Direct Belief: An Essay on the Semantics, Pragmatics, and Metaphysics of Belief 9781614510826, 9781614510901

Citation preview

Direct Belief

Mouton Series in Pragmatics 23

Editor Istvan Kecskes

Editorial Board Reinhard Blutner Universiteit van Amsterdam The Netherlands N. J. Enfield Max-Planck-Institute for Psycholinguistics Nijmegen The Netherlands

Ferenc Kiefer Hungarian Academy of Sciences Budapest Hungary Lluı´s Payrato´ University of Barcelona Spain

Raymond W. Gibbs University of California Santa Cruz USA

Franc¸ois Recanati Institut Jean-Nicod Paris France

Laurence R. Horn Yale University USA

John Searle University of California Berkeley USA

Boaz Keysar University of Chicago USA

De Gruyter Mouton

Deirdre Wilson University College London Great Britain

Direct Belief An Essay on the Semantics, Pragmatics, and Metaphysics of Belief by

Jonathan Berg

De Gruyter Mouton

ISBN 978-1-61451-090-1 e-ISBN 978-1-61451-082-6 ISSN 1864-6409 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at http://dnb.dnb.de. ” 2012 Walter de Gruyter GmbH & Co. KG, Berlin/Boston Printing: Hubert & Co. GmbH & Co. KG, Göttingen ⬁ Printed on acid-free paper Printed in Germany www.degruyter.com

Preface

This project goes back to when I was a graduate student, trying to get a better understanding of the de re/de dicto distinction. After poring over Tyler Burge's "Belief De Re" and the related literature, I eventually came to the conclusion that talk of the interchangeability of coreferential names suffered from a kind of modal ambiguity: whether or not one could substitute one name for another depended on whether the substitution was to preserve the truth of the sentence, or the appropriateness of the utterance; and so I argued in a paper I wrote for Keith Donnellan in 1979, called "Two Kinds of Substitutivity." This led me to take more seriously an intriguing remark that David Kaplan had once tossed out in a seminar, to the effect that direct reference theorists might explain away substitution failure by appeal to Grice's theory of conversational implicature. Just how such an explanation might go was the subject of a graduate students' colloquium talk I gave at UCLA the following spring and has continued to occupy me ever since. Having since presented bits and pieces of my work on this subject in a number of papers and talks, I eventually came to realize that nailing things down persuasively requires a comprehensive presentation of the whole extended argument. That is the purpose of this book. The first chapter is mainly a critical review of theories of the semantics of belief ascriptions that I reject. After setting up the problem as a question of how to accommodate both de re and de dicto uses of belief ascriptions, I argue against treating belief ascriptions as ambiguous, indexical, or semantically indeterminate. As an alternative I introduce the Theory of Direct Belief, which treats having a belief about an individual as an unmediated relation between the believer and the individual the belief is about. The positive argument comes in the second chapter, where I show how the import of de dicto interpretations of belief ascriptions can be accounted for pragmatically, as conversationally implicated, rather than semantically, as belonging to what is actually said. This leads to a discussion in the third chapter of the nature of belief, and in particular of the relation between beliefs and conceptions. The Theory of Direct Belief is contrasted in these terms with the Medium View of belief, according to which beliefs are held in a medium of conceptions. Two arguments for the Medium View (and hence, against the Theory of Direct Belief), apparently stemming from the common idiom of how a person thinks what he thinks (on analogy with

vi Preface how a person says what he says), are then shown to be unsuccessful. And in the final sections I turn to what I take to be the principle motivation behind the Medium View—the "Inner Speech" Picture of Thought—which I try to show is unwarranted.

Acknowledgements

I am grateful, first of all, to Tyler Burge, Keith Donnellan, and David Kaplan, who guided my earliest work on this project, and who each contributed so profoundly to my philosophical development. Having endeavored relentlessly to save me from philosophical folly, their support and encouragement must not be construed as an endorsement of any of my views. I am also grateful for generous support which enabled me to work on this project at the following venues: the Center for Philosophy of Science, University of Pittsburgh, 1990-1; Jerry Fodor and Ernie Lepore's NEH summer seminar, Rutgers University, 1992; the Department of Philosophy, Simon Fraser University, 1995-6; the Center for Studies in Language and Information, Stanford, 1996; the Department of Philosophy and the Institute for Advanced Studies, University of Bristol, 1999; the Institute of Philosophy, School of Advanced Study, University of London, 2008-9; and above all, the Department of Philosophy and the Faculty of Humanities here at the University of Haifa. And I am especially thankful for all the helpful comments I have received along the way, from Fred Adams, Michael Antony, Kent Bach, Anne Bezuidenhut, John Biro, Jessica Brown, Steven Davis, Anthony Grayling, Mitch Green, Philip Hanson, Thomas Hofweber, David Israel, Asa Kasher, Igal Kvart, Kirk Ludwig, Diego Marconi, Adam Morton, Peter Pagin, John Perry, Bjorn Ramberg, François Recanati, Nathan Salmon, Stephen Schiffer, Barry Smith, Mark Steiner, Josef Stern, Corliss Swain, Alberto Voltolini, Ruth Weintraub, Andrew Woodfield, Takashi Yagisawa, and Ed Zalta, as well as some very helpful anonymous referees. Portions of this work derive from the following publications, whose publishers I wish to thank for their permission to make use of these works here: "The Pragmatics of Substitutivity," Linguistics and Philosophy, 11 (1988), 355-70; "On the Very Idea of Thinking in Words," Belgian J. of Linguistics, 8 (1993), 9-19; "In Defense of Direct Belief: Substitutivity, Availability, and Iterability," Lingua e Stile, 33 (1998), 461-70; "Troubles with Neo-Notionalism," Philosophia, 27 (1999), 459-82; and "Must We Know What We Say?" in Saying, Meaning, Implicating, ed. Georg Meggle and Christian Plunze, Leipzig: Leipzig University Press, 2003, pp. 64-71. Above all, I am thankful to my wife, Sharon, and my children, Noam, Nava, and Nadav, for all their love and encouragement.

ȱȱ

Contents

Preface.......................................................................................................... v Acknowledgements .................................................................................... vii Introduction .................................................................................................. 1 Chapter 1: The instability of belief ascriptions (and how not to explain it) 1. From language to thought ...................................................................7 2. Accommodating shifty intuitions......................................................10 3. The appeal to ambiguity....................................................................12 3.1. Lexical ambiguity .............................................................................12 3.2. Syntactic ambiguity ..........................................................................17 3.3. Multigrade status...............................................................................21 3.4. The persistence of shifty intuitions ...................................................24 4. The indexical view............................................................................24 4.1. Implicit modes of presentation..........................................................24 4.2. Hidden-indexical semantics ..............................................................27 4.3. Articulated indexicality.....................................................................36 5. Semantic indeterminacy....................................................................40 5.1. Incompleteness..................................................................................40 5.2. Similarity ..........................................................................................42 5.3. Hopelessness .....................................................................................46 6. Direct belief ......................................................................................48 7. Summary ...........................................................................................49 Chapter 2: The pragmatics of substitutivity 1. Truth and appropriateness.................................................................54 2. Conversational implicature ...............................................................56 3. Implicated normalcy .........................................................................57 4. Normalcy for belief ascriptions.........................................................60 5. Variations on verbatim acceptability ................................................63 6. Identity beliefs ..................................................................................70 7. Availability .......................................................................................75 8. Semantic intuitions............................................................................82 9. Iterability...........................................................................................91

x

Contents

10. 10.1. 10.2. 11.

Other pragmatic accounts of substitution failure ..............................95 Soames and what is said ...................................................................95 Thau and what is implicated .............................................................99 Summary .........................................................................................10ϯ

Chapter 3: Conceptions, belief, and “inner speech” 1. The medium view of conceptions ...................................................108 2. The behavior problem .....................................................................111 2.1. The problem ....................................................................................111 2.2. The Higher Order View of conceptions..........................................113 2.3. A solution to the problem ...............................................................114 3. Suspended belief .............................................................................115 4. The inner speech picture of thought................................................12ϯ 5. Thinking in words...........................................................................130 5.1. Silent uttering..................................................................................130 5.2. Imagining ........................................................................................134 6. Two paradigms of belief .................................................................137 7. Summary .........................................................................................140 References ................................................................................................ 143 Index......................................................................................................... 153

Introduction

What is it for someone to believe that so-and-so is such-and-such—say, for Barack Obama to believe that Osama bin Laden is dead? In one form or another, this question—going to the very heart of how words and thoughts and things in the world are all related to each other—has been the central question in the philosophy of language ever since Frege set the agenda over a century ago. The question has mainly been addressed as one of semantics, asking for the meaning (or truth conditions) of belief ascription sentences (such as ‘Barack Obama believes that Osama bin Laden is dead’). But, as has become increasingly appreciated, what an expression means and how it is used are separate, albeit related. And so, whatever we are to say about the semantics of belief ascription sentences must be distinguished from, and coordinated with, what we say about the pragmatics of utterances of those sentences. Moreover, the semantic question cannot be divorced from the substantive one—we cannot seriously consider what the word ‘believes’ means without considering what belief is. It is my aim in this essay to address the question at hand in a way that takes into account all three of these related aspects—the semantics, the pragmatics, and the metaphysics of belief. The first chapter concerns the semantics of belief and is mainly a critical review of theories I reject. Although not all the objections I raise are new, I try to lay them out in a novel and illuminating way. In the first section I review how questions about meaning quickly lead to questions about thought—in particular, how questions about the meaning of proper names lead to questions about the semantics of belief: e.g., if Lois believes Clark Kent is a reporter, and Clark Kent is Superman, does it follow that she believes that Superman is a reporter? Focusing on this as a question about the truth of the sentence S:

Lois believes Superman is a reporter

(assuming the truth of the original Superman story) allows for a convenient exhaustive partitioning of theories of the semantics of belief ascriptions: (a) those on which S is true, (b) those on which S is false, (c) those on which S is either, depending on the context, and (d) those on which S is neither.

2

Introduction

This comprehensive taxonomy proves especially useful when we take into account not only the conventional Fregean intuition that S is false, but also the Quinian observation that S seems to be true when taken in the so-called “relational sense.” Deeply impressed by the overwhelming evidence for theories of direct reference (well enough known that I do not review it), I favor the first option—Lois does indeed think that Superman is a reporter.1 More broadly, I argue for the Theory of Direct Belief, according to which having a belief about an individual is an unmediated relation between the believer and the individual the belief is about.2 Reconciling this position with the standard Fregean intuitions obviously takes some explaining, which comes mainly in the following chapters. In the first chapter I look at the other options. Disregarding the second option—the Fregean assumption that it is false that Lois thinks Superman is a reporter—due to its prima facie inability to accommodate the relational sense, I concentrate in this chapter on the only remaining possibilities, the third and fourth options.3 In preparation for this I review in Section 2 the various ways in which standard semantic theory accommodates context-dependence: a sentence may seem true in some contexts and false in others because it is ambiguous, because it is indexical, or because it is just semantically indeterminate. These three possibilities are considered in turn in the following sections. 1.

2.

3.

The main arguments in favor of theories of direct reference, from the loci classici of Keith Donnellan (1970), David Kaplan (1978, 1979, 1989), Saul Kripke (1972), and Hilary Putnam (1975), are nicely summed up by Nathan Salmon (1986: ch. 5) and Scott Soames (2002: ch. 2). Exactly what counts as “having a belief about an individual” and as “unmediated” will be discussed at length; to forestall possible misunderstandings in the meantime one might think of having a belief about an individual as its being the case that there is some individual of whom one believes that he is soand-so. Thus, where one believes, e.g., that the 44th president of the United States is clever, one has a belief about a certain individual—Barack Obama— that he is clever; and my claim is that one thereby stands in a binary relation with Barack Obama which does not essentially involve conceptions, modes of presentation, or other alleged components of a medium in which beliefs are held. See also Chapter 2, note 1, and Chapter 3, Section 1, especially note 4. I do not presume that the prima facie problems confronting the Fregean approach are more than just prima facie; I myself have suggested elsewhere how Fregeans might get around the problem of “substitution success” (1988: 358). But I find the evidence adduced in the following chapters for the first option more compelling.

Introduction 3

In Section 3 I consider the claim that belief ascriptions are ambiguous. First I show how they fail to satisfy any of the standard tests for lexical ambiguity, and then I argue against taking them as syntactically ambiguous. I also argue that Quine’s proposal (1977) to take the belief predicate as having multigrade status does not establish that belief ascriptions are ambiguous, but actually seems to assume it—thereby running afoul of the cited linguistic evidence. The section concludes with a brief indication of why no two-fold distinction at all, such as that between relational and notional senses of ‘believes’, could adequately handle the linguistic data. In Section 4 I argue against treating belief ascriptions as indexical. The analysis of belief ascriptions as containing “hidden indexicals,” referring to modes (or types of modes) of presentation, suffers from problems not only with the appeal to implicit modes of presentation, but especially with the semantics of hidden indexicals and the related idea of “unarticulated constituents” that are paradoxically part of what is said. Mark Richard’s indexical alternative (1990), taking the word ‘believes’ itself as indexical, is also seen to suffer from some hard problems, particularly with regard to ordinary intuitions about indexical reference. And finally, in Section 5 I look at how belief ascriptions might be considered semantically indeterminate, in the sense that the sentence fails to determine a particular proposition, even after all ambiguities are resolved and the referents of all singular terms are determined. One way a sentence can be semantically indeterminate in this sense is by being semantically incomplete, but such a view of belief ascriptions is seen to encounter most of the troubles facing the hidden indexical view. Another way to construe belief ascriptions as semantically indeterminate is by treating them as similarity claims, as proposed by Stephen Stich (1983), on the basis of Donald Davidson’s treatment of indirect discourse (1968); but this approach is shown to rely on questionable premises about belief ascriptions and about similarity claims. Lastly in this section I consider the possibility that belief ascriptions are just too irregular to admit of a coherent semantic theory. My main reply to this is the rest of the book. In the second chapter I spell out a pragmatic account of substitution failure—that is, I explain how the Fregean intuition that Lois does not believe that Superman is a reporter derives not from the falsity of the sentence S:

Lois believes Superman is a reporter,

4

Introduction

but from the falsity of what can be conveyed by an utterance of S. In particular, I show how in uttering S we can conversationally implicate the false proposition that Lois would accept S as true.4 The chapter opens with a brief review of the basics of Grice’s theory of conversational implicature (1989). In Section 3 I derive the Principle of Implicated Normalcy, according to which speakers generally implicate that the circumstances regarding whatever they are speaking of are not abnormal in any significant, unanticipated, unindicated way. This leads me to conclude in the next section that the utterer of a belief ascription normally implicates that his ascription is acceptable verbatim, in the sense that the person to whom the belief is ascribed would accept that very ascription, word for word, as true. Thus, in a normal utterance of S it would be conversationally implicated that the utterance is acceptable verbatim, e.g., that Lois would accept S as true— which, of course, she would not—hence the mistaken Fregean intuition that S is false. In Section 5 I explain how to generalize this approach to cases where we ascribe beliefs to individuals who clearly do not understand the words we use, and hence, for whom the ascription clearly cannot be acceptable verbatim, and in Section 6 I show how the approach applies to belief ascriptions regarding identity propositions and their negations. The balance of Chapter 2 addresses the principle objections and counterproposals to my approach. Perhaps the leading objection to pragmatic accounts of substitution failure has been the charge of counterintuitivity. In Section 7 I argue against François Recanati’s Availability Principle, which says, “In deciding whether a pragmatically determined aspect of utterance meaning is part of what is said, that is, in making a decision concerning what is said, we should always try to preserve our pre-theoretic intuitions on the matter” (1993: 248); and this leads to a more general discussion in Section 8 of the role of intuitions in semantic theory. Section 9 resolves an objection raised by Stephen Schiffer (1987a), concerning certain belief ascriptions where the ascribed belief itself contains a belief ascription. And then in the last section of the chapter I explain how my account is prefer-

4.

This is where I part ways with others who agree that substitution failure is to be accounted for pragmatically. Scott Soames (2002, 2005, 2006, 2008) and Michael Thau (2002) offer pragmatic accounts very different from my own, which I take up in Section 10, and Nathan Salmon (1986) refrains from committing himself to any particular pragmatic account at all. Earlier suggestions to account for substitution failure pragmatically have been made by J. O. Urmson (1968), Tom McKay (1981), and Jon Barwise and John Perry (1981).

Introduction 5

able to alternative pragmatic accounts of substitution failure, offered by Scott Soames (2002, 2005, 2006, 2008) and by Michael Thau (2002). In Chapter 3 I focus on belief itself, reformulating the question as one about the structure of belief. According to the Medium View, beliefs are always held in a medium of conceptions of some sort or another. Thus, Lois’s belief that Superman is a reporter must be mediated somehow by a conception she has of Superman (it being left open whether or not the conception is actually a component of the belief). This is clearly incompatible with the Theory of Direct Belief, which takes having a belief about an individual as an unmediated relation between the believer and the individual the belief is about.5 Of course, the most influential argument for the Medium View (and against the Theory of Direct Belief) has been the Fregean argument based on the premise that substitution of coreferential names does not always preserve truth—a premise shown in the second chapter to be unfounded. In the third chapter I first consider and reject two other arguments for the Medium View, based on the idiom of “thinking-of-as,” and then I argue against what seems to be the pre-theoretical motivation behind the view. Section 2 concerns the Behavior Problem: why is it that Lois behaves one way when she sees Superman in his Superman outfit and another way when she sees him in his Clark Kent outfit? According to the Medium View her beliefs about this individual are in some way mediated by, and hence differentiated by, her distinct conceptions of him—when she believes he is standing in front of her, it is either via her Superman conception of him or via her Clark Kent conception of him—“thinking of him as” one or the other. But no such explanation is available to the Direct Belief theorist. How else, then, can we explain Lois’s behavior? And if conceptions are not components of beliefs (or of vehicles by which we have beliefs), then what are they? Dealing with the second question first, I argue for the Higher Order View of conceptions, on which it is conceptions that are to be explained in terms of beliefs, rather than the other way around. And then I 5.

I believe Thau (2002) and I share the same bottom line here, but argue for it very differently, in accordance with different motivating considerations and different ways of carving up logical space. What he refers to as “pure Millianism” seems to be the Theory of Direct Belief, and the Medium View seems to be the disjunction of two views he treats separately as “Fregeanism” and “guise Millianism”—the latter being Salmon’s view, which we both reject. Unfortunately, his arguments against the Medium View are couched in a rich conceptual framework that is too elaborate for me to review here.

6

Introduction

solve the Behavior Problem by rejecting the assumption that Lois’s reactions to Superman must be accounted for entirely on the basis of her beliefs about him, rather than beliefs she has about other things—such as his cape. In Section 3 I discuss a problem originally raised by David Kaplan, of how to reconcile belief with “suspended belief”—for if names are indeed directly referential, it seems that one might continue to believe a certain proposition while at the same time having ceased to believe it. It is on the basis of such data that Salmon argues for his own version of the Medium View, according to which believing a proposition consists in being inclined to assent to it inwardly when apprehending it in a certain way. I argue instead that we can describe cases of suspended belief perfectly well without resort to talk of inward assent and modes of apprehension. Having disarmed the arguments against the Theory of Direct Belief, I turn in the final sections to a critical look at what I take to be the main source of the Medium View’s wide appeal—the underlying picture of thought as “inner speech.” In Section 4 I show how fleshing out the Inner Speech Picture of Thought quickly runs into seemingly insurmountable obstacles, since the metaphor of thought as language suffers from so many significant disanalogies, especially in cases of “nonrepresentational” belief. I suggest in Section 5 that the attractiveness of the inner speech picture of thought might be the result of a common inclination to take “thinking in words” as a paradigm of believing. But, I argue, the underlying analogy between asserting a proposition by uttering a sentence and thinking a thought by silently uttering a sentence is deeply misguided, probably resulting from confusion between “silent” utterance and imagined utterance. In the final section I consider briefly how the clash between the Inner Speech Picture of Thought and the Theory of Direct Belief might ultimately boil down to a clash between two distinct pre-theoretical paradigms of belief, differing in how closely they tie belief to verbal behavior; and I urge resisting the natural inclination to give verbal behavior more attention than it deserves.

Chapter 1 The instability of belief ascriptions (and how not to explain it)

1. From language to thought What is the meaning of a word? What, in particular, is the meaning of a name?1 Could it be simply the thing named? This possibility is presumably ruled out by the following Fregean considerations: if the meaning of a name were just the thing named, then names of the same thing would be freely interchangeable; but this is not the case when speaking of what someone thinks, since someone might think that A is so-and-so without thinking that B is so-and-so, even when A and B are the same thing.2 1.

2.

In speaking of a “name” I mean to be speaking of a proper name, and in speaking of the “meaning” of a name I mean to be speaking of its semantic content—what it contributes (in virtue of its semantic properties) to the proposition expressed by a sentence in which it occurs. Frege’s own argument, from the opening lines of “On sense and reference” (1893), is this: (1) a = a and a = b have different cognitive value. (2) If identity were a relation between objects, then a = a would not differ from a = b (provided a = b). (3) Identity is not a relation between objects. I would make the following exegetical assumptions (cf. Burge (1977: 354): (i) P and Q have different cognitive value iff it is possible to believe one without believing the other; (ii) identity is a relation between objects iff identity statements (composed of two names flanking an equal sign) express a relation between the bearers of the names flanking the equal sign; and (iii) if the meaning of a name is just the thing named, then identity statements express a relation between the bearers of the names flanking the equal sign. Thus, the following argument would at least be Fregean, if not Frege’s: (1) It is possible to believe that a is F without believing that b is F, even if a = b. (2) If the meaning of a name is just the thing named then not-(1). (3) The meaning of a name is not just the thing named.

8

The instability of belief ascriptions

Evidence of such “substitution failure” in belief ascriptions seems to abound. Suppose, for instance, that the Superman story were true—in particular, that Lois Lane, mistakenly thinking that the names ‘Superman’ and ‘Clark Kent’ refer to two different people, is inclined to answer affirmatively when asked, “Is Clark Kent a reporter?” but to answer negatively when asked, “Is Superman a reporter?” Then we would be inclined to accept (1)

Lois Lane believes Clark Kent is a reporter

but to reject (2)

Lois Lane believes Superman is a reporter,

even though the two sentences differ only in which of two names of the same individual is used.3 And so, goes the argument, the names must differ in meaning, even though they name the same individual. However, as Quine has shown (1956:187), the Fregean evidence of substitution failure is not as conclusive as it might first appear, for we can easily imagine circumstances—with no change in thestory—where (2) would be acceptable.4 For example, marveling at Superman’s skill at disguising himself, a fan might say, Look at what a master of disguise that Superman is! Why, when he puts on that suit and strolls into his office at the Daily Planet, he fools everybody— Mr. White, Jimmy Olson; even Lois Lane believes Superman is a reporter.

Questions about meaning thus lead to hard questions about belief—does Lois Lane think Superman is a reporter or does she not?5 Belief ascriptions

3. 4.

5.

The word ‘believes’ in my examples may be replaced throughout with ‘believes that’, ‘thinks’, ‘thinks that’, etc. It is important not to confuse contexts of utterance with states of affairs. The claim about (2) is that even in the same state of affairs (given by the story), and even given the same referents for the names ‘Lois Lane’ and ‘Superman’ and for the predicate ‘is a reporter’, there are conversational contexts in which an utterance of (2) would be acceptable (in addition to the more obvious contexts in which it would not). Questions about meaning are, of course, not the only route to the hard questions about belief, which arise as well from considerations about Leibniz’s law (Section 6).

From language to thought 9

are notoriously unstable, in that “Fregean” intuitions pull us one way, while “Quinian” intuitions pull us another.6 Logical space allows four positions here: (2) can be true, false, both, or neither.7 I shall argue for the first, and more generally, for the view that coreferential names are indeed interchangeable in belief ascriptions salva veritate. To do this I shall explain how the problem with (2) is that in uttering it we may conversationally implicate, without explicitly saying, something false—viz., that Lois Lane would confirm our report of her belief—which detracts not from the truth of the sentence, but only from the appropriateness of its utterance.8 And I shall argue that this position allows for the best account of belief and meaning. First, however, I shall in this chapter briefly review some of the problems facing other views.9

6.

7.

8. 9.

Though Quine discusses intuitions in both directions, I use the term ‘Quinian intuitions’ to refer (however ironically) to the ones which sometimes lead us to accept (2). I am talking here about the sentence (2), in the given state of affairs, with fixed referents for the names ‘Lois Lane’ and ‘Superman’ and for the predicate ‘is a reporter’. In terms of truth relative to a context of utterance, the four possibilities are that (2) is true relative to all contexts, that it is false relative to all contexts, that it is true relative to some contexts and false relative to others, and that it is neither true nor false relative to any context. I take “what is a said” in an utterance as what is said explicitly. Braun (1998, 2001, 2002) looks at the issue differently, as a question of how to defend theories of direct reference (holding, roughly, that the meaning of a proper name is just its bearer) against the objection that they lead to the strange result that a rational person can simultaneously believe a proposition and its negation. I agree with him that although this result obtains, it is not objectionable—a rational person can indeed simultaneously believe a proposition and its negation. Braun argues against certain pragmatic accounts of how this is possible, proposing instead an explanation in terms of “ways of believing”—a person can believe a proposition in one way and believe its negation in another way. For reasons to be explained later (partly in Section 4.1 and mainly in Chapter 3) I would not accept his proposal about ways of believing, but neither would I advocate any of the pragmatic accounts he considers. Rather, I think that the possibility of a rational person’s believing a proposition and its negation is clearly evident from all the cases where a person is unaware of such things as the coreferentiality of two names—Lois clearly believes that Clark Kent is a reporter and that Superman is not, and so with regard to this individual sometimes called ‘Clark Kent’ and sometimes called ‘Superman’, she believes the proposition that he is a reporter, as well as its

10

The instability of belief ascriptions

2. Accommodating shifty intuitions “The obvious next move,” writes Quine (1956: 188), upon having observed the divergent intuitions a belief ascription might arouse, “is to try to make the best of our dilemma by distinguishing two senses of belief”—the “relational” or “de re” sense, and the “notional” or “de dicto” sense.10 That is, a natural way to account for our conflicting Fregean and Quinian intuitions about (2)

Lois Lane believes Superman is a reporter

would be by declaring the word ‘believes’ ambiguous, rendering the sentence true on one reading, but not on the other.11 But just how obvious is the move to ambiguity?

negation, that he is not. As for the question I raise, of how to account for our shifting intuitions about the truth-values of belief ascriptions such as (2), Braun does not seem to offer an answer. 10. Against what he acknowledges is the standard interpretation, Crawford (2008) argues that the two senses of belief Quine proposes here do not correspond to the relational/notional distinction, and that neither of these distinctions corresponds to the de re/de dicto distinction (where de dicto belief is Quine’s dyadic belief and de re belief is Quine’s triadic and higher degree belief, to be discussed in section 3.3). Since I would not be able to do justice to Crawford’s argument in a brief remark, and since the exegetical point is not essential to my case, I will stick to the standard interpretation without argument. The details of my interpretation of the distinction (which could just as well be construed as nothing more than my own definitions) will be provided at the beginning of the next section. 11. An expression is ambiguous in the relevant sense if it has more than one sense (meaning). (I will not be concerned here with possible distinctions between polysemy, homonymy, etc.) Cf. Stephen Stich, for whom a term is ambiguous if (i) speakers view the term’s extension as divided into two or more distinct categories, (ii) speakers using the term typically intend to say something concerning things of (only) one of these categories, and (iii) such intentions are common or expected for each of these categories (1986: 128–30). One problem with this account is that the third clause leaves no room for obscure senses. Also, these conditions may not be sufficient for ambiguity. On the basis of this account Stich concludes that utterances containing indefinite descriptions are ambiguous, between one reading having the form of a mere ex-

Accommodating shifty intuitions 11

First it should be noted that shifty truth-values (that is, contextdependent intuitions about the truth-value of a sentence) are no proof of ambiguity. That is, a given sentence in a given state of affairs may seem true in one context of utterance and false in another without being ambiguous.12 One possibility is that the sentence is indexical. For example, She is hungry can obviously vary in truth-value, even with no change in the state of affairs. This is not because the word ‘she’ has more than one meaning, but because the meaning it has can determine different referents in different contexts. Conflicting intuitions about truth-values can also be due to implicit qualifications (or other implicit elements) that are understood only from the context. For instance, whether or not Everybody trusts Emily is taken as true will depend on what is taken as the domain of discourse; whether or not Emily is big is taken as true will depend on what is taken as the relevant comparison class.13 This is not because the words ‘everybody’ or ‘big’ are ambiguous, but because they are usually implicitly qualified. In cases such as these the sentence by itself does not determine the truth-conditions by which we interpret an utterance of it. And of course, whenever it is hard to assign truth-values, there is also the possibility that it is because there are none to assign; belief ascriptions may simply be too unruly to admit of any semantical theory. istential generalization and another reading involving an intended reference to a particular individual (1986: 137–40). 12. On problems with taking the presence of ambivalent intuitions about truthconditions as a test of ambiguity see Atlas (1989: 32–34), Cruse (1986: 82, n. 12), and Stich (1986: 133–136). 13. Even given a particular comparison class, the application of a vague predicate may also depend on the point of the comparison, which typically would also be only implicit.

12

The instability of belief ascriptions

Assuming that these are the only ways one could possibly provide a semantic account of “shifty intuitions”—in terms of ambiguity, indexicality, or indeterminacy—I shall consider each of these ways of accounting for our conflicting intuitions about (2).14 An important caveat here is that this chapter is intended not as a general survey of the literature on the semantics of belief ascriptions, but rather as a survey only of certain regions of logical space as I have carved it up. Hence, particular theories will be addressed only as examples of the population of a particular region of logical space, and many important and interesting theories and issues in the semantics of belief will not be addressed directly. Moreover, in discussing the various positions I reject, my aim will be only to highlight the most significant problems they face; I do not claim to provide any conclusive refutations.

3. The appeal to ambiguity 3.1. Lexical ambiguity The view offered by Quine as “the obvious next move”—that our shifty intuitions about (2)

Lois Lane believes Superman is a reporter

correspond to two senses of the word ‘believes’—is widely assumed without argument (or even acknowledgement). How exactly might such a move be made? The “notional” or “de dicto” sense of ‘believes’—according to which (2) would be false—can be characterized as that for which the following three conditions obtain: (a) reverse disquotation is valid—that is, if 14. This taxonomy is meant to leave room for everybody—including “semantic contextualists,” who would treat belief ascriptions as either indexical or semantically indeterminate (in the relevant sense, whereby sentences are semantically determinate only if they determine truth-conditions, at least relative to disambiguation and assignment of referents to indexicals; some might say ‘underdeterminate’ rather than ‘indeterminate’). I leave it open for now whether what is literally and explicitly said in an utterance might depend on the context for more than disambiguation and the assignment of referents to indexicals. I argue elsewhere (2002) for a version of “semantic minimalism,” some of which will surface in the ensuing discussion.

The appeal to ambiguity 13

Lois Lane believes Superman is a reporter, then (under appropriate circumstances) she would sincerely say (or at least would not sincerely deny) ‘Superman is a reporter’; (b) exportation is not valid—that is, from the premise that Lois Lane believes Superman is a reporter, it does not follow that there is someone who Lois Lane believes is a reporter; and (c) substitution (of coreferential names) is not valid—that is, from the premises that Lois Lane believes Clark Kent is a reporter and that Clark Kent is Superman, it does not follow that Lois Lane believes Superman is a reporter. And then the “relational” or “de re” sense of ‘believes’—according to which (2) would be true—can be characterized as that for which the aforementioned conditions do not obtain, i.e., that according to which reverse disquotation is not valid, but exportation and substitution are valid.15 What would show that the word ‘believes’ is ambiguous in this way? One sign of ambiguity in an expression is the existence of distinct “disambiguating” expressions capturing each of the putative senses.16 The strongest case would be where the language itself contains distinct synonyms (not synonymous with each other) for an ambiguous word in each of its senses—as the word ‘board’, for example, is synonymous with ‘plank’ in one sense and with ‘committee’ in another sense.17 Alternatively, disam15. Of course, ‘believes’ might be ambiguous in other, irrelevant ways, which may be ignored. Note that as David Kaplan points out, Quine construes the relational sense as admitting both substitution and also quantifying in, and the notional sense as admitting neither (1986: 237). Although Kaplan shows how these two criteria can come apart—in that there could be a context excluding substitution while allowing quantifying in—we may assume for the present discussion that the two senses divide as envisaged by Quine, substitution and quantifying in going together, since the issue here has been set up in terms of substitution, not quantifying in (and of course, Kaplan does not claim that substitutivity can hold where quantifying in is excluded). In any case, my characterization of the notional sense is more robust than I really need; for present purposes it would suffice to make the distinction on the basis of subsitutivity alone. 16. Despite common parlance, there are no decisive tests of ambiguity; but what are referred to as “tests” may at least serve as signs, or prima facie indications. 17. As Cruse (1986: 54) explains, the relation between the disambiguating expressions and the disambiguated expression in its two senses need not be synonymy. It could be another paradigmatic semantic relation, such as antonymy, as the word ‘light’ is disambiguated by ‘heavy’ and ‘dark’; or a paronymic relation, as the word ‘race’ is disambiguated by ‘racial’ and ‘racing’.

14

The instability of belief ascriptions

biguating expressions might be found in another language—as two senses of ‘know’ are distinguished by the French words ‘sais’ and ‘connais’. Or, instead of actually checking languages we know, we might simply ask ourselves whether the discovery of such expressions in another language would be expected or surprising; presumably we would expect the different senses of an ambiguous word, such as ‘bank’, to have different translations in an arbitrary foreign language, and we would be surprised if it did not (Kripke 1977: 19). But the word ‘believes’ does not seem to pass any of these tests for ambiguity: English does not contain synonyms of ‘believes’ which, when substituted in (2)

Lois Lane believes Superman is a reporter

make the sentence alternately true and false; nor, as far as I know, do any other languages; and I, for one, would be surprised to discover such a language.18 Another sign of ambiguity derives from the fact that, as D. A. Cruse explains, an ambiguous expression “resists as it were, the simultaneous activation of more than one of its senses” (1986: 62).19 One manifestation of such resistance would be the impossibility of “crossed interpretations” for sentences reduced by conjunction or ellipsis.20 For instance, the sentence 18. Appealing to the expressions ‘believes de dicto’ and ‘believes de re’ would clearly just beg the question, much as stipulating meanings for the terms ‘believes1’ and ‘believes2’; disambiguating expressions need to be found, not stipulated. Also, although there seem to be disambiguating expressions for the sentence taken as a whole—viz., formulations of the relational and notional readings—this does not establish the lexical ambiguity of any of the sentence’s components. 19. Cf. Stich’s test (1986: 129) in terms of intentions, according to which it would be odd to use an ambiguous term without having just one of the senses in mind; e.g., A: We are going to have some trouble with the banks. B: Do you mean the river banks or the financial institutions? A: Well, I hadn’t thought about that. Some of each, I suppose. Such tests are not to be applied, of course, in cases of deliberate double entendre, as in puns and poetry. It is not clear, though, whether this proviso can be formulated noncircularly. 20. See Atlas (1989: 44–65). Atlas also cites Ryle (1949) and Chomsky (1972: 33).

The appeal to ambiguity 15

Jack and Jill began to cry could be interpreted as meaning that they both began to weep or that they both began to shout, but not as meaning that one began to weep and the other began to shout. On the other hand, a sentence such as Jack and Jill each have one child is understood as being true not only if they each have a son or each have a daughter, but also if one has a son and the other has a daughter— presumably showing that the word ‘child’ is not ambiguous but merely “general” with regard to gender. Once again, ‘believes’ does not seem to pass the test, since crossed interpretations do not seem ruled out for Lois Lane and I both believe that Superman is a reporter. I might say, for example, Lois Lane and I both believe that Superman is a reporter, though only I, but not she, would put it in just those words.21 Although these tests of ambiguity are not conclusive, they certainly do not show that ‘believes’ is ambiguous; if anything, they suggest just the oppo-

21. It may be that the reduced-conjunction test cannot be used decisively where one of the putative senses entails the other, since the crossed interpretation would entail the uncrossed interpretation in the more general sense, thereby making it hard to tell them apart. For instance, although ‘dog’ in one sense applies only to male dogs, it is not clear that the crossed interpretation is impossible for ‘Dick and Jane each have a dog’. See Cruse (1986: 63–65), and Atlas (1989: 48–53). The same problem seems to arise for Stich’s multipleintention test (1986: 129) and for a quantification test proposed by Wiggins (1971a: 28). The possibility of a crossed interpretation for ‘believes’ might be more apparent in a case of VP-ellipsis: I believe Superman is a reporter, and so does Lois (though she wouldn’t put it in those words).

16

The instability of belief ascriptions

site.22 Moreover, if ‘believes’ has two senses in the way imagined, then so too—for similar reasons—must all the other verbs and verb-phrases expressing propositional attitudes (‘knows’, ‘doubts’, etc.). This undermines the appeal to ambiguity in three ways. First of all, it is hard to see exactly where to locate the putative ambiguity in compound verb phrases such as ‘is of the opinion that’ or ‘holds the view that’. Surely the verbs ‘is’ and ‘holds’ are not ambiguous in the relevant way, nor do I see how the nouns ‘opinion’ and ‘view’ could be. Secondly, we tend not to ascribe ambiguity en masse; rather, when a particular duality of use ranges over a wide class of expressions, we usually prefer a unitary explanation. We would not posit multiple senses to account for the type/token distinction, for instance, or the distinction between soldiers and toy soldiers.23 Furthermore, the positing of senses must be subject to the same constraints as the positing of other theoretical entities. For one thing, such theoretical posits should be kept to a minimum; hence, Grice’s Modified Occam’s Razor (MOR): “Senses are not to be multiplied beyond necessity” (1989: 47).24 This becomes especially significant as the parallel between ‘believes’ and other propositional attitude expressions serves to up the ontological ante. Moreover, although criteria of ambiguity such as those discussed may be supported by general theoretical considerations, in the end the positing of a theoretical entity must be evaluated as part of a complete

22. Indeed, although some of these tests are thought of as giving necessary conditions for ambiguity, all they really do is have us replace intuitions about ambiguity with related intuitions. 23. There may be exceptions to this, such as the dual purpose vocabulary of explanation and justification. It is at least arguable that words such as ‘because’ are ambiguous between explanatory and justificatory senses. But recognizing such exceptions does not require abandoning the general theoretical preference to avoid them. Cf. Zalta (1988: 166–167, 172–173), who attempts to accommodate both Fregean and Quinian intuitions about belief ascriptions by treating all terms occurring in attitude contexts as ambiguous between “objectual” content (ordinary denotation) and “cognitive” content. 24. Charges of ambiguity must not be made lightly—lest every dispute, philosophical and otherwise, dissolve into mere divergence of senses. As Kripke observes, “It is very much the lazy man’s approach in philosophy to posit ambiguities when in trouble” (1977: 19). Although Occam’s Razor is customarily taken as applying to postulated entities, MOR could be understood as applying to semantic rules rather than semantic contents.

