Philosophical Essays. Volume 1 Philosophical Essays, Volume 1: Natural Language: What It Means and How We Use It [Course Book ed.] 9781400837847

Citation preview

Philosophical Essays

This page intentionally left blank

PHILOSOPHICAL ESSAYS Volume 1 natural language: what it means and how we use it Scott Soames

princeton university press p r i n c e t o n a n d ox f o r d

Copyright © 2009 by Princeton University Press Published by Princeton University Press, 41 William Street, Princeton, New Jersey 08540 In the United Kingdom: Princeton University Press, 6 Oxford Street, Woodstock, Oxfordshire OX20 1TW All Rights Reserved Library of Congress Cataloging-in-Publication Data Soames, Scott. Philosophical essays : natural language : what it means and how we use it / Scott Soames. p. cm. Includes bibliographical references and index. ISBN 978-0-691-13680-6 (v. 1 : hardcover : alk. paper)—ISBN 978-0-691-13681-3 (v. 1. : pbk. : alk. paper) 1. Language and languages—Philosophy. 2. Linguistics. 3. Semantics. I. Title. P107.S67 2009 410.9—dc22 2008019492 British Library Cataloging-in-Publication Data is available This book has been composed in Sabon Printed on acid-free paper. ∞ press.princeton.edu Printed in the United States of America 10 9 8 7 6 5 4 3 2 1

T H E S E V O L U M E S A R E D E DI C AT E D T O T H E P E O P L E I L OV E

My wife, Martha My sons, Greg and Brian and to the memory of my parents Bill and Ruth Soames

This page intentionally left blank

Contents

The Origins of These Essays

ix

Introduction

1

Part One Presupposition

21

Essay One A Projection Problem for Speaker Presuppositions

23

Essay Two Presupposition

73

Part Two Language and Linguistic Competence

131

Essay Three Linguistics and Psychology

133

Essay Four Semantics and Psychology

159

Essay Five Semantics and Semantic Competence

182

Essay Six The Necessity Argument

202

Essay Seven Truth, Meaning, and Understanding

208

Essay Eight Truth and Meaning—in Perspective

225

Part Three Semantics and Pragmatics

249

Essay Nine Naming and Asserting

251

Essay Ten The Gap between Meaning and Assertion: Why What We Literally Say Often Differs from What Our Words Literally Mean

278

viii • Contents

Essay Eleven Drawing the Line between Meaning and Implicature—and Relating Both to Assertion

298

Part Four Descriptions

327

Essay Twelve Incomplete Definite Descriptions

329

Essay Thirteen Donnellan’s Referential/Attributive Distinction

360

Essay Fourteen Why Incomplete Definite Descriptions Do Not Defeat Russell’s Theory of Descriptions

377

Part Five Meaning and Use: Lessons for Legal Interpretation

401

Essay Fifteen Interpreting Legal Texts: What Is, and What Is Not, Special about the Law

403

Index

425

The Origins of These Essays

Essay One

Linguistic Inquiry 10 (1979): 623–66. Reprinted by permission of Linguistic Inquiry.

Essay Two

D. Gabbay and F. Guenthner, eds. Handbook of Philosophical Logic, vol. 4: Topics in the Philosophy of Language. Dordrecht: Reidel, 1989. 553–616. Reprinted with the kind permission of Springer Science and Business Media.

Essay Three

Linguistics and Philosophy 7 (1984): 155–79. Reprinted with the kind permission of Springer Science and Business Media.

Essay Four

Jerrold J. Katz, ed. The Philosophy of Linguistics. Oxford: Oxford University Press, 1985. 204–26. Reprinted by permission of Oxford University Press.

Essay Five

James E. Tomberlin, ed. Philosophy of Mind and Action Theory. Philosophical Perspectives 3. Atascadero, Calif.: Ridgeview. 575–96. Reprinted by permission of Blackwell Publishing.

Essay Six

Linguistics and Philosophy 14 (1991): 575–80. Reprinted with the kind permission of Springer Science and Business Media.

Essay Seven

Philosophical Studies 65 (1992): 17–35. Reprinted with the kind permission of Springer Science and Business Media.

Essay Eight

Midwest Studies in Philosophy 32, forthcoming. Reprinted by permission of Midwest Studies in Philosophy and Blackwell Publishing.

Essay Nine

Zoltan Szabo, ed. Semantics vs. Pragmatics. New York: Oxford University Press, 2005. 356–82. Reprinted by permission of Oxford University Press.

Essay Ten

Written for the conference “Asserting, Meaning, and Implying,” in honor of Jay David Atlas, at Pomona College, April 2004. Previously unpublished.

x

•

The Origins of These Essays

Essay Eleven

Noûs, forthcoming. Reprinted by permission of Blackwell Publishing.

Essay Twelve

Notre Dame Journal of Formal Logic 27 (1986): 349–75. Reprinted by permission of the Notre Dame Journal of Formal Logic and Duke University Press.

Essay Thirteen

Philosophical Studies 73 (1994): 149–68. Reprinted with the kind permission of Springer Science and Business Media.

Essay Fourteen Teorema 24, no. 3 (2005): 7–30. Reprinted by permission of Teorema. Essay Fifteen

Written for the International Conference on Law, Language, and Interpretation at the University of Akureyri, Akureyri, Iceland, April 2007. Previously unpublished.

Philosophical Essays

This page intentionally left blank

Introduction

The fifteen essays in this volume span twenty-eight years of thinking about linguistic meaning—what it is, how we use it, and what questions should be answered by empirical theories dealing with it. A central task undertaken in the essays is to distinguish different, but intricately related, kinds of linguistically expressed information—including (i) what a sentence means (ii) what is presupposed in uttering it, (iii) what is asserted, and (iv) what is merely implicated, or suggested. An overarching theme, starting in part 1, is that the relationships among these are intricate, complex, and nontransparent—with far-reaching implications for theories of linguistic meaning and language use. As competent speakers, we do, of course, understand most of the sentences we use, and most of the utterances we hear. However, I argue, this understanding is not the source of privileged, systematic, and reliable intuitions about the semantic properties of sentences—such as when two of them mean the same thing—or about pragmatic properties of their use— such as which information carried by an utterance is due to the meaning of the sentence uttered, and which is due to other factors. In short, though the distinctions between the different types of information (i)–(iv) are theoretically important, the ability to speak and understand a language provides, at best, an incomplete and a highly fallible ability to reliably recognize instances of each. Nothing could be further from the truth than the old saw that theories of meaning (semantics) and use (pragmatics) are theories of the untutored linguistic intuitions of competent speakers. The same negative point can be made about the still popular view that linguistic theories of natural language in general, and semantic theories in particular, are attempts to explicate the unconscious internalized theories employed by speakers in understanding and using their language. Natural languages are, of course, human products. Both their causal origins, and their continued existence, stem from causally efficacious action— of an as yet largely unknown sort—in the minds of speakers. It does not follow, however, that the only sensible empirical inquiry about language is inquiry into the cognitive architecture that generates and sustains them. Sentences, and other expressions, have grammatical structures and representational contents that can be studied in abstraction from questions about how they initially came to have those structures and contents,

2 • Introduction

what psychological states and processes are responsible for their retaining them, or how speakers come to know whatever they do know about them. This nonpsychologistic perspective is fleshed out in the first two essays of part 2. The central notion in a theory of meaning is representational content. Complications aside, to know the meaning of a sentence is to know the ways in which uses of it represent the world to be. Thus, to a first approximation, the meaning of a sentence is a function from contexts of utterance to ways the sentence represents the world as being—with the meaning of subsentential expressions being that which they contribute to the meanings of sentences in which they occur. Since to represent the world as being a certain way is to impose conditions that must be satisfied if the world is to be way it is represented to be, the representational content of a sentence in a context determines its truth conditions (as used in the context). Hence, a central task of a theory of meaning for a language is to specify the truth conditions of its sentences. How far this takes us, and what more may be required of an adequate semantic theory, are the central preoccupations of the last three essays in part 2. The essays in parts 3 and 4 are concerned with the relationship between meaning, assertion, and conversational implicature. The main conclusions are: (i) that what is said or asserted by an utterance often exceeds the semantic content of the sentence uttered, (ii) that Gricean maxims responsible for conversational implicatures sometimes supplement semantic content in determining what is said or asserted, (iii) that sometimes the semantic contents of sentences are not among the propositions asserted by normal (nonironic, nonmetaphorical) utterances of them, (iv) that nevertheless there is an intimate and systematic relationship between the two, (v) that articulating the nature of this relationship makes possible solutions to previously outstanding semantic and pragmatic problems involving names, definite descriptions, and propositional attitude ascriptions, and (vi) that semantic and pragmatic theories incorporating these solutions fit naturally into the abstract, nonpsychologistic conception of linguistic theories sketched in part 2. The final essay, in part 5, shows how practical consequences about the interpretation of legal texts can be drawn from this theoretical perspective.

Part 1: Presupposition The first essay, “A Projection Problem for Speaker Presuppositions,” originally appeared in 1979. The projection problem, as it was then conceived, was that of determining the presuppositions of compound sentences on the basis of the presuppositions of their component clauses,

Introduction

•

3

and the propositions they express. Though the essay is nearly three decades old, it contains four items that may still be of some interest. First, it articulates a framework for extracting predictions about speaker presupposition from approaches that use different theoretical conceptions— (i) presuppositions as conditions for avoiding truth-value gaps, (ii) presuppositions as Gricean conventional implicatures, and (iii) presuppositions as requirements for incrementing sets of propositions previously assumed or established in the context. The empirical predictions extracted from these approaches can then be treated as a kind of common currency that allows us to compare and evaluate them. Second, the essay specifies data demonstrating the inadequacy of an important approach, advocated by Lauri Karttunen and Stanley Peters, in which Montague semantics is extended to cover non-truth-conditional aspects of meaning by dividing the semantic content of an expression into two parts—its contribution to truth conditions and its contribution to conventional implicatures (which were identified with presuppositions). (A related negative result for three-valued systems is obtained simultaneously.) Third, the essay argues that a positive solution to the projection problem must incorporate conversational implicatures in a way that had previously been disallowed. However, it also shows that the simplest strategies for doing so fail to accommodate all relevant data. Finally, the essay poses the challenge of finding a descriptively adequate, and theoretically explanatory, way of combining needed features of the semantic (threevalued or conventional implicature) approach with essential pragmatic features.1 Three years later, I tried, unsuccessfully, to answer this challenge in “How Presuppositions Are Inherited: A Solution to the Projection Problem” (1982). Though that paper discusses a great deal of data, and provides a snapshot of the strengths and weaknesses of the then leading approaches, in retrospect it is clear that the proposed solution was both descriptively incomplete, and theoretically unsatisfying. This, along with many other things, is discussed in essay 2, “Presupposition.” There, I distinguish several notions of presupposition drawn from the philosophical literature: “logical presupposition,” defined as a necessary condition for having a truth-value,2 “expressive presupposition,” defined as a necessary condition for a sentence’s expressing a proposition,3 and “pragmatic presupposition,” defined as a requirement that a 1

One minor descriptive problem with the discussion in the essay is its incorrect description of the presuppositions of certain classes of sentences, including A Ved, too, A Ved again, and It was A that Ved. These errors are corrected in the final section of essay 2. 2 Derived from Gottlob Frege. 3 Derived from one of two lines of thought found in, but conflated by, Peter Strawson.

