes. We need a grammatical statement of when contrasts arise between forms, as a function of particular syntactic environments and particular linguistic alternatives.9 Even Chomsky’s explicitly syntactic Binding Theory fits this description to some degree. The linguistic nature of the phenomenon is clear, but the role for general interpretive principles with the generality and abstraction of Horn and Levinson’s Q, R, M, and I is not. Assessing the philosophical import of this research is complicated by the fact that it tends to rely on informal and programmatic statements of its pragmatic principles. Horn and Levinson themselves do not offer a formal framework to predict when pragmatic principles apply and what default interpretations they trigger. Much the same is true of other explorations of interpretive contrasts within the linguistic pragmatic literature—for example, in the interpretation of stressed pronouns as in Kameyama (1999), the interpretation of null pronouns as in Dimitriadis (1996), and the form of more complex referring expressions as in Gundel et al. (1993). They tend to describe alternative elements and the syntactic environments in which they contrast in surprisingly fine-grained ways, while relying on an intuitive account of the division of pragmatic labor to characterize the interpretive effects of these contrasts. Thus, it seems that more precision is required to address the key theoretical questions—What are the kinds of implicatures that we can and should explain with linguistic rules? And what do those rules actually look like? Formal linguistic work promises to clarify the situation, and so we will wait to address these questions in detail until surveying a range of analyses of related interpretive effects in Part II of the book. However, we can tip our hand: We are inspired by Horn and Levinson to endorse the claims that implicated content is linguistically determined, and that general pragmatic principles play a key role in disambiguation. But we will ultimately reject the linguistic interpretive principles that are key to Horn and Levinson’s accounts, and explain both the origin of implicated content and the process of disambiguation in fundamentally different ways. 9 The need for linguistic rules is underscored by the perspectival use of reflexives illustrated in Henry James’s For herself, nothing was changed . . . (from Portrait of a Lady 1917). Henry James’s use of herself here accompanies discourse that indirectly reports a main character’s first-person thoughts. In general, ordinary pronouns are also possible in syntactic environments where reflexives carry this suggestion of point of view and emphasis. It’s not surprising that English offers speakers the resources to choose whether or not to signal this meaning. In fact, we may even want to say that the pronoun is the unmarked alternative in these environments, so that here the reflexive gets an interpretive twist through the division of pragmatic labor!
FOR INDEXING
t h e l i n g u i s t i c tu r n 59 However such arguments are resolved, there is a clear limit on the scope of linguistic approaches: the inherent limits of all conventions. Horn and Levinson give particular weight to Grice’s distinction between generalized CIs and particularized CIs. Generalized CIs are regularly associated with types of linguistic expressions; particularized CIs depend on idiosyncratic features of the conversational context and purposes. To the extent that particularized CIs are creative nonce inferences invented ad hoc by speakers, they cannot be grounded in conventions and so are not suitable for linguistic analysis. Flouting, in particular, is a telling diagnostic of the specific notion of rationality at work in Grice’s framework. Linguistic approaches to implicature cannot explain implicatures via the flouting of interpretive principles. That would in effect require giving a rule that says how to break all the rules! With Gricean reasoning, by contrast, interlocutors are free to exploit interpretive principles in creative and even counterintuitive ways. This viewpoint suggests that we should not expect linguistic approaches to implicature to extend straightforwardly to the various phenomena that other researchers have characterized in terms of flouting the Cooperative Principle—for example, cases of metaphor and irony. Linguistic approaches, then, capture at most part of traditional pragmatics.
FOR INDEXING
4 The Psychological Turn
In Grice’s framework, pragmatics is different from semantics because pragmatics is decisively concerned with a speaker’s choices in contributing content to conversation. Grice himself had a particular idea about the nature of the principles that describe speaker’s choices. For Grice, speakers appeal to the norms of rationality to add layers of meaning to what they say. We’ve seen that this is not the only way to characterize speakers’ choices, however. The linguistic theories surveyed in Chapter 3 suggest that it is often the rules of language that describe speakers’ choices. For example, linguistic rules can distinguish unmarked forms, which should be used to describe typical situations via R principles, from marked forms, which should be used to describe exceptional situations via Q principles. Such rules might give a way to broaden the insights of Grice’s framework while accounting for the apparent empirical variability of implicated meaning. In this chapter, we explore yet a different understanding of speakers’ choices. Here we think of choices not as targeting an idealized notion of rationality, as in Grice, not as following linguistic rules, as in Horn and Levinson, but simply as exhibiting the empirical psychological principles through which people turn out to perceive, reason, and act. The locus classicus of this psychological approach to pragmatics is Sperber and Wilson’s (1986) Relevance Theory. Relevance Theory explores the analogy between making sense of an utterance and using other kinds of information about the world. As we explore in section 4.1, Relevance Theory suggests that speakers’ contributions to conversation go beyond conventional semantics because of the reasoning hearers inevitably use to connect what happens to the concerns that interest them. Sperber and Wilson couch their theory in terms of a Principle of Relevance, but, as we shall see, this is not an analogue, revision, or refinement of Grice’s Cooperative Principle (CP) or Horn’s or Levinson’s R principles. It’s a fundamentally different kind of construct—a psychological law that represents an empirical discovery about human cognition.
FOR INDEXING
th e p s ych o l o g i c a l tu r n 61 Meanwhile, a range of other researchers, such as Pinker, Nowak and Lee (2008), explore the analogy between linguistic decision-making and the heuristic, biased, and bounded reasoning processes that describe people’s decisions more generally. As we explore in section 4.2, these researchers suggest that pragmatic meaning can amplify semantics because hearers understand the strategies that speakers have actually used in formulating utterances—with all the flaws and limits of those strategies. As with the linguistic theories of Chapter 3, these psychological theories are instructive because they generalize the scope of conversational implicatures to a range of apparently similar cases. And they are also instructive because they reveal the weaknesses of Gricean explanations couched purely in terms of the CP and the maxims. In particular, psychological approaches attempt to ground interpretation in empirical models of inference, which, they argue, provide a better explanation. Despite these contributions, these theories cannot sustain Grice’s conception of the relationship between semantics and pragmatics, as we emphasize in section 4.3. Most importantly, Grice is prepared to characterize the contributions that speakers make through the semantics of their utterances, and the contributions that speakers make through implicature, in terms of public propositional content added to the common ground of the conversation. Psychological approaches must abandon this idea. For psychological approaches, communicated meaning must invariably involve an open-ended, uncertain distribution over ideas, since what matters is ultimately the information that the speaker brings to their audience’s attention through an utterance. To retain a parallel between implicatures and other contributions, psychological approaches must deny the idea of public propositional content altogether.
4.1 The Psychology of Interpretation: Relevance Theory Relevance Theory is broadly inspired by Grice’s (1957) suggestion that meaning is fundamentally a psychological phenomenon. The goal of Relevance Theory is to replace Grice’s idealized, rationalistic account of meaning with an empirical, naturalistic account grounded in the science of the mind. This is a large and ambitious research program—an approach to representation and communication, not just to pragmatics—and our review of Relevance Theory will therefore focus selectively on specific aspects of the theory. Our key sources in discussing it will be Relevance: Communication and Cognition (Sperber and Wilson 1986) and Meaning and Relevance (Wilson and Sperber 2012). Nevertheless, while Relevance Theory makes a range of proposals that do not
FOR INDEXING
62
the psycholo g ica l tu r n
directly pertain to our investigation in this book, we think it’s important to introduce the overall framework of the approach in some breadth. Readers who approach Relevance Theory specifically for its insights into implicature and other inferential aspects of interpretation might not fully appreciate the depth of its differences with other approaches otherwise. Sperber and Wilson’s approach is grounded in three distinctive key ideas. The first is a psychological conception of efficient inference, which, Sperber and Wilson argue, characterizes all human information processing. This conception is cashed out in the technical notion of relevance, which we explore next. Because speakers plan utterances as intentional efforts to get information across, this kind of efficient inference is at the heart of communication. The second key idea of Relevance Theory is an inferential explanation of disambiguation. Sperber and Wilson take particular pains to characterize the psychological processes that lead hearers to specific understandings of ambiguous or underspecified utterances. Their view is that ambiguity resolution promotes efficient inference. We continue this section by reviewing the reasons why this conception of ambiguity resolution is both substantive and original. The third key idea is that interlocutors use efficient inference to specialize and adapt their knowledge of language to their purposes in specific interactions. In particular, Relevance Theory calls attention to the explicature of an utterance—a pragmatically derived meaning that enriches and transforms whatever semantics is specified by grammar. We finish our discussion of Relevance Theory by showing how it appeals to the concept of explicature to streamline pragmatics and avoid both linguistic stipulations, on the one hand, and such Gricean mechanisms as flouting, on the other.
4.1.1 Relevance as Efficient Inference For Sperber and Wilson, the essential problem of language understanding is to figure out what an utterance means for you. In key ways, they argue, this problem recapitulates the broader problem of understanding the world and identifying what any ongoing event means for you. In particular, in both cases you must draw inferences to relate what is happening to what you already know and to your practical concerns. Moreover, in both cases, it is hopeless to pursue these inferences exhaustively. There are indefinitely many possible conclusions: some may require prodigious effort to derive; others may offer only marginal value in informing future decisions. Accordingly, Sperber and Wilson suggest, general psychological mechanisms must be in place to focus the inferences we draw in understanding. Because these mechanisms respond to empirical limits in our
FOR INDEXING
th e p s ych o l o g i c a l tu r n 63 ability to process information, they have a distinctively psychological character. For example, they reflect the dynamics of human attention, the heuristics of human problem solving, and the biases of human information gathering. At the same time, Sperber and Wilson suggest, these mechanisms are normally effective in allowing us to draw useful inferences efficiently; in particular, they underpin the inferences interlocutors draw from utterances, including inferences that go beyond the grammatically specified meanings of the words used. Sperber and Wilson (1986) propose to characterize our inferential abilities in terms of the notion of relevance, their key theoretical construct. Relevance is a graded notion that measures the inferential demand of a stimulus an agent encounters. Sperber and Wilson characterize this inference in terms of contextual effects, modifications of an agent’s information state that synthesize incoming assumptions with remembered information. Their definition is: (67) (68)
Extent condition 1: an assumption is relevant in a context to the extent that its contextual effects in this context are large. Extent condition 2: an assumption is relevant in a context to the extent that the effort required to process it in this context is small. (Sperber and Wilson 1986: 125 Ex 10)
In general, in understanding the world, it’s in our interest to draw lots of inferences that connect new events to their implications for us. Other things being equal, a stimulus is more relevant the more inferences it prompts. At the same time, at a certain point, we have to cut off inferences, because it’s no longer worth the effort to derive further conclusions. Other things being equal, a stimulus is more relevant the faster we can get to this point. A very relevant stimulus, then, is one that prompts many ideas very easily. Conversely, cognition is most efficient when it maximizes relevance, by attending to stimuli that matter and quickly finding the important consequences of those stimuli. We can illustrate the importance of maximizing relevance in inference by considering what a diner might think about when she finds herself at dinner in a restaurant, and the server brings out a dish of food to be shared at her table. She might realize the group that she is out with can commence eating. She might notice how tasty, spicy, and warm the food looks. She might think about what she needs to do in her immediate future: pass the dish around, serve it for herself, make space on her plate, etc. But there’s also much she knows about the food that she is unlikely to think about when the dish is brought to her, because it just doesn’t matter. Suppose, for example, the dish features finely diced green peppers—the diner can infer from their presence that somewhere in the back earlier that day someone was chopping and washing and preparing those peppers.
FOR INDEXING
64
the psycholo g ical tu r n
Similarly, she knows that after she’s done eating, the waiters will take the plates away and someone in the back will wash them. But she is probably not going to think about this now; it’s not important. She’s just going to eat. Sperber and Wilson would like us to suppose that the same sort of inferences occur to an audience when they understand an utterance as when they understand what happens around them in the world. The simplest way to think about this is to imagine reading a narrative describing a dinner. As you read along, you find that the next sentence is (69). (69)
The server brought out a dish of food.
In reading this, you are going to take the perspective of the characters eating that dinner and draw inferences similar to those just described. You know they are preparing to eat; you can think about what the food must be like for them. And just as you wouldn’t think about the details of the mechanics of restaurants yourself if you were eating in a restaurant, you won’t think about them when you’re reading this story. There’s a commonality between the inferences you draw normally when events happen and the inferences you draw if you’re trying to understand a text.1 This simple hypothesis already leads to the most important consequence of Relevance Theory—an explanation of why CIs are often fuzzy. In Grice’s theory, a speaker’s CI is based on a particular bit of extra information that the speaker is trying to signal to her audience indirectly; in Horn’s and Levinson’s theories, CIs come from rules of language that add specific information to an utterance when it is used in certain ways. In Relevance Theory, something looser and less precise is going on. Speakers are trying to prompt relevant inferences for their audience, and their audiences are trying to follow up on what the speakers are saying by drawing relevant inferences. This process is open-ended, because audiences can draw fewer or many inferences; and it’s also approximate, because a speaker doesn’t necessarily know in advance which inferences the hearer is going to draw. Sperber and Wilson often point to cases of nonverbal communication to illustrate this aspect of Relevance Theory:
1 You might worry about the disanalogy between events that take place around us in the world and the content of an utterance we interpret, which might be true, but might be false, or might be fiction. In fact, most of our general knowledge is probabilistic; it admits exceptions. That means all events can bring defeasible evidence of how things are, and we will always have to distinguish between the new arguments that events prompt us to consider and the conclusions we ultimately endorse. Such considerations restore the parallel between utterances and other events. In all cases, we want to focus our inference on arguments that matter to us, as we construct an interpretation of new stimuli that summarizes all the new evidence it gives us, and all the new arguments we must therefore consider.
FOR INDEXING
th e p s ych o l o g i c a l tu r n 65 (70)
Mary and Peter are newly arrived at the seaside. She opens the window overlooking the sea and sniffs appreciatively and ostensively. When Peter follows suit, there is no one particular good thing that comes to his attention: the air smells fresh, fresher than it did in town, it reminds him of their previous holidays, he can smell the sea, seaweed, ozone, fish; all sorts of pleasant things come to mind, and while, because her sniff was appreciative, he is reasonably safe in assuming that she must have intended him to notice at least some of them, he is unlikely to be able to pin her intention down any further. (Sperber and Wilson 1986: 55)
Sperber and Wilson emphasize that our utterances often exhibit this indefinite character. For example, in conversation with us, Sperber has suggested that an exclamation of the utterance Wow! has much the same interpretation as the sniff of (70). Imagine that the speaker and the audience are watching a sporting event, and one of the athletes does something extremely impressive; the speaker comments with an utterance of Wow! Her utterance doesn’t mean anything specific; there’s not a particular proposition she wants the hearer to appreciate, or that she wishes to communicate. Relevance Theory can describe it as follows: The speaker is intending to prompt a range of relevant inferences on the part of the hearer about how impressive what just happened is. And the hearer, in turn, can understand this utterance by pursuing whatever such inferences come to mind. There’s no single particular proposition to communicate or to recover in these sorts of cases. Rather, as described by Relevance Theory, there is an indefinite set of propositions the speaker imagines that the audience might conclude from the utterance. This approximate, open-ended communication is normal and unproblematic for Relevance Theory; it’s even consistent with a central place for communicative intentions in semantics and pragmatics. In particular, Sperber and Wilson assume that speakers often have only general or approximate intentions about the inferences that their audience will draw, and count their intentions as accomplished as long as the audience does entertain a sufficient range of such inferences. Of course, we often have something more specific to say to each other than, Wow! even if we don’t narrow what we have to say all the way down to a single proposition. However, Relevance Theory argues that approximate and open-ended interpretive inference is always important, and that most of our utterances have a rough-and-ready character. Relevance Theory offers a vocabulary to describe the interpretation of these utterances, and to explain why these utterances are so natural. For example, consider the approximation involved in an utterance of (71).
FOR INDEXING
66 (71)
the psycholo g ica l tu r n The lecture starts at five o’clock. (Wilson and Sperber 2012: 54 Ex 7)
In using (71), the speaker is not committed to the precise claim that the lecture starts at exactly five—something that would be false if the lecture started at 5:05 say. Rather, (71) simply serves as a ballpark estimate for when the lecture starts. The way to describe this in Relevance Theory is that the speaker is trying to prompt inferences about when the lecture starts; namely, all the inferences the audience draws when something starts at or shortly after five. There are many ways a speaker could encode a message that prompts the same inferences. In particular, the speaker could choose to make her knowledge of the start time more explicit, as in (72) for example. (72)
The lecture starts at or shortly after five. (Wilson and Sperber 2012: 55 Ex 11)
Sperber and Wilson argue that such more exact descriptions of the situation turn out to be less relevant, because they require the hearer to draw inferences that are more involved and therefore more difficult. If the audience draws some inferences after hearing (71) and draws the same inferences from the more cumbersome but exact description (72), then it’s actually the approximate, rough, usage that’s more relevant, because this is the utterance that gets the desired information across with the least work for the hearer. Sperber and Wilson describe exaggeration in an analogous way. Consider (73). (73)
Bill is the nicest person there is. (Sperber and Wilson 1986: 235 Ex 106a)
We don’t take (73) as a literal report that Bill surpasses everyone else alive in niceness. Rather, the speaker wants the hearer to know that Bill is very nice indeed, and invites him to go through for himself all the inferences that might embody Bill’s surpassing niceness: that the speaker ranks Bill at the top of their mutual acquaintances; that the speaker is at a loss to describe how nice Bill is, and so forth. The hearer arrives at those inferences naturally when the speaker uses the hyperbolic utterance. By making those salient inferences so easy for the hearer to draw, the speaker can communicate these inferences in a highly relevant way. There’s no expectation that the speaker should have used a literally true statement instead.
4.1.2 Interpretation and Optimal Relevance For Sperber and Wilson, relevance has a particularly powerful role to play in language understanding because of interlocutors’ cooperative effort. Part of the very idea of communicating, Sperber and Wilson suggest, is that the
FOR INDEXING
th e p s ych o l o g i c a l tu r n 67 speaker expects that the audience will attend to the communicative act and draw inferences from it. The speaker’s interests in signaling, and the audience’s interests in understanding, are both best served if these inferences are as easy to draw as possible—that is, if the utterance is the most relevant possible stimulus for prompting the inferences the speaker intends. Now, our psychological endowments generally succeed in making us efficient communicators; this means, in part, that our utterances will in fact be optimally relevant in this way. Our examples have already hinted at this, in explaining approximation as in (71) and exaggeration as in (73). Thus, for Relevance Theory, every utterance communicates the presumption of its own optimal relevance: This, finally, is the Principle of Relevance. This simple and general principle places surprisingly strong constraints on the process of disambiguation. We summarize the result in (74). (74)
Disambiguation by the Principle of Relevance: Because utterances carry the presumption of their own optimal relevance, they are to be disambiguated in the most relevant way, that is, in the way that prompts the most inferences with the least effort.
To appreciate the force of this suggestion, imagine that you are at a meal, or even reading about one, and you find one of the diners saying (75): (75)
It looks good.
The Principle of Relevance tells you to interpret it and good in a way that allows you to draw the most relevant inferences. Your attention is on the food, and you have all of that inferential procedure in place for assessing the food and what you are going to do with it. When you consider that it could be the food, and good could be tasty, you’re readily prompted to all of those inferences that naturally follow specifically when the food looks tasty. This is a highly relevant interpretation. Relevance Theory and Levinson’s pragmatics thus agree that pragmatics resolves meaning. Both approaches reject the Gricean idea that pragmatics starts from literal meaning, and instead assume that pragmatic principles operate across a much broader scope. For Relevance Theory, the audience is always drawing inferences in order to make an utterance relevant. As part of this process, the audience interprets ambiguous or underspecified words in sensible and useful ways. However, the specifics of Relevance Theory’s account of pragmatic disambiguation are distinctive. Interpreters do not apply general interpretive schemes like Grice’s maxims or Levinson’s principles. Nor do they reason in a general
FOR INDEXING
68
the psycholo g ica l tu r n
way about all of the open-ended background knowledge that might potentially help explain the utterance. Instead, Relevance Theory says that disambiguation arises through adjustments you make to your understanding of an utterance in order to find conclusions that matter to you. As we’ll soon see, this approach to disambiguation leads to distinctive accounts of the flexibility of interpretation seen particularly in forms of figurative speech. Moreover, the approach suggests that inference in interpretation is targeted and efficient, which seems to accord with the speed and ease with which we understand utterances in discourse. At the same time, this approach is quite restrictive; disambiguation on many theoretical approaches involves more expressive linguistic rules and more flexible inference than Relevance Theory allows. As we’ll explore more systematically in Part II, the empirical phenomena that motivate these alternative approaches bring a variety of challenges for Relevance Theory.
4.1.3 Explicature, Implicature, and Grammar Relevance Theory makes one last assumption about the dynamics of interpretation. Sperber and Wilson argue that the mapping between form and meaning in language is not a code that gives a fixed translation between words and their real-world referents. Instead, grammar is organized so that words merely give evidence about the entities, properties, and relations that figure in the speaker’s meaning. Thus, a key role of the Principle of Relevance in disambiguating an utterance is to derive specific new meanings in context for the grammatical elements that make it up. To call attention to the inference required to resolve the meanings of the speaker’s words, Relevance Theory introduces the notion of the explicature of an utterance. The explicature is a level of interpretation, derived by inference from the logical form of the utterance, which spells out its enriched pragmatic meaning. Sperber and Wilson suggest that many figures of speech which might traditionally be understood in terms of Grice’s notion of implicature are better analyzed in terms of this new notion of explicature. Our discussion of (75) It looks good anticipates this point. For Relevance Theory, the fact that good means tasty in this context, where it describes the food, is not an implicature of the utterance, calculated from a general meaning for good, but is part of the explicature that indicates what the speaker means by good in the first place. Similarly, Relevance Theory suggests that the interpretive effects of alleged scalar implicatures are actually explicatures in many cases. Take (76). (76)
I drank some of the coffee.
FOR INDEXING
th e p s ych o l o g i c a l tu r n 69 For Relevance Theory, a speaker who utters (76) aims to describe the particular quantity of coffee that she drank. The word some she uses is not a grammatically encoded existential quantifier; it simply gives evidence about the quantity the speaker has in mind. In this case, we infer that the quantity of coffee is substantial but partial: enough to start feeling some extra energy, but not all that was available, and not just a sip either. For Relevance Theory, this is an explicature. It appeals to a precise new sense for some to capture the specific range of quantities at issue in (76).2 Note that this explicature captures interpretation more precisely than a theory of generalized implicatures that simply predicts that (76) implicates that the speaker did not drink all of the coffee. This seems to be an advantage of a flexible approach to adapting interpretation, such as Relevance Theory’s of enriching interpretation to facilitate inference. Sperber and Wilson even extend the idea of explicature to the case of metaphor. A metaphor they use to illustrate their point is (77). (77)
Robert is a bulldozer. (Sperber and Wilson 1986: 236 Ex 107)
In Relevance Theory, a speaker uses a metaphor, just like any other utterance, to give evidence of the consequences she wants her audience to draw. Metaphors thus contribute new specific meanings to the explicature of an utterance.3 Take (77). For Sperber and Wilson, this metaphor invites a range of inferences about Robert: that he’s forceful in action, that he will knock over anything that gets in his path, that he’s like a machine in getting stuff out of the way, or in getting what he wants. Prompting these sorts of inferences is the point of the utterance. It’s obvious that the utterance is literally false; Robert is a person, and not a piece of heavy machinery. Sperber and Wilson argue that all of the consequences that the audience might consider from a literal understanding of bulldozer are easily discarded. The hearer knows Robert is not a piece of machinery; that he doesn’t require gasoline in order to run; that he doesn’t have a driver who manipulates or decides his courses of action. Indeed, the hearer knows that the speaker did not intend him to think about Robert this way; and that the speaker could anticipate that he would recognize that he should not do so. The cost of considering and throwing away these inferences is small, whereas the many genuine insights that the hearer gets from this utterance leap easily to mind. Even a metaphorical utterance, therefore, can be optimally relevant if you think of the inferences it can prompt for you. The result is an explicature: a new, more specific 2 Our source here is Noveck and Sperber’s (2007) discussion of Some of the guests have arrived, reprinted as (Wilson and Sperber 2012: 314 Ex 15). 3 This suggestion has been developed in a variety of ways in Relevance Theory and related approaches. See Carston (2002) and Sperber and Wilson (2008), as well as related accounts in Bezuidenhout (2001) and Recanati (2004).
FOR INDEXING
70
the psycholo g ica l tu r n
meaning for bulldozer that underwrites just those inferences that spring to mind with the metaphorical interpretation of (77). As evidenced by (75), (76), and (77), explicature is thus a far-reaching theoretical construct for loosening the link between semantics and pragmatics.
4.1.4 Relevance Theory and the Psychology of Interpretation The psychological perspective of Relevance Theory represents a new approach to the semantics–pragmatics divide; it overlaps with each of the other theories we have considered but brings crucial differences from all of them. Like Horn and Levinson’s views, Relevance Theory has much simpler pragmatic principles than Grice’s, and uses them in more diverse ways in interpretive reasoning. However, Relevance Theory sides with Grice in assuming that interlocutors derive their interpretations on the fly, case by case, by reasoning about what makes sense in context. Its interpretative principles are not generalized features of the rules of language, and this is a contrast between Relevance Theory and Horn’s or Levinson’s approach. In the end, of course, Relevance Theory offers a unique characterization of interpretive inference. One difference concerns the public nature of implicatures. For Grice, the conversational context and the maxims are often strong enough for the audience to know exactly which proposition or propositions the speaker wants to get across. Relevance Theory challenges Grice’s perspective by opting for a looser notion of communication grounded in psychological principles. Speakers try to prompt relevant inferences, and audiences try to pursue them, but some mismatch in their thinking is inevitable. In many ways, Relevance Theory gets its power and interest from its claim that we can simplify our conception of linguistic knowledge by recognizing the creativity of interpretive inference, particularly once we appreciate that this inference aims at drawing interesting conclusions efficiently. In particular, Relevance Theory makes the case that its general interpretive mechanisms lead to both more precise and more diverse understandings of utterances than any of the other theories we have explored. The key question, then, which we return to explicitly in Part II and III, is whether it really is possible to dispense with richer characterizations of linguistic knowledge or with special-purpose accounts of interpretive inference across these cases.
4.2 Social Reasoning and the Psychology of Choice Where Relevance Theory attempts to ground a theory of implicatures in the psychology of understanding, it’s also possible to ground a theory of implicatures in the psychology of choice. In many respects, this strategy seems to preserve
FOR INDEXING
th e p s ych o l o g i c a l tu r n 71 more of Grice’s inferential framework. For example, it characterizes interpretation in terms of a richly psychological understanding of intention recognition. Hearers observe an utterance, reason back to the commitments of the speaker who made the utterance, and in so doing come to recognize additional contributions that the speaker puts forward indirectly. The key difference, however, is that psychological theories assume that hearers do not necessarily or exclusively draw on the Cooperative Principle and the maxims in reasoning about a speaker’s decisions about what to say. Rather, hearers draw on everything they know about the possibly limited, heuristic, and biased ways that people in fact tend to make all their decisions. We shall see that this simple difference has far-reaching consequences. Pinker et al. (2008) offer an instructive overview of the possibilities of this approach. The paper proposes to account for three particular kinds of interpretive effects by showing that speakers’ choices are not subject to general principles of rationality but instead reflect predictable but “irrational” patterns of decision-making. This broader perspective on decision-making also involves relaxing Grice’s strong conception of the cooperative nature of conversation. A crucial fact about Pinker et al.’s (2008) explanations is that interlocutors have interests that possibly diverge in key respects. Pinker et al. (2008) do not claim that the mechanisms they propose account for all implicatures, but their cases nevertheless underwrite a useful tour of psychological theories of human decisions.4 When people choose what to say, Pinker et al. (2008) argue, they make the same kinds of assumptions, simplifications, and errors as when they make all their other decisions, for the same kinds of reasons. Lee and Pinker (2010) report a series of experimental results in which the decisions of human speakers conform to the predictions of these empirical strategic models. Pinker et al.’s (2008) work thus represents a critical assessment both of intention recognition and of cooperation as invoked by Gricean Theory. The first case that Pinker et al. (2008: 833) consider is an implicated bribe. They imagine a speeder who has just been pulled over by a traffic officer and who says (78). (78)
Gee officer, is there any way I could take care of this right now?