The appeal to ambiguity 17

theory.25 The great expense of positing a whole range of extra senses would seem especially extravagant, if not prohibitive, insofar as the data can be explained—as I shall argue it can—by means of some independently established theoretical apparatus.26

3.2. Syntactic ambiguity To make the appeal to ambiguity without multiplying senses (of individual words), one might try to construe belief ascriptions as ambiguous syntactically, rather than lexically.27 This was first proposed by Russell, whose treatment of belief ascriptions such as (2)

Lois Lane believes Superman is a reporter

consists of two stages. First, common proper names are treated as abbreviations of definite descriptions. If, for instance, ‘Superman’ abbreviates ‘the superhero from Krypton’, then (2) becomes (3)

Lois Lane believes the superhero from Krypton is a reporter.

(We may ignore here the analogous treatment of the name ‘Lois Lane’.) Then, sentences containing definite descriptions are analyzed as existential generalizations. In sentences such as (3), where a definite description occurs in an embedded sentence, there are thus two analyses: (4a) Lois Lane believes that one and only one entity is a superhero from Krypton, and that that entity is a reporter, and

25. As Wiggins (1971a: 25; 1971b: 48, 50) and Alston (1971: 36–37) both observe, senses are posited as part of a broader semantic theory. 26. Such considerations may have been behind Quine’s quick dismissal of the lexical ambiguity proposal (1956: 188). And they are explicitly behind Kripke’s preference (1977: 18) for a “unitary” account (positing no semantic ambiguity) of Keith Donnellan’s referential/attributive phenomenon (1966). 27. I am speaking of syntactic ambiguity broadly, so as to include what some might prefer to think of as nonsyntactic, logical ambiguity.

18

The instability of belief ascriptions

(4b) One and only one entity is a superhero from Krypton, and Lois Lane believes that that entity is a reporter; which might be formalized, respectively, as (5a) B(L, x[Sx & (y)(Sy ĺ y = x) & Rx]) and (5b) x[Sx & (y)(Sy ĺ y = x) & B(L, Rx)].28 So construed, belief ascriptions such as (2) suffer from an ambiguity of scope: the existential quantifier—introduced by the definite description abbreviated by a name in the embedded sentence—may occur in the scope of the belief operator, or the other way around. Rather than explore Russell’s theory at length, I wish only to point out that each stage of his analysis is highly controversial. It is far from evident that common proper names abbreviate definite descriptions; nor is it at all clear that any sentence containing a definite description is false if the description is not uniquely satisfied.29 Is there a less controversial way to treat belief ascriptions as syntactically ambiguous? Perhaps a better way to get a scope ambiguity in (2) would be by exporting the embedded name, assuming that a sentence with a proper name—say, of the form ‘A is F’—is equivalent to an existential generalization, to the effect that there exists something which is identical to A and which is F. That is, on the basis of the simple, unobjectionable logical truth FA ļ x(x=A & Fx), one might try adopting a general principle of exportation to the effect that for any formula ij, term Į, and variable ȕ, 28 Belief of a proposition is represented here by a binary operator on ordered pairs consisting of a term and a formula. I emphasize that the sentences might be formalized this way, since there are, of course, many other ways the formalization might go; and I offer these and the formalizations below only as rough heuristic devices, on which not too much weight should be put. 29 See Donnellan (1966, 1970), Kripke (1972), Strawson (1950), and related works. However, for opposing views cf. Bach (1987: chs. 5–8) and Matushansky (2008).

The appeal to ambiguity 19

EX

ij(...Į...) ļ ȕ(ȕ=Į & ij(Į/ȕ)).30

Again, since the name ‘Superman’ occurs in (2) in an embedded sentence, the existential quantifier—this time introduced by exportation in accordance with the proposed principle—may occur in the scope of the belief operator, or the other way around. That is, (2) could be rendered as either (6a) Lois Lane believes that there is someone who is Superman and is a reporter, or (6b) There is someone who is Superman and who Lois Lane believes is a reporter; which might be formalized, respectively, as (7a) B(L, x[x=S & Rx]) and (7b) x(x=S & B[L, Rx]). But however reasonable it may be to take (2) as ambiguous in this way, such a move cannot solve the problem of shifty intuitions, since the two proposed readings are equivalent. In order to allow the second, “relational,” reading, the belief operator—which, barring lexical ambiguity, must be the same in both readings—must itself be relational, so both readings are equally committed to the existence of Superman and equally uncommitted to any facts about Lois’s use of the name ‘Superman’.31

30. This general principle will not be as unobjectionable, of course, as the logical truth which inspired it; but I shall suppose for the sake of argument that the principle is acceptable. 31. Cf. Burge (1977: 345), who claims to express the de re/de dicto distinction in terms of logical form, but nevertheless uses two different operators (or predicates), one for de dicto belief and one for de re belief—thereby falling back on lexical ambiguity. On the reasonable assumption that relational belief is closed under EX, the equivalence of (7a) and (7b) can be shown as follows:

20

The instability of belief ascriptions

Indeed, for any similarly motivated analysis of (2) in terms of an ambiguity between the scope of a quantifier and that of the belief operator, when the latter takes narrow scope it must be relational (to permit quantifying in); so barring a charge of lexical ambiguity (which would raise the problems discussed in the previous section), the belief operator must remain relational when it takes wide scope; so the only straightforward way to get the notional reading would be by analyzing the name in terms of some predicate.32

(7a) (i) (ii) (iii) (7b)

B(L, x[x=S & Rx]) y[y=S & B(L, x[x=y & Rx])] Ry ļ x[x=y & Rx] y[y=S & B(L, Ry)] x(x=S & B[L, Rx])

(7a), EX EX (i), (ii), closure of belief under EX (iii), alphabetic variance.

(Since the belief operator is relational, we may assume that it allows EX.) And in light of the hypothesis of EX, the result should come as no surprise. 32. (a) Note that the standard Montagovian analysis (e.g, Montague (1974: Chs. 4, 6) and Dowty et al (1981: Ch. 6)), rendering (2) as ambiguous between B(l,^[R(s)]) and Ȝx[B(l,^[R(x)])](s), fails to capture the intended distinction between relational and notional senses, since Montague’s belief predicate allows substitution of co-intensional names (and even simply co-referential names, on the assumption that names are rigid designators). It may still be possible, of course, to construct a grammatical theory on which (2) is derivable from two deep structures differing only in quantifier scopes and corresponding to the relational and notional readings; but that could establish the alleged ambiguity of (2) only if the grammatical theory had sufficient independent motivation. I am grateful to Per-Erik Malmnas for bringing this to my attention. (b) Even if names give way to predicates, there may still be a question about substitutivity for coextensional predicates. Bach argues that the scope ambiguity between (5a) and (5b) does not account for a distinction between relational and notional readings of (5a) with regard to the predicate ‘S’ (1987: 209). (c) For an entirely different way of appealing to a scope distinction, involving reinterpretation of the quantifiers as ranging over possibilia, see Hintikka (1962, 1969).

The appeal to ambiguity 21

Of course, nothing I have said so far shows that belief ascriptions such as (2) cannot be syntactically ambiguous between notional and relational interpretations.33 Rather, I have argued only that what is possibly the most popular way of seeing a scope ambiguity in (2) cannot readily be patched up to avoid well-known serious problems. That is, without appeal to Russell’s theories of definite descriptions and of common proper names, it does not seem that the sort of scope ambiguity Russell finds in (2) can yield the relational and notional interpretations. There is, however, a more conclusive (and simpler) argument against taking (2) as syntactically ambiguous. Although Modified Occam’s Razor may not be relevant here (since positing syntactic ambiguity involves positing neither extra entities nor extra rules), claims of syntactic ambiguity do seem subject to the simultaneous activation test. For example, Thelma and Louise like racing cars does not seem to allow the crossed interpretation according to which Thelma likes to race cars and Louise likes cars that are (for) racing. If belief ascriptions fail this test—as suggested in the previous section—then they would seem to be unambiguous not only lexically, but also syntactically.

3.3. Multigrade status Rather than pursue the “obvious” move to ambiguity, Quine immediately turns to “a more suggestive treatment” (1956: 188), according to which a belief ascription such as (2)

Lois Lane believes Superman is a reporter

may express either a dyadic relation between a believer and a proposition, as in (8a) B(Lois, the proposition that Superman is a reporter),

33. Nor have I argued against positing a scope ambiguity in belief ascriptions where the embedded clauses contain definite descriptions rather than proper names.

22

The instability of belief ascriptions

or a triadic relation among a believer, an object, and an attribute, as in (8b) B(Lois, Superman, the attribute of being a reporter). Quine stresses that the belief relation of (8b) is to be viewed not as dyadic belief between Lois and the proposition that Superman has the attribute of being a reporter, but rather as “an irreducibly triadic relation among the three things” (1956: 189)—Lois, the attribute of being a reporter, and Superman. The relational sense of (2) is thus distinguished from the notional sense, in that Superman is one of the relata in (8b), but not in (8a). Thus, ‘believes’ would be ambiguous, between the dyadic relation and the irreducibly triadic relation. Indeed, it would be ambiguous between infinitely many relations, since for every n>1, Quine posits an n-ary relation, holding among an intension of degree n-2 and n-1 individuals (the believer and the n-2 objects of which the intension is believed)—in order to allow for embedded clauses with indefinitely many “referential positions” (where a term serves simply to refer to its object, and hence, would legitimately be subject to substitution and exportation). But taking belief ascriptions as (multiply) ambiguous in this way just brings us back to the objections raised above in the discussion of lexical ambiguity (Section 3.1). Quine subsequently explains, however, that ‘believes’ is to be formalized by a single predicate, having “multigrade” status—figuring as an nplace predicate for each n>1. He further explains that “the adoption of a multigrade predicate involves no logical anomaly or any infinite lexicon. It can be viewed as a one-place predicate whose arguments are sequences” (1977: 114). For example, the two multigrade formalizations of (2) corresponding to (8a) and (8b), respectively, would be as a predicate applied to a sequence of a believer and a proposition (9a) B(), and as the very same predicate applied to a sequence of a believer, an object, and an attribute, (9b) B(). Instead of the distinct dyadic and triadic predicates of (8a) and (8b), both (9a) and (9b) use just the same single monadic predicate. Thus, the multigrade treatment seems to avoid positing two (or more) senses of ‘believes’,

The appeal to ambiguity 23

since it construes belief ascriptions in terms of a single multigrade predicate. But how a natural language expression is (to be) formalized must be distinguished from whether or not it is (taken as) ambiguous. In particular, formalizing a natural language expression by a single predicate does not preclude taking the expression to be ambiguous, since we can always concoct a single formal predicate for as many nonsynonymous natural language predicates—or distinct senses of a single predicate—as we like. For example, we can define a formal one-place predicate that is true both of pencils and also of ordered pairs in which the first member is a parent of the second; or a single predicate that is true of banks in both senses.34 This can be contrasted with relatively natural multigrade treatments of natural language predicates such as ‘agreed’, ‘swarmed’, or ‘assembled’, which seem to be univocally of indefinitely many places. The question of ambiguity is thus orthogonal to the question of formalization. Indeed, Quine’s proposal is best understood as being about how to formalize (2) on the assumption that it is ambiguous. Whether (2) is treated as having notional and relational senses formalized by (9a) and (9b), or by (8a) and (8b), either way it is being construed as ambiguous. So although the multigrade treatment may successfully address Quine’s formal concerns—of how to avoid the logical anomaly of predicates of variable arity and an infinite lexicon (as well as the imagined anomaly of quantifying in)—it does not save his proposal from the objections to treating ‘believes’ as ambiguous.35 34. For any predicates F1, ..., Fn, we can define a single binary predicate F such that F(k,s) iff s is a sequence satisfying Fk. 35. (a) Quine notes that his multigrade treatment was anticipated in a footnote by David Kaplan (1969: 239, n. 3), where Kaplan describes a procedure he subsequently refers to as “the logician’s trick” of “re-ambiguation” (1986: 237– 238). As Kaplan warns, “Re-ambiguation is a notational unification of what is conceptually disparate … we should not let delight in the handiwork blind us to the underlying question of conceptual coherence.” (b) It should be clear that Quine’s efforts are directed not towards the semantics of natural language, but towards the construction of a suitably rigorous language for scientific discourse—a language in which he eventually concludes that belief de re has no place: “we can perhaps see our way to writing off the de re cases of propositional attitude for purposes of scientific language. … The indexicals and necessity and possibility are convenient in daily discourse, to the point of virtual indispensability; and we can say the same of propositional attitudes de re … while admitting none of these idioms to absolute or scientific discourse” (1995: 357–358). See Section 5.3.

24

The instability of belief ascriptions

3.4. The persistence of shifty intuitions Perhaps the strongest reason against trying to accommodate shifty intuitions about belief ascriptions by an appeal to ambiguity is that it just won’t be enough. Suppose belief ascriptions are indeed ambiguous between notional and relational readings. Suppose further (adapting Kripke’s Paderewski example) that having read some sonnets written by Clark Kent, Lois Lane has come to believe that there are two very different individuals named ‘Clark Kent’, one an exciting poet, the other an unexciting reporter. Then shifty intuitions persist, even just with regard to the notional sense. For the ascription Lois Lane believes Clark Kent is exciting —taken notionally—will seem acceptable in a discussion of Lois’s literary tastes, but not in a discussion of her interaction with fellow workers. So even if belief ascriptions were ambiguous between notional and relational readings, that would not account for all our shifty intuitions about them.36

4. The indexical view 4.1. Implicit modes of presentation Rather than account for the instability of belief ascriptions by treating them as ambiguous, one might suppose that the reason they seem to vary in truthvalue from context to context is that they are indexical; that is, belief ascriptions might depend on context just as the sentence (10) She is hungry

36. Cf. Mark Crimmins (1992: 167–168) who seems to make much the same point, though without explicitly concluding anything about ambiguity. Of course, taking the name ‘Clark Kent’ as lexically ambiguous, referring to something like various Fregean senses corresponding to Lois’s different ways of talking about him, might arguably explain the various notional interpretations, but not the relational interpretation. An adequate theory must account for the varying notional interpretations in addition to the relational interpretation.

The indexical view 25

does. In the case of (10), of course, it is easy to locate the indexicality in the single word ‘she’, referring to different individuals in different contexts; whereas none of the words in (2)

Lois Lane believes Superman is a reporter

seems to vary in this way across the contexts in question.37 How, then, could the instability of (2) be construed as a case of indexicality? According to the “hidden-indexical” view as formulated by Stephen Schiffer (1977, 1992), which for present purposes does not differ significantly from the theory expounded by Mark Crimmins (1992) and by Crimmins and John Perry (1989), the logical form of (2) can be represented as (m)[Tm & B(Lois, , m)], where T is “an implicitly referred to and contextually determined type of mode of presentation” (1992: 503), and where B is the triadic relation holding between an agent, a proposition, and a mode of presentation whenever the first believes the second under the third.38 So (2) might be paraphrased as

37. We may assume that in the contexts in question the bearers of the names and the time of utterance all remain constant. 38. Crimmins and Perry distinguish two ways of using a belief ascription, each with its own logical form: a specific mode of presentation may be “provided,” in which case that mode of presentation is what is implicitly referred to; or the mode of presentation may be merely “constrained,” in which case what is implicitly referred to is that constraint on modes of presentation—like the type of mode of presentation figuring in Schiffer’s analysis. Arguably Crimmins and Perry thus diverge from Schiffer in taking belief ascriptions as ambiguous (raising the objections discussed in Section 3). Schiffer avoids positing a second logical form by noting that the implicitly mentioned type of mode of presentation may be so narrow as to include just one mode of presentation. However, it is not clear that Schiffer’s account thereby renders belief ascriptions any more univocal than does Quine’s multigrade treatment. Crimmins and Perry also diverge from Schiffer in taking the implicit modes of presentation (which they call “notions” and “ideas”) as “concrete cognitive particulars,” but this difference does not matter for present purposes. Cf. Graeme Forbes (1996), who construes belief ascriptions as sometimes containing a

26

The instability of belief ascriptions

Lois Lane believes the proposition that Superman is a reporter under some mode of presentation of type T. And presumably the type of mode of presentation referred to in a context where (2) seems true would differ from the type referred to in a context where (2) seems false. (For the sake of clarity I shall henceforth speak simply in terms of “modes of presentation” rather than “types of modes of presentation.”) Schiffer rightly observes that one challenge facing the hidden-indexical theorist is to explain what “modes of presentation” are (1992: 510-512; 1990). This would be especially significant if, as Schiffer maintains, the notion of a mode of presentation were strictly a technical notion.39 For surely, in invoking a technical notion, a theorist incurs a special burden of explanation, not shared by theorists who do not invoke that notion. Arguably, however, the notion of a mode of presentation is an ordinary notion, manifesting itself in ordinary talk of “conceptions” or “ways of thinking” of something (analogous to talk of seeing something from different “views”). It would be natural to say, for instance, that ancient astronomers thought of Venus in two different ways, or that Lois Lane has two different conceptions of the individual we call ‘Superman’ (as I shall suppose in Chapter 3). To the extent that such talk is an important part of how we explain things, it will require no less attention than our practice of belief ascription—especially if, as seems likely, the two are connected. So the hidden-indexical theorist may be no more obliged to provide an account of this notion than anyone else is. This is not to say, however, that the hidden-indexicalist’s appeal to modes of presentation comes with no strings attached. For one thing, if beliefs and other propositional attitudes are explained in terms of modes of presentation, then on pain of circularity, modes of presentation cannot be explained in terms of any of those propositional attitudes (as Schiffer seems to do by defining modes of presentation in terms of Frege’s constraint, which is formulated in terms of belief and disbelief, as observed in the pre“suppressed” indexical referring “logophorically” to expressions in the complement clause. 39. Schiffer takes modes of presentation to be contextually defined by “Frege’s constraint,” according to which a rational individual can simultaneously believe and disbelieve the same proposition only by believing it under distinct modes of presentation without recognizing that they are modes of presentation of the same thing (1992: 502–503).

The indexical view 27

vious note). So, for example, what it is for Lois to have two conceptions of Superman could not be explained by the hidden-indexicalist noncircularly in terms of any of her beliefs about Superman.40 Secondly, quite apart from explaining what a mode of presentation is, the hidden-indexicalist must explain what it is to believe a proposition under a given mode of presentation. However ordinary the notion of a mode of presentation may be, the notion of believing under a mode of presentation is not nearly as ordinary, if at all. People seldom, if ever, speak of such a thing, and there seems to be no natural way of doing so.41

4.2. Hidden-indexical semantics Apart from whatever problems are raised by the appeal to modes of presentation, the hidden-indexical view faces tough problems concerning the semantics of hidden indexicals. To account for the fact that a belief ascription 40. One might argue that modes of presentation and propositional attitudes are on equal footing, to be explicated together; but this is to concede that neither can be explained noncircularly in terms of the other. 41. (a) It should be noted that the obscurity of the notion of believing a proposition under a given mode of presentation is a problem not just for the hiddenindexicalist, but for anyone at all who treats belief ascriptions by appeal to this ternary notion. And it is a problem for them not only becuase they have to explain the notion, but also because they have to justify ascribing beliefs involving it to ordinary speakers. Thus, David Braun has argued persuasively that such a notion could not be part of what speakers conversationally implicate in uttering belief ascriptions, since it is so doubtful that speakers entertain or intend to convey such things altogether (1998: 567–568). (Modes of presentation will be discussed more fully in Chapter 3.) (b) Another objection Schiffer raises against the hidden-indexical view is that whereas it treats the relevant mode of presentation in a belief ascription as one of the relata of a triadic belief relation, an explicit specification of the mode of presentation would take the form of an adverbial qualifier, e.g., ‘under mode of presentation M’ (1992: 518–519). However, as is widely known, a sentence such as ‘Col. Mustard did it in the library with the candlestick’ can easily be understood as expressing a triadic relation between a person, a place, and a weapon—despite the fact that the second and third relata are indicated by expressions occurring in adverbial qualifiers. For a different reply to this objection of Schiffer’s see Peter Ludlow (1995) and the ensuing discussion in Schiffer (1996) and Ludlow (1996).

28

The instability of belief ascriptions

such as (2) contains no expression referring to a mode of presentation, the hidden-indexicalist likens belief reports to weather reports; it is argued that just as a speaker who says ‘It’s raining’ implicitly refers to a place (typically but not always the place of utterance), so, too, the utterer of a belief ascription refers implicitly to a mode of presentation—which Perry (1986) calls an “unarticulated constituent” of the proposition expressed. One problem with this explanation is that the unarticulated constituents of weather reports (as well as the other examples of unarticulated constituents given by Perry and by Crimmins) are at least articulable; someone who says ‘It’s raining’ can easily specify the place that he did not explicitly mention. But the utterer of a belief ascription would be hard put to articulate the particular mode of presentation to which he presumably referred.42 In support of the claim that unarticulated constituents need not be articulable, Perry imagines an isolated, climatically uniform region he calls “Zland,” whose inhabitants have no idea of any place else and, hence, no name (or other expression) for their own land (1986: 211). So, when it rains anywhere in Z-land, it rains everywhere in Z-land; and so, when the Zlanders say ‘It is raining’ they do not say where—and indeed, could not. (Z-landese is apparently partially homophonic with English.) Assuming Zland is an unarticulated constituent of the proposition expressed by a Zlander who says ‘It is raining’, one may conclude that unarticulated constituents can indeed be unarticulable. Nevertheless, Perry does concede that since Z-lander semanticists themselves share the Z-landers’ view of the world, they would with good reason treat ‘raining’ as expressing a monadic predicate of times, rather than a dyadic predicate of times and places. Now, given that the semantics of a natural language is determined by the linguistic practices and intentions of the speakers of that language, it would seem that we, too, in giving the semantics of Z-landese—not English—should treat ‘raining’ as monadic, for the very same reasons that the Z-lander semanticist does; and in that case Z-land would not be an unarticulated constituent of the proposition 42. There may be some cases in which an allegedly unarticulated constituent is not immediately articulable by the speaker, such as when a young child says what time it is without being able to specify the relevant time zone. But even in such cases the speaker would be able to articulate the unarticulated constituent if suitably tutored; and the unarticulated constituent is at least articulable by others—neither of which conditions seems to obtain in the case of unarticulated modes of presentation. Cf. Schiffer’s “meaning-intention problem,” (1992: 512).

The indexical view 29

expressed by a Z-lander’s utterance of ‘It is raining’. Even though the truth-conditions of Z-landese weather reports in some sense involve climatic zones—’It is raining’ is true in Z-landese at time t iff it is raining in Z-land at time t—it does not follow that climatic zones are constituents of the propositions expressed by those reports. Although the Z-landers’ weather reports could be construed as ascribing a dyadic predicate to a time and a place—Raining(t, Z-land)—they could also be construed as ascribing a monadic predicate to a time—Raining-in-Z-land(t)—which more accurately reflects the Z-landers’ semantic intentions and beliefs.43 In any case, regardless of what we may say of the Z-landers’ weather reports, when we give the semantics of belief in our own language, we do not have the special vantage point that we enjoy when looking at the Z-landers “from the outside,” with our richer understanding of climatic regions. Rather, we are in the position of the Z-lander semanticists, who rightfully would recognize no reason to posit constituents they could not articulate.44 A second problem with hidden-indexical semantics is that it seems to miss the distinction between what a speaker means and what the speaker actually says—more broadly, between pragmatic content and semantic content.45 For if unarticulated constituents are indeed constituents of the proposition expressed in the utterance of a sentence, then there would be no 43. I am assuming that a semantic theory of a natural language is supposed to capture the communicative intentions of the speakers of that language, reflecting their way of looking at the world. 44. It may be telling that to provide an example of unarticulable unarticulated constituents Perry must resort to bizarre imaginary worlds. 45. The distinction may be hard to formulate precisely, but it is well established by examples. It is clear how in uttering sentences such as ‘Matilda wasn’t drunk this morning’ or ‘Laverne is a pneumatic drill’ one can mean something quite different from what the sentences literally mean. (See Chapter 2, Section 1.) Following Grice (1989), I use the expression ‘what is said’ to refer to what is said explicitly and literally. (I diverge from Grice in those places where he restricts what is said to what the speaker means.) In terms of Robert Stainton’s helpful discussion of the terminology, I use ‘what is said’ to refer to “the merely locutionary content” that results from “disambiguated expression meaning plus reference assignment to slots” (2006: 225). This is to be contrasted with broader uses of ‘what is said’, to refer to what one is taken to mean or assert (Soames (2002: ch. 3), Herman Cappelen and Ernie Lepore (2005: 4)), or to include “explicatures” (Robyn Carston (2002)) or the results of pragmatic processes other than dismabiguation and assignment of referents to indexicals (François Recanati (2004)). See Chapter 2, Section 1, note 3.

30

The instability of belief ascriptions

difference in semantic content between sentences differing only in whether or not a particular constituent is articulated. Crimmins offers various kinds of examples of unarticulated constituents, and this problem is possibly clearest in his examples of implicit restriction (1989: 18-19). Crimmins holds that in a typical utterance of the sentence (11) Cars start when the key is turned in the ignition the proposition expressed is that in normal or relevant circumstances cars start when the key is turned in the ignition. However, although this may well be what the speaker meant, it must be distinguished from what the speaker actually said; the semantic content of (11) must be distinguished from that of (11') In normal or relevant circumstances cars start when the key is turned in the ignition, even though the latter may express what the speaker means in a typical utterance of (11).46 The same goes for implicit relata: if the proposition expressed in an utterance of ‘It’s raining’ included a location as one of its constituents, then the sentence would not differ in semantic content from an expanded sentence in which the relevant location were given explicitly, e.g., ‘It’s raining here’. However, although the expanded sentence may well be used to express what the speaker means in uttering the former sentence, the two sentences cannot plausibly be construed as having the same semantic content.47 Why would anyone think in the first place that the unmentioned location in a typical utterance of ‘It’s raining’ must be part of what is said, rather than just what is meant? One reason might be that rain is relative to location, in this sense: (a) it cannot be raining unless it is raining in some location, and (b) it could be raining in one location at the same time that it is not raining in another (Perry (1986: 206)). That is, the specification of different locations may yield non-equivalent expansions of the sentence, such 46. I am assuming that the actual world has a fixed domain of discourse, with regard to which nonmodal sentences are to be interpreted (and likewise for other possible worlds and modal sentences pertaining to them). 47. Also, as pointed out to me by Hilla Pollak, if unarticulated constituents are indeed part of what is said, then articulating them explicitly—as in (11') and the sentence ‘It’s raining here’— should sound redundant; but it does not.

The indexical view 31

as ‘It’s raining in Palo Alto’ and ‘It’s raining in Murdock’. Thus, a general test for hidden indexicals might be formulated as follows: Relativity Test of Hidden Indexicality A sentence S contains a hidden indexical referring to an element of (type) T iff (a) S is true only if there is some x  T such that S is true with respect to x; and (b) for some x, y  T, S is true with respect to x but not with respect to y.48 However, the sentence ‘Rachel is eating’ passes the Relativity Test just as well as ‘It’s raining’—Rachel cannot be eating unless she is eating something, and she could be eating an orange without at the same time eating an apple. It does not follow, however, that what Rachel is presumed to be eating is part of what is said in uttering the sentence ‘Rachel is eating’.49 In the case of adverbial qualifiers—which is what the hiddenindexicalist claims is implicit in most belief ascription sentences—the construal of what is unmentioned as part of what is said is especially strange. One cannot walk, for instance, without walking at a particular speed, in a particular direction, etc.; and one could walk at one speed in one direction without walking at another speed in another direction. But we would not 48. To avoid questions about how to interpret the expession ‘true with respect to’, the hidden indexicalist might prefer to formulate the criterion along roughly these lines: A sentence S contains a hidden indexical referring to some x  T iff (a) S is true only if there is some x  T, such that the result of expanding S to include explicit reference to x is true; and (b) for some x, y  T, the result of expanding S to include explicit reference to x is true, but the result of expanding S to include explicit reference to y is not. 49. Kenneth Taylor (2001) considers a similar objection, with regard to an utterance of ‘Laura danced the tango all night last night’, which he readily admits clearly does not have a location as an unarticulated constituent. He argues that what makes this sentence different from ‘It’s raining’ is that “When Laura is dancing in a place, it is not the place that undergoes the dancing” (2001: 54). Clearly, though, the same cannot be said of Rachel’s eating and what she ate—what she ate underwent eating no less than a rainy location undergoes raining.

32

The instability of belief ascriptions

conclude—as the Relativity Test would have us conclude—that part of what is said in an utterance of the sentence ‘Samantha is walking’ is that she is walking at such-and-such a speed, in such-and-such a direction, etc.50 Or to take an example closer to the issue at hand: one cannot assert a particular proposition except by uttering words (albeit perhaps in a broad sense of ‘uttering words’); and one could assert a proposition by uttering certain words without uttering other words. But we would not conclude that part of what is said in an utterance of the sentence Samantha asserted that tailgating is dangerous is that she uttered such-and-such particular words. Alternatively, according to Jason Stanley and Zoltan Szabo’s “argument from binding” (2000: 94-95), an unmentioned location must be part of what is said in an utterance of ‘It’s raining’, on the grounds that Wherever there are rain clouds, it’s raining is naturally interpreted as For every location l at which there are rain clouds, it’s raining at l. (See also Stanley (2007: 51-54, 225-227). I have adjusted their argument to the present example.) If so, then by the same token, since There are several ways in which Jack has been kind to Jill is naturally interpreted as There are several w such that w is a way of being kind and Jack has been kind to Jill in way w, it would presumably follow that what is said in an utterance of ‘Jack has been kind to Jill’ includes an unmentioned way of being kind. I take this to be a reductio ad absurdum of Stanley and Szabo’s argument.51

50. Cappelen and Lepore make a similar point (2005: 74–75). 51. In “Language in Context” Stanley (2007) replies to several objections of “over-generation,” but none of those replies apply to the example I give here.

The indexical view 33

Yet another reason for construing the intended location as part of what is said in a weather report could be that the utterer of a weather report typically does have in mind a particular location with respect to which he expects his utterance to be interpreted—it would be strange (though not impossible) to say ‘It’s raining’ meaning simply that it is raining somewhere.52 The general idea might be formulated thus: Having-in-Mind Test of Hidden Indexicality A sentence S contains a hidden indexical referring to some element of (type) T iff an utterer of S typically has in mind a particular element of T with respect to which the utterer expects the utterance to be interpreted. Weather reports differ in this respect from the examples in which direct objects or adverbial qualifiers are not mentioned; the utterer of ‘Rachel is eating’ and ‘Samantha is walking’ would typically (but not necessarily) mean only that Rachel is eating something and that Samantha is walking in some way. This difference, however, is irrelevant, because what is typically meant by uttering a particular sentence is not necessarily what is said. This is evident from tautological sentences, such as ‘War is war’, as well as metaphorical uses of sentences that are literally false, such as ‘Elvis Costello is from another planet’. Simply to assume that what a speaker typically means in uttering a particular sentence is what is said in an utterance of the sentence is just to miss the distinction between what is meant and what is said. Thus, I diverge sharply here from Stanley, Szabo, and Jeffrey King, for whom “accounting for our ordinary judgements about the truth-conditions of various sentences is the central aim of semantics. … these judgements are the data of semantic theorizing…” (2000: 90), endorsed by Stanley and King (2005: 160). On my view semantics is primarily concerned with the truth-conditions themselves, not our ordinary judgments about them. Accounting for our ordinary judgments about the truth-conditions of various sentences is the job of an overall theory of communication, of which se52. (a) This might be what lies behind Taylor’s assertion that “The semantic incompleteness [of ‘It’s raining’] is manifest to us as a felt inability to evaluate the truth value of an utterance of [the sentence] in the absence of a contextually provided location (or range of locations)” (2001: 53). (b) Recanati (2007) presents an example of how one could say ‘It’s raining’ meaning simply that it is raining somewhere.

34

The instability of belief ascriptions

mantics is only one part. Semantic theories, about what words mean, must be coordinated with pragmatic theories, about how words, given their meanings, are used. This is not to say, of course, that what is typically meant in uttering a particular sentence has no bearing at all on what is said in uttering the sentence; clearly they are related (somehow) and sometimes even coincide. (I will discuss semantic intuitions at greater length in Chapter 2, Section 8.) By treating “unarticulated constituents” as constituents of what is said, hidden-indexical semantics violates not only the distinction between semantic and pragmatic content, but also the standard view of semantic content as compositional. As Crimmins rightly observes, it is generally assumed that “a principle of compositionality worthy of the name” (1992: 9) would require that the semantic content of a complex expression depend only on the arrangement and semantic contents of its component expressions.53 But the unarticulated constituents of hidden-indexical semantics are by their very nature beyond the contents of component expressions.54 The problems with hidden-indexical semantics become even more apparent when the method is generalized to other implicit elements of utterances. Consider, for instance, the various ways in which a speaker might intend an utterance to be interpreted—literally, sarcastically, metaphorically, etc. A speaker will generally intend his utterance to be interpreted according to a particular one (or more) of these modes of interpretation, and interpretation by different modes would generally yield nonequivalent propositions. Al53. Crimmins formulates this slightly differently, omitting the role of syntactic structure and relativizing content to statements, but the differences are not significant in this discussion. 54. I am assuming that hidden indexicals do not count as component expressions of a compound expression. If the hidden-indexicalist means to preserve compositionality by rejecting this assumption, the “compositionality” thereby achieved would be decidedly unworthy of the name. Apart from the disputed examples of unarticulated constituents—which as counterexamples to the standard view of compositionality can only beg the question— Crimmins provides no grounds for rejecting the standard view of compositionality, nor does he offer any alternative. Stanley and Szabo maintain that apart from the “phonological sentence” that is “articulated” in an utterance there is also an associated “grammatical sentence” that is “uttered,” but not all of whose components are necessarily actually pronounced (2000: 77–78). Locating hidden indexicals in the grammatical sentence allows for compositionality of the posited unarticulated grammatical sentences, but not of the phonological sentences, that we actually articulate.

The indexical view 35

though the speaker could specify the intended mode of interpretation explicitly, usually it is left implicit, discernible from the context. But if items implicitly affecting the apparent truth values of a speaker’s utterance (the utterance’s truth-value according to the “ordinary judgements” about truthconditions that Stanley, Szabo, and King aim to preserve in semantic theorizing) are to be taken as the referents of hidden indexicals in the sentence uttered, then these modes of interpretation—usually thought of as paradigmatically extrasemantic—would always be part of what is said. So, for instance, when I say ‘Marvin is a movie star’, I would be referring not only to Marvin and movie-stardom, but also, by means of a hidden indexical, to the way in which my utterance is to be interpreted—literally or sarcastically. Similarly, instead of indexicalizing the modes of interpretation, one could just as well indexicalize the interpretations they yield. A speaker will generally intend any expression he uses to be interpreted as having a certain sense—whether it be one of the finitely many literal senses of the expression, or one of indefinitely many nonliteral senses—and the intended senses of someone’s words are generally recognized without explicit specification. By the method of hidden indexicals, however, the intended senses of expressions could be picked out by hidden indexicals, so the particular interpretation a speaker’s words are to be given would always be determined as such by what the speaker says. Indeed, Crimmins and Perry themselves present the English construction ‘in the sense that …’ as an example of a way of “articulating commonly suppressed constituents of a claim” (1989: 701). But if a specification of intended senses were generally part of the semantic content of utterances, then no utterance could ever be semantically ambiguous. For instance, if I say ‘Betty is at the bank’, I could be said to be referring by means of a hidden indexical to my intended sense of ‘bank’—thereby resolving any ambiguity. Indeed, even nonliteral content can be construed this way as part of what is said. If I say ‘Betty is at the bank’ meaning that she’s “at the bank” in the sense that she is working on her finances, then according to Crimmins and Perry’s view of the construction ‘in the sense that’, this nonliteral sense of my words is an unarticulated constituent of what I said.55

55. Cf. François Recanati on “transfer” (2004: 263–266). For a positive account of alleged cases of unarticulated constituents, see Section 5.2, note 66.

36

The instability of belief ascriptions

4.3. Articulated indexicality Rather than posit hidden, unarticulated indexicals, Mark Richard (1990) argues that the sentence (2)

Lois Lane believes Superman is a reporter

is indexical on the grounds that the very word ‘believes’ is indexical, referring to different relations in different contexts. Roughly, Richard construes (2) as saying the following: The sentence ‘Superman is a reporter’ represents a sentencethat Lois Lane accepts, where “the nature of the representation relation changes from context to context” (1990: 149). For example, in some contexts the sentence ‘Superman is a reporter’ represents the sentence ‘Clark Kent is a reporter’ (which Lois Lane accepts), rendering (2) true; in others it does not, rendering (2) false. Whether or not one sentence represents another in a given context is a matter of whether or not the second is the result of certain contextually restricted uniform substitutions of coreferential expressions in the first. For example, the sentence ‘Superman is a reporter’ comes from substituting the name ‘Superman’ for the coreferential name ‘Clark Kent’ in the sentence ‘Clark Kent is a reporter’, so the second sentence represents the first in those contexts where this substitution is among those determining the nature of the representation relation. Taking the set of such substitutions determined by a given context as a correlation, (2) can be expressed as follows: The sentence ‘Superman is a reporter’ represents--under a contextually suitable correlation--a sentence that Lois Lane accepts. Richard’s theory is actually far more intricate than what I have so roughly described (and the final version is still more intricate than what follows, extracted from (1990: 137). First, an annotation is an ordered pair of an expression and its Russellian referent, e.g., , , . Then the Russellian annotated matrix, or RAM, of a sentence is a sequence of annotations, one for each constituent of the sentence, e.g., is the RAM of ‘Clark Kent is a reporter’. A correlation is a function mapping annotations to annotations having the same Russellian referent. A RAM p= represents a RAM q under the correlation f iff q = . A believer’s representational system, or RS, is the set of RAMs of all the sentences she accepts (or of all “the mediators of her beliefs” (1990: 137), or of all the “sentences ‘believed-true’” (1990: 175)). Then what (2) says is this: The RAM of the sentence ‘Superman is a reporter’ represents a RAM in Lois Lane’s representational system, this representation being under a correlation appropriate to the context.56 Richard’s proposal raises some challenging questions. One concerns the ways in which the semantics of a language might be constrained by the conceptual wherewithal of the language’s speakers. How plausible could it be to suppose that when we ascribe beliefs we are actually speaking of something as exotic as Russellian annotated matrixes? Richard dismisses such questions as irrelevant, explaining that his aim is to give only “a correct and illuminating account of the truth conditions” (1990: 157, n. 14) of belief ascriptions, not a “meaning-preserving” analysis (and apparently not reference-preserving either). But if that is the case, then if we seek a semantic theory of belief ascriptions, and if such a theory is to characterize the meanings of such sentences, Richard’s proposal is not relevant. Another question concerns the characterization of a believer’s “representational system.” Although Richard sometimes presents it as the set of RAMs of the sentences a person accepts, this is clearly only an approximation, given that nonverbal creatures might have beliefs (as Richard allows). When first introducing his notion of a representational system, Richard defines it more generally as the set of RAMs of “all of the mediators of her [the believer’s] beliefs.” One challenge here is to explain what these mediators are. Richard speaks of them as sentential psychological objects having parts or aspects that represent or determine individuals and properties analogously to how expressions of natural language do; and having one 56. David Chalmers (2011) offers what he considers a descendant of Richard’s theory, according to which (2) means that Lois “endorses” an “enriched proposition” that is “coordinated” (relative to the context) with the enriched proposition expressed by ‘Superman is a reporter’. However, Chalmers does not address the question at hand of whether sentences such as (2) are ambiguous, indexical, or indeterminate.