4 • Introduction

use of a sentence places on the existing conversational record (including the set of shared background assumptions) at the time of utterance.4 The question is raised as to how, if at all, these notions apply to the vast outpouring of descriptive data on presupposition gathered by linguists in the 1970s and early 1980s. It is argued that (some) requirements placed by pragmatic presuppositions on the conversational record must be defeasible in two different ways: (i) by de facto accommodation—whereby certain apparent violations of presuppositional requirements result in the conversational record being updated to accommodate them—and (ii) by de jure accommodation—in which the presuppositional requirements are cancelled, or rendered inoperative, by other pragmatic implicatures, or features of the context, with which they may conflict. It is further argued that the need for these two forms of accommodation undermines both logical presupposition and conventional implicature as sources of at least some pragmatic presuppositions. More strongly, it is argued against Stalnaker, on independent grounds, that logical presuppositions—in the sense of necessary conditions on the avoidance of truth-value gaps—are never the source of pragmatic presuppositions. On a more constructive note, the rudiments of Irene Heim’s ingenious use of Discourse Representation Semantics (an extension of Montague semantics to provide semantic interpretations of whole discourses) to solve the projection problem are discussed. After pointing out a number of the attractive features of the system, I offer some descriptive and explanatory criticisms. The essay closes with a brief discussion of the varieties of presupposition, and the suggestion that presupposition may not be a unitary phenomenon, after all.5

Part 2: Language and Linguistic Competence These essays are concerned with how syntactic and semantic theories are related to psychological theories of the linguistic competence of language users. Essays 3 and 4—“Linguistics and Psychology” and “Semantics and Psychology”—argue that, despite being relevant to cognitive psychology, syntax and semantics are not themselves branches of that discipline. Some linguistic facts, including certain facts about truth and reference, are directly relevant to theories of natural language in ways they are not 4

Derived from Robert Stalnaker. The reader interested in a more recent, comprehensive account of presupposition in natural language should consult Beaver (2000). Another fascinating piece, directly relevant to essays 1 and 2, is Kripke (n.d.). 5

Introduction

•

5

relevant to psychological theories of linguistic competence.6 Similarly, psychological facts about speakers’ linguistic performance bear on psychological theories of competence in ways they don’t bear on syntactic and semantic theories. Since theories in the two domains answer different questions and are responsible to different facts, they are conceptually distinct. In addition, I argue, there is little reason to expect the formal structures posited by linguistic theories to be isomorphic to cognitive structures posited by psychological theories of linguistic competence. Although natural language is a quintessential human product, linguistic theories are only indirectly related to questions about the cognitive architecture that produces it. For example, syntactic and semantic universals limiting the formal systems qualifying as possible natural human languages constrain feasible hypotheses about the cognitive structures responsible for language acquisition and use, but they don’t identify such structures. In general, theories in syntax and semantics are rich sources of psychological questions, even though they typically don’t answer them. Essay 5, “Semantics and Semantic Competence,” extends the nonpsychological conception by arguing that semantic competence does not arise from knowledge of the semantic properties of expressions characterized by a correct semantic theory. First, I explain why knowledge of truth conditions is neither necessary nor sufficient for understanding. Next I sketch a conception of structured propositions, and outline a semantic theory in which meanings of sentences are functions from contexts to such propositions. Although this semantic theory is an improvement over a strict truth-conditional account, knowledge of its theorems, linking sentencecontext pairs to propositions, is neither necessary nor sufficient for understanding.7 Thus, the gap between semantics and semantic competence remains, even on the most promising semantic theory. The argument is 6 This is, I think, a better way of putting the point than was done in the section “Truth Conditions” of essay 3, where an unduly narrow characterization of psychological theories was suggested that was all-too-easily (mis)read as excluding so-called wide contents of mental states. Today I would state the argument differently. The semantic facts of a language are constituted by facts about how its terms were initially introduced, the beliefs and intentions of subsequent language users, the norms and presuppositions governing their interactions with one another, and the relations they bear to things in their environments. Neither the semantic facts of the language, nor many of the underlying facts constituting them, are facts about which psychological theories of individual agents make claims or predictions. Rather, many are facts on which claims about wide-psychological content depend. 7 I would now formulate the argument for this point slightly differently. Whereas I was formerly inclined to accept assumption (13a) in the essay, Richard (1993) has convinced me that it is more questionable than I first realized. (See my discussion in Soames n.d.) Without (13a), instances of what is called “schema K” in the essay cannot be derived from the semantic theory, in which case it is obvious that knowledge of the relevant theorems is neither necessary nor sufficient for understanding the language.

6 • Introduction

capped by a point about explanatory priority. In many cases in which competent speakers do have knowledge of the semantic properties of sentences and other expressions, their competence doesn’t derive from that knowledge; rather their semantic knowledge is derived from, and explained by, their competence. In essay 6, “The Necessity Argument,” I distinguish the basis of my opposition to Noam Chomsky’s psychological/conceptualist conception of linguistics from Jerrold Katz and Paul Postal’s “realist”/Platonist criticism of that position. The issue concerns theorems—essential to any adequate semantic theory—stating that certain sentences are necessary truths/falsehoods, and that certain sentences follow necessarily from other sentences. The criticism by Katz and Postal is that since the theorems of empirical psychological theories—of the sort Chomsky takes linguistic theories to be—must be contingent, they cannot include semantic claims about the modal status of sentences. My essay explains why this is not so. Opposition to the Chomskian reduction of linguistics to psychology need not depend on one’s views about the ontology of linguistic objects, one’s views of their modal properties, or one’s conception of how we find out about them. Essays 7 and 8, “Truth, Meaning, and Understanding” and “Truth and Meaning—in Perspective,” raise a foundational question about Donald Davidson’s program in semantics. What justification, if any, can be given for taking a Davidsonian truth theory to be a theory of meaning? The two essays canvass the main attempts to answer this question, and explain why—despite the attractiveness of the program, and the reasons for initial optimism about providing a positive answer—none of the major attempts to do so can be accepted. In addition, it is argued, this series of attempts has led to a mismatch between the proposed justificatory requirements of the program and the empirical research done to advance it. The proposed requirements that must be satisfied by true, compositional, truth theories, if they are to count as candidates for theories of meaning, have been tightened to the point that important Davidsonian analyses of linguistic constructions advanced by working semanticists don’t satisfy them. Together, the problems of justification and mismatch constitute a crisis in this approach to the theory of meaning.

Part 3: Semantics and Pragmatics The next three essays sketch a new conception of the relationship between the theory of linguistic meaning (semantics) and the theory of the use of sentences to assert and convey information (pragmatics). Although similar in some respects to recent thoughts of Kent Bach (1994),

Introduction

•

7

Robyn Carston (2002), Stephen Neale (2007), and François Recanati (1993, 2003), my conception grew out of the position taken in Beyond Rigidity (Soames 2002). There, I proposed that a speaker who utters a sentence S often asserts several propositions, including some that are pragmatically enriched, and that the semantic content of S is a proposition which, for every normal context C in which S is uttered, is among those asserted. These points were used to explain how substitution of coreferential names in simple sentences may change the propositions asserted by utterances of those sentences, even when it doesn’t change the propositions semantically expressed. Extending this lesson to sentences used to ascribe assertions and beliefs allowed me to reconcile a Millian semantic analysis of linguistically simple names with Fregean intuitions about the assertions made, the beliefs ascribed, and the information conveyed by utterances of sentences differing only in the substitution of coreferential names. In short, I argued that a new conception of the relationship between semantics and pragmatics could be used to make an important contribution to the solution of Frege’s puzzle. In essay 9, “Naming and Asserting,” I explain in detail why the Beyond Rigidity account of semantics and pragmatics has to be revised. According to the updated view, the semantic content of S is not always a complete proposition, but rather is a set of conditions that constrains the candidates for assertion, while allowing speakers a measure of freedom for pragmatic enrichment within those constraints. On this view, assertions must be pragmatic enrichments of semantic content. When the semantic content of S in C is a complete proposition it qualifies as a degenerate case of pragmatic enrichment (of itself). However, it counts as asserted by an utterance of S only if it is a relevant and obvious consequence of salient presuppositions in C plus the (potentially) enriched propositions that the speaker primarily intended to assert. In addition to being independently motivated, this revision is crucial in extending the ability of the Millian to accommodate the pretheoretic intuitions driving Frege’s puzzle. The essay closes with illustrations of how the view generalizes to other cases, including indexicals and possessives. The new conception of meaning as least common denominator is further developed in essay 10, “The Gap between Meaning and Assertion: Why What We Literally Say Often Differs from What Our Words Literally Mean.”8 According to this conception, the semantic content of S is that which is common to what is asserted by utterances of S in all normal contexts. Although the content of S is often a complete proposition, 8 Delivered in April of 2004 at Pomona College at a conference, “Asserting, Meaning, and Implying,” in honor of Jay David Atlas. Originally intended for a conference volume that never materialized, the essay appears in print for the first time here.

8 • Introduction

and, hence, a proper candidate for being asserted and believed, in some cases it is only a skeleton, or partial specification, of such a proposition. In many contexts, the semantic content of S—whether it is a complete proposition or not—interacts with information that is presupposed to generate a pragmatically enriched proposition that it is the speaker’s primary intention to assert. Other propositions count as asserted only when they are relevant, unmistakable, and a priori consequences of the speaker’s primary assertions, together with salient presuppositions of the conversational background. This framework is used to illuminate Kripke’s “Puzzle about Belief,” plus several related problems involving definite descriptions. Essay 11, “Drawing the Line between Meaning and Implicature—and Relating Both to Assertion,” integrates Gricean conversational maxims into the new semantic-pragmatic model. I argue that, in addition to generating implicatures, the maxims play an important role in determining what is asserted. Sentences containing bare numerical quantifiers—of the form n Fs—are used to illustrate the point. The crucial fact is that utterances of these sentences in different contexts result in the assertion of different propositions. Sometimes the content contributed by ⎡n Fs⎤ to the assertion made by an utterance of a sentence containing it is the same as that of the fully specified quantifier ⎡at least n Fs⎤. At other times it is the content of ⎡exactly n Fs⎤, ⎡at most n Fs⎤, or ⎡up to n Fs⎤—among others. This isn’t semantic ambiguity. Instead, I argue, the semantic contents of sentences containing bare numerical quantifiers are incomplete, requiring pragmatic enrichment in order to provide a complete proposition to be asserted. Conversational maxims help determine what is asserted by narrowing the class of possible enrichments to those that most effectively advance the conversation. When several enrichments are otherwise feasible, the maxims dictate that the agent select the strongest, most informative, and relevant propositions among them for which the agent can be presumed to have adequate evidence. In this way, the maxims contribute to the truth conditions of (what is said by) utterances, over and above their role in generating conversational implicatures. This extension of the integrated semantic-pragmatic model has a methodological corollary. In tracing part of the information asserted or conveyed by an utterance to the conversational maxims—instead of the meaning of the sentence uttered—we are showing that the information is rationally extractable from the utterance, together with the least common denominator conception of meaning, even if real, nonidealized, speakers don’t employ the maxims either consciously or unconsciously. On this conception, the meaning of a sentence is the minimal information associated with it that must be mastered by a rational agent—over and above the ability to communicate intelligently, efficiently, and cooperatively

Introduction

•

9

with other members of the linguistic community. Consequently, no matter what idiosyncratic processes individual speakers actually go through in extracting information from utterances, the question of what their sentences mean—and of what part of that which is asserted or conveyed is due to meaning, and what part is due to other, pragmatic factors—is determined by idealized rational reconstruction, not psychological research. In this way, the integrated semantic-pragmatic model sketched in part 3 extends the nonpsychological conception of linguistic theories articulated in part 2. It also provides a diagnosis of the problematic source of what has been called “Grice’s paradox.” It is a basic presupposition of the classical Gricean model that the generation of conversational implicatures requires speaker-hearers to have a clear, reliable grasp of the meaning of the sentence uttered, which is closely, and transparently, related to what is said by an utterance of it. If this presupposition were correct, it would seem that we should be able to rely on that grasp to determine whether a piece of information conveyed by an utterance is, or is not, conversationally implicated. If it is part of what is said, this should be recognizable from our understanding of the sentence uttered. If it is conversationally implicated, there should be a canonical derivation demonstrating it to be. Either way, it would seem, there is little room for serious theoretical controversy about what is, and what isn’t, implicated. In point of fact, however, there is no shortage of such controversy. That is Grice’s paradox. The error leading to the paradox is, I think, that of too closely identifying what is said or asserted by an utterance with the meaning (or semantic content) of the sentence uttered. On the semantic-pragmatic account developed in part 3, semantic content is too theory-laden to be the object of reliable, systematic knowledge on the part of competent speakers. Although Grice-like derivations are crucial to the role of the maxims in determining both what is asserted and what is conversationally implicated, these derivations are not routinely reconstructable by ordinary speakers, because the semantic contents that initiate the derivations are not psychologically available to them, simply by virtue of their linguistic competence.

Part 4: Descriptions The next three essays deal with singular definite descriptions, with special focus on incomplete descriptions like ‘the cook’ and ‘the table’—which are often used to assert truths about one particular individual, despite the fact that the world contains many cooks and tables, and nothing in the descriptions distinguishes one from all the rest. Essay 12, “Incomplete

10

•

Introduction

Definite Descriptions,” critiques the lessons drawn from these expressions by Jon Barwise and John Perry for their theory of meaning—situation semantics. That theory, roughly put, is what you get by starting with the idea that the proposition semantically expressed by a sentence (relative to a context of utterance) is the set of circumstances in which it comes out true, and insisting that these circumstances are not different possible states of the world as a whole—but different states of parts of the world. Two linguistic constructions that played important roles motivating this approach were propositional attitude ascriptions and incomplete definite descriptions—both of which seemed to require partial circumstances of evaluation to resolve what were then taken to be intractable problems. In essay 1 of volume 2, I argue that, in fact, the problems posed by attitude ascriptions can’t be solved in this way. In essay 12 of this volume, I argue for a similar conclusion about incomplete descriptions. The guiding idea in the latter case was that true, unproblematic uses of ⎡ The F is G⎤, when ⎡the F⎤ is incomplete, require the sentence as a whole to be evaluated in partial circumstances in which the uniqueness condition carried by ‘the’ is satisfied because one and only one thing satisfies F in the circumstances—despite the fact that many things satisfy F in reality as a whole. Essay 12 shows in detail why this idea is incorrect, as originally formulated in Barwise and Perry (1983), and why the modification of the basic framework later given in Barwise and Perry (1985) doesn’t change the result. The central positive conclusion drawn is that the use of incomplete descriptions to assert truths is explained by the contextual supplementation of their contents by salient objects, properties, events, or states of affairs—which render the descriptions contextually complete.9 Is this contextual supplementation semantic or pragmatic? At the time, I was inclined to think in some cases—when the utterance of ⎡The F is G⎤ results in the assertion of something true and nothing false—it was semantic. The argument for this, given in note 5 of the essay, was that if the supplementation of the description was pragmatic then, although the speaker could be regarded as saying something true (to which the utterance of the description contributed a completed assertive content), the speaker would also have to be regarded as saying something false, since the semantic content of the sentence (to which the description contributed an incomplete semantic content) would be false. The unspoken presupposition of this argument was that whatever else may be asserted 9 Although this conclusion can be implemented within situation semantics, it does not require partial situations as circumstances of evaluations for sentences (as opposed to fullyfledged possible world-states). Thus it removes one of the central motivations for that framework.