The veiled suggestion of (78) is that the speaker would be willing to pay the officer off to avoid a formal citation. For Pinker et al., it’s crucial that this suggestion is deniable. One can at least conceive that (78) would be uttered by a naive speaker, 4 The landmark study in this tradition is Kahneman and Tversky (1979). See Camerer (2005) for a recent tutorial and Camerer and Fehr (2006) for a scientific review, and Ariely (2008) or Kahneman (2011) for surveys for a popular science audience.
FOR INDEXING
72
the psycholo g ical tu r n
out of a wishful belief that immediate action could legitimately lessen the consequences of the infraction. So there is always some doubt about what the speaker is really after. This diagnosis is incompatible with Grice’s understanding of implicatures. To see this, ask yourself whether the naive speaker’s strategy is a rational one. If it is, then the suggested bribe does not fit Grice’s definition of CI. It is not required to maintain the assumption that the speaker is rational, since we could also explain (78) as a rational move by assuming that the speaker made it naively. Of course, nothing in Grice’s theory prevents us from supposing that the speaker is canny, sophisticated, or on the make, and using that assumption to shape our understanding of what the speaker says. But the inferences we draw are not CIs, because CIs for Grice require a specific basis, namely, the CP and rationality. In fact, however, Pinker et al. (2008) suggest that (78) is not really a rational discourse move for a naive speaker after all. The goal of the naive speaker is to ask an honest question. She should be aware that (78) is likely to be understood— or perhaps misunderstood—as offering a bribe, and for that reason she should avoid the utterance. This error in judging her audience seems like part of what makes the naive speaker naive. She has no idea that other people might aim their utterances at such dishonest purposes, and so she cannot anticipate how her utterance will sound to a more jaded audience. In social interactions, however, idealized rationality requires agents to take one another’s perspective when necessary, and to think through others’ reactions to what they will do. The naive speaker’s mistake comes from not thinking the utterance through deeply and carefully enough. This is the sense in which it is irrational. If speakers regularly make such errors in deciding what to say, we need to apply a different standard in interpreting utterances than Grice’s. In particular, for cases such as these, we needn’t confine ourselves, when recognizing implicatures, to what’s required to maintain the assumption that the speaker is rational. We may also have to consider plausible but irrational moves by speakers. It should not be surprising that people might decide naively on uncalculated moves in strategic interactions such as these. Behavioral economics reveals lots of similar errors in people’s decision-making. One revealing experimental case concerns beauty contest games, where subjects must strategize about the aggregate behavior of groups they are members of. These games take their name from Keynes’s metaphorical explanation of speculative bubbles in asset prices. Speculators buy stocks not because of the inherent value of the company’s balance sheet and future revenues, but just because of their expectation of what subsequent buyers will be willing to pay for the stock. As Keynes sees it, by considering how good stocks look to others, rather than the underlying fundamentals, they
FOR INDEXING
th e p s ych o l o g i c a l tu r n 73 turn the stock market into a kind of beauty contest. Famously, speculation can create a bubble where assets are traded at distorted high prices simply because traders expect prices to rise further. Corrections can be very slow in coming and can take the form of devastating crashes in prices. As Keynes is purported to have put it, “the market can remain irrational longer than you or I can remain solvent.” A typical beauty contest game, as studied experimentally by Nagel (1995), asks a group of subjects each to choose a number between 0 and 100. The subjects whose guesses are closest to two-thirds of the average value across the group share a large payoff; the other players get nothing. A player who thinks the problem through to its logical conclusion, and assumes that everyone else will do the same, should pick 0 as her choice, expecting everyone else to pick 0 as well.5 Of course, there’s little reason to expect people to make these idealized choices. Naive players might not think their choices through so carefully. They might simply identify a range of numbers they consider low, and choose a number from this range more or less at random. Once there are such players in the mix, it’s no longer optimal to play 0. Instead, optimal players should make a higher guess, in part because they anticipate the high guess of naive players, and in part because they now expect analogous, higher guesses from the other optimal reasoners in the group. The presence of naive reasoners also lessens the advantage of following the optimal strategy. To produce a winning guess, the optimal strategy needs to have accurate information about the distribution of naive players in the group. If we think the group is too smart, we guess too low. If we think the group is too naive, we guess too high. But the naive approach can also turn out winning guesses, if the threshold it uses is appropriately responsive to behavior of the broader group. What matters, it seems, is learning a good model, not having a good decision-making mechanism, and many heuristics seem equally good in this respect. Not surprisingly, experiments find that human subjects consistently play higher scores than 0 in this beauty contest game, and accordingly that players need to play a higher score than 0 in order to win (Nagel 1995; Ho, Camerer and Weigelt 1998). The conclusion is that there are limits to idealizations of rationality in explaining human behavior. To describe what people actually do, we often need to allow for naive people that make simple, heuristic decisions that fail to produce good responses in communities of ideally rational agents but that can often yield effective results in practice against other fallible agents. 5 Here’s the proof. The rational choice, if you expect some distribution of moves by the other players, is to play two-thirds of that average. But if the average is higher than 0, the strategy you’re playing will give other players a reason to lower their guess downwards, in anticipation of the move you will make. Only if everyone plays 0 are the players’ choices stable.
FOR INDEXING
74
the psycholo g ical tu r n
Pinker et al. (2008) suggest that this same generalization applies to language use. In particular, it’s important to consider the naive heuristic of using utterances literally, oblivious to possible indirect interpretations that more sophisticated speakers might consider. In the case of a veiled bribe as in (78), Pinker et al. (2008) present a mathematical model of a case where the indirect utterance really is deniable. In response to the utterance, an honest officer is left with substantial uncertainty about whether he’s dealing with a dishonest briber or an honest but naive speeder who just seems to be offering a bribe. His best course of action is to ignore the offer. By contrast, a corrupt officer is left with enough evidence of the speaker’s complicity to go along with the bribe. The good outcome in both cases makes the indirect bribe the best course of action for the sophisticated speaker. Pinker et al. (2008) suggest that a wide range of offers and threats are commonly veiled this way. Differing sophistication of speakers is only one reason Pinker et al. (2008) offer for speakers’ use of calculated indirection. Other cases, they suggest, are continuous with the apparently paradoxical choices that are endemic in the maintenance of human social relationships. As an example, Pinker et al. (2008: 835) consider a customer at a busy restaurant who tips the host to jump the queue to get a table. In this setting, it seems very awkward to say something explicit like (79) while handing over a $20. The customer would be more likely to use an utterance like (80), which seems much more comfortable. (79) (80)
Please could you seat us right away. I was wondering if you might have a cancelation.
By using (80), and appealing indirectly to the solicitousness of the restaurant in a vague and general way, the speaker seems to undermine the otherwise clear meaning of offering the tip. This use of an utterance intentionally to invest an interaction with ambiguity falls outside the Gricean perspective on CIs—which instead enrich a speaker’s meaning in recognizable ways. For Pinker et al. (2008), this interpretive effect is a reflection of our social feelings, since it is social feelings rather than rational calculations that motivate many of our interactions with one another. Here, Pinker et al. (2008) suggest, the interlocutors prefer indirection because of the distinctive value people experience in participating in smooth relationships, over and above the outcomes those relationships provide. In particular, a direct utterance like (79) suggests that the host is at the customer’s disposal; the utterance is in conflict with an alternative conception of their relationship where the two have limited responsibilities to one another but do favors for one another as they can. On this interpretation, (79) will engage the customer and the host in a fraught renegotiation of
FOR INDEXING
th e p s ych o l o g i c a l tu r n 75 their relationship. The indirect utterance, by contrast, allows interlocutors with differing senses of their relationship to leave their differences unresolved and keep their relationship on its current terms. They judge this a better outcome than understanding each other but discovering an awkward misalignment. We can draw an analogy to a pattern of results in behavioral economics typified by experiments on ultimatum games, where it’s also important to appeal to social feelings as well as rational calculations in explaining human decisions. The classic experiment pairs two anonymous subjects who interact only for a single interaction (Güth, Schmittberger and Schwarze 1982). The experimenter offers the first participant a sum of money, which he can split as he chooses with the second participant. The second participant must then make a decision: she can accept the offer, in which case the offer becomes permanent and both players keep the money; or she can reject the offer, in which case both participants get nothing. Game theory suggests that the first participant ought to offer the minimum possible sum to the second, and the second should always accept: any offer leaves the second participant ahead of the status quo, and so she has no reason to reject it. But empirically, people do reject offers that strike them as unfairly low, and in any case people tend to make offers that might be regarded as generous or at least fair. Scientists disagree about the best explanation of human choices in ultimatum games (Camerer and Fehr 2006). Anger at a low offer may lead to an irrational rejection; seemingly generous players may just be cold-hearted rationalists hedging against such possible outcomes. Players may value fairness, over and above the monetary payoffs they receive. Those values may make observed play a rational equilibrium for real people. Participants may not optimize their outcome, but simply aim for an outcome that is good enough. There are difficulties with all these explanations, and they are not mutually exclusive. But any of these explanations points to the importance of tracking social choices not solely in terms of practical outcomes, but also in terms of the human feelings and conceptualizations that are implicated in the choices. It’s this broad point that Pinker et al. (2008) appeal to in their explanation of (80). With their third case, Pinker et al. (2008: 837) argue that our indirections sometimes simply reflect the different framing of utterances as literal and public or indirect and off-the-record. In particular, they suggest that sexual innuendo is often motivated this way. It’s the end of a date, and the couple are walking their way home. One of them, arriving at the door of the apartment building where he lives, asks the question in (81). (81)
Do you want to come upstairs and see my etchings?
FOR INDEXING
76
the psycholo g ical tu r n
Pinker et al. (2008) don’t suggest explaining the indirection of (81) by appeal to the uncertainties associated with naive speakers or problematic relationships: (81) leaves too little doubt about the speaker’s intentions. Instead, they suggest, what matters is the speaker’s choice to leave his overtures off the record. If we think, for example, that—however ambiguous or vague our literal meanings are, and however obvious our innuendo is—what we say is certain and out there for good, whereas what we implicate is uncertain and deniable, then that view will shape how we count the costs and benefits of literal versus indirect utterances. Such framing effects abound in behavioral economics. A typical case explores how institutions can most effectively shape people’s behavior through pricing. For example, Homonoff (2013) considers the example of a metropolitan area that wants to decrease customers’ use of disposable plastic bags. (She studies Washington, DC, in particular.) One option is to give customers a discount for bringing their own bags. Another is to discount the prices but charge customers a tax for any bags they need. Customers fare exactly the same under the two schemes: if you walk into the store without a bag, you pay slightly more for your basket of goods; if you walk into the store with one, you pay slightly less. But the second scheme proves more effective in getting customers to bring their own bags. It turns out people are more averse to outcomes that are framed as losses than they are keen on outcomes that are framed as gains. We’ll let you draw your own conclusions about how that might apply to utterances like (81). In taking this psychological approach to implicature, They break with Grice in two important ways. They depart from the idealization inherent in Grice’s Cooperative Principle by emphasizing the biased and heuristic reasoning that may well underpin speakers’ choices. However, Pinker et al. (2008) also depart from the generality inherent in the Cooperative Principle by emphasizing the empirical background knowledge that informs hearers’ understanding of speakers. To understand indirection, they suggest, you have to know how sophisticated your interlocutor is likely to be, what your interlocutor is likely to believe and desire, and what your interlocutor is likely to expect of you. This represents an important way that empirical psychological facts can enter into interlocutors’ understanding of one another, even in situations where speakers reliably make good decisions. Of course, Grice’s understanding of implicature is compatible with drawing on interlocutors’ shared background knowledge, including the context and accepted purposes of the conversation. Geurts (2010), for example, emphasizes how important background assumptions about the speakers’ knowledge and interests are to calculating implicatures on a Gricean approach. We suspect, however, that as one appeals more and more to empirical knowledge of this
FOR INDEXING
th e p s ych o l o g i c a l tu r n 77 kind in drawing inferences about the significance of an utterance, one gets a theory of implicatures that is increasingly divorced from Grice’s larger theory of communication. For example, this appeal to psychology makes uncertainty a crucial feature of alleged implicatures. Hearers are uncertain about what speakers know, what they want, and what options they have considered in attempting to make a contribution to conversation through a particular utterance. Different assumptions about the speaker can lead to different conclusions about the content of their communicative intentions. This uncertainty gives implicatures an indefinite, open-ended character, similar to the open-ended and approximate interpretations of underspecified utterances in Relevance Theory. Russell (2012) explores this idea in detail, by spelling out a Bayesian probabilistic model of utterance understanding. He uses the formalism to argue for a new and broader definition of implicature which—unlike Grice’s definition—accommodates the uncertainty associated with inference and indirection. Such issues cannot be ignored on a psychological approach.
4.3 Assessing the Psychological Turn It would be very surprising if we could not learn a lot about conversation by investigating the psychological mechanisms behind it. In fact, as interlocutors we can often use our social knowledge in real time to gain insights into our interlocutors. We find particularly compelling instances of such possibilities in Difficult Conversations, a guide to success in problematic interactions, compiled from the fieldwork and interventions of the Harvard Negotiation Project (Stone, Patton and Heen 1999). If your goal is to communicate effectively, they argue, you have to learn to listen for the unstated problems that really motivate people’s utterances. You have to explicitly address the real, underlying problems, even if that means foregoing your response to what others actually say. As an illustration, consider Stone et al.’s (1999) advice for dealing with outbursts that accuse you of purposefully causing harm. They cite (82) as an example of such an accusation: (82)
You act like you’re my father or something. You have this need to control me or put me down. (Stone et al. 1999: 45)
This utterance—a representative case from their fieldwork—catches the speaker lashing out in embarrassment after being criticized in public by her partner. Communicating effectively, Stone et al. (1999) suggest, involves focusing on the
FOR INDEXING
78
the psycholo g ical tu r n
feelings the speaker exhibits, rather than the substance of the accusation she might be making. (83)
Listen Past the Accusation for the Feelings. Remember that the accusation about our bad intentions is always made up of two separate ideas: (1) we had bad intentions and (2) the other person was frustrated, hurt, or embarrassed. Don’t pretend they aren’t saying the first. You’ll want to respond to it. But neither should you ignore the second. And if you start by listening and acknowledging the feelings, and then return to the question of intentions, it will make your conversation significantly easier and more instructive. (Stone et al. 1999: 56)
We have found this to be useful advice—but we cite this example for its theoretical interest. If Stone et al. (1999) are right about how conversations work, we can’t use the Cooperative Principle and the Maxims to understand the real subtext of our interlocutors’ utterances. We have to understand and engage with the peculiar psychological biases and emotional dynamics that lead to these problematic utterances. In particular, according to Stone et al. (1999), speakers’ accusatory utterances are mistakes. They do not serve their speakers’ interests. Under the circumstances, speakers would do better simply to state the feelings that resulted from interlocutors’ actions, and then enter into an open dialogue about what interlocutors’ intentions might have been. In other words, these are utterances where a speaker’s emotions leak through their choices. In effect, frustration and anger reveal themselves through the psychological mechanisms that prompt prereflective and counterproductive choices of accusatory messages, just as those emotions often reveal themselves, prereflectively and counterproductively, in other kinds of outbursts. But listeners’ instincts to defend themselves against such accusations are just as much mistakes. Listeners must learn to listen better. Indeed, Stone et al. (1999) suggest that interlocutors need better perspectives and social skills, not just better ways to track the conversation or plan their contributions to it. If your conversation depends either on hinting about problems or lashing out in pain, it’s already off the rails. They write: (84)
Don’t Rely on Subtext.. . . Bringing it up by not quite bringing it up seems a happy medium between avoiding and engaging. It is a way of doing neither and doing both. The problem is, to the extent you are doing both, you’re doing both badly. You end up triggering all of the problems you worried you’d create by bringing it up, without getting the benefit of clearly saying what you want to say. (Stone et al. 1999: 191)
FOR INDEXING
th e p s ych o l o g i c a l tu r n 79 Stone et al.’s (1999) critique of rhetorical questions—presented through the hypothetical example (85)—illustrates the ineffectiveness of implicature in a way that rarely makes it into the pragmatics literature. (85) (86)
Is golf so important that you need to play it this often? This is a classic example of a statement masquerading as a question. It’s obvious that the meaning of the comment is conveyed in the subtext. What is less obvious is what the meaning is supposed to be. Your tone conveys anger or frustration. But it’s not clear what is causing the anger or what your husband is supposed to do about it. Are you angry that your husband is engaged in a meaningless sport rather than community service or household chores? Are you angry that he’s not taking you along? Are you angry that you aren’t spending enough time together? How would he know? (Stone et al. 1999: 192)
This perspective helps to bring home the opposition between psychological approaches to conversation and the Gricean appeal to the Cooperative Principle. Stone et al.’s (1999) analysis makes sense only if interlocutors are normally ignorant of many inferences they can and should be drawing about their partners’ mental states—inferences that science can elucidate. Indeed, grounding an account of interpretation directly in psychology rules out key parts of Grice’s pragmatic picture. For example, the psychological turn makes it very difficult to describe utterances as flouting pragmatic principles. For Grice, pragmatic principles reflect norms that agents can choose to follow or not to follow. But for Sperber and Wilson (1986) or Pinker et al. (2008), pragmatic principles represent fundamental generalizations about how people process information. One can no more opt out of the laws of psychology for effect, than one can opt out of the laws of physics for effect. As Sperber and Wilson (1986: 162) put it, “Communicators do not ‘follow’ the principle of relevance; and they could not violate it even if they wanted to.” Thus, we have to explain figurative utterances—metaphor, irony, exaggeration, and the like— with reference to the same psychological principles that govern the enrichment of literal meaning. Relevance Theory, among other psychological approaches, is explicit in responding to this challenge. Grounding interpretation in psychology also undermines the sharp distinction in Grice’s pragmatic picture between the meaning a speaker intends to communicate with an utterance and the other information a hearer gets from observing it. Psychological approaches agree that the audience recognizes speakers’ communicative intentions, and can distinguish between the information the speaker wants to get across and the other information that the speaker
FOR INDEXING
80
the psycholo g ical tu r n
reveals through his actions. But psychological approaches cannot use this distinction to develop an account of content and inquiry of the sort advanced by philosophers such as Stalnaker (1973, 1979) and described in Chapter 2. For the most radical of psychological approaches, even the attribution of public propositional content to utterances is problematic. Instead, psychological approaches cast all interpretive inference as inherently uncertain, and inherently intertwined with the audience’s ramifying understanding of unfolding events. Thus, even if psychological approaches offer profound insights into the nature of conversation, the relationship of psychology to utterance interpretation remains problematic. To explain utterances through psychological mechanisms is to contest the appeal to linguistic principles made by theorists like Horn and Levinson. It is also to assume that particular understandings of creative utterances never recruit specific interpretive mechanisms. And it blurs the lines between the meaning of an utterance and the information the speaker conveys. These claims are both surprising and tendentious. As always, we will address these claims in detail in what follows. For now, we note that while there is strong evidence for some of the empirically based psychological reasoning we have explored in this chapter, we will ultimately want to distinguish it both from semantics and from collaborative reasoning. In particular, on our view, conventional grammar, imaginative engagement, and even cooperative disambiguation are best described in other theoretical terms. It’s only by doing so, we suggest, that we can acknowledge the psychological insights we have surveyed in this chapter while maintaining an account of communication founded in terms of interlocutors’ efforts to negotiate a mutually agreed answer to questions of interest.
FOR INDEXING
Summary of Part I and Projection
Pragmatic approaches to meaning urge us to attend closely to the strategies of speakers in conversation and the corresponding interpretive reasoning of their audiences. Our intuitions about utterances seem, in part, to be intuitions about idealized speakers’ choices, goals, and plans. They seem, in part, to reflect our broader psychology as people. And they also seem to depend on linguistic rules that function to balance the interests of the different participants in a communicative situation. Refining these common-sense truisms leads to a diverse set of principles for explaining utterance interpretations. Where these principles apply, utterance interpretations will go beyond conventional content and reveal additional meaning—conversational implicatures. The theories we have surveyed in Part I show broad similarities in the interpretive phenomena they target and in the interpretive principles they propose. They do not, however, establish a single coherent framework for pragmatics. In fact, we find deep scientific and philosophical disagreements. Grice’s original proposal, as described in Chapter 2, integrates linguistic knowledge and conversational implicature within one, overarching approach to meaning. For Grice, utterance interpretations record the content the speaker intends to contribute to the conversation. Some of these contributions happen to be conventional: established by arbitrary links between linguistic elements and their interpretations that are in common currency in the community. Others turn out to be creative: calculated, on the fly, by appeal to normative rational principles that speakers usually follow but sometimes flout. Ultimately, however, speakers’ communicative intentions are planned and recognized by common mechanisms whatever information they draw on. Such mechanisms represent the point of contact between linguistic knowledge and psychological processes. By contrast, all of the theorists whose work we considered in Chapter 3 and Chapter 4 contest this characterization of the semantics–pragmatics interface. Take Horn (1984) or Levinson (2000). They propose interpretive principles
FOR INDEXING
82
summar y of par t i and projection
similar in character to Grice’s Maxims, but they avoid positing a distinctive connection between linguistic knowledge and the psychological mechanisms of rational action. Instead, they argue that the rules of language say when and how the interpretive principles apply. The result is that implicatures are straightforwardly part of language. It’s reasonable for a philosopher to assess such proposals as radically different from Grice’s. Alternatively, consider psychological approaches like those of Sperber and Wilson (1986) or Pinker et al. (2008). These theories explore interlocutors’ open-ended abilities to reason in conversation, but they don’t ground linguistic meaning in speakers’ communicative intentions the way Grice does either. Fundamentally, they ask what speakers reveal in the choices they make, and what inferences they ultimately prompt on the part of their audience. A philosopher might reasonably be skeptical that such information is tied to Grice’s normative notion of rationality, or that it contributes to conversation in the way Grice hoped to characterize with his notion of speaker meaning. As we saw, psychological accounts may ultimately have to abandon the idea of having a definite propositional content that speakers contribute to conversation. Even if they do not, psychological accounts do not see pragmatics as contributing information that’s comparable to conventional meaning. Though neither linguistic approaches nor psychological ones allow pragmatics to augment semantics the way Grice does, as a matter of fact the two camps demarcate linguistic meaning in very different ways. Linguistic theories generally group the CIs with linguistic meaning, while the psychological theories we survey in Chapter 4 generally do not. We can now appreciate the challenge of getting clear on the relationship between semantics and pragmatics. We have difficult empirical questions to answer about the nature and scope of the rules of language. But it’s not enough to distinguish the interpretive effects of different kinds of knowledge and reasoning in language use. The way we interpret the empirical results depends on our take on diverse philosophical issues—on the nature of content and representation, on the relationship of meaning and agency, and on the bases for human interaction and collaboration. Our philosophical conclusions will commit us not only to specific answers about the mechanisms of language use but also to a broader perspective in the philosophy of language and mind. Our agenda for the rest of the book is thus to bring empirical characterizations of utterance interpretation into closer contact with philosophical issues and arguments. In particular, in Parts II and III of the book, we survey what’s known about the status of various interpretive effects without prejudice to the overarching theoretical and philosophical questions, and develop a broader and
FOR INDEXING
su m m a r y of p a r t i a n d pro j e c t i on 83 better-informed perspective. Part IV offers a philosophical context to make sense of what we find. We begin by exploring a range of interpretive principles that enrich semantics but vary from language to language and are needed in order to recover logical form. The goal here is to characterize the totality of linguistic knowledge that goes into interpretation. We will suggest that we can be very eclectic about the linguistic rules that factor into interpretation; we needn’t assume they must take the form of overarching pragmatic principles, rules for licensing syntactic structures, or bare-bones semantic rules for reference and truth. We can allow that the rules of language are more diverse. As a matter of fact, we should all be for this conclusion, because the linguistics literature offers lots of evidence in support of it. Afterwards, in Part III, we will argue that we also have good evidence that the inferential mechanisms recruited in language interpretation are quite diverse and eclectic. We’ll try to convince you that you can’t put together the inferences required for metaphor with the inferences required for irony or those required for hinting, and so on for other figures of speech. Since all of these practices are dramatically distinctive, as we will argue, it follows that overarching frameworks like the Cooperative Principle or the Principle of Relevance can’t be the whole story. But there was never any reason to expect this kind of theoretical uniformity; we are creative, flexible, idiosyncratic thinkers, and we know how to guide one another in using all our skills. By the end of Part III, then, we will have completed our break with the Gricean and neo-Gricean programs, a break adumbrated already over these last few chapters. We find eclectic linguistic rules that explain many alleged CIs, and we find genuinely heterogeneous interpretive reasoning principles that explain many others. Thus, in Part IV, we will invite you to conclude that the category of conversational implicature does no theoretical work. Pragmatics can be, at most, a theory of disambiguation; pragmatic reasoning never contributes content to utterances. However, because a complete picture involves such heterogeneous linguistic rules and such diverse imaginative engagement—and the ever-present background of collaborative reasoning—we cannot endorse a single crisp semantics–pragmatics divide. Instead, we will acknowledge multiple competing standards we can usefully appeal to in characterizing collaboration, inquiry, meaning, and understanding in conversation.
FOR INDEXING
FOR INDEXING
PA RT II
The Interpretive Effects of Linguistic Rules
FOR INDEXING
FOR INDEXING
Introduction to Part II
As we saw Part I, extant theories differ greatly in their accounts of the relationship between the interpretation of utterances and linguistic meaning. Grice inaugurated a philosophical approach, grounded in appeals to rationality and the norms of cooperation. By contrast, the linguistic approach, epitomized by the work of Horn and Levinson, appeals to distinctive interpretive principles that are part of the language faculty. Yet another approach, associated with Sperber and Wilson and many others, appeals to empirical facts about the psychological processes through which we make decisions and understand the world. Despite their differences, these approaches all assume that a suitable mechanism of pragmatic inference plays a pervasive and powerful role in deriving the rich and detailed interpretations for utterances in context. In particular, they invariably give accounts of the interpretation of specific utterances in context that start just from a bare-bones specification of linguistic meaning. In this Part II of our book, we challenge this bare-bones approach to linguistic meaning. When interlocutors derive the interpretation of utterances, we suggest, they must factor in a wide range of interpretive constraints over and above semantics as traditionally assumed in discussions of implicature—constraints that are associated in essentially arbitrary ways with the linguistic elements of the utterance. In diagnosing these conventions, we will draw on results from linguistics to emphasize the diverse kinds of interpretive constraints involved. Conventional meaning, it seems, goes far beyond truth-conditional meaning, or even the simple kinds of conventional implicatures that Grice invoked, for example, to distinguish but from and (Grice 1989: 25–6). Some rules, governing discourse coherence, seem to explicitly specify the possible actions that particular utterances can be used for; see Chapter 6. Other rules, governing presupposition, seem to explicitly specify the possible contexts in which particular utterances can be used; see Chapter 7. Still more rules, governing information structure, seem to explicitly specify what relationships particular utterances
FOR INDEXING
88
int ro duc t ion to p a r t ii
can bear to salient alternatives that have been or might be uttered in the ongoing conversation; see Chapter 8. Indeed, conventional meaning seems so eclectic and variable that we begin Part II in Chapter 5 with a brief methodological justification explaining how linguists have been able to characterize these interpretive effects and how we can argue that they are, in fact, conventional. We then turn to the cases with which we open the book—indirect speech acts in Chapter 6, temporal interpretation in Chapter 7, and relationships of contrast or disagreement in Chapter 8. We think these cases provide clear examples that motivate representative classes of linguistic knowledge, but we have no general arguments that we can apply to the category of CIs as a whole. Indeed, some of the most important cases are undoubtedly more complex than the clear cases with which we begin. For example, we conclude Chapter 8 by taking stock of the well-studied case of scalar implicatures. Surprising controversy surrounds the question of when a scalar implicature genuinely figures in the content a speaker intends to communicate, in part because such judgments run up against the psychological complexities we briefly explored in Chapter 4. However, key cases of scalar implicature do seem to contribute public content to conversation. Explaining these cases will require us to appeal to the full gamut of linguistic knowledge—truth-conditional meaning, discourse relations, presupposition, and information structure—and to acknowledge the important sense in which these conventions allow interlocutors to communicate efficiently. These more problematic cases serve as a reminder that the project of characterizing the linguistic rules that apply to particular forms and phenomena is an empirical one, which we should not expect to yield the same results in all cases. We cannot hope to show that any of the approaches to pragmatic interpretations from Part I is incoherent or impossible. But if we are correct in the linguistic conventions we identify in Chapters 6–8, these approaches are substantially diminished in their coverage. For, we will argue, in all the cases we consider, where alternative approaches have postulated pragmatic processes of enrichment, what’s really going on is disambiguation: finding the right reading of the utterance, understood as a grammatically specified pairing of form and meaning. We’ll have to look more broadly into the range of interpretive effects, in Part III, if we are to find evidence of more substantive kinds of interpretive inference. But we’ll also have to look more deeply into the place of meaning in cognition and collaboration, in Part IV, to understand whether we should view disambiguation, which we observe everywhere in language use, ultimately as a vindication of the traditional Gricean understanding of the relationship of semantics and pragmatics, or as a challenge to it.