38

The instability of belief ascriptions

consists in being in a certain sentential state (1990: 40-41, 252). Even if this sufficiently clarified what mediators are, the definition of a representational system as the RAMs of the mediators of a person’s beliefs seems to require a prior notion of belief itself, in order to distinguish the mediators of a person’s beliefs from the mediators of her other propositional attitudes. Although Richard discusses many aspects of this picture, it is not clear that his position can be adequately clarified in a noncircular way—i.e., that instead of explaining belief ascriptions in terms of the sentences a person accepts, he could explain them in terms of something else, that correlates with belief states but is not characterized in terms of belief. Of course, as long as Richard aims only to give truth conditions of belief ascriptions, this kind of circularity may not be a problem—although, if the truth condition the theory gives for (2) is ultimately something like this: The RAM of the sentence ‘Superman is a reporter’ represents a RAM that Lois Lane believes, then there may be a question about how illuminating the theory is. (The plausibility of analyzing belief in terms of sentential mediators will be the subject of Chapter 3.) More hard questions arise regarding the generalization of Richard’s theory to other propositional attitude reports. Clearly, if ‘believes’ is indexical in the way Richard says it is, then as noted with regard to the view that belief ascriptions are ambiguous, the phenomenon must be widespread, occurring in all the “psychological” verbs—’wishes’, ‘hopes’, ‘wants’, ‘fears’, etc. But the theory cannot stop there, because propositional attitudes can be ascribed without psychological verbs: Lois Lane has the belief that Superman is a reporter. Lois Lane is of the opinion that Superman is a reporter. Where would the indexicality lie in ascriptions such as these? If not in the verbs, then in the nouns ‘belief’ and ‘opinion’ (and hence also in all other “psychological” nouns)? Or would ‘has the belief’ and ‘is of the opinion’ be treated as separate lexical items, synonymous with ‘believes’ and semantically noncomposite? And then there are also propositional attitude ascriptions such as these: According to Lois Lane, Superman is a reporter.

The indexical view 39

For Lois Lane, Superman is a reporter. Is ‘according’ indexical? Is ‘for’? Perhaps the greatest challenge facing Richard’s approach is to reconcile it with ordinary intuitions about indexicality. Avoiding the unorthodoxy of indexicals that are hidden, Richard nevertheless posits indexicality that is hidden; for even if the word ‘believes’ is indexical, it is surely not straightforwardly so—Richard’s proposal is highly innovative and surprising.57 The word ‘believes’ comes out indexical on Richard’s view only because it involves tacit reference to correlations; it refers to a representation relation that changes from context to context, because the representation relation is representation under a certain correlation, that depends on the context. This runs against ordinary intuitions about indexicality in at least three ways. First of all, we expect indexicality to be apparent, not covert. Indexical expressions ordinarily (if not always) wear their indexicality on their sleeves; moderately well-trained speakers can recognize indexicality fairly easily, especially where the indexical referent is to be determined by an intention of the speaker’s (as would seem to be the case in Richard’s proposal).58 Secondly, we expect speakers using indexicals to be able to tell us what they are referring to. Even when the indexical referent is somewhat vague (e.g., ‘It’s too noisy here’ or ‘This is beginning to make me feel uncomfortable’), we expect the speaker to be able to specify at least roughly what it is that he is talking about. But it is doubtful that speakers ascribing belief could even begin to specify which correlation or representation relation they are indexically referring to, especially if they do not realize that they are referring to such things in the first place and do not even have any idea of what such things are.59 Thirdly, Richard deviates from ordinary intuitions most sharply in proposing a kind of indexical variation that is not at all systematic. It is a hallmark of indexical reference that it shifts across contexts systematically—’I’ refers in each context to the speaker of the utterance, ‘now’ refers in each context to the time of the utterance (perhaps in conjunction with a directing 57. Cappelen and Lepore thus refer to such an approach as the “Surprise Indexical Strategy” (2005: 8). 58. As Cappelen and Lepore put it, “If expression e is context sensitive, then it’s obviously context sensitive” (2005: 112). 59. Even if Richard’s aim is only a correct and illuminating account of truth conditions, his appeal to indexicality cannot ignore the ways in which indexicality is intuitively tied to speaker’s intentions.

40

The instability of belief ascriptions

intention to refer to a particular longer or shorter range of time that includes the time of the utterance). Even when indexical variation seems least constrained—e.g., ‘this’—the variation is still systematic at least in being determined by a directing intention. But the proposed indexical reference to correlations (or the various representation relations they determine) cannot be secured by a directing intention of a speaker who knows nothing of such referents; and there is no other way to characterize the purported indexical referent in a systematic way. Indeed, if indexicality can be attributed so easily, with no concern for systematicity or the cognitive limitations of speakers’ intentions, it provides a lazy man’s approach to any question about context-dependent intuitions about the truth-value of a sentence. When I say George is an elephant, I might mean, literally, that George is an elephant, or I might mean that George is obese, or that he seldom forgets, or that he is a Republican, etc. We could then say that the predicate ‘is an elephant’ is indexical, referring to these various properties in various contexts. More generally, we could say for any sentence S that it expresses (or at least is equivalent to) the proposition that the RAM of S represents a true RAM, where the representation relation is relative to a contextually determined correspondence (not necessarily a reference-preserving correlation in Richard’s sense) between annotations of the RAM of S and annotations of the true represented RAM; and then we could say that the contextually determined correspondence (or the representation relation it determines) is referred to indexically by the main verb in S. We could—but of course, no one would, and I think the reason why is that we have a natural conception of indexicality, based on paradigmatic indexicals such as ‘I’, ‘here’, and ‘this’, whereby indexical reference is apparent, articulable by the speaker, and systematic.

5. Semantic indeterminacy 5.1. Incompleteness Another way to account for the instability of belief ascriptions is by taking them to be semantically indeterminate, in the sense that the sentence fails to determine a particular proposition, even after all ambiguities are resolved

Semantic indeterminacy 41

and the referents of all singular terms are determined. One way a sentence can be semantically indeterminate in this sense is by being semantically incomplete—by having, in Kent Bach’s words, “a semantic ‘hole’” (1987: 76), as in the case of his example, (12) Pepsi is better. This kind of semantic indeterminacy allows for a proposition’s being determined once the semantic hole is filled—perhaps on the basis of features of the context of utterance. Thus, it may be apparent in one context that the utterer of (12) means that Pepsi is better than Coke, and in another context that the utterer means that Pepsi is better than a cheeseburger.60 Here there are two possibilities: either the completed proposition is taken as the semantic content of the semantically incomplete sentence or it is not. If it is, then perhaps the difference between semantic incompleteness and hidden-indexicality is only terminological. For, whatever is taken as filling a semantic hole in a semantically incomplete sentence could be taken instead as the referent of a hidden indexical, and vice versa.61 Indeed, although Crimmins and Perry (1989) liken the context-dependence of sentences with unarticulated constituents to that of sentences containing indexicals, and although Crimmins suggests that belief reports are “prime candidates for semantically motivated posits of covert expressions” (1992: 17), they do not share Schiffer’s terminology of “hidden indexicals.” “In a case of underarticulation,” they explain, “there is no expression to determine the constituent” (1989: 700) as there is in a case of indexical reference.62 So perhaps their position should be discussed under the heading of

60. Bach argues (though has since modified his view) that belief ascriptions are “semantically indeterminate” in that the sentence alone does not determine what “elements of the content” (1987: 213), or “way[s] of thinking” (1987: 199), if any, are being ascribed. Similarly, François Recanati holds that the sentence “underdetermines” what is said, in that the latter may include modes of presentation which can be determined only in context (1993: 355–357). 61. Thus, Cappelen and Lepore treat what they call “Surprise Indexicalists, Hidden Indexicalists, and proponents of Unarticulated Constituents” (2005: 12) as all holding different versions of the same view. 62. Cf. Bach: “An indexical is there in the sentence,” not hidden (1994: 133). Crimmins and Perry and Bach are apparently talking about expressions in what Stanley and Szabo call the “phonological sentence,” that is actually pro-

42

The instability of belief ascriptions

this section rather than that of the last (even though they most surely would not think of it this way). However, insofar as hidden-indexicality and semantic incompleteness are equivalent in the way explained, the considerations presented above would apply as well either way. The second possibility is that the completed proposition is not taken as the semantic content of the semantically incomplete sentence. This line may have the advantage of avoiding some of the troubles of hidden indexical semantics, but as long as belief ascriptions are taken as semantically incomplete, there is still the matter of what elements are to be added to the semantic content of the semantically incomplete sentence, and how, in order to obtain a complete proposition. As long as one maintains that what is missing in belief ascription sentences is something like modes of presentation, functioning as something like ways of having beliefs, one is faced with the same objections raised earlier to the appeal to implicit modes of presentation (Section 4.1, and more to come in Chapter 3). Moreover, whether what is allegedly missing from the sentence is construed as part of the semantic content or not, it remains doubtful that belief ascription sentences do not explicitly express complete propositions (as argued above with regard to hidden indexical semantics, Section 4.2).63

5.2. Similarity Stephen Stich (1986: 136) offers a different way of construing belief ascriptions as semantically indeterminate: On my view, a sentence of the form ‘S believes that p’ ascribes a certain sort of cognitive state to S and goes on to identify that state by comparing it to a hypothetical state of the ascriber. … To determine which belief-state is nounced, rather than some “grammatical sentence” that is presumably “uttered” but not pronounced (2000: 77–78). 63. More recently Bach has argued that “An utterance of ‘A believes that S’ is true iff A believes a certain thing which requires the truth of the proposition that S” (1997: 238). Still maintaining that belief reports are semantically incomplete, he apparently thinks that this “certain thing” (requiring the truth of the proposition that S) cannot be determined independently of context; and he seems to rule out its being anything like a way of taking propositions or a mode of presentation. However, he does not explain what this certain thing is, or how it can require the truth of a proposition (such as the proposition that S).

Semantic indeterminacy 43 being ascribed to S, we must imagine that the content sentence were to be said in earnest by the speaker in a setting akin to the one in which the belief sentence is uttered. We must then infer back to the belief-state which would cause such an utterance. The belief-state being ascribed to S is one which is similar to this one, the degree and dimensions of similarity being largely determined by the context of the utterance.64

Thus, Stich accounts for the instability of belief ascriptions by holding that “belief ascriptions are similarity claims, and similarity claims are context dependent” (1983: 106). Neither of these contentions, however, is entirely right. It is clear that to interpret an utterance of the sentence (13) Tel-Aviv is similar to New York City,

64. As a “rough paraphrase” of ‘S believes that p’ Stich (1983: 88) offers the following: p. S is in a belief state similar to the one which would play the typical causal role if my utterance of that had had a typical causal history. Stich is explicitly building here on Donald Davidson’s paratactic account of indirect discourse (1968: 105), according to which ‘Galileo said that the earth moves’ means The earth moves. (x) (Galileo’s utterance x and my last utterance make us samesayers). This account might be naturally extended to belief ascriptions, so that ‘Galileo believes that the earth moves’ would mean The earth moves. (x) (Galileo’s belief state x samesays my last utterance). (Cf. Lepore and Loewer (1989: 352–353).) However satisfying such a theory may be, it does not address the present question, since the instability of belief ascriptions would be inherited by assertions about samesaying; questions about belief and belief ascriptions would simply return as questions about samesaying and about assertions about samesaying. As Davidson himself points out, his proposal concerns only the logical form of certain sentences, not their semantic content (1967: 30–31; 1968: 104, n. 14); “a theory of the kind proposed,” he explains, “leaves the whole matter of what individual words mean exactly where it was” (1967: 32–33).

44

The instability of belief ascriptions

one must know what significance to attach to the various ways (respects, dimensions) in which the two cities might be compared, and that this may vary across contexts. For this reason one might suppose that such sentences are semantically incomplete, having a semantic hole for the relevant weighting of respects of similarity. But this leads to the same problems raised by hidden-indexical semantics with regard to the distinction between what a speaker means and what the speaker actually says. In speaking of “similarity claims” we must distinguish between the claim a speaker intends to make in uttering a sentence such as (13)—what the speaker means—and the claim that is the proposition expressed by the sentence (in that context)—what the speaker actually says. It may well be that in uttering (13) one might mean exactly what would be expressed by (a literal use of) the sentence (13') Tel-Aviv is similar to New York City with regard to being a national financial capital; but it does not follow that the sentences (13) and (13') have the same semantic content (or express the same proposition, in that or any other context).65 So if in saying “similarity claims are context dependent” Stich means only the pragmatic claim, that a sentence such as (13) can be used with the intention of making different claims in different contexts, then his claim is surely unobjectionable. But if what he means is the semantic claim, that sentences such as (13) (or literal utterances of such sentences) express different propositions in different contexts—that what is said in utterances of such sentences depends on the context—then his claim is false.66 65. We may ignore here any variation in the time of utterance and in the referents of the names. 66. If (13) does not (ever) express the same proposition as (13'), then what proposition does it express? One plausible answer is that what is said in an utterance of (13) is nothing—or at least nothing that is a proposition. (13)—as well as other semantically incomplete sentences, such as (12)—might thus be viewed as the uttered fragment of a longer sentence which is in some sense intended though not fully pronounced. Despite the grammaticality of (13), it would resemble in this respect nonsentential expressions that can nevertheless be successfully used to convey complete propositions—such as the one-word exclamation ‘Amazing!’ said with regard to something unmentioned but suitably salient. Alternatively, the proposition expressed by (13) may be just the existential generalization that Tel-Aviv is similar to New York City in

Semantic indeterminacy 45

A failure to distinguish what is said from what is meant may also be responsible for Stich’s construal of belief ascriptions as similarity claims. There is no doubt that belief ascriptions involve similarity judgments, some of which Stich elucidates. But this feature is not peculiar to belief ascriptions, for every assertion involves a similarity judgment—whenever we apply a predicate, we do so at least partly because we judge the thing to which we are applying the predicate similar (in the relevant ways) to the things to which the predicate is typically applied. We would not call something a “chair,” for instance, unless we deemed it similar (in the relevant ways) to the things typically called “chairs.” However, assuming that not every assertion is a similarity claim, these implicit similarity judgments could not be what is said in the assertions. That we mean to use our words in ways that are similar to how they have been used before is not what we actually say every time we speak, but just a fact of language. Indeed, if belief ascriptions were similarity claims, then the ascription (2)

Lois Lane believes Superman is a reporter

would express nothing more than (2*) Lois Lane is similar (with respect to belief states) to someone who believes Superman is a reporter.67 Though (2*) may be implicit in (or entailed by) (2), it surely does not express what (2) expresses.68 some respect. Although this would render all unqualified similarity sentences trivially true, it may be possible to explain pragmatically how we can utter a trivially true sentence, such as (13), to mean something not at all trivial, such as (13') (as in a typical utterance of ‘War is war’, for example). Similar treatments can be given for ‘It’s raining’ and other such purported examples of sentences with unarticulated constituents. (Emma Borg (2004: 240; 2007: 350–351) calls this “the hidden argument place version of minimalism,” for which she argues.) 67. I am assuming that a person S is in a belief state similar to the belief state b, iff S is similar (with respect to belief states) to someone in a belief state similar to b. 68. Stich has conceded in conversation that his analysis of belief ascriptions was intended to capture what speakers would typically say is conveyed in a belief ascription—what I refer to as pragmatic content (Section 2.6)—as opposed to

46

The instability of belief ascriptions

5.3. Hopelessness In light of the many failed attempts to reach a satisfactory theory of the semantics of belief ascriptions, it may be tempting to conclude that belief ascriptions are semantically indeterminate, in the sense that the usage of such sentences is just too irregular to be characterized by any proper semantic theory. Quine (1977: 122), for instance, after having worked through a number of such failed attempts, concludes that “our renunciation must extend to all de re belief, and similarly, no doubt, for the other propositional attitudes. ... I see the verb ‘believe’ even in its de dicto use as varying in meaningfulness from sentence to sentence.”69 Kripke (1979: 378) seems similarly inclined, at least with regard to those belief ascriptions which have been especially recalcitrant to semantic theory, warning of “an area where our normal practices of interpretation and attribution of belief are subjected to the greatest possible strain, perhaps to the point of breakdown.” And on the basis of the many difficulties encountered by theories of the semantics of belief ascriptions, Schiffer (1987b) abandons the view that natural languages have compositional semantic theories altogether. It will be helpful to distinguish between two different ways of arriving at the view that belief ascriptions are semantically indeterminate (in the strong sense being considered). One way is to argue on the basis of the sort of “breakdown” of which Kripke warns, noting extraordinary cases to which the regularities we recognize in our normal usage of belief ascriptions do not seem to extend. However, as with the implicit similarity judgments discussed above, this breakdown phenomenon is not unique to belief ascriptions—all our linguistic practices are rooted in normal cases, so they all break down in suitably bizarre circumstances. Perhaps the best known exposition of this point is Wittgenstein’s (1958: §142; see also §80): It is only in normal cases that the use of a word is clearly prescribed; we know, are in no doubt, what to say in this or that case. The more abnormal the case, the more doubtful it becomes what we are to say. And if things were quite different from what they actually are—if there were for instance something such as their semantic content, strictly construed—a concept of which he is skeptical. 69. Still more recently, Quine (1995: 358) concludes that “the requirement that distinguishes de re from de dicto, namely knowing who or what, is a function of the contextual situation and not a general distinction,” and hence, that the idiom of propositional attitudes de re is not to be admitted to scientific discourse.

Semantic indeterminacy 47 no characteristic expression of pain, of fear, of joy; if rule became exception and exception rule; or if both became phenomena of roughly equal frequency—this would make our normal language-games lose their point.

Thus, as we move from paradigmatic cases of the use of an expression to cases that are more and more bizarre, we are bound to reach an area where our practices regarding the use of the expression break down, in the sense that we can no longer discern regularities allowing us to extrapolate (in a principled, consistent way) from standard uses of the expression to such radically abnormal cases.70 Now, if it is assumed that the semantic theory of a particular expression must cover all cases in which the expression could be used, no matter how nonstandard, then from the breakdown of linguistic practice with regard to a particular expression, one may conclude that there can be no semantic theory of that expression, so that sentences containing that expression would be semantically indeterminate. And if every expression of natural language suffers from such breakdown, so much the worse for the semantics of natural language, which would thus be impossible. (One might misinterpret Wittgenstein as arguing this way.) However, we need not assume (and I do not think Wittgenstein assumed) that semantic theories must be so comprehensive as to cover even the most deviant cases. Indeed, the prevalence of breakdown is good reason to think that this would be too much to expect of semantics. Insofar as the semantics of a natural language is determined by phenomena as amorphous as regularities in the behavior of the speakers of the language, it would not be reasonable to demand of semantic theories that they provide completely comprehensive, mathematically precise models. This is not to say, of course, that hard cases may be ignored. As with theories in any other science, the more comprehensive a semantic theory is, the better. However, though semantic theorizing may be limited by the fact that linguistic practice breaks down in strange enough cases, it is not thereby precluded.71 Rather than argue for the semantic indeterminacy of belief ascriptions on the basis of breakdown, one might do so inductively. That is, one’s 70. This breakdown phenomenon goes hand in hand with that of the aforementioned implicit similarity judgments. For if when we apply a predicate to something, we must assume, and thereby indicate, that we take the thing to be suitably similar to the things to which the predicate is paradigmatically applied, then the less similar the case at hand is to paradigmatic cases, the more problematic the application of the predicate would be. 71. For more on this, see my (2002).

48

The instability of belief ascriptions

grounds for concluding that no semantic theory of belief ascriptions can be correct might be simply that none of the theories one has considered are correct. (This seems to be Quine’s strategy, and also Schiffer’s—with the important difference that Schiffer makes his argument deductively valid by claiming to have considered all the possibilities.) The strength of such an argument would depend, of course, on whether any theories were neglected or incorrectly rejected. In the following section I shall turn to such a theory.

6. Direct belief Frege seems to have considered sentences such as (2)

Lois Lane believes Superman is a reporter

false, which seems to ignore the Quinian intuition that in certain contexts (2) is acceptable.72 On the other hand, taking (2) to be both true and false (depending on the context), or neither true nor false, leads to the problems sketched in the previous sections. The only other possibility is that (2) is indeed true, Fregean intuitions notwithstanding. But the view that (2) is true has more behind it than just the shortcomings of the other possibilities. It is hard to deny Leibniz’s Law, that if a=b, then whatever is true of a must be true of b; which would seem to entail that names of the same individual can be substituted for one another without changing the truth-values of the sentences in which they occur—which, indeed, is often given as just another formulation of Leibniz’s Law.73 If so, then the truth of (2) is guaranteed by that of (1)

Lois Lane believes Clark Kent is a reporter,

given that Superman is Clark Kent. Moreover, well known observations by Keith Donnellan (1970), Saul Kripke (1972), David Kaplan (1978, 1979, 1989), and Hilary Putnam 72. Cf. my (1988: 358–362), where I show how pragmatic considerations might help to reconcile Quinian intuitions with the view that (2) is false. 73. Leibniz’s Law would actually (and not just seemingly) entail that names are interchangeable salva veritate only where the names are used strictly referentially, only to refer to their bearers, and not to refer in any way to themselves (or to serve as the referents of other expressions).

Summary 49

(1975) provide strong evidence for the natural view that the meaning of a proper name is nothing but the name’s bearer, in the sense that the semantic contribution of a name to the proposition expressed by a sentence containing the name is simply the individual named. This, too, seems to support the formulation of Leibniz’s Law allowing substitution of coreferential names, thereby yielding (2) as above. Motivated by these considerations, the Theory of Direct Belief treats having a belief about an individual as an unmediated relation between the believer and the individual the belief is about. In particular, (2) is taken at face value, as expressing a relation between Lois Lane and the singular proposition consisting of the individual, Superman, and the property of being a reporter. Thus, since Superman is Clark Kent, (1) and (2) express the same proposition. The Theory of Direct Belief has been widely dismissed or ignored mainly because of the healthy Fregean intuition that (2) is false—or more generally, that substitution of coreferential names in the embedded sentence of a belief ascription may fail to preserve the truth-value of the ascription. However, in the next chapter I shall explain how such intuitions come from failing to distinguish what is said in a belief ascription from what is otherwise conveyed by uttering the ascription. The other main objection to the Theory of Direct Belief comes from those who argue that the semantics of belief ascriptions must involve something like modes of presentation. In Chapter 3 I explain how those arguments fail, and I argue more generally against adverting to modes of presentation in the semantics of belief ascriptions.

7. Summary The chapter began with a puzzle about conflicting intuitions about belief ascriptions. On the one hand, as Frege taught us, coreferential names do not always seem to be interchangeable in belief contexts salva veritate—e.g., it seems that (1)

Lois Lane believes Clark Kent is a reporter

can be true while (2)

Lois Lane believes Superman is a reporter

50

The instability of belief ascriptions

is false, even given that Clark Kent and Superman are the very same individual. On the other hand, as Quine observes, coreferential names sometimes are interchangeable in belief contexts salva veritate—e.g., without changing any of the details of the Superman story, we can easily imagine a context of utterance where (2) seems true: Look at what a master of disguise that Superman is! Why, when he puts on that suit and strolls into his office at the Daily Planet, he fools everybody—Mr. White, Jimmy Olson; even Lois Lane believes Superman is a reporter. So the question is how to reconcile the Fregean intuitions about sustitution failure with Quinian intuitions about substitution success. Putting the question this way conveniently allows a natural division of logical space into four areas with regard to a particular belief ascription such as (2): I. II. III. IV.

(2) is true, despite the Fregean intuitions about substitution failure; or (2) is false, despite the Quinian intuitions about substitution success; or (2) is both true and false, depending somehow on the context of utterance, or (2) is neither true nor false altogether.

Favoring option I (and dismissing option II as more or less a nonstarter— though see Section 6, n. 72), I have presented in this chapter a critical review of various ways of pursuing options III and IV. I began by observing that belief ascriptions such as (2) do not meet the standard criteria for lexical ambiguity. The word ‘believes’ does not seem to have disambiguating expressions in known languages, nor do we expect to find disambiguating expressions for it in unknown languages. Nor do belief ascriptions such as (2) pass the “crossed interpretation” test. Moreover, if ‘believes’ were ambiguous in the way being considered, so too would a wide range of expressions, including all synonyms of all propositional attitude verbs—including complex expressions, such as ‘holds the view that’, where it is hard to see where the putative ambiguity might be located. And if we aim to respect something like Grice’s Modified Occam’s Razor (“Senses are not to be multiplied beyond necessity”), then we should

Summary 51

not posit an ambiguity when the data can be explained—as I shall argue it can—by means of some independently established theoretical apparatus. A more popular way of taking belief ascriptions such as (2) as ambiguous has been Russell’s method, but following Russell requires accepting not only his theory of definite descriptions, but also his theory of common proper names—both of which lead to familiar problems. Alternatively, I considered a less controversial way of finding a scope ambiguity in belief ascriptions, without incurring the costs of Russell’s tendentious claims about proper names and definite descriptions—but this scope ambiguity was seen to yield only equivalent interpretations. Moreover, since belief ascriptions such as (2) fail the “crossed interpretations” test, that would seem to rule out syntactic ambiguity as well. I also argued that despite Quine’s claim to the contrary, his treatment of belief ascriptions does not avoid the assumption that they are ambiguous, and hence does not evade the objections I have presented against that assumption. And in the end I explained how Paderewski-type examples render any merely binary ambiguity insufficient for accounting for the indefinitely many shifty intuitions that a belief ascription may generate. Next I considered the possibility that the reason our intuitions about the truth of belief ascriptions such as (2) seem to depend on the context of utterance is that belief ascriptions contain indexical expressions, referring in different contexts to different “modes of presentation.” This approach was seen to incur two formidable tasks right from the start. The first is to explain what modes of presentation are; and the second is to explain the ternary relation in which they are presumed to take part: what exactly is it to believe something under a particular mode of presentation? Moreover, if modes of presentation are introduced as part of an explanation of what belief is, then on pain of circularity, one would need to explain modes of presentation without adverting to the notion of belief. The indexical approach was also seen to suffer from the problem of what to point to as the allegedly indexical expression. Since none of the expressions in (2) appear to be indexical (in the right way), hidden indexical theorists suggest that belief reports are like weather reports, in that just as a speaker who says ‘It’s raining’ implicitly refers to a place, so too, the utterer of a belief ascription refers implicitly to a mode of presentation. Against this I argued that the analogy is significantly flawed, in that someone who says ‘It’s raining’ can easily specify the place that he did not explicitly mention, but the utterer of a belief ascription would be hard put to

52

The instability of belief ascriptions

articulate the particular mode of presentation to which he presumably referred. More significantly, I argued against the method of hidden-indexical semantics in general. First I showed how this method seems to miss the distinction between what a speaker means and what the speaker actually says, since it fails to distinguish between the semantic contents of sentences differing only in whether or not a particular constituent is articulated. Then I considered various proposed criteria for the presence of hidden indexicals—the Relativity Test, the argument from binding, and the Having-inMind Test—all of which were seen to fail. In conclusion I pointed out some of the more bizarre consequences of hidden-indexical semantics that seem to make it clearly incompatible with any normal understanding of semantic content. I then turned to Mark Richard’s proposal that belief ascriptions such as (2) contain an indexical that is not hidden, viz., the word ‘believes’. Richard’s proposal was seen to face several challenges, much like those facing the hidden-indexical view. First, it is doubtful that when we ascribe beliefs, we are actually speaking of something as exotic as Russellian annotated matrixes. Secondly, Richard owes us an explanation of his notion of “mediators,” in terms of which he defines his idea of a “representational system”—which itself stands in need of a noncircular explanation. Thirdly, if ‘believes’ is indexical in the way Richard says it is, then (as noted with regard to the ambiguity view) so too must be all synonyms of all psychological verbs, as well as other expressions used for ascribing propositional attitudes, such as ‘according to’. Fourthly, Richard’s approach fares no better than the hidden-indexical-view in conforming to our pre-theoretical intuitions of indexicality, whereby indexical reference is apparent, articulable by the speaker, and systematic. Deviating from the standard conception of indexicality was shown to provide a lazy man’s approach to any question about context-dependent intuitions about the truth-value of a sentence. So much for option III (whereby (2) is true in some contexts, false in others). I then considered three ways of pursuing option IV (whereby (2) is neither true nor false altogether). The first was by taking belief ascriptions such as (2) as semantically incomplete, having a semantic hole that needs to be filled in order to determine a complete proposition. If the completed proposition is taken as the semantic content of the semantically incomplete sentence, then the difference between semantic incompleteness and hiddenindexicality was seen to be insignificant, since whatever is taken as filling a

Summary 53

semantic hole in a semantically incomplete sentence could be taken instead as the referent of a hidden indexical, and vice versa. In that case, all the objections to the hidden-indexical view would apply as well to the view that belief ascriptions are semantically incomplete. If, on the other hand, the completed proposition is not taken as the semantic content of the semantically incomplete sentence, then as long as what is allegedly missing in belief ascription sentences is something like modes of presentation, functioning as something like ways of having beliefs, one still faces the same objections raised with regard to implicit modes of presentation. Moreover, as argued with regard to hidden-indexical semantics, it remains doubtful that belief ascription sentences do not explicitly express complete propositions. A different way of taking belief ascriptions as semantically indeterminate is proposed by Stephen Stich, arguing that “belief ascriptions are similarity claims, and similarity claims are context dependent” (1983: 106). Both of these claims were rejected as implausible if taken as claims about semantic content. Finally I considered the view that belief ascriptions are semantically indeterminate in the sense that the usage of such sentences is just too irregular to be characterized by any proper semantic theory. One way to arrive at this view is on the basis of Kripke’s conclusion that the most recalcitrant puzzles of belief seem to lie in “an area where our normal practices of interpretation and attribution of belief are subjected to the greatest possible strain, perhaps to the point of breakdown” (1979: 378). Here I argued that this breakdown phenomenon pervades all our linguistic practices, but we need not assume that semantic theories must be so comprehensive as to cover even the most deviant cases; we should be able to live with a little semantic indeterminacy around the edges. Another way to arrive at the view that belief ascriptions are semantically indeterminate is inductively, on the basis of failed attempts to get their semantics right. This is where I offer a theory that succeeds--the Theory of Direct Belief.

Chapter 2 The pragmatics of substitutivity

1. Truth and appropriateness According to the Theory of Direct Belief, having a belief about an individual is an unmediated relation between the believer and the individual the belief is about.1 In particular, (1)

Lois Lane believes Clark Kent is a reporter

is taken at face value, as expressing a relation between Lois Lane and the singular proposition consisting of the individual, Clark Kent, and the property of being a reporter. Thus, assuming Clark Kent is Superman, (1) expresses the same proposition as (2)

Lois Lane believes Superman is a reporter.

Ever since Frege, however, examples such as this have widely been taken as showing that coreferential names are not always interchangeable salva veritate, for it is generally assumed that given the traditional Superman story, (1) is true and (2) is false. But the assumption that (2) is false is mistaken. In this chapter I shall explain how the inclination to reject (2) is to be accounted for not by the alleged falsity of the sentence, but by the inappropriateness of its utterance.

1.

The requirement that the belief relation be unmediated makes the Direct Belief theorist an exception to common views of “(neo-)Russellian” theories— e.g., Richard: “Although Russellians take attitudes like belief to be relations to Russellian propositions, they allow that such relations are mediated relations” (1990: 120); Recanati: “neo-Russellian” theorists generally define belief in terms of a ternary relation involving something like modes of presentation (1993: 43, n. 10). This divergence between the theory of Direct Belief and Russellian theories employing a ternary belief relation will be largely insignificant in this chapter, though it will be central in the next. See Chapter 3, Section 1, especially note 4.

Truth and appropriateness 55

There are many ways a sentence may be true while its utterance would not be appropriate. The utterance of a true sentence may be impolite, for instance. Or, as in the case of (2), the utterance of a sentence may be inappropriate because of the falsity not of the sentence uttered, but of something conveyed by the sentence’s utterance.2 I am relying here on a distinction between what is actually said in an utterance, on the one hand, and all that the utterance may convey (whether intended by the speaker or not), on the other. Such a distinction is hard to spell out but still harder to deny, since there are so many clear cases of saying one thing while conveying something else.3 For example, by saying sarcastically that Hedda is a great cook, I can convey the proposition that Hedda is a horrible cook; by saying metaphorically that Rudyard is a refrigerator, I can convey the proposition that Rudyard is stubborn. Moreover, part of what is conveyed in any normal utterance of mine is that I believe what I am saying, that my listener can hear me, that we both know the language I am using, etc.—but normally, of course, none of this is part of what is said. Grice’s theory of conversational implicature (1989) is about one way of conveying something without saying it; and it is in this way that an utterance of (2) would normally convey something false, thereby rendering the utterance inappropriate. After a brief review of Grice’s theory I shall explain how it applies to belief ascriptions.4

2. 3.

4.

Nathan Salmon (1986: 59) makes the distinction in terms of what is “semantically encoded” and what is “pragmatically imparted.” For now we may take what is said in the utterance of an indicative sentence to be the proposition (or propositional components) expressed by the sentence— the sentence’s semantic content—relative to a particular way of resolving any ambiguities in the sentence and of determining the referents of any (explicit) indexicals. Cf. Recanati (1993: 263, 265), who sometimes prefers to call this “the primary semantic value;” see below, Section 7. Also cf. Soames (2002), who maintains that what is said or asserted in uttering a sentence typically transcends the sentence’s semantic content; see below, Section 10.1. See also Chapter 1, note 45. (a) I diverge here from Salmon (1986), who refrains from offering any particular account of how utterances of belief ascriptions pragmatically impart what they do. (b) Wayne Davis (1998) puts forth a range of interesting objections to Grice’s theory, but although I present my view in Gricean terms, I do not believe that I make essential use of any of the Gricean claims that Davis rejects.

56

The pragmatics of substitutivity

2. Conversational implicature Viewing a large part of our linguistic practice as a kind of cooperative, goal-directed behavior, Grice (1989: 26) observes that participants in a talk exchange may be expected (ceteris paribus) to observe the following principle: The Cooperative Principle Make your 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 you are engaged.5 This general principle is elucidated by the following more specific maxims (1989: 26-27): Maxims of Quantity 1. Make your contribution as informative as is required (for the current purposes of the exchange). 2. Do not make your contribution more informative than is required. Maxims of Quality Try to make your contribution one that is true. 1. Do not say what you believe to be false. 2. Do not say that for which you lack adequate evidence. Maxim of Relation Be relevant. Maxims of Manner Be perspicuous. 1. Avoid obscurity of expression. 2. Avoid ambiguity. 3. Be brief (avoid unnecessary prolixity). 4. Be orderly. Grice then defines conversational implicature as follows (1989: 31):

5

A talk exchange may well have more than one accepted purpose or direction, but following Grice I will use the singular.

Implicated normalcy 57 A man who, by (in, when) saying (or making as if to say) that p has implicated that q, may be said to have conversationally implicated that q, provided that (1) he is presumed to be observing the conversational maxims, or at least the Cooperative Principle; (2) the supposition that he is aware that, or thinks that, q is required in order to make his saying or making as if to say p (or doing so in those terms) consistent with this presumption; and (3) the speaker thinks (and would expect the hearer to think that the speaker thinks) that it is within the competence of the hearer to work out, or grasp intuitively, that the supposition mentioned in (2) is required.

For example, if a motorist tells me he has run out of gas, and I reply by telling him that there is a gas station around the corner, I thereby conversationally implicate that the gas station is (or at least may be) open—for if I did not think so, my contribution to the conversation would be irrelevant to the purpose of the exchange, in violation of the maxim of Relation.

3. Implicated normalcy We routinely distinguish between what is normal and what is not. It is normal, for instance, for fresh grapes to be juicy, for people to feel uncomfortable in very hot weather, and for trees to grow toward the sky; it is not normal for classrooms to contain elephants, for apples to taste like chocolate, or for people to enjoy pain. Sometimes, of course, things are not normal—some people are masochists—but it is trivial that normally things are normal. Consequently, we normally presume that things are normal. This presumption is sometimes suspended, of course, as in cases of known or suspected abnormality. But since things are normally normal, normalcy is normally presumed.6 The presumption of normalcy generates conversational implicatures in the following way:

6.

As Jason Stanley observes, “when someone tells us that John ate this morning, we assume he did so in the normal way” (2007: 204). He points to such assumptions as clear examples of content that is conveyed pragmatically, not entering into the truth-conditions of the utterance, though he does not offer an account, as I am about to, of how.