Introduction

•

11

by an utterance of S, the semantic content of S (the proposition it semantically expresses) is always asserted, unless something special about the utterance—irony, sarcasm, or an obviously canceling conversational implicature—indicates otherwise. Though this assumption was standard at the time, it is, of course, rejected by the new conception of the relationship between meaning and assertion developed in part 3. Applying that conception to cases involving incomplete descriptions—the details of which are explained in essays 10 and 14—leads to the conclusion that the needed contextual supplementation of incomplete definite descriptions pointed out in essay 12 is pragmatic, rather than semantic. This result fits nicely with what I had to say about the distinction between attributive and referential uses of definite descriptions. When a description is used referentially, the speaker’s utterance results in the assertion of a singular proposition about the real or presumed designation of the description. When it is used attributively no such singular proposition is asserted. In the section “Attributive/Referential” of essay 12, I argued that, contra Barwise and Perry, this distinction does not rest on a semantic ambiguity, but is purely pragmatic. The point is obvious when the singular proposition asserted by a referential use is about the presumed, rather than the real, designation of the term used referentially. Since the pragmatic mechanism at work in these cases doesn’t cease to operate when the presumed designation coincides with the semantic one, it is argued that no semantic ambiguity is needed to account for the facts about assertion that characterize referential uses. This point, which naturally generalizes to uses of incomplete definite descriptions, is reprised in essays 10 and 14, where the same mechanisms that explain successful uses of indefinite descriptions also explain the distinction between attributive and referential uses. Essay 13, “Donnellan’s Referential/Attributive Distinction,” begins by explicating the contribution made by Keith Donnellan’s (1966) introduction of the distinction to the theory of direct reference and the assertion of singular, Russellian propositions. It then returns to the question of semantics versus pragmatics. This time, however, the focus is on cases involving discourse anaphora like those in (1).10 (1) A man came to the office this morning. He / The man tried to sell me an encyclopedia. Roughly put, Donnellan’s thesis was that the antecedent in (1) is used referentially to pick out a certain individual who is both the semantic referent and content of the pronoun or description that is anaphoric on it in the

10

These are discussed in Donnellan (1978).

12

•

Introduction

discourse. In cases like this, he argued, descriptions (like ‘the man’) are directly referential terms the contents of which are the individuals that speakers use them, referentially, to pick out. I maintained—based on considerations parallel to those Donnellan (1979) himself used in another connection—that this was not so. Instead, I suggested, the anaphoric terms are to be understood along the lines first proposed by Gareth Evans (1977), and later refined by Martin Davies (1981) and Stephen Neale (1990). Essay 14, “Why Incomplete Definite Descriptions Don’t Refute Russell’s Theory of Descriptions,” is the latest, and most complete, statement of my views on incomplete descriptions. There, I defend a modified version of Russell’s theory of descriptions in which ⎡the F⎤ is treated as a generalized quantifier, and sentences containing it are assigned Russellian truth conditions. The task is to explain why this theory is not defeated by successful uses of incomplete definite descriptions, when it is taken as a semantic account of ordinary English. Doing this requires the conception of meaning as least common denominator, and the resulting distinction between what a sentence means in the common language versus what a speaker means by it, and uses it to assert, on a given occasion. As I emphasize in the essay, this was not how Russell himself looked at the matter. His main task was not to give a theory of the meaning of sentences in the common language of a linguistic community, but to specify what an individual means in using those sentences on a given occasion. From this individualistic perspective, incomplete definite descriptions don’t pose serious problems. Since pragmatic enrichment of utterances is ubiquitous, what a speaker means by a superficially incomplete description ⎡the F⎤ on a given occasion is typically not incomplete. From this perspective, the fact that the assertive content of ⎡the F is G⎤ differs widely from speaker to speaker, and context to context—and is considerably richer than the semantic content of the sentence in the common language—is of no particular concern. It becomes a concern only when one tries to give a semantic theory of what that content is, and how it relates to what the sentence is used to assert on particular occasions. To adopt a Russellian theory of descriptions in a semantic theory of this sort, one must not, I argue, assume that when speakers assertively utter a sentence S containing an incomplete description, they mean, and assert, the proposition that is the semantic content of S (in the context). Instead, they typically assert a proposition which is a pragmatic enrichment of that content. An alternative view—to which I devote a little, but not enough, attention in essays 10 and 14—agrees that contextual supplementation is the key to understanding successful uses of incomplete descriptions, but argues that it cannot be pragmatic. Instead, the view maintains, such supplementation must be understood as the contextual filling out of a hidden semantic parameter. The key argument for this view, given

Introduction

•

13

by Jason Stanley (2000, 2002), comes from apparent instances of variablebinding, like those in (2) and (3). (2) (3)

Every student answered every question. Every student answered every question on her exam.

Consider a case in which (2) is uttered with the intention of asserting (what is expressed by) (3). Stanley argues that this can’t come from free pragmatic enrichment, but rather results from the contextual interpretation of a hidden underlying variable. On his view, the underlying logical structure of (2) is something along the lines of (4). (4)

[Every student x][Every question y: f(x)] x answered y

The context in question supplies the interpretation of ‘f’—assigning it a function from individuals to the property of being on the individual’s exam. The effect is to give the interpretation is a question on x’s exam to the formula restricting the range of the second quantifier. Thus, after contextual interpretation, (2) semantically expresses the same proposition as (3). Stanley’s point, apparently, is that since the proposition expressed by (2) in the context is one that can arise only from interpreting linguistic material containing a variable bound by the initial quantifier, (2) itself must contain that linguistic material in logical form. This can be so only if occurrences of nouns are quite generally accompanied by hidden variables of this sort. Thus, (6) is the underlying logical structure of (5), and the need for contextual completion is built into the linguistic structure of all relevant examples. (5) (6)

Every bottle is in the fridge. [Every bottle y: f(y)] y is in the fridge.

Why, even if this were so, contextual enrichment could not be pragmatic is, I think, not fully clear. Even less clear, however, is why the use of (2) to assert the proposition expressed by (3) should require a hidden variable in logical structure. Stanley’s argument for this seems to depend on the idea that what needs to be added to (2) to get the proposition expressed by (3) is not an independent semantic unit, but rather is something that interacts semantically with an operator that is already syntactically represented. Since the supposed unarticulated constituent [what needs to be added to get the content of (3) from (2)] is not the value of anything in the sentence uttered, there should be no readings of the relevant linguistic constructions in which the unarticulated constituent varies with the values introduced by operators in the sentence uttered. Operators in a sentence can only interact with variables in the sentence that lie within

14

•

Introduction

their scope. . . . Thus . . . [the] interpretation [of the unarticulated constituent] cannot be controlled by operators in the sentence. (Stanley 2000, 410–11; my emphasis)11 The point here seems to be that what needs to be added to (2) to get the interpretation (3) is not a single semantic value, but a formula, ‘is on x’s exam’, the semantic value of which varies with different assignments to ‘x’. Because of this, Stanley thinks, pragmatic enrichment cannot result from pragmatically enriching a constituent of the structured proposition semantically expressed by the hidden-variable-free formula (7). (7)

[Every student x] [Every question y] x answered y

Thus, he assumes, if pragmatic enrichment were to occur at all, it would have to involve adding extra words to interact with the words and phrases already there. After giving arguments against this, he concludes that since pragmatic enrichment of (2) cannot explain the use of (2) to assert the proposition expressed by (3), (2) must itself have a hidden variable in logical structure requiring contextual interpretation. This argument is flawed in several ways.12 The most important point, for our purposes, is that it is perfectly possible to get the proposition expressed by (3) by pragmatically enriching the proposition semantically expressed by (2)—on the hypothesis that the logical structure of (2) is given by (7), rather than (4). On this hypothesis, the structured proposition semantically expressed by (2) is a complex consisting of the higher-order property expressed by the quantifier ‘[every student x]’ plus the property contributed by the formula ‘[Every question y] x answered y’ to which it attaches.13 The former is the property of being instantiated by every instantiator of the property of being a student (i.e., 11 Note, at the beginning of the passage the unarticulated constituent is spoken of as a kind of content, or semantic value; at the end it is spoken of as a linguistic constituent that gets semantic values. This conflation contributes to the confusion in the argument. 12 At bottom, I suspect it involves a misunderstanding of quantification. In standard semantics, the truth-value of a quantified formula ⎡Qx (Rxy)⎤, relative to an assignment A, is computed not from the extension of ⎡Rxy⎤, relative to A, but from its extension relative to a class of assignments related to A. As a result, extension is not strictly compositional. A similar remark holds for semantic content. This may make it appear as if occurrences of formulas to which quantifiers attach don’t contribute properties to the semantic contents of larger quantified formulas (relative to assignments). However, this is a mistake. In general, an occurrence of such a formula contributes a property P (or something that determines P), of which the higher-order property contributed by the quantifier occurrence is predicated. There is no bar to pragmatically enriching a proposition of this sort; one simply replaces P by some enrichment P* of it. For an explanation of the technical points about quantification, see Salmon (2006a, 2006b). 13 This talk of properties can be treated as Russellian functions from arguments to structured propositions about those arguments, along the lines indicated in essay 5.

Introduction

•

15

being instantiated by every student). The latter is the property of being one who answered every question. Pragmatic enrichment consists in replacing the latter property with an enriched—more restricted—version of it. In this case, the enriched property is that of being one who answered every question on one’s exam. No new variables or any other words are introduced. We simply have an operation on contents. Thus, Stanley’s variable-binding argument, on which he places so much weight, has no force against the view of pragmatic enrichment advocated here. Much more beyond this (plus the hints sketched in essays 10 and 14) would have to be said to establish that the pragmatic view is definitely superior to Stanley’s contextually semantic approach. However, it may be worth mentioning a final point to which I attach some weight. The semantic content of a sentence—including (8)—can be, but does not have to be, thought of as a proposition. (8)

Every F is G.

Instead, it can be treated as a set of constraints requiring the proposition asserted to be one in which some restriction/enrichment of the property of being instantiated by everything that instantiates FP (the property expressed by F) is predicated of some restriction/enrichment of GP (the property expressed by G).14 Whenever we have a quantifier phrase we get a constraint on what is asserted requiring some contextually relevant restriction/enrichment of the semantic content of that phrase. In Stanley’s system this is reflected in having one or more hidden domain variables inserted into every quantifier phrase—no matter how much the range of the quantifier has already been restricted by the linguistic material that is overtly present. That strikes me as redundant. Suppose we have a very long restricting clause that appears overtly on the quantifier. Of course, no matter how long it is, we could have always said more, but surely this is no reason to think that we have in effect already said more by including unpronounced variables in our sentence.

Part 5: Meaning and Use Essay 15, “Interpreting Legal Texts: What Is, and What Is Not, Special about the Law,” applies lessons learned from the study of language to a question the resolution of which has practical consequences for our lives. Most prominent among these lessons are those involving the nondescriptionality of names, natural kind terms, and other expressions previously thought to be definable, the gap between meaning and assertion, and a 14

We here let every property count as a (vacuous) restriction/enrichment of itself.