FOR INDEXING
5 The Scope of Linguistic Conventions
To diagnose the role of linguistic knowledge in interpretation, we need to be able to factor out the possible contributions of pragmatics—on any of the diverse and far-reaching models we saw in Part I. This is obviously a difficult empirical problem, which calls for subtlety and care. Fortunately, as we explain in section 5.1, a wide range of research characterizes aspects of utterance interpretation in precise, but largely theoretically neutral ways. This research lets us to pinpoint specific interpretive regularities, without necessarily explaining why these interpretive regularities arise. It’s only by surveying and critiquing the specific analyses that we can assess the role that conventional linguistic knowledge might play in any particular case. As we outline in section 5.2, we think there is decisive evidence that conventions are involved in key cases. Our arguments diagnose the presence of linguistic conventions by appeal to interlocutors’ need to coordinate on the public contributions utterances make to conversation. As we review in section 5.3, this methodology aligns both with the understanding of convention going back to Lewis (1969) and with Grice’s own tests for the presence of conventional implicatures. Moreover, we argue that the presence of coordination is incompatible with extant models of general interpretive principles and the general motivations of Gricean and neo-Gricean pragmatic programs.
5.1 Empirical Research: Insights and Frameworks To explain the empirical resources that we use to characterize interpretation, we begin with a brief tour of Coherence Theory. Coherence Theory grows out of the theory of interpretation as abduction initially developed by AI researcher Jerry Hobbs (1990). We follow the exposition in Andrew Kehler’s
FOR INDEXING
90
the scop e of ling u i s t i c conven t i on s
Coherence, Reference and the Theory of Grammar (2001).1 (sdrt), a particular formal approach to discourse coherence that systematizes a wide range of fine-grained descriptive insights into utterance interpretation in a flexible and relatively non-committal formal framework. Coherence Theory resembles Relevance Theory in that both approach interpretive inference as an application of general mechanisms that let people understand the world around them. However, the two approaches offer different overarching frameworks for this fundamental problem of understanding. Relevance Theory supposes that people interpret events by drawing inferences that matter for them. Coherence Theory, by contrast, assumes that agents explain events in terms of prior background knowledge. Something is coherent if it fits in with what you know already; if it’s easy to explain with a few assumptions; if it exemplifies patterns familiar to you. Take the case of a waiter bringing out a dish of food in a restaurant, whose understanding on Relevance Theory we discussed in connection with example (69). Coherence says that you make sense of such an event by fitting it in with your expectations, perhaps thinking: “Hey, that’s the food we ordered; at restaurants they always bring the food you order to the table. And so, this is not surprising; it makes sense.” Maximizing coherence is different from maximizing relevance. A puzzling event may be incoherent—and thus problematic from the standpoint of Coherence Theory—even if it’s obvious that the event has no consequences that affect your interests, and so is unproblematic from the standpoint of Relevance Theory. Conversely, we may readily arrive at an explanation of a familiar event, maximizing coherence, while failing to notice the implications of this event for our present plans, and thus failing to maximize relevance. For Hobbs and Kehler, the audience understands an utterance so as to maximize coherence. In effect, the audience must come up with the most sensible answer to the question, “Why would the speaker have said that?” Disambiguation is a side-effect of this process. To the extent that a specific resolution of ambiguity leads to an explanation that fits more naturally with what the audience knows about the speaker, the language, the context, and the conversation, the resolution figures in a maximally coherent interpretation. How the audience finds coherence is a flexible and eclectic process. There is no limit to the background knowledge, the familiar patterns, or the explanatory assumptions that the audience can invoke in explaining the speaker’s utterance. For example, if you are interpreting an utterance, you might use background 1 We are also heavily influenced by Asher and Lascarides’s (2003) Segmented Discourse Representation Theory.
FOR INDEXING
the s cop e of l i n g u i s t i c conven t i on s 91 knowledge about grammar: it makes sense that a speaker would use linguistic constructions that are familiar from your experience. You might use background knowledge about patterns of use and the things that people tend to do in the language: it makes sense that a speaker would act in familiar ways. And you might use particular knowledge about the speaker and his goals and beliefs and circumstances: it makes sense that a speaker would act in ways consonant with the interests and perspectives you’re already aware of. In particular, then, Coherence Theory admits many different approaches to CIs. It is compatible with Horn’s view that CIs are linguistically encoded: if Horn is right, audiences know that certain expressions are subject to specific interpretive principles, and audiences can therefore appeal to those principles to get a coherent explanation of certain utterances. Coherence Theory is also compatible with Relevance Theory’s idea that CIs require creative exploratory inference; audiences know that speakers may want to trigger a broad class of inferences, and audiences can make sense of an utterance by finding such inferences and assuming that the utterance was intended to prompt them. What’s more, Coherence Theory is compatible with Grice’s view that you retrieve a CI by recognizing a specific presumption the speaker is intending you to make; assuming that the speaker intends you to recognize some specific assumption— and otherwise appealing to powerful and general principles like the CP—can lead to a simple and coherent explanation of a speaker’s utterance. In short, Coherence Theory is actually neutral about the factors that influence disambiguation or lead to CIs—a point that echoes the contrast between Gricean theory and common-sense views of disambiguation like Blackstone’s (1765). In developing predictions to cover specific cases, Coherence Theorists aim to characterize the knowledge the hearer uses as perspicuously as possible; generality is a secondary consideration. Coherence Theorists thus face little pressure to commit to the status of the inferences they posit. For example, consider the dynamics of reference across discourse, as exhibited by (1) and (2): (1) (2)
The President nominated Jones. He expected a quick confirmation. Jones was nominated by the President. He expected a quick confirmation.
In the absence of overriding cues from the broader context, He in (1) is the President, and He in (2) is Jones. We can approach this in Coherence Theory by simply observing that—at least in certain circumstances—there is a preference to resolve a subject pronoun in one sentence to the subject of the preceding sentence. We don’t have to say what the reason for this preference is. It might turn out to be the only reasonable choice, given the kinds of information and goals that people typically have in conversation. It might turn out to be a side-effect
FOR INDEXING
92
the scop e of lingu i s t i c conven t i on s
of the processing mechanisms people have for attending to entities as they track contributions to conversation. Or it might just turn out to be rule of English, one that doesn’t even carry over to other languages. In documenting and systematizing interpretation, such broader theoretical questions are secondary. In fact, we may not be in the position to answer them until we understand more about range of possible interpretations, within and across languages, and the general constraints imposed by human cognitive architecture and the norms of rationality. Of course, Coherence Theorists hope that accumulating neutral models of a diverse and representative set of phenomena will ultimately make it possible to revisit those questions more authoritatively.
5.2 A Tour of the Data Although formal theories of interpretation can often avoid substantive commitments about the relationship of semantics and pragmatics, they do offer a range of accumulated evidence about interpretation. In particular, we will argue, this evidence shows that much of what researchers have classified as CIs are in fact expressions of interlocutors’ linguistic knowledge. The cases with which we began this book provide a road map to the knowledge involved. In Chapter 6 we explore speech acts, particularly the indirect speech acts epitomized by (25), the polite request repeated here: (25)
Can I have the French Toast?
Such examples surely showcase interlocutors’ abilities to collaborate with one another and get things done together. But, we argue, they just as surely showcase interlocutors’ expertise at coordinating with one another—acting in accord with their shared expectations for how their activity should proceed, and thereby aligning their actions to fit one of the many possible strategies that would lead to a good outcome. These shared expectations, we argue, constitute conventions and are in fact tied to particular linguistic forms. Some of these conventions, we argue, describe the kinds of acts that individual utterances accomplish. Others describe the possibilities a particular language allows for organizing discourse into patterns of inquiry, argument, and negotiation. This goes beyond syntax and semantics as usually conceived, but, we will argue, it is part of speakers’ linguistic grammar nonetheless. In cases of ambiguity, general pragmatic principles may help an audience to identify what action the speaker is undertaking. But it is a mistake to attribute such interpretive effects to the action of pragmatic principles alone.
FOR INDEXING
the s cop e of l i n g u i s t i c conven t i on s 93 In Chapter 7, meanwhile, we explore the dynamics of presupposition and anaphora, as epitomized by the kinds of inferred temporal relationships between successive clauses we find in narratives such as (26), repeated here. (26)
Oil prices doubled and demand for consumer goods plunged.
Anaphors are expressions which pick up values from the discourse context; presuppositions are constraints on their possible values. We will defend a widely held linguistic theory, on which anaphors and presuppositions are associated with particular linguistic items by the grammar, so these ingredients of interpretation are a matter of linguistic convention. Presuppositions differ from the speech act conventions we survey in Chapter 6 in that presuppositions specify information about the entities evoked by the utterance that must be supported by the context in which the utterance is used. Thus, when presuppositions are satisfied, they are not directly visible in the contribution the utterance makes to conversation. The interpretive effects of presuppositions are visible indirectly, however: they force anaphoric reference and other interpretive ambiguities to be resolved so as to satisfy them. As when we disambiguate speech acts, we may need pragmatic principles to recognize the preferred resolution of anaphora and presupposition. But accounts of pragmatic inference err when they attribute the specifics of interpretation to the action of the principles themselves. The content of interpretation must be licensed by the conventional, rule-governed dynamics of discourse anaphora. Chapter 8 presents the third case study: we consider information structure, the linguistic devices we have to signal what’s important about what we say and how it fits in to the dynamics of interaction. Information structure is typically encoded in variations of word order, syntactic elaboration, or intonation. We will analyze the sense of contradiction in Kripke’s (1978) example (27), repeated here, as a prototypical manifestation of information structure. (27)
Well, it looked red.
Information structure represents yet another dimension of conventional meaning in utterances. It describes the inferential connections that are significant to a speaker: What is her utterance about? What question does it answer? Who posed that question and who will follow it up? What alternative answers might one have expected? And how would those alternatives differ? These constraints describe both the content that we associate with an utterance and the action that we understand the speaker to be taking with it, and can have powerful and indirect ramifications for what meanings seem most plausible.
FOR INDEXING
94
the scop e of ling u i s t i c conven t i on s
But the constraints involved are extremely abstract, marking general parallels and contrasts, issues and expectations, and thus have qualitatively different effects on interpretation from the speech act conventions that describe the possible contributions of utterances and the presuppositions that link utterances to the context.
5.3 The Argument for Convention The dialectic of Part II thus centers on the contrast between two kinds of views. On the one hand, we have the views that the specific interpretations we find in cases such as (25), (26), and (27) derive from the interaction of simple meanings with general pragmatic principles. These pragmatic principles may be characterized and conceived in different ways in different theories, but on the received views pragmatic principles always contribute content to interpretation. On the other hand, we have our view: that the specific interpretations we find in these cases are always a matter of linguistic knowledge that associates forms with interpretive constraints which completely determine the content of interpretation. Pragmatics merely disambiguates. As we anticipated in Chapter 2, we are struck by how precise speakers seem to be in the “extra” meaning they are able to get across with their utterances in cases such as (25), (26), and (27). We are struck by how essential the uptake of this “extra” meaning is to an audience’s understanding of these utterances. This precision alone, we suspect, is good reason to think that the meaning is regimented by shared, public standards. But we will argue that we can find ample evidence for grammatical encoding, once we articulate a clear boundary between two possible patterns of explanation. The challenge is to come up with working diagnostics for grammatical associations of form and meaning that embrace speech act knowledge, presupposition and anaphora, and information structure. Elaborating the informal descriptions of section 5.2, we will suggest that these rules do not straightforwardly contribute to truth-conditional meaning, and so are not readily captured by the notion of what is said by which Grice characterizes CIs. Nor can these interpretive effects be characterized via the notion of explicature from Relevance Theory or Grice’s own notion of conventional implicature. This new theoretical possibility forces us to clarify our understanding of CIs. Recall from Chapter 2 the diverse ingredients involved in calculating conversational implicatures on Grice’s view, as summarized in (24) and excerpted in (3) again here. (3)
To work out that a particular conversational implicature is present, the hearer will rely on the following data; (1) the conventional meaning of
FOR INDEXING
the s cop e of l i n g u i s t i c conven t i on s 95 the words used together with the identity of any references that may be involved; (2) the cooperative principle and its maxims; (3) the context, linguistic or otherwise, of the utterance; (4) other items of background knowledge and (5) the fact (or supposed fact) that all relevant items falling under the previous headings are available to both participants and both participants know or assume this to be the case. (Grice 1975, 50). The expansive and wide-ranging linguistic knowledge that we think underwrites the interpretation of cases like (25), (26), and (27) may not be what Grice has in mind by conventional meaning in item (1) of (3). Nevertheless, we cannot count it as part of the other items of background knowledge Grice allows in (4) of (3). To do so undermines Grice’s distinction between conversational and conventional implicatures, it undermines Grice’s conception of the distinction between semantics and pragmatics, and it undermines the arguments Grice and others have appealed to in characterizing CIs. We must be on the alert to implicit appeals to linguistic knowledge in alleged derivations of CIs. The best way to do that, we think, is to keep in mind the key features of Lewis’s (1969) theory of conventions. We will return in more detail to Lewis’s (1969) theory in Chapter 14, where we acknowledge some of the weaknesses of his specific model of linguistic conventions and offer a notion, grounded in the empirical and theoretical work we do in Parts II, III, and IV that we think may serve us better. But we think Lewis’s (1969) view is right in its key features, and we need only Lewis’s (1969) broad perspective to get clear on the limits of CIs. Lewis starts from an abstract characterization of the strategic settings in which mutual expectations are decisive in agents’ decision-making. These are coordination problems: situations that make available multiple, incompatible courses of action for groups of agents to achieve their goals. Lewis offers a range of examples: two people meeting in one person’s office, the other’s office, or a cafe; drivers driving on the right side of the road or driving on the left; parties on a phone call deciding whether the original caller or the original recipient should call back when the call is dropped; and many others. These situations require agents to act in concert, arriving at one compatible course of action. But there’s an element of arbitrariness about which course of action the agents take, because the strategy they choose to follow is underdetermined by the practical logic of their situation. Lewis’s (1969) key insight is that when agents know how to coordinate reliably with one another, they have a kind of convention. They share an arbitrary choice about what to do. This choice is grounded in their knowledge of one
FOR INDEXING
96
the scop e of lingu i s t i c conven t i on s
another. Since each knows that the others follow the convention, that gives her a decisive reason to follow the convention as well. Lewis observes that the kinds of convention he envisages can arise gradually, without explicit agreement. We start by using purely general considerations in plausible but unreliable attempts to coordinate. If we succeed, we come to share precedents that help us to coordinate further. Eventually, we coordinate reliably through specific knowledge of one another. As we explain in Chapter 14, we think Lewis’s (1969) account does need some revision to account for grammatical conventions, because they are tacit and subject to the deep universals of the language faculty. However, in Parts II and III, our arguments depend only on the idea that conventions are solutions to coordination problems. If we find a coordination problem that people reliably solve, then we have good reason to think that they solve it through some kind of convention, and not through general mechanisms grounded in abstract functional considerations—the principles of rationality, the functional pressures of speaking, the mechanisms evolved for efficient cognition. Coordination problems have multiple solutions that are equally good. There are no abstract functional considerations that distinguish one from another.2 This understanding of the dialectic fits Grice’s own discussion of the characteristics of CIs. This is perhaps most clearly illustrated through his notion of non-detachability (Grice 1975, 43–6), an aspect of his theory we have not yet mentioned. For Grice, it should not be possible to eliminate a CI from an utterance (that is, to detach the CI from the utterance) merely by reformulating the utterance using alternative expressions that are to, all intents and purposes, equivalent. Specifically, Grice asks us to imagine two sentences that are used to say the same thing; that is, that agree in truth-conditional meaning; and further, he asks us to imagine that the sentences are equally fluent and perspicuous; nothing in the way either is formulated would render it uncooperative of a speaker to use it with its intended meaning. Then, of course, there can be no difference in the way the Cooperative Principle applies to either of the two sentences. We must
2 Matters are much more delicate when it comes to the Chomskyan picture of language as involving aspects of our psychological makeup that are universal but essentially arbitrary. We think it’s most useful scientifically and philosophically to capture all the arbitrary ingredients of language—innate or learned—under a single category. If biological constraints are key to the arbitrary patterns by which people communicate, convention may not be the best word for them—they certainly wouldn’t fit Lewis’s (1969) definition. But progress in science has a way of giving us better concepts. See Millikan (1998) and our discussion in section 14.3. As it happens, the view that the interpretive principles of pragmatics are the innate results of contingent processes of evolution is an empirical hypothesis that so far has few defenders. All of the theories we surveyed in Part I assume that the universal interpretive principles of pragmatics—far from being biological quirks—are strongly motivated by functional considerations.
FOR INDEXING
the s cop e of l i n g u i s t i c conven t i on s 97 make the exact same presumptions when we observe a speaker uttering either, and so, the two hypothetical utterances must carry the exact same CIs.3 We will build on Grice’s notion of non-detachability to uncover evidence for the existence of conventionality in the interpretation of speech acts, presupposition, and information structure. In each of these cases, native speakers’ intuitions separate the conversational roles of equally fluent, apparently synonymous sentences. That suggests that the alleged Gricean and neo-Gricean derivations of the implicatures of (25), (26), and (27) we surveyed in Part I mask implicit appeals to conventions of various kinds. Moreover, it suggests that the conventions in question are conventions of language, tied to speakers’ knowledge about the sentences themselves. Grice’s notion of non-detachability provides just one of many diagnostics of coordination on interpretation. Equally important are comparisons across languages and speech communities. An interpretive effect may be non-detachable for some speakers, and thus might seem potentially derivable from general interpretive principles. But it may just as well be due to general interpretive conventions that those speakers share. If corresponding utterances in other languages—or other utterances of the same sentence in other communities—are interpreted differently, we find evidence of that speakers are in fact coordinating on the interpretation they select. It must therefore be a matter of convention, and plausibly, a matter of grammar.
3 When Grice talks about non-detachability he speaks canonically of synonymies, but we obviously need to have some complications to cover CIs that arise from the Maxim of Manner. It’s difficult to know exactly how unperspicuous things have to be in order to be uncooperative; as we’ll see later, the Maxim of Manner is particularly insidious at obscuring conventional constraints on how people can use their utterances.
FOR INDEXING
6 Speech Act Conventions: Indirection and Relevance
As English speakers, we know that questions about ability, like (89), can be taken a couple different ways. (89)
Can you play Chopin’s E minor prelude?
A speaker might use (89) to inquire about how a friend’s piano lessons have progressed. Or a speaker might use (89) to put in a request for the piece that should come next in a friend’s performance. The difference between these two interpretations provides our first test of the predictive power of general pragmatic principles. Griceans and neo-Griceans argue that the question meaning is primary. In fact, it’s the only meaning that figures in speakers’ knowledge of language. The request meaning is derived, when appropriate, by principles that let interlocutors flesh out what others say into messages that make sense on independent grounds. We will call this the reinterpretation account. We will argue, by contrast, that (89) represents a straightforward, commonplace case of ambiguity—or, more precisely, polysemy. The word can simply has two distinct but closely related meanings, both of which English speakers must learn. We survey the ample evidence for the ambiguity account in section 6.1. Our discussion allows us to observe, in particularly instructive ways, why appeals to general pragmatic principles can be attractive but misleading. In section 6.2, we consider in more detail the form that linguistic knowledge must take on the ambiguity view. We draw on Coherence Theory (introduced in Chapter 5) to argue that speakers command a grammar of speech acts that spells out what moves are available to them in discourse and how those moves can be organized into coherent larger patterns. This broader perspective suggests that linguistic knowledge in fact plays a crucial role in a very wide range of alleged relevance implicatures.
FOR INDEXING
sp e e ch a c t conven t i on s 99
6.1 The Ambiguity of Indirect Speech Acts Fundamentally, the reinterpretation account of indirect speech acts disagrees with the ambiguity account about the extent of linguistic knowledge. The two accounts seem to offer very different answers for what English speakers must have learned, in order to recognize the two interpretations of (89). For the ambiguity account, there are two senses of can. For the reinterpretation account, the ambiguity reflects social knowledge on the part of interlocutors. There are two interpretations because there are different purposes that speakers can have—asking versus requesting—and because different contexts impose different practical constraints that may make one or the other of these moves appropriate. But we need to unpack these explanations more precisely to understand where precisely the two accounts disagree. As ambiguity theorists, we admit that speakers’ purposes vary to fit the context—they clearly do. What we deny is the suggestion of the reinterpretation account that these purposes explain how utterances get their interpretations. We can explain the reinterpretation view and its appeal most compellingly by considering situations that clearly favor the indirect interpretation. For example, let’s return to one of the examples with which we opened the book. The server at a restaurant approaches a table where a customer has just put down her menu. The customer says (25). (25)
Can I have the French Toast?
The natural interpretation here is the indirect one: The customer has just placed an order for French Toast. Broadly speaking, it seems right to summarize the inference that pinpoints the right interpretation in the traditional way suggested by Grice or Searle, as summarized in (90). (90)
a. An utterance of (25) looks as a matter of literal meaning to be asking a question about what’s possible. b. But, here, in a restaurant, the menu already says what’s possible, so there’s nothing to be gained by asking about it. c. If, though, the speaker intends to be placing an order, then her utterance is a cooperative move for her to make in this conversational context. d. And, as a matter of fact, the sentence she uttered is associated with a request. e. Therefore, we infer the speaker intends to be making a request.
FOR INDEXING
100
sp e e ch a c t conven t i on s
As we saw in Chapter 2, Searle (1975) (among others) suggests that the prevailing circumstances of (25)—the fact that a restaurant can normally serve the items on its menu but only when a customer requests them; the fact that the question is a precondition for this request—should allow the interlocutors to triangulate easily on the intended interpretation. We do not deny that the argument in (90) might underwrite the audience’s understanding of (25). More generally, we accept that the choice to use an indirect speech act is usually a collaborative one. Indirect speech acts can be a good way for interlocutors to streamline their interactions with one another. Famously, for example, the indirect meaning allows for a literal answer even on its request meaning (Bach and Harnish 1979). Thus, when the customer uses (25), she gives the server a way to acknowledge and accept her utterance even while declining her request, as in (91). (91)
No, sorry, we’re sold out today.
One reason people may use indirect speech acts is that they keep the direction of conversation more open this way (Brown and Levinson 1987). As we saw in Chapter 4, this can avoid potentially complicated work—negotiating conflicts in relationships and sorting out differences in perspectives. But these issues of intention recognition and collaboration aren’t really at issue in the difference between the reinterpretation account and the ambiguity account. As we’ll see in much more detail in Chapter 13, any account of semantics can be linked with a theory of intentions in collaboration; this connection will inevitably relate disambiguation to intention recognition. Instead, the fundamental disagreement centers on the nature of the association that links the utterance to its interpretation. In (90), in particular, the critical point is what information (90d) appeals to. For Searle (1975), this information is the natural associations between actions and their preconditions. Any such connection is potentially available. To use (90) to summarize the interpretation of (25) is therefore to show how the utterance gets a new meaning from the social reasoning of interlocutors. We believe, however, that (90d) cites a convention—another lexical sense for the word can. Thus, for us, (90d) shows how interlocutors triangulate on the most plausible of the grammatically specified analyses of the utterance. In fact, we are skeptical that general pragmatic processes ever play the role (90) seems to envisage on the Gricean story. Armed with this more precise understanding of the opposition between the two views, let’s turn to the data. First, we will consider the grammar of the word please, drawing on Lakoff (1973); Sadock (1974), and Horn (1989). It’s
FOR INDEXING
sp e e ch a c t conven t i on s 101 felicitous to use please in direct, explicit requests, including ones formulated with imperatives such as (92). (92)
Please turn the music down.
You can also use please unproblematically on many indirect requests. (93)
I would like to have the Eggs Benedict please.
As (93) illustrates, please can fit naturally even in utterances that seem not to mention the addressee or the action they are to perform. On the other hand, there are utterances that clearly invite the addressee to respond, but where please is inappropriate. Contrast (94) and (95) for example. (94)
#I’m thirsty please.
(95)
I’d like a drink please.
Being thirsty is pretty much the same thing as looking favorably on a hypothetical drink. You can even imagine a regular saying “Boy am I thirsty today!” upon entering a bar, and getting his usual order from the bartender. Still, using please on such an utterance doesn’t work. What is please latching on to? It seems like please can accompany requests, as long as they are marked as requests in the usual way, rather than left implicit. In other words, please is diagnostic of a level of linguistic representation in which direct and indirect requests have the same status. This argues that premises like (90d) are not just associations but are part of speakers’ grammars. There’s also evidence of the grammatical status of indirect requests in the differing interpretation of pairs of apparently similar sentences in English. Searle (1975) himself observes the contrast between formulations like (25) with can and alternative formulations like (96) involving be able. (96)
Am I able to have the French Toast?
(25) makes for a much better indirect request than (96). In fact, (96) actually sounds like a serious question. If you overheard a customer uttering (96), you might well assume that the customer was following up discussion of a previously mentioned food allergy with a question about the ingredients of this dish. The difference between can and be able is easily substantiated empirically. We’ve seen that one diagnostic for the request interpretation is that it can be accompanied by please. A Google search for please can you reports 120m hits, roughly one in six of the 740m hits for can you; a Google search for please are you able reports 220,000 hits, less than one in four hundred of the 97m hits for
FOR INDEXING
102
sp e e ch a c t conven t i on s
are you able.1 English speakers somehow know that (25), rather than (96), is the ordinary formulation of an indirect request—and this factors strongly into how we produce and interpret utterances involving the terms. Again, this suggests that premises like (90d) are not associations, but are part of speakers’ knowledge of language. Finally, indirect speech act interpretations vary across languages—as observed early and influentially by Wierzbicka (1985). English utterances sometimes carry indirect meanings that are not available for their translations. (97) (98)
Would you like a beer? Miałbys ochot˛e na pivo?
An English speaker can use (97) to offer a drink to a guest. A Polish speaker cannot use its translation (98) this way (?). If used indirectly (which Wierzbicka claims is unlikely), the Polish question would be taken as a suggestion that the speaker and the addressee go out for a beer somewhere together. Conversely, other languages offer formulations whose literal translations cannot achieve the same speech acts in English. A Hungarian speaker would find ´ (99) a natural way to apologize (Suszczynska 1999). (99) (100)
Ne haragudjon. Don’t be angry.
An English speaker cannot apologize with its translation (100). Or consider the Japanese request in (101), as described by (Horvat 2000: 117) and translated literally as (102). (101)
Kikasete itadakemasen ka?