58

The pragmatics of substitutivity

The Principle of Implicated Normalcy Speakers generally conversationally implicate that the circumstances regarding whatever they are speaking of are not abnormal in any significant, unanticipated, unindicated way. The basis for this principle is Grice’s first maxim of Quantity, requiring speakers to be adequately informative for the purpose of the talk exchange. This requires preventing misunderstandings that may significantly interfere with the accomplishment of that purpose. Because of the presumption of normalcy, listeners will assume that the circumstances regarding the subject of conversation are normal—unless, of course, it is already apparent or suspected that they are not. Therefore, they will be mistaken with regard to any unanticipated, unindicated abnormalities. So, to the extent that such misunderstandings are significant, a cooperative speaker who is aware of them would attempt to prevent them. Therefore, a speaker who does not indicate significant, unanticipated abnormalities conversationally implicates that there are none. For example, if in the course of recounting the day’s events I mention to my wife, without elaborating, that I saw the dean today, I would thereby conversationally implicate that the dean was alive at the time. For it would be extremely abnormal for me to see someone—especially a colleague— dead, and so I know that my remark about having seen the dean today would lead my wife to surmise that he was alive. So given that the purpose of the conversation was, at least in part, to report the day’s matters of interest, I would be less than adequately forthcoming if I allowed my wife to persist in thinking that he was alive when I knew that he was not. Thus, the assumption that I was observing the Cooperative Principle yields the conclusion that I must have believed the dean was alive, which is thereby conversationally implicated. Note that according to this reasoning, we routinely conversationally implicate a huge number of things that we are not even thinking of. If, for instance, in telling you about my visit to a restaurant I say, “the waiter brought me my hamburger very soon after I ordered it,” I would typically be conversationally implicating that the hamburger was not delivered “encased in a cubic yard of solid lucite plastic so rigid that it takes a jack hammer to bust it open.”7 For if the hamburger had been delivered in such

7.

Searle (1978: 217)—Searle’s words, though his example is different and is used to opposite ends.

Implicated normalcy 59

an unusual way, I would be less than adequately informative were I to withhold such interesting information. This clearly contradicts the assumption that one can conversationally implicate only what one consciously intends to convey. But this assumption is wrong, for two reasons. First, Grice himself seems to think otherwise, as in the example given earlier (1989: 32): A is standing by an obviously immobilized car and is approached by B; the following exchange takes place: (1) A: I am out of petrol. B: There is a garage around the corner. (Gloss: B would be infringing the maxim “Be Relevant” unless he thinks, or thinks it is possible, that the garage is open and has petrol to sell; so he implicates that the garage is open, or at least may be open, etc.)

There is no indication here that B is consciously thinking about what Grice says he implicates. The last word of Grice’s gloss—’etc.’—is especially telling, as it suggests that the list of conversational implicata is open. Grice explicitly acknowledges this (1989: 40): Since, to calculate a conversational implicature is to calculate what has to be supposed in order to preserve the supposition that the Cooperative Principle is being observed, and since there may be various possible specific explanations, a list of which may be open, the conversational implicatum in such cases will be [a] disjunction of such specific explanations; and if the list of these is open, the implicatum will have just the kind of indeterminacy that many actual implicata do in fact possess.

In any case, despite this seemingly decisive evidence, I shall not rely on my intrepretation of Grice’s view.8 It will suffice simply to stipulate that the word ‘implicate’ will be used in such a way that one need not mean or even be explicitly aware of what one implicates. Even if this is not Grice’s idea of implicature, it will suffice for my purposes. Even though the Principle of Implicated Normalcy thus has us conversationally implicating on a massive scale, it is may be useful to keep in mind how the principle is nevertheless restricted. Obviously, speakers seldom (if ever) conversationally implicate that the circumstances regarding whatever they are speaking of are not abnormal in any way at all. A particular normality with regard to what one speaks of is conversationally implicated according to the principle only if the corresponding abnormality, if it ob8.

For detailed exegesis of Grice see Jennifer Saul (2002).

60

The pragmatics of substitutivity

tained, would be significant, unanticipated, and unindicated. It must be signifcant enough, and in the right way, so that the listener’s being mistaken about it would hinder the shared purpose of the conversation. And the maxim of quantity would not require mentioning an abnormality that was anticipated or otherwise indicated.

4. Normalcy for belief ascriptions Belief ascriptions are often acceptable verbatim, in the sense that the person to whom the belief is ascribed would accept that very ascription, word for word, as true. Indeed, I take it as an obvious fact that correct belief ascriptions are normally acceptable verbatim—as long as the reputed believer knows the language of the ascription, as well as enough about the context of the ascription in order adequately to resolve ambiguities and to determine the referents of indexical expressions.9 If I said to you, for instance, (3)

Kripke thinks Gödel was the greatest logician of the twentieth century,

you would normally suppose that if what I said is true, then if you were to go up to Kripke with a transcript of our conversation and ask him, pointing at (3), “Is that so?” he would answer affirmatively (assuming, of course, that he chose to cooperate). Insofar as verbatim acceptability is normal for belief ascriptions, it follows by the Principle of Implicated Normalcy that the utterer of a belief ascription normally implicates that his ascription is acceptable verbatim. In particular, in a normal utterance of (2) 9.

Lois Lane believes Superman is a reporter Arguably, cross-language belief ascriptions (ascriptions in a language not spoken by the reputed believer) are less than fully normal. For in the simplest, most common cases of belief ascription, the ascriber and the ascribee are cognitively much alike, having not merely the same form of life, but speaking the same language (and even having largely the same concepts and beliefs). In any case, I do not mean to rely here on anything more controversial than the platitude that what a person believes corresponds more or less with what he would say—not always straightforwardly, but fairly predictably. A closer look at cross-language ascriptions comes in the next section.

Normalcy for belief ascriptions 61

it would be conversationally implicated that the utterance is acceptable verbatim, i.e., that Lois would accept that very utterance, word for word, as true. For the speaker knows that the listener is inclined to presume that the case at hand is normal, and hence, to infer that Lois would accept the utterance word for word; so, assuming the speaker is being cooperative and, hence, adequately informative, the speaker would prevent the listener from being misled by this presumption; so it must be supposed that the speaker does not think the presumption of normalcy yields any false conclusions in this case, and in particular, that the speaker does not think there is anything wrong with the conclusion that Lois would accept the utterance verbatim; so in uttering (2) the speaker conversationally implicates that the utterance is acceptable verbatim.10 On the other hand, verbatim acceptability is not conversationally implicated in suitably abnormal cases—such as when (2) is uttered in a conversation about how well Superman fools Lois when he disguises himself as Clark Kent (Chapter 1, Section 1). In such circumstances the presumption of normalcy as applied to verbatim acceptability of belief reports is suspended, since the talk of Lois’s being fooled makes it clear that the case is not normal and in particular that we cannot rely on her judgment about sentences such as (2). Hence, there is no danger in such circumstances that the listener will erroneously infer that Lois would accept the report verbatim, so there is no such potential misunderstanding (to the effect that the ascription is acceptable verbatim) that a cooperative speaker would be

10. (a) For ease of exposition I have been discussing the simplest sort of cases, where the believer would rightly be expected to have no trouble interpreting the belief ascription—(2) and (3) are formulated in the language of the believer and have no significantly ambiguous or indexical expressions. If, on the other hand, I say to you She thinks you are light, we would expect the reputed believer to accept this report only if she knew the referents of the indexicals ‘She’ and ‘you’ and the relevant sense of the ambiguous word ‘light’; so what would be conversationally implicated would be the more qualified claim, that she would accept the ascription were she duly informed of the referents of the indexicals and the relevant sense of the ambiguous expression. (b) Although the present discussion is only about belief, ascriptions of other propositional attitudes—doubt, fear, etc.—seem also to be normally acceptable verbatim and, hence, amenable to similar treatment.

62

The pragmatics of substitutivity

obliged to prevent; so the normal conversational implicature about verbatim acceptability does not arise.11 Thus, the inclination to reject (2)—when there is such an inclination— arises not because (2) itself is false, but because the utterance of (2) would be inappropriate, because it would carry the false conversational implicatum that (2) is acceptable verbatim.12 And in the special cases where we are 11. One might object that talk about Lois’s being fooled is not sufficient to cancel the implicature. Suppose all the theorems attributed to Godel were actually proved by Schmidt, whom Godel plagiarized. Then in talking about how well Godel has fooled everybody—even Kripke—one might say (3) Kripke thinks Gödel was the greatest logician of the twentieth century. But here, even in the context of talking about Kripke’s being fooled, the normal implicature is not cancelled; we would still expect Kripke to accept (3) verbatim. The difference between these two cases of talk about someone being fooled is that in Lois’s case it is clear that she is being fooled not only into thinking Superman is a reporter, but also into thinking that the besuited reporter she is looking at is named ‘Clark Kent’ and not ‘Superman’—as a result of which we would not expect her to accept the ascription (so the normal implicature is cancelled by the context). On the other hand, Kripke is being fooled only into thinking that Godel proved the incompleteness of arithmetic—he’s not being fooled about the name ‘Godel’ in a way that would call into question his use of it; so there is no reason for us to doubt his assessment of the ascription (so the normal implicature is not cancelled). 12. Jennifer Saul agrees that the Fregean intuition that (2) is false is mistaken, but she insists that this is not due to any confusion between semantic content and conversational implicata. Rather, she suggests that the inclination to reject (2) comes from having good evidence against it. Such evidence could be that Lois says things like ‘Clark Kent is a reporter but Superman isn’t’, and that she asks Clark Kent, be-suited, if he wrote yesterday’s Superman story, while only asking Superman, be-caped, if he read it. “We will simply have more confidence” (2007: 163), she explains, in our judgment of (2)’s falsity on the basis of this evidence, than in any inferences to the contrary on the basis of any other evidence. But why would we take this evidence as such strong evidence against (2)? For one thing, this kind of evidence owes its strength to the normal cases, which do not involve the kind of confusion that makes Lois’s case interesting; knowing as we do of her mistaken belief that the two names are not coreferential, we should be less impressed by this evidence than we normally would be. And in any case, to say that we take this evidence as weighing against (2) amounts merely to saying that we think that (2) is false under these circumstances—so the claim that we reject (2) because we

Variations on verbatim acceptability 63

inclined to accept (2)—despite Lois’s disposition to reject it—it is because there is no significant danger of the listener’s being misled into thinking that the ascription is acceptable verbatim.13

5. Variations on verbatim acceptability The particular case of (2)

Lois Lane believes Superman is a reporter,

is somewhat special, of course, since Lois knows the language in which her belief is being ascribed. Sometimes, however, we ascribe beliefs to individuals who clearly do not understand the words we use, and hence, for whom the ascription clearly cannot be acceptable verbatim—in which case the ascription would not be conversationally implicated to be acceptable verbatim. Nevertheless, the treatment of (2) can be adapted to such cases on the basis of similar conversational implicata.14 For instance, consider a context in which we would be inclined to accept are confident that we have strong evidence against it is just a reformulation of the Fregean intuition, not an explanation of it. 13. If Recanati is right that “Implicature Theorists typically invoke a conversational principle or maxim of faithfulness” (1993: 332)—e.g., “In reporting someone’s belief, try as much as possible to use the words the believer would use” (not Recanati’s formulation)—then apparently my account is not typical, since I do not invoke any such principle. Rather, the principle I invoke is the Principle of Implicated Normalcy, which is derived from Grice’s first maxim of Quantity. The principle of faithfulness, on the other hand, is not derivable from Grice’s maxims (pace Recanati (1993: 333))—and with good reason, as speakers do not in fact generally follow it (as seen in the earlier example where (2) is uttered in a conversation about how well Superman fools Lois when he disguises himself as Clark Kent (Chapter 1, Section 1)). 14. Thus, it should be clear that I do not claim that whenever substitution seems to fail in a belief ascription, it is because the utterance of the ascription conversationally implicates that the ascription is acceptable verbatim to the purported believer. More generally, as will become more evident later in this section, I do not claim that there is any unique conversational implicatum that is responsible for every case of apparent substitution failure in a belief ascription. In any such case there will be a variety of false implicata, not all having to do with verbatim acceptability.

64 (1f)

The pragmatics of substitutivity

Lois Lane croit que Clark Kent est un journaliste

but reject (2f)

Lois Lane croit que Superman est un journaliste.

Assuming Lois Lane does not know French, neither ascription would be acceptable verbatim; and assuming further that Lois’s ignorance of French is common knowledge among the conversants, utterances of the ascriptions would not conversationally implicate that they are acceptable verbatim. Rather, something that would be conversationally implicated (in the given sort of context) by utterances of (1f) and (2f), would be that (1)

Lois Lane believes Clark Kent is a reporter

and (2), respectively, are acceptable verbatim. For when ascribing beliefs to someone who does not speak the language of the ascription, what is normal is that the person would accept verbatim not the ascription itself, but an appropriate translation of it into that person’s own language. (The appropriateness of a translation depends, of course, on the context; and in the given context (1) and (2) are appropriate translations of (1f) and (2f), respectively, given that the names ‘Superman’ and ‘Clark Kent’ are each used the same in French as they are in English.) A slightly more complicated case would be where Lois has not yet learned the names ‘Superman’ and ‘Clark Kent’, but nevertheless has already come to believe that there are two individuals, one of whom wears a cape, leaps tall buildings in a single bound, etc., the other of whom wears a suit, works at the desk next to hers at the “Daily Planet,” etc. Knowing this we might be inclined to reject (2) and (2f) on the grounds that, say, (2d) Lois believes the guy who wears a cape and is able to leap tall buildings in a single bound is a reporter would not be acceptable verbatim—in which case the relevant false conversational implicatum accounting for our rejection of (2) and (2f) would be that (2d) is acceptable verbatim (given that (2d) would be an appropriate reformulation of (2) and (2f) in Lois’s idiolect). These considerations point to a general strategy for adapting the treatment of (2) to cases where the individual to whom belief is ascribed does

Variations on verbatim acceptability 65

not understand the words used in the ascription. Suppose that two belief ascriptions differ only in which of two coreferential names occur in the embedded sentence, and that we are inclined to accept one, (1g) x believes N is F, but to reject the other, (2g) x believes M is F. Clearly our reasons for doing so must have something to do with the purported believer, x, and the two different names, M and N. In the simplest case it is that x herself uses these names differently, accepting one of the ascriptions while denying the other, even though speakers do not normally distinguish between coreferential names in this way. In other cases, where x does not understand all the words used in the ascriptions, the general strategy will be to identify whatever it is about x’s behavior in virtue of which we find (2g), but not (1g), objectionable, and take that as what is conversationally implicated by an utterance of (2g) in accordance with the Principle of Implicated Normalcy. These implicata might be about x’s inclinations with regard to various translations or reformulations of the relevant ascriptions, as in the cases of (2f) and (2d), respectively, or they might even be about nonverbal behavior. For instance, as a sort of limiting case, suppose that Rover, a dog, displayed behavior in virtue of which we would accept (1r)

Rover thinks Superman is his friend

but reject (2r)

Rover thinks Clark Kent is his friend;

suppose, say, that Rover consistently wags his tail and so on in the presence of Superman dressed in the Superman outfit, but consistently growls and so on in the presence of Superman dressed in the Clark Kent outfit. Here there is obviously no room at all to talk about what belief ascriptions Rover might accept or reject. Nevertheless, generalizing from the treatment of the Lois Lane case, and assuming that normal friendly dog behavior includes tail-wagging but not growling, we can account for the unacceptability of

66

The pragmatics of substitutivity

(2r) as deriving from the falsity of the conversational implicatum that in the presence of Superman in his Clark Kent outfit Rover would wag his tail and not growl. Jennifer Saul (1998: 378) raises doubts about whether this sort of strategy can work where a generalized belief ascription adverts at once to a whole range of individuals with possibly widely varying linguistic capacities.15 Suppose, for instance, that Superman in his Superman outfit is observed entering a phone booth by a crowd of spectators including the following: (i) English speakers thinking the names ‘Superman’ and ‘Clark Kent’ are not coreferential; (ii) non-English speakers familiar with names in the language they speak which correspond to the names ‘Superman’ and ‘Clark Kent’ as they are used in English and which they think are not coreferential; (iii) speakers not familiar with the names ‘Superman’ and ‘Clark Kent’, or with corresponding names in another language, but who think there are two distinct individuals, one having properties we standardly associate with the name ‘Superman’, and the other having properties we standardly associate with the name ‘Clark Kent’; and (iv) nonspeakers who nevertheless believe there are two distinct individuals as described in the previous clause. Thus one might well accept (a) The spectators think Superman is in the phone booth without accepting (b) The spectators think Clark Kent is in the phone booth. But if this difference between (a) and (b) is because the utterance of (b), but not of (a), carries a false conversational implicatum, what exactly would that implicatum be? It could not be simply that the spectators would accept the report verbatim, since that would be false for (a) as well, given the presence of spectators who do not speak English altogether. In fact, given the varied linguistic capacities of the spectators, no single one of the various methods described above will be sufficient, since each method is aimed at a particular class of reputed believers defined by certain aspects of the

15. Saul considers utterances of ‘No ancient astronomers believed that Hesperus is Phosphorus’; but since I will take up identity beliefs and negations of belief ascriptions only in the next section, I will frame the discussion here in terms of a different example.

Variations on verbatim acceptability 67

extent of their linguistic knowledge (corresponding to the subgroups of spectators specified in clauses (i)-(iv)). The key to answering this question lies in recognizing that normalcy is relative—what is normal for Lois Lane’s believing something is not what is normal for Rover’s believing it. And so, what the utterance of a belief ascription conversationally implicates in accordance with the Principle of Implicated Normalcy will also be relative—one thing for Lois, quite another for Rover. Thus, the relevant implicatum of (b) might be formulated as follows: (b*) For any spectator s: (i) if s speaks English, s would accept the report verbatim; (ii) if s does not speak English but does speak another language having names which correspond to the names ‘Superman’ and ‘Clark Kent’ as they are used in English, then s would accept verbatim an appropriate translation of the report into that language; (iii) if s speaks only a language without names corresponding to the names ‘Superman’ and ‘Clark Kent’, but thinks there are two distinct individuals, one having properties standardly associated with the name ‘Superman’ and the other having properties standardly associated with the name ‘Clark Kent’, then s would accept verbatim a suitable reformulation of the report in s’s language in terms of any description standardly associated with the name ‘Clark Kent’; (iv) if s speaks no language at all but nevertheless displays two distinct patterns of behavior—say, S-behavior and C-behavior—in virtue of which we would say s believes there are two distinct individuals, each having properties standardly associated with the names ‘Superman’ and ‘Clark Kent’, respectively, then s would display Cbehavior with regard to whoever is in the phonebooth; and so on. Lest (b*) seem too elaborate to be conversationally implicated, it should be noted first that (b*) can easily be broken down into more manageable pieces, which can be viewed as distinct conversational implicata; and each clause of (b*) is easily derivable in the way explained earlier in each of the corresponding cases. Moreover, recall Grice’s point about the openness and indeterminacy of conversational limplicature (1989: 40): Since, to calculate a conversational implicature is to calculate what has to be supposed in order to preserve the supposition that the Cooperative Principle is being observed, and since there may be various possible specific explanations, a list of which may be open, the conversational implicatum in

68

The pragmatics of substitutivity such cases will be [a] disjunction of such specific explanations; and if the list of these is open, the implicatum will have just the kind of indeterminacy that many actual implicata do in fact possess.

Conversational implicature is notoriously open-ended, sometimes extending well beyond what the speaker may be prepared to articulate; just think of all that I might implicate by saying, “She is his Juliet.” So we should not assume that the speaker must always be able to provide a handy, simple formulation of all that is conversationally implicated in a given utterance. Nothing like (b*) has to be in anybody’s head; it is enough for the conversants to accept that the utterance of (b) would lead them to expect the kind of consequences elaborated in (b*). In any case, (b*) is really nothing more than one way of spelling out part of the putative notional or de dicto sense of (b). Although such a putative sense is not really another full-fledged semantic content of the sentence (b), alongside the relational or de re sense (as argued in Chapter 1, Section 3), it certainly might be an additional interpretation of an utterance of (b). If so, then another way to generalize the treatment of (2) to cases such as (b) would be by simply characterizing the relevant conversational implicatum of (b) as the notional interpretation of the utterance (what is said being the relational interpretation).16 In the most ordinary cases, such as (2), the most salient feature of the notional interpretation is verbatim acceptability; in less ordinary cases, such as (2f), (2d), and (2r), it is the various kinds of variation on verbatim acceptability suggested above. The dialectical point here is worth restating and clarifying. My argument with those who offer semantic analyses of “notional” belief is not primarily about the details of notional belief (although I certainly do have my reservations about the appeal to anything like modes of presentation), but about whether those details concern the semantic content of belief ascription sentences, or the pragmatic import of utterances of those sentences. Just as I agree with Searle that part of what we should understand 16. This seems to be where Saul ends up, eventually proposing that “an utterance of [a] belief-reporting sentence, A believes that S, will standardly carry the implicature that A believes that S under a guise … the speaker would take … to be … appropriate … for the purposes of her utterance (upon being sufficiently informed)” (1998 383). As plausible as this claim may be, it is vulnerable to the same objection that Saul herself raises against another proposal, namely, “there isn’t much of an explanation of the nature of guises” (1998: 373)—nor, I would add, of the ternary notion of believing under a guise (as I argued against the indexical view, Chapter 1, Section 4.1).

Variations on verbatim acceptability 69

from a typical utterence of the sentence ‘the waiter brought me my hamburger’ is that the hamburger was not encased in a cubic yard of solid lucite plastic, yet disagree with him about whether this enters into the truthconditions of what was said by uttering the sentence—so, too, I agree that often what we should understand from an utterance of (2) is that Lois would accept (Richard (1990)) or endorse (Chalmers (2011)) some sentence or proposition or representation that is related to the embedded sentence in a contextually suitable way, but I do not agree that this condition belongs to the (or a) semantic content of the sentence (2). What philosophers since Frege have tried to codify as belonging to the essence of belief (de dicto), I propose to treat as what is merely normal, but not essential, for belief (de re). An advantage to my approach is that I can settle for ad hoc, case-by-case accounts, in terms of whatever conversational implicata might be relevant in particular cases, without having to spell out any general, allinclusive characterization of some single, coherent concept of notional belief. Indeed, I take the elusiveness of the latter as evidence in support of my position. A different but ultimately equivalent way of accommodating ascriptions of belief to individuals who cannot understand them would be by saying that what is normally conversationally implicated in such cases is that the belief ascription is counterfactually acceptable verbatim, in that the ascription would be acceptable verbatim if the purported believer did understand it. (Of course, understanding two coreferential names would not require knowing that they are coreferential; it would be enough to have them both in one’s lexicon.) Thus, our inclination to reject (2f)

Lois Lane croit que Superman est un journaliste

would be because its utterance conversationally implicates that it would be acceptable verbatim if Lois knew French; and our inclination to reject (2r)

Rover thinks Clark Kent is his friend

would be because its utterance conversationally implicates that it would be acceptable verbatim if Rover spoke English; and so on.17 Clearly this ap17. This is not to be confused with a counterfactual semantic analysis of belief ascriptions, according to which a belief ascription in a language L is true if and only if it would be acceptable verbatim if the reputed believer knew L. The proposal is only that the counterfactual is conversationally implicated by

70

The pragmatics of substitutivity

proach leads to hard questions about the semantics of counterfactuals—a topic well beyond the scope of this work. But on the plausible assumption that any satisfactory theory of counterfactuals would analyze the relevant counterfactuals along the lines of (b*), the two approaches are equivalent.

6. Identity beliefs Since my treatment of (2)

Lois Lane believes Superman is a reporter

does not rely on any peculiarities of the embedded sentence, it generalizes straightforwardly: belief ascriptions do not violate the principle that names of the same thing can be substituted for each other salva veritate; when such substitution seems to fail, it is only because the names are not interchangeable salvis implicatis.18 This applies in particular to the sort of cases involving identity claims with which Frege began this discussion. For example, since (4)

Lois Lane believes Superman is Superman

is surely true, and since Clark Kent is Superman, it follows that (5)

Lois Lane believes Clark Kent is Superman

is true. Of course, (5) may well seem false, but this is only because in uttering it we would normally implicate something false—viz., that it is acceptable verbatim. And since (5) is true, its negation (6)

Lois Lane does not believe Clark Kent is Superman,

as well as the stronger and more colloquial an utterance of the belief ascription, not that it is equivalent to the ascription. (So the truth of the counterfactual does not entail the truth of the ascription.) 18. The distinction between substitution that preserves semantic content and substitution that preserves pragmatic import was the subject of my “Two Kinds of Substitutivity” (1979).

Identity beliefs 71

(6')

Lois Lane doesn’t realize that Clark Kent is Superman,

must be false. Of course, (6) and (6') may well seem true. Indeed, it would be natural and fitting to utter (6') in telling the story and in explaining Lois’s behavior (see Section 3.2). But the acceptability of utterances of (6) and (6') is easily accounted for pragmatically, without the supposition that the sentences are actually true. One reason these sentences seem true is simply that in uttering them we would normally implicate something that is true, viz., that (6) is acceptable verbatim.19 Moreover, they are the negations of sentences that are typically used to convey—and hence, that seem to express—something false. Since an utterance of (5) would typically carry the conversational implicatum that (5) is acceptable verbatim, this proposition (or something like it) is naturally taken (albeit incorrectly) to be the semantic content of (5)—to the point that the negation of this proposition would naturally be conveyed by the negation of (5), i.e., (6). Thus, the case at a hand is an instance of the more general, familiar phenomenon of inherited pragmatic content. Pragmatic content—what is conveyed in an utterance without necessarily being said—can (sometimes) be inherited, in the following way: if an expression that can be used with a certain pragmatic content is embedded in a compound utterance, the pragmatic content of the compound utterance may incorporate the pragmatic content—rather than the semantic content—of the embedded part.20 Inherited pragmatic content is perhaps most easily recognizable in metaphorical discourse. Suppose, for instance, I use the sentence (7)

Ilana is a real drill sergeant!

to indicate that Ilana is very strict. A listener understanding what I meant might express disagreement by using the negation of (7): (7')

Ilana is not a real drill sergeant!

In uttering (7') my interlocutor would clearly mean not the negation of (7) (that is, of its semantic content), but rather, the negation of what I meant in 19. Of course, the factitivity of ‘realize’ precludes the verbatim acceptability of any ascription of the form ‘X doesn’t realize that P’. 20. A formal account of pragmatic inheritability might follow the lines of Peter Pagin and Francis Jeffry Pelletier’s account of what they call “modulation” (2007: 46–50).

72

The pragmatics of substitutivity

uttering (7)—its pragmatic content. This could be the case even if my interlocutor believed that (7) is literally true; its semantic content would be irrelevant. Likewise, the pragmatic content conversationally implicated in the utterance of a belief ascription may be inherited by utterances in which the belief ascription is embedded. Thus, since (5) may be uttered with the pragmatic content that the ascription is acceptable verbatim (or, say, that Lois thinks the names ‘Clark Kent’ and ‘Superman’ are coreferential), its negation, (6), may be used to negate (pragmatically, not literally) this pragmatic content of (5), rather than its semantic content.21 Mitchell Green objects that my treatment of belief ascriptions cannot advert to the inheritability of pragmatic content without providing an explicit criterion distinguishing cases where pragmatic content is inherited from cases where it is not (1998: 79). A quick look suggests that the cases where pragmatic content is inherited are those where the pragmatic content is what is commonly heard as “what is said”—something I have characterized as “what is said loosely” (2002: 357)—including, inter alia, metaphorical interpretations and generalized conversational implicata, as well as pragmatic contents resulting from “pragmatic ellipsis” (where the words uttered are only part of a longer sentence that literally expresses what the speaker wishes to say), or from what Recanati calls “free enrichment” (1993: 261), or from what relevance theorists call “explicature” (Dan Sperber and Deirdre Wilson (1986: 182) and Robyn Carston (2002: 116)).22 Nice as it would be to formulate and justify a precise criterion for the inheritability of pragmatic content, it is enough for my treatment of belief ascriptions to observe the fact that the relevant pragmatic contents of belief ascriptions are indeed inherited, even if the question of why these, but not all pragmatic contents, are inherited is left unanswered. What reason might there be for thinking that my position incurs the burden of providing such a criterion? Green does not say. The demand would 21. This position is to be contrasted with any proposal of “metalinguistic negation” according to which inherited pragmatic content actually enters into the semantic content of the sentence uttered. See Recanati (1993: 273, 344) and Cohen (1971: 58). 22. Of course, if one simply assumes to begin with that what is said must be what is commonly heard as being said, then given that what is implicated is not said, it follows that nothing that is commonly heard as what is said can be analyzed as conversationally implicated. But starting from such an assumption merely begs the question. (Indeed, I would take this inference as a reductio ad absurdum of the initial assumption.)

Identity beliefs 73

be more understandable were I relying some way on such a criterion. Suppose I argued thus: (a) (b) (c)

Pragmatic contents are inherited iff they have property P. The conversational implicata that I attribute to belief ascriptions have property P. The conversational implicata that I attribute to belief ascriptions are inherited.

Then it would be easy to see how I would be obligated to spell out and justify the first premise. But I make no such argument. In fact, I think of (c) as an observation, not a conclusion—or if as a conclusion, than as a conclusion merely from observations about what we typically mean in uttering various sentences, not a conclusion from anything like (a). Therefore, I do not any see any justification for Green’s demand. A more direct objection is the “scope argument,” due initially to L. J. Cohen and formulated here by François Recanati (1993: 271-272, with reference to Anscrombe and Ducrot): (a) (b)

(c)

(d)

(e)

Conversational implicatures are pragmatic consequences of an act of saying something. An act of saying something can be performed only by means of a complete utterance, not by means of an unasserted clause such as the antecedent of a conditional. Hence, no implicature can be generated at the sub-locutionary level, i.e. at the level of an unasserted clause such as the antecedent of a conditional. To say that an implicature falls within the scope of a logical operator is to say that it is generated at the sub-locutionary level, viz. at the level of the clause on which the logical operator operates. Hence, no implicature can fall within the scope of a logical operator.

Presumably this shows that conversational implicata cannot be inherited in the way I suggest.23

23. “Presumably,” because (d) might be just a stipulative definition, in which case (e) might not amount to anything contradicting my position. (And if (d) is a substantive claim, it is not clear what it states.)

74

The pragmatics of substitutivity

The main problem with this argument is that what follows from (a) and (b) is not (c) (at least not taken in a way that makes the rest of the argument work), but rather (c')

Conversational implicatures are generated only by complete utterances (no conversational implicature can be generated by an unasserted clause).24

But (c') does not prevent complex utterances from inheriting the pragmatic contents that could have been conversationally implicated by utterances of their parts. Recall how the Gricean listener works out what is conversationally implicated—he asks himself (as it were), “What must the speaker be thinking, given that he uttered S and is observing the Cooperative Principle?” Now, if S is a complex sentence, such as a conditional or a negation, he might wonder about a part of it, “Why did the speaker formulate the antecedent [or express what was negated] with those particular words?” In addressing such questions the listener will certainly need to take into account the complete utterance—the fact that the part in question is being negated, or being used as the antecedent with regard to a certain consequent, will help him interpret the listener. And a plausible strategy for interpreting those utterance parts is simply to consider what pragmatic contents they would have had if uttered on their own—especially if they typically have those pragmatic contents.

24. Other problems with the argument are, first, that premise (a) as it stands is not quite right, since Grice’s definition of conversational implicature also applies to acts of merely “making as if to say.” Secondly, regardless of how the original definition was formulated, it can easily be expanded in natural ways so as to apply to all conversational contributions, not just complete acts of saying (or making as if to say). This includes both non-indicative speech acts, such as questions and commands, as well as parts of speech acts, such as referring. Moreover, as Green notes, “Nothing in the Gricean apparatus precludes calculation of what a speaker could have intended to convey in uttering a clause that is not put forth with any illocutionary force” (1998: 87, n. 40, emphasis mine).

Availability 75

7. Availability François Recanati rejects the kind of account I have given of substitution failure, primarily on the grounds that it violates the Availability Principle: In deciding whether a pragmatically determined aspect of utterance meaning is part of what is said, that is, in making a decision concerning what is said, we should always try to preserve our pre-theoretic intuitions on the matter. (1993: 248)

Assuming that our pre-theoretic intuitions about belief ascriptions are that what I have construed as mere conversational implicata are actually part of what is said, Recanati points to the Availability Principle as ruling out my account. Recanati derives the Availability Principle from the Availability Hypothesis: we have distinct conscious representations for ‘what is said’ and ‘what is implicated’ by a given utterance: both are consciously accessible, and are consciously accessible as distinct.25 (1993: 245)

And Recanati argues for the Availability Hypothesis on the basis of two premises: (i) speakers and hearers must be capable of working out what is conversationally implicated, and (ii) working out what is conversationally implicated requires consciously grasping what is said and what is implicated (1993: 245). The first premise is Recanati’s interpretation of Grice’s observation that conversational implicatures must be calculable: “The presence of a conversational implicature must be capable of being worked out” (1989: 31). Recanati interprets this as meaning that the speaker and the hearer—not just the theorist—must be capable of working out the implicatures (1993: 245).

25. Cf. (1993: 248). More recently Recanati (2004: 20) characterizes availability in terms of the following “constraint:" Availability What is said must be intuitively accessible to the conversational participants (unless something goes wrong and they do not count as ‘normal interpreters’). Assuming that Recanati does not mean to distinguish between intuitive accessibility and conscious accessibility, this constraint amounts to the half of the Availability Principle that concerns just what is said, not what is implicated.

76

The pragmatics of substitutivity

But Grice’s own explanation of the calculability requirement suggests otherwise (1993: 31): The presence of a conversational implicature must be capable of being worked out; for even if it 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.

Clearly, Grice infers the calculability requirement from the distinction between conversational implicature and other kinds of implicature, which he formulates as follows: A man who, by (in, when) saying (or making as if to say) that p has implicated that q, may be said to have conversationally implicated that q, provided that (1) he is presumed to be observing the conversational maxims, or at least the Cooperative Principle; (2) the supposition that he is aware that, or thinks that, q is required in order to make his saying or making as if to say p (or doing so in those terms) consistent with this presumption; and (3) the speaker thinks (and would expect the hearer to think that the speaker thinks) that it is within the competence of the hearer to work out, or grasp intuitively, that the supposition mentioned in (2) is required. (1989: 31, last emphasis mine)

Since Grice explicitly allows that it be in the competence of the hearer merely to grasp intuitively the requisite connection between the implicatum and the Cooperative Principle, the calculability requirement he infers from this distinction could not be that the speaker and hearer themselves must actually be capable of articulating this connection. Rather, the demand that the intuition be replaceable by an argument—presented as the difference between conversational implicature and conventional implicature—requires only that the presumption that the Cooperative Principle is being followed could serve as grounds for inferring that the speaker must believe the implicatum.26 26. (a) According to Recanati, “Grice clearly had in mind the participants in the talk-exchanges themselves” (1993: 245) as those who must be capable of working out conversational implicatures. But Grice explicitly acknowledges the fact that theorists not lacking in ordinary linguistic competence nevertheless disagree in many cases about the presence of conversational implicatures (1989: 49). (b) Cf. Michael Thau, who argues against Recanati’s Availability principle on the grounds that our knowledge of what is conversationally implicated comes from some noninferential ability, and that we engage in linguistic communica-

Availability 77

Even if it were true that conversational implicata must be capable of being worked out by the speakers and hearers themselves, it is not true that working them out requires consciously grasping what is said. For one thing, it is natural to suppose (as does Recanati) that many cognitive processes are not entirely conscious, and there is no prima facie reason to suppose that the calculation of what is implicated is not such a process—especially in light of Grice’s remarks (above) about intuitive grasping. (Our semantic knowledge could be just as implicit as our syntactic knowledge, for instance.)27 In any case, it is clear from Grice’s definition that what is actually said need not be part of the calculation, as conversational implicata can be derived from what the speaker merely makes as if to say. That is, what is said is not an essential input for the working out of what is conversationally implicated; what one merely makes as if to say can serve as well. So, working out what is conversationally implicated need not involve any recognition, conscious or otherwise, of what is actually said.28 A different argument for the Availability Hypothesis emerges in Recanati’s discussion of how what is said fits into the broader picture of “what is communicated” (1993: 247): I suggest that we consider ‘what is communicated’ as simply a name for the level at which we find both what is said and what is implicated—the top level, characterized by conscious accessibility . … On this view, the conscious availability of what is said no longer is a mystery: if what is commu-

tion in accordance with the conversational maxims without necessarily thinking of them (2002 Chapter 4, Sections 3–7). 27. Recanati has more recently relaxed the demand for conscious accessiblility of what is said and what is implicated (premise (ii) in support of the Availability Hypothesis), allowing that the inference of what is conversationally implicated from what is said may be tacit, as long as “the subject herself has the reflective capacities for making the inference explicit” (2004: 50). But he has still not shown why conscious accessibility, even in this weaker form, is required. 28. Elsewhere Recanati (2004: 19) has adopted a broader sense of ‘says’, including what Grice referred to as what one merely makes as if to say : S says that p, in the broad sense, iff he either says that p (in the strict sense) or makes as if to say that p (again, in the strict sense of ‘say’). On the assumption that what one makes as if to say includes what one intuitively seems to say, the broader sense of ‘says’ renders the Availability Principle trivial and consistent with my account of substitution failure.

78

The pragmatics of substitutivity nicated, which is consciously accessible, consists of what is said and what is implicated, then what is said cannot but be consciously accessible.

Here the Availability Hypothesis seems to be supported by the following reasoning: What is communicated is consciously accessible. What is said is part of what is communicated. What is said is consciously accessible. Taken at face value, according to the ordinary meanings of the words, neither of these premises seems true. Speakers are often not aware of all that they communicate—as is so often the case when a speaker is misunderstood. And what one communicates does not always include what one literally says—as in the case of sarcasm.29 Perhaps, though, what should be taken at face value is not these premises, but the language of Recanati’s suggestion, which appears to be a stipulative definition of the expression ‘what is communicated’. (“I suggest that we consider ‘what is communicated’ as simply a name for …”) If this expression is defined as applying only to something consciously accessible, then the first premise is trivial; likewise, if it is defined as applying in part to what is said, then the second premise is trivial. Now, although one can surely choose to use words however one wants, stipulative definitions should at least be consistent—one must not define something as having properties that are incompatible. In particular, “what is communicated” cannot be defined as being both (a) consciously accessible and (b) inclusive of what is said, unless what is said is consciously accessible. But then since this is what is at issue, the premises beg the question. In light of all these problems with the various reasons offered in support of it, perhaps the Availability Hypothesis should be thought of not as the conclusion of any argument, but as a genuine hypothesis. As such, it must be a hypothesis about “what is said” either (a) in the Gricean sense, or (b) in some other, non-Gricean sense. Now, as a claim about what is said in the Gricean sense, the Availability Hypothesis does not seem plausible, not 29. Recanati believes it is because of such cases that Grice defines conversational implicature in terms of saying or making as if to say, and this is why Recanati stipulates the broader sense of ‘says’ described in the previous note. I interpret Grice’s “making-as-if-to-say” option as allowing conversational implicata to be derived from what one says loosely, not just what one says strictly.