16

•

Introduction

new understanding of vague terms. The essay incorporates these lessons into a model for (i) determining the content of positive law from the linguistic texts used to enact it, and the circumstances surrounding its enactment, and (ii) the principles used to derive legally correct results about particular cases from the content of the governing law. The content of a law is what is asserted or stipulated by the lawmakers in adopting the legal text. As with assertion (or stipulation) generally, this content is only partially determined by the linguistic meaning of the relevant text. Other factors include the purposes, presuppositions, and communicative intentions of the relevant actors. It is shown how, in particular legal cases, failure to take these factors into account—and thereby to conflate what a legal text means with what it is used to assert or stipulate—leads to the content of the law being misidentified, and to the cases being wrongly decided. One example of this sort involves referential uses of language—on analogy with referential uses of descriptions discussed in essays 10 and 12–14. In these cases, lawmakers or other rule makers assert or stipulate something about particular things they have in mind, over and above the semantic contents of the sentences they use (which may, or may not, themselves be asserted or stipulated). In a different kind of case, lawmakers use sentences the semantic contents of which are incomplete, and require contextual supplementation. For example, suppose that, in passing a law, the lawmakers say, “The use of firearms within city limits is prohibited.” Whenever an instrument is used, it must, of course, be used in some way, or for some purpose. However, the meaning of the lawmakers’ sentence is silent about which uses are to be prohibited. To determine this crucial aspect of the law, one must ascertain what uses the lawmakers understood themselves to be prohibiting. The mistake to be avoided is in thinking that because the linguistic meaning of ‘the use of firearms’ is minimal and nonspecific, its contribution to what they asserted or stipulated must also be minimal and nonspecific—something along the lines of ‘the use of firearms in any way’. Although one could, in very special contexts, use the sentence to forbid all uses of firearms—including as wall decorations, subjects of photographs, museum exhibits, and the like—more normally the sentence would be used to prohibit firearms as weapons. Thus, to interpret the law one must consult not just linguistic meaning, but also communicative intent. This point, though obvious from a correct understanding of the relationship between meaning and assertion, has proven too difficult for some of our highest courts. Part (ii) of the model examines how legally correct resolutions of particular cases are related to the content of the governing law. Here the main points of interest are the ways in which legal interpretation may

Introduction

•

17

justifiably create new law. Surprising as it may initially seem, sometimes the content of existing law—that which the lawmakers have asserted or stipulated—determines an outcome that is legally incorrect in a particular case. The reason for this, I argue, lies in a proviso giving judges (and other officials) authority to make minor adjustments in law to avoid transparently undesirable results in cases not previously contemplated. Although explicitness in formulating laws is desirable, it is impossible to anticipate every eventuality, and, inevitably, one reaches a point of diminishing returns. Recognizing this, we implicitly give judges a limited prerogative to make changes in the law they interpret. In this respect legal interpretation is special, and different from other ways of interpreting texts. Another way in which legal interpretation sometimes creates new laws occurs when the resolution of a case requires deciding issues that had previously been left vague. Often, this occurs when some crucial term in a legal text is governed by linguistic rules that determine items to which it applies and items to which it does not apply, while being silent about a range of intermediate items. If the case at hand involves a crucial item for which the term is undefined, and if the lawmakers themselves did not adopt a special contextual standard resolving the indeterminacy, the court itself is called upon to do so. Although there are linguistic and other constraints on how this can, legitimately, be done, there is typically a range of freedom within which judges have discretion. The exercise of this discretion is justified by the fact that the court has no alternative, if it is to decide the case at all, and by the fact that legislators are well aware of this aspect of adjudication, and often deliberately employ vague language with future judicial precisification in mind—thereby implicitly ceding limited legislative authority to courts.

Conclusion There was a time, just a few decades ago, when philosophy and the philosophical study of language were thought to be one and the same. Then, every significant philosophical question was thought to be a linguistic question. Thankfully, that is no longer so. Although the philosophy of language still has much to contribute to every area of philosophy, its main contemporary focus—and that of the essays in this volume—is on what it can tell us about language, representation, cognition, and communication, as independent, but interrelated, subjects in their own right. This development is part of a story that is as old as philosophy itself. At the birth of the discipline, the study of philosophy and the pursuit of theoretical knowledge were essentially one and the same. Over time, one discipline

18

•

Introduction

after another gradually emerged from its philosophical womb, and matured into an independent field of study. The most notable example in the last century was symbolic logic. Today, linguistic semantics and pragmatics, cognitive science, and the general study of language and information are struggling to emerge as productive and interconnected areas of scientific inquiry. The philosophical study of language has a central role to play in the development of this exciting, but still embryonic, intellectual enterprise. More than any other, this is the enterprise to which the essays in this volume have aspired to make contributions.

References Bach, Kent. 1994. “Conversational Impliciture.” Mind and Language 9:124–62. Barwise, Jon, and John Perry. 1983. Situations and Attitudes. Cambridge: MIT Press. ———. 1985. “Shifting Situations and Shaken Attitudes.” Linguistics and Philosophy 8:105–61. Beaver, David. 2000. Presupposition and Assertion in Dynamic Semantics. Stanford, Calif.: CSLI Publications. Carston, Robyn. 2002. Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell. Davies, Martin. 1981. Meaning, Quantification, and Necessity. London: Routledge and Kegan Paul. Donnellan, Keith S. 1966. “Reference and Definite Descriptions.” Philosophical Review 75:281–304. ———. 1978. “Speaker Reference, Descriptions, and Anaphora.” In Syntax and Semantics, vol. 9, Pragmatics, ed. Peter Cole, 47–68. New York: Academic Press. ———. 1979. “The Contingent Apriori and Rigid Designators.” In Contemporary Perspectives in the Philosophy of Language, ed. Peter A. French, Theodore E. Uehling, and Howard K. Wettstein, 12–27. Minneapolis: University of Minnesota Press. Evans, Gareth. 1977. “Pronouns, Quantifiers, and Relative Clauses (I).” Canadian Journal of Philosophy 7:467–536. Kripke, Saul A. n.d. “Presupposition and Anaphora: Remarks on the Formulation of the Projection Problem.” Unpublished manuscript, City University of New York, Graduate Center. Neale, Stephen. 1990. Descriptions. Cambridge: MIT Press. ———. 2007. “On Location.” In Situating Semantics: Essays on the Philosophy of John Perry, ed. Michael O’Rourke and Corey Washington, 251–393. Cambridge: MIT Press. Recanati, François. 1993. Direct Reference. Oxford: Blackwell. ———. 2003. Literal Meaning. Cambridge: Cambridge University Press. Richard, Mark. 1993. “Articulated Terms.” In Language and Logic, ed. James E. Tomberlin, 207–30. Philosophical Perspectives 7. Atascadero, Calif.: Ridgeview.

Introduction

•

19

Salmon, Nathan. 2006a. “Terms in Bondage.” Philosophical Issues 16, no. 1: 263–74. ———. 2006b. “A Theory of Bondage.” Philosophical Review 115:415–48. Soames, Scott. 1982. “How Presuppositions Are Inherited: A Solution to the Projection Problem.” Linguistic Inquiry 13:483–545. ———. 2002. Beyond Rigidity: The Unfinished Semantic Agenda of “Naming and Necessity.” New York: Oxford University Press. ———. n.d. “What Are Natural Kinds.” Forthcoming in Philosophical Topics. Stanley, Jason. 2000. “Context and Logical Form.” Linguistics and Philosophy 23:391–434. ———. 2002. “Making It Articulated.” Mind and Language 17:149–68.

This page intentionally left blank

PA RT O N E

Presupposition

This page intentionally left blank

ESSAY ONE

A Projection Problem for Speaker Presuppositions

Introduction Ten years ago, in “The Projection Problem for Presuppositions,”1 Terence Langendoen and Harris Savin posed a problem that has been of central concern to theorists working on presupposition ever since. That problem was to determine “how the presuppositions and assertions of a complex sentence are related to the presuppositions and assertions of the clauses it contains.”2 In formulating this problem, Langendoen and Savin assumed that (1) Complex sentences bear presuppositions; (2) The presuppositions of a complex sentence are a function of the presuppositions and assertions of the clauses that make it up; and (3) A speaker who assertively utters a sentence that presupposes P indicates that he is presupposing the truth of P.3 1 Langendoen and Savin (1971). Although published in 1971, the paper was presented at the 1969 Spring Semantics Festival held at the Ohio State University in Columbus, Ohio. 2 Langendoen and Savin (1971), 55. 3 Although Langendoen and Savin do not explicitly concern themselves with what a speaker presupposes, they do claim that presuppositions express conditions “which must be satisfied (be true) for the sentence as a whole to be a statement, question, command, and so forth” (1971, 55). This characterization of presupposition can be connected with claims about speaker presuppositions via he following assumptions:

(i) Speakers (standardly) intend the sentences they utter to make statements (formulate questions, express commands, etc.). (ii) Speakers (standardly) intend their audience to recognize that the sentences they utter make statements (formulate questions, etc.). (iii) Competent speakers and hearers recognize what the sentences of their language presuppose (and they know this about each other). It follows from (i)–(iii), plus the Langendoen-Savin characterization of presupposition, that speakers (standardly) utter a sentence that presupposes P only if they presuppose P—that is, only if they take P to be true and assume their audience does, too. Since Langendoen and Savin presumably would agree with (i)–(iii), they would also accept (3) above. Many later theorists have given up the view that the presuppositions of a sentence express

24

•

Essay One

Proposed solutions to the projection problem have focused on attempts to specify the compositional function mentioned in (2). In their paper, Langendoen and Savin proposed a remarkably simple solution which became known as the Cumulative Hypothesis. The Cumulative Hypothesis (4) Compound (multi-clause) sentences inherit all of the presuppositions of their constituent clauses.4 The following examples illustrate how the hypothesis was supposed to work: (5) It wasn’t Alex who solved the problem. (6) If it wasn’t Alex who solved the problem, then he won’t be awarded the fellowship. (7) If the problem was difficult, then it wasn’t Alex who solved it [the problem]. (8) Someone solved the problem. Since (5) presupposes (8), and (5) is a constituent of (6) and (7), the Cumulative Hypothesis predicts that (6) and (7) also presuppose (8). Thus, Langendoen and Savin were able to account for the fact that a speaker who utters (5), (6), or (7) presupposes that the proposition expressed by (8) is true.5 However, many cases could not be handled by their hypothesis. (9) a. If there is a king of France, then the king of France is in hiding. b. There is a king of France. (10) a. If John has children, then all of his children are probably intelligent. b. John has children.

necessary conditions for it to make a statement. Some have gone further, rejecting the view that the presuppositions of a sentence are necessary conditions for it to make a true statement. Nevertheless, (3) has remained a commonplace among theorists who posit sentential presuppositions. 4 This is a somewhat simplified statement of Langendoen and Savin’s hypothesis. For a more detailed discussion, see Langendoen and Savin (1971, 57–60). 5 Langendoen and Savin did not explicitly consider cleft sentences like (5). However, examples analogous to (5)–(8) can he constructed for virtually any presupposition-bearing sentence. For example: (i) All of John’s children are intelligent. (ii) If all of John’s children are intelligent, then intelligence is probably hereditary. (iii) If intelligence is hereditary, then all of John’s children are probably intelligent. (iv) John has children.

A Projection Problem

•

25

(11) a. If the problem has been solved, it wasn’t Alex who solved it. b. Someone solved the problem. In each case, the consequent of (a) presupposes (b). Consequently, the Cumulative Hypothesis predicts that (a) itself presupposes (b). According to (3), this means that a speaker who assertively utters (a) indicates that he presupposes the truth of (b). Since this prediction is false, the LangendoenSavin account had to be given up. Nevertheless, their conception of the projection problem has remained. The conventional wisdom has been that (1)–(3) are correct, but that Langendoen and Savin’s algorithm for computing the presuppositions of compound sentences is incorrect. The task has been to come up with more complicated “inheritance” algorithms that will avoid the counterexamples to the original account. Two leading theorists who have been engaged in this enterprise are Lauri Karttunen and Stanley Peters. Their latest paper, “Conventional Implicature,” provides a detailed statement of one of the most explicit, systematic, and influential theories of presupposition available today.6 The central achievement of “Conventional Implicature” is the extension of the framework of Montague semantics to cover non-truth-conditional aspects of meaning. In this framework, the semantic content of an expression is divided into two components: its contribution to truth conditions, and its contribution to conventional implicatures. Accordingly, each phrase in English is associated with two expressions of intensional logic: an extension expression which specifies the contribution to truth conditions, and an implicature expression which does the same for implicatures. In keeping with the general theme of Montague semantics, the account is strongly compositional. For each basic phrase, the appropriate extension and implicature expressions are listed in the lexicon with the phrase. To each phrase derived by means of a syntactic rule, the paired semantic rule assigns an extension expression and an implicature expression as a function of the extension and implicature expressions of the phrases it is derived from.7 The result is the first fully formal and unified account of truth-conditional and non-truth-conditional aspects of meaning within the tradition of formal (model theoretic) semantics. 6 For earlier statements of the theory, see Karttunen (1973, 1974); Peters (1975, 1977, 1979); and Karttunen and Peters (1975, 1976). See also Stalnaker (1973) and (1974) for discussions of the notion of pragmatic presupposition, for commentary on some of Karttunen’s early views, and for an attempted solution to the projection problem for truthfunctional compounds that parallels that given in Karttunen (1974). 7 Karttunen and Peters (1979), 16.