(102)
Can you do us the favor of having us listen?
(101) is a fluid way of asking someone to tell a story, but Horvat has found that native English speakers need special instruction to master the ins and outs of this aspect of Japanese. Such cases provide yet more evidence that speakers’ grammars specify links between utterances and the speech acts they perform. We think these data constitute a knock-down argument against the Gricean view that these interpretations are conversational implicatures—and against the broader neo-Gricean view that these interpretations are derived by general 1 Search results obtained on Nov. 9, 2012. Please are you able is likely even rarer in native English discourse than the counts suggest; cursory examination reveals many accidental coincidences, as Can you try to reboot your system please? Are you able to access the page now? and many clear malapropisms, as Please are you able to advice on changing headlight bulbs . . . The data are even more skewed for other formulations. About one in eight of the 487m hits for Can I is Please can I (62m hits). Just one in 700 of almost 20m hits for Am I able is Please am I able (28,000 hits).
FOR INDEXING
sp e e ch a c t conven t i on s 103 pragmatic principles. In fact, however, these data are familiar in the literature on speech acts in linguistics. And, perhaps surprisingly, their theoretical significance remains in dispute. Researchers’ reactions to these data shows how entrenched Grice’s assumptions about language and meaning have become. We will briefly signpost the dialectic here and return to the broader issues in Part IV. Since Searle (1975), a major goal of research on indirect speech acts has been to explain the constrained variation we see in indirect speech acts across languages. It’s thought that pragmatic calculations along the lines of (90) would help to do this, because premises like (90d) are grounded in the regularities that agents already need to be aware of to plan complex interactions in conversation. If such connections must link utterances to their indirect meanings, we have an explanation for what indirect speech acts are possible, how they arise, and how they are learned. Searle’s point is a good one. But we do not see how it distinguishes the reinterpretation account from the ambiguity account. Conventions are not inexplicable or unmotivated. To the extent that conventions evolve from agents’ spontaneous and improvised successes at acting together—as we’ve seen, Lewis (1969) suggests this is the usual case—conventions are likely to exhibit constrained variation. Instead, what makes them conventions, following Lewis (1969), is just that they have one reasonable alternative—because they do not follow exclusively from the practical logic of agents’ situations. The rest of semantics confirms this expectation. The obviously conventional meanings of words and expressions exhibit a panoply of motivations and generalizations (Goldberg 2006). For example, somebody who hears the word corkscrew for the first time can probably guess what’s being referred to—something shaped like a screw that’s used for corks. Likewise for the less familiar term wine key. But the contrast shows the element of convention: corkscrews include all openers that work by inserting a helical worm into a stopper so that it can be pulled out; wine keys are specifically those corkscrews with the form of a folding pocketknife with a worm and a lever attached to a central handle by hinges. The terms corkscrew and wine key are both expressions of English, and their reference is a matter of semantics (of convention), and not of calculability. A French speaker facing a gap in her English vocabulary might be inspired by her own native tire-bouchon and a presumption of rationality to ask for a plug pull. She might even be understood. But it so happens that pulling the plug in English means something else. Just the same, treating can as polysemous in (89) allows us to observe that the meanings involved fit together in partly—but not completely—predictable ways. Some authors (including (Morgan 1978; ?) among others), offer a different kind of defense of general pragmatic principles in the face of contrasts like
FOR INDEXING
104
sp e e ch a c t conven t i on s
that between (25), where indirection is possible, and (96), where it is not. They suggest that the indirect interpretation is available by default. But the existence of a routine or ordinary formulation of the indirect speech act preempts the derivation of the indirect interpretation in other cases. (In Chapter 3, we observed that unusual expressions take on marked meanings in many approaches to linguistic pragmatics.) Again, we think these proposals are not in fact compatible with the Gricean program and its understanding of the division between semantics and pragmatics. At the same time, it’s easy to acknowledge the spirit of such proposals on the ambiguity account, by understanding them to describe the organization of speakers’ grammar and lexicon. Researchers often cast preemption as an application of Grice’s Maxim of Manner. In other words, it’s uncooperative—irrational—to use (96) to express the indirect speech act meaning, because there’s a better way to express it. That’s why we suspect that the speaker must have intended the unusual question meaning instead of the more natural request meaning. We must be careful reasoning this way. Grice’s standards of calculability allow no appeal whatsoever to conventional information beyond what goes into determining what the speaker says—the function of calculability in the theory is to distinguish conversational implicature from conventional implicature.2 Thus, we can’t appeal to Grice’s Maxim of Manner to distinguish interpretations for equivalent forms just because one is conventional and one is not. Grice’s Maxim of Manner only kicks in when an utterance is so cumbersome or awkward that no speaker could sensibly use it without some ulterior motive. In fact, awkwardness is no obstacle for indirect requests generally. Elaborate utterances are regularly understood to make especially polite indirect requests—provided they are conventionally used to do so: (103)
Could you do me the kindness of passing the salt?
How could the Maxim of Manner call for an exceptional interpretation of (96) and not treat (103) the same? Again, it’s useful to draw an analogy with other well-known cases of preemption in semantics. For example, in specifying the English lexicon, it might be a good idea to include a default rule that the name of an animal is also the name of its meat (Copestake and Briscoe 1995). This rule must be preempted, 2 Several commentators have asked us whether conventionality and calculability must be incompatible. The answer must be a resounding “Yes.” Conventions require alternatives, but the existence of alternatives means it would be rational that the conventional case didn’t hold, and so, you can’t calculate anything from it. “The maxims are not conventions at all. They are a ‘part’ of what it is to be verbally cooperative and so lose the feature of arbitrariness so dear to most accounts of conventions” (Harnish 1976, 357). See Lewis (1969) on convention, about which we’ll have more to say in Part IV.
FOR INDEXING
sp e e ch a c t conven t i on s 105 however, for a small set of cases—beef, pork, veal, venison—where the meat gets a special name. For example, since the association between chicken and its meat is endorsed by the lexicon, we understand (104) as an ordinary event. But, by contrast, since the association between cow and its meat is not, we understand (105) to describe a fantastic monster whose appetite extends to whole animals. (104) Zul ate a chicken sandwich. (105) Zul ate a cow sandwich. The received view is that this is an interesting fact about polysemy in semantics— even though it yields patterns of interpretations strikingly parallel to those in (25) and (96). Thus, suppose we handle indirect speech acts through the mechanism Copestake and Briscoe propose for polysemy in general: with default rules in the lexicon allowing items with certain kinds of meaning to have certain kinds of interpretations. These rules could be preempted based on the existence of particular alternative forms with alternative meanings. The result will be an instance of a familiar pattern in the organization of meaning, and as good an explanation of indirect speech acts as we can expect. Dan Sperber (pers. comm.) defends reinterpretation accounts from yet another perspective. He suggests that both (25) and (96) ultimately have the same range of interpretations. The formulation in (96) may be rare, but speakers, if confronted with it, could only really understand it as an indirect speech act in the end. Of course, we get enough experience with unfamiliar uses of language, because there is enough variability across communities, that we may suspect that an utterance has been formulated with unfamiliar expectations about how it will be interpreted. In such circumstances, we may be well served by considering what it would have made sense for the speaker to say. Even in these cases, however, we suspect that hearers must make an intuitive guess about a conventionalized indirect speech act, and so the listener’s inference lacks the content of a Gricean calculation. On the ambiguity view, the question to ask is (106). (106) What is a plausible convention that I could postulate to assign this utterance a likely intended interpretation? It’s not the Gricean question (107). (107) How do principles of rationality and collaboration explain the creative use to which the speaker has put this utterance with its known interpretation?
FOR INDEXING
106
sp e e ch a c t conven t i on s
Even if interlocutors can sometimes understand indirect speech acts used creatively—just as they can sometimes understand utterances with words that are unfamiliar or used in unfamiliar senses—it is a mistake to conclude from such lucky guesses, and the motivations and generalizations behind them, that they somehow establish that the meaning involved is calculable in Grice’s sense, or that pragmatic principles are involved in determining meaning. When we delimit knowledge of language, we must bracket the partial explanations that fit different aspects of grammar together and listeners’ creative abilities to anticipate or reconstruct what others are trying to say to them. We do not try to provide a pragmatic theory of cooperative rationality to describe that part of meaning that “makes sense,” and restrict semantics just to that part of meaning that doesn’t. We’ll have much more to say about the interrelationships of different kinds of interpretive reasoning in Part IV of the book.
6.2 Speech Acts as Discourse Relations If the question and request interpretations of (89) represent a genuine ambiguity, we must be able to characterize the linguistic knowledge involved. In the 1970s, generative semanticists beginning with Ross (1970) attempted to describe speech act interpretations with the traditional tools of transformational syntax and truth-conditional semantics. But the relevant syntactic evidence proved equivocal, while deep problems quickly emerged in understanding utterance content after characterizing speech acts in semantic terms. See Sadock (2004). The failure of this program established that the interpretive difference among readings of (89) is not a matter of syntactic structure or semantic content. It’s a matter of what the speaker is doing. To characterize linguistic knowledge, we must therefore specify the contributions that speakers can make with their utterances, and map out how grammar associates these functions with specific linguistic forms. We will follow the approach to the linguistics of discourse developed in Coherence Theory, and particularly Hobbs (1990); Kehler (2001) and Asher and Lascarides (2003). To appreciate the place of the linguistics of discourse in Coherence Theory, recall its key assumption: making sense of an utterance is inferring an explanation of what the speaker was doing in using it. Good explanations feature interpretive connections that link the utterance to previous ones. In (108) for example, as discussed by Hobbs (1979) and Kehler (2001), we understand the second sentence as an explanation of the first. (108)
John took the train from Paris to Istanbul. He has family there. (Kehler 2001: 2 Ex 1)
FOR INDEXING
sp e e ch a c t conven t i on s 107 The suggestion is not only that John has family in Istanbul but that this fact makes sense of why John undertook this specific trip. Perhaps he went to visit them, for example. You might think, following Grice and his followers, that such connections just derive from the rationality of speakers. When speakers produce discourse, we should expect speakers to tie together a chain of related ideas, and we should expect their audience to recognize the resulting connections.3 But Coherence Theorists argue that the interpretive connections we find in discourse are stronger than mere requirements of rationality can impose. Consider (109), for example. (109) John took the train from Paris to Istanbul. He likes spinach. (Hobbs 1979: 67) As interpreters, we find it extremely compelling to take a followup, generic stative sentence as an explanation for a prior eventive sentence. So in (109), we are even tempted to imagine that there is some explanatory connection between spinach and train trips to Istanbul. This absurd interpretation is not the only rational way to link these two sentences together, however. In fact, speakers might have perfectly good reasons to give these two separate bits of information about John. For example, in (110), the discourse cites these two facts as examples of John’s old-fashioned tastes. (110) John has old-fashioned tastes. He took the train from Paris to Istanbul. He likes spinach. Why is it that reasonable interpretations, like that of (110), don’t suggest themselves when we read (109) on its own, and instead interpreters think of bizarre interpretations? Coherence Theory suggests that language users rely a particular repertoire of cues and structures in organizing their discourse. It’s these patterns that enjoin us to connect the spinach and the train trip to one another in (109), and that exclude the possibility signaled in (110). These patterns thus constitute a specific dimension of interlocutors’ knowledge of language, over and above the dictates of rationality in connecting sentences together in discourse. Coherence Theory describes the possibilities as a set of forms that establish the structure of discourse and simultaneously link together the propositions expressed in individual sentences into argumentative relations. Fish (2011) draws an explicit analogy between the formal structures that establish compositional 3 Though we have often encountered this intuitive suggestion in discussion, we have not seen it developed systematically with attention to the relevant data.
FOR INDEXING
108
sp e e ch a c t conven t i on s
semantics within the sentence and the formal structures that establish rhetorical structure across discourse. (111)
The rhetorical structures are structures of argument . . . [Like syntactic structures] they too are formal—abstract, contentless—but rather than being the forms that make random words into propositions (sentences), they are the forms that link propositions together into complex units. (Fish 2011, 29)
Rhetorical structures encompass not only the patterns of argument familiar from logic, but a wide variety of other strategies for organizing and presenting related ideas. Many of these strategies seem to be part of the conventionalized inventory of particular kinds of discourse. For example, inspired by Graff and Birkenstein (2007), Fish considers a strategy that allows speakers “to summarize conventional wisdom on a topic on the way to disagreeing with it”. His example is (112). (112)
They say that money talks, but I say money corrupts.
Note how much the interpretation of (112) depends on the form that organizes the discourse, rather than the individual propositions that make it up. Logically, it’s open to the speaker to endorse and go beyond the conventional wisdom. Both halves of (112) could be true: money could give people the influence to get things done and inspire them to abuse their authority as a result. But the formal organization of (112) implies a deeper contrast. We understand the speaker to reject the common-sense importance of money, because it is so hard to use money for good. This formal organization is, moreover, a conventional one. (113)
You might think that disagreeing is a “natural” act, requiring no formal skills. But disagreeing is a learned activity; it is more than just saying no; it involves giving reasons for your rejection of what others have said . . . Organizing a discourse around views you oppose is an achievement of artifice, and it is made possible by a form you have to acquire through practice. (Fish 2011, 29)
One sign of its conventional status is the fact that there is no comparable pattern I say/they say as illustrated in (114). (114)
I say that money corrupts, but they say money talks.
(114) does not read as a summary of the speaker’s argument against commonsense views of the importance of money. In particular, (114) seems to lack the suggestion, contributed by the formal organization of (112), that “they” are wrong. One sign of the difference is that (114) reads, at least to us, like a disillusioned report of the ineffectiveness of the speaker’s idealism. The difference in
FOR INDEXING
sp e e ch a c t conven t i on s 109 interpretation points to the conventional knowledge that underpins the special interpretation of (112). Though there might be good reasons for speakers to move from familiar ideas to new ones, it is still difficult to see how general considerations of rational action could dictate that speakers who contrast A and B must have such different purposes from speakers who contrast B and A. Fish’s analysis of arguments of the form “they say/I say” suggests that rhetorical structures are composed from a wide variety of specific patterns that speakers know. A similar moral might be drawn from the patterns surveyed by Graff and Birkenstein (2007), the extensive inventories of rhetorical relations of Mann and Thompson (1988) and Knott (1996), or the guidelines used by Prasad, Dinesh, Lee, Miltsakaki, Robaldo, Joshi and Webber (2008) to annotate rhetorical relations in the Penn Discourse Treebank. At the same time, other researchers have highlighted broad categories of relations that can underpin coherence in text. These categories give an instructive sense of the range of implicit meanings that are established by speakers’ discourse competence. For example, Kehler (2001: 187) distinguishes the three kinds of coherence illustrated in (115–117). (115) Max spilt a bucket of water. He tripped on his shoelace. (116) Max spilt a bucket of water. He spilt it all over the rug. (117) Max spilt a bucket of water. John dropped a jar of cookies. In (115), we find a cause–effect relation. We understand that tripping contributed to Max’s falling. Such inferential connections underpin many kinds of explanatory discourse. In (116), we find what Kehler calls an occasion relation. We understand that the second sentence says more about what ensued in the situation described by the first. This also illustrates a typical pattern, in which speakers elaborate extended descriptions of a common theme across multiple sentences. Finally, (117) shows a resemblance relation. We understand that the speaker has chosen to describe these two events together to highlight the similarities and differences among them. Of course, speakers know how to organize discourse in terms of parallels and contrasts. Hobbs (1985, 1990) and Asher and Lascarides (2001) identify additional kinds of moves that connect utterances across speakers in interactive language use. For example, conversation includes a distinctive category of adjacency pairs (Schegloff 1972; Schegloff and Sacks 1973), such as mutual greetings or linked questions and answers where one speaker meets the expectations established by another. (118) is a representative example (Asher and Lascarides 2001: 189 Ex 13). (118)
a. Q: Did John fail his exams? b. A: He got 60%.
FOR INDEXING
110
sp e e ch a c t conven t i on s
A’s answer in (118b) informs Q that John has actually passed his exams, with an aggregate score that just meets the threshold. The dialogue thus constitutes an indirect question–answer pair—an exchange where the response settles the question by inference given certain presupposed background information. Another general category is what Hobbs calls evaluation, where utterances are related by proposing and motivating goals for the discourse (Hobbs 1985: 12). Hobbs gives (119) and (120) as examples of evaluation for practical and communicative goals. (119) (120)
Did you bring your car today? My car is at the garage. Do you know what I mean?
In (119), the fact that the speaker’s car is at the garage motivates the discourse goal behind the speaker’s question: finding other plans for getting around. In (120), the fact that the speaker does not know whether she has been understood motivates the discourse goal of her question: addressing the issue explicitly. Researchers often find overlapping connections in discourse, as in (121) from Baldridge and Lascarides (2005: Ex 3). (121)
a. Q: Can we meet next week? b. A: I’m free on Tuesday.
Here A’s response offers an indirect answer to Q: since A is free on that Tuesday it will be possible to meet next week. This is an instance of the same kind of connection we find in (118). But the discourse is probably not just about what’s possible. Q’s question serves indirectly to request a meeting. And A’s response accepts that proposal and offers a suggestion to meet on Tuesday, advancing the interlocutors’ practical goals by the kind of connection Hobbs calls evaluation. In thinking of connections in discourse, it’s common to present a hierarchy of possible relations (Knott 1996; Miltsakaki, Robaldo, Lee and Joshi 2008), reflecting contributions to the conversation along a variety of dimensions, possibly simultaneously (Bunt 2011). Linguistic rules describe not only what rhetorical structures there are but how they fit together. Discourses, like sentences, have complex, nested structures that are built incrementally by the individual contributions that speakers make. We can see this hierarchical structure in (122) from Webber, Stone, Joshi and Knott (2003: 549 Ex 8). (122)
a. On the one hand, Fred likes beans. b. Not only does he eat them for dinner. c. But he also eats them for breakfast and snacks. d. On the other hand, he’s allergic to them.
FOR INDEXING
sp e e ch a c t conven t i on s 111 In (122), the discourse as a whole is organized as a contrast between Fred’s passion for beans and his problems with them. The description of Fred’s avid consumption of beans in (122b–c) supports the speaker’s main claim in (122a) by offering illustrative cases that corroborate and delineate Fred’s obsession. The examples in (122b–c) are themselves alternatives that parallel one another but stand in a scalar relationship: There is the expected case of dinner in (122b) and in addition the extreme cases of breakfast and snacks in (122c). The overall organization of (122) is thus the tree structure diagrammed in (123).
(123)
contrast HH HH H 122d support HH H 122a scalar-parallel HH 122b 122c
Such trees let us appreciate Fish’s suggestion in (111) more precisely, by formalizing the parallel between the organization of coherent discourse and the syntax of sentences. It’s worth noting in this connection that purely rational grounds do not dictate that all purposeful activity should inevitably exhibit such hierarchical organization. In fact, it’s often necessary in managing practical real-world activity to interleave actions that contribute to different goals (Sacerdoti 1975; HayesRoth and Hayes-Roth 1979). Think of how you fold together multiple errands into an efficient tour, or think of what must happen in the kitchen if all the dishes for dinner are to be ready at the same time. Although we perform and understand interleaved practical actions, we interleave actions in discourse in much more constrained ways. Webber et al. (2003: 550 Ex 10) illustrate this with (124): (124)
a. On the one hand, Fred likes beans. b. Not only does he eat them for dinner. c. On the other hand, he’s allergic to them. d. But he also eats them for breakfast and snacks.
(124) does not have the interpretation of (122). If it’s even intelligible at all, (124) specifically sets up breakfast bean-eating as a contrast to John’s allergy, rather than as an extreme illustration of his love for beans.
FOR INDEXING
112
sp e e ch a c t conven t i on s
As Webber et al. (2003) show, linguists can easily accommodate these interpretive constraints by assuming that discourse coherence is signaled by connectives like on the one hand and on the other hand, and that the structural relationships induced by these connectives cannot exhibit crossing dependencies. To accept this proposal is to endorse a distinctively linguistic explanation for the interpretive contrast between (122) and (124). It confirms the perspective of Coherence Theory that rhetorical structure is part of language, not just a consequence of general rationality. Cross-linguistic research on discourse structure is in its infancy, but Coherence Theory naturally predicts that speakers’ resources for articulating and organizing contributions to discourse will vary across languages and communities. We seem to find hints of such differences in the varying techniques languages seem to use, for example, in negotiating problematic interactions. ´ Thus, Suszczynska (1999: 1061–2 Ex 1 and 2) contrasts the prototypical English apology in (125) with the prototypical Hungarian one in (126). (125)
Oh, I’m so sorry! Are you all right? Let me help you with your things.
(126)
Kérem, ne haragudjon. Nagyon ügyetlen voltam. Jöjjön, segitek összeszedni a csomagokat. Nem ütötte meg magát? Remélem nem sülyos. Please, don’t be angry. I was very clumsy. Come, I will help you pick up the packages. Haven’t you got hurt? I hope it is not serious.
The examples show speakers apologizing with the distinct conventionalized indirect forms of their languages—but also in some cases differing in the moves they make to elaborate their apologies and in the order they contribute them. ´ Suszczynska’s conclusions are based a corpus of exemplars from subjects of the different speech communities—and extend to Polish, where interlocutors’ strategies are different yet again. It’s conceivable that we might find underlying principles describing the goals of speakers and their practical backgrounds that could clarify the rational underpinnings of such effects, but we are skeptical. The preponderance of the evidence, as we have surveyed it here, is that the rhetorical moves that speakers make and the discourse structures that result are functions of speakers’ linguistic competence, not their practical rationality.
6.3 Summary: Signaling the Organization of Discourse The discussion in section 6.2 allows us to recognize the indirect speech acts we surveyed in section 6.1 not just as manifestations of linguistic knowledge, but as part of the larger system speakers have to organize discourse. In particular, we
FOR INDEXING
sp e e ch a c t conven t i on s 113 have argued that speakers use conventional relational connections both to signal the implicit inferences that make discourse coherent and, relatedly, to signal their goals and intentions in organizing discourse. This broader perspective allows us to revisit and clarify a wide range of cases from Part I of the book. The key interpretive effects, we think, derive from specific linguistic knowledge rather than general interpretive principles. Consider first (46), repeated here: (46)
a. Wife:
Bob’s having a party; do you want to go?
b. Husband:
OK.
c. Wife [Later]:
Are you sure you want to go?
d. Husband:
OK, let’s not go. I’m tired anyway.
Recall that (46) involves a miscommunication. The wife intends her utterance (46c) to carry only its straightforward contribution as a question, but the husband takes it as a suggestion not to attend the party. For Horn (1984), this is evidence of an ambiguity in pragmatics. In particular, Horn analyzes (46) in terms of a conflict between general pragmatic principles—the Q principle that a speaker means no more than what she said; and the R principle that a speaker is committed to all the ordinary consequences of what she said. The perspective of Coherence Theory also allows us to characterize the alternative interpretations of (46c) as a linguistic ambiguity—but it is a much more prosaic one. According to Coherence Theory, speakers have multiple options for what contributions they can make to discourse and how they can express those contributions linguistically. In this case, the grammatical options allow for both a question interpretation and a suggestion interpretation for (46c). No deeper principles need be involved, either in how the alternative interpretations arise or in how speakers resolve them. That’s as it should be, we think: we are able to capture the ambiguity here the same way one would describe any other kind of ambiguity, as reflecting alternative possibilities for analyzing an utterance according to speakers’ knowledge of language. Coherence Theory also suggests that we should capture scalar implicatures, in part, through suitable relations of answerhood (Asher and Lascarides 2001). A question establishes standards for how much information is required in an answer. A question inherently distinguishes adequate or complete answers from partial or incomplete ones. Part of making sense of an answer is recognizing whether the speaker intends it as a full answer or as a partial answer to the question at issue—disambiguating the coherence relation that connects the answer to the discourse. This disambiguation brings interpretive consequences.
FOR INDEXING
114
sp e e ch a c t conven t i on s
In recognizing a move as a full answer, we understand the information presented as implicitly exhaustive, as providing everything the audience needs to know that’s relevant to the question. So, for example, we can infer that the speaker implicitly excludes those possibilities she does not explicitly mention, even if the literal meaning of the utterance might seem to leave them open. In recognizing a move as an incomplete answer, on the other hand, we understand the information presented as exhausting the speaker’s relevant knowledge about the question. So we can infer that the speaker is not in a position to give additional details or to address other relevant instances. Thus, in Coherence Theory, the specifics of scalar interpretation come from resolving ambiguity in discourse rather than from applying general interpretive principles. We pursue the connection between questions and scalar implicatures further in Chapter 8 in the context of the grammar of information structure. Even Grice’s famous example of relevance implicature (6b), repeated here, looks on our view like an ambiguity resolved in context to its most plausible interpretation. (6)
a. A : I’m out of gas. b. B : There’s a gas station around the corner.
Coherence Theory suggests that (6) illustrates a conventional rhetorical strategy for responding to a reported difficulty with a suggestion about how to resolve it. Given that background, it makes sense why we would understand (6) as we do. But our conventions for working together might have played out differently. For example, depending on the varying standards and expectations of the speech community, (6b) might have been a rude rejection of (6a)’s request for gas, a brusque rebuke for disturbing the speaker, an invitation for speaker and addressee to go to the gas station together, and so forth. All these alternatives should, we think, be regarded as part of speakers’ knowledge of language, since they describe how sentences are conventionally interpreted. The fundamental moral, we suggest, is just this: In identifying speech acts, interlocutors rely on mutual expectations that are not determined by the dictates of rationality, or any other general interpretive principles. Our interpretive choices reflect our knowledge of the arbitrary communicative practices of our community. They are conventions. Purported arguments to the contrary, we have suggested, fail to appreciate the empirical variability of indirection across languages, the distinctively linguistic character of the rules involved, and the explanatory motivations linguistic rules always have.
FOR INDEXING
7 Presupposition and Anaphora: The Case of Tense and Aspect
Speakers’ discourse often juxtaposes descriptions of successive events, as in (26), repeated here. (26)
Oil prices doubled and demand for consumer goods plunged.
It’s natural to understand the events described in (26) as taking place one after the other: first, prices double; then, demand plunges. More precisely, (26) involves a contingent relationship between the successive events (Hobbs 1985; Steedman 1997). The second event depends on the first, in that the first somehow sets the stage for the second event. In particular, (26) seems to evoke an unfolding cascade of interrelated shocks at the start of an economic downturn. Such contingent relationships are at the heart of the narrative connections that characterize so much of human discourse. Narration is one of the central relations in Asher and Lascarides’s (2001) taxonomy; for Hobbs (1985: 10) and Kehler (2001: 192), it’s an important special case of the occasion relation. Common though they may be, narrative interpretations are not inevitable. Take (1) for example. (1)
Short sellers aim to buy low and sell high, but not in that order. They sell shares they don’t own now (they are short of these shares) in hopes of buying them cheaper later.1
As (1) reports it, buying low and selling high are two aims that traders may undertake in parallel. The selling need not depend on or even follow the buying. Of course, the famous injunction, “Buy low and sell high,” does normally report a narrative sequence. The authors of this passage, Peter Kennedy and John Saunders, adapt the expression here to remind readers why there can be profit 1 Example from Peter Kennedy and John Saunders, “GeneMax Battles Short Sellers,” The Globe and Mail, Toronto, Nov. 18, 2002
FOR INDEXING
116
presupp osit ion an d a n a ph or a
in the transactions they describe, even though the trades actually begin with the sale—a promise to supply something in the future—and conclude with the purchase that will make good on that promise. The variability of temporal interpretation in discourse gives us a second test of the predictive power of general pragmatic principles. Grice and neo-Griceans argue that the narrative discourse seen in (26) and the parallel discourse seen in (1) both involve a linguistic meaning that’s neutral about the temporal location of the described events. The narrative interpretation is derived, when appropriate, by the general reasoning that speakers use to make sense of what others say. For example, as we saw in Chapter 2, Grice argues that the contingency in the interpretation of (26) follows from the prevailing interest of cooperative speakers in describing events in the order they occur. Again, we reject this explanation in favor of an alternative view—with many antecedents in the linguistic literature—that the difference reflects an ambiguity. This view holds that logical form—any formal statement of the meaning of utterances as delivered by a grammar—determines a specification of a set of temporal relations among described eventualities. As a result, it’s logical form that settles whether a sentence has a narrative reading or another possible interpretation. We begin this chapter in section 7.1 by framing the two alternative views, and offering some key data that distinguishes them. We continue by briefly sketching the general theoretical picture that inspires our more linguistic view. We explain in section 7.2 why logical form must specify interpretive dependencies across sentences in many cases, including the temporal dependence exhibited in (26). We explain in section 7.3 how dimensions of linguistic meaning, including the dynamics of presupposition and anaphora, let speakers signal the interpretive dependencies they have in mind. This general picture underscores the crucial role for knowledge of language in settling the interpretations of utterances in context—including, prominently, the temporal interpretation of discourses such as (26). We conclude in section 7.4 by considering the broader dialectic which researchers have used to defend a CI analysis. Grammatical explanations of temporal interpretation, and other dependencies in discourse that depend on the dynamics of presupposition and anaphora, derive their ambiguities rather than stipulating them, and so do not fall prey to the criticisms that are often voiced against ambiguity accounts in the Gricean program.