Availability 79

only because of Grice’s own explicit remarks (above, especially note 26), but also because of the broader context of Grice’s program. For Grice developed the theory of conversational implicature with the express purpose of showing that sometimes what is actually said is not what some would suppose (1989: “Prolegomena”); and if that is so, then what is said is not always consciously accessible as such. On the other hand, as a claim about what is said in some other, non-Gricean sense, the Availability Hypothesis would simply not be relevant (in a discussion of what is said in the Gricean sense).30 The main problem with Recanati’s appeal to the Availability Principle is that the principle itself is either too weak or too strong. Here are two ways of taking the Availability Principle: Weak AP Theories about what counts as part of what is said should agree with our intuitions as much as possible. Strong AP Pre-theoretic intuitions of competent speakers about what it is that is said must always be preserved; i.e., what is said must be whatever competent speakers would say is said. Weak AP is just a restricted version of the general methodological platitude that philosophical theories should agree with our intuitions as much as possible. This is a trivial corollary of the view (which I share) that philosophical theories are supposed to explain our intuitions. This does not mean, however, that pre-theoretic intuitions are holy. Pre-theoretic intuitions may be revised or discounted if they are outweighed by other considerations or 30. One might object here that both Recanati and Grice mean to be talking about “what is said” in the ordinary sense. However, there are many ordinary senses of ‘what is said’ (some of which are nicely enumerated by Robert Stainton, 2006: 224–225). In developing his theory of conversational implicature, Grice explicitly intended to speak of what is said in just one particular sense, to be understood from the context of the exposition of his theory (1989: 24–25, 118). So, although he may be understood as speaking of what is said in an ordinary sense, it must be an ordinary sense which is compatible with what he says about it, since it is only what he says about it that determines which sense it is. In particular, the relevant sense of ‘what is said’ is apparently not one in which the Availability Hypothesis holds.

80

The pragmatics of substitutivity

can be suitably accounted for. In particular, as long as the Gricean account of substitution failure (as laid out above, Sections 1-6) is accompanied by an acceptable account of whatever common intuitions it fails to preserve, Weak AP does not rule it out. So, although Weak AP is eminently plausible, it is not strong enough to carry the burden Recanati tries to place on it.31 Moreover, the natural conclusion to be drawn from the Availability Hypothesis is not Weak AP, but Strong AP. For if we always have distinct conscious representations for what is said—if what is said must be consciously accessible as such—then a competent speaker’s intuitions about what it is that he said must always be right. Understood this way, as Strong AP, the Availability Principle could indeed serve the purpose to which Recanati puts it in his rejection of the Gricean account of substitution failure.32 But Strong AP is too strong to be plausible, because (as I shall show in the following section) it is incompatible with the familiar fact that competent speakers sometimes disagree about what is said in a given utterance.33 So, if the Availability Principle is to fulfill the dialectical role Recanati assigns to it—as both a consequence of the Availability Hypothesis 31. Weak AP could, of course, be the basis for an objection to the Gricean account of substitution failure, given that the Gricean account does not agree with all our intuitions. However, assessments of how well a theory agrees with our intuitions must be made holistically, taking into account all our intuitions about all the theory’s consequences. In particular, the intuitivity of consequences regarding the truth conditions of particular belief ascriptions must be considered together with the intuitivity of consequences regarding the notions of ambiguity, indexicality, indeterminacy, etc.—as discussed in the previous chapter. 32. (a) Recanati needs Strong AP elsewhere as well, as in his rejection of “Minimalism” (1993: 245, 256). (b) Further evidence for the Strong AP interpretation of the Availability Principle is provided by Recanati’s more recent paraphrase of it: “‘what is said’ must be analyzed in conformity to the intuitions shared by those who fully understand the utterance—typically the speaker and the hearer, in a normal conversational setting” (2004: 14). 33. Strong AP also seems to rule out Recanati’s own analysis of belief ascriptions. For according to Recanati, the that-clause in a belief ascription in which a singular term seems to occur opaquely refers to an enriched quasi-singular proposition. But, whatever pre-theoretic intuitions competent speakers may have about what is said, surely these do not include intuitions about quasisingular propositions. Cf. Schiffer (1992: 512; 1995: 101).

Availability 81

and also a premise for the rejection of the Gricean account of substitution failure (and of Minimalism, n. 32)—then it must be too strong to be acceptable; but if it is taken in a way that renders it reasonable, then it is too weak to yield the conclusions Recanati would like to draw from it. It is interesting that despite Recanati’s views about “what is said,” he explicitly allows for semantic products that are not “available” to conscious apprehension. In particular, he refers to what I have characterized as “what is said” in the utterance of an indicative sentence—the proposition expressed by the sentence relative to a particular way of resolving any ambiguities in the sentence and of determining the referents of any (explicit) indexicals—as “the primary semantic value” (1993: 263, 265) of the sentence.34 And he readily grants that these primary semantic values may enter into unconscious cognitive processes (e.g., as the “input proposition” for the “primary pragmatic process” of “enrichment,” (1993: 243, 260). It could be, therefore, that Recanati’s objection to the Gricean account of substitution failure is only terminological (as suggested also by the apparently stipulative nature, noted above, of Recanati’s definition of ‘what is communicated’). For whatever I have claimed about “what is said” in belief ascriptions could be reformulated as a claim about their “primary semantic values.” And whatever Recanati claims about “what is said” could be reformulated in the terminology I have been using as a claim about “what competent speakers would commonly say ‘has been said.’” Moreover, if “conversational implicature” is restricted by definition so that conversational implicata must be derivable by competent speakers from what they would say has been said, then the use of the term can be extended—or a different term can be adopted—to apply in such a way that “conversational implicata*” may be derivable (in the same fashion) merely by theorists, from merely primary semantic values.35 34. Elsewhere he calls this “the minimal proposition expressible by the sentence” (1993: 243), “the input proposition” (1993: 243), “the basic level interpretation” (1993: 315, 318), and “the proposition Normally expressed” (1993: 318). 35. Indeed, Recanati has subsequently come to allow that ‘what is said’ might have two senses, one requiring conscious accessibility—in accordance with his own preferred usage—and the other corresponding to my own usage and not requiring conscious accessibility. And I have made a distinction along similar lines, between strict semantic content and loose semantic content (2002). While granting that the notion of “what is said” corresponding to my preferred usage (as strict semantic content) “makes sense,” Recanati main-

82

The pragmatics of substitutivity

8. Semantic intuitions Recanati’s reliance on the Availability Principle reflects a common, general assumption about the role of intuitions in semantic theory: Semantic Intuitionism Semantic theories should conform to “semantic intuitions”—what competent speakers say about the meanings (or truth-conditions) of the words and sentences they use. Though Semantic Intuitionism is typically assumed without argument, it might be supported by the following premises: 1. 2.

Competent speakers ipso facto know what they say. The semantic intuitions expressed by people who know what they say—that is, their sincere pronouncements about what is said (or about its truth-conditions)—are ipso facto reliable.

But how plausible are these premises? In particular, do competent speakers always know what they say?36 The short answer to the question is, “Yes and no.” The longer answer is that there are different senses in which one might “know” what one says, and one might know what one says implicitly without fully knowing it explicitly. This calls for some clarification, both of the difference between implicit and explicit knowledge of what one says, and also of the difference between full and less than full explicit knowledge.37 Knowing implicitly what one says consists in merely being able to use words appropriately, to convey what one wishes to convey. This is to be contrasted with knowing explicitly what one says, in the sense of being able to articulate and recognize correct paraphrases and to make correct judgments about other semantical claims (regarding truth-conditions, entailtains that it is “a theoretically useless entity which plays no role in communication” (2004: 86). Elsewhere, though, he argues that it does not even make sense: “There is no such thing as ‘what the sentence says’ … There is a single notion of what is said, and that is a pragmatic notion” (2004: 58). 36. Some of this discussion appears in my “Must We Know What We Say?" 37. Talk of knowing what is said can also be expected to display the same sort of pragmatic sensitivity as any other talk of knowing what or knowing who (see Boer and Lycan), but for present purposes this complication may be ignored.

Semantic intuitions 83

ment, etc.). So, whereas the former is mainly a matter of knowing how to use words successfully in linguistic interaction, the latter is more a matter of being able to distinguish true semantic claims from false ones. Competent English speakers certainly know how to use the word ‘knows’, for example, in sentences such as S:

Obama knows that Mexico City is the capital of Mexico,

so they at least have implicit knowledge of what they say in uttering such sentences. But few would be able to formulate a correct, nontrivial paraphrase of S; and many would not be able to judge correctly whether S is equivalent (semantically or logically) to S':

Obama believes truly and with good reason that Mexico City is the capital of Mexico.38

A distinction must also be made between different degrees of explicit semantic knowledge (that is, explicit knowledge of what is said). Some explicit knowledge of what is said is trivial—e.g., that the following sentence is true: ‘Obama knows that Mexico City is the capital of Mexico’ is true (relative to this context of utterance) iff Obama knows that Mexico City is the capital of Mexico. On the other hand, some explicit knowledge of what is said is substantial— e.g., the knowledge that S concerns, inter alia, Obama’s psychological state. If S and S' are synonymous, then explicit semantic knowledge of that would be even more substantial. Full explicit knowledge of what is said may be thought of as explicit semantic omniscience—knowing everything there is to know about the semantics of the words uttered. That is, a speaker has full explicit semantic knowledge of what he says if and only if he can correctly determine the truth-value of any claim about the semantics of his utterance. 38. Cf. James Higginbotham’s distinction between the meaning of an expression for a speaker and the speaker’s views about the expression’s meaning (1998: 430). Implicit knowledge of what one says corresponds (more or less) to knowledge of meanings in this sense, and explicit knowledge of what one says corresponds to correct views about meanings.

84

The pragmatics of substitutivity

So, the first premise of the argument for Semantic Intuitionism is plausible, but only with regard to implicit knowledge. Competent speakers must at least have implicit knowledge of what they say—otherwise, they could hardly be considered competent. Moreover, we expect competent speakers to have at least some substantial explicit knowledge of what they say; they should be able to formulate at least some correct informative claims about the meanings of their words, even if they cannot come up with precise definitions. Clearly, though, it is very doubtful that anyone ever has full explicit knowledge of what he says. Competent speakers often cannot provide precise, nontrivial formulations of what they say; and even when they can, there are bound to be some semantic facts about their utterances (what they are synonymous to, what truth-conditions they have, what they entail, etc.) that the speakers themselves do not recognize as such. It is easy to point out cases where people who use sentences perfectly well are mistaken about their semantic content: I live in a hilly Mediterranean city that has a big port and starts with an ‘H’. All seven-eyed philosophers can fly. All that glitters is not gold. The name ‘Haifa’ has five letters in it, one of which occurs twice. If dogs are animals, then bachelors are unmarried. The present king of France is bald. I think the vast majority of English speakers get the semantics of most of these sentences wrong; in any case, it is clear that for each of these sentences there are competent speakers who disagree about the sentence’s truth conditions. Russell and Strawson, for example, both know perfectly well how to use sentences containing definite descriptions, such as K:

The king of France is bald;

so both know, at least implicitly, what they say in uttering such sentences. And both have some substantial explicit knowledge of what is said in uttering such sentences—both know, for instance, that the following is true: One can say something true in uttering K only if there exists a king of France.

Semantic intuitions 85

But they clearly disagree about the semantics of those sentences—e.g., about whether the following is true: If there is no king of France, then in uttering K one says something that is false. Since at least one of them must be wrong, at least one of them does not have full explicit knowledge of what he says in uttering sentences such as K.39 Thus, it is clear that competent speakers are not necessarily semantically infallible; there are always bound to be some claims about what they say that they cannot accurately assess.40 So, although competent speakers must at least have implicit knowledge of what they say (and can even be expected to have some substantial explicit knowledge of what they say), they seldom if ever have full explicit knowledge of what they say. But it is the latter that is required for the second premise of the argument for Semantic Intuitionism—it is only when speakers have full explicit knowledge of what they say that all their declared semantic intuitions are reliable. Hence, if the premises of the argument for Semantic Intuitionism are rendered plausible, the argument rests on equivocation between implicit knowledge of what is said, in the first premise, and explicit knowledge of what is said, in the second. The fact that linguistic competence, in the sense of knowing how to speak a language, does not guarantee perfect semantic knowledge, in the sense of being able to articulate (in an illuminating way) the semantic contents—or even truth conditions—of sentences of the language, should come as no surprise. For one thing, there is no reason to suppose that our explicit knowledge of semantics would be any greater than our explicit knowledge of syntax—which clearly lags far behind our practical syntactic ability. And just as basic syntactic notions—such as that of a sentence—are quasi-

39. The prevalence of semantic disagreement even among professionals wellversed in the difference between semantics and pragmatics undercuts the objection raised by Mark Richard (1990: 125) and endorsed by David Braun (1998: 570–571), that the sort of pragmatic account I give of belief ascriptions is doubtful, because it is not as widely and readily accepted as many other pragmatic accounts (e.g., of the temporal order sometimes associated with the word ‘and’). 40. I am thankful to Jacek Jadacki for helping me to clarify this.

86

The pragmatics of substitutivity

technical, so too for semantic content.41 The notion of semantic content is only quasi-technical, since there is an ordinary, pre-theoretical distinction between what a sentence literally means and what a speaker means in uttering the sentence on a particular occasion.42 And the semantic content of a sentence is certainly closely related (albeit not as simplistically as some would have it) to what competent speakers mean in uttering that sentence. But the notion of semantic content is technical enough that figuring out exactly what the semantic content of an expression is sometimes takes a lot of work—and even then, smart people, who have given it a lot of thought, might disagree (as so often happens not only in philosophical analysis, but even in lexicography). Another reason why linguistic competence should not be expected to guarantee perfect explicit semantic knowledge is the obvious yet widely overlooked fact about language use, that what really counts when we speak is often not what we actually say. As exact transcripts of real conversations plainly show, sometimes we get our point across by uttering a string of words that does not even constitute a sentence—and hence does not express a proposition altogether. Other times we utter a sentence expressing one proposition when our main aim is to convey some other proposition. And the proposition expressed by the sentence actually uttered may well be false— even clearly so—as is often the case when we speak “loosely” or metaphorically. So it is not as strange as it might seem at first that the acceptability of an utterance often does not correspond to the truth-value of the string of words actually uttered. Having seen that semantic theories need not conform to common semantic intuitions, may we conclude that semantic intuitions may be ignored? Of course not. What, then, is the place of semantic intuitions in semantic theory? To answer this question it is helpful to consider first just what it is that semantic intuitions reflect. What competent speakers can be expected to know merely in virtue of their being competent speakers is how to use the expressions they use. Thus, their claims about the meaning of an expression are first and foremost an indication of the pragmatic import of utterances of the expression. That is, it is knowledge of pragmatic content, not seman41. The quasi-technical nature of the notion of semantic content is reflected in Grice’s gloss of his talk of “what the speaker has said” as meant in “a certain favored, and maybe in some degree artificial, sense of ‘said’” (1989: 118). 42. I think Thau exaggerates when he declares that “the vast majority of competent speakers don’t have any views about semantic values at all” (2002: 165).

Semantic intuitions 87

tic content, that is most directly reflected in the articulated semantic judgments of competent speakers.43 Of course,

43. (a) Recanati (1993: 246) seems to make a similar point, in terms of his distinction between “linguistic meaning” and “what would be said or communicated” (corresponding, respectively, to what I am calling “semantic content” and “pragmatic content”): To be sure, users of the language claim to have intuitions concerning what the sentences in their language mean; but these intuitions are not directly about their purported objects—linguistic meanings. They do not bear on the linguistic meanings of sentences, which are very abstract and inaccessible to consciousness, but on what would be said or communicated by the sentence were it uttered in a standard or easily accessible context. However, (i) I would not go so far as to say that common intuitions about what sentences mean “do not bear on” sentence meaning; for they certainly do, though not as directly as one might suppose. More importantly, (ii) when it comes to the reliability of common intuitions, I would group “what is said” with sentence meaning, not with what is communicated. That is, the thing that common semantic intuitions most directly bear on is what is communicated, not what is said. (b) Similarly, Soames (2002: 68) suggests that the reason ordinary speakers are not always right about semantic content is that the semantic content of a sentence is, roughly, the information invariantly asserted and conveyed by utterances of the sentence in normal contexts, whereas ordinary speakers tend to focus on the information that is merely typically conveyed. This may be true, but the relevant notion of semantic content remains to be elaborated noncircularly. (c) See also Ralph Walker: “the ascriptions of truth-value one … makes for technical reasons are liable to diverge substantially from those that people ordinarily and unreflectively make. … ascriptions of truth and falsity are themselves context-relative and can be made for different purposes” (1975: 139). I would add that these different purposes include the theorist’s somewhat esoteric objective of talking about the semantic content of a sentence, on the one hand, and the common speaker’s everyday objective of talking about the pragmatic import of its utterance, on the other. Although Walker does not explicitly rely on this distinction, he does recognize a notion of literal truth, “used by those who wish to exclude everything but sense and reference from the determination of truth-value” (1975: 140); and what he says about literal truth seems close to what I say about semantic content:

88

The pragmatics of substitutivity

pragmatic and semantic contents are not unrelated—it is in virtue of an expression’s semantic content that its utterance can have a particular pragmatic content, and it is on the basis of the pragmatic contents of the utterances of an expression that we arrive at its semantic content.44 But knowing (explicitly) what an expression’s semantic content is, is not just a matter of knowing how to use the expression; it requires abstracting from a wide variety of uses, in a way that is in harmony with our understanding of the myriad of relations between semantic and pragmatic contents. Thus, taking semantic intuitions into account does not require accepting them at face value; semantic theories may diverge from semantic intuitions, as long as the divergence is suitably accounted for. What counts in the end is that the semantic contents of expressions are determined in accordance not only with the pragmatic contents of their utterance—typically reflected in explicit semantic intuitions—but also with the many ways in which semantic and pragmatic contents are related.45 Given its dubious underpinnings, why is Semantic Intuitionism so popular? Why is it so common to neglect the distinction between implicit and

Our intuitions about literal truth are often unclear, which is not surprising since it is not a notion we normally use much; but it is a notion of considerable utility for the theory of meaning. A theory which sets out to assign truth-conditions to the sentences of a language may be expected to assign conditions for literal truth. 44. It is for just this reason that the more an expression is used with a particular pragmatic content other than its semantic content—as in the case of belief ascriptions—the more likely speakers are to mistake that pragmatic content for semantic content. 45. Cf. Mark Richard (1990: 125–126): When sophisticated speakers have what amounts to an unshakable conviction about truth conditions—and our anti-Russellian convictions approach this—we have a compelling reason to look for a theory that honors those intuitions, even if there already exists a coherent theory that denies the intuitions. If such a theory is to be had, we should, all else being equal, embrace it. Though easily overlooked, the condition of “all else being equal” is essential—and seldom, if ever, fulfilled. Semantic intuitions can be honored without being accepted as correct, as long as it is shown how they arise and how they fit into the big picture, including both semantic and pragmatic theories.

Semantic intuitions 89

explicit knowledge of what one says, and to assume that the semantic pronouncements of competent speakers are reliable? This general tendency may derive from an underlying assumption about the relation between meaning and saying: The Coincidence View of Meaning and Saying What competent, sincere speakers mean is typically what they say. Although sometimes a speaker may say one thing while meaning something else, it is much more typical for speaker-meaning to coincide with sentence-meaning—for speakers to mean exactly what they say (as most would insist they do!). Given this way of looking at meaning and saying, interpreting utterances— determining what a speaker means—is mainly a matter of determining the semantic contents of the words uttered; thus, a theory of linguistic interpretation is mainly a semantic theory for the language in use. So on the Coincidence View, competence in linguistic interpretation largely guarantees explicit semantic knowledge; competent speakers know explicitly what they say, in which case semantic theories should conform to their avowed intuitions. The Coincidence View is to be contrasted with the Divergence View: The Divergence View of Meaning and Saying What competent, sincere speakers mean is typically not exactly what they say. Although sometimes a speaker means exactly what she says, it is much more typical for speaker-meaning to diverge from sentence-meaning (for speakers to mean something other than what they actually say). According to the Divergence view, interpreting utterances is not mainly a matter of determining the semantic contents of the words uttered; a theory of linguistic interpretation includes much more than a semantic theory. Thus, competence in linguistic interpretation does not guarantee explicit semantic knowledge; competent speakers know what they and others mean, but they often do not know (explicitly) what they say. Despite the popularity of the Coincidence View, prima facie evidence for the Divergence View abounds. One such kind of evidence is the sort of semantic disagreement discussed above. Since competent speakers sometimes disagree about what they say in uttering certain sentences, some of

90

The pragmatics of substitutivity

them must be mistaken about what they say; and at least where those who are mistaken about what they say mean what they think they say, what they mean diverges from what they say. If for example, Russell is wrong in thinking that what he says in uttering a sentence containing a definite description is a certain existential generalization, then where he means, in uttering the sentence, what he (incorrectly) thinks he is saying (the existential generalization), what he means diverges from what he says. (And if he is right, then it is in the utterances of those who reject his theory that what is meant diverges from what is said.) More evidence favoring the Divergence View lies in the prevalence of syntactic error. Competent speakers often pronounce sequences of words that are not even syntactically well-formed, e.g., when they fail to complete sentences or alter them midway. Assuming that they nevertheless mean something, and that well-formedness is necessary for semantic content, these would be cases where what is meant could not be what is said, since actually nothing is said (that is, no proposition is the semantic content of the words uttered). Perhaps the strongest evidence for the Divergence View consists in what might be thought of as pragmatic ellipsis, where we seem to omit words that we expect the listener to be able to restore on the basis of pragmatic considerations:46 Uttered: Meant:

The chair is broken. The chair you are about to sit on is broken.

Uttered: Meant:

Everybody got drunk. Everybody at the party I went to last night got drunk.

Uttered: Meant:

Jill is bigger. Jill is bigger than her brother.

46. Some kinds of pragmatic ellipsis have been referred to as “explicature” (Sperber and Wilson (1986: 182); Carston (2002: 116)) or (in an entirely different framework) “conversational impliciture” (Bach (1994)). Pragmatic ellipsis is to be distinguished from syntactic ellipsis, as in Jack fell down, and Jill did too, where the “missing” words can be recovered without any pragmatic ingenuity.

Iterability 91

Uttered: Meant:

This book cost $40. This book cost about $40.

Uttered: Meant:

FLOTATION DEVICE UNDER SEAT. There is a flotation device under this seat.

Uttered: Meant:

Having a great time; wish you were here. I am having a great time; I wish you were here.

Uttered: Meant:

Traffic jam. I am late because of a traffic jam.

For some of these—most notably those toward the top of the list—attempts have been made to construe the semantic content of the shorter sentence as including that of the longer one.47 But even if such counterintuitive proposals were to succeed, they apply to only a small portion of the rich variety of ways in which speakers clearly seem to mean more than what they actually say. Considering these data en masse, with due regard to the full range of such cases—including those lower down on the list—it is hard to deny the Divergence View. Once the inadequacy of the Convergence View of Meaning and Saying is recognized, the pitfalls of Semantic Intuitionism are more easily avoided. The semantic intuitions of competent speakers are surely important—but they are not a reliable guide to semantic content. Common semantic intuitions constrain semantic theory only indirectly—in the context of an overall theory of interpretation, including theories of how strict semantic content is related to pragmatic content and other kinds of pragmatic import.48

9. Iterability Iterated applications of the belief operator may seem to pose a tougher challenge to the Theory of Direct Belief. If Floyd is one of the people who insists that (6')

Lois Lane does not realize that Clark Kent is Superman

47. See, e.g., Jason Stanley and Zoltan Szabo (2000), and Robert Stainton (2006). 48. For further arguments against Semantic Intuitionism see Bach (2001).

92

The pragmatics of substitutivity

is true and uses it accordingly, it would seem that (8)

Floyd believes Lois Lane does not realize Clark Kent is Superman

must also be true. (And if you think (6') is true, it would seem that (9)

You believe Lois Lane does not realize Clark Kent is Superman

must also be true.) But since Floyd surely believes that Lois Lane realizes Clark Kent is Clark Kent, it follows on the Theory of Direct Belief that Floyd believes Lois Lane does realize Clark Kent is Superman must be true. So, as Stephen Schiffer has observed (1987a), the Direct Belief theorist would have Floyd both believing and disbelieving (believing false) the same proposition, even though he does not share Lois Lane’s confusion. However, appearances notwithstanding, (8) is actually false (as well as (9)!).49 Certainly it would be natural to say (8), in order to account for Floyd’s sincere utterances of (6'). But as already noted (in the previous section), what is natural—or even proper—to say is not necessarily what is true. In particular, the pragmatics of belief ascriptions embedded in (iterated) belief ascriptions is much like that of belief ascriptions embedded in negations, in that the pragmatic content of the embedded part may be inherited by the whole (Section 6). In the case at hand, there are two stages of inheritance. First (as in the case of (5) and (6) in Section 6), since (5')

Lois Lane realizes Clark Kent is Superman,

would typically be used to convey, by conversational implicature, that (say) Lois realizes that the names ‘Clark Kent’ and ‘Superman’ are coreferential, an utterance of its negation, (6'), could inherit this content, so that the 49. The truth-values of belief ascriptions according to the Theory of Direct Belief may be easier to recognize and accept if one keeps in mind that the ascription is to be interpreted relationally or de re (with respect to proper names)—the theory takes the de re interpretation as the sentence’s only semantic content, and the notional or de dicto interpretation comes out only as pragmatic content.

Iterability 93

pragmatic content of the utterance of (6') would be the negation not of (5') itself, but of the pragmatic content of an utterance of (5')—i.e., so that the pragmatic content of the utterance of (6') would be that Lois does not realize that the names ‘Clark Kent’ and ‘Superman’ are coreferential. Then, this pragmatic content may in turn be inherited by an utterance of (8), so that the pragmatic content of the latter would be not that Floyd believes (6'), but that Floyd believes the pragmatic content of a typical utterance of (6')—i.e., that Floyd believes Lois Lane does not realize that the names ‘Clark Kent’ and ‘Superman’ are coreferential. Thus, (8) is understandably thought to be true, even though what it literally means—that Floyd believes the semantic content of (6')—is false (as the story goes).50 Against this Schiffer (1987a: 465) insists that (8) is true, arguing as follows: (a) (b) (c) (8)

Floyd believes (6') is true Floyd knows that if (6') is true then Lois Lane does not realize Clark Kent is Superman. Floyd knows modus ponens. Floyd believes Lois Lane does not realize Clark Kent is Superman.

The problem with this argument is that the second premise is false. We must distinguish carefully here between what Floyd knows and what he does not. We may assume that he recognizes the complement of (b), (b')

If (6') is true then Lois Lane does not realize Clark Kent is Superman,

as an instance (more or less) of the disquotational schema If ‘ij’ is true then ij,

50. It is easy to see how the inheritability of pragmatic content similarly accounts for cases in which belief ascriptions are embedded in other ways, such as Ernie Lepore and Barry Loewer’s example, ‘If Lois Lane believes that Superman = Clark Kent, then she will report it in the Daily Planet’ (1995: 192), as well as the sort of cases, raised by James Higginbotham (1995: 120), where the idiom of belief ascriptions occurs in nonassertoric speech acts, such as questions and suppositions.

94

The pragmatics of substitutivity

and hence knows that the sentence (b') is true. However, (b) itself does not follow—for believing that a sentence is true does not necessarily entail believing what the sentence expresses. The problem here is that Floyd (like most people) does not fully know the semantics of belief (nor of the word ‘realizes’). Floyd’s semantic ignorance manifests itself first with regard to (6')—although he certainly knows how to use the sentence (at least in typical circumstances), he does not really know what the sentence, strictly speaking, means. Hence, he does not really know the meaning of sentences in which (6') is embedded, such as (b'). So his believing that the sentence (b') is true does not entail his believing what (b') expresses.51 Indeed, knowing enough about the semantics of belief to be able to infer (b') from the disquotational schema will generally be incompatible with thinking that (6') is true in the first place. Clearly, in rejecting (8) I am rejecting the disquotational principle of belief, according to which, roughly, people believe whatever is expressed by the sentences they sincerely assert.52 Despite the popular view of the disquotational principle as a “self-evident truth” (Kripke (1979: 361); Salmon (1986: 130)), it should be no less evident that the principle cannot be plausible if it is not restricted to cases where the person in question fully understands the sentence in question. For insofar as a person is mistaken about 51. Kripke makes a similar point (though in a different context) (1972: 69–70): Of course, anyone who knows the use of ‘is called’ in English, even without knowing what the statement means, knows that if ‘quarks’ means something then ‘quarks are called “quarks"‘ will express a truth. He may not know what truth it expresses, because he doesn’t know what a quark is. But his knowledge that it expresses a truth does not have much to do with the meaning of the term ‘quarks’. Likewise, anyone who knows the use of ‘is true’ in English knows that if (6') is a meaningful declarative English sentence, then (b') expresses a truth; but he will not know exactly what truth it expresses, if he does not know exactly what ‘realize’ means. And his knowledge that (b') expresses a truth does not have much to do with the meaning of the word ‘realize’ (or the word ‘believe’). 52. Kripke stated the principle as follows (1979: 360): where ‘p’ is to be replaced, inside and outside all quotation marks, by any appropriate standard English sentence [lacking indexical or pronominal devices or ambiguities]: ‘If a normal English speaker on reflection, sincerely assents to “p,” then he believes that p.’

Other pragmatic accounts of substitution failure 95

what a sentence means, his utterance of the sentence will be an indication not of his believing what is expressed by the sentence, but of his believing what he thinks is expressed by the sentence. In particular, due to Floyd’s confusion about the meanings of (5') and (6'), his disposition to utter (6') is evidence not of his believing what is expressed by (6')—i.e., not of (8)— but rather, of his believing what he thinks is expressed by (6')—i.e., of his believing that Lois Lane does not realize that the names ‘Clark Kent’ and ‘Superman’ are coreferential.53

10. Other pragmatic accounts of substitution failure 10.1. Soames and what is said According to Scott Soames, the problem with (2)

Lois Lane believes Superman is a reporter

lies not in what is conversationally implicated by uttering it, but in what is actually said.54 Soames contends that what is said in an utterance typically transcends the semantic content of the words uttered. In particular, although the semantic content of (2) is only that Lois believes the singular proposition consisting of Superman and reporterhood—which is clearly true— Soames holds that what one actually says in uttering (2) might be something such as (10) Lois Lane believes that the famous superhero Superman is a reporter,

53. Nathan Salmon, who points to the disquotation principle in suppport of his view that (8) is true, seems to recognize that the principle must be restricted in the way I have suggested, as he explicitly assumes that Floyd fully understands (6') (1989: 264, 278, n. 20). He thinks this assumption is clearly true, but I have explained why I think it is not. The principle also raises much deeper questions about the nature of belief, to be discussed in Chapter 3. 54. Although Soames usually speaks in terms of assertion, he explicitly takes saying as a way of asserting (2002: 57), often using the two words interchangeably. For the sake of uniformity I will speak in terms of saying. And of course, he does not explicitly speak of (2); the claims I attribute to him about (2) are extrapolated from what he says about similar examples.

96

The pragmatics of substitutivity

which unlike (2) is clearly false. This raises some hard questions. First of all, if what is said in uttering a sentence is not its semantic content (the proposition, if any, expressed by the sentence) then what is it? Secondly, how exactly is it that by uttering (2) one can convey what is expressed by (10)? With regard to the first question Soames offers several answers. At first he characterizes what is said in uttering a sentence (in standard cases, where the utterer is not speaking metaphorically, sarcastically, ironically, or nonliterally, and is not conversationally implicating anything incompatible with the semantic content) as the semantic content of the sentence along with all mutually recognized, direct, immediate, relevant consequences of it together with background knowledge (2002: 57). Troubled by some cases where the semantic content is false or unintended, Soames later modifies the account so as to include as what is said only “acceptable completions"of the semantic content—which may or may not include the semantic content itself (2005: 365). These acceptable completions are subsequently described as “proper pragmatic enrichments” of the semantic content (2006: 6), pragmatic enrichments being the contents of syntactic expansions of the uttered words (2008: 456). Although Soames explicitly refrains from offering a theory of what makes a pragmatic enrichment proper (at least as of (2006)), he considers a pragmatic enrichment part of what is said only if it is “obvious and relevant.”55 Soames adds that Grice’s maxims help determine what is said in the following way: “When several enrichments are otherwise feasible, the maxims dictate that one select the strongest, most informative, and relevant propositions among them for which one has adequate evidence” (2008: 456). It is hard to see, however, why we should abandon the traditional Gricean view, that if a speaker said that p, he must have uttered a sentence that means that p (1989: 87-88). Returning to Kent Bach’s example, if I utter just the words Pepsi is better, making no explicit reference to Coke, then although

55. (2008: 456). In the case of the semantic content of the sentence uttered, what is required for being (part of) what is said is to be “an obvious and relevant consequence of the enriched proposition that it is the speaker’s primary intention to assert.” But it is not clear that this is supposed to be a general constraint on what is said.

Other pragmatic accounts of substitution failure 97

Pepsi is better than Coke may well be an acceptable completion of my remark (and even what I meant), it does not seem that I actually said that Pepsi is better than Coke— or that I said anything about Coke at all.56 Regardless of whether or not completions such as these are strictly speaking part of what is said in the utterances they presumably complete, and regardless of whether or not what is said can include more than the semantic content of the words uttered—the second question remains unanswered: how exactly is it that by uttering (2) one can convey what is expressed by (10)? Merely labeling (10) as “said” or “asserted” in an utterance of (2) does not explain how it gets conveyed. If uttering (2) conveys (10), how does it happen? All we can surmise from what Soames says about what is said is that (10) is said by uttering (2) because (a) it is an obvious and relevant expansion of (2), and

56. It is interesting that Soames explicitly acknowledges that his expansive notion of what is said is not needed for a pragmatic account of substitution failure: Those Millians who are unwilling to take this further step [of locating the false information conveyed by an utterance of (2) in what is said] may nevertheless accept the main thrust of my argument by developing an account of linguistic communication that emphasizes the importance of the information that speakers primarily intend to convey (in those cases in which it diverges from what is asserted). (2002: 361, n. 7, emphasis mine) Taking together (a) this explicit recognition of the adequacy for present purposes of the notion of the information that speakers primarily intend to convey, (b) Soames’s inclusion of speaker’s reference—introduced by Kripke (1977: 14) as a special case of Grice’s notion of utterer’s meaning—as part of what is said (2006: 14–15), and (c) the relative plausibility of taking pragmatic completions as meant rather than said—it is tempting to think that what Soames has in mind as what a speaker says or asserts is actually something closer to what the speaker means. I discuss Soames’s semantics of indirect discourse in more detail in “What is Said to Be Said” (2007).

98

The pragmatics of substitutivity

(b) it differs from other enrichments in that Grice’s maxims dictate that it be selected as the strongest, most informative, and relevant proposition among them for which one has adequate evidence.57 Now condition (a) by itself is clearly not sufficient, as it is satisfied just as well by the negation of (10)—also an expansion of (2), which would be no less obvious and relevant. In order to weed out unintended expansions, condition (b) is essential, restricting what is purportedly said to suitable interpretations of the speaker. But how does (b) do this? Strictly speaking, Grice’s maxims cannot dictate how to select an interpretation, because all they dictate is how to contribute to a conversation; they tell us how to speak, not how to interpret. The way we use Grice’s maxims in interpretation is by assuming that they are being followed. Understood in this way, what (b) tells us to select would be enrichments that can most plausibly be ascribed to the speaker on the assumption that the speaker is following the maxims. But this seems to amount to nothing other than what is conversationally implicated (according to Grice’s definition, above, Section 2.2). It seems then that Soames ultimately shares my view that what is false about (2) is not its semantic content, but what an utterer of (2) might conversationally implicate—except that, having adopted his nonstandard conception of “what is said” as including more than the semantic content of the words uttered, Soames construes the false information conveyed by an utterance of (2) not as conversationally implicated, but as said. And he, but not I, insists that this false information be the content of a syntactic expansion of (2). But perhaps the most significant difference between us here is that Soames has not offered any account of how the guilty information is conveyed. He has suggested what form it might take, but has not spelled out how, by the mechanism of conversational implicature or otherwise, an utterance of (2) actually gets the false information across.58 57. Soames actually puts it in the plural: “When several enrichments are otherwise feasible, the maxims dictate that one select the strongest, most informative, and relevant propositions among them for which one has adequate evidence” (2008: 456, emphasis mine); so he may have meant that what one selects might be a number of such propositions, not just one, as I have put it. But the difference is not significant for present purposes. 58. As he himself admits, “Much about the conception of linguistic meaning and language use to be sketched remains incomplete, including the development of precise theories of how pragmatic enrichment takes place, and the formulation of constraints on what constitutes a proper pragmatic enrichment of the

Other pragmatic accounts of substitution failure 99

10.2. Thau and what is implicated Michael Thau agrees with me that the problem with (2)

Lois Lane believes Superman is a reporter

lies in what is implicated by uttering it, not in what is said. But Thau suggests that the false information conveyed by an utterance of (2) is implicated conventionally, not conversationally. Grice (1989: 25) explains the notion of conventional implicature with the example of someone uttering (11) He is an Englishman; he is, therefore, brave and thereby implicating (12) His being brave is a consequence of his being an Englishman. Grice maintains that (12) is only implicated, not said, since it could be false without (11) being false (strictly speaking).59 But it is implicated conventionally, not conversationally, because it is determined by the conventional meaning of the words (particularly the word ‘therefore’), rather than being derived as something the speaker must believe on the assumption that he is following the Cooperative Principle. Thau argues that names such as ‘Superman’ and ‘Clark Kent’ (and ‘Hesperus’ and ‘Phosphorus’, etc.) differ from ordinary proper names, in that a speaker is not competent in using them unless he associates certain descriptions with them; a speaker who does not associate the right descrip-

semantic content of a sentence. Although these matters are critical, at this point, I have no general theory of them” (2006: 6). In an earlier discussion Soames formulates several claims about what is conveyed (in his words, “said” or “asserted”) in various cases of belief ascription utterances, but there too he does not explain how (2002: 212–222). 59. An earlier example of Grice’s might be less objectionable: by saying ‘She is poor but honest’ one conventionally implicates, roughly, that there is some contrast between poverty and honesty, which presumably could be false without the uttered sentence being false (1961: 90). For an opposing view see Kent Bach’s “Myth of Conventional Implicature” (1999).