26

•

Essay One

Of central concern to Karttunen and Peters is the way in which putative presuppositions are handled in their system. They argue persuasively that some examples that linguists have treated as presuppositions are, in fact, either generalized or particularized conversational implicatures.8 However, they also claim that many putative presuppositions are conventional implicatures that should be handled by their compositional mechanisms. Such putative presuppositions include those associated with the definite article, quantifiers, particles like even, too, only, implicative verbs such as manage and fail, factives, cleft and pseudocleft sentences, and Karttunen’s filters and holes. Particularly interesting is the way in which Karttunen’s filters and, or, and if, then go over in the new system. Their contributions to truth conditions are given by the standard two-valued truth tables for ‘&’, ‘∨’, and ‘→’.9 Their contributions to conventional implicatures are given by (12)–(14). (Where α is any sentence, αi is an expression that represents the conventional implicature of α, and αe is an expression that represents its extension, or truth conditions.) (12) (A and B)i = (Ai & (Ae → Bi)) (13) (If A, then B)i = (Ai & (Ae → Bi)) (14) (A or B)i = ((Ae ∨ Bi) & (Be ∨ Ai))10 8 The notion of a conversational implicature is due to Paul Grice. According to Grice, a speaker often suggests more by assertively uttering a sentence than can be gathered from the proposition that the utterance expresses. Many of these suggestions can be explained on the assumption that conversation is guided by principles of cooperation and rational communication that regulate the efficient exchange of information. Grice calls such principles conversational maxims and uses them to define the notion of conversational implicature. According to Grice, a speaker who assertively utters a sentence S conversationally implicates a proposition Q, if and only if he is presumed to be observing the conversational maxims; the supposition that he believes Q is required in order to make his utterance of S consistent with this presumption; and the speaker knows that the hearer can intuitively grasp that this supposition is required. Conventional implicatures (also due to Grice) are similar in that they are suggestions that can be gathered from an utterance of a sentence over and above the proposition that the utterance expresses. However, conventional implicatures do not arise from conversational maxims, but rather from the meanings of particular lexical items. For further discussion of these notions, see Grice (1965, 1967, 1975). 9 Karttunen and Peters (1979, 34). The truth tables are:

(31) (i)

(ii) &

B T F

A T F

T F F F

A

(iii) ∨

B T F

T F

T T T F

A

→

B T F

T F

T F T T

10 ‘A’ and ‘B’ are metalinguistic variables that range over English sentences. i is a function that assigns to each English sentence an expression in the intensional logic which

A Projection Problem

•

27

(12)–(14) are inheritance conditions intended (along with conditions for other constructions) to replace Langendoen and Savin’s Cumulative Hypothesis. The way these conditions work can be illustrated by showing how (13) accounts for examples (6)–(11). First consider (5) and (6). (5) It wasn’t Alex who solved the problem. Ae: Alex didn’t solve the problem. Ai: Someone solved the problem. (=(8))11 (6) If it wasn’t Alex who solved the problem, then he won’t be awarded the fellowship. According to Karttunen and Peters, (5) expresses the proposition that Alex didn’t solve the problem (Ae) and conventionally implicates the proposition that someone solved the problem (Ai). Condition (13) tells us that this proposition (Ai) is also conventionally implicated by (6). Thus, Karttunen and Peters predict that (6) conventionally implicates that someone solved the problem. Note that on this account the conventional implicature in question is not entailed by the propositions that (5) and (6) express.12 Thus, the truth of the implicature is not a necessary condition for an utterance of (5) or (6) to make a true assertion. Nevertheless, Karttunen and Peters represents its conventional implicatures. e is a function that assigns to each English sentence an expression in the logic that represents its truth conditions. ‘&’, ‘∨’, and ‘→’ are treated as denoting functions from pairs of formulas in the logic to formulas in the logic. Where x and y are formulas of the logic, the value of the function denoted by ‘&’ at the arguments x and y is the conjunction of x and y; similarly for ‘∨’ and ‘→’. The English words ‘and’, ‘or’, and ‘if, then’ are treated analogously in (12)–(14), except that the domains and ranges of the functions they denote consist of English sentences. I am indebted to Richard Cartwright for a discussion of this way of explaining the notion in (12)–(14). 11 Strictly speaking, Ai also includes the proposition that Alex exists. However, this does not affect the present discussion and can safely be ignored. 12 Karttunen and Peters recognize other cases in which the conventional implicatures are entailed. (i)

The king of France is wise. Ai: There is a king of France. (ii) It was Alex who solved the problem. Ai: Someone solved the problem. However, even in these cases the implicatures resulting from embedding the sentences as antecedents of indicative conditionals are typically not entailed by the conditionals themselves. (iii)

If the king of France is wise, then the French are lucky. Ai: There is a king of France. (iv) If it was Alex who solved the problem, then he will be awarded a fellowship. Ai: Someone solved the problem.

28

•

Essay One

claim that a speaker who assertively utters a sentence commits himself to the truth of its implicatures. On their view, a speaker who utters a sentence S, which conventionally implicates P, indicates that he is presupposing P.13 Thus, Karttunen and Peters predict that a speaker who utters (5) or (6) indicates that he is presupposing (8). Next consider (9), (10), and (11). (9)

If there is a king of France, then the king of France is in hiding. Ae = Bi: There is a king of France. (10) If John has children, then all of his children are probably intelligent. Ae = Bi: John has children. (11) If the problem has been solved, it wasn’t Alex who solved it. Ae = Bi: Someone solved the problem (the problem has been solved). In each case the proposition expressed by the antecedent (Ae) is identical with the proposition conventionally implicated by the consequent (Bi).14 Thus, the implicatures of the form (15) that condition (13) associates with these sentences are tautologies. (15) Ae → Bi Consequently, although Karttunen and Peters predict that a speaker who utters (9)–(11) commits himself to the truth of the corresponding instance of (15), they regard this commitment as trivial. Moreover, they correctly predict that such a speaker does not implicate that he is assuming the truth of Bi. Examples like these illustrate why Karttunen and Peters chose (13) rather than (16) as their inheritance algorithm for indicative conditionals. (16) (If A, then B)i = (Ai & Bi) Since (16) is just a special case of the Cumulative Hypothesis, (9)–(11) are counterexamples to it. Karttunen and Peters avoid this problem by rejecting (16) in favor of (13). This makes the account of (7) a little more complicated than one might first imagine. (7) If the problem was difficult, then it wasn’t Alex who solved it.

13 See Karttunen and Peters (1979, 11–15). For a related discussion of pragmatic presupposition, see Stalnaker (1973) and (1974). 14 For purposes of exposition, I am ignoring any other conventional implicatures that B (of (9), (10), or (11)) might have.

A Projection Problem

•

29

Ae: The problem was difficult. Bi: Someone solved the problem.15 According to condition (13), (7) conventionally implicates (presupposes) the proposition expressed by (17): (17)

The problem was difficult → Someone solved it (Ae → Bi)

Thus, Karttunen and Peters predict that a speaker who utters (7) indicates that he is presupposing (17). However, they admit that such a speaker would normally also be taken to be presupposing Bi. Their explanation of this is that the speaker’s commitment to Bi is based on his commitment to (17). A priori, there are three reasons why one might believe that (17) is true: (I) One might assume that its antecedent (Ae) is false. (II) One might assume that its consequent (Bi) is true. (III) One might not know the truth-values of antecedent or consequent, but have reason to think that the truth of the antecedent would provide good grounds for taking the consequent to be true. In the case of a speaker who utters (7), (I) can be eliminated immediately. The proposition expressed by Ae is not only the antecedent of what the speaker conventionally implicates; it is also the antecedent of the proposition he asserts. We may presume that he would not have asserted this conditional if he were assuming its antecedent were false.16 Consequently, a speaker who utters (7) implicates that he is not assuming the falsity of Ae. For such a speaker, (III) can usually also be eliminated. If one does not know in advance whether or not a problem has been solved, finding out that it is difficult would hardly constitute good grounds for thinking that someone has solved it.17 Consequently, Karttunen and Peters are able to reason as follows: A speaker who utters (7) conventionally implicates (presupposes) (17). Since his reason for taking (17) to be true (and for 15 Here again, I am ignoring any other conventional implicatures that B might have. In discussing this example, I will also not be concerned with the implicatures of A. 16 Karttunen and Peters (1979, 8–9) explain this presumption as the result of a conventional implicature of indicative conditionals. However, if, as they assume, the propositions expressed by such sentences are the same as those expressed by material conditionals of logic, then the presumption can be seen as a generalized conversational implicature resulting from the maxim of Quantity. 17 Perhaps one could construct special contexts in which this would be good grounds. In such contexts, Karttunen and Peters would claim that a speaker uttering (7) does not implicate that he is assuming Bi. However, in normal contexts a speaker who utters (7) is taken to be assuming this. See Karttunen and Peters (1979, 37–39) for a discussion of an analogous example.

30

•

Essay One

taking his audience to be prepared to accept it) is not (I) or (III), it must be (II). Thus, the speaker indicates that he is presupposing Bi. Since this is what had to be explained, Karttunen and Peters claim that condition (13) accounts for (7), as well as (6), (9), (10), and (11).18 Conditions (12) and (14), for conjunctions and disjunctions, function in essentially the same way. It should be noted, however, that there is a significant difference between (12) and (13) on the one hand and (14) on the other. The former are asymmetric regarding the contributions that A and B make to the conventional implicatures of conjunctions and conditionals. In each case the compound sentence is said to inherit all of the conventional implicatures of A, but not B. (14), on the other hand, is symmetric regarding the contributions of A and B. The reason for this symmetry is provided by examples like (18a, b): (18)

a. Either there is no king of France, or the king of France is in hiding. b. Either the king of France is in hiding, or there is no king of France.

These sentences are disjunctive counterparts of (9). In neither case does a speaker who utters the sentence implicate that there is a king of France.

18 On this account, (7) differs from (i) in that a speaker who utters the former, but not one who utters the latter, indicates that he believes (17) and (in most contexts) Bi to be true.

(i) If the problem was difficult, then Alex didn’t solve it. Bi: Someone solved the problem. Not all minimal pairs are accounted for by Karttunen and Peters in this way. (ii) If the problem was difficult, then it was Alex who solved it. (iii) If the problem was difficult, then Alex solved it. According to Karttunen and Peters, (ii) and (iii) express the same proposition; they differ only in that (ii) conventionally implicates (17), whereas (iii) does not. Note, however, that both (ii) and (iii) entail (17). Thus, a speaker who utters either one indicates that he is taking (17) to be true. In contexts in which the speaker’s commitment to (17) cannot plausibly be attributed to (III) above (that is, in contexts in which it is not plausible to suppose that he has non-truth-functional grounds for (17)), his utterance of (ii) or (iii) will indicate that he is assuming Bi. Thus, according to Karttunen and Peters, (ii) and (iii) do not differ regarding what they indicate that the speaker believes to be true. Rather, they differ in that a speaker who utters (ii) indicates that he regards (17) and (in most contexts) Bi to be uncontroversial, whereas a speaker who utters (iii) indicates only that he regards them to be true. This analysis may be questioned on the grounds that it seems to understate the difference between (ii) and (iii). However, since this issue is not directly related to the points I wish to discuss, I will leave open the question of how to correctly characterize these sentences.

A Projection Problem

•

31

In order to account for this, the inheritance condition for disjunction has been made symmetric.19 This, then, is the Karttunen-Peters solution to the projection problem for truth-functional compounds. Its significance lies not only in its handling of a substantial range of data, but also in its role as part of a unified and systematic semantics. This combination of empirical motivation and theoretical attractiveness makes it important to determine whether or not the Karttunen-Peters account is correct. In what follows, I will bring out several difficulties that demonstrate that it is not. Although some of these difficulties can be avoided by modifying the present inheritance conditions, others appear intractable and seem to call for a new approach to the presuppositions arising from truth-functional compounds.20 It should be emphasized that more is at stake than the fate of inheritance conditions (12)–(14). Karttunen and Peters have noted that there is an interesting relationship between their conditions and certain threevalued truth tables found in theories of logical presupposition.21 By examining this relationship further, I will illuminate certain problems with the Karttunen-Peters theory and show that some of these problems apply equally to theories based on logical presupposition. Thus, if my critique is correct, the two major accounts of the presuppositions of truth-functional compounds can both be shown to suffer from inherent inadequacies. The upshot of this is a challenge not only to various theories of presupposition, but also to the conventional wisdom about the projection problem and what must be done to solve it. Traditionally, the task has been seen as that of specifying the presuppositions of compound sentences as a function of the presuppositions and “assertions” of their constituents. If a compound sentence S presupposes P, it has been assumed that a speaker who utters S indicates that he is presupposing P. If S does not presuppose P, the speaker does not indicate that he is presupposing P. What this approach overlooks is that there are not two, but three cases to consider: (i) Cases in which the speaker indicates that he is presupposing P (positive cases). 19

There has been considerable controversy about whether the inheritance conditions for conjunctions and conditionals should also be symmetric. For more on this, see Wilson (1975), Soames (1976), and Karttunen and Peters (1979). 20 But not necessarily a new approach to the presuppositions (implicatures) of other sentences. For present purposes, I am leaving it open whether the Karttunen-Peters treatment of the presuppositions (implicatures) of other sentences is correct. 21 See Peters (1979); Karttunen and Peters (1979), 44–46.

32

•

Essay One

(ii) Cases in which the speaker indicates that he is not presupposing P (negative cases). (iii) Cases in which the speaker indicates neither that he is, nor that he is not, presupposing P (neutral cases). The central difficulty with the standard approaches to the projection problem is that they ignore the difference between (ii) and (iii). I will argue that the projection problem for truth-functional compounds should be reformulated in terms of (i)–(iii). The problem is to predict the positive, negative, and neutral cases from a variety of factors, including: (a)

The presuppositions of the constituent clauses of the sentence uttered, (b) The proposition expressed by the sentence, (c) The background assumptions of the speaker and hearer at the time of the utterance, and (d) What the speaker conversationally implicates that he is, and is not, assuming. In effect, the projection problem must be solved, not for truth-functional compounds directly, but for speakers who utter such compounds. Approaching the problem in this way opens up new alternatives for explaining the phenomena that linguists and philosophers have been concerned with. In particular, it allows me to sketch an intuitively plausible view that avoids the difficulties that plague Karttunen and Peters, while accounting for much of the data that motivated their theory. I must add, however, that the program I sketch is incomplete in certain respects and faces obstacles of its own before it can be considered fully adequate. Nevertheless, if I am right about Karttunen and Peters, we need not only a new theory, but also a new way of looking at the presuppositions that arise from utterances of truth-functional compounds. The program I propose is an attempt to provide such a perspective.