FOR INDEXING
pre su pp o s i t i on a n d a n a ph or a 117
7.1 The CI Account and its Alternatives According to the Gricean, recall from Chapter 2, the temporal relationship associated with (26) is supposed to be anchored in the assumption that the speaker is adhering to the Maxim of Manner. This speaker can be treated as a cooperative conversationalist only if we assume she is being orderly, by which the Griceans mean only if we assume she is presenting events in their actual temporal order; otherwise, her contribution is uncooperative. The standard Gricean derivation, then, would go something along the lines of (2). (2)
a. The speaker has uttered the first conjunct, “Oil prices doubled,” and continued with a second conjunct, “and demand for consumer goods plunged,” and we presume that this second contribution is appropriate for the context in which it is uttered. b. The speaker must therefore intend the second conjunct to provide a relevant continuation of the first. c. Given what the speaker has said, and the presumption that she is obeying the Cooperative Principle, the only possible such relation is that the plunge followed (and was likely contingent upon) the doubling. d. So, the speaker means that the events occurred in the order that she presented them.
What we are claiming is that this particular derivation hides a covert appeal to a convention, in this case at step (2c), and that the need for this convention shows that the derivation is not a calculation of a CI from what is said by principles of rationality alone. Instead, it is a record of how one might disambiguate (26) to its most natural, narrative reading. To appreciate the role of convention in this derivation, we, once again, invoke Grice’s own test of non-detachability. Consider the alternative utterance in (3). (3)
Oil prices have doubled and demand for consumer goods has plunged.
(3) describes the events with the English present perfect have doubled and has plunged rather than the simple past doubled and plunged. This is a difference in aspect, the grammatically encoded perspective that the speaker takes in describing an eventuality. The simple past describes things from a perspective situated amid the flow of unfolding past events. The present perfect describes things from the perspective of the actual present situation (Reichenbach 1947; Moens and Steedman 1988).
FOR INDEXING
118
presupp osit ion an d a n a ph or a
This difference in perspective seems not to carry any truth-conditional contrast. Both a past tense description of an event and a corresponding present perfect description seem to be true just in case such an event has occurred. There is an interpretive contrast between (26) and (3) however. With (3), we are much less sure about whether temporal succession is necessarily involved. We readily interpret (3), unlike (26), as presenting two separate and potentially unrelated reasons as to why the economy is in trouble. The use of (3) does not constitute a violation of the Cooperative Principle, or any of its four maxims. In fact, its felicity suggests an alternative interpretation of how we can understand orderliness: (4)
Orderliness sometimes involves presenting events in their order of importance.
Apparently, Grice’s injunction “Be orderly” is less precise than derivations like (2) suggest. With this foil in mind, we can return more precisely to the interpretive reasoning of the audience. As we saw in Chapter 6, in order to understand (26) or (3), the audience needs to recognize the coherence relation that the speaker uses to link her two clauses together. In fact, in (26), we understand that the speaker is using the second conjunct to continue a narrative. By contrast, with (3), it seems perhaps more likely that the speaker has chosen to describe two past events in succession because they are similar, rather than because they stand in a particular temporal relation. Locating the perspective point in the past with doubled and plunged on the one hand, as opposed to in the present with have doubled and have plunged on the other, is a signal of the interpretive difference. Thus, in both cases, the speaker’s choice of the conventional elements with which she articulates the utterance, in particular, her choice of the particular tense and aspect on these verbs, guides the audience towards the interpretation she has in mind. The rest of the chapter explains exactly how the process works. In Section 7.2, we review evidence that a simple past tense verb like doubled is interpreted temporally much like a pronoun. It is understood to describe what happened during a contextually specified interval, in much the same way as a pronoun is understood to describe a contextually specified individual (Reichenbach 1947; Partee 1973; Webber 1988). Crucially, as we see in section 7.3, language offers dimensions of conventional meaning that strongly constrain—and indeed perhaps even determine (Stojnic, Stone and Lepore 2013)—what variables appear where in logical form. These rules include centering systems that set up preferences over available variables and presuppositions that reorient discourse to variables whose values
FOR INDEXING
pre su pp o s i t i on a n d a n a ph or a 119 have particular properties. This information, which varies from language to language and from utterance to utterance, is what determines the range of possible interpretations for discourse. With tense, then, as with speech act ambiguities, understanding involves choosing the reading that makes the most sense from the candidates delivered by the grammar. Understanding does not involve enriching those readings to transform them from ones that do not make sense to ones that do.
7.2 Temporal Interpretation as Discourse Reference We begin with the phenomenon of discourse reference (Karttunen 1969; Kamp 1981; Heim 1982; Webber 1983), as illustrated in (5). (5)
a. A man walked in b. He sat down.
There is a broad consensus that discourse (5) is true just in case some man walked in and sat down during some relevant temporal interval established in the discourse context. In particular, (5) is meaningful and true even when several men walked in and sat down during this time. Moreover, in such a case, a speaker can use (5) truly, without being acquainted with any particular man about which (5) is true (or is claimed to be true). This interpretation makes he in (5b) dependent on a man in (5a). We can’t represent he by a constant representing a specific individual; no such individual need be available. Nor can we represent he by a Russellian definite description. That would require us to identify the referent of he uniquely by some distinguishing characteristics. The discourse doesn’t provide any such characteristics and doesn’t even seem to entail that there must be any. The simplest way to capture this dependency is to represent he by a variable in logical form, and to develop a regime that allows this variable to be bound by the quantifier a man—that is, allows the variable to be interpreted in tandem with the quantifier—even though the quantifier occurs in an earlier sentence. Such variables are known as discourse referents and the regime for interpreting them across discourse is known as dynamic semantics. The terminology reflects a particular metaphorical understanding of discourse interpretation. The idea is that when an audience understands a pronoun, they associate it with an interpretive element (a variable) that links it to its antecedent (a previous noun phrase, but perhaps a quantificational one without a specific real-world referent). This interpretive element is properly thought of as part of the discourse rather than something in the world: a
FOR INDEXING
120
presupp osit ion an d a n a ph or a
discourse entity or discourse referent. Meanwhile, to formalize the interpretive dependency between quantifiers and variables across discourse, it’s useful to model the changing information state of the audience as they accept each sentence in turn. This dynamic perspective on information growth across discourse gives dynamic semantics its name. We won’t have much to say about the details of discourse referents and dynamic semantics here.2 Dekker (2012) offers a comprehensive introductory treatment. For now we simply observe that such formalisms allow us to represent the meaning of (5) as (6). (6)
a. [x|man(x), walked-in(x)] b. |sat-down(x)
The pronoun he is translated by the same variable x that is introduced by the quantifier a man, and the interpretive dependency results in the correct existential truth-conditions for the discourse as a whole. Pronouns are not the only linguistic expressions that need to be modeled with variables that are interpreted across discourse. Famously, Partee (1973, 1984) draws an analogy between pronouns and tensed verbs. Her argument is that the time at which we understand an event to take place shows much of the same interpretive variability we see with pronouns. For example, speakers can use a tensed verb to refer to an understood time, like the last time the speaker left the house in (7), just as a speaker can use a pronoun to refer to an understood individual as in (8). (7) (8)
I didn’t turn off the stove. (Partee 1984: 244 Ex 1a) She left me. (Partee 1984: 244 Ex 1b)
Conversely, the temporal interpretation of a tensed verb can be bound by an adverb of quantification, as in (9), much as a pronoun can be bound by a quantified noun phrase, as in (10). (9) (10)
Whenever Mary telephoned, Sam was asleep. (Partee 1984: 246 Ex 5a) Every woman believes that she is happy. (Partee 1984: 246 Ex 4a)
In the context of such examples, the narrative dependency of tensed verbs seen for example in (26) seems analogous to the interpretive dependency of 2 Our current research continues to explore the philosophical issues involved. For example, Stojnic et al. (2013) argue that dynamic semantics makes it possible to formalize conventions for the interpretation of context-dependent expressions in powerful and philosophically revealing ways. Ultimately, however, we expect such arguments to complement and strengthen the characterization of interpretive dependencies we offer here.
FOR INDEXING
pre su pp o s i t i on a n d a n a ph or a 121 pronouns as in (5). That is, we understand the plunge in (26) to follow the doubling, no matter when the doubling takes place, much as we understand the man who sat down in (5b) to be the same as the man who came in, no matter who it is. Partee’s argument suggests that tensed verbs should be represented in logical form with variables that range over temporal intervals and serve to locate the described eventualities in time. A wide range of research in semantics has fruitfully followed this suggestion. One important discovery of this line of research has been a better understanding of the range of dependent interpretations that are possible across discourse. Flexibility is crucial (Webber 1983). To illustrate, consider the pronoun they. Unlike he, they does not always have a unique direct antecedent. For example, (11) shows that they can have split antecedents: they in (11b) depends on a man and on a woman in (11a). (11)
a. A man met a woman. b. They went to the movies.
Alternatively, they can refer to a kind of thing that’s been implicitly evoked, as with Labradors in (12). (12)
a. Kim got a Labrador puppy. b. They are very friendly dogs.
In other words, they is interpreted against a set of background possibilities that draw both on explicitly evoked entities and on a common-sense ontology of associations. In much the same way, descriptions of past events introduce new related, intervals into the discourse context. Once we know something happened, we can talk about what happened in the run-up to the event, or about what results the event brought about (Kamp 1979; Partee 1984; Hinrichs 1986; Webber 1988). Thus, in temporal reference, we find cases like (13) where we understand successive tensed verbs to describe simultaneous events, so the tense of the second verb depends on the time of the first event. (13)
John played the piano. Mary played the kazoo. (Webber 1988: 66 Ex 12ab)
But we also find cases like (14), where we understand a second tensed verb to describe what happened immediately leading up to the time of the previously mentioned event. (14)
John bought Mary some flowers. He picked out three red roses, two white ones, and one pale pink. (Webber 1988: 67 Ex 14ab)
FOR INDEXING
122
presupp osit ion an d a n a ph or a
The narrative continuation in (26) is straightforwardly explained as another instance of this general pattern. In narrative, the second tense verb describes what happened immediately following some previously mentioned event. Temporal reference is not just a feature of grammatical tense; it is a ubiquitous part of interpretation (Hobbs 1985; Enç 1986). Whenever sentence elements describe actions and states, implicitly or explicitly, there seems to be the possibility of locating their interpretation within a temporal interval that’s established in the context. Thus, consider the noun phrase a jogger in Hobbs’s (15). (15)
A car hit a jogger in Palo Alto last night. (Hobbs 1985: 7 Ex 4)
On its most natural interpretation, (15) suggests that the person involved in the accident was hit while he was jogging, not just that the person involved had a generic preference for jogging as exercise. These temporal dependencies in interpretation aren’t straightforward reflexes of structural relationships in the sentence involving syntax or surface scope. Consider (16), for example. (16)
A car hit a Stanford lecturer in Palo Alto last night. (A variant of Hobbs 1985: 7 Ex 5)
Here we do understand lecturer as a generic description of the crash victim—(16) need not involve a car barreling into a classroom! A different pattern of temporal reference is evident in (17). (17)
All rich men were obnoxious children. (Enç 1986: 207 Ex 3)
As Enç observes, one possible interpretation of (17) is that rich men, past and present, were obnoxious when growing up. In other words, rich men is understood to pick out anyone who fits the description during some independently established interval of times under consideration; the tense of were must vary with each man, much as we find in (9), to pick out the interval during which he was a child. Examples like (16–17) suggest that temporal interpretation is not determined directly by scope relations or other mechanisms of compositional semantics. But some authors have suggested that they are merely a side-effect of interpretive reasoning. For example, Wilson and Sperber (1998) suggest that temporal interpretation follows from listeners’ efforts to establish the optimal relevance of utterances. Meanwhile, Lascarides and Asher (1993) and Kehler (2000) argue that the coherence relations between clauses already necessitate many of the temporal relationships among events we find in interpretation.
FOR INDEXING
pre su pp o s i t i on a n d a n a ph or a 123 Hobbs (2010) extends this idea to the coherence relationships that apply within individual clauses. Of course, the temporal interpretation of utterances must be compatible with the constraints of coherence—just like any other aspect of interpretation. But we are convinced, for example, that the simultaneous reading of (13) represents a more specific interpretation than is required merely to establish coherence. Playing the piano and playing the kazoo will stand in a resemblance relation no matter when the events take place: compare the coherent, temporally unspecified interpretation of (3). Relevance Theorists have suggested that such specificity can be captured through open-ended processes of “pragmatic enrichment.” However, Stanley and Szabó (2000) offer convincing arguments that such specific interpretations must be a reflex of interpretive dependencies that specify values for variables in logical form. Accordingly, we think that the balance of the evidence supports the traditional approach as we have sketched it, which captures dependent temporal interpretation using mechanisms of discourse reference.
7.3 Linguistic Knowledge and Discourse Reference As we characterized it in section 7.2, a wide range of expressions in natural language involve discourse reference to temporal intervals. What makes tense special, then, is not that tense leads to dependent temporal interpretations in discourse. Rather, what makes tense special is the conventional meaning it has. Tense is associated with non-truth-conditional constraints that shape how the temporal interpretation of verbs is resolved across discourse. One such dimension of meaning is presupposition. We can think of presuppositions, following Stalnaker (1973), as a constraint on the common ground of the conversation. However, in this context it’s useful to observe, following van der Sandt (1992) and Roberts (2003), that these backgrounded elements of meaning often serve to constrain the resolution of interpretive dependencies in discourse. For example, consider (18). (18)
A man and a boy walked in. The man sat down.
The noun phrase the man in (18), like the pronoun he in (5), has an interpretation that depends on the interpretation of a man in the previous sentence. But it wouldn’t be felicitous to replace the man by he in (18), since a man and a boy both introduce possibilities for reference and there’s nothing in context that would distinguish one or the other as he. We can think of the meaning of the full noun phrase the man as serving to establish the interpretive link. That is, we
FOR INDEXING
124
presupp osit ion an d a n a ph or a
associate the man with a presupposition, which ensures that we resolve its interpretation using an established discourse referent for a man. The context in (18) does establish a unique such discourse referent (Roberts 2003). The past tense of plunged in (26) has a parallel significance. As we’ve argued in section 7.2, a dependent simple past tense verb is interpreted with respect to a discourse referent over temporal intervals. The verb locates the event it describes in time during this interval. The past tense contributes a further presupposition about this interval: it must be located before the time of utterance. A second such dimension of meaning is attention in discourse (Grosz and Sidner 1986; Grosz, Joshi and Weinstein 1995). We’ve seen with (1) versus (2) that realizing meanings through specific linguistic forms can conventionally shift interpretive preferences for subsequent dependent elements. The examples are repeated here: (1)
The President nominated Jones. He expected a quick confirmation.
(2)
Jones was nominated by the President. He expected a quick confirmation.
In cases like (1) and (2), the subject proves more prominent than an object as a candidate for coreference with a subsequent pronoun. Again, we find something similar with tenses. Using the simple past tense in English to describe a past event draws attention to the interval immediately afterwards, which becomes the default reference for subsequent tenses, in much the same way that the subject of a sentence establishes the default reference for subsequent pronouns (Webber 1988; Lascarides and Asher 1993). If a speaker wishes to describe a different temporal interval, the speaker must provide additional evidence of this alternative discourse relationship should he wish his audience to recognize the interpretation he has in mind (Lascarides and Asher 1993). Aspect carries a different kind of conventional meaning. For example, Moens and Steedman (1988) analyze perfect aspect in terms of a kind of conventional implicature. They argue that speakers of English use the present perfect to reference particular consequences of an event located indefinitely in the past, and to present those consequences as still holding in the present. This fits Oil prices have doubled, for example, where the speaker references the relevant result that oil prices remain much higher than they were at some indefinite point in the recent past. Similarly, with demand for consumer goods has plunged, the speaker references the result that demand remains lower than it was in the recent past. As a matter of convention, then, the discourse is knit together as a description of two present economic challenges, and not the contingent relationship between successive events.
FOR INDEXING
pre su pp o s i t i on a n d a n a ph or a 125 When we describe the verb doubled or plunged as the simple past form in English, or describe the verb have doubled or have plunged as the present perfect form, we are implicitly describing both the morphological structure of the verb and the interpretive constraints with which it is conventionally associated. Crosslinguistic evidence reminds us that verbs with similar morphological structure need not be associated with the same interpretive constraints. For example, the French verb in the passé composé has a closely analogous morphological structure to an English present perfect verb: it consists of a present tense auxiliary, typically but not always the cognate avoir of have, together with the perfect participle of the verb. Despite this analogy, the French verb is recognized as a past tense form because it prefers to participate in narrative chains via temporal reference; it carries no suggestion of present relevance (Kamp 1979; Kamp and Rohrer 1983). Verbs in the Slavic languages typically organize discourse dynamics by a primitive contrast between the perfective and the imperfective, which does not align with the contrast between, say, the simple, progressive, and perfect verb forms of English (Kagan 2007; Altshuler 2010). In short, the inventory of meanings for verbal morphology varies from language to language, and each language carries a further, arbitrary choice about which morphological element is mapped onto which meaning. This is what it means to say the interpretive constraints are conventionalized. We can now see precisely how our view undermines Grice’s view that the narrative interpretation of (26) is a CI. For us, the narrative interpretation of (26) reflects linguistically represented interpretive dependencies, in temporal reference, which are also resolved through interpretive constraints of presupposition and attention that are grammatically associated with the tense and aspect of English verbs. But the non-narrative interpretation of (3) also reflects linguistically represented interpretive dependencies, which are resolved through grammatically encoded interpretive constraints. In short, the difference in interpretation of (26) and (3) is a consequence of conventional meaning, not general interpretive principles.
7.4 Grice’s Arguments in Perspective The theoretical context we have offered in sections 7.2 and 7.3 substantiates the broad argument we sketched in section 7.1: Temporal interpretation is a matter of disambiguating the logical form of discourse, in a way that satisfies the interpretive constraints conventionally associated with the linguistic expressions the speaker has used. It is not a matter of Gricean elaboration of communicated meaning with additional, implicated meaning derived by principles of rationality.
FOR INDEXING
126
presupp osit ion an d a n a ph or a
This interpretive pattern—involving discourse reference, presupposition, and salience—is, we believe, a powerful and general mechanism for deriving precise interpretations for utterances in context. Researchers must be sensitive to the possibilities of such analyses, and their advantages, in considering interpretive effects involving alleged CIs. Indeed, it’s particularly important to appreciate the power of discourse reference, presupposition, and salience, because the framework sidesteps many of the arguments that are traditionally used to arbitrate between semantic and pragmatic contributions to interpretation. In particular, we can concur with the Gricean that the interpretive inference that leads to the conclusion in (2d) is not underwritten straightforwardly by a conventional specification of the truth-conditional meaning of the speaker’s utterance. The simple past defaults to narrative chain by introducing and referencing successive intervals. The present perfect conventionally encodes the present relevance of described events. These particular conventional differences are usually captured as a combination of presupposed content, rules for the dynamics of discourse referents, and constraints on discourse relations (Asher and Lascarides 2003). But many linguists argue that the constraints themselves are not naturally described in truth-conditional terms (Kamp and Rohrer 1983). These particular kinds of conventions escape many of the standard arguments deployed within the Gricean program. This point is worth explaining in detail. Recall from Chapter 2 that Grice specifically discusses how temporal order is communicated in an effort to criticize Strawson’s view that and in English has a distinct sense in which it means the same as and then. Grice observes that there are utterances where default temporal order is explicitly denied or canceled by the speaker, as with the short-sellers’ buy low and sell high from (1). Grice presents us with a theoretical choice between between a single meaning for and and general principles on the one hand, or two distinct conventionalized meanings for and and the same principles of rationality on the other. By utilizing what’s now called “Grice’s razor”—Grice called it “the Modified Occam’s Razor” Grice (1989: 47–50)—we must prefer the first account. The mechanisms invoked are already needed and independently motivated, but they allow us to dispense with Strawson’s alleged ambiguity. Grice is certainly right about there not being an ambiguity in and, but the arguments he presents do not carry over to effects like the pronoun-like interpretation of temporal restriction in the verbal domain. What we see going on is that the authors of (1) are exploiting referential ambiguity. There are two salient intervals, the trading interval in which the short sellers’ aims are to be realized, and the specific subinterval that follows the purchase. To interpret the
FOR INDEXING
pre su pp o s i t i on a n d a n a ph or a 127 second conjunct, we must locate the selling within one or the other of these two intervals, resolving the interpretive dependency in the way that the writers have in mind. We of course naturally default to the narrative resolution, but additional information can make us override this interpretation, and call for the other resolution. As far as the verb goes, the very same convention is at play in both cases. Since there is only one convention, there is no parsimony in deriving one or the other of the readings via an implicature; we can derive them both by a univocal, context-dependent meaning. In general, for this kind of case, where context-dependence is a possible ingredient, arguing about “cancelability” seems misguided, and Grice’s razor does not apply. To bring this point home, consider (19). (19)
But within this lengthy discussion, Stewart allowed Obama the time to frame responses the way he (Obama, that is) wanted to.3
John Grier, the writer of (19), uses a parenthetical remark to follow up his utterance of he and revise its default interpretation. Grier here uses he to refer to Obama. Grier might have instead used he in this context to refer to Stewart. Recognizing this potential ambiguity, Grier provides a parenthetical to indicate the interpretation he has in mind. Of course, for Grice, the interpretation of he is part of what is said. One cannot always describe a parenthetical that clarifies the interpretation the speaker has in mind as “canceling an implicature”. What’s happening in (19) is something else. But we see an exact parallel between (Obama, that is) in (19) and but not in that order in (1). Both parentheticals disambiguate the interpretive dependencies of discourse reference. The more we see elements of interpretation as conventional, the more pervasive we find this kind of explanation to be. Defenders of the derivation of utterance content from pragmatic principles have, we think, been insufficiently appreciative of its power.
3 John Grier, “Obama on ‘Daily Show’: Did he Best John Stewart,” Christian Science Monitor, Oct. 28, 2010.
FOR INDEXING
8 Information Structure: Intonation and Scalars
Our next exploration concerns the effect of yet another dimension of meaning: information structure. The grammar of information structure foregrounds what’s important about each new utterance, contrasts it with other relevant possibilities, and marks its place in the dynamics of the interaction. If that seems too abstract, consider this, more down-to-earth characterization of Information structure: Information structure is what English encodes with intonation. The rules that link intonation to information structure in English are not universal. Other languages mark information structure in very different ways. Information structure is distinct from the grammar of speech acts we considered in Chapter 6, which specifies the different kinds of moves that speakers can make with language, and it’s distinct from the grammar of discourse reference we considered in Chapter 7, which specifies how utterance interpretations take elements from context. Nevertheless, in all three cases, we believe, researchers have been misled by failing to appreciate the expressive resources of grammar. In this chapter, we will discover how often researchers have incorrectly attributed the interpretive effects of information structure to general pragmatic reasoning. Information structure is subtle, and has therefore been a matter of deep theoretical uncertainty in linguistics. However, there is increasing agreement on the broad outlines of how to think about information structure. We describe the key principles in section 8.2. Before getting into the complexities of the theory, however, we think it’s clearest to begin with a key example that epitomizes the dialectic—Kripke’s magic case (27) repeated here. (27)
Well, it looked red.
We will suggest that it’s information structure that gives (27) its meaning of disagreement. Information structure, in the full richness we see in Section 8.2,
FOR INDEXING
in for m a t i on s t r u c tu re 129 is just a generalization of the specific, grammatically encoded distinctions that let the speaker signal disagreement with (27). Once we understand the interplay of information structure and interpretation in (27) and analyze what we find in general terms, it will become clear that a broad range of apparent implicatures are in fact best explained as reflexes of encoded information structure. Chief among these, we will suggest, are the famous scalar implicatures we first explored in Chapter 3 and return to in detail in section 8.3.
8.1 A Case Study For Kripke, it seems clear that the sense of disagreement we find in (27) is not marked grammatically, but instead ought to be attributed to general pragmatic principles. In particular, Kripke concludes “the speaker [of (27)] meant that it [the handkerchief] may not have been red” because: (146) a plausible general principle of human discourse would have it that if a second speaker insists that a stronger assertion should be replaced by a weaker one, he thereby wishes to cast doubt on the stronger assertion. (Kripke 1978: 14) (Grice 1961 reasons in an analogous way.) Kripke’s discussion confirms our intuitive sense that what’s happening in this interaction makes sense. But the principles Kripke appeals to assume a particular way of understanding the contribution the second speaker is making. In what sense is the second utterance weaker than the first? How does the audience know it’s intended as a replacement? Why isn’t it understood as presenting complementary information? And why doesn’t the audience take it as an elaboration? We will suggest that we naturally understand the second speaker as we do because the speaker has used the grammar of English to mark her utterance as an expression of partial agreement, highlighting what she can and can’t accept about the previous speaker’s utterance. Thus, it’s grammar, not pragmatics, that’s responsible for the interpretation of the example. To see why this is so, notice how natural it is, in reading this example, to imagine it performed with a rise-fall-rise intonation contour, where only the word looked is prosodically prominent. Pierrehumbert and Hirschberg (1990) would write this way of delivering the utterance as in (147). (147) It looked L*+H red LH% The diacritics indicate the phonological status of the intonation contour as regimented by the grammar of English. English distinguishes two extremes of
FOR INDEXING
130
infor mat ion st r u c tu re
prominence in prosody, which are normally associated with excursions in pitch (among other cues). They are written L for low and H for high. English organizes its prosodic phrases into complex tunes by assigning prominence in two ways. The stressed syllables of key words can have an associated accents (marked by *), and the phrases as a whole can have associated tones (marked by %), which describe the prosodic contour of the phrase after its final accented syllable. So, the prosody diagrammed in (147) indicates that there is a rapid, late rise on looked (L*+H). This excursion is followed by an immediate fall to a low tone that’s sustained over the rest of the phrase until it is released at the end with a small rise (LH%). Although this rise occurs on the word red, it’s associated with the phrase as a whole and so does not give the word red prominence. This rise-fall-rise contour is conventionally associated in English with limited agreement (Pierrehumbert and Hirschberg 1990). In particular, as we will see in more detail in what follows, the accent on looked indicates that there are relevant alternatives to the red appearance of the handkerchief. The overall tune indicates that the speaker is uncertain about these alternatives. In short, the intonation encodes the speaker’s judgment that she is uncertain about what has been said because things can look one way but be another. To see how closely this convention cleaves to intonation in English, imagine the second sentence performed with a different intonation pattern, one with a high rising accent on It and a slightly less high rising accent on red, as in: (148)
It L+H* looked red L+!H* LH%
This particular pattern seems more like an expression of shocked agreement, signifying something like, “It certainly was red; I saw it with my own eyes.”1 This difference in function reflects the need to interpret the conventional meaning of this different intonation contour. For Pierrehumbert and Hirschberg (1990: 296), for example, the shape of the prosodic contour here signals the speaker’s expectation that the information presented is already mutually believed. At the same time, the accents continue to contrast this shared understanding with other possibilities one might imagine: particularly the other handkerchiefs, in other colors, the magician must have had tucked away. In each of these two cases, we described the first speaker as asserting that the handkerchief is red and the second as asserting that it looks red; there is no disagreement between these two propositions. However, contingent on which 1 The ! annotation in (148) indicates a compression of pitch range, called downstep or catathesis. The notation is due to Ladd (1983). While Pierrehumbert and Hirschberg (1990) analyze catathesis in a different way, Ladd’s approach has become a standard part of subsequent annotation of English intonation (Silverman, Beckman, Pitrelli, Ostendorf, Wightman, Price and Pierrehumbert 1992).