100

The pragmatics of substitutivity

tions with the right names suffers from “linguistic confusion or ignorance” about those names. “To see what I’m getting at,” he says, suppose that someone has the descriptions associated with the names Superman and Clark Kent reversed: like most other people, this person falsely believes that the super-powered protector of Metropolis and the bespectacled Daily Planet reporter are distinct, but he also believes that Superman is the name of the bespectacled Daily Planet reporter, and that Clark Kent is the name of the super-powered protector of Metropolis. It seems to me that such a person doesn’t really know how to use the names Superman and Clark Kent; that is to say, such a person suffers from linguistic confusion about the names. (2002: 173)

Thau concludes that the content of these associated descriptions is part of what is conventionally implicated in utterances of sentences containing them. Thus, by uttering (2) one conventionally implicates something false such as (13) Lois Lane believes the super-powered protector of Metropolis is a reporter. One problem with this approach is that conventional implicatures— deriving only from general linguistic features of the words uttered and not from anything peculiar to the context of utterance (such as the assumption in a particular context that the speaker is following the Gricean maxims)— are not easily canceled (if at all). There doesn’t seem to be any way to say (11), for instance, without also committing oneself to (12); an utterance such as this: He is an Englishman; he is, therefore, brave—though of course I don’t mean to imply that his being brave is a consequence of his being an Englishman would at least be anomalous, even if not quite self-contradictory. But this does not seem to be the case with (2) and (13); it seems perfectly acceptable to affirm the first and deny the second: Lois Lane believes Superman is a reporter—though of course I don’t mean to imply that she thinks the super-powered protector of Metropolis is a reporter. After all, she doesn’t realize that the reporter and the superhero are one and the same.

Other pragmatic accounts of substitution failure 101

Nor does anything like (13) get implicated in the sort of contexts generating Quine’s relational intuitions, as in the example I gave where (2) seems true: Look at what a master of disguise that Superman is! Why, when he puts on that suit and strolls into his office at the Daily Planet, he fools everybody—Mr. White, Jimmy Olson; even Lois Lane believes Superman is a reporter. The lesson of such contexts is critical: the false information that can be conveyed by an utterance of (2) is not always conveyed by an utterance of (2). Taking the false information as conversationally implicated, depending on the context, accommodates this fact; taking it as conventionally implicated, not depending on the context, does not. An even greater problem with Thau’s conventional implicature approach is that there is no way to apply it to names that do not have any descriptive content conventionally associated with them. Thau anticipates this objection, recognizing that many would think one could believe that Cicero was a great orator without believing that Tully was. Thau’s reply is to dismiss the data: While I wouldn’t balk at saying the person in the example has the one belief and not the other, it seems equally intuitive to say that such a person does believe that Tully was a great Orator; he just doesn’t know that Cicero was called Tully. So the Fregean and Millian intuitions are equally compelling in this case; … this is because the names don’t have distinct descriptive contents that most people familiar with the names associate with them. The only names for which the Fregean intuitions dominate—and, hence, the only ones that provide an argument for a Fregean solution to the puzzle— are the ones in which the relevant descriptions are so closely associated with the names that they become part of what’s conventionally implicated by sentences containing the names. (2002: 174)

So according to Thau, names with conventionally associated descriptions are the only ones for which Fregean intuitions (that substitution fails) are more compelling than Millian intuitions (that substitution succeeds—what I called “Quinian” intuitions); and Fregean intuitions need be considered only where they are more compelling than Millian intuitions. Both claims, I shall argue, are wrong. For me, Fregean intuitions are no less compelling in the Cicero/Tully case than in the Superman/Clark Kent case; but I won’t argue over intuitions. Rather, I shall consider a different example involving names without

102

The pragmatics of substitutivity

conventionally associated descriptions, where I think Thau would have to agree that the Fregean intuitions are as compelling as in the Superman/Clark Kent case. Suppose a covert bigamist maintains one household with Emma, in a town where he is known only as “Brian Smith,” and another household with another wife, in a town where he is known only as “Joe Young;” neither wife has the slightest clue of her husband’s double life, and in particular, neither is familiar with the name he uses with the other. Suppose now that Emma hears on the radio that “Joe Young was killed in a traffic accident.” Upon hearing the news, she thinks it’s a shame but—as to be expected—is not deeply affected. Why not? The natural answer would be (14) Emma believes that Joe Young was killed, but not that Brian Smith was. But there is no reason to suppose that the Fregean intuitions in support of (14) are any less compelling than the Fregean intuition in support of (14') Lois Lane believes that Clark Kent is a reporter, but not that Superman is; just as there is no reason to suppose that the Fregean intuitions against Emma believes Brian Smith was killed are any less compelling than the Fregean intuitions against (2)

Lois Lane believes Superman is a reporter.

Clearly, though, the names ‘Joe Young’ and ‘Brian Smith’ have no descriptive content conventionally associated with them. Hence, the Fregean intuitions about substitution failure do not depend on conventionally associated descriptive content. In any case, it is wrong to suppose that Fregean intuitions need to be considered only insofar as they dominate. It is true, of course, that a Fregean treatment of belief ascriptions (where (1)

Lois Lane believes Clark Kent is a reporter

Summary 103

and (2) are not logically equivalent, because they concern not Superman, but different modes of presentation of Superman) would seem to be best supported by those cases where Fregean intuitions are strongest. But the question at hand is not simply whether the prima facie strongest Fregean arguments from Fregean intuitions about substitution failure are sound. It is, rather, how to account all at once for both the Fregean intuitions that substitution fails as well as the Millian intuitions that it does not. Even when Millian intuitions dominate over Fregean intuitions, let alone when they are equally compelling, the Fregean intuitions remain part of the data to be explained.

11. Summary In this chapter I showed how coreferential names are indeed interchangeable salva veritate, despite any Fregean intuitions to the contrary. Fregean intuitions were seen to result from confusing the truth of a belief ascription sentence with the appropriateness of its utterance, the latter but not the former being affected by what is conversationally implicated by the utterance. My pragmatic account of substitution failure began with the observation that we normally presume that things are normal. From this I derived, on the basis of Grice’s first maxim of Quantity, the Principle of Implicated Normalcy: speakers generally conversationally implicate that the circumstances regarding whatever they are speaking of are not abnormal in any significant, unanticipated, unindicated way. Given that correct belief ascriptions are normally acceptable verbatim (with some minor qualifications), in the sense that the person to whom the belief is ascribed would accept that very ascription, word for word, as true, it follows by the Principle of Implicated Normalcy that the utterer of a belief ascription normally implicates that his ascription is acceptable verbatim. In particular, in a typical utterance of (2)

Lois Lane believes Superman is a reporter

it would be conversationally implicated that the utterance is acceptable verbatim, i.e., that Lois would accept that very utterance, word for word, as true. Mistaking this conversational implicatum for part of the semantic content of (2) is what leads to the Fregean intuition that substitution fails.

104

The pragmatics of substitutivity

On the other hand, verbatim acceptability is not conversationally implicated in suitably abnormal cases—such as when (2) is uttered in a conversation about how well Superman fools Lois when he disguises himself as Clark Kent (Chapter 1, Section 1). In such circumstances the presumption of normalcy as applied to verbatim acceptability of belief reports is suspended, since the talk of Lois’s being fooled makes it clear that the case is not normal and in particular that we cannot rely on her judgment about sentences such as (2); so the normal conversational implicature about verbatim acceptability does not arise, and the Fregean intuition that substitution fails gives way to the Quinian intuition that it succeeds. After giving this account of our conflicting intuitions about (2), I showed how the account could be adapted to cases where we ascribe beliefs to individuals who clearly do not understand the words we use, and hence, for whom the ascription clearly cannot be acceptable verbatim. Rejecting the assumption that there is any unique conversational implicatum (or schema thereof) that is responsible for every case of apparent substitution failure in a belief ascription, I provided a general strategy for identifying relevant false implicata underlying Fregean intuitions in different cases and explained how this strategy works in a problem case posed by Jennifer Saul (1998). I then showed how my account of (2) generalizes to identity belief ascriptions, such as (5)

Lois Lane believes Clark Kent is Superman

and even (6)

Lois Lane does not believe Clark Kent is Superman.

Since (5) is true, following by substitution from (4)

Lois Lane believes Superman is Superman,

its negation (6) must be false. The mistaken Fregean intuition that (6) is true was explained as a case of inherited pragmatic content: if an expression that can be used with a certain pragmatic content is embedded in a compound utterance, the pragmatic content of the compound utterance may incorporate the pragmatic content—rather than the semantic content—of the embedded part. Thus, the pragmatic content of (6) is the negation of the

Summary 105

pragmatic content of (5)—i.e., that (5) is not acceptable verbatim. This view of the inheritability of pragmatic content was then defended against objections by Mitchell Green (1998), L. J. Cohen (1971), and François Recanati (1993). A detailed reply to an objection by Recanati (1993, 2004) on the basis of his Availability Principle then led to a broader discussion of the role of intuitions in semantic theory. I considered how the more general position of Semantic Intuitionism (of which the Availability principle may be seen as one version), according to which semantic theories should conform to what competent speakers say about the meanings (or truth-conditions) of the words and sentences they use, might be supported by the premises that (a) competent speakers ipso facto know what they say, and (b) the semantic intuitions expressed by people who know what they say—that is, their sincere pronouncements about what is said (or about its truth-conditions)—are ipso facto reliable. After distinguishing between implicit and explicit knowledge of what one says, and between full and less than full explicit knowledge of what one says, I showed that the first premise is plausible only with regard to implicit knowledge of what one says (or only partial explicit knowledge), while the second premise requires full explicit knowledge of what one says—so the argument rests on equivocation between these different senses of “knowing what one says.” Thus rejecting Semantic Intuitionism, I argued that the fact that knowing how to speak a language does not guarantee being able to articulate the semantic contents (or truth conditions) of sentences of the language, should come as no surprise. For one thing, there is no reason to suppose that our explicit knowledge of semantics would be any greater than our explicit knowledge of syntax. Moreover, it should be obvious that what really counts when we speak is often not what we actually say. Thus it was seen that it is knowledge of pragmatic content, not semantic content, that is most directly reflected in the articulated semantic judgments of competent speakers. I then offered a diagnosis, suggesting that the popularity of Semantic Intuitionism comes from an inclination to assume the Coincidence View of Meaning and Saying: what competent, sincere speakers mean is typically what they say. Against this I argued for the Divergence View of Meaning and Saying: what competent, sincere speakers mean is typically not exactly what they say. Next I replied to an objection by Stephen Schiffer (1987a), involving iterated applications of the belief operator. If Floyd insists that

106

The pragmatics of substitutivity

(6')

Lois Lane does not realize that Clark Kent is Superman

is true and uses it accordingly, it would seem that (8)

Floyd believes Lois Lane does not realize Clark Kent is Superman

must also be true. But since Floyd surely believes that Lois Lane realizes Clark Kent is Clark Kent, it follows on the Theory of Direct Belief that Floyd believes Lois Lane does realize Clark Kent is Superman must be true. So Floyd winds up both believing and disbelieving the same proposition, even though he does not share Lois Lane’s confusion. Here I explained how (8), though understandably thought to be true, is actually false. I considered an argument by Schiffer in support of (8) and showed that it relied on an implausible version of the principle of disquotation. I concluded the chapter with an examination of two alternative pragmatic accounts of substitution failure. Scott Soames (2002) locates the problem with (2)

Lois Lane believes Superman is a reporter

not in what is conversationally implicated by uttering it, but in what is actually said, though where “what is said” in an utterance typically transcends the semantic content of the words uttered. Against Soames’s position I raised two objections: first, that he apparently abandons the traditional Gricean view, that if a speaker said that p, he must have uttered a sentence that means that p; and secondly, that he fails to provide a satisfactory account of how an utterance of (2) yields its false pragmatic content. Michael Thau (2002) suggests that the false information conveyed by an utterance of (2) is implicated conventionally, not conversationally. Arguing that names such as ‘Superman’ and ‘Clark Kent’ are associated by competent speakers with certain descriptions, Thau concludes that the content of these associated descriptions is part of what is conventionally implicated in utterances of sentences containing them—so that by uttering (2) one conventionally implicates something false such as (13) Lois Lane believes the super-powered protector of Metropolis is a reporter.

Summary 107

I gave three objections to this view. First, it violates the requirement that conventional implicatares not be cancelable. Secondly, the theory does not account for why nothing like (13) gets implicated in the sort of contexts generating the Quinian intuition that (2) is true. Thirdly, there is no way to apply Thau’s proposal to names that do not have any descriptive content.

Chapter 3 Conceptions, belief, and “inner speech”

1. The medium view of conceptions Despite its semantic and pragmatic virtues, the Theory of Direct Belief may arouse some doubts based on certain assumptions about the nature of belief. To highlight the relevant considerations, I would like to reformulate the discussion in terms of a question about belief and conceptions. Lois Lane thinks (let us suppose) that there are two distinct individuals, one of whom is called ‘Superman’, wears a cape, flies, etc., the other of whom is called ‘Clark Kent’, wears a suit, works as a reporter, etc. We may say, then, that she has two conceptions of Superman (assuming, of course, that Superman is Clark Kent). What is the relation between Lois’s conceptions of Superman and her beliefs about him?1 The Medium View is that conceptions belong to some kind of medium in which beliefs are held.2 Of course, this view (or views—what Wittgenstein might call a “picture”) comes in many widely varying versions. The best known version of the Medium View is the classical Fregean view, according to which conceptions—modes of presentation—are the very stuff beliefs (thoughts that we believe) are made of. Lois’s belief that, say, Superman can fly, is said to be composed of (or at least determined by) the sense of the name ‘Superman’ and the sense of the predicate ‘can fly’; and her different conceptions of Superman may be thought of as the two different senses of the names ‘Superman’ and ‘Clark Kent’, out of which her beliefs about Superman are constructed. (Of course, I am restricting my attention

1.

2.

I am using the word ‘conception’ in the pre-theoretical sense characterized by this example. For a comprehensive discussion of many other senses of the word, see Andrew Woodfield (1991). I use the term ‘beliefs’ to refer to propositions believed. Cf. Mark Richard’s “psychological sententialism” (1990: 38). Psychological sententialism would not be as restrictive as the Medium View, if the analogues of terms and predicates that are the constituents of psychological states could be the very individuals and properties that they “represent;” e.g., if “Russellian” propositions could individuate psychological states.

The medium view of conceptions 109

to the cognitive role of Fregean senses, as modes of presentation, ignoring for now their metaphysical and semantical aspects.) But Frege’s is not the only Medium View. According to vehicular versions of the Medium View, Lois’s conceptions of Superman belong to vehicles, of which her beliefs about Superman are the contents. These vehicles of belief are seen as intermediaries by means of which we believe what we believe. Thus, one might say that Lois’s conceptions of Superman (variously called “modes of presentation,” “notions,” “guises,” “Mentalese expressions,” “ways of believing,” etc.) belong not to what she believes, but to how she believes it; that it is via her various conceptions of Superman that Lois has her various beliefs about him, much as it is via his various names that she talks about him.3 Either way, whether conceptions belong to belief vehicles or to beliefs themselves, the Medium View of conceptions is just as much a view of beliefs—as being held in a medium consisting of conceptions. According to the Medium View, to say that Lois thinks Superman can fly is to say that she stands in some relation to a conception she has of him. More generally, having a belief about some individual consists, on the Medium View, in bearing some relation to a conception of that individual. (This is not to rule out other relata, to which some versions of the Medium View advert.) Adopting the Medium View of conceptions (and belief) would thus require rejecting the Theory of Direct Belief, since the latter takes having a belief about an object as bearing some unmediated relation directly to that object (e.g., for Lois to think Superman can fly is, according to the Theory of Direct Belief, for her to bear some relation not to a conception of Superman, but to Superman himself).4 Before considering what reasons there might be for adopting the Medium View, a couple of dialectical clarifications may be helpful. First, the 3.

4.

Vehicular versions of the Medium View are often called “representational(ist)” theories of belief (as in Schwitzgebel (2006)). In other terms, the Medium View includes what Michael Thau (2002) calls “Fregeanism,” as well as what he calls “Guise Millianism,” which would be a vehicular Medium View. Thau shares my rejection of these views, but for very different reasons, too intricate to discuss here. Of course, in claiming that having a belief about an object consists in bearing some “unmediated” relation directly to that object, I do not mean to suggest that no other objects are in any way involved, but only that the relation is not essentially mediated by components of a medium of belief. (See Chapter 2, n. 1.)

110

Conceptions, belief, and “inner speech”

issue at hand concerns only our ordinary, “folk” concepts of belief and conception. That is, this discussion concerns the semantics of the relevant natural language expressions, the aim being to clarify the pre-theoretical (or folk-theoretical) concepts underlying our ordinary use of these expressions. Therefore, empirical discoveries that cognitive scientists might make are not relevant (at least not directly). Since my concern here is with the semantics of belief—a theory of what exactly it is that we say when we attribute beliefs—I am talking not about what belief happens to be—in the particular case of human beings, being as they are—but about what belief has to be, in general—for angels, aliens, and anything else to which we might ascribe beliefs. Thus, familiar arguments for something like the Medium View, but where this is taken as a contingent fact, to be supported by empirical evidence, must not be misunderstood as supporting the Medium View itself. In fact, on any view according to which it is even just possible to have a belief about some individual without bearing some (significant) relation to a conception of that individual, the Medium View is false.5 Secondly, although I think the Medium View is wrong, I shall not argue against it until later in this chapter. For now I shall only show how arguments offered in favor of the Medium View fail to establish their conclusion.6 For some, of course, the Medium View is simply self-evident. One source of such thinking might be the assumption that the immediate objects of mental activity could only be mental objects. For Frege it was simply

5.

6.

So as opposed to what one might have expected, Jerry Fodor’s position, at least as sometimes formulated, actually seems to preclude the Medium View, since he considers the Language of Thought Hypothesis (LOT) empirical: “It is, I take it, an empirical question whether psychological processes are computational processes” (1975: 47); “It goes without saying that RTM [the Representational Theory of Mind, = LOT] needs to make an empirical case” (1987: 26). And as Kim Sterelny explains, “A language of thought is only one possible realization of intentional states” (1990: 29). Richard makes much the same point about the irrelevance of the empirical theories of cognitive scientists (1990: 43–45), but his dialectical stance towards psychological sententialism is in a sense just the opposite of mine (with regard to the Medium View). He sees psychological sententialism, which he endorses, as being commonly considered implausible, in response to which he tries to show that the common arguments against it fail; whereas I see the Medium View, which I reject, as being commonly considered self-evident, in response to which I try to show that the arguments for it fail.

The behavior problem 111

absurd to suppose that the thought that Mount Blanc is 4000 meters high would contain Mount Blanc itself, with all its snowfields. “A thought,” Frege matter-of-factly declares, “is something immaterial and everything material and perceptible is excluded from this sphere” (1918: 20). But this is no argument. Rather, what would generally be taken as Frege’s argument in favor of the Medium View (and what has probably been the most influential argument for the Medium View) would be one based on the premise that substitution of coreferential names does not always preserve truth—a premise shown in the last chapter to be unfounded.7 Even without relying on any appeal to substitution failure, one might be led to the Medium View by the common adverbial idiom of belief ascription—we readily speak of how a person thinks what he thinks, on analogy with how a person says what he says. For instance, regarding Lois’s beliefs about Superman one might say, “thinking of him as Superman, she thinks he can fly.” Leaving for later a discussion of the analogy of thought to speech, for now I would like to examine two arguments in favor of the Medium View apparently inspired by this idiom of “thinking-of-as.”

2. The behavior problem 2.1. The problem The most common post-Fregean arguments for the Medium View are based on ordinary explanations of behavior by appeal to conceptions and by the use of the “thinking-of-as” idiom.8 To account, for instance, for the fact that Lois behaves one way when she sees Superman in one of his roles, and a completely different way when she sees him in his other role, it is natural to 7.

8.

Frege introduces modes of presentation (and subsequently “senses”) on the basis of the considerations in the first paragraph of (1893). These include the opening argument (presumably against taking equality as a relation between objects), which is built on the premise that “a = a and a = b are obviously statements of differing cognitive value” (1893: 142). I take this premise as equivalent to the claim that it is possible to believe that a = a without believing that a = b—which is to say that in certain belief ascriptions coreferential names are not interchangeable salva veritate (Chapter 1, note 2.) A good example of this kind of argument is given by Mark Richard (1990: 126; 1997). See also David Braun (2001); Crimmins (1992: 32–34); Fodor (2008: 67–68, 86); and François Recanati (1993: 341).

112

Conceptions, belief, and “inner speech”

say (as above) that she has two conceptions of him. Moreover, assuming her reactions to him are to be accounted for in terms of her beliefs about him, there seems little choice but to construe these conceptions as part of her beliefs about him, or at least as part of the means by which she has those beliefs. Suppose, for example, that whenever he greets her in his Superman role she reacts by enthusiastically exclaiming, “Hi, Superman!” but whenever he greets her while assuming his Clark Kent role she reacts by rolling her eyes and saying, “Bye, Clark.” Since her thinking that he has greeted her is common to both situations, it cannot by itself account for her different reactions. However, her different reactions seem tied to her different conceptions of him. Thus, when she says “Hi, Superman!” one might account for her behavior by saying that she thinks, thinking of him as Superman, that he has greeted her; and when she says “Bye, Clark” one might account for this different behavior by saying that she thinks, thinking of him as Clark Kent, that he has greeted her. Whether her conceptions of him (indicated by the “thinking-of-as” clauses) are actually part of her beliefs about him, or merely part of the vehicles by which she has those beliefs, they may be used either way to account for her two distinct patterns of behavior when faced with the same individual.9 So the Theory of Direct Belief faces two challenges. If conceptions are not parts of beliefs or of belief vehicles, then (i) what are they? And (ii) how else are we to explain the kind of behavior described?

9

Thau seems to argue that if substitution failure can be accounted for pragmatically, then the behavior problem can be solved in the same way (Thau: 104–107). For if the reason that we react differently to utterances of the sentences ‘Superman just said “Hello"‘ and ‘Clark Kent just said “Hello"‘ is because they differ not in semantic content, but rather in the information pragmatically imparted, then so, too, when we use those sentences to explain Lois’s behavior, her behavior will be explained not by the identical semantic content of the sentences, but by the different information pragmatically imparted by each sentence’s utterance. This may work on a metalinguistic level, as an account of how we talk about behavior, but not on the object level of how actions are caused (in part) by beliefs. The question remains, if the belief that Superman just said “Hello” is the very same as the belief that Clark Kent just said “Hello,” what then is the cause of the observed variation in Lois’s behavior?

The behavior problem 113

2.2. The Higher Order View of conceptions Meeting the first of these challenges amounts to providing an alternative to the Medium View of conceptions. This is not hard, since the Medium View of conceptions, despite its overwhelming popularity, is not the only view of conceptions. Instead of construing beliefs or belief vehicles as constructed out of conceptions, one might think of having a conception as consisting in having certain beliefs. On this Higher Order View of conceptions it is conceptions that are to be explained in terms of beliefs, rather than the other way around. There are many ways such an explanation might go.10 In the case of Lois’s two conceptions of Superman the key is the belief of hers that led us in the first place to say that she has these two conceptions—viz., her belief that there are two distinct individuals, one of whom is called ‘Superman’, wears a cape, flies, etc., the other of whom is called ‘Clark Kent’, wears a suit, works as a reporter, etc. Her believing this existential generalization is alone enough to account for her having these two conceptions. If we wanted further to explain what a single conception is, we could say that it is a set of predicates (or properties) that one believes are jointly instantiated.11 Thus, the set of the predicates ‘is called ‘Superman’’, ‘wears a cape’, ‘flies’, etc., would be one of Lois’s conceptions; the set of the predicates ‘is called ‘Clark Kent’, ‘wears a suit’, ‘works as a reporter’, etc., would be another.12 However the explanation may go (which I leave open), the Higher Order View of conceptions provides advocates of the Theory of Direct Belief with a fully compatible alternative to the Medium View; for it allows us to distinguish Lois’s two conceptions of Superman without saying she has her beliefs via them.

10. Cf. Woodfield (1991). 11. This idea of conceptions is like Grice’s idea of dossiers (1969: 141) and Perry’s idea of files (1980: 84–89). 12. Note that these particular sets are not disjoint; Lois’s two conceptions of Superman have many predicates in common, such as ‘has dark hair’. Also, these particular conceptions happen to be unique conceptions, in that Lois believes each set of predicates to be uniquely satisfied.

114

Conceptions, belief, and “inner speech”

2.3. A solution to the problem Since the Higher Order View does not locate conceptions in beliefs or belief vehicles, the need remains for an account of each of Lois’s two different reactions to Superman’s greetings; if her beliefs about Superman are none other than her beliefs about Clark Kent (all being beliefs about one and the same object), then why does she sometimes react to him one way and other times react to him another way? This problem is not hard to solve once we drop the unwarranted assumption that Lois’s reactions to Superman must be accounted for entirely on the basis of her singular beliefs about him. Although her singular beliefs about him are the same in both situations, her total cognitive state in each situation may be distinguished by other beliefs she has. In the situation where she says “Hi, Superman!” for example, she has beliefs about his cape—e.g., that it is hanging from the shoulders of the man currently facing her—that she does not have in the situation where she says “Bye, Clark.” Likewise, in one situation she believes the name ‘Superman’ refers to the man currently facing her, whereas in the other she does not; and so on. Alternatively, her cognitive state in each situation can also be distinguished by certain general beliefs of hers. In the situation where she says “Hi, Superman!” for example, she has the general belief that there is someone in front of her named ‘Superman’, not named ‘Clark Kent’, wearing a cape, not a suit, etc.—a belief that she does not have in the situation where she says “Bye, Clark.” To see more clearly how the proposed solution to the behavior problem works, recall the alternative solution provided by the Medium View. Lois’s “Superman behavior”—saying “Hi, Superman!” and so on, when he greets her in his Superman role—was to be accounted for by saying that thinking of him as Superman she thinks that he has greeted her; and her “Clark Kent behavior”—saying “Bye, Clark” and so on, when he greets her in his Clark Kent role—was to be accounted for by saying that thinking of him as Clark Kent she thinks that he has greeted her. But then how exactly is her thinking of him as Superman supposed to lead to her Superman behavior, rather than her Clark Kent behavior? Presumably, thinking of him as Superman (and not as Clark Kent) would lead her to ascribe to the person who just greeted her the various properties belonging to her Superman conception of him—is called ‘Superman’, wears a cape, flies, etc.—as opposed to the properties belonging to her Clark Kent conception. And so, thinking that the person who just greeted her is called ‘Superman’ (not ‘Clark Kent’),

Suspended belief 115

wears a cape (not a suit), flies (is not earthbound), etc., she would naturally react with her Superman behavior rather than her Clark Kent behavior. Notice, however, that what ultimately leads to Lois’s Superman behavior—e.g., her belief that the person who just greeted her is called ‘Superman’ (not ‘Clark Kent), wears a cape (not a suit), flies (is not earthbound), etc.—are beliefs that we can easily attribute to her without resorting to any talk of “thinking of as.” In fact, it is only in virtue of her having such beliefs (that there is someone in front of her wearing the Superman cape, named ‘Superman’, etc.) that we might be inclined to say that she has these beliefs about the person in front of her thinking of him as Superman. So, the Medium View is not necessary in order to account for the fact that someone might display distinct patterns of behavior corresponding to distinct conceptions of the same individual. This is not to say, of course, that the Medium View is wrong. But it does mean that arguments for the Medium View on the grounds that it is the only view that can account for this sort of data are not sound.

3. Suspended belief Another argument for the Medium View that is based on the “thinking-ofas” idiom is the argument from suspended belief, originally due to David Kaplan (1969) and later put forward in a different form by Nathan Salmon (1986: 92-101), whose version I will discuss here. In the case Salmon considers, Elmer believes at one point that there are two individuals named ‘Bugsy’, both of whom are dangerous; and subsequent events lead him to change his mind, in a way he might well express by saying, “One of the guys named ‘Bugsy’ is dangerous, but the other may or may not be—I no longer know what to think.” Unbeknownst to Elmer, there is in fact only one individual named ‘Bugsy’ whom he is thinking of and referring to in the relevant thoughts and utterances, respectively, for Bugsy has assumed a disguise by which Elmer has been fooled. So on the one hand—“thinking of Bugsy in one way”—Elmer continues to consider this individual dangerous; while on the other hand—“thinking of Bugsy in another way”—he says he has changed his mind and is now uncertain of whether this individual is dangerous. How are we to reconcile Elmer’s continued belief about Bugsy with his change of mind and ensuing suspension of belief? Salmon frames the discussion in terms of singular propositions. The singular proposition that Bugsy is dangerous is (or can be thought of as) the

116

Conceptions, belief, and “inner speech”

ordered pair of which Bugsy is the first member and the property of dangerousness is the second member: . Suppose—as the Theory of Direct Belief requires—that having a belief about something consists simply in bearing a certain binary relation to a singular proposition containing that thing as its first member. So for Elmer to think Bugsy is dangerous would be for Elmer to stand in the belief relation to the ordered pair consisting of Bugsy (the individual himself) and the property of dangerousness. But then for Elmer to change his mind about whether Bugsy is dangerous and become uncertain of it would be for him to change his mind about and become uncertain of the very same singular proposition. How could he believe this proposition on the one hand and also suspend belief in it on the other? Salmon’s response is to bring in “guises” or “modes of apprehension” that mediate between us and the singular propositions we believe. While retaining the view that to have a belief about something is to bear a certain binary relation toward a singular proposition containing the thing the belief is about, he goes on to analyze that relation in terms of a ternary relation between believers, propositions, and modes of apprehension—thus rejecting the Theory of Direct Belief for a vehicular version of the Medium View. Elmer continues to believe the singular proposition that Bugsy is dangerous, Salmon explains, in that there is a mode of apprehension for that proposition to which Elmer and the proposition remain suitably related: (x)[Elmer grasps that Bugsy is dangerous by means of x & BEL(Elmer, that Bugsy is dangerous, x)], where Elmer, the proposition, and a mode of apprehension are suitably related (by BEL) if, approximately, Elmer is inclined to assent inwardly to the proposition when apprehending it in that way. At the same time, Elmer also comes to withhold belief from the very same proposition, in that there is a(nother) mode of apprehension for that proposition, for which it comes to be no longer the case that Elmer, the proposition, and this other mode of apprehension are in the same way suitably related: (x)[Elmer grasps that Bugsy is dangerous by means of x & ~BEL(Elmer, that Bugsy is dangerous, x)]. (1986: 112) Salmon maintains that nothing short of such a move to the Medium View will do:

Suspended belief 117 Without some relativized, ternary notion, and the resulting distinction between withholding belief and failure to believe, the attempt to describe Elmer’s complex doxastic state with respect to the relevant singular proposition breaks down. The only thing one can say using the binary notion of belief—to wit, that Elmer does believe the proposition that Bugsy is dangerous—is highly misleading at best. (1986: 112-113)

Salmon is right, of course, that it would be highly misleading to say merely that Elmer believes Bugsy is dangerous and to leave it at that. But that is not all that can be said without resorting to modes of apprehension and a ternary notion of inclination to inward assent. Just as the force of the behavior problem comes from mistakenly assuming that an agent’s behavior with regard to another individual must be accounted for entirely on the basis of her singular beliefs about that individual, so too, the force of Salmon’s argument here comes from the assumption that an agent’s doxastic state with regard to another individual must be describable entirely in terms of her beliefs directly about that individual. However, aside from the singular proposition that (a)

Bugsy is dangerous,

Elmer has many other beliefs involved in his change of mind. Before his change of mind he also believed that (b) (c) (d) (e) (f)

he knows two guys named ‘Bugsy’, the first guy he met called ‘Bugsy’ is dangerous, it is reasonably certain that the first guy he met called ‘Bugsy’ is dangerous, the second guy he met called ‘Bugsy’ is dangerous, and it is reasonably certain that the second guy he met called ‘Bugsy’ is dangerous.

Suppose now, for ease of exposition, that Elmer would say that it is the second guy he met named ‘Bugsy’ whose dangerousness he has become uncertain of. Then after his change of mind, Elmer continues to believe (a), (b), (c), and (d), but no longer believes (e) and (f); and although he has come to believe that (f) is false, he does not believe (e) is false. Thus, it is actually (e), not (a), with regard to which he suspends belief. It may be tempting to locate Elmer’s suspended belief in (a), since that is perhaps the most salient singular proposition about Bugsy. But there is no reason to

118

Conceptions, belief, and “inner speech”

assume that Elmer’s befuddlement must be described in terms of his relation to (a) rather than (e). Thus, as long as we do not restrict our attention to singular propositions about Bugsy, we can give a perfectly adequate description of Elmer’s complex doxastic state without resort to any talk of modes of apprehension or inward assent. Moreover, by locating Elmer’s suspended belief in (e) rather than (a), this account of Elmer’s befuddlement has the additional virtue of enabling us to take suspended belief straightforwardly, as the conjunction of neither believing nor disbelieving a certain proposition—rather than resorting to a non-standard construal of suspended belief, as Salmon proposes, in terms of modes of apprehension and inclinations to inward assent, whereby suspended belief consists in grasping a proposition p by a mode of aprehension by which one “is disposed neither to inward agreement nor to inward disagreement with respect to p” (1986: 172, n. 1). Nevertheless, one might object that it is precisely by locating Elmer’s suspended belief in (e) rather than (a) that my account fails to address the question at hand, viz., how could Elmer both continue to believe, and at the same time also suspend belief in, the very same singular proposition—(a)? As Salmon puts it, the challenge is “to describe Elmer’s complete doxastic state with respect to the relevant singular proposition” (emphasis mine)— that Bugsy is dangerous—whereas what I have described is Elmer’s doxastic state with regard to another individual—Bugsy. On the face of it the story of Elmer presents us with a contradiction about (a): Elmer believes it and also suspends belief in it, so assuming that suspending belief in a proposition precludes believing it, Elmer believes (a) and does not believe (a). How does shifting attention to Elmer’s attitudes to (e) address the question about (a)?13 13. Cf. Kripke’s objection to certain replies to his puzzle about belief: It is no solution in itself to observe that some other terminology, which evades the question whether Pierre believes that London is pretty, may be sufficient to state all the relevant facts. I am fully aware that complete and straightforward descriptions of the situation are possible and that in this sense there is no paradox. … But none of this answers the original question. (1979: 369) Similarly, according to the objection being considered, it is no solution to the puzzle about Elmer’s suspended belief in (a) to provide a complete and straightforward description of the situation by observing his attitudes to some other propositions.

Suspended belief 119

Although this objection is misguided, it raises some interesting questions. It is misguided, because it misconstrues the question at hand as being primarily about Elmer’s attitude to the proposition that Bugsy is dangerous. But as Salmon himself explains, The most important aspect of Elmer’s Befuddlement is the fact that Elmer has changed his mind about something and withholds belief where he once had an opinion. This aspect of the example—the change of mind from having an opinion to suspension of judgment—poses the most pressing and difficult philosophical problems, and is at the same time the most philosophically illuminating feature of the example. (1986: 170, n. 1)

The fundamental question here is how to account for Elmer’s change of mind—never mind regarding exactly what.14 It is only if one assumes from the start that this change of mind was in regard to a particular proposition, that one would think of the question as being about his attitudes to that particular proposition. But making such an assumption just begs the question of how best to describe Elmer’s change of mind. In any case, my account actually does address the question of what to say about Elmer’s attitudes to (a), for the simple reason that locating Elmer’s suspension of belief in (e) forestalls the apparent contradiction that comes from assuming that it is in (a). But that cannot be all of it. Recall the inference implicit in the story: (i) (ii) ?

Elmer believes (a) and also suspends belief in (a). Suspending belief in a proposition precludes believing it. Elmer believes (a) and does not believe (a).

In taking (e) rather than (a) as the object of Elmer’s suspension of belief, I have blocked the inference by rejecting the first premise—specifically, the second conjunct of that premise, that Elmer suspends belief in (a)—while accepting the second premise. I do so on the basis of what I shall call the simple construal of suspended belief, according to which suspending belief in a proposition consists simply in believing neither it nor its negation. This way of thinking about suspended belief clearly entails premise (ii), so avoiding the self-contradictory conclusion requires rejecting premise (i); and since I agree that Elmer believes (a), I therefore reject—on this simple construal of suspended belief—that Elmer suspends belief in (a). 14. This is where the puzzle about Elmer differs from Kripke’s puzzle: the latter is about a particular belief ascription, but the former is not.

120

Conceptions, belief, and “inner speech”

But it may be plausible—even more plausible—to construe suspended belief differently. We might start by taking Elmer’s professed ambivalence with regard to (a) as a case of suspended belief par excellence. That is, recognizing that there is something interesting going on in Elmer with respect to (a)—something that manifests itself in his saying, “Maybe Bugsy is dangerous, maybe he isn’t; I no longer know what to think”—we might start by taking this as “suspended belief,” ex hypothesis, and then consider what it comes to. This would make premise (i) true (assuming still that Elmer believes (a)), which would then require rejecting premise (ii). I shall call such a construal of suspended belief—according to which Elmer does suspend belief in (a)—a robust construal of suspended belief. If it is reasonable to suppose that there is a coherent robust construal of suspended belief, then my account of suspended belief according to the simple construal is not complete without a complementary account of robust suspended belief—which, given that Elmer continues to believe (a), would have to be one that makes premise (ii) false. Why should we think in the first place that there is a coherent robust construal of suspended belief? Why not just construe suspended belief the simple way and leave it at that? It is not hard to understand why someone might feel dissatisfied with the simple construal of suspended belief. Suspending belief sounds as if it ought to involve something other than just neither believing nor disbelieving—after all, there is much that we neither believe nor disbelieve that we have not even thought of, or that we are not even capable of thinking of. If we suspend belief in a proposition, it feels as if there should be some positive epistemic component—perhaps some kind of considering, or wondering, or deliberate refraining from judgment. Intuitively it seems that there must be something going on in Elmer’s suspended belief that is not just a failure to believe.15

15. Thus, after rejecting the simple construal of suspended belief, Jane Friedman considers (and rejects) various ways of strengthening it. However, since premise (ii) must be true on any construal that makes suspended belief in a proposition a sufficient condition for believing neither it nor its negation, and since premise (i) must be true on any robust construal of suspended belief (ex hypothesis, assuming Elmer believes (a)), none of the options Friedman considers can be a coherent robust construal. Interestingly, Friedman explains that she restricts her attention to strengthened versions of the simple construal only for the sake of argument, and that she would eventually prefer a robust construal that makes premise (ii) false.