Analogous Systems It is useful to begin by comparing the Karttunen-Peters theory to two analogous systems. One is the “filtering” theory of Karttunen (1974). The other is a theory of logical presupposition which involves three-valued truth tables for and, or, and if, then. In order to compare these systems, one needs to define a set of notions in terms of which the comparisons can be made.

A Projection Problem

•

33

Speaker Presupposition (19) A speaker X presupposes a proposition P (at time t) iff (at time t) a. X assumes P, and b. X assumes that his audience either already accepts P, or is prepared to treat it as uncontroversial. Sentential Presupposition (20) A sentence S of a language L presupposes a proposition P iff uttering S in L indicates that one is presupposing P.22 Logical Presupposition (21) A proposition P logically presupposes a proposition Q iff the truth of Q is a necessary condition for P to be either true or false. For purposes of comparison, the crucial notion is sentential presupposition. The inheritance conditions (12)–(14) are stated in terms of conventional implicature. However, Karttunen and Peters maintain that their theory of conventional implicature is meant to capture essentially the same phenomena as theories of “pragmatic presupposition” based on definitions like (19) and (20). Thus, one may take it that whenever Karttunen and Peters claim that S conventionally implicates P, they are also claiming that S presupposes P in the sense of definition (20).23 With this in mind, one can easily see that the inheritance conditions (12)–(14) are simply the Karttunen (1974) filtering conditions restated in a new terminology.24 The (1974) conditions were parts of an inductive definition of a relation, satisfies-the-presuppositions-of, that held between sets of propositions (called “contexts”) and sentences. Contexts were taken to be sets of background assumptions of speakers and hearers. A speaker who uttered a sentence S indicated he was assuming that the set of background assumptions common to him and his audience satisfied the presuppositions of S.25 22

Definitions (19) and (20) are roughly patterned after those given in Stalnaker (1972, 1973, 1974). 23 See Karttunen and Peters (1979, 13–15) for a discussion of the relationship between conventional implicature and speaker and sentential presupposition. 24 The only difference is that in Karttunen (1973) and (1974) the condition given for disjunction was asymmetric and was equivalent to (i): (i) (A or B)i = (Ai & (Ae ∨ Bi)) The possibility that the conditions for disjunctions (and conjunctions) might be symmetric was mentioned in footnote 5 of Karttunen (1974), where (24) was given. 25 More precisely, the speaker indicated that he assumed that the set of background assumptions, augmented with propositions that his audience was prepared to treat as uncontroversial (if they did not accept them already), satisfied-the-presuppositions-of S. Although

34

•

Essay One

The (1974) conditions were (22)–(24): (22)

A context X satisfies-the-presuppositions-of (A and B) iff a. X satisfies-the-presuppositions-of A, and b. X ∪ {A}26 satisfies-the-presuppositions-of B. (23) A context X satisfies-the-presuppositions-of (If A, then B) iff a. X satisfies-the-presuppositions-of A, and b. X ∪ {A} satisfies-the-presuppositions-of B. (24) A context X satisfies-the-presuppositions-of (A or B) iff a. X ∪ {−A} satisfies-the-presuppositions-of B, and b. X ∪ {−B} satisfies-the-presuppositions-of A. Whether or not a context satisfies-the-presuppositions-of a sentence depends on what it entails. Where S is any sentence, let Sp express a proposition whose entailment by a context is a necessary and sufficient condition for the context to satisfy-the-presuppositions-of S. Intuitively, Sp represents the full presuppositional requirements of S. Thus, where Ap and Bp represent the full presuppositional requirements of A and B, respectively, (22)–(24) are equivalent to (25)–(27). (25) A context X satisfies-the-presuppositions-of (A and B) iff a. X entails Ap, and b. X entails (A → Bp).27 (26) A context X satisfies-the-presuppositions-of (If A, then B) iff a. X entails Ap, and b. X entails (A → Bp). (27) A context S satisfies-the-presuppositions-of (A or B) iff a. X entails (−A → Bp) or equivalently (A ∨ Bp), and b. X entails (−B → Ap) or equivalently (B ∨ Ap). The difference between these conditions and the inheritance conditions (12)–(14) is essentially terminological. Instead of talking about the presuppositional requirement, Sp, of a sentence, Karttunen and Peters now talk of its conventional implicature, Si. Instead of saying that, ideally, sentences can be appropriately uttered only in contexts that entail their presuppositional requirements, they now say that “ideally, every conventional

Karttunen did not explicate the content of his (1974) system in precisely this way, I take it that this is an accurate reconstruction of his central idea. 26 In stating these conditions, I am not bothering to distinguish between sentences and propositions. Strictly speaking, we should have the union of X with the singleton of the proposition expressed by A. However, this is a matter of detail that need not concern us here. 27 The conventions governing this notation parallel those in footnote 10. Again, I am not worrying here about the difference between sentences and propositions.

A Projection Problem

•

35

implicature ought to belong to the common set of presumptions which the utterance of the sentence is intended to increment.”28 Thus, the new conditions are just the old ones in new dress.29 This can easily be seen by comparing the inheritance condition (13) with the filtering condition (26). According to (13), an indicative conditional, (28), conventionally implicates and, hence, presupposes (29): (28) If A, then B (29) Ai & (Ae → Bi) Since Si in the new system corresponds to Sp in the old system, and since A and Ae have the same truth conditions, (13) predicts that (28) presupposes (30): (30) Ap & (A → Bp) The filtering condition (26) makes the same prediction. According to it, a speaker uttering (28) indicates that he assumes that the set of background assumptions entails (30). Both Karttunen (1974) and Karttunen-Peters (1979) implicitly adopt an idealization in which speakers and hearers recognize the logical consequences of their beliefs (and realize this about each other). Thus, according to (26), a speaker uttering (28) indicates that he assumes that the set of background assumptions includes (30). This is tantamount to saying that the speaker accepts (30) and assumes that his audience does, too—that is, the speaker presupposes (30). Consequently, the inheritance condition (13) and the filtering condition (26) both predict that (28) presupposes (30) (in the sense of definition (20)). Analogous predictions can also be made on the basis of certain theories of logical presupposition. Consider, for example, a theory based on the three-valued truth tables in (31).30 (31) a. and T A F * 28

b. T T F *

B F F F *

* * F *

if, then T A F *

c. T T T *

B F F T *

* * T *

A

or T F *

B T T T T

F T F *

* T * *

Karttunen and Peters (1979), 14. Karttunen and Peters point this out (1979, 35–36). The reader may note that (12)–(14) use ‘Ae’ and ‘Be’ in places where (25)–(27) just use ‘A’ and ‘B’. This difference is immaterial. Since the conditions are concerned with entailment relations which are defined in terms of truth conditions, no substantial difference results from this change in terminology. (The terminology of (12)–(14) is, of course, clearer and more explicit in this respect.) 30 Karttunen and Peters (1979, 44) present similar truth tables in their discussion of Hausser (1976). The analogy between the inheritance conditions (12)–(14) and certain systems of logical presupposition is noted on pp. 44–47. 29

36

•

Essay One

Where S is any sentence, let S p express a proposition the truth of which is a necessary and sufficient condition for the proposition expressed by S to be either true or false. The proposition expressed by Sp will then be logically equivalent to the set of logical presuppositions of the proposition expressed by S. (In effect, Sp will be a “maximal” logical presupposition.) This feature of Sp allows one to use it to formulate the “inheritance” conditions that are implicit in the truth tables of (31). For example, according to (31c) the proposition expressed by a disjunction, (32), logically presupposes the proposition expressed by (33): (32) (33)

A or B − (−A & − Bp) & − (−B & − Ap)31

Since (33) is logically equivalent to (34), (34) (A ∨ Bp) & (B ∨ Ap)

31

In (33), ‘−’ is “external” negation. (i) S −S T F F T * T The first conjunct of (33) tells us that we are not in either of the circled positions in (ii): (ii)

A

or T F *

T T T T

B F T F *

* T * *

The second conjunct of (33) tells us that we are not in either of the circled positions in (iii): (iii) B or T F * T T T T A F T F * * T * * Consequently, (33) as a whole tells us that the disjunction (32) has a truth-value. In other words, (33) expresses a “maximal” logical presupposition of (32). Note that the truth tables in (31) interpret the English connectives and, or, and if, then. They are not intended to interpret ‘&’, ‘∨’, and ‘→’. In what follows I will have several occasions to compare the Karttunen-Peters inheritance algorithms with algorithms implicit in three-valued truth tables for the English connectives. Such algorithms will be expressed in the standard formalism which includes ‘&’, ‘∨’, ‘→’, and ‘−’. In all such cases, these formal connectives can safely be read as resulting in T or F for any assignment of T, F, * to A and B (with * being treated exactly like F).

A Projection Problem

•

37

(31c) predicts that (32) logically presupposes (34). This is analogous to the output of the Karttunen-Peters inheritance condition for disjunctions. There is, of course, an important difference between the KarttunenPeters system based on (12)–(14) and the system of logical presupposition based on (31). The former deals with the conventional implicatures of sentences, whereas the latter is concerned with the logical presuppositions of propositions. Nevertheless, the predictions of the two systems can be made comparable. I have shown how predictions about conventional implicatures can be used to generate predictions about sentential presuppositions. By making certain additional assumptions, one can use logical presuppositions in the same way. The assumptions in question are given in (35). (35)

a. A speaker who asserts a proposition P (at time t) indicates that he assumes that P has a truth-value in all possible worlds compatible with the common background assumptions of the conversation (at t). b. A speaker who asserts a proposition P (at t) indicates that he assumes that the common background assumptions of the conversation (at t) entail the logical presuppositions of P. c. A speaker who asserts P (at t) indicates that he presupposes (in the sense of (19)) the propositions that are logically presupposed by P.

In a system that adopts an idealization in which (A) and (B) are true, (A) Speakers and hearers recognize the logical consequences of their beliefs (and realize this about each other); and (B) Speakers and hearers recognize the logical presuppositions of the propositions expressed by the sentences of their language, (35a), (35b), and (35c) come to essentially the same thing.32 Given such an idealization, a theory of logical presupposition incorporating (31) and (35) makes predictions about sentential presupposition that are analogous to those made by Karttunen and Peters. In particular, both systems predict that disjunctions presuppose propositions of the form (34). Similar results hold for conjunctions and conditionals.

32 The idea of connecting logical presupposition to speaker and sentence presupposition via (35a–c) is implicit in Stalnaker (1972, 1973, 1974).

38

•

Essay One

Negation and Symmetric Inheritance Conditions So far, I have considered three theories: (i) The Karttunen-Peters theory of conventional implicature (ii) The “pragmatic” theory of Karttunen (1974) (iii) A theory of logical presupposition based on (31) and (35). These theories make analogous predictions about the sentential presuppositions of truth-functional compounds. However, in the case of (i) and (iii) it is important to note that the predictions are only analogous, not identical. For example, (i) and (iii) predict that disjunctions presuppose propositions of the form (34) (repeated here): (34) (Ae ∨ Bi) & (Be ∨ Ai) (i) (A ∨ Bp) & (B ∨ Ap) (iii) However, the elements Ae, Ai, Be, and Bi have different logical properties in the Karttunen-Peters system than the elements A, Ap, B, and Bp have in theories of logical presupposition. This difference is illustrated by the following formulas: (36) a. (A ∨ Bp) & (B ∨ Ap) b. (A ∨ Bp) & (B ∨ Ap) & (Ap ∨ Bp) (37) a. (Ae ∨ Bi) & (Be ∨ Ai) b. (Ae ∨ Bi) & (Be ∨ Ai) & (Ai ∨ Bi) (36a) and (36b) are logically equivalent in systems of logical presupposition. In such systems, A entails Ap and B entails Bp. Thus, A, Bp, B, and Ap each entail the final conjunct of (36b). As a result, this conjunct is redundant and does not add anything to what is already stated in (36a). The situation is different with Karttunen and Peters. Since Ae and Be do not always entail Ai and Bi, (37a) and (37b) are not equivalent. Thus, a question arises for Karttunen and Peters that does not arise in the corresponding theory of logical presupposition. Should the algorithm for determining the presuppositions (conventional implicatures) of disjunctions result in (37a), as Karttunen and Peters maintain, or should it result in (37b)? The question at issue involves the fate of conventional implicatures (presuppositions) that are common to both disjuncts. The final conjunct of (37b) represents precisely these implicatures. This follows from two things: (i) In the Karttunen-Peters system, the set of conventional implicatures of a sentence is closed under entailment (that is, any proposition entailed by a conventional implicature of S is itself a conventional implicature of S).