FOR INDEXING
in for m a t i on s t r u c tu re 131 intonation pattern the second speaker presents her sentence with, she expresses either doubt or support. Note how general this convention is. Consider the alternative dialogue in (149), where the speaker describes the handkerchief in the opposite way. (149)
a. And then he put the handkerchief that was not red on the side of the table. b. Well, it looked red.
If the second speaker still replies with the same characteristic intonation as in (27), she still signals disagreement. In sum, when somebody utters (27), intending to correct the preceding speaker, the audience cannot retrieve the intuition of disagreement just from the semantics. Red things can look black and white things can look red in the right light2 General knowledge about rationality is not enough either. Instead, understanding depends on recognizing a convention about how people tend to show what they are doing when they use particular utterances in their shared language. English grammar dictates how to employ intonation to encode these meanings in its peculiar way. Such rules fix the use speakers make of these constructions; and carry inferences tightly constrained by linguistic knowledge at a range of levels, from phonology to discourse.
8.2 The Grammar of Information Structure Our analysis of Kripke’s (27) is indicative of the power of information structure to signal a speaker’s assumptions about the contribution she is making to conversation. It’s important to appreciate that such effects are pervasive, not just isolated examples. At the same time, languages are surprisingly variable in the resources they recruit to achieve them. Some languages encode information structure with word order, e.g., Czech (Sgall, Hajio¸vá and Panevova 1986); others use syntax to mark information structure, e.g. Catalan (Vallduvi 1992); and still others use morphological elements, e.g. Japanese (Kuno 1973; Kuroda 1965) or Kallalissut (Bittner 2014). Moreover, the grammar of information structure—like all of language—is ambiguous. There can be uncertainty about what form a speaker has used and what meaning she has in mind. We’ll largely pass over this complication, but will return to it briefly at the end of the chapter. The literature on information structure calls attention to three broad classes of meanings that languages offer to allow speakers to frame the contributions 2 “What could be a clearer case of something which looks blue to me than the sky on a clear day?” (Grice 1989: 10).
FOR INDEXING
132
infor mat ion st r u c tu re
they make with their utterances. The first we will consider is the contrastive dimension, which lets a speaker indicate where her utterance differs from other possibilities that might have been expected. The second is the relational dimension, which lets a speaker distinguish the material that serves to advance the conversation from other material in the utterance that serves merely connect it to what comes before. The third is the interactive dimension, which lets a speaker position her contribution with respect to the evolving network of expectations and commitments which define the conversational context: Does it open up something new or set limits on what’s happened already? Does it take the initiative or leave matters in the hands of the audience? The different dimensions of information structure often overlap in the meanings of particular forms. Prosodic contours of English utterances, for example, encode particular meanings along all these dimensions simultaneously. Nevertheless, English—and other languages—offer resources to systematically vary each of these dimensions of meaning. This richness matters for interpretation. The reasoning we illustrated with (27), for example, involves all three dimensions. Thus, as we emphasize in our discussion of alleged scalar implicatures in section 8.3, researchers who approach pragmatic phenomena without a systematic perspective on information structure can easily miss the contributions of grammar to interpretation.
8.2.1 Contrast The most salient dimension of information structure for English speakers is the contrastive one. In English, the speaker has great freedom to differentiate an utterance from its possible alternatives by placing prosodic accents for emphasis. For example, if the context has established a choice among two articles of clothing, a red sweater and a red jacket, the speaker would naturally indicate her choice as in (150). (150)
I want the red jacket H* LL%
The notation H* LL% indicates that the utterance is to be performed with the canonical rise-fall tune associated with unproblematic assertions in English. The speaker articulates it with a single accent, on the word jacket. By contrast, if the speaker needs to express a choice between a red jacket and a blue jacket, as established alternatives, she might instead use (151). (151)
I want the red H* jacket LL%
FOR INDEXING
in for m a t i on s t r u c tu re 133 The difference in prosody between (150) and (151) lies in the placement of prominence. In (150), jacket is prominent and red is not; in (151), red is prominent and jacket is not. With more alternatives, richer intonation might be called for, as in (152) or (153). (152) I want the red H* jacket H* LL% (153) I want the red H* jacket !H* LL% (152) shows the default way to spread prominence across a phrase in English, with a series of accents with gradually increasing pitch across the phrase, and a corresponding sense of maximal emphasis at the end of the phrase. (153) describes a realization that gives balanced weight to red and jacket by realizing the second word in a slightly compressed pitch range. The variations in (150–153) suggest that we need to associate the utterances with differences both in form and in meaning. Thus, (151), in contrast with (150) or (152), indicates that the speaker’s preference for the red jacket differs from other potential preferences she might have expressed in this context not in that she wants a jacket, but specifically which color the jacket she wants is. Note that speakers do produce this intonation consistently in circumstances that match this interpretive constraint. Likewise, hearers apply this constraint in understanding utterances they hear, and have difficulty processing utterances if circumstances do not support the intonation they hear. (See Hirschberg 1993 for English or Swerts, Krahmer and Avesani 2002 for the very similar grammar of Dutch.) For English speakers, the pattern of accentuation in (150–153) seems almost inevitable. So it’s tempting to try to account for it on Gricean lines, as a matter of conversational implicature. The speaker chooses to emphasize one thing in (150), and another thing in (151). It’s only rational to emphasize what’s important in context—hence the apparent difference in the contexts where (150) and (151) are appropriate. As Sperber and Wilson (1986: 203) put it in developing this idea, “stress is a sort of vocal equivalent of pointing, a natural means of drawing attention to one particular constituent in an utterance.” For Sperber and Wilson, such stress amounts to an instruction to devote additional processing effort to the focused constituent. By their principle of optimal relevance, speakers and hearers will prefer interpretations where this processing effort brings relevant effects; thus stress is possible only when the stressed constituent is particularly important to the message the speaker aims to convey. However, we needn’t invoke Grice to observe that the particular kind of emphasis at issue in (150) and (151) is a way to draw distinctions about alternatives in context—rather we take this as the definition of emphasis. Crucially,
FOR INDEXING
134
infor mat ion st r u c tu re
moreover, it turns out to be a convention of English that the performance of (150–153) signals differences of emphasis in this sense. There are many languages where speakers cannot signal emphasis this way. Italian is often cited as an example. Swerts et al. (2002), for example, elicited noun phrase object descriptions from Dutch and Italian speakers. The experiment manipulated the context, much as we have done in (150–153), so that different material served to distinguish the target object from its salient alternatives. Dutch speakers consistently varied their intonation to signal this contrast, in ways analogous (for Dutch) to what we describe in (150–153). Italian speakers, however, did not. Acoustic analysis and perceptual experiments failed to find any association between the referential context and the utterances Italian speakers used.3 Since Italian speakers cannot use prosodic accents analogous to those in (150–153) with the contrastive interpretation found in English, this distinctive pairing of form and meaning must be part of the grammar of information structure in English.
8.2.2 Relation Next we turn to the relational dimension of information structure: The expressive resources that frame some parts of an utterance as a link to what has come before and mark other parts of the utterance as moving the conversation forward. We can get some intuition for the difference already in (150). The context has set up a choice for the speaker, and the speaker has chosen to announce her decision. So the sentence frame I want recapitulates the direction the utterance is taking in relation to this context. There are a range of ways of characterizing how this material hooks the utterance onto its context: By indicating what the utterance is about (Vallduvi 1992), by showing what question the utterance answers (Ginzburg 2012; Roberts 2012), or by anchoring the contribution of the utterance in the flow of information in conversation (Halliday 1967; Sgall et al. 1986). The theoretical differences are substantial, but need not concern us here. We call the constituents in an utterance that serve this function the theme of the utterance, following Halliday (1967) and Steedman (2000). The theme of an utterance is distinguished from the rheme—the part that makes a distinctive contribution, answers the relevant question, or, alternatively, most advances the dynamic of the conversation. 3 Note that Swerts et al. (2002) do not show that Italian speakers cannot signal this dimension of information structure. As they note, theories of information structure in Italian suggest that the relevant distinctions are expressed in variation in word order and syntactic structure at the level of the sentence, not at the noun phrase, and so weren’t available to subjects in their experimental protocol.
FOR INDEXING
in for m a t i on s t r u c tu re 135 If we understand utterances like (150) as answers to simple direct questions—for (150), what do you want?—we underestimate the interpretive significance of making the theme of an utterance explicit. Normally, a speaker has a range of avenues for continuing the discourse. This might include answering the main question on the table, answering (or partially answering) one of many subsidiary or ancillary questions, or addressing the next issue, which may unfold in a range of ways. Here it’s useful to recall from Chapter 6 the diversity of coherence relations in discourse. Speakers can continue a discourse using cause–effect relations, resemblance relations, occasion relations, evaluation relations, and adjacency pairs. Each case determines an implicit question that the subsequent utterance will answer (van Kuppevelt 1995). Say the discourse starts with the description of some event. Then cause– effect relations allow us to answer how the event arose or what the event brought about. Resemblance lets us answer how the event compares to other events. Occasion lets us answer what else happened along with the event, and specifically what happened next in narrative sequence. In each case, the theme of the subsequent utterance will be whatever material shows what question the utterance addresses. Since there are so many possibilities, the theme announces the speaker’s substantive new decisions about the organization of the discourse. The theme will not be determined uniquely by the discourse context or predictable in advance by the audience—a point advocated already by Bolinger (1972). This flexibility is an important reason to think of information structure as its own dimension of meaning in utterances. Information structure abstracts away from coherence relations; its effect is to organize the speaker’s contribution rather than to to specify the contribution itself. Nevertheless it goes hand in hand with that contribution; it’s not just presupposed background against which that contribution is interpreted. In English, the articulation of an utterance into theme and rheme, like the marking of contrast, is typically accomplished through intonation. In particular, English speakers use phrasing to delimit parts of utterances with consistent theme–rheme functions. For example, in a context where a group has been asked what color each would like to wear, it would be natural for them to answer, one at a time, with utterances with the intonation annotated in (154). (154) I L+H* want LH% the red H* jacket. LL% The prosody of (154) comes with a sense of perceived disjuncture between the words I want (the theme of 154, accent on I), and the words the red jacket (the rheme of 154, accent on red). The break between the phrases may be realized with a pause, but even without the pause, English speakers can use changes in vowel
FOR INDEXING
136
infor mat ion st r u c tu re
duration, pitch movements, and other prosodic cues to signal the boundary between phrases. The distinction between theme and rheme in English is marked by the kind of tune that speakers use on the phrase. In (154), we imagine the red jacket performed with the tune that accompanies rheme phrases that are contributed unproblematically to the discourse. This is a simple contour with a single slow rise followed by a sharp fall at the last accent—what we’ve been writing with H* accents, an L phrase tone and L boundary tone. In fact, this is the tune we have found on all the utterances of (150–153). (Steedman 2000 argues that the grammar of English intonation allows speakers to leave the theme unmarked.) By contrast, the first phrase I want of (154) is annotated for the typical contour by which English speakers mark themes, with a series of quick rises at each point of emphasis, followed by a low rise at the end of the phrase—what we’ve written with L+H* accents, an L phrase tone, and a H boundary tone. Note that the prosodic accent on I marks this theme as one of a set of alternative themes that a contribution might felicitously address at this stage in the conversation. This fits the context we have set up: As part of answering the general question of what everyone will wear, each of the participants may at any point address the question of what she herself will wear. Again, this use of intonation to mark theme and rheme—compelling as it is for English speakers—is an arbitrary fact of English grammar. Even closely related languages differ, by realizing these meanings with different tunes. German, for example, marks the rheme with a series of sharp falling accents (written H+L*), in place of the sustained high (H*) accents of English. Further afield, languages use a diverse range of cues to signal the relational dimension of information structure; see Drubig (2003) and Büring (2010). Languages often use word order and syntax, so that a particular position in the sentence is reserved for thematic material, while another might be reserved for rhematic material. É. Kiss’s (1987) analysis of Hungarian and Vallduvi’s (1992) analysis of Catalan are influential examples of this kind. Other languages mark constituents with particles—functional morphemes—that indicate their information status in relation to the broader discourse. Akan (a language of Ghana), for example, marks the answers to questions by fronting the material that provides the answer, and following it by the focus particle na (Boadi 1973; Drubig 2003). All this variability underscores the grammatical status of the particular means that English happens to use to encode such distinctions.
FOR INDEXING
in for m a t i on s t r u c tu re 137
8.2.3 Interaction The final dimension of information structure marks the status of speakers’ contributions within the dynamics of interaction. This too is specified by intonation in English. For example, the choice of tune in English can indicate whether a contribution stands complete on its own or whether the speaker intends to continue it with additional related material. The choice of tune can also indicate that the speaker regards the contribution as definitive, or that the speaker is waiting for the addressee to acknowledge the contribution before going forward, or even that the speaker expects the addressee to commit independently to the contribution. Pierrehumbert and Hirschberg (1990) includes a classic overview of this dimension of English intonation. To start, consider the English rising contour, written L* HH%, which indicates that the speaker’s contribution is not to make a predication. As Pierrehumbert and Hirschberg (1990) observe with (155), part of their example (17), this is used not only to ask questions but to express incredulity. (155) I L* should H apologize L* H H% The intonation of (155) accompanies words that ostensibly apologize, but with this delivery, the speaker uses it to express shock that he would be asked to do such a thing. A more subtle example of the same dimension of meaning is the contrast Pierrehumbert and Hirschberg (1990) draw between (156) and (157), their examples (59) and (60). (156)
a. My new car manual is almost unreadable LL% b. It’s quite annoying LH%
(157)
c. I spent two hours figuring out how to use the jack LL% a. My new car manual is almost unreadable LH% b. It’s quite annoying LL% c. I spent two hours figuring out how to use the jack LL%
The difference is that (156b) and (157a) end with rising tones, where (156a) and (157b) end with falling tones. The interpretive contrast is that (156b) and (157a) are presented as incomplete without a subsequent continuation. By contrast, the falling contour marks (156a) and (157b) as complete on their own. We naturally make sense of the distinction in information structure by resolving the referent of it differently: “whereas the referent of it in (60) [(157)] is likely to be interpreted as my new car manual, the referent in (59) [(156)] is likely to be understood to be
FOR INDEXING
138
infor mat ion st r u c tu re
‘my spending two hours figuring out how to use the jack”’ (Pierrehumbert and Hirschberg 1990: 305). Such interactive dimensions of information structure are again realized with different resources in different languages. For example, in the linguistics of Mandarin Chinese, there is a tradition of describing the meanings of the sentence-final particles ne, ba, ma, and a in terms of the organization of discourse and the interaction of speaker and hearer. Though the meanings of these terms are, at best, elusive (Li and Thompson 1981), Li (2006) offers a review and analysis that explicitly compares the functions of these particles to that of intonation in English, in terms of modulating the degree of commitment of the speaker to an utterance and indicating its relationship to the ongoing discourse.
8.2.4 The Meaning of Information Structure We can see, from the characterization that we have provided here, that information structure is different from truth-conditional meaning, as well as from the speech act meaning we consider in Chapter 6 and the presupposed meaning we consider in Chapter 7. The contrast between (158) and (159) makes this clear. (158) (159)
Bill L*+H* likes oatmeal L*+H cookies LH% Bill L+H* LH% likes oatmeal H* cookies LL%
(158) is a variant with the intonation of (27). It has late rising accents on Bill and oatmeal, embedded within an overall rise-fall-rise contour. As we’ve seen, this might be used to challenge a previous claim—perhaps, that nobody likes oatmeal cookies anymore. (159), meanwhile, showcases the structured question–answer intonation we saw with (154). Accordingly, it is a natural partial answer to a general question, such as What is everyone’s favorite cookie? There’s no truth-conditional difference between (158) and (159). Both utterances are assertions that contribute the same proposition to the discourse. The utterances do relate differently to the context—(158) is a challenge in a way that (159) is not—but it’s best to think of the intonation as helping to disambiguate the speech act in this case, rather than as specifying it directly. It’s the different rhetorical connections that set up the implicit questions and expectations that organize the discourse. These must be compatible with intonation, but, as we’ve seen, the same theme–rheme articulation is possible for utterances that contribute to the discourse in a wide range of ways. Finally, it makes sense to distinguish information structure from presupposition. Presupposition involves interpretive constraints that describe the context
FOR INDEXING
in for m a t i on s t r u c tu re 139 in which the utterance is used. By contrast, information structure characterizes the utterance itself and the message selected by the speaker. In this sense, information structure, unlike presupposition, involves choices made by the speaker rather than an established context that speakers and other interlocutors jointly track. Of course, as always, adding additional meaning to an utterance can have complex interpretive effects. Understanding—including its associated processes of disambiguation and accommodation—involves complex, integrative inference. We prefer to think of the alleged truth-conditional effects of association with focus (Rooth 1985, 1992) in these terms. At issue is the truthconditional contrast between (160) and (161). (160) We only groom dogs. (161) We only groom dogs. We use our ordinary textual conventions to present emphasis in (160) and (161), eliding the exact prosodic realization of the alternative utterances (which is unfortunately probably only an ambiguous indication of information structure in this case). If a kennel manager says (160), she rules out other services that the shop might offer for dogs—boarding them, walking them, training them, and so forth. When a kennel manager says (161), by contrast, she rules out grooming other sorts of animals—horses, cats, rabbits, or whatever. The difference coincides with the placement of the accent in (160) and (161). What the speaker accents is what differs across the alternatives that the speaker rules out. Semanticists beginning with Rooth (1985) have sometimes aimed to capture differences such as those in (160) and (161) directly with compositional rules. But we think the non-truth-conditional meanings we have presented suffice to explain the difference, provided we consider the effects of meaning on disambiguation. The key idea is that only is a quantifier over a contextually specified domain (Rooth 1992; von Fintel 1994; Stanley and Szabó 2000). Thus, both (160) and (161) say that grooming dogs is the only one of its contextual alternatives that the speaker and her colleagues do. To find a more precise interpretation, we must reconstruct the speaker’s context more precisely. At the same time, to understand (160) and (161), we must interpret the speaker’s intonation as an appropriate signal of information structure in context. For (160), the utterance must differ from alternative possibilities in involving the action of grooming as opposed to other actions, although the speaker takes it for granted that the object of all these actions can be expected to be dogs. By contrast, for (161), the speaker assumes that the actions under discussion can be expected
FOR INDEXING
140
infor mat ion st r u c tu re
to be grooming, but that different objects might be expected in context. Again, to get a precise interpretation, we must recognize how the context could establish these expectations. In effect, this is a problem of inferring the correct joint resolution of a family of ambiguities in an utterance. All the pragmatic theories we have considered assume that such resolutions are inferred in a coherent and economical way. Concretely, the interpreter needs to hypothesize a set of alternatives to interpret the domain of quantification for only. The interpreter needs to hypothesize a set of alternatives to interpret the contrastive intonation as well. The most economical hypothesis is that these alternatives are the same: Then the interpreter can interpret the utterance with reference to a single set of alternatives. Thus for (160) we assume that the relevant alternatives for only are things done to dogs, and for (161) we assume that they are grooming different animals. This is of course the interpretation we observe. We will have much more to say about disambiguation in Chapter 13. For now we note that the explanation we have just sketched could be formalized straightforwardly in many pragmatic theories. This is particularly true of Hobbs, Stickel, Appelt and Martin’s (1993) implementation of Coherence Theory in their model of interpretation as abduction. This model explicitly predicts that the most likely interpretations are generally achieved by merging redundancies in this way. Though this account is consistent with meaning and interpretation as we have characterized it throughout Part II of the book, it might seem suspiciously indirect. Examples like (162) show that some indirection is needed, however. (162)
People who [grow]f rice generally only [eat]f rice. (Rooth 1992: Ex 70)
The notation here indicates that the emphasis in (162) is on grow and eat. In particular, (162) is felicitously performed with prosodic prominence on the word eat and no prosodic prominence on its argument rice. Nevertheless, (162) means that rice is the only (relevant) thing rice-growers eat. It does not mean what you’d get if you strictly associate the interpretation of only with intonation: That eating is the only relevant thing that rice-growers do with rice. This is in line with an abductive account of interpretation. With (162), unlike (160) and (161), we can explain the intonation with the contrast that’s overtly presented in the utterance, between growing rice and eating rice. No accommodation of additional contextual alternatives is needed to interpret the intonation. We must still recover the domain of quantification for only, as in (160) and (161), but we make this assumption based on the most likely direction for the speaker’s message, without any pressure to link the domain of quantification for only to
FOR INDEXING
in for m a t i on s t r u c tu re 141 the speaker’s prosody. That’s key here—since the cases in questions are precisely ones of people both growing rice and eating it.
8.3 Information Structure and Scalar Implicatures The phenomenon of scalar implicature is one of the prototypical cases of CI in Grice and one of the most extensively researched and hotly debated areas in the study of language use. Views remain wildly divergent. Geurts (2010) defends a classic Gricean account, while Horn (2006) emphasizes the continued importance of broadly Gricean interpretive principles in the neo-Gricean account he favors. By contrast, Chierchia (2004) and Chierchia, Fox and Spector (2012) argue for representing the associated interpretive effects in logical form, in tandem with the semantics delivered by the grammar, and without reference to general interpretive principles. We cannot hope to survey the research or settle the issues here—that would take a book of its own. But we believe that the grammatical views have more to recommend them than is usually appreciated. The debate is generally regarded to be an empirical one, with the key issue turning on the status of embedded implicatures, of the sort discussed by Levinson (2000); see Chapter 3.2. Of course, some interpretive effects could be regarded as decisive evidence for accounts that derive them in tandem with compositional meaning. But there is, in fact, broad agreement on the idea, shared by Horn and Levinson and explored throughout Chapter 3, that scalar implicatures are constrained by linguistic rules. The question is what form the rules take, and particularly where general interpretive principles come in. Answering the question requires us to put the data and the theories into perspective.
8.3.1 Scalars: Delimiting the Phenomenon Much of the literature is insufficiently precise about delimiting scalar implicature. As Bach (2005) emphasizes, in considering alleged implicatures, it’s always important to restrict our attention to meaning that the speaker is committed to getting across. In Chapter 4.3, we saw how important this is with implicatures in general, but this is a particularly fraught issue in the case of the scalars. The exchange in (163) illustrates the difficulty. (163)
a. A: You eat a lot of Brussels sprouts. What do you like best about them? b. B: They are very nutritious.
Speakers’ choices may always reveal their implicit assumptions and expectations. In (163b), for example, B probably would have said that she thought Brussels sprouts were delicious if she did. So A can reasonably infer that B wouldn’t eat
FOR INDEXING
142
infor mat ion st r u c tu re
them just for the taste. But there’s no reason to assume that this has to part of B’s meaning in (163b). In fact, B might be surprised and even embarrassed that she has inadvertently let this slip through her description of the food. Similarly, there may be cases where the speaker knows that what she says will be potentially revealing, but is simply indifferent as to whether her audience is going to catch on. Consider (164) for example. (164)
I was feeling generous so I gave a whole bunch of the cookies away.
Since the speaker describes the quantity she gave away as a whole bunch of the cookies, it’s probably safe to assume that the speaker didn’t knowingly share all of the cookies she had. But this assumption seems immaterial to the point the speaker wants to make here, and the speaker may not be committed in any way to the audience recognizing this assumption. What seems to matter in (164) is the exuberance suggested by the absolute number of cookies she shared. This is not something that an alternative expression like all of the cookies I baked would necessarily convey—unless the audience already knows that it was a big batch. Since the assumption is not part of the speaker’s communicative intention and is not necessary to explain the cooperativeness of the speaker’s utterance, this is not an implicature either, in Grice’s sense. To fit Grice’s definition, the interpretation has to be something the speaker is committed to getting across—something that’s required for her to be cooperative. This does happen with many cases, as in the exchange in (165). (165)
a. A: Did I eat all of the cookies? b. B: You ate some of the cookies.
(165b) is the same sentence (42) considered by Horn, but now put into a specific context. (In particular, we imagine A suddenly worried that the cookies B left out might have been destined for the party, not just surplus snacks.) Unless we recognize (165b) as contributing that A ate some but not all of the cookies, we can’t take (165b) as answering A’s question, even partially. The utterance wouldn’t make sense in context. Such cases are plentiful, but like Kripke’s (27), we will suggest, these cases have special features which show that the enriched interpretation is linguistically marked rather than simply inferred via general interpretive principles. We believe that the discussion of scalar inferences that hearers may draw from utterances is not always careful about the distinction between what speakers communicate and what they simply reveal. But the distinction is important. What speakers reveal from their choices is a matter of pure psychology, not a
FOR INDEXING
in for m a t i on s t r u c tu re 143 matter of cooperative rationality at all. We made this point earlier in Chapter 4, in connection with utterances like (82), repeated here: (82)
You act like you’re my father or something. You have this need to control me or put me down.
This utterance allows the listener to infer that the speaker was frustrated, hurt or embarrassed: only someone who felt that way would say something like this. But it doesn’t implicate that. Whenever an agent considers and rejects an action, that agent must have appraised that action as unsuitable for achieving the agent’s goals—regardless of whether those goals are cooperative and regardless of whether the agent intends to share her appraisal about the action in question. To the extent that there’s anything systematic to say about such inferences at all—and, despite the success of Pinker, Nowak and Lee (2008), some skepticism may be justified (Fodor 2000)—implicature is the wrong paradigm for getting clear on them.
8.3.2 The Status of Competing Explanations Any Gricean approach to scalar inferences faces the so-called symmetry problem (Block 2008). The calculation of the scalar inference depends on the alternatives that are taken into account in assessing the speaker’s utterance. Take Horn’s (42). One set of alternatives that we can use for assessing this utterance is the pair ⟨some, all⟩. All is stronger than some. Grice says, if the speaker could have said all she would have. But she didn’t, so she must have judged it unsatisfactory. Thus, she is distancing herself from the claim that the addressee ate all of the cookies. But here’s another set of alternatives that we can use for assessing this utterance: ⟨some, some but not all⟩. Again some but not all is stronger than some. So you could argue: if the speaker could have said some but not all she would have. But she didn’t, so she must have judged it unsatisfactory. Thus, this speaker is distancing herself from the claim that the addressee ate some but not all of the cookies. If the speaker knows how many cookies the addressee ate, it must have been all of the cookies. The symmetry problem is a way of exposing an assumption hidden in the Gricean logic that explains alleged scalar implicatures. The assumption is that all is an option the speaker considered and rejected in choosing to say (42), whereas some but not all is not. It seems strong indeed to claim that this is a fact about general rationality, rather than a fact about grammar, as embodied in the language faculty and speakers’ mutual expectations about how to use language
FOR INDEXING
144
infor mat ion st r u c tu re
to convey specific messages in context.4 In fact, as we saw in Chapter 3, neoGriceans do advert to linguistic knowledge and processes in explaining the scalar calculation. Horn (1972, 1984) has scales, determined by the lexicon, which say what the alternatives are. Levinson (2000) goes further, hinting at a processing account where scale structure is closely related to the discipline that the language faculty has for choosing words and building complex linguistic structures. See also Block (2008) for a review of other formal proposals deriving alternatives through grammatical mechanisms. The point is that these theories do see scalar inferences as part of grammar, not really as implicatures in the Gricean sense. It just happens to be a part of grammar that takes a particular form that looks rather different from semantics as we typically understand it. From this point of view, Chierchia’s (2004) alternative “semantic” account of the scalars does not seem like a particularly radical departure. Chierchia assumes that the mechanisms that trigger the enriched interpretations associated with scalars are represented as operators in logical form, and the alternatives for comparison are contextual parameters for these operators. Thus, Chierchia’s difference with Horn is not in the grammatical status of the mechanisms and knowledge through which the interpretations are derived. Instead, it’s a difference in how these mechanisms are formalized and how they interface with other aspects of grammar. If operators in logical form and parameters for context dependency are the best tools we have to capture these kinds of interpretive effects, as von Fintel (1994) and Stanley and Szabó (2000) argue, then Chierchia’s approach is probably the right one to go with. This verdict can seem unsatisfying. Horn’s and Levinson’s presentation of their views sounds principled and general, while Chierchia’s approach sometimes looks just like a technical reimplementation of the basic facts. But this impression overlooks what is deep and general about Chierchia’s work—namely, its links with formal models of information structure, as explored in section 8.2.