Suspended belief 121

If Elmer robustly suspends belief in (a) while at the same time believing (a), what could this robust suspension of belief consist in? Salmon answers this question by providing a robust construal of suspension of belief in terms of modes of apprehension—thereby retaining premise (i) but rejecting premise (ii). But is this, as he seems to insist, the only way to construe suspended belief robustly? Sean Crawford offers an alternative treatment of suspended belief, one that does not require resorting to anything like modes of apprehension. On his view, suspending belief in a proposition requires not only believing neither it nor its negation, but also believing that one believes neither it nor its negation. So according to Crawford, Elmer “has not in fact suspended judgement; he only thinks he has” (2004a: 189; 2004b: 227).16 Making it false that Elmer suspends belief in (a), this construal of suspended belief is not robust—but it does suggest one that is: A suspends belief in p if and only if A believes that A believes neither p nor its negation. Or, better yet, on the assumption that suspension of belief is more about uncertainty than a mere lack of belief: A suspends belief in p if and only A believes that neither p nor its negation is reasonably certain.17 This way of thinking about suspension of belief is robust—it makes it true that Elmer suspends belief in (a), thereby capturing Elmer’s professed ambivalence. Yet, it avoids the conclusion that Elmer believes and does not believe (a), by denying premise (ii)—that suspending belief in a proposi-

16. I am extrapolating on the basis of what Crawford says regarding a slightly different example. Crawford has since suggested that it might be too strong to say Elmer thinks he has suspended belief in (a), but it would be enough to note that Elmer wrongly believes that he does not believe (a). 17. This seems to be the sort of analysis sought by Friedman (2011), whose declared goal is “to show that a subject suspends judgment only if he has an indecision-representing attitude,” where indecision-representing attitudes are those “that express or represent or are a subject’s indecision about the truth of some propositions.”

122

Conceptions, belief, and “inner speech”

tion precludes believing it (on the assumption, of course, that one can believe a proposition that one thinks is not reasonably certain).18 Yet another way to construe suspended belief robustly without relying on modes of apprehension and the like would be by taking it as a more or less independent primitive—a sui generis attitude resembling wondering whether or being disinclined to affirm or deny. Taken this way, suspended belief would not be defined simply in terms of belief, as it customarily has been, nor even in the same terms in which belief is defined (as Salmon proposes). The tradition of taking suspended belief as trivially definable in terms of belief seems to compel Salmon to try at least to keep the two straightforwardly definable in the same terms; but such a constraint is not necessary. In either case, whether robust belief is analyzed in terms of beliefs about certainty or understood as an independent sui generis attitude, the real problem here is how to account for the logical and nomological links between propositional attitudes. Of course, not all the attitudes are closely linked to each other; believing that something is the case, for example, does not seem closely connected to wanting it to be the case. But belief does seem closely connected to some other attitudes, especially disbelief and suspended belief. Naturally it is tempting to express such close links as logical ties, whereby one attitude could be seen as a logical function of another (and its content)--hence the traditional simple analysis of suspended belief as believing neither the proposition nor its negation. Recognizing the need to render the connection between the two somewhat looser, Salmon tries to keep the connection as tight as he can by construing both attitudes as simple logical constructions out of the same primitives (guises and inward assent). But often the links between closely connected attitudes are not strict logical links, neither direct nor via selected primitives, but rather, strong yet defeasible ceteris paribus links. Other things being equal, belief and suspended belief are indeed mutually exclusive; but where the believer is unaware of relevant information, other things are not necessarily equal. Thus, the assumption that belief precludes suspended belief can be rejected without recourse to any version of the Medium View.

18. Crawford apparently prefers to preserve the intuition behind premise (ii), at the expense of the intuition that Elmer has indeed suspended belief in (a).

The inner speech picture of thought 123

4. The inner speech picture of thought In light of the weakness of the arguments in favor of the Medium View, I would like to turn to what I take to be the true source of the Medium View’s wide appeal, and that is the underlying picture of thought as “inner speech”—the Inner Speech Picture of Thought.19 There is a long and venerable tradition of thinking of thought as a kind of “private assertion,” a sort of unpronounced “talking to oneself.” Plato, for instance, describes thinking as “a talk which the soul has with itself” (1990: 190a) and judgment as “speech spoken, not aloud to someone else, but silently to oneself” (1990: 190a). More recently, in one of the twentieth century’s most influential works in the philosophy of mind, Wilfrid Sellars suggests that “overt verbal behavior is the culmination of a process which begins with ‘inner speech,’” (1956: 318). Moreover, modern cognitive science has developed largely on the basis of some form or another of Jerry Fodor’s “language-of-thought” hypothesis (connectionism notwithstanding), according to which “having a propositional attitude is being in some relation to an internal representation” (1975: 198). Against this background it is not surprising that Salmon would rely on what he calls “the traditional conception of belief as inward assent” (1986: 130). Although the general picture of thought as inner speech underlies a wide variety of particular views, it is possible to specify some of the picture’s more characteristic features. First and foremost, there must be some kind of psychological analogues to expressions—“senses,” “modes of presentations,” “Mentalese expressions,” etc. As with linguistic expressions, these expression-analogues must be governed by some sort of “syntax” by which they can be compounded and can be sorted into analogues of sentences and analogues of subsentential parts—including analogues of referring expressions and analogues of predicate expressions. Likewise, there must be some sort of “semantics” assigning contents to them (preferably compositionally—so that the contents of compound expression-analogues are determined by the contents of their components and the structure by which they 19. I do not claim that the Medium View entails the Inner Speech Picture of Thought, or vice versa, or that proponents of the first actually argue for it on the basis of the second, but only that the second is likely in many cases to underly one’s inclination toward the first—just as Wittgenstein sees the Augustinian picture of language as underlying many claims he rejects. Regarding explicit arguments for the Medium View on the basis of inner speech, see Edouard Machery (2005).

124

Conceptions, belief, and “inner speech”

are compounded).20 Moreover, the contents assigned to sentence-analogues should be the same sort of contents as those assigned as the semantic contents of sentences, so that it will be straightforwardly possible to say what someone thinks and to believe what someone says. And there must be acts to perform with them analogous to referring and asserting. This picture is most clearly associated, of course, with language-of-thought hypotheses, but it equally underlies congruent systems of senses, modes of presentation, etc. (whether as the stuff of beliefs or of belief vehicles).21 The rough picture I have sketched of thought as inner speech is just that—a rough picture, not a theory. As long as the corresponding metaphor is not adequately fleshed out, it remains only an extended metaphor, not a serious model. How, then, is the metaphor to be filled out and cashed in? To start with, what exactly are the psychological analogues to expressions? It is notoriously difficult to say anything about these entities that goes beyond the metaphor itself.22 And the enormity of the task grows as we look for mental analogues to referring and asserting. Talk of such things as “tokenings of Mentalese sentences” and “inward assent” does not explain the analogy but only extends it. This already points to an important disanalogy, in that words and utterances are relatively concrete and familiar; they are easy to spot and to talk about, even pre-theoretically—unlike their analogues under the thought-as-inner-speech metaphor. This disanalogy manifests itself at every level of language—syntactic, semantic, and pragmatic. Although we can easily see how expressions are combined to form sentences—as a temporal sequence of events in spoken language, as a spatial sequence of characters in written language—even this most superficial syntactic structure cannot be seen among the expressionand sentence-analogues in thought. At the level of semantics the disanalogy is even stronger: how a word in natural language comes to have its meaning clearly has to do with social customs regarding the use of that word; but in the head there is no room for such a thing.23 Thus, the explanation of how

20. See, e.g., Fodor (1985: 18); Horst (1996: 25); Fodor and Pylyshyn (1988: 12). 21. As in the discussion of the Medium View, the issue here is conceptual, not empirical; the Inner Speech Picture of Thought is to be understood as a picture of what thought is in general, not merely of what thought happens to amount to in the particular case of Homo sapiens. 22. See, e.g., Schiffer (1990). 23. The point here does not depend on any version of Wittgenstein’s private language argument. Rather, it is just that there cannot be within one’s mind any

The inner speech picture of thought 125

these expression-analogues acquire and possess their contents must be radically different from that of the meaning relation for (natural) language. And at the level of pragmatics the analogy seems to break down altogether, for speech acts such as assertion are essentially tied to communicative intentions—such as intending to induce a certain belief in the hearer—for which the mind is too sparsely populated to provide any counterparts. As long as the metaphor of thought as language is not adequately fleshed out, its theoretical usefulness remains severely limited. As helpful as it may be as a heuristic for cognitive scientists interested in developing models of particular cognitive processes, it cannot go very far towards providing a theory of belief or meaning, since it only shifts the basic questions that such a theory is supposed to answer. If believing a particular proposition consists in being disposed to bear some relation that is an analogue of asserting to some entity that is an analogue of a sentence, then questions about belief reappear as questions about the mental analogue of asserting, and questions about the meaning of a sentence reappear as questions about the content of sentence-analogues.24 Moreover, it is questionable whether the metaphor of thought as language can be fleshed out altogether, due to cases of what might be thought of as “nonrepresentational” belief, where the believer seems to have a belief about something without having any inner representation of that thing. In general, the Inner Speech Picture of Thought runs into trouble wherever one might have a thought for which there seems to be no “inner sentence” (or utterance), analogous to the natural language sentence we would use in reporting the thought. Though it will suffice for my argument to present just one example of a thought without an inner sentence-analogue, I shall nevertheless present several kinds of such examples, not only to increase my chances of hitting on a convincing one, but also because I think that the of the social custom or interaction in virtue of which words come to have their meanings. 24. My point here is not quite the same as the charges of circularity commonly raised against computational or representational theories of mind (cf. Horst (1996: 119); Sterelny (1990: 32–33). Whereas those theories are supposed to be theories of intentionality, the theories I address aim to account only for belief (or for the semantics of belief ascriptions). But a theory that purports to explain believing a proposition in terms of some even more obscure relation (such as an inclination toward inward assent) to some even more obscure object (such as an inner representation of a proposition) does not provide much explanatory value.

126

Conceptions, belief, and “inner speech”

weight of the examples as a whole sheds more doubt on the Inner Speech Picture of Thought than would any single contestable counterexample by itself. A natural starting point for many would be thoughts about sense data, the paradigmatic subjects of Russellian (singular) propositions.25 Consider a thought about a particular sense datum—say, that it is interesting. What might serve as the inner name-analogue for the sense datum in this thought? We may assume that the individual possesses no description, in natural language or otherwise, uniquely picking out the sense datum in question, for it may be entirely new to the individual—a previously unexperienced shade of orange, say—and it may be one of a number of such new sense data all simultaneously experienced. It may be tempting to suggest some trivial description, saying, for instance, that the sense datum is thought of as “what I am now thinking of.” But it seems possible that the individual might not be thinking of the sense datum as the thing he is thinking of; he need not be thinking about his own thinking at all (just as it should be possible to refer without making mention of one’s referring.) Alternatively one might hold that such a thought is thought via a simple demonstrative such as ‘this’ (or an inner analogue of ‘this’). But how could such a demonstrative secure its purported demonstratum? In actual speech there are conventions of pointing and of exploiting conversational salience and other public contextual features to tie the use of a demonstrative to its referent, but in private thoughts about private experiences there seems to be no room for such mechanisms.26 However tempting it may be to argue that an internal ‘this’ can be directly tied to an internal demonstratum simply by some internal salience (whatever that may be), this would amount to assuming that one’s thinking of something ipso facto affords one an inner expression by means of which the thing is thought of—which amounts to begging the question. Moving more into the open, consider thoughts about newly encountered sensory stimuli that may be considered more or less public (or at least shared), such as sounds and smells. Suppose, for instance, upon tasting a new, exotic food, I notice a number of new flavors, thinking of one of them 25. Not assuming that there really are such things as sense data, or any other private sensory objects, I direct my remarks here only to those who are not convinced that there are no such things. 26. I do not mean to be urging the more severe Wittgensteinian view, which precludes “private” phenomena altogether; mine is just a trivial point about the public and social aspects of demonstrating.

The inner speech picture of thought 127

that it is exceptionally delicious. Again, as in the case of new private sensory experiences, there seems to be no nontrivial description by means of which I have this thought. Some would assume that if I could, indeed, individuate this flavor, so that it could be the object of my thought—if I could discriminate between it and others—that just means that I have some internal representation of it. And confronted with the difficulty of actually finding such a representation, some would simply define it as whatever it is in virtue of which the discrimination is possible. Clearly this, too, begs the question.27 Prima facie evidence against the Inner Speech Picture of Thought is not confined to the sensory realm. Other thoughts of objects for which internal representations seem to be lacking include certain thoughts of feelings, such as the feeling of deja-vu, or the feeling I have when I am coming down with the flu. Although such feelings can be thought of after the fact as “the feeling I had when … ,” or in some cases by name (deja-vu, nausea, melancholy, etc.), the first time one experiences such a feeling no such name or nontrivial description is available, yet one might already be thinking about it—thinking, say, how unpleasant it is.28 If it still seems doubtful that one might think about an object without yet having a fully formed representation of it, by means of which the object is thought of, it may help to consider some modes of thought prior to belief. What does it take merely to focus one’s attention on an object? Surely this could be done without at the same time thinking anything about the object. That is, there is a nonpredicative way of thinking of, a sort of noticing or attending to, that does not involve believing the object to be any particular way or another. In fact, this kind of thinking may in some cases be preparatory for acquiring beliefs about an object. This nonjudgmental focusing of 27. (a) As Wittgenstein urges (albeit in a different context), “Don’t say: ‘There must be …’— but look and see whether there is …” (1958: §66). Cf. Russsell’s admonition: “The method of ‘postulating’ what we want has many advantages; they are the same as the advantages of theft over honest toil. Let us leave them to others and proceed with our honest toil” (1919: 71). (b) Clearly I am assuming that representations represent things to an individual that interprets them, since that is how they must be if they are to be the objects of inward assent. For an alternative view of representations, see Haugeland (1981). 28. Thinking how unpleasant it is must be distinguished, of course, from thinking (saying to oneself, as it were) “it is unpleasant”—which would be an instance of the previous case, of thoughts purportedly via inner demonstratives.

128

Conceptions, belief, and “inner speech”

one’s attention could be thought of as a mental analogue to the linguistic act of referring, whereby an object is specified for the sake of subsequently predicating something of it. To require something like a subject term for having a belief about the object would just be to carry the referencepredication analogy too far (in effect presupposing the Inner Speech Picture of Thought). Further and perhaps stronger evidence against the Inner Speech Picture of Thought can be found in our practice of ascribing belief for the sake of explaining complex patterns of behavior. The simplest such cases are like the one discussed at the beginning of Chapter 1, where Lois Lane believes Superman is a reporter is accepted as an indication of how well Superman can disguise himself. In this particular case, however, one may hold that Lois Lane has this belief of Superman only in virtue of her having it via another representation of him, viz., the name ‘Clark Kent’. But the embedded sentence in a correct belief ascription is not always at most just one substitution (of coreferential terms) away from a sentence the believer would accept. For the connection between one’s beliefs and one’s behavior, verbal and otherwise, is not always so simple and straightforward. For example, despite the controversy over ascriptions of grammatical knowledge, they at least illustrate well this kind of belief talk. When a developmental psycholinguist says, for instance, that native English speakers usually realize that English is head-first by the time they are twenty-four months old, but not by the time they are two months old, she clearly need not suppose that twenty-four-month-old English speakers have any sort of representation of the concept of being head-first, or even of the English language. Rather, she may ascribe this belief to them merely in order to account for a complex pattern in their (verbal) behavior. Even though it is controversial in the particular case of grammatical knowledge whether any nonlinguist really believes the various grammatical rules that linguists formulate, it would be hard to deny the prevalence and importance of this kind of belief talk in general. I might say of an infant that he used to think that physical objects cease to exist when they go out of view, though now he realizes that physical objects can exist unseen, but I need not (and would not) suppose that he has representations of (or even concepts of) physicality, existence, or visibility; my point might be only that he reacts in certain ways when objects go in and out of his view. Similarly, I am certain that many nonphilosophers think that Leibniz’s Law (in

The inner speech picture of thought 129

at least one version) fails in epistemic contexts, though I wouldn’t suppose that they know what Leibniz’s Law is or what epistemic contexts are. Unlike in the previous case, those to whom I attribute this belief about substitution failure typically do have a rich enough language to define the expressions in terms of which I ascribe the belief. But although this suggests that they could learn the relevant concepts, it does not show that they have these concepts (or that representations of these concepts are playing any part in their having the belief). Indeed, there are even true belief ascriptions where the subject does have all the relevant concepts and representations, and yet it cannot be said that the belief is held in any way through or in virtue of them—as when one explains a child’s behavior by saying that she thinks crying will make her parents give in, even though she would sincerely deny it, or when a therapist describes a patient as believing that his mother never loved him, even though the patient sincerely disagrees. Our liberal practice of ascribing purposes and goals provides further illustration of how beliefs can be ascribed to explain behavior without assuming or implying the existence of any inner sentence-analogue. For often where we would say that a person is doing something in order to achieve a certain goal, we would say that she thinks that her doing that thing will facilitate her achieving that goal. We might even say that she believes the conditional, that if she does that thing, then she will (probably) achieve her goal. Yet we frequently do this without expecting the subject to agree with our account of her behavior, and sometimes without even expecting the subject to have any idea at all of the motives we attribute to her. So, for instance, if someone is behaving belligerently in order to enhance her social status, we could just as well say that she thinks that if she behaves belligerently, it will enhance her social status; yet we need not at the same time suppose that she has any concept or representation of belligerence or social status. Although it may be possible to concoct some sort of inner sentenceanalogues for some of the kinds of belief discussed above, on the whole these examples seem to point away from the Inner Speech Picture of Thought.29

29. There is, of course, much more that can and has been said against various versions of the Inner Speech Picture of Thought, as in the work of Daniel Dennett and Robert Stalnaker, the adequate discussion of which would take me too far afield.

130

Conceptions, belief, and “inner speech”

5. Thinking in words30 5.1. Silent uttering Why, then, is the metaphor of thought as language so popular? Why is it so tempting and so easy to slip into talk of “tokenings of mental sentences,” for instance, or “inward assent to modes of apprehension?” I think the attractiveness of the inner speech picture of thought is to be explained at least in part as the result of a common inclination to take “thinking in words” as a paradigm of believing.31 What, after all, could be a clearer instance of believing that Superman can fly than when someone silently says to himself the sentence ‘Superman can fly’? Silent uttering seems to be a way of thinking something in the same way that vocal uttering is a way of asserting something.32 For instance, just as someone can assert that Belgian waffles are delicious by uttering the sentence ‘Belgian waffles are delicious’, so too, it would seem, one can think the thought that Belgian waffles are delicious by silently uttering the same sentence. But this analogy—which I take to be the heart of the matter—turns out to be deeply misguided. The trouble with the analogy in question arises in light of the distinction between, on the one hand, merely uttering a sentence, and on the other 30. The material in this section derives from my “On the Very Idea of Thinking in Words” (1993). 31. David Braddon-Mitchell and Frank Jackson similarly suggest that the language of thought hypothesis draws its inspiration from the fact that “when we contemplate what we believe, we at least sometimes seem to be involved in some kind of ‘saying to ourselves’; so we might think of belief as a kind of inner assertion that may or may not issue in public assertion” (1996: 179). 32. Of course, this so-called ‘silent uttering’, being silent, is not really uttering— but it is very much like uttering, in that they both involve some process of going through the sentence word by word. (And the notion can clearly be extended to include cases where only some of the words in a sentence are reviewed, or where images are used in place of words, as when we review an inner picture of a person, instead of the person’s name.) Presumably the only difference between silent uttering and actual uttering is that the latter includes the actual phonetic production of the sentence, of which the former stops just short; but apart from this, silent uttering is just like its audible counterpart (and may even include some degree of the neuro-muscular activity characteristic of speech production). In any case, I shall continue to use expressions such as ‘silent uttering’ with the understanding that I am using them metaphorically.

Thinking in words 131

hand, actually asserting something by uttering a sentence; i.e., the distinction between the relatively modest phonetic act of mere utterance or pronunciation and the full-blown speech act of assertion. For instance, one might utter the sentence ‘Belgian waffles are delicious’ in order to practice one’s enunciation, or to see what the sentence sounds like, or to give an example of iambic meter; but in none of these cases would one thereby have asserted (in the relevant sense) that Belgian waffles are delicious. Now the question is, how does this distinction with regard to speech carry over with regard to thought? And the simple answer is that just as mere uttering is distinguished from assertion, so, too, mere silent uttering is to be distinguished from thinking. For instance, one might silently utter the sentence ‘Belgian waffles are delicious’ in order to determine how many words it has, in order to make a point of remembering what someone else said, or in order to mull it over; but in none of these cases would one thereby have thought (in the relevant sense) that Belgian waffles are delicious. But is it really so simple? How exactly are these distinctions to be made? Consider first the case of speech. What exactly is the difference between asserting something by uttering a sentence and merely uttering the sentence? Given that the former includes the latter, we may ask, what does assertion consist in over and above the mere uttering of a sentence? That is, what does it take in addition to uttering a sentence to be asserting what the sentence expresses? According to John Searle the act of assertion has an “essential condition” which “has to do with the fact that the utterance is an attempt to inform the hearer and convince him of its truth” (1965: 53). According to Paul Grice in order to assert something—indeed, in order to mean anything at all by one’s words—one must intend, among other things, to produce a certain effect in some audience. Whatever reservations one may have about the details of Searle’s theory of speech acts or of Grice’s theory of meaning, it seems safe to assume that the act of assertion is at least in part characterized by certain ‘communicative’ beliefs and intentions—beliefs and intentions concerning one’s act of assertion as an act of communication, aimed at affecting some audience. The problem, then, for the analogy in question is that there is no way of carrying over such a distinction to the realm of thought. That is, although we may distinguish assertion from mere uttering on the basis of the speaker’s beliefs or intentions with regard to an intended audience, we can not similarly distinguish thinking from mere silent uttering—for the simple

132

Conceptions, belief, and “inner speech”

reason that in thought there is no intended audience with respect to which one might have such beliefs or intentions. The point here, in other words, is that thinking is essentially disanalogous to asserting, in that asserting is essentially a matter of communication, which essentially involves an audience—which in thought is essentially absent. It is not just that asserting and thinking happen to differ with respect to a particular feature, viz., communicative role—after all, even in the best of analogies there will always be some respects in which the analogues differ—but that the basis for the distinction in speech between mere uttering and asserting can have no analogue in thought for the allegedly analogous distinction, between mere silent uttering and thinking. One might try to get around this objection by arguing (as Asa Kasher has suggested) that speech acts such as assertion are not essentially communicative, and in particular, that it is possible to construe assertion in such a way that acts of assertion do not necessarily involve an audience. Note that this is not an easy move, since speech acts so strongly appear to be acts of communication. In the light of the overwhelming prima facie evidence, the burden of proof— and a heavy one, at that—would fall squarely on those who would imagine that speech acts and illocutionary force might be explained without reference to communication and to intentions with regard to an audience. And how might such an argument go? One nonstarter would be the suggestion that the difference between assertion and mere uttering is just that in the case of assertion one actually means what one utters. This just begs the question, reformulating it in the way it is addressed by Grice, viz., what is it for someone to mean, by making a certain utterance, that p? Another nonstarter would be to point out what might look like cases of assertion without any audience—such as writing in a secret diary or putting up a sign that might never be read. But it does not take much imagination to come up with audiences for such cases. Sometimes the audience is merely nonspecific or unknown to the speaker; sometimes it is only hypothesized by the speaker. Either way, even if the audience does not actually exist, there is at least a possible (or unknown) audience serving as the object of one’s communicative intentions—which, of course, there cannot be for silent uttering.33 And in the case of notes to oneself the audience is simply

33. See Grice (1989: 113). Though he speaks of “internal” utterances, he uses scare quotes, and he is probably best understood as treating internal speech as imagined speech.

Thinking in words 133

oneself-at-a-later-time. Though oneself-at-a-later-time does not seem to be an audience distinct from oneself, it actually is—at least in the sense that since we are in different cognitive states at different times (and in particular, since we are capable of forgetting), it makes sense for the purposes of communication to view oneself-at-a-later-time as a distinct individual, capable of serving as well as anybody else as the object of communicative intentions, such as the intention to inform. But clearly one could not similarly intend to inform oneself—that is, one’s current self—of anything by silent uttering. Alternatively, one might concede that assertion is a matter of communication and is distinguished from mere utterance accordingly, while maintaining that the analogous distinction between thinking and mere silent uttering is to be made on the basis of some kind of private cognitive act that is not specific to communication. But what cognitive act might this be that distinguishes, say, actually thinking that Belgian waffles are delicious from merely silently uttering the sentence? As noted with regard to utterance and assertion, we do not get anywhere by saying that when we think by silently uttering a sentence, it is because we mean what we silently utter. For then the question just becomes, what is it to mean what we silently utter? I would answer that to mean what we silently utter is just to think it. The problem is that any other likely candidate for the cognitive difference between thinking and mere silent uttering seems bound to be just as circular. Is it that thinking by silently uttering a sentence involves judging true what we silently utter? Or accepting it, or believing it? Or having certain other beliefs about what we silently utter? Clearly such answers do not get us very far if thinking (believing) is what we are trying to explain. Of course, I do not deny that there is a cognitive difference between thinking and silent uttering. The question is whether this difference mirrors the one between asserting and actual uttering. And we have seen, first, that the distinction between asserting and mere uttering, in terms of communicative intentions and beliefs, can have no place in the inherently private realm of thought; and secondly, that attempts to locate the difference between thinking and mere silent uttering in some noncommunicative cognitive act tend to be circular, amounting to no more than the trivial claim that what distinguishes thinking from mere silent uttering is that in the case of the former we actually think what we silently utter—in which case it is hard to see silent uttering as a way of thinking something, altogether, in the way that vocal uttering is a way of asserting something (a point on which I shall elaborate later).

134

Conceptions, belief, and “inner speech”

Another problem with the analogy between asserting something by uttering a sentence and thinking something by silently uttering a sentence is the explication of the notion of silent uttering. Although we know that silent uttering is supposed to be the analogue in thought to vocal uttering, this does not explain very much when it is the whole analogy between asserting and thinking in words that is in question. What exactly is it that we actually do to a sentence when we utter it silently? One thing for certain is that we do not literally utter it. Hence, any explanation of silent uttering as a kind of saying, talking, pronouncing, and so on, can at best be metaphorical. But how is this metaphor to be explained? When we actually utter a sentence, we go through the sentence word by word, pronouncing each of the words; and it is clear enough what this pronouncing consists in. Now also when we silently utter a sentence, we go through the sentence word by word; but it is not at all clear what we do to the words we silently utter— except, of course, that we do not pronounce them. (Indeed, there cannot be any pronouncing in thought.34) On the one hand, silent uttering is characterized primarily by its resemblance to uttering, while on the other hand, it differs from uttering precisely in that what is uttered only silently is not uttered (not pronounced). Silent uttering is thus like a ham and cheese sandwich without the ham.

5.2. Imagining Perhaps what makes silent uttering like uttering is that both consist in producing tokens of expressions. When we utter a sentence we produce a phonetic token of it, just as when we write a sentence we produce a written token—an inscription—of it. And so, perhaps, in silently uttering a sentence we produce some kind of mental token of it. What, then, could the mental token of an expression be, and what would it be to produce one? Though we cannot literally speak in thought, any more than we can literally write in thought, and hence cannot literally produce spoken or written tokens of expressions in thought, we can—literally—imagine spoken or written tokens of expressions. That is, just as we can imagine ourselves eating a waffle, or imagine a waffle topped with whipped cream, so, too, we can 34. Recognizing that we obviously cannot pronounce words in thought, one might suppose at first that what we do to words when we silently utter them is simply think them. But a little further reflection shows that we do not actually think words, except in the sense of thinking of them.

Thinking in words 135

imagine ourselves uttering a sentence, or imagine an utterance or an inscription of a sentence produced by someone else.35 What I am suggesting, in short, is that what we have been calling “silent” utterance is nothing other than imagined utterance.36 The recognition of silent utterance as imagined utterance can clear up a lot of the confusion resulting from talk of “thinking in words.” For one thing, however much we have yet to learn about imagination, it is something relatively clear in comparison to the mysterious, oxymoronic notion of silent utterance. To understand talk of imagined utterances we do not need to interpret any metaphors, or to rely on any special new concepts—as in the case of talk of silent utterance. Secondly, the recognition of silent utterance as imagined utterance prevents us from mistaking silent utterance, or thinking in words, for a kind of speech. For just as an imagined waffle is not a waffle, and to imagine oneself eating a waffle is not to eat a waffle, so, too, an imagined utterance is not an utterance. Of course, to imagine oneself saying something and actually to say it will have some important things in common—most notably the sentence that is in one case said and in the other case imagined to be said. But it should be clear that the act of speaking and the act of imagining (in particular, imagining oneself speak) are two entirely different kinds of acts, neither of which can be explained as an instance of the other. What about the distinction between actually thinking what one silently utters, and merely silently uttering it (without thinking it)—on analogy with actually asserting what one utters, and merely uttering it without asserting it? Clearly we can merely imagine an utterance—even one of our own— without at the same time thinking what we imagine being uttered. (Thinking a thought in the relevant sense is obviously to be distinguished from entertaining a thought.) For instance, I can easily imagine myself uttering the sentence ‘Belgian waffles are awful’ without, at the same time that I am imagining this, also thinking that Belgian waffles are awful. Indeed, we can imagine ourselves uttering something we believe to be false (as we might do when contemplating lying) or uttering something about which we are 35. I believe this is what Benny Shanon calls “enactment” (1990: 11). 36. To see that this is so, one need only imagine oneself uttering something and then compare this with silent uttering. As John Heil puts it: “When you rehearse an argument in your head, you are engaged in an important form of mental imagery: verbal imagery. You hear words in your head, or, more likely, you both feel and hear yourself uttering words. This is, if anything is, a robust species of imagery” (1998: 213).

136

Conceptions, belief, and “inner speech”

undecided (as we might do when mulling something over). What does it take over and above imagined utterance to be “thinking in words?” What does it take over and above imagining myself uttering the sentence ‘Belgian waffles are delicious’ to be “thinking in words” that Belgian waffles are delicious? I think all it takes—certainly all it takes formally—is that I have the requisite belief, e.g., that I indeed think that Belgian waffles are delicious. That is, “thinking in words” that Belgian waffles are delicious “by silently uttering” the sentence ‘Belgian waffles are delicious’ consists in (i) imagining an utterance of the sentence ‘Belgian waffles are delicious’ while at the same time (ii) simply thinking (i.e., believing) that Belgian waffles are delicious.37 One may object here that by explaining thinking in words in terms of imagined utterance and thinking, I am just begging the question—that the analysis I propose is vulnerable to the very same objection I raised earlier, against explaining thinking in words as silent utterance accompanied by some belief-like propositional attitude with regard to what is silently uttered. However, whether or not an explanation is viciously circular depends (in part) on what it is that is supposed to be explained. In particular, an explanation of thinking in words in terms of thinking simpliciter is thereby circular only to the extent that the notion of thinking in words is to be used in the account of thinking simpliciter. This is what I take to be the intention not only of Plato, in the passage cited, but more generally of anyone who would make use of the alleged analogy, between thinking in words and assertion, in support of any larger claims about thought and its relation to language. But I have no such intention. Even if one disavows all intentions to theorize about thought in general, there is what might be viewed as another kind of circularity, which my analysis helps to bring out, in the very idea of thinking in words. For although I have been using the expression ‘thinking in words’ merely to refer to that familiar experience that I described above and have since been discussing, the common idea of thinking in words is one according to which thinking something in words is a way of thinking it simpliciter. But if thinking a certain thought in words is, as I maintain, nothing but thinking that thought while imagining an utterance of a sentence expressing it, then it seems preposterous to say that this is a way, or a mode, of thinking the

37. This can be compared with thinking out loud, in which, though we actually utter something, we do not actually assert anything, but only pretend to, or make as if. (“Did you say something?” “No, just thinking out loud.”)

Two paradigms of belief 137

thought. For by the same reasoning, thinking a thought while eating a waffle and thinking a thought while riding in a train would also be distinct ways or modes of thinking a thought. The problem is even more glaring in terms of silent utterance. According to the common idea of thinking in words, silent utterance is a way of thinking something in just the same way that vocal utterance is a way of asserting something. Granted, it is possible to utter something silently without actually thinking it, just as it is possible to utter something aloud without actually asserting it. But if the way to think something by silently uttering it is just, as I maintain, to silently utter it while thinking it, then it is absurd to consider silent uttering a way of thinking. For by the same token, eating a waffle and riding a train would be ways of thinking something— to think a thought by eating a waffle, just eat a waffle while thinking the thought! Once we recognize that “silently uttering” a sentence is not really uttering it at all, but just imagining an utterance of it, and that the familiar experience we call “thinking in words” is just thinking a thought while imagining an utterance of a sentence expressing the thought, it becomes apparent that the part of “thinking in words” where the thinking lies is just the thinking; the part involving words—the imagined utterance of a sentence—is no more than what Wittgenstein calls a “characteristic accompaniment” (1958: §152), essentially independent of the thinking of the thought. And that is what is wrong with the very idea of “thinking in words,” which would have us think just the opposite.38

6. Two paradigms of belief The clash between the Inner Speech Picture of Thought and the Theory of Direct Belief might ultimately boil down to a clash between different pre38. Of course, nothing I have said should be taken as a denial of the important connections between thought and language, of which surely there are many. Indeed, my explanation of thinking in words provides for a straightforward account of its cognitive significance: when we imagine something being said, we are thereby able to consider it and react to it—to “hear” it— as if it were actually being said. From here it is easy to see how our linguistic capabilities may well affect what we think (as described, for instance, by Shanon). But (as Shanon has since suggested in correspondence) this way of making use of words in thought is better thought of as thinking with words—as tools—rather than in them—as a medium.

138

Conceptions, belief, and “inner speech”

theoretical paradigms of belief—that is, between different ways of answering the question, “What would be a clear, sure example of someone’s having a particular belief?" One kind of answer: suppose that Nadav has sincerely asserted the sentence, Adolphe Sax invented the saxophone; that when asked, “Did Adolphe Sax invent the saxophone?” he would sincerely answer “Yes;” that he is a normal, competent English speaker, fully understanding all the words in the sentence mentioned; and so on. This, then, would be a clear case of Nadav’s believing that Adolphe Sax invented the saxophone. Such a case may be classified as one of verbal belief, in that a belief is ascribed to a verbal being on the basis of his verbal behavior. Another kind of answer: suppose that a dog—Fido, say—is facing the door, head up, tail wagging; that his master has just driven up the driveway and is now putting his key in the lock; that as soon as his master gets the door open, Fido will jump up on him; that this scenario repeats itself every day when Fido’s master comes home; and so on. Then this would be a clear case of Fido’s believing that his master is behind the door. Such a case may be classified as one of nonverbal belief, in that a belief is ascribed to a nonverbal being (at least as far as we know) on the basis of his nonverbal behavior.39 Now, which kind of answer is better? Which is more paradigmatic of belief—verbal belief or nonverbal belief?40 One reply—the verbalist reply—is that verbal belief is more paradigmatic of belief than nonverbal belief is. Being the highly verbal creatures that we are, it is easy to see how we might be predisposed toward verbalism; for it is only natural for us to pay greater attention to the verbal than to

39. Brenda Judge calls this kind of belief “non-representational;” and she distinguishes a dog’s non-representationally thinking that the cat is up the tree from his having the thought ‘the cat is up the tree’ (Thinking about Things, p. 177). 40. (a) Tempting as it may be to suppose that verbal and nonverbal belief correspond to two senses of the word ‘belief’, this move runs afoul of the considerations put forth in Section 3. (b) Cf. Robert Stalnaker’s distinction between the “linguistic picture” and the “pragmatic picture” (1984: 4–5).

Two paradigms of belief 139

the nonverbal.41 This may take the form of focusing on verbal dispositions and behavior rather than nonverbal dispositions and behavior; on verbal thinkers rather than nonverbal thinkers; or on verbalized or verbalizable thought rather than nonverbalized or nonverbalizable (or nonrepresentational) thought. And then it is easy to see how verbalism (especially, but not necessarily, reinforced with the thinking-in-words paradigm of belief) might predispose us to the Inner Speech Picture of Thought in general and the Medium View of conceptions and belief in particular. But despite our natural verbalist inclinations, there is no prima facie reason to favor the verbal paradigm of belief over the nonverbal paradigm—no reason to think of verbal belief as any more basic or characteristic of belief than nonverbal belief. Verbal behavior is just one kind of behavior; verbal thinkers are just one kind of thinker; verbalizable thought is just one kind of thought. But our practice of ascribing belief is aimed at all kinds of behavior, among all kinds of thinkers, thinking all kinds of thoughts. Verbalism myopically overlooks nonverbal belief; so if the latter is to be given its full due, verbalism must be rejected.42 And once verbalism is rejected, the Medium View loses much of its initial attractiveness.

41. Cf. John Heil: “Philosophers are perhaps by nature inclined to play down the significance of imagery. This may be due, in part, to philosophers’ fixation on arguments and theses expressed in language. When we turn our thoughts to such things, we typically do so in a linguistic frame of mind. We rehearse arguments, try out theses, and formulate replies, all in language” (1998: 213). 42. Cf. Ruth Barcan Marcus. I share her desire “to escape the strong languagecentricity of prevalent theories about epistemological attitudes and their objects” (1983: 330). I agree with her that linguistic behavior is only a part of the behavior to be explained by belief-talk, and that “lingua-centric” (1983: 332) theories of belief exaggerate the status of certain speech acts as belief indicators (1983: 335). And in particular I agree that such theories are prima facie refuted by the cases she mentions of ascriptions of belief to animals and preverbal children (1983: 331, 332; 1981: 509) and to patients undergoing psychoanalysis (1983: 333). However, I do not share her fundamental intuition that the objects of belief must be possible states of affairs (1983: 324; 1981: 505), nor her related claims that it is impossible to believe that Hesperus is not Phosphorus (1983: 330), that belief does not factor out of conjunction (1983: 337), and that belief ascriptions are relative to sets of conditions (1981: 508).