A Projection Problem

•

39

(ii) The set of entailments common to two propositions (e.g., Ai and Bi) = the set of entailments of their disjunction (e.g., (Ai ∨ Bi)). Thus, (37b) guarantees that all conventional implicatures that are common to both disjuncts will be inherited by the entire disjunction. Since (37a) does not guarantee this, we can decide between the two conditions by finding common implicatures that (37a) does not preserve. Examples of this kind are provided by disjunctions of type (38). (38) a. Ai and Bi each entail some proposition C. (C is conventionally implicated by both disjuncts.) b. Neither Ae nor Be entails C. (The propositions expressed by the disjuncts do not entail the common conventional implicature.) c. − Ae and − Be each entail C; or, equivalently, − C entails Ae and Be. (The “external” negation of the common implicature entails the propositions expressed by the disjuncts.)33 In order to grasp the significance of (38), it is useful to divide (38a) into two cases.34 Case 1 Ai = Bi = C (The common implicature exhausts the implicatures of each disjunct.) Case 2 Ai = (P & C); Bi = (Q & C) (The disjuncts share an implicature, but this implicature does not exhaust the implicatures of each disjunct.) In Case 1, (37a) is logically equivalent to (39), (39) (−Ae → C) & (−Be → C) 33

Strictly speaking, (38a) and (38b) are sufficient to characterize examples in which (37a) predicts that the common implicature C is not inherited by the entire disjunction. Where these conditions are satisfied, but (38c) is not, (37a) predicts that the disjunction implicates (i) or (ii) or both (but not C itself). (i) Ae ∨ C (ii) Be ∨ C However, the availability of arguments like the one given for example (7) makes these cases hard to distinguish from those in which the theory predicts that the disjunction conventionally implicates C. This difficulty is avoided by adding (38c). If a disjunction satisfies (38a–c), then (37a) predicts that the common implicature C is completely eliminated. (See below.) 34 The ‘=’ sign in Cases 1 and 2 should be read “is identical with or logically equivalent to.” In Case 2 it is assumed that either P is not logically valid or Q is not logically valid or both are not logically valid.

40

•

Essay One

which, by (38c), is logically valid (and hence trivial). In Case 2, (37a) is logically equivalent to (40), which, in turn, is equivalent to (41): (40) (−Ae → (Q & C)) & (−B e → (P & C)) (41) (−Ae → Q) & (−Ae → C) & (−Be → P) & (−Be → C) Since (38c) guarantees the validity of the second and fourth conjuncts, they can be dropped, leaving (42) or, what amounts to the same thing, (43): (42) (−Ae → Q) & (−Be → P) (43) (Ae ∨ Q) & (Be ∨ P) In both cases, the implicature C has been completely eliminated. Thus, an algorithm incorporating (37a) predicts that a disjunction of type (38) does not inherit the common implicature C. An algorithm incorporating (37b) predicts that C is inherited. The latter prediction is the one wanted. In order to see this, one must consider the way in which specific examples are treated by Karttunen and Peters. According to them, ordinary examples of negation like (44) and (45) are instances of classical, two-valued, “external” negation. (44) The chairman of the department isn’t a skeptic about the external world. (45) The chairman of the department isn’t very sensible. Where ‘−’ is the familiar two-valued operator and α and β represent the truth conditions of (46) and (47), (46) The chairman of the department is a skeptic about the external world. (47) The chairman of the department is very sensible. −α ⎤ and ⎡−β⎤ represent the truth conditions of (44) and (45). Thus, Karttunen and Peters would hold that the propositions expressed by (44) and (45) do not entail the proposition expressed by (48): ⎡

(48)

The department has a chairman.

Nevertheless, a speaker who assertively utters (44) or (45) commits himself to the proposition expressed by (48). According to Karttunen and Peters, the reason for this is that (44) and (45) conventionally implicate (presuppose) that proposition. This implicature arises from two features of their system. First, (46) and (47) are taken not only to express propositions that entail the proposition expressed by (48); they are also taken to conventionally implicate (presuppose) it. Second, ordinary negation is treated as a “hole” for conventional implicatures. Thus, negative

A Projection Problem

•

41

sentences like (44) and (45) are said to share the conventional implicatures (presuppositions) of their positive counterparts. Now consider (49): (49) Either the chairman of the department isn’t a skeptic about the external world or the chairman of the department isn’t very sensible. Ae : − (The chairman of the department is a skeptic about the external world) Be: − (The chairman of the department is very sensible) Ai = Bi = C: The department has a chairman. Intuitively, a speaker who assertively utters (49) implicates (presupposes) that the department has a chairman. However, Karttunen and Peters fail to account for this. According to their inheritance condition (14), (49) conventionally implicates (37a), (37a) (Ae ∨ Bi) & (Be ∨ Ai) which is equivalent to (39): (39) (−Ae → C) & (−Be → C) Since (39) is logically valid, Karttunen and Peters incorrectly predict that (49) does not implicate that the department has a chairman. Any disjunction of type (38) can be used to make the same point. Further examples of this type in the Karttunen-Peters system are: (50) Either the king of France isn’t wise, or the French generals aren’t under the king’s control. C: There is a king of France.35 35 Could the implicature C plausibly be regarded as conversational rather than conventional? A speaker who asserts a disjunction conversationally implicates (via the maxim of Quantity) that he has non-truth-functional grounds for his assertion. Note, however, that given the Karttunen-Peters account of the truth conditions of (50), a speaker who believed (i)

(i)

The French generals are following an unwise policy

could have non-truth-functional grounds for the truth of the proposition expressed by (50), even if he were not sure whether France had a king. Thus, if C is conversationally implicated, more than the maxim of Quantity (or Quality) is involved. It is conceivable that the implicature could be reconstructed as conversational by using Grice’s other maxims, or entirely new ones. However, it is not obvious that this could be done, and, moreover, it is unlikely that such a strategy could be used to save (14). In order to maintain the system for treating putative presuppositions as conventional implicatures, one must countenance some negations, disjunctions, etc., whose presuppositions are not conversational, but rather are purely conventional. If there are such cases, there will be examples among them of type (38).

42

•

Essay One

(51) Either it wasn’t Harry who lost the money, or he doesn’t realize that he lost it. C: Someone lost the money.36 (52) Either it wasn’t Martha who tapped Bill’s phone, or he will never discover that she is the one who did it. C: Someone tapped Bill’s phone.37 (53) Either every radical in the house will escape, or every one of them will die C: There is at least one radical in the house.38 (54) Either all of John’s children are asleep, or none of his children are asleep. C: John has at least one child.39 In all of these cases the implicature C is incorrectly eliminated by inheritance condition (14). In order to avoid this difficulty, one must reject (14) in favor of the new inheritance condition (55): (55) (A or B)i = (Ae ∨ Bi) & (Be ∨ Ai) & (Ai ∨ Bi) Modifying the system in this way allows one to capture the fact that conventional implicatures (presuppositions) common to both disjuncts are always inherited by the entire disjunction.40 36

C is a conventional implicature of the right disjunct in virtue of the fact that it is entailed by (i) (i)

Harry lost the money

which, according to Karttunen and Peters, is conventionally implicated by the right disjunct. 37 In Karttunen and Peters’s system, the proposition expressed by (52) is logically equivalent to the proposition expressed by its right disjunct. Why, then, would the disjunction ever be used? To avoid the implicature that Martha tapped Bill’s phone, which is carried by the right disjunct alone. Note, however, that (52) retains the implicature C. If one wanted to avoid even this implicature, while making the same assertion as would be made with (52), one could use (i): (i) 38

Either Martha didn’t tap Bill’s phone or he will never discover that she did.

According to Karttunen and Peters, (i) is (vacuously) true if there are no Ns.

(i)

Every N is P.

Sentences of this form are said to conventionally implicate (ii): (ii)

There is at least one N.

I assume that in the right disjunct of (53), every one of them has the force of ‘every radical in the house’. 39 Karttunen and Peters do not give explicit treatments of all and none. I am here assuming that they would handle them in a way that parallels their treatment of every. 40 The question of whether other constructions require symmetric inheritance conditions is controversial. However, if they do, analogous arguments can be constructed to show that they must include (i): (i) (Ai ∨ Bi)

A Projection Problem

•

43

Conflicting Presuppositions The Problem I now turn to a more serious problem,41 one which affects all of the theories considered so far: the Karttunen-Peters theory (with either (14) or (55)), the Karttunen (1974) theory, and the theory of logical presupposition based on (31) and (35). The problem involves disjunctions whose disjuncts bear conflicting presuppositions. Where Ai and Bi are logically contradictory, the putative presupposition (conventional implicature) (56) entails (57): (56) (Ae ∨ Bi) & (Be ∨ Ai) (57) (Ae ∨ Be) However, (57) is the proposition expressed by (58): (58)

A or B

Thus, when Ai and Bi are contradictory, all of the theories considered so far predict that the disjunction presupposes the proposition it expresses

For example, although Karttunen and Peters are not convinced by putative counterexamples to their asymmetric condition for indicative conditionals, they cite (ii) (ii) (If A, then B)i = [Ae → Bi] & [Be→Ai] as the appropriate condition if a symmetric principle proves to be required. However, this will not do. Counterexamples to (ii) can be constructed by using conditionals in which (iii) and (iv) obtain: (iii) Both A and B conventionally implicate some proposition C. (iv) Both express propositions which entail C. Conditionals satisfying (iii) and (iv) can be formed from (50)–(52) by making the positive counterpart of the left disjunct the antecedent, and making the positive counterpart of the right disjunct the consequent. Examples corresponding to (49), (53), and (54) can easily be constructed by making appropriate changes in content. Thus, if the condition for indicative conditionals proves to be symmetric, it should be (v): (v) (If A, then B)i = [Ae→Bi] & [Be→Ai] & [Ai ∨ Bi] For further discussion of this issue plus an argument that conjunctions with but require symmetric “filtering” within the Karttunen (1974) framework, see Soames (1976, 352–61). For an argument that (24) (and, by extension, (14)) cannot be saved by changing the semantics to include “internal” as well as “external” negation, see Soames (1976, 344–52). 41 The problem noted in this section was first developed in Soames (1976, 367–72). See Wilson (1975) for a discussion of a different class of cases involving conflicting presuppositions and their relevance to certain theories of presupposition not considered here.

44

•

Essay One

(asserts).42 When one looks at English, however, one finds that this is not so.43 For example, consider (59): (59) Either the loan company repossessed Bill’s only car, or it repossessed his second car. Ai: Bill had exactly one car. Bi: Bill had at least two cars. Since Ai and Bi are incompatible, Karttunen and Peters predict that a speaker who assertively utters (59) indicates that he presupposes the proposition he asserts.44

42 In order to reconstruct the argument for the theory of logical presupposition discussed in the section “Analogous Systems” above, substitute A, Ap, B, and Bp for Ae, Ai, Be, and Bi, respectively. Note that the counterpart to (57) will then be (i):

(i) A ∨ B Because of the three-valued interpretation of or (and the two-valued construal of ∨), (i) will not represent the very same proposition as (58). However, the proposition expressed by (i) will entail the proposition expressed by (58) (see truth table (31c)). Since the theory of logical presupposition based on (31) and (35) characterizes the set of presuppositions of a sentence as closed under entailment, it predicts that (58) presupposes the proposition it expresses (asserts). 43 Note that there is nothing incoherent about the claim that a sentence presupposes the proposition it expresses. (i) (ii)

The king of France is a king. There is a (unique) king of France.

The propositions expressed by (i) and (ii) entail one another. Since (i) presupposes the proposition expressed by (ii), and since this proposition entails the proposition expressed by (i), it is plausible to hold that (i) presupposes the proposition it expresses. The present criticism of the Karttunen-Peters theory, and of the theory of logical presupposition discussed in the section “Analogous Systems,” is not that they predict that there are sentences which presuppose what they express. Rather, the criticism is that they wrongly characterize many disjunctions as being of this type, when, in fact, they are not. 44 An inheritance condition incorporating (56) predicts that (59) presupposes (conventionally implicates) (i) and (ii): (i) The loan company repossessed Bill’s only car ∨ Bill had at least two cars. (ii) The loan company repossessed Bill’s second car ∨ Bill had exactly one car. The conjunction of (i) and (ii) is logically equivalent to the proposition expressed by (59). In general, if Ai and Bi are incompatible, and if A and B entail Ai and Bi (as in (59)), then (iii) and (iv) are logically equivalent (i.e., mutually entailing). (iii) (Ae ∨ Bi) & (Be ∨ Ai) (iv) A or B

A Projection Problem

•

45

This prediction is false. To say that a speaker indicates that he presupposes P is to say that he indicates that P is uncontroversial, that he takes it for granted, and that he assumes his audience does, too. However, someone who assertively utters (59) does not do any of these things. He does, of course, indicate that he believes the proposition he asserts. However, he does not indicate that it is uncontroversial, that he takes it for granted, or that he assumes his audience already accepts it. Thus, (59) does not presuppose (conventionally implicate) the proposition it expresses.45 This can also be shown by embedding (59) in larger contexts. (60) If the loan company either repossessed Bill’s only car or repossessed his second car, then he will sue the company.