8.3.3 Information Structure Enriches Scalar Meanings Van Kuppevelt (1996) proposes that alleged scalar implicatures are actually an interpretive consequence of the grammar of information structure. To motivate this account, consider how the theory of information structure applies to the question–answer pair in (165). Information structure marks a theme that specifies the question under discussion that the utterance provides information 4 Geurts (2010) offers an attempt to justify implicatures on grounds of general rationality, but we think it doesn’t work. It appeals implicitly to conventional asymmetries between different forms—since the argument effectively requires interpreters to pick a preferred “pragmatic” interpretation in cases where no grammatical option is licensed.
FOR INDEXING
in for m a t i on s t r u c tu re 145 about. For (165b) the question is what the speaker ate, and, in particular, what fraction of the cookies it was. Meanwhile, information structure marks a rheme as providing the speaker’s answer to that question. Contrast within the rheme indicates how the answer the speaker is giving differs from the other answers that one would naturally entertain in this context. In (165b), the rheme is some of the cookies, and (as you’ve no doubt imagined in reading the utterance to yourself) some must be marked for contrast. In other words, in (165b), information structure marks this utterance, as a matter of grammar, as specifying the fraction of cookies the speaker ate by giving the answer some of the cookies to the exclusion of other salient answers, in particular all of the cookies. If this is the right way to think about the meaning of the intonation in the English articulation of (165b), then the enriched interpretation of some we find there is contributed not by interlocutors’ open-ended reasoning about rationality and cooperation, but simply as a matter of grammatical rules. Chierchia’s approach, drawing as it does on the formal tools that semanticists have developed to characterize information structure, seems an eminently reasonable way to capture this. Van Kuppevelt offers extended arguments that alleged scalar implicatures are indeed a consequence of information structure. We refer readers to van Kuppevelt (1995, 1996). But to a large extent, his work explores in more detail principles that are already in evidence from our discussion of (165). Once you have excluded nebulous cases of hearers’ psychological inference from consideration, as we’ve seen you must, the cases where scalars get enriched interpretations are closely tied to specific kinds of discourse contexts. In these cases, coherence places strong constraints on the information structure that speakers could possibly use to articulate their contribution to the ongoing discourse. But information structure applies in these utterances with the same general interpretive force that’s demonstrated in many other kinds of utterances (including utterances without scalar interpretations). The meaning of information structure of these utterances in these contexts, however, has the effect of enriching the interpretation of scalar terms. We needn’t endorse the details of van Kuppevelt’s proposal to observe that the line of inquiry is a compelling one. We think that the promise of this way of thinking undermines what appeal the Gricean story might at first have promised.
8.4 Summary In this chapter, we have surveyed marking of the emphases and contrasts in utterances, marking of the topics of utterances and the issues speakers address,
FOR INDEXING
146
infor mat ion st r u c tu re
and signals of the negotiated back-and-forth of dialogue. These distinctions of information structure involve the functional organization of contributions to conversation, and so apparently invite explanations in rational terms. Indeed, the literature in linguistics and philosophy offers numerous cases, from (27) as described in section 8.1, to (165) and the other scalar cases described in section 8.3, where researchers have often emphasized Gricean explanations for what we argue are in fact the effects of information structure. Though we’ve offered a tour of the kinds of meanings researchers have proposed for information structure, the topic remains extremely controversial in linguistics. Researchers disagree about fundamental issues in information structure and its interface with grammar. One question is the extent to which information structure is specified holistically or compositionally. For English, for example, some research seems to isolate specific tunes with specific functions, such as Ward and Hirschberg (1985) or Hirschberg and Ward (1995). At the same time, researchers also look for general principles that apply compositionally to yield the interpretation of specific tunes from the accents and tones that make them up, as in Pierrehumbert and Hirschberg (1990). It’s unclear which perspective is most fruitful. Related problems surround the fundamental notions at play in information structure, including how best to characterize the relevant notions of contrast, theme, and interaction. Finally, controversy surrounds the mapping between surface forms and the underlying distinctions of meaning, since surface forms often seem to be ambiguous or underspecified for information structure in certain respects. Characterizing these ambiguities is particularly challenging because researchers make different assumptions not only about the elements involved and their interpretations, but also about the broader relationships between phonetics, phonology, syntax, and semantics. Readers should be prepared that ongoing research does not offer consensus on these issues. Though researchers must be sensitive to these open issues in developing future theories, we think the take-home message is inevitable. Information structure is a matter of grammar. Speakers encode distinctions of information structure using the particular grammatical resources of their native language. Kripke’s sense of disagreement in (27) is a consequence of the distinctive rules of English, and not just because that way of articulating a contribution makes sense. General pragmatic principles do not contribute this content to interpretation.
FOR INDEXING
Summary of Part II and Projection
Though truth-conditions, and the lexical and compositional rules required to derive them, are crucial to meaning in language, that’s not all there is. In Part II, we have seen evidence that the rules for meaning in language govern a wide range of other interpretive effects. Chapter 6 argued that linguistic rules specify what contributions utterances can make and how those contributions fit together into coherent discourse. Chapter 7 observed that these contributions are often parameterized by elements drawn from the discourse context, and argued that grammatical rules constrain how parameter values are obtained. Chapter 8 argued that speakers use additional grammatical devices to mark distinctions in information structure, including what parts of the utterance answer implicit open questions, what parts contrast with alternatives, and what parts need to be followed up.5 Our exploration of these linguistic domains has served three purposes. First, it has allowed us to challenge a wide range of claims from Part I of the book that specific interpretive effects are the result of general interpretive principles. Examples have included the indirection of many speech acts (Chapter 6.1), the inferential connections involved in establishing a coherent relationship between an utterance and its context (Chapter 6.3), the temporal flow of narrative discourse (Chapter 7.3), the many inferred comparisons associated with marked prosody (Chapter 8.1), and scalar implicatures in particular (Chapter 8.3). This covers a wide swath of the implicatures from Grice’s group a (Grice 1989: 32), which he saw as derived straightforwardly from the maxims without any apparent violation of the CP. It also covers a wide swath of the data addressed by other 5 We make no claim that these dimensions of meaning exhaust truth-conditional meaning. For example, we have had little to say about Grice’s familiar category of conventional implicatures. More generally, linguistic research such as Tonhauser, Beaver, Roberts and Simons (2013) suggests that elements of conventional meaning that are somehow backgrounded differ greatly not only in what kind of contribution they make but how that contribution feeds into compositional meaning. We have explored just a few of the many possible cases.
FOR INDEXING
148
summar y of par t ii and projection
researchers in the Gricean tradition, as surveyed in Part I. No matter what theoretical conclusions you might draw, the empirical considerations of Part II seem to substantially diminish the scope of CI. The second point of Part II has been to start to put the relationship between semantics and rational inference in critical perspective. In particular, we have come to see disambiguation as a powerful mechanism that inevitably connects utterance interpretation to speakers’ goals and points of view in conversation. The idea is that the rules of language fix meanings for sentences in context only under a specific grammatical analysis. When a speaker makes an utterance, there is often ambiguity. If the speaker is cooperative, however, it will be clear how to resolve the ambiguity, and the resulting interpretation will be one that fits the accepted purpose or direction of the conversation. We have used disambiguation as a foil to clarify what’s distinctive about Grice’s view and the related approaches we surveyed in Part I. They have interpretive principles that do not simply disambiguate but additionally enrich or reinterpret the linguistically determined meaning. The example of disambiguation shows what is at issue in accepting pragmatic enrichment or reinterpretation. There is no dispute that interlocutors are normally cooperative. There is no dispute that interlocutors can use their presumption of cooperation to inform utterance interpretation. The dispute is instead about the nature and scope of the rules of language. We have argued in Part II than they are richer than one might have at first suspected—especially in light of the bare-bones linguistic rules invoked by the accounts we considered in Part I. The third and final point of Part II has been to critique the specific arguments of Gricean approaches. Researchers have often argued that interpretive effects are derived by general principles on the grounds that the effects are not always present: the effects are present exactly when they make sense. We do not think this gives even prima facie evidence for a reinterpretation account. A disambiguation account makes exactly the same prediction. Researchers have often argued that interpretive effects are derived by general principles on the grounds that it requires less stipulation than a rule-governed account. In so doing, they have generally underestimated the potential of rulegoverned accounts not only to stipulate ambiguities but to derive and predict them. Finally, theorists have been quick to assume that interpretive effects are derived by general interpretive principles simply because they are not triggered by truth-conditional aspects of meaning. But there are lots of different aspects of meaning, and their contribution to interpretation is both wide-ranging and
FOR INDEXING
su m m a r y of p a r t i i a n d pro j e c t i on 149 subtle. In short, to convincingly rebut a disambiguation account, the case for general interpretive principles needs to be made in much more empirical and formal detail than is often attempted. None of this, however, so far at least, requires a dramatic change in the general Gricean division between semantics and pragmatics. Grice recognized that the conventional meaning of a sentence often goes beyond what is said (Grice 1989: 122); that is, there are aspects of conventional meaning that play no role whatsoever in fixing what the speaker says with his utterance. In particular, Grice drew a “principal difference between a conventional implicature and a conversational implicature” in “that the [former] depend upon the presence of some particular conventional device . . . whereas [the latter] do not” (Neale 1992: 524; compare also Grice 1989: 25–6, 31, 359–68).6 One classic example of such a device is the connective term but. The difference between uttering (166) and (167) is that anyone who utters (166) will very likely be taken to be drawing some sort of contrast between poverty and honesty, whereas this is not the case with an utterance of (167) (166) She is poor but she is honest. (167) She is poor and she is honest. For Grice, the opposition of poverty and honesty is not part of what the speaker says with her utterance of (166); that is, it is not part of the truth-conditions for that utterance. Thus, according to Grice, utterances of (166) and (167) say the same thing. But, just the same, the opposition of but is conventionalized. Perhaps, then, the conclusion of these last few chapters should be no more than that the devices for conventional implicatures are much more pervasive and flexible than has been hitherto recognized.7 If this is so, then Grice’s story about the privileged place of strategic reasoning in connecting content to conversation continues to be plausible—it is certainly possible for an audience to understand a speaker’s choices, as well as her words, and it is certainly possible for a speaker to anticipate and exploit this understanding. Sustaining Grice’s story, however, would require us to motivate the 6 Of course, there are those, Blakemore (1987, 2002) and Bach (1999), who criticize the notion of conventional implicature and offer alternative accounts. 7 Grice’s idea was that conventional implicatures were essentially lexical in nature, with minimal interaction with compositional semantics and non-linguistic context. This understanding remains in recent work on conventional implicature, e.g. Potts (2005) and Williamson (2009). By contrast, linguistic work on dimensions of non-truth-conditional meaning, such as presupposition, has emphasized the interaction of non-truth-conditional meaning with context and compositional semantics; see e.g. Kamp and Rossdeutscher (1994b,a), or Tonhauser et al. (2013). Work in computational linguistics also emphasizes the integrative inference that is needed to resolve ambiguity (e.g. Hobbs et al. 1993; Asher and Lascarides 2003). A Gricean would need a much more expansive understanding of conventional implicature to cover all these cases. Nevertheless, the revision does not seem to involve a deep architectural change.
FOR INDEXING
150
summar y of par t ii and projection
viability of the category of CI through a different set of cases. In the next part of this book, therefore, we will assess the prospects for doing this by turning to cases where the inferences in question are less amenable to conventionalized treatments. Cases that remain are, characteristically, instances of hints or suggestions. In these cases, speakers clearly intend their utterances to get something across to their audiences, but it is not so clear how to delineate precisely what the information is, or what reasoning an audience needs to do to appreciate it. Indeed, the phenomena we will explore seem intrinsically open-ended. Of course, in some of these cases (e.g. of hinting and suggesting), we can see audiences reasoning about the speaker and her rational choices. But these cases seem to be just one interpretive practice among many. Indeed, we think other figures that theorists ordinarily catalogue as cases of CI—metaphor, sarcasm, irony, and humor—recruit distinctive ways of thinking about, and appreciating, the imaginative world implicit in our understanding of the speaker’s utterance. To call these effects pragmatic is to miss both their heterogeneous origins and their non-propositional effects. Given the broader space of meaning and interpretation, we are quite skeptical that any useful work can be done specifically by the category of conversational implicature; and so, to the extent that the traditional division between semantics and pragmatics relies on the category of CI, we are deeply skeptical about this division. We begin to mount our case by turning to the sorts of alleged cases of CI that seem not to rest on convention.
FOR INDEXING
PA RT III
Varieties of Interpretive Reasoning
FOR INDEXING
FOR INDEXING
Introduction to Part III
In Part II, we explained a range of interpretive effects by appeal to particular linguistic rules rather than generative interpretive principles. If we are right, these effects are not CIs; they escape the explanations of the theories we reviewed in Part I. But the majority of the analyses in Part I—and all the most challenging and exciting ones—still remain. It’s to these that we now turn. The subject of Part III of the book is cases whose interpretations, all agree, outstrip the rules of language. We are particularly concerned with cases of figurative language, including metaphor, sarcasm, and irony, along with other kinds of evocative language, such humor and hinting, whose import also escapes conventional meaning, broadly construed. These tropes lead to clearly open-ended, even creative interpretations. The question is what guides these interpretive effects. The theories of Part I have suggested a range of reasoning mechanisms that might account for them. One possibility is that the effects reflect a coordinated appeal to standards of rationality, as characterized in Chapter 2. Another is that they follow either from general mechanisms of understanding or from hearers’ reasoning about speakers, as surveyed in Chapter 4. Indeed, these are not mutually exclusive possibilities. All could figure in important ways in language use. (However, if these effects are not governed by linguistic rules, then the linguistic mechanisms explored in Chapter 3 do not apply.) What we will challenge about these reasoning mechanisms here, however, is their generality. The mechanisms we explored in Part I are all intended to apply in more or less uniform ways across a wide range of figurative and evocative language. This seems a strong assumption. We will now consider whether the particularities of figurative and evocative language can sustain such uniform explanations. We will suggest that they cannot. As with Part II, we will emphasize the rich particularities of different interpretations. We will argue that these need to be explained on their own terms,
FOR INDEXING
154
int ro duc t ion to p a r t iii
so we will suggest that pragmatics can do no more than disambiguate independently provided candidate interpretations. At the same time, however, we will emphasize some empirical and interpretive contrasts that, we believe, set the interpretive effects we survey in Part III apart from those in Part II. We will argue that there are no linguistic signals of the interpretive effects we survey in Part III. Moreover, specific processes of imaginative engagement are essential to understanding them. Thus, there is a real divide between the cases of Part II and those of Part III. We need new tools to make sense of it: Part IV will describe the approach we recommend. We begin, as in Part II, with some methodological preliminaries and a tour of the data.
FOR INDEXING
9 The Scope of Interpretive Reasoning
Linguistic rules, we have seen, are responsible for a diverse range of contributions to interpretation. So what, if anything, does reasoning contribute to interpretation? What’s crucial about interpretive reasoning, on our view, is that its interpretive effects are not due to convention. That raises a methodological quandary. Recall from Part II that we delimit linguistic knowledge by drawing closely on Lewis’s (1969) lightweight understanding of conventions. Conventions are simply the mutual expectations that let some group of agents successfully coordinate. On this understanding, it is very difficult to rule out the role of convention in any particular linguistic exchange. People in conversation are rarely complete strangers. They have normally enjoyed a substantial history of interactions with one another. Accordingly, they will likely share a complex tapestry of precedents and mutual expectations that can inform their understandings of one another. Especially when examples are simple— and repeated unchanged across the scientific and philosophical literature—the possibility looms large that our intuitions are informed by meanings that are idiosyncratic but nevertheless conventional in our sense. As we proceed, we will therefore strive, methodologically, to focus, as much as we can, on clear examples of creative interpretation. Accordingly, each of the chapters that follows begins by offering an argument that the interpretive effects at issue do not get their interpretations by established conventions. However, our point is not just that these interpretive effects are improvised: Our point is that they also differ from one another. These differences don’t just seem to be a matter of context or background knowledge: We see no way to capture them without appealing to distinctive imaginative processes that the audience brings to the interpretive enterprise. This, for us, underscores the importance of treating these interpretive effects in a qualitatively different way from the effects we explored in Part II.
FOR INDEXING
156
the scop e of inter pre t ive re a s on i n g
9.1 The Cases We begin by introducing several of sorts of examples that provide plausible cases of non-conventionalized patterns of reasoning: Metaphor (which we explore in Chapter 10); non-serious utterances, including cases of sarcasm, irony, and humor (which we explore in Chapter 11); and hinting (which we explore in Chapter 12).
9.1.1 Metaphor According to Donne’s famous line: (168)
No man is an island entire of itself; every man is a piece of the continent, a part of the main.1
Taken literally, Donne’s text appears to characterize people as a certain kind of territory—not islands but fragments of larger wholes. In actuality, people are neither. But literal truth is not what Donne is after in this passage. In Donne’s metaphor, people are places, so that the interpersonal connections that people have with one another are reimagined as geographical relationships. Donne’s (168) charts the topography of this metaphorical world, where nobody is fully autonomous, and everybody is in part defined by those beside them. Donne’s point is for his audience to explore this imagery, and come to an appreciation of our social bonds that is informed by it.
9.1.2 Sarcasm Consider Anne and Dana’s sarcastic dialogue in (169), reported by Gibbs (2000: 6). (169)
a. Anne: Isn’t it so nice to have guests here? b. Dana: Totally! c. Anne: I just love it, you know, our housemates. They bring in the most wonderful guests in the world and they can totally relate to us. d. Dana: Yes, they do.
If Dana’s (169b) were sincere, it would be an expression of complete agreement that the house guests are exceedingly agreeable; if Anne’s (169c) were sincere, it would be a hyperbolic commendation of her housemates’ sociability. But in this conversation, Anne and Dana are not speaking sincerely. Instead, Anne and Dana’s point is a complete inversion of what we’d find in those sincere uses. The guests are unpleasant and difficult; the housemates are inconsiderate and incompatible. The point of Anne and Dana’s sarcasm seems to be to acknowledge their 1 Part of Meditation 17, from Devotions Upon Emergent Occasions (1624).
FOR INDEXING
the s cop e of i n ter pre t ive re a s on i n g 157 mutual displeasure at their living situation in a way that reinforces their sense of intimacy and rapport with one another.
9.1.3 Literary Irony Consider the description of intellectual progress in the eighteenth century in (170), presented in Jon Stewart’s mock textbook America (The Book) as the caption for a gratuitous diagram of the human brain (2004: 11). (170) The Enlightenment’s focus on reason raised the profile of the brain, which became the organ considered most important to human character, overtaking the spleen. Like many examples of literary irony, this utterance evokes and criticizes an imagined author for effect. The utterance is not exactly wrong: Enlightenment thinking did privilege rationality, did bring a scientific understanding of the functions of the body in which the brain was particularly prominent, and did displace an ancient and widely held theory of bodily humors, in which the spleen did have a special place as the seat of anger and bad temper. But because the utterance is based on these truths, we can see all the more clearly the misplaced efforts of Stewart’s imagined textbook author in simplifying the facts for easy understanding. In (170), scientific progress is described in the language of popularity and polling: the brain gets a “raised profile” and “overtakes” the spleen. It’s as if genuine explanation is too cumbersome—or the author thinks science really is just a beauty contest. But the resulting “simple” story is almost absurdly incomprehensible. Thus, the real point of (170) seems to be how selfdefeating feigned simplicity ultimately can be. It’s a point that Stewart makes often in critiquing journalists and their “horse-race” and “sound-bite” coverage of American politics.
9.1.4 Humor Though examples (169) and (170) involve not only sarcasm and irony but also humor, we shall see that sarcastic and ironic utterances need not be intended as humorous. Meanwhile, humor occurs in many cases that are not sarcastic or ironic the way (169) and (170) are. As an example, consider the conversation in (171), as reported by Lampert and Ervin-Tripp (2006: 54). (171) Friends talk about Diet Pepsi. a. Vi: There is a problem with aspartame, but I can’t remember what it is. b. Joe: It affects your memory. c. Vi: Hah! That’s a good one.
FOR INDEXING
158
the scop e of inter pre t ive re a s on i n g
This exchange is part of a conversational thread prompted by the observation that the Diet Pepsi that Vi is drinking contains aspartame. Conventionally, Joe’s utterance carries the content that aspartame affects memory. But Joe does not seem to be using his utterance to make such a claim. Instead, Joe and Vi understand Joe’s contribution as a joke. Joe’s utterance evokes the possibility that Vi’s difficulty in recalling what she has heard about aspartame is in fact a symptom of the very problem she wants to talk about. Humorous imagery characteristically offers a surprising perspective that shows its topic in a new, negative light. That’s certainly the case here: Joe’s joke undercuts both the dangers of aspartame and Vi’s competence as a conversationalist. Joe’s point seems to be to ridicule Vi and her worries, rather than to contribute directly or even indirectly any information to the conversation at all.
9.1.5 Hinting Finally, there are utterances whose import is allusive and indirect, without being figurative or insincere. An evocative case is that of the examiner who, after casually looking over a completed final exam handed in to her by a student, responds with an utterance of (172). (172)
You might want look at your answer to problem three again.
Under these circumstances, the point of (172), though the speaker seems to hold back from actually saying so, is to invite the student to revisit his answer, notice his error, and correct it. In interpreting this utterance, it seems, the student must work out what the examiner might be up to, and find an appropriate response.
9.2 The Gricean Explanation As we saw in Part I, research on the semantics–pragmatics divide generally assumes that interpretive effects such as those surveyed in section 9.1 should be captured—primarily or entirely—through the action of general pragmatic principles. This is particularly true of Grice and his successors, who characteristically explain such effects as CIs via the mechanism of flouting. Recall from Chapter 2 that flouting the maxims is a communicative strategy that exploits an apparent violation of the CP to give an utterance a creative reinterpretation. Theoretically, the problem is to characterize what happens when the speaker uses an utterance whose ordinary contribution would be defective, if general background knowledge associates that contribution with a sensible alternative. The theory of flouting predicts that interlocutors can use the speaker’s utterance in these circumstances to coordinate on the alternative, associated interpretation.
FOR INDEXING
the s cop e of i n ter pre t ive re a s on i n g 159 Grice and his successors are often explicit that the tropes of section 9.1 are cases of flouting the maxims.
9.2.1 Metaphor As we saw in Chapter 2, Grice (1975) and Searle (1979) explain metaphor as a case of flouting. With (5) Richard is a lion, for example, we reinterpret lion in terms of its associations of fierceness, leadership, and pride, because these associations make a better contribution for (5) than its literal meaning.
9.2.2 Sarcasm Grice uses flouting to explain sarcastic uses of utterances such as (173). (173) X is a fine friend. (Grice 1989: 34) He imagines (173) uttered after it emerges that X has betrayed the speaker, so its literal meaning obviously fails to conform to the Maxim of Quality. At the same time, a prominent associated idea, the contrary proposition that X is a terrible friend, springs naturally to mind. The logic of flouting lets the speaker use (173) to implicate that contrary proposition.
9.2.3 Irony Grice’s example of literary irony considers the reviewer who writes (174), as a way of suggesting some “hideous defect” in a singer’s performance. (174) Miss X produced a series of sounds that corresponded closely with the score of “Home sweet home”. (Grice 1989: 37 Ex b) For Grice, to utter (174) flouts the Maxim of Manner. The reviewer’s circumlocution is too cumbersome to be used sincerely. But (174) inevitably prompts us to imagine alternative, more prosaic descriptions that the reviewer has ostentatiously failed to use. The logic of flouting turns (174) into a denial of these alternatives: all must be inadequate to Miss X’s indescribably bad recital.
9.2.4 Hinting A prototypical hint is Grice’s recommendation letter: (175) Dear Sir, Mr X’s command of English is excellent, and his attendance at tutorials has been regular. Yours, etc. (Grice 1989: 33) Uttering (175) brings an apparent violation of the Maxim of Quantity. We expect writers to offer favorable remarks on their students’ accomplishments.
FOR INDEXING
160
the scop e of inter pre t ive re a s on i n g
In confronting (175), we immediately observe that the writer has not said anything of the kind. But suppose that the writer has chosen (175) to get across that she has nothing good to contribute. On this interpretation, according to Grice, (175) says exactly what it needs to in order for the audience to conclude for themselves that Mr X is weak. As we saw in Chapter 4, not all approaches to pragmatic reasoning afford explanations in terms of flouting. On these theories, indirection, such as that found in (173) or (174), does require us to say something special. However, as we see more in Chapter 11, general principles remain front-and-center in indirection, even without flouting. These theorists suppose that speakers are distancing themselves from the utterance they use, and so propose an echoic (Sperber and Wilson 1981) or layered (Clark 1996) interpretive process. General principles apply once to understand the utterance as if used sincerely, and again to understand what the speaker is doing by distancing herself from it. As with the explanations that we considered for cases (25–27) in Parts I and II, these accounts are designed, in part, to pinpoint the role of general principles in recognizing a speaker’s intention in using an utterance. As with the explanations in Parts I and II, it does seem reasonable to assume that disambiguation in such cases as or (168–172) or (173–175) can proceed through processes of intention recognition, in part based on expectations of cooperation; we’ll return to this suggestion in Chapter 13. However, appeals to general interpretive principles go further. They purport to show not just that the specific communicative effect that the speaker aims is something that makes sense for them to do, but also that the effect is calculable: To derive it, the audience need only know the meaning of what the speaker uttered, general background knowledge, and overarching pragmatic principles. This is what we will contest.
9.3 Overview of the Argument The main goal of Part III is to establish the heterogeneity of interpretive reasoning. Ultimately, the data in section 9.1 challenge theorists to explain how speakers can come to share insights and make points that are so far removed from the ordinary meanings of what they say. Pragmatic theories propose to explain the effects by undifferentiated mechanisms. The idea is that interlocutors coordinate by associations, general background knowledge, and their shared focus on what matters to them. We will argue that such explanations elide important, qualitative distinctions among the different practices interlocutors have for engaging with imagery.
FOR INDEXING
the s cop e of i n ter pre t ive re a s on i n g 161 Our argument thus depends on a methodology for drawing distinctions among interpretive effects. To do this, we will explore different practices in rich detail, attempting to be as precise as possible in our characterization of how they work. As a result, we will be able to highlight specific interpretive contrasts that distinguish them. We will argue that the different practices have their own bases, boundaries, and dynamics that can only be explained by appealing to separate inferential mechanisms. Our argument culminates in the presentation of a kind of minimal pair. We give examples where a single utterance is understood in its context to introduce an ambiguity between qualitatively different interpretations, derived according to different interpretive practices by qualitatively different kinds of imaginative engagement. Thus, it must be the practices themselves—rather than the contextual background or general interpretive principles—that enable speakers to use these utterances to get their points across. Both the arguments we make in Part III and the arguments we make in Part II undermine pragmatic explanations. However, what we find in Part III is very different from what we find in Part II. The interpretive effects of Part III are always non-detachable, in Grice’s sense. In Part III, we will find no linguistic signals that conventionally distinguish different interpretations. Indeed, the interpretive practices we will consider, such as metaphorical thinking, do not seem limited to utterances or to linguistic meaning. Moreover, we will suggest that the practices are not easily characterized in terms of propositional content. When we engage our imaginations in these different ways, it seems essential that there are dynamic processes at work. How one thinks seems as important a part of understanding as what one thinks. This too contrasts with the contributions to conversation we consider in Part II. Thus, while we accept that interlocutors coordinate on the imaginative practices by which they engage with utterance content, we think it’s important to separate these practices, and the insights they prompt, from meaning and grammar as revealed in Part II. Our work here in characterizing the practices thus lays the groundwork for our broader thinking about the place of meaning in language use.