140

Conceptions, belief, and “inner speech”

7. Summary I started this chapter by reformulating the discussion in terms of a question about belief and conceptions. Given that Lois Lane may be said to have two conceptions of Superman, what is the relation between Lois’s conceptions of Superman and her beliefs about him? I defined the Medium View as holding that conceptions belong to some kind of medium in or by which beliefs are held, in which case having a belief about some individual consists in bearing some relation to a conception of that individual—in direct opposition to the Theory of Direct Belief. After setting aside the nonargument that the Medium View is just obvious, as well as Frege’s argument from substitution failure (refuted in the previous chapter), I considered two arguments in favor of the Medium View, based on the common practice of speaking of how a person thinks what he thinks, on analogy with how a person says what he says. The first argument of this sort was based on the Behavior Problem: if Lois’s various conceptions of Superman do not enter into her beliefs about him, then her beliefs about Superman are none other than her beliefs about Clark Kent, in which case there would be no way to account for her behaving towards Superman in one way when he appears in his “Superman” costume and in another way when he appears in his “Clark Kent” costume. But instead of taking conceptions as constituents of belief (or belief vehicles), I proposed the Higher Order View of conceptions, on which having a conception consists in having certain beliefs. Then I explained how Lois’s distinct patterns of behavior can be accounted for on the basis of other beliefs she has, not about Superman, but about his (Superman) cape, his (Clark Kent) glasses, etc. The other argument I considered in support of the Medium View is the argument from suspended belief, as advanced by Nathan Salmon, who tells the story of someone who in the end seems simultaneously to believe a certain proposition while also suspending belief in the very same proposition. Despite Salmon’s claim that the scenario cannot be explained adequately without the Medium View, I presented three ways of doing so. Having shown the unsoundness of the arguments for the Medium View, I pointed to the Inner Speech Picture of Thought as the likely source of the view’s wide appeal. Against this picture I argued that the metaphor of thought as language has yet to be fleshed out, and that cases of apparently nonrepresentational belief make it questionable whether it ever can be. By way of diagnosis I proposed that the popularity of the Inner Speech Picture

Summary 141

of Thought might be explained at least in part as the result of a common inclination to take “thinking in words” as a paradigm of believing, on the basis of an analogy between asserting something by uttering a sentence and thinking something by “silently uttering” a sentence. Against this analogy I argued that there is no way in the realm of thought to accommodate the distinction between merely uttering a sentence and actually asserting something by uttering a sentence, and I proposed that “silent uttering” is better understood as imagined uttering. The chapter concluded with a brief look at how the clash between the Inner Speech Picture of Thought and the Theory of Direct Belief might ultimately boil down to a clash between different pre-theoretical paradigms of belief. Among paradigmatic cases of belief I distinguished cases of verbal belief, where a belief is ascribed to a verbal being on the basis of his verbal behavior, from cases of nonverbal belief, where a belief is ascribed to a presumably nonverbal being on the basis of his nonverbal behavior. Observing that it is easy to see how we might be predisposed to the view that verbal belief is more paradigmatic of belief than nonverbal belief is, and how such a view might in turn predispose us to the Inner Speech Picture of Thought in general and the Medium View of conceptions and belief in particular, I claimed that there is no prima facie reason to favor the verbal paradigm of belief over the nonverbal paradigm. Having refuted the main arguments in favor of the Medium View, and having identified and shed doubt on the verbalist thinking that seems to underlie it, this is where I might be expected to offer a positive, nonverbalist alternative. I will not do so, however, because the Theory of Direct Belief allows for a wide range of more particular theories of belief, and as my aim has been to defend the theory in all its generality, I shall remain neutral with respect to the theory’s various possible versions.43 I certainly recognize that if Lois’s believing that Superman is a reporter does not consist in her bearing some relation to him that is mediated by some conception, mode of presentation, or other component of a medium in which beliefs are held, then it remains to be said what it does consist in. If the relation between a believer and the individual her belief is about is not a mediated relation, then what is it? That is a question for another book. Though this may leave the reader less than fully satisfied, it should do so 43. Some ideas about how particular versions of the Theory of Direct Belief might go may be found in the work of F. P. Ramsey (1931), D. M. Armstrong (1973), H. H. Price (1969), Lynne Rudder Baker (1995), and Eric Schwitzgebel (2002).

142

Conceptions, belief, and “inner speech”

no more than versions of the Medium View which leave unexplained the relations between believer and mediating element, and between mediating element and object of belief. What I have done is as follows. First, I explained why we should not accept theories of the semantics of belief ascriptions which would account for the shifty intuitions aroused by certain belief ascriptions by appeal to ambiguity, indexicality, or semantic indeterminacy. Secondly, I showed how such intuitions can be accounted for pragmatically. In particular, I showed how by uttering a belief ascription that is, strictly speaking, true, one may, given some innocuous assumptions about normalcy, conversationally implicate something false, which might easily be mistaken for what one actually said—and I defended this account against various objections and showed why it is preferable to other pragmatic accounts of substitution failure. Thirdly, I defended the Theory of Direct Belief against metaphysical objections embodied in the Medium View, showing how the leading arguments for the Medium View fail and how a certain picture which seems to lie behind the Medium View is deeply misguided. Though this all falls far short of a nice knockdown argument for the Theory of Direct Belief, I hope it is enough to make the reader doubt that there is any nice knockdown argument against the Theory of Direct Belief.

References

Alston, William P. 1971 How does one tell whether a word has one, several or many senses? In Semantics, Danny D. Steinberg and Leon A. Jakobovits (eds.), 35–48. Cambridge: Cambridge University Press, 1971. Anderson, C. Anthony, and Joseph Owens (eds.) 1990 Propositional Attitudes: The Role of Content in Language, Logic, and Mind. Stanford: CSLI. Anscombre, J.-C., and O. Ducrot 1978 Echelles argumentatives, echelles implicatives et lois de discours. Semantikos 2: 43–67. Armstrong, David M. 1973 Belief, Truth and Knowledge. Cambridge: Cambridge University Press, Atlas, Jay 1989 Philosophy Without Ambiguity. Oxford: Oxford University Press, Bach, Kent 1987 Thought and Language. Oxford: Oxford University Press. 1994 Conversational impliciture. Mind and Language 9: 124–162. 1997 Do belief reports report beliefs? Pacific Philosophical Quarterly 78: 215–241. 1999 The myth of conventional implicature. Linguistics and Philosophy 22: 327–366. 2001 Seemingly semantic intuitions. In Meaning and Truth, Joseph Campbell, Michael O’Rourke, and David Shier, (eds.), 21–33. New York: Seven Bridges Press. Baker, Lynne R. 1995 Explaining Attitudes. Cambridge: Cambridge University Press. Barwise, Jon, and John Perry 1981 Semantic innocence and uncompromising situations. Midwest Studies in Philosophy 6: 387–403. Berg, Jonathan 1979 Two kinds of substitutivity. Ms. 1988 The pragmatics of substitutivity. Linguistics and Philosophy 11: 355–370. 1993 On the very idea of thinking in words. Belgian J. of Linguistics 8: 9– 19. 1998 In defense of direct belief: Substitutivity, availability, and iterability. Lingua e Stile 33: 461–470.

144 References 1999 2002 2003

Troubles with neo-notionalism. Philosophia 27: 459–482. Is semantics still possible? Journal of Pragmatics 34: 349–359. Must we know what we say? In Saying, Meaning, Implicating, Georg Meggle and Christian Plunze (eds.), 64–71. Leipzig: Leipzig University Press. 2007 What is said to be said. Fifth Barcelona Workshop on Issues in the Theory of Reference, Barcelona. Boer, Steven, and William G. Lycan 1975 Knowing who. Philosophical Studies 28: 299–344. 1986 Knowing Who. Cambridge: MIT Press, Borg, Emma 2004 Minimal Semantics. Oxford: Oxford University Press, 2007 Minimalism versus conextualism in semantics. In Context-Sensitivity and Semantic Minimalism, Gerhard Preyer and Georg Peter (eds.), 339–359. Oxford: Oxford University Press. Braddon-Mitchell, David, and Frank Jackson 1996 The Philosophy of Mind and Cognition. Oxford: Blackwell. Braun, David. 1998 Understanding belief reports. Philosophical Review 107: 555–595. 2001 Russellianism and explanation, Philosophical Perspectives 15: 253– 289. 2002 Cognitive significance, attitude ascriptions, and ways of believing propositions. Philosophical Studies 108: 65–81. Burge, Tyler 1977 Belief de re. Journal of Philosophy 69: 338–362. Cappelen, Hermann, and Ernie Lepore 2005 Insensitive Semantics. Oxford: Blackwell. Carston, Robyn 2002 Thoughts and Utterances. Oxford: Blackwell. Chalmers, David J. 2011 Propositions and attitude ascriptions: A Fregean Account. Noûs 45: 595–639. Chomsky, Noam 1972 Language and Mind. New York: Harcourt Brace Jovanovich. Cohen, L. Jonathan 1971 Some remarks on Grice’s views about the logical particles of natural language. In Pragmatics of Natural Languages, Y. Bar-Hillel (ed.), 50–68. Dordrecht: D. Reidel. Crawford, Sean 2004a Pure Russellianism. Philosophical Papers 33: 171–202. 2004b A solution for Russellians to a puzzle about belief. Analysis 64: 223– 229.

References 145 2008

Quantifiers and propositional attitudes: Quine Revisited. Synthese 160: 75–96. Crimmins, Mark 1992 Talk About Beliefs. Cambridge: MIT. Crimmins, Mark, and John Perry 1989 The prince and the phone booth: Reporting puzzling beliefs. Journal of Philosophy 56: 685–711. Rpt. in The Problem of the Essential Indexical: and Other Essays. Oxford: Oxford University Press, 1993. Cruse, D. A. 1986 Lexical Semantics. Cambridge: Cambridge University Press. Davidson, Donald 1967 Truth and meaning. Synthese 17: 304–323. Rpt. in Inquiries into Truth and Interpretation, Donald Davidson, 17–36. Oxford: Oxford University Press, 1984. 1968 On saying that. Synthese 19: 130–146. Rpt. in Inquiries into Truth and Interpretation, Donald Davidson, 93–108. Oxford: Oxford University Press, 1984. Davidson, Donald, and G. Harman (eds.) 1972 Semantics of Natural Language. Dordrecht: D. Reidel. Davidson, Donald, and Jaakko Hintikka (eds.) 1969 Words and Objections. Dordrecht: D. Reidel. Davis, Wayne 1998 Implicature. Cambridge: Cambridge University Press. Donnellan, Keith 1966 Reference and definite descriptions. Philosophical Review 75: 281– 304. 1970 Proper names and identifying descriptions. Synthese 21: 335–358. Rpt. in Semantics of Natural Language, Davidson, Donald and G. Harman (eds.), 356–379. Dordrecht: D. Reidel, 1972. Dowty, David R., Robert E. Wall, and Stanley Peters 1981 Introduction to Montague Semantics. Dordrecht: Kluwer. Fodor, Jerry 1975 The Language of Thought. New York: Crowell. 1985 Fodor’s guide to mental representation. Mind (New Series) 94: 55– 97. Rpt. in A Theory of Content and Other Essays, Jerry Fodor. Cambridge: MIT Press, 1990. 1987 Psychosemantics. Cambridge: MIT Press,. 1990 A Theory of Content and Other Essays. Cambridge: MIT Press. 2008 LOT 2: The Language of Thought Revisited. Oxford: Oxford University Press. Fodor, Jerry, and Z. Pylyshyn 1988 Connectionism and cognitive architecture: A critical analysis. Cognition 28: 3–71.

146 References Forbes, Graeme 1996 Substitutivity and the coherence of quantifying in. Philosophical Review 105: 337–372. Frege, Gottlob 1893 Uber sinn und bedeutung. Zeitschrift fur Philosophie und Philosophische Kritik, 100: 25–50. Translated as On sense and reference, in Translations from the Philosophical Writings of Gottlob Frege, Peter T. Geach and Max Black (eds.), 56–78. Oxford: Blackwell, 1952. Rpt. in Basic Topics in the Philosophy of Language, Robert M. Harnish (ed.), 142–160. Englewood Cliffs, NJ: Prentice Hall. 1918 Der gedanke: Eine logische untersuchung in Beiträge zur Philosophie des Deutschen Idealismus I: 58–77. Rpt. as The thought: A logical inquiry, A. M. and Marcelle Quinton (trans.), Mind 65 (1956): 289–311. Rpt. in Philosophical Logic, P. F. Strawson (ed.), 17–38. Oxford: Oxford University Press, 1967. Also rpt. in Basic Topics in the Philosophy of Language, Robert M. Harnish (ed.). Englewood Cliffs, NJ: Prentice Hall, 1994. French, Peter A., Theodore E. Uehling Jr., and Howard K. Wettstein, eds. 1979 Contemporary Perspectives in the Philosophy of Language. Minneapolis: University of Minnesota Press. Friedman, Jane 2011 Suspended judgment. Philosophical Studies, DOI: 10.1007/s11098011-9753-y (2011). Green, Mitchell 1998 Direct reference and implicature. Philosophical Studies 91: 61–90. Grice, Paul 1961 The causal theory of perception. Proceedings of the Aristotelian Society, Supplement 35: 121–152. Rpt. in The Philosophy of Perception. G. J. Warnock, (ed.), 85–112. Oxford: Oxford University Press. 1989 Studies in the Way of Words. Cambridge: Harvard University Press. 1969 Vacuous names. In Words and Objections, Donald Davidson and Jaakko Hintikka (eds.), 118–145. Dordrecht: D. Reidel. Harnish, Robert M. (ed.) 1994 Basic Topics in the Philosophy of Language. Englewood Cliffs, NJ: Prentice Hall. Haugeland, John 1981 Mind Design. Cambridge, Mass.: MIT Press. Heil, John 1998 Philosophy of Mind. London: Routledge. Higginbotham, James

References 147 1995

The place of natural language. In On Quine, Paolo Leonardi and Marco Santambrogio (eds.), 113–139. Cambridge: Cambridge University Press. 1998 On knowing one’s own language. In Knowing Our Own Minds, Crispin Wright, Barry C. Smith, and Cynthia McDonald (eds.), 429– 442. Oxford: Oxford University Press. Hintikka, Jaakko 1962 Knowledge and Belief. Ithaca: Cornell University Press. 1969 Semantics for propositional attitudes. In Models for Modalities, Jaakko Hintikka, 87–111. Dordrecht: D.Reidel. Horst, Steven W. 1996 Symbols, Computation, and Intentionality. Berkeley: University of California Press. Judge, Brenda 1985 Thinking about Things. Edinburgh: Scottish Academic Press. Kaplan, David 1969 Quantifying in. In Words and Objections, Donald Davidson and Jaakko Hintikka (eds.), 206–242. Dordrecht: D. Reidel. 1978 Dthat. In Pragmatics, P. Cole (ed.). New York: Academic Press. Rpt. in Contemporary Perspectives in the Philosophy of Language, Peter A. French, Theodore E. Uehling Jr., and Howard K. Wettstein, (eds.), 383–400. Minneapolis: University of Minnesota Press, 1979. 1979 On the logic of demonstratives. In Contemporary Perspectives in the Philosophy of Language, Peter A. French, Theodore E. Uehling Jr., and Howard K. Wettstein, (eds.), 401–414. Minneapolis: University of Minnesota Press. 1986 Opacity. In The Philosophy of W. V. Quine, Lewis Edwin Hahn and Paul Arthur Schilpp (eds.), 229–289. LaSalle: Open Court. 1989 Demonstratives. In Themes from Kaplan, Joseph Almog, John Perry and Howard Wettstein, (eds.), 481–563. New York: Oxford University Press. Kripke, Saul 1972 Naming and necessity. In Semantics of Natural Language, Donald Davidson and G. Harman (eds.), 253–355, 763–769. Dordrecht: D. Reidel. Rpt., revised, as Naming and Necessity. Cambridge: Harvard University Press, 1980. 1977 Speaker’s reference and semantic reference. Midwest Studies in Philosophy, 2: 255–276. Rpt. in Contemporary Perspectives in the Philosophy of Language, Peter A. French, Theodore E. Uehling Jr., and Howard K. Wettstein, (eds.), 6–27. Minneapolis: University of Minnesota Press. 1979 A puzzle about belief. In Meaning and Use, Avishai Margalit (ed.), 239–283. Dordrecht: Reidel. Rpt. in Basic Topics in the Philosophy

148 References of Language, Robert M. Harnish (ed.), 352–392. Englewood Cliffs, NJ: Prentice Hall. Leonardi, Paolo, and Marco Santambrogio (eds.) 1995 On Quine. Cambridge: Cambridge University Press. Lepore, Ernest, and Barry Loewer 1989 You can say that again. Midwest Studies in Philosophy 14: 338–356. 1995 Quine and the attitudes. In On Quine, Paolo Leonardi and Marco Santambrogio (eds.), 186–205. Cambridge: Cambridge University Press. Ludlow, Peter 1995 Logical form and the hidden-indexical theory: A reply to Schiffer. Journal of Philosophy 92: 102–107. 1996 The adicity of ‘believes’ and the hidden indexical theory. Analysis 56: 97–101. Machery, Edouard 2005 You don’t know how you think: Introspection and language of thought. British Journal for the Philosphy of Science 56: 469–485. McKay, Tom 1981 On proper names in belief ascriptions. Philosophical Studies 39: 287–303. Marcus, Ruth Barcan. 1981 A proposed solution to a puzzle about belief. Midwest Studies in Philosophy 6: 501–510. 1983 Rationality and believing the impossible. Journal of Philosophy 80: 321–338. Matushansky, Ora 2008 On the linguistic complexity of proper names. Linguistics and Philosophy 31: 573–627. Montague, Richard 1974 Formal Philosophy. Richmond Thomason (ed.). New Haven: Yale University Press. Pagin, Peter, and Francis Jeffry Pelletier 2007 Content, context, and composition. In Context-Sensitivity and Semantic Minimalism, Gerhard Preyer and Georg Peter (eds.), 25–62. Oxford: Oxford University Press. Perry, John 1980 A problem about continued belief. Pacific Philosophical Quarterly 61: 317–332. Rpt. in The Problem of the Essential Indexical and Other Essays, John Perry, 69–90. Oxford: Oxford University Press, 1993. 1986 Thought without representation. Proceedings of the Aristotelian Society, Supplement 60: 137–152. Rpt. in The Problem of the Essen-

References 149

1993 Plato 1973 1990

tial Indexical and Other Essays, John Perry, 205–225. Oxford: Oxford University Press, 1993. The Problem of the Essential Indexical and Other Essays. Oxford: Oxford University Press. Theaetetus. Trans. John McDowell. London: Oxford University Press. Theaetetus. Trans. M.J. Levett. Revised, Miles Burnyeat. In Theaetetus of Plato, Miles Burnyeat. Indianapolis and Cambridge: Hackett.

Price, H. H. 1969 Belief. London: Allen & Unwin. Putnam, Hilary 1975 The meaning of ‘meaning’. In Philosophical Papers II: Mind, Language, and Reality, Hilary Putnam, 215–271. Cambridge: Cambridge University Press. Quine, W. V. O. 1956 Quantifiers and propositional attitudes. Journal of Philosophy 53: 177–187. Rpt. and revised in Ways of Paradox and Other Essays, W. V. O. Quine, 185–196. Cambridge: Harvard University Press, 1976. 1977 Intensions Revisited. Midwest Studies in Philosophy 2: 5–11. Rpt. in Theories and Things, W. V. O. Quine, 113–123. Cambridge: Harvard University Press, 1981. 1995 Reactions. In On Quine, Paolo Leonardi and Marco Santambrogio (eds.), 347–361. Cambridge: Cambridge University Press. Ramsey, Frank P. 1931 ‘General propositions and causality.’ In his Foundations of Mathematics and other Logical Essays, Frank P. Ramsey, 237–255. London: Kegan Paul, Trench & Trubner. Recanati, François 1993 Direct Reference. Oxford: Blackwell. 2001 What is said. Synthese 128: 75–91. 2004 Literal Meaning. Cambridge: Cambridge University Press. 2007 It is raining (somewhere). Linguistics and Philosophy 30: 123–146. Richard, Mark. 1990 Propositional Attitudes: An Essay on Thoughts and How We Ascribe Them. Cambridge: Cambridge University Press. 1997 Propositional attitudes. In A Companion to the Philosophy of Language, Bob Hale and Crispin Wright (eds.), 197–226. Oxford: Blackwell. Russell, Bertrand 1905 On denoting. Mind 14: 479–493.

150 References 1919 Introduction to Mathematical Philosophy. London: Allen & Unwin. Ryle, Gilbert 1949 The Concept of Mind. London: Hutchinson. Salmon, Nathan 1986 Frege’s Puzzle. Cambridge: MIT Press. 1989 Illogical belief. Philosophical Perspectives 3: 243–285. Saul, Jennifer M. 1998 The pragmatics of attitude ascription. Philosophical Studies 92: 363– 389. 2002 Speaker meaning, what is said, and what is implicated. Noûs 36: 228–248. 2007 Simple Sentences, Substitution, and Intuitions. Oxford: Oxford University Press. Schiffer, Stephen 1977 Naming and knowing. Midwest Studies in Philosophy 2: 28–41. 1987a The ‘Fido’-Fido theory of belief. Philosophical Perspectives 1: 455– 480. 1987b Remnants of Meaning. Cambridge: MIT Press. 1990 The mode-of-presentation problem. In Propositional Attitudes: The Role of Content in Language, Logic, and Mind. Anderson, C. Anthony, and Joseph Owens (eds.), 249–268. 1992 Belief ascription. Journal of Philosophy 89: 499–521. 1995 Review of Direct Reference, François Recanati. Linguistics and Philosophy 19: 91–102. 1996 The hidden-indexical theory’s logical-form problem: A rejoinder. Analysis 56: 92–97. Schwitzgebel, Eric 2002 A phenomenal, dispositional account of belief. Nous 36: 249–275. 2006 Belief. In The Stanford Encyclopedia of Philosophy, Edward N. Zalta (ed.). http://plato.stanford.edu/entries/belief/ Searle, John 1965 What is a speech act? In Philosophy in America, Max Black (ed.), 221–239. London: Allen & Unwin. Rpt. in Philosophy of Language. John Searle (ed.), 39–53. London: Oxford University Press. 1978 Literal meaning. Erkenntnis 13: 207–224. Sellars, Wilfrid 1956 Empiricism and the philosophy of mind. In Minnesota Studies in the Philosophy of Science 1, Herbert Feigl and Michael Scriven (eds.), 253–328. Minneapolis: University of Minnesota Press. Shanon, Benny 1987 On the place of representations in cognition. In Thinking: The Second International Conference, David N. Perkins, Jr., Jack Lochhead,

References 151 and John Bishop (eds.), 33–49. Hillsdale, NJ: L. Erlbaum Associates. 1990 Why do we (sometimes) think in words. In Lines of Thinking: Reflections on the Psychology of Thought, Vol.1, K. J. Gilhooly, M. T. G. Keane, R. H. Logie, and G. Erdos, (eds.), 5–14. New York: Wiley. Sperber, Dan, and Deirdre Wilson 1986 Relevance. Cambridge: Harvard University Press. Soames, Scott 2002 Beyond Rigidity. Oxford: Oxford University Press. 2005 Naming and asserting. In Semantics versus Pragmatics, Zoltan Gendler Szabo (ed.), 356–382. Oxford: Oxford University Press. 2006 The gap between meaning and assertion. Ms. http://www.rcf.usc.edu/~soames/forthcoming_papers/Meaning_and_ Assertion.pdf Forthcoming, in Asserting, Meaning, and Implying, Martin Hackl and Robert Thornton (eds.). Rpt. in Philosophical Papers, Vol. 1, Scott Soames. Princeton: Princeton University Press, 2008. 2008 Drawing the line between meaning and implicature–and relating both to assertion. Nous, 42: 440–465. Stainton, Robert 2006 Words and Thoughts. Oxford: Oxford University Press. Stalnaker, Robert 1984 Inquiry. Cambridge, Mass.: MIT Press. Stanley, Jason 2007 Language in Context. Oxford: Oxford University Press. Stanley, Jason, and Jeffrey King 2005 Semantics, pragmatics, and the role of semantic content. In Semantics versus Pragmatics, Zoltan Gendler Szabo (ed.). Oxford: Oxford University Press. Rpt. in Language in Context, Jason Stanley, 133– 181. Oxford: Oxford University Press, 2007. Stanley, Jason, and Zoltan Szabo 2000 On quantifier domain restriction. Mind and Language 15: 219–261. Rpt. in Language in Context, Jason Stanley, 69–110. Oxford: Oxford University Press, 2007. Steinberg, Danny D., and Leon A. Jakobovits (eds.) 1971 Semantics. Cambridge: Cambridge University Press. Sterelny, Kim 1990 The Representational Theory of Mind. Oxford: Blackwell. Stich, Stephen P. 1983 From Folk Psychology to Cognitive Science. Cambridge: MIT Press. 1986 Are belief predicates systematically ambiguous? In Belief, R. Bogdan (ed.), 119–148. Oxford: Oxford University Press.

152 References Strawson, P. F. 1950 On referring. Mind 59: 320–344. Szabo, Zoltan Gendler (ed.) 2005 Semantics versus Pragmatics. Oxford: Oxford University Press. Taylor, Kenneth 2001 Sex, breakfast, and descriptus interruptus. Synthese, 128: 45–61. Thau, Michael 2002 Consciousness and Cognition. Oxford: Oxford University Press. Urmson, J. O. 1968 Criteria of intensionality. Proceedings of the Aristotelian Society, Supplementary Volume 42: 107–122. Walker, Ralph C. S. 1975 Conversational implicatures. In Meaning, Reference and Necessity, Simon Blackburn (ed.), 133–181. Cambridge: Cambridge University Press. Wiggins, David 1971a On sentence-sense, word-sense and difference of word-sense: Towards a philosophical theory of dictionaries. In Semantics, Danny D. Steinberg and Leon A. Jakobovits (eds.), 14–34. Cambridge: Cambridge University Press. 1971b A reply to Mr Alston. In Semantics. Danny D. Steinberg and Leon A. Jakobovits (eds.), 48–52. Cambridge: Cambridge University Press. Wittgenstein, Ludwig 1958 Philosophical Investigations. 3rd ed. Trans. G.E.M. Anscombe. New York: Macmillan. Woodfield, Andrew 1991 Conceptions. Mind, 100: 547–571. Zalta, Edward N. 1988 Intensional Logic and the Metaphysics of Intentionality. Cambridge: MIT.

Index

Adams, Fred, vii Almog, Joseph, 147 Alston, William P., 17 n. 25, 143, 152 ambiguity, 10, 12–24, 25 n. 38, 35, 61 n. 10, 80 n. 31 crossed interpretations, 14–16 disambiguation, 13–14 lexical, 12–17 multigrade predicates, 21–23 syntactic, 17–21 Anderson, C. Anthony, 143, 150 Anscombe, G.E.M., 152 Anscombre, J.-C., 143 Antony, Michael, vii appropriateness, v, 9, 54–56 Armstrong, David M., 141 n. 43, 143 articulability, 28–35 Atlas, Jay, 11 n. 12, 14 n. 20, 15 n. 21, 143 availability, 75–81

Theory of Direct Belief, 2, 5–6, 48–49, 54, 92 n. 49, 108–115, 137–139, 140 verbal/nonverbal, 138 Berg, Jonathan, 143–144 Bezuidenhut, Anne, vii binding, argument from, 32 Biro, John, vii Bishop, John, 151 Black, Max, 146, 150 Blackburn, Simon, 152 Boer, Steven, 82 n. 37, 144 Bogdan, R., 151 Borg, Emma, 45 n. 66, 144 Braddon-Mitchell, David, 130 n. 31, 144 Braun, David, 9 n. 9, 27 n. 41, 85 n. 39, 111 n. 8, 144 Brown, Jessica, vii Burge, Tyler, v, vii, 7 n. 2, 19 n. 31, 144 Burnyeat, Miles, 149

Bach, Kent, vii, 18 n. 29, 20 n. 32, 41, 42 n. 63, 90 n. 46, 91 n. 48, 96, 99 n. 59, 143 Baker, Lynne R., 141 n. 43, 143 Bar-Hillel, Y., 144 Barwise, Jon, 4 n. 4, 143 behavior problem, 111–115 belief identity beliefs, 70–74 relational/notional (de re/de dicto), 2–3, 10, 13–14, 19–24, 40, 46 n. 69, 68–69, 92 n. 49 suspended belief, 115–122 robust construal, 120

Campbell, Joseph, 143 Cappelen, Hermann, 29 n. 45, 32 n. 50, 39 n. 57, 39 n. 58, 41 n. 61, 144 Carston, Robyn, 29 n. 45, 72, 90 n. 46, 144 Chomsky, Noam, 14 n. 20, 144 Cohen, L. Jonathan, 72 n. 21, 73, 105, 144 compositionality, 34 Coincidence View of Meaning and Saying, 89 conceptions, 5, 26, 108–115 Cooperative Principle, 56–61, 67, 76, 99 Crawford, Sean, 121–122, 144

154 Index Crimmins, Mark, 24 n. 36, 25, 28, 30, 34, 35, 41, 111 n. 8, 145 crossed interpretations, 14–16, 51 Cruse, D. A., 11 n. 12, 13 n. 17, 14, 15 n. 21, 145

Grice, Paul, v, 4, 16, 29 n. 45, 50, 55, 56–59, 63 n. 13, 67, 74, 75–81, 86 n. 41, 96, 97 n. 56, 98–101, 103, 106, 113 n. 11, 131, 132, 144, 146 Guise Millianism, 5 n. 5, 109 n. 3

Davidson, Donald, 3, 43 n. 64, 145, 146, 147 Davis, Steven, vii Davis, Wayne, 55 n. 4, 145 disambiguation, 13–14 disquotational principle, 94, 95 n. 53, 106 Divergence View of Meaning and Saying, 89–91, 105 Donnellan, Keith, v, vii, 2 n. 1, 17 n. 26, 18 n. 29, 48, 145 dossiers, 113 n. 11 Dowty, David R., 20 n. 32, 145 Ducrot , O., 73, 143

Hackl, Martin, 151 Hahn, Lewis Edwin, 147 Hale, Bob, 149 Hanson, Philip, vii Harman, Gilbert, 145, 147 Harnish, Robert M., 146, 148 Having-in-Mind Test of Hidden Indexicality, 33 Haugeland, John, 127 n. 27, 146 Heil, John, 135 n. 36, 139 n. 41, 146 Higher Order View, 113 Higginbotham, James, 83 n. 38, 93 n. 50, 146–147 Hintikka, Jaakko, 20 n. 32, 145, 146, 147 Hofweber, Thomas, vii Horst, Steven W., 124 n. 20, 125 n. 24, 147

Erdos, G., 151 Feigl, Herbert, 150 files, 113 n. 11 Fodor, Jerry, vii, 110 n. 5, 111 n. 8, 123, 124 n. 20, 145 Forbes, Graeme, 25 n. 38, 146 Frege, Gottlob, 1–5, 7–10, 16 n. 23, 24 n. 36, 26, 48–50, 54, 62 n. 12, 69, 70, 101–104, 108–111, 140, 144, 146, 150 Fregeanism, 5 n. 5, 109 n. 3 French, Peter A., 146, 147 Friedman, Jane, 120 n. 15, 121 n. 17, 146 Geach, Peter T., 146 Gilhooly, K. J., 151 Grayling, Anthony, vii Green, Mitchell, vii, 72–73, 74 n. 24, 105, 146

implicature conventional, 99–105 conversational, 9, 56–63 indexicality, 24–40 hidden, 27–36; see also articulability; binding, argument from; compositionality; Having-inMind Test of Hidden Indexicality; Relativity Test of Hidden Indexicality incompleteness, 40–42 indeterminacy, 40–48 inner speech, 123–137; see also silent utterance Inner Speech Picture of Thought, 123–129 intuitions, 2, 7–9, 82–91 Semantic Intuitionism, 82

Index 155 Israel, David, vii Jackson, Frank, 130 n. 31, 144 Jakobovits, Leon A., 143, 151, 152 Judge, Brenda, 138 n. 39, 147 Kaplan, David, v, vii, 2 n. 1, 6, 13 n. 15, 23 n. 35, 48, 115, 147 Kasher, Asa, vii Keane, M. T. G. , 151 King, Jeffrey, 33, 151 Kripke, Saul, 2 n. 1, 14, 16 n. 24, 17 n. 26, 18 n. 29, 24, 46, 48, 53, 60, 62 n. 11, 94, 97 n. 56, 118 n. 13, 119 n. 14, 147–8 Kvart, Igal, vii Language of Thought Hypothesis, 110 n. 5 Leonardi, Paolo, 147, 148, 149 Lepore, Ernie, vii, 29 n. 45, 32 n. 50, 39 n. 57, 39 n. 58, 41 n. 61, 43 n. 64, 93 n. 50, 144, 148 Levett, M. J., 149 Lochhead, Jack, 150 Loewer, Barry, 43 n. 64, 148 Logie, R. H., 151 Ludlow, Peter, 27 n. 41, 148 Ludwig, Kirk, vii Lycan, William G., 82 n. 37, 144 McDonald, Cynthia, 147 McDowell, John, 149 Machery, Edouard, 123 n. 19, 148 McKay, Tom, 4 n. 4, 148 Marconi, Diego, vii Marcus, Ruth Barcan, 139 n. 42, 148 Margalit, Avishai, 147 Matushansky, Ora, 18 n. 29, 148 Medium View, 108–111 Meggle, Georg, vii, 144 modes of presentation, 2 n. 2, 3, 24, 25 n. 38, 26–27, 28 n. 42, 41 n. 60,

49, 51, 53, 68, 103, 108, 109, 111 n. 7, 124 Modified Occam’s Razor, 16, 21 Montague, Richard, 20 n. 32, 145, 148 Morton, Adam, vii multigrade, 21–23 normalcy, 57–70 O’Rourke, Michael, 143 Owens, Joseph, 143, 150 Pagin, Peter, vii, 71 n. 20, 148 Pelletier, Francis Jeffry, 71 n. 20, 148 Perkins, Jr., David N., 150 Perry, John, vii, 4 n. 4, 25, 28, 29, 30, 35, 41, 113 n. 11, 143, 145, 147, 148–149 Peter, Georg, 144, 148 Peters, Stanley, 145 Plato, 123, 136, 149 Plunze, Christian, vii, 144 pragmatic content, 29, 34, 45 n. 68, 71–74, 86–88, 91, 92, 93, 104–106 Preyer, Gerhard, 144, 148 Price, H. H., 141 n. 43, 149 Principle of Implicated Normalcy, 57–63 psychological sententialism, 108 n. 2 Putnam, Hilary, 2 n. 1, 48, 149 Pylyshyn, Zenon, 124 n. 20, 145 Quine, W. V. O., 3, 8, 9, 10, 12, 13 n. 15, 17 n. 26, 21–23, 25 n. 38, 46, 48, 50, 51, 101, 145, 147, 148, 149 Quinton, A. M., 146 Quinton, Marcelle, 146 Ramberg, Bjorn, vii Ramsey, Frank P., 141 n. 43, 149 rationality, 9 n. 9

156 Index Recanati, François, vii, 4, 29 n. 45, 33 n. 52, 35 n. 55, 41 n. 60, 54 n. 1, 55 n. 3, 63 n. 13, 72, 73, 75–81, 87 n. 43, 105, 111 n. 8, 149, 150 Relativity Test of Hidden Indexicality, 31–32 representational theory of mind, 110 n. 5, 151 representationalism, 109 n. 3 Richard, Mark, 3, 36–40, 52, 54 n. 1, 69, 85 n. 39, 88 n. 45, 108 n. 2, 110 n. 6, 111 n. 8, 149 Russell, Bertrand, 17–21, 51, 54 n. 1, 84, 88 n. 45, 90, 149-50 Russellian annotated matrix, 36–40 Ryle, Gilbert, 14 n. 20, 150 Salmon, Nathan, vii, 2 n. 1, 4 n. 4, 5 n. 5, 6, 55 n. 2, 55 n. 4, 94, 95 n. 53, 115–122, 123, 140, 150 Santambrogio, Marco, 147, 148, 149 Saul, Jennifer M., 59 n. 8, 62 n. 12, 66–68, 104, 150 Schiffer, Stephen, vii, 4, 25, 26, 27 n. 41, 28 n. 42, 41, 46, 48, 80 n. 33, 92–94, 105–106, 124 n. 22, 148, 150 Schilpp, Paul Arthur, 147 Schwitzgebel, Eric, 109 n. 3, 141 n. 43, 150 scope argument, 73–74 Scriven, Michael, 150 Searle, John, 58 n. 7, 68, 131, 150 Sellars, Wilfrid, 123, 150 Shanon, Benny, 135 n. 35, 137 n. 38, 150–151 Shier, David, 143 silent utterance, 130–137 similarity, 42–46 Smith, Barry C., vii, 147 Soames, Scott, ix, 2 n. 1, 4 n. 4, 5, 29 n. 45, 55 n. 3, 87 n. 43, 95–98, 106, 151 speaker meaning, 29–30, 33–34

Sperber, Dan, 72, 90 n. 46, 151 Stainton, Robert, 29 n. 45, 79 n. 30, 91 n. 47, 151 Stalnaker, Robert, 129 n. 29, 138 n. 40, 151 Stanley, Jason, 32–35, 41 n. 62, 57 n. 6, 91 n. 47, 151 Steinberg, Danny D., 143, 151, 152 Steiner, Mark, vii Sterelny, Kim, 110 n. 5, 125 n. 24, 151 Stern, Josef, vii Stich, Stephen P., 3, 10 n. 11, 11 n. 12, 14 n. 19, 15 n. 21, 42–45, 53, 151 Strawson, P. F., 18 n. 29, 84, 146, 152 Swain, Corliss, vii Szabo, Zoltan Gendler, 32–35, 41 n. 62, 91 n. 47, 151, 152 Taylor, Kenneth, 31 n. 49, 33 n. 52, 152 Thau, Michael, ix, 4 n. 4, 5, 76 n. 26, 86 n. 42, 99–103, 106–7, 109 n. 3, 112 n. 9, 152 Thomason, Richmond, 148 Thornton, Robert, 151 truth, 54–56 Uehling Jr., Theodore E., 146, 147 unarticulated constituents, 3, 28–29 Urmson, J. O., 4 n. 4, 152 verbalism, 138–139 verbatim acceptability, 4, 63–70 Voltolini, Alberto, vii Walker, Ralph C. S., 87 n. 43, 152 Wall, Robert E., 145 Warnock, G. J., 146 Weintraub, Ruth, vii Wettstein, Howard K., 146, 147

Index 157 what is said, ix, 3, 4, 9 n. 8, 29 n. 45, 30–35, 41 n. 60, 44–45, 49, 55, 68, 72, 75–81, 82–91, 95–98, 105, 106, 144, 149, 150 Wiggins, David, 15 n. 21, 17 n. 25, 152 Wilson, Deirdre, 72, 90 n. 46, 151 Wittgenstein, Ludwig, 46–47, 108, 123 n. 19, 124 n. 23, 126 n. 26, 127 n. 27, 137, 152

Woodfield, Andrew, vii, 108 n. 1, 113 n. 10, 152 Wright, Crispin, 147, 149 Yagisawa, Takashi, vii Zalta, Edward N., vii, 16 n. 23, 150, 152