FOR INDEXING
10 Perspective Taking: Metaphor
We begin our survey with the case of metaphor, where, we believe, the dialectic is particularly clear. We have already seen to different attempts to explain metaphorical interpretation as the exercise of general interpretive principles. In Chapter 2, we explored Grice’s (1975) and Searle’s (1979) idea that metaphors exploit natural associations and interlocutors’ cooperative reinterpretations of one another’s utterances. In Chapter 4, we briefly sketched Sperber and Wilson’s (1986) proposal that metaphors illustrate interlocutors’ general ability to improvise creative, contextually appropriate meanings that result in useful inferences. The main point of this chapter is to pinpoint the weaknesses of these views. We will argue, particularly following Black (1955); Davidson (1978), and Camp (2003, 2008, 2009), that metaphorical interpretation involves a distinctive process of perspective taking.1 Metaphor invites us to organize our thinking about something through an analogical correspondence with something it is not. Any explanation of the import of metaphorical utterances will need to appeal to this distinctive perspective-taking operation. General interpretive principles will not explain metaphor on their own. We frame this argument in response to the development of Part II and in anticipation of the development of Part IV. We begin in section 10.1 by introducing metaphor and explaining why we think the insights of metaphor, and the distinctive import of words used metaphorically, frequently go beyond the conventional rules of language. We gesture towards a positive account of metaphor in section 10.2, and explain why the mechanisms involved cannot be general ones. Finally, in section 10.3, we make our case for the open-ended, dynamic, and improvisational nature of metaphorical thinking. In Part IV, we 1 There is a vast and contentious literature on metaphor across a range of fields, and much of the most insightful work bridges the perspectives of different disciplines. We refer readers to Hills (2012) as a starting point for a systematic survey of the topic. Instead, as with the linguistics of Part II, we offer an opinionated introduction to the work that has most influenced us.
FOR INDEXING
p ers p e c t ive t a k i n g : m e t a ph or 163 will suggest that these considerations can underwrite a theoretically informed judgment that the creative, imaginative endeavor of metaphor escapes not only the conventions of language but the very kinds of meanings that those conventions establish.
10.1 The Creativity of Metaphor So far, we have seen only simple examples of metaphor: We used (5) in Chapter 2 to introduce Searle’s approach to metaphor; we used (77) in Chapter 4 to introduce Sperber and Wilson’s. (5)
Richard is a lion.
(77)
Robert is a bulldozer.
These are relatively formulaic examples. Perhaps, one might argue, they just involve idiosyncratic but familiar meanings for lion and bulldozer. That would make them dead metaphors, a special kind of conventional polysemy. Dead metaphors seem to be processed like other cases of semantic ambiguity. See Glucksberg (2001) or Bowdle and Gentner (2005), who illustrate this lexicalization with the case of gold mine in (176). (176) An encyclopedia is a gold mine. To get clear on the creativity of metaphor we need better examples. We like to cite comedian Matt Groening’s (177) in this connection. (177) Love is a snowmobile racing across the tundra; and then suddenly it flips over, pinning you underneath. At night, the ice weasels come. (177) is an instance of a general metaphorical strategy noted by Lakoff and Johnson (1980: 44): Love is a journey. The idea is to see the temporal progression of a relationship, particularly its affective progression, in terms of the analogous progression of a trip from a starting point to a destination. Lakoff and Johnson (1980) illustrate the pervasiveness of the metaphor through a number of stock examples, such as those in (178–180). (178) We are at a crossroads. (179) We can’t turn back now. (180) We’re spinning our wheels. Groening’s example shows that the metaphor is productive, allowing speakers and their audiences to explore these analogies in open-ended ways. Groening seems to invite us to imagine love affairs as starting with a sense of adventure
FOR INDEXING
164
p ersp e c t ive takin g : m e t a ph or
and excitement but eventually going horribly wrong and leaving us trapped with gnawing feelings of torture and pain. It really does seem like that what’s doing the work here is an ability to understand Groening’s sentence literally, and then, to make an analogy between the experience of being in love and the kind of history he describes. Henceforth in this chapter we’ll exclude the conventional, dead metaphors from consideration, and reserve the term metaphor for these active, creative cases. If we were to approach such creative metaphors in ways parallel to the interpretive effects we explored in Part II, we would hope to uncover new rules of language to describe the new interpretations. One option would be to suggest that the words of (177) have alternative, grammatically sanctioned meanings that compose to derive Groening’s message about love. Another option would be to appeal to context. On this view, with metaphorical imagery, speakers creatively evoke not the ordinary literal meanings of words but relevant, analogical ones, which compose together to determine the metaphorical meaning of an utterance (Stern 2000). Our assumption about the creativity of metaphors, however, is that there are no linguistic cues and no linguistic reflexes for the insights that the speaker is offering; the audience must develop a fresh understanding through analogical thinking. That analogy proceeds not constituent-by-constituent, but holistically, across extended discourses like (177). We will go on to argue this in more detail. If that’s right, then if we can decompose the insights of a metaphor into specific meanings for its component expressions, we have all but showed that the metaphor is dead. We need another approach to explain how speakers produce metaphors and hearers understand them.
10.2 Metaphorical Perspective Taking Like many cognitive scientists, we believe that metaphor is not only creative but distinctive. Thus we see metaphor as falling outside the purview of the pragmatic theories of Part I and the linguistic rules we considered in Part II. The key to our understanding of metaphor is an idea stated particularly clearly by Davidson (1978). Davidson suggests that the point of metaphor involves a special kind of perspective taking. Moreover, we shouldn’t think of this perspective taking as being propositional in nature or even as carrying information. To explain his view Davidson (1978: 218) draws an analogy with Wittgenstein’s famously ambiguous image of a duck–rabbit. To experienced viewers, the image has no surprises; they are familiar with its shapes, patterns, and textures and have experienced its realignments into a duck figure and a rabbit figure countless times. Indeed, this information is always salient as they inspect the
FOR INDEXING
p ers p e c t ive t a k i n g : m e t a ph or 165 figure. Nevertheless, they can still see the figure as a duck or see it as a rabbit. This makes a difference to their experience, but it is not a difference in content. It’s natural to describe the difference as one that shapes and organizes their thinking, and it’s natural to describe metaphor in the same way. But what does it actually mean to think of one thing as another? What does it mean to shape or organize thinking? You might be tempted, with Aristotle, to describe it in terms of noticing similarities. For Aristotle, metaphors are just similes in disguise: (181) The simile is also a metaphor. The difference is but slight. When the poet says of Achilles that he “Leapt on the foe as a lion” this is a simile; when he says of him “the lion leapt,” it is a metaphor. [Similes] are to be employed just a metaphors are employed, since they are really the same thing except for the difference mentioned. (Roberts 1924: iii. 4) In other words, if Aristotle is right, when you say Love is a snowmobile, all you mean is that love is like a snowmobile. But that’s clearly not enough, as Black (1955) already observed. The concept of similarity is unconstrained and variable (Goodman 1972; Tversky 1977). Love is like a snowmobile in countless ways, and we can appreciate any of those similarities without thinking of love as a snowmobile. Black’s way to get more precise was his interaction theory of metaphor. Metaphorical interpretation, for Black, starts with a tension and incompatibility between what we know about the source domain (lions, let’s say, or snowmobiles) and what we know about the target domain (Richard, love). In characterizing what we know, Black emphasizes the ramifying texture and social relevance of the facts that we must consider; he called these facts systems of associated commonplaces. For example, it’s salient about lions that they are ferocious, formidable opponents, and brave. This distinctive pattern of ideas is an important part of what we have to reconcile with Richard in order to make sense of the metaphor (5) Richard is a lion. Black proposed that this tension is resolved by a basic cognitive operation; this operation constitutes the distinctive perspective taking of metaphor. We take those aspects of the source that fit the target—a set of properties, let’s say, that are prototypically associated with the source domain, and then, we filter out what we’re thinking about the target by emphasizing these aspects, which may be things we know about the target or not, but which prompt us to reassess and reinterpret what we know about the target in a way that highlights consequences of these important features of the source domain.
FOR INDEXING
166
p ersp e c t ive takin g : m e t a ph or
Black’s view is a good start, but it falls short in some key respects. First, Black ties the imaginative engagement of metaphor to the recognition of a deviant literal meaning. But we can interpret utterances metaphorically even when their literal meaning is not anomalous, as we find in (178–180) for example. Black is right that the metaphorical thinking required to interpret these cases does involve overcoming an incompatibility, namely the incompatibility between relationships of love and physical journeys. But this incompatibility is not always visible in the literal meaning of the metaphorical utterance. A related problem with Black’s view is that it ties metaphorical interpretation too closely to the words the speaker uses. Lakoff and Johnson (1980) emphasize how the same underlying metaphors surface in systematically varying ways across a wide range of examples. Just as talk of snowmobiles in (177) evokes the general metaphor of love as a journey, so (5) evokes a general metaphor describing people in terms of the analogous behavioral propensities of animal species. Anyway, metaphors are not confined to single words, or even to isolated sentences in texts. We have to abstract away from language. In Donne’s (168), repeated below, the metaphor extends across clauses: (168)
No man is an island entire of itself; every man is a piece of the continent, a part of the main.
Interpreting (168) does not just involve triggering new interpretations for island or continent. That misses the breadth and interconnections of Donne’s imagery. The underlying metaphor sees people as places, and social connections as geographical connections. The whole discourse evokes this one figurative possibility. Metaphorical thinking is as extensive in scope as metaphorical discourse. Metaphorical thinking often requires us to find many analogical correspondences simultaneously. It maps collections of objects, properties, and relationships in a source domain systematically onto corresponding ones in a target domain (compare Lakoff and Johnson 1980; Gentner and Bowdle 2008). Love isn’t just a snowmobile, in (177), for example. Understanding the metaphor involves further correspondences: Having the thoughts and affect of love is analogous to being in the snowmobile, the development of love is analogous to movement in the snowmobile, and so forth. This complexity has systematic consequences for our attention and inference. For example, we can transfer inferences that we draw readily in the source domain to establish complex new relationships in the target domain. And we may find new salience in complex interrelationships in the target domain that may be familiar but insufficiently appreciated (Camp 2009). In addition to enriching Black’s understanding of the structure of metaphor, we can be more precise than Black about how metaphorical thinking works.
FOR INDEXING
p ers p e c t ive t a k i n g : m e t a ph or 167 Camp (2009: 112 Ex 7) uses the example of (182) to help clarify the creative engagement that we bring to reconciling the different domains in this interpretive process. (182) I am Anna Karenina. To appreciate a speaker’s point in using this particular metaphor, the addressee must again place important properties and relationships in a source domain in correspondence with those of a target domain. This imaginative effort again allows the similarities and analogies across domains to deepen and enrich an understanding of the target. In the metaphorical interpretation of (182), for example, Camp observes, the task is “matching important features of Anna’s to features of mine that are similar to hers in some significant respect” (Camp 2009: 112). She might entertain the way she, like Karenina, faces conflicts between social forces of constraint and conformity, and her own independent sense of her identity and purposes. As Camp points out, this is a very different process from literally imagining oneself to be Anna Karenina. In understanding the metaphor, Camp does not, and indeed should not, imagine herself to be actually inhabiting nineteenth-century Russia, or to be actually having the specific conflicts between her love for her son and her passion for her lover, features which drive Karenina’s story. At best, she is supposed to find correspondences, in her actual life, to these aspects of Karenina’s condition. As Camp puts it, the “direction of fit” of the metaphor in (182) maps Karenina onto its speaker, and not the other way around. Though metaphorical thinking requires the audience to recognize how one thing is like another, metaphorical thinking is not just problem solving—it shapes the listener’s imaginative engagement.2 Camp (2008) explains this by juxtaposing metaphors with telling details. Telling details, like metaphors, are elements the speaker provides in setting a scene that are intended to be richly evocative and affectively laden. The speaker’s point in providing these details is also, in part, to invite the audience to explore their consequences. In so doing, the audience can deepen their appreciation of how they are meant to react to the scene, who they are supposed to empathize with, and what sort of feelings should be prompted. The Chinese poet Li Po is someone who used telling details influentially in poetry. Camp (2008: 19) cites the line (183) in Ezra Pound’s translation: (183) I watch the moon through the clear autumn. 2 Others speak of “picturing,” “imagistic,” or “framing” effects–(Moran 1989; Guttenplan 2005; Stern 2008)—and “brainstorming”, and “imaginative play” (Reimer 2008; Hills 2008).
FOR INDEXING
168
p ersp e c t ive takin g : m e t a ph or
In the context of this poem, we know this line is uttered by a woman whose lover has failed to meet an appointed assignation. The moonlight conveys the stillness and loneliness of knowing she has been rejected, while the clear autumn night underscores the deliberateness of her lover’s absence. Camp emphasizes that these descriptions of telling details align in many respects with the predictions of Relevance Theory, and thus show how Relevance Theory can capture part of the interpretation of a metaphor. We naturally expect that we can color our interpretation of a text by presenting telling details that prompt relevant inferences. When we consider metaphor, however, we find something missing from such general interpretive reasoning, and the evocative conclusions it might lead us to. What’s missing is the distinctive fit involved in metaphorical perspective taking, and seen in cases like (182). To drive the point home, let’s return to Lakoff and Johnson’s (178), We’re at a crossroads. We can take this as a telling detail, observed along a physical journey. To do so frames the present moment as a particularly decisive and important one, reminds us that the choices we make now may affect both where we’re going and how we get there, and invites us to contrast our journey so far with what happens afterwards. But something more—something special—has to happen for us to invoke the metaphor love is a journey with (178), and thereby leverage such thinking to make sense of a relationship. There is no way to explain metaphorical interpretation, Camp (2008) and others suggest, without a special theory of this kind of metaphorical thinking. The considerations we have advanced so far distinguish metaphor from other strategies for utterance interpretation, and establish the role for a certain kind of perspective taking in understanding metaphors. In presenting our case, we have implicitly organized the development around the information carried by metaphor. But Black already observes that that metaphorical thinking gives special insights that can’t be paraphrased. When you say (5) Richard is a lion, you’re not just saying he is a courageous or a formidable opponent. You are also thinking of him as a lion, and this imagery, this perspective you are taking on Richard, is as important to the effect and the point of the metaphor as any propositional information it may convey. Even if interlocutors are familiar with Richard, lions, and their attributes, (5) can be a worthwhile and effective utterance simply by prompting the hearer to organize his thinking about Richard—in the dynamic sense we have been exploring, which is not necessarily tied to content—as the metaphor directs. This thread, which runs from Black, through Davidson, to Camp, emphasizes that metaphors can shape our responses and guide our thinking, because of the particular kind of perspective taking
FOR INDEXING
p ers p e c t ive t a k i n g : m e t a ph or 169 they involve, without conveying information in the usual sense. This, more than anything else, is why we think metaphors must be explained in a distinctive way.
10.3 Metaphor and the Dynamics of Discourse Our goals so far in this chapter have been very modest. We have argued simply that metaphor is creative rather than conventional, and that it involves a distinctive kind of thinking, not just the general interpretive principles interlocutors always use. One response to such arguments would simply be to fold the results of metaphorical thinking into meaning as we ordinarily understand it. This is the strategy defended, in different ways, by Stern (2000) and Camp (2008). On our view, however, this strategy fails to come to grips not only with the power of metaphor itself but also with the explanatory resources of traditional pragmatic theories. We have already considered two general pragmatic accounts of metaphor, in Chapter 2 and Chapter 4. Let’s revisit the essence of these accounts with reference to (177). The Gricean proposal is that metaphors are cases of conversational implicature, where a speaker’s meaning diverges from the literal meaning of an utterance.3 This account holds that Groening uses (177) to implicate that love is a snowmobile*, where snowmobile* names the relevant property, associated with snowmobiles, that captures the import of the metaphor.4 The audience recognizes this interpretation by applying the Cooperative Principle. The CP reminds us that the speaker must have aimed for a contribution that was appropriate to the conversation. The literal meaning is just one source of evidence about what that contribution is supposed to be. We need to consider all the information we have. One salient fact is that the literal meaning instructs us to imagine an exciting course of events, an unexpected turn, and its prolonged and painful repercussions. This imaginative associate would be a cooperative characterization of the experience of love. Since the speaker has made us think of this association, and has done so using knowledge he would expect to be available to us, that must be the contribution he intended to make. Thus we derive an implicated meaning for (177). 3 See Searle (1979) as well as Martinich (1984), Bach and Harnish (1979) and Reimer (2008) for related accounts. 4 We should note we are implicitly exploiting developments within Relevance Theory in framing this account; in particular, the practice of referring to a metaphorical meaning with starred expressions is due to Carston (2002). She calls these ad hoc concepts; i.e., something not specifically given by any of the conventions of language, but something you can recover in context, because of the constraints imposed by a communicative intention; or imposed by the pull of Relevance. More on this later.
FOR INDEXING
170
p ersp e c t ive takin g : m e t a ph or
For Relevance Theory, meanwhile, language users always adjust the interpretation of words in context to draw useful inferences. Refining the meaning of snowmobile to snowmobile* is just an extreme example of this phenomenon. More precisely, as we think through (177), we know that not everything involved in this snowmobile ride could possibly be true of love; some important things, however, might also true of love; and these importantly related ideas are where we focus our reasoning efforts: We derive inferences, of course, about the fun that new events can bring, about the constant threat of calamity, about the hidden threat of additional dangers. The concept snowmobile* captures the relevant snowmobile-inferences that we are prompted to by this line of thinking and the speaker’s utterance. In short, according to both the Gricean and the Relevance Theorist, even though metaphors are creative, there’s nothing special about interpreting them. The audience disambiguates the metaphorical speaker’s meaning through the same general psychological principles that are involved in all cases of communication. The audience discovers the import of the metaphor through an undifferentiated process of inference or association that is not interestingly distinguished from any other appeal to background knowledge in understanding. Thus, it’s a mistake to try to theorize about metaphors in more fine-grained ways, by systematizing and abstracting pervasive metaphors, as in Lakoff and Johnson (1980), or by explicating the mechanisms of analogical thinking, as in Bowdle and Gentner (2005) or Camp (2008, 2009). There are no real distinctions to support such projects. For a theorist who defends the distinctiveness of metaphor, we think the only effective response to such challenges is to follow Davidson. Meaning as we understand it—public content that underwrites interlocutors’ joint inquiry into how things are—doesn’t seem to capture what metaphor is doing most of the time. The insights of metaphor crucially involve perspective taking, seeing things in certain ways, which is a fundamentally different enterprise. Davidson’s view constitutes a radical and yet plausible critique of the very idea of a pragmatic approach to metaphor, as traditionally understood. In short, we know that metaphor is not just general interpretive reasoning—we know that it is a distinctive and creative kind of interpretive engagement—because the insights it brings are often not propositional in nature and so cannot be characterized in the inferential framework of all traditional pragmatic theories. In fact, we think there’s surprisingly little evidence to suggest that metaphor ever contributes information—if we understand information in the semantic sense that we have emphasized in Parts I and II of publicly accessible content that supports inquiry. Rather, metaphors seem to help us to attune ourselves to
FOR INDEXING
p ers p e c t ive t a k i n g : m e t a ph or 171 distinctions in the world, before we delineate and name them precisely, before we recruit them into our shared practices of inquiry. We briefly recount one of our favorite cases from Lepore and Stone (2010), to give the flavor of the dialectic. Metaphorical instructions are frequent and effective, but also deeply puzzling. Matthew remembers (184) from learning to swim the butterfly growing up. (184) With each stroke, let a wave flow along the length of your body, from your head down your back and through your legs. (Lepore and Stone 2010: 176 Ex 9) The butterfly is very difficult. Your hips naturally sink in the water, but counterintuitively, if you let them drop, it becomes harder to swing your arms out of the water in your recovery, and harder to keep your mouth high enough to breathe. The imagery of (184) can help. You have to relax and flow through it: After you raise your head to breathe, you have to let your head dive under so your hips float up. This wave is obviously just an imaginary one, but it’s a useful metaphor for the orchestration of movements that you have to perform in order to swim this stroke successfully. How is it, then, that a coach can succeed in telling you what to do with a metaphor if this metaphor doesn’t contribute a meaning by which it informs you of what it is you are supposed to do? We take the language at face value. Suppose, for example, we say that the instruction just asks the swimmer to move and to use the image of a wave to guide the orchestration of her movement, that is, to approach her action and experience itself through metaphorical thinking.5 This seems to offer just as good an explanation of how we follow a metaphorical instruction as would some kind of implicated meaning. Metaphorical thinking is all we can do when we try to follow the instruction. But metaphorical thinking is just what we need to succeed. We find Davidson’s objections to metaphorical meaning compelling. Davidson’s view also comports with our philosophical prejudices. Pragmatic theories of metaphorical meaning appeal to too simple an understanding of language and of metaphor. At the same time, we think they elide too much of the subtlety and sophistication of speakers’ interactions with one another. People can 5 Emerson (1836: 34–5) reminds us that we regularly bring metaphorical thinking directly to our experience, not just to language. “Who looks upon a river in a meditative hour,” he asks, “and is not reminded of the flux of all things? Throw a stone into the stream, and the circles that propagate themselves are the beautiful type of all influence . . . It is easily seen that there is nothing lucky or capricious in these analogies, but that they are constant, and pervade nature . . . man is an analogist, and studies relations in all objects.”
FOR INDEXING
172
p ersp e c t ive takin g : m e t a ph or
make meaning with language, but we think it would be strange and remarkable if they did not use language in other ways, for other purposes. We will leave this, for now, as a tentative suggestion, and move on to other cases of figurative language. We need a broader perspective, both empirical and theoretical, to fully assess the role of imaginative engagement in interpretation, meaning, and understanding.
FOR INDEXING
11 Presenting Utterances: Sarcasm, Irony, and Humor
Our investigation of metaphor showcased the imaginative engagement involved. We believe that other kinds of figurative and evocative language likewise elicit distinctive kinds of imaginative engagement. In this chapter, we extend this idea to a range of further cases that are generally thought of as non-serious, echoic, or pretenseful uses of utterances. In particular, section 11.1 explores sarcasm, which we understand as an invitation to explore the contrast between the apparent contributions of an utterance and the way things actually are. Section 11.2, meanwhile, explores literary irony, which we understand as an invitation to engage with an imagined speaker. Finally, section 11.3 covers humor, which seems to involve, among other things, an invitation for the hearer to appreciate a surprising pair of contrasting perspectives. The goal of describing such utterances as CIs, we shall see, had deep consequences for the way Grice and his followers formulated their ideas. In fact, many of the proponents of echoic and pretenseful account of non-serious language, including Sperber and Wilson (1981) and Clark and Gerrig (1984) are also known for their defenses of general pragmatic principles. Thus, they expect the bulk of the interpretive effects associated with sarcasm, irony, and humor to follow from those general interpretive principles. The key new assumption they make is just that speakers can exhibit utterances for effect, in addition to sincerely using them. By contrast, we see the interpretive effects of sarcasm, irony, and humor, in line with metaphor, as involving heterogeneous kinds of imaginative engagement. These are slippery cases; the phenomena and the terminology are both difficult to pin down. We will be as clear as possible about the particular interpretive effects that we are after. We think the differences we find among those effects make it worthwhile to describe each case on its own terms. Section 11.4
FOR INDEXING
174
present ing utter a n ce s
develops this argument further in the context of the broader program of the book.
11.1 Sarcasm We delimit sarcasm following Grice (1975), Clark and Gerrig (1984), and Camp (2012). Sarcasm, intuitively, is when a speaker uses an utterance that means one thing to get across to her audience its opposite.1 This has always been a paradigm case of Grice’s (1975) notion of flouting the CP. In accounting for sarcastic examples such as (173) X is a fine friend, Grice gestures at the insight that has been developed in all subsequent approaches. The speaker doesn’t use (173) directly to make his contribution; instead, the speaker makes as if to say something, distancing himself from the utterance as he presents it. In short, we should think of such cases, like quotations, as offering words that are not endorsed as the speaker’s own, not reflecting the speaker’s beliefs, and not embodying the speaker’s perspective. The suggestion of pragmatic theories is that an appropriate characterization of the possibly echoic (Sperber and Wilson 1981) and possibly pretenseful (Clark and Gerrig 1984) relation of the speaker to the utterance will trigger the interpretations we find for utterances such as (173) by general interpretive principles. We must test such claims against the subtlety and complexity of sarcasm. To start, sarcasm cannot be captured just by taking the negation or contrary of what the speaker literally says (Kumon-Nakamura, Glucksberg and Brown 1995; Camp 2012). Sarcasm can target any of the contributions of an utterance. If a speaker says (185) to cut down a show-off with all the answers: (185)
You sure know a lot. (Kumon-Nakamura et al. 1995: 4)
she inverts the implicature that her interlocutor’s informed interventions have been productive. If a speaker says (186) after the addressee has just allowed a door to close in the speaker’s face: (186)
Thanks for holding the door. (Kumon-Nakamura et al. 1995: 4)
the reversal infects the speech act that the speaker is performing, turning (sincerely) thanking into sarcastically “thanking”—somehow indicting the hearer’s thoughtlessness in a moment of inconsiderate behavior. (Like KumonNakamura et al. 1995, we are at a loss to provide an alternative formulation of 1 There are other understandings of sarcasm: e.g., disparaging and non-serious comments are often described as sarcastic even when the utterance is used exclusively for its literal meaning. At the same time, this trope is sometimes referred to as irony: as will become clear, by irony we have something different in mind.
FOR INDEXING
pre s en t i n g ut ter a n ce s 175 this inverted speech act that could accomplish the same effect through its literal or conventional interpretation.) Moreover, we must appreciate the creativity of sarcasm. In particular, sarcastic speakers often develop a foil to how things are through extended patterns of description. Thus, consider the utterance of (187) as part of a sarcastic mock-news segment2 by comedian Aasif Mandvi on the Western corporations that supply toxic asbestos to the construction industry in India. (187) This CBC documentary. . . clearly shows Indian workers taking every precaution. (188) [Footage of two men moving big tufts of asbestos as casually as if it were hay.] (189) That is a real bandana. (190) [Cuts to close up of one of the men, whose mouth is covered by a bandana.] This is a sarcastic discourse—indeed, a sarcastic multimedia presentation. Mandvi describes the workers as though they are working carefully with expensive equipment to ensure they are not injured by asbestos, where it is transparently clear the exact opposite is so. The second sentence of (187), “That is a real bandana,” shows that Mandvi develops one overarching sarcastic image across his extended description. The coherence of Mandvi’s discourse demands that the bandana be understood as an example of the precautions the workers are taking, while the adjective real presupposes contrasting “imitation” bandanas that workers without these sarcastically imagined “safety precautions” might be forced to use. The description of the bandana thus amplifies the incongruity of Mandvi’s remarks, by specifically highlighting the divergence between what the workers are wearing and the protective suits and other gear that we are accustomed to seeing in cases of work with hazardous materials. In responding to discourses such as (187), we are inclined to agree with Camp (2012), contra Clark and Gerrig (1984), that sarcasm has a point. The utterance is not merely an affective or expressive act that displays Mandvi’s negative feelings for the conditions of Indian asbestos workers, or distances him from a description of their working conditions as good. Rather we think that Mandvi’s point is, in part, that the workers (and their employers) are not taking every possible practical precaution for themselves: they are indeed taking few, if any, precautions. 2 From The Daily Show, Thurs May 12, 2011:
thu-may-12-2011/ored-to-death>.