{

}

(61) It is possible that the loan company either repossessed Bill’s may be only car or repossessed his second car. (62) No, it didn’t. [Said in response to (59)] (63) Did it really? [Said in response to (59)] According to Karttunen (1974) and Karttunen and Peters (1979), the presuppositions (conventional implicatures) of (59) are inherited by (60)–(63). Thus, they predict that a speaker who utters one of these sentences implicates that the proposition expressed by (59) is true. However, this is incorrect. A speaker who utters (60)–(63) conversationally 45 Karttunen (1974) cites cases in which the main point of a speaker’s remark is to communicate what he is presupposing.

(i) We regret that children cannot accompany their parents to commencement exercises. If regret is factive, (i) conventionally implicates (presupposes) the truth of the complement. Nevertheless, (i) might be uttered by someone in authority for the express purpose of informing his audience that the complement is true. Karttunen correctly observes that in cases like this, where the shared background assumptions of the conversation do not already include all of the conventional implicatures (presuppositions) of our utterance, we “communicate indirectly, convey matters without discussing them.” He says (1974, 191–92), When we hear a sentence such as [(i)], we recognize that it increments contexts which entail that children are not permitted at commencement exercises. These are the only contexts that satisfy the presuppositions of [(i)]. So if we have not realized already that we are supposed to be in that kind of context, the sentence lets us know indirectly. Perhaps the whole point of uttering [(i)] was to make us conclude this for ourselves so that we would not have to be told directly. The situation with (59) is different. Here we are told something directly: that the loan company repossessed a car of Bill’s. Thus, the way (59) functions in communication cannot be understood in terms of Karttunen’s discussion of (i). Consequently, it should not be analyzed as conventionally implicating (presupposing) the proposition it expresses.

46

•

Essay One

implicates that he is not assuming the truth of (59).46 Consequently, the Karttunen-Peters theory is incorrect and cannot be accepted as it stands. It should be noted that (59) is not an isolated counterexample. For example, consider (64) and (65): (64) Either the lawyer located Susan’s only heir or he found the oldest of her (several) heirs. Ai: Susan had exactly one heir. Bi: Susan had more than one heir. (65) Either Bill met the king of Slobovia or he met the president of Slobovia. Ai: Slobovia had a king. Bi: Slobovia had a president. (64) is just like (59). (65) is similar, except the conflicting implicatures are not logically incompatible. However, one can easily imagine (65) being uttered in contexts in which (66) is taken for granted. (66)

Countries with kings don’t have presidents.

In such contexts, Ai and Bi conflict in that their conjunction is logically incompatible with the common background assumptions of the conversation. Inheritance conditions for disjunctions which incorporate (56) (56) (Ae ∨ Bi) & (Be ∨ Ai) predict that (65) presupposes (conventionally implicates) (67) and (68): (67) Bill met the king of Slobovia ∨ Slobovia had a president. (68) Bill met the president of Slobovia ∨ Slobovia had a king. Although the conjunction of (67) and (68) does not entail the proposition expressed by (65), the conjunction plus (66) does. Since (66) may be a shared background assumption, (65) gives rise to essentially the same problems as (59) and (64) do.47 46 A speaker who uttered (60) or (61) while assuming the truth of the proposition expressed by (59) would violate the maxim of Quantity. Such a speaker would be guilty of making a weaker statement when he could have made a stronger one—namely, the conjunction of the antecedent and consequent of (60), or, in the case of (61), (59) itself. A speaker who uttered (62) while assuming the truth of the proposition expressed by (59) would be guilty of violating the maxim of Quality—saying something which he believed to be false. Presumably (63) involves an extension of the maxim of Quality to rule out asking a question that one already thinks one knows the answer to. 47 Ironically, (65) is cited in Karttunen (1974) in support of his theory. Although Karttunen considered utterances of (65) in contexts containing (66), he did not point out that in such contexts, (65) is essentially equivalent to the conjunction of its putative presuppositions. However, he apparently did think that (65) could be used, in the manner of (i) in footnote 45, to convey the information contained in its presuppositions. An argument that the communicative

A Projection Problem

•

47

Examples like this allow one to expand the characterization of the present class of counterexamples. This class consists of utterances of disjunctions in which (i) A presupposes (conventionally implicates) a proposition PA; (ii) B presupposes (conventionally implicates) a proposition PB; (iii) PA, together with shared background assumptions, is incompatible with PB. Inheritance algorithms incorporating (56) predict that in all such cases the speaker indicates that he presupposes the proposition he asserts. Since this prediction is false, the Karttunen-Peters theory, the theory of Karttunen (1974), and the theory of logical presupposition based on (31) and (35) are all incorrect.48 Complicating the Inheritance Conditions One way of trying to handle this problem is to modify the inheritance algorithm for disjunctions. However, it is not easy to see how to do this. The problem is particularly severe for the theory of logical presupposition based on (31) and (35). Since the inheritance algorithm implicit in this theory arises from (31c), (31) c.

A

or T F *

T T T T

B F T F *

* T * *

role of (65) cannot be understood in this way can be constructed which precisely parallels the argument in footnote 45. A further argument that (65) does not conventionally implicate (presuppose) (67)–(68) can be constructed by embedding (65) in contexts analogous to (60)–(63). 48 The problem discussed above is not limited to disjunctions. (i) If the loan company didn’t repossess Bill’s only car, it repossessed his second car. (ii) If the lawyer didn’t locate Susan’s only heir, he located the oldest of her (several) heirs. (iii) If Bill didn’t meet the king of Slobovia, he met the president of Slobovia. If the inheritance condition for indicative conditionals is symmetric (see footnote 40), then (i)–(iii) pose the same problem as (59), (64), and (65). If the inheritance condition is asymmetric, as in (13), (23), or truth table (31b), then the theory incorrectly predicts that (i)–(iii) presuppose (conventionally implicate) (i′)–(iii′): (i′) Bill had exactly one car. (ii′) Susan had exactly one heir. (iii′) Slobovia had a king. Although I am using disjunctions to illustrate the problem posed by conflicting presuppositions, it should not be forgotten that this problem applies to conditionals as well.

48

•

Essay One

changing the algorithm would mean modifying (31c). However, there is no completely satisfactory way of doing this. If one rejects (31c) in favor of (31d), (31) d.

A

or T F *

T T T *

B F T F *

* * * *

one incorrectly predicts that the proposition expressed by (59) logically presupposes a contradiction (the conjunction of Ap and Bp) and that the proposition expressed by (69), (69) Either there is no king of France or the king of France is in hiding, logically presupposes that there is a king of France. If one rejects (31c) in favor of (31e), (31) e. B or T F * T T T T A F T F F * T F * one captures only presuppositions common to both disjuncts. Thus, one fails to account for cases in which uttering a disjunction indicates that the speaker is presupposing something that is presupposed by a single disjunct. (70) Either Boris will support the party or the party will see to it that he is re-educated, too. Bp: Someone other than Boris has been re-educated. (71) Either it’s Bill who has the money or we are out of luck. Ap: Someone has the money. (72) Either all of John’s children are bald or baldness isn’t hereditary. Ap: John has children.49 49

A theory of logical presupposition incorporating (31c) and (35) treats the presuppositions of a speaker who utters (70)–(72) in exactly the same way that Karttunen and Peters do. (70) is claimed to presuppose (i): (i) A ∨ Bp (71) and (72) are claimed to presuppose (ii):

A Projection Problem

•

49

Other modifications of the truth table for or are equally inadequate. In every case one either predicts that the speaker indicates that he presupposes P, when in fact he does not; or one fails to predict that he indicates that he presupposes P, when in fact he does. As a result, no theory of logical presupposition in which (i) or is truth functional and (ii) A speaker who asserts P indicates that he presupposes whatever P logically presupposes can account for the range of data that Karttunen and Peters are concerned with.50

(ii) B ∨ Ap Arguments of the form given in connection with example (7) are then used to predict the speaker’s commitment to Bp in the case of (70) and Ap in the case of (71) and (72). The present point is that even if one accepts these arguments in the case of (31c), they cannot be employed with (31e). According to (31e), (70)–(72) presuppose not (i) or (ii), but (iii): (iii) Ap ∨ Bp In the case of (70), it is easy to show that the speaker must believe Ap (that is, that Boris and the party exist). Since this explains his commitment to (iii), the theory fails to explain why he would normally be taken to be committed to Bp as well. A similar point holds for (71) and (72). Thus, opting for (31e) renders the theory incapable of accounting for a large range of cases—those in which a presupposition of a disjunct is not “filtered.” 50 Hausser (1976) claims to account for the full range of this data with the following three-valued truth tables: A T F *

not A F T *

A

and T F *

T T F *

B F F F F

* * F *

or T A F *

T T T T

B F T F *

* T * *

if, then T A F *

T T T T

B F F T *

* * T *

These tables, together with (35) (i.e., (ii) above), make virtually the same predictions about speaker and sentential presuppositions as Karttunen and Peters do. (See Karttunen and Peters (1979, 44–46) for further discussion.) Since Hausser’s table for or = (31c), disjunctions with conflicting presuppositions pose the same problem for him as they do for Karttunen and Peters. The argument given above shows that Hausser cannot solve this problem by picking another truth table for or. Nor will it do to simply drop (35) (i.e., (ii)) without putting anything in its place. If this were done, there would be no way of using the truth tables to explain what a speaker indicates that he presupposes on the basis of the sentence he utters. It is, of course, conceivable that three-valued truth tables might have a justification independent of claims about speaker or sentential presupposition, as defined in (19) and (20). Nothing in my argument either assumes or denies the possibility of such a justification.

50

•

Essay One

Each three-valued truth table for or corresponds to a possible inheritance condition for disjunctions. In claiming that no such truth table is adequate, I am claiming that none of the corresponding inheritance conditions is either. However, there are possible inheritance conditions that do not correspond to any three-valued truth table for or. One might wonder whether Karttunen and Peters could save their theory by substituting one of these conditions for (55). The problem is to avoid false predictions about examples with conflicting implicatures. Thus, the most promising strategy would seem to involve simply adding a clause to cover the problematic cases, while leaving everything else as it was. The result of this strategy is (73). (73) (A or B)i = the conjunction of a. (Ai ∨ Bi) and b. −(Ai & Bi) ∨ ((Ae ∨ Bi) & (Be ∨ Ai)) If Ai and Bi conflict (either because they are logically incompatible or because their conjunction is logically incompatible with a shared background assumption), the implicature represented by (73b) will be trivial (that is, it will either be logically valid or entailed by shared background assumptions). Thus, (73) correctly predicts that a speaker uttering (59), (64), or (65) conventionally implicates the proposition expressed by (Ai ∨ Bi), but not those expressed by Ai, Bi, (59), (64), or (65). It might also be argued, on behalf of Karttunen and Peters, that where Ai and Bi do not conflict, the right disjunct of (73b) is the operative one, in which case nonproblematic disjunctions will be treated as before.51 Unfortunately, there are problematic cases that cannot be handled in this way. Such cases involve disjunctions in which (i) The disjuncts have conflicting conventional implicatures but (ii) One disjunct has a further “unfiltered” conventional implicature that the other does not. If a disjunction satisfies (i), (73) predicts that the only (nontrivial) conventional implicatures it inherits are those shared by both disjuncts. Thus, it predicts that the implicature mentioned in (ii) will not give rise to a (nontrivial) conventional implicature on the part of the entire disjunction.

Rather, my claim is that even if we find some set of three-valued truth tables which are acceptable on other grounds, they will not provide an adequate basis for explaining the speaker and sentential presuppositions that Karttunen, Peters, Stalnaker, and others have been concerned with. 51 If such an argument can be given, presumably it will parallel the one given above in connection with example (7).

A Projection Problem

•

51

This prediction is incorrect. A speaker who assertively utters (74) implicates the proposition expressed by (75): (74) Either the lawyer located Susan’s only heir or his assistant located the oldest of her several heirs. (75) The lawyer has an assistant. This implicature is not predicted by (73). If a speaker’s commitment to (75) is the result of a conventional implicature, then (74) is a counterexample to (73).52 Of course, it might be claimed that the proposition expressed by (75) does not result from a conventional implicature (presupposition) of (74), but rather is a conversational implicature of it. The maxim of Quantity requires a speaker to have non-truth-functional grounds for his assertion. Since it is not easy to imagine a speaker having such grounds without assuming (75), the claim that this is a conversational implicature is not implausible. However, this is not enough. To say that an implicature can be explained conversationally is not to deny it might be explained, at least in part, conventionally. There is no reason why both should not be the case.53 Moreover, there is evidence that (75) does result from a conventional implicature of (74); it behaves like a conventional implicature when (74) is embedded in a larger construction which is a “hole” for such implicatures. (76) If the lawyer located Susan’s only heir or his assistant located the oldest of her several heirs, then the case will be settled soon.

52 In the Karttunen-Peters system a speaker’s commitment to the proposition expressed by (75) might be the result of a conventional implicature even though (74) docs not conventionally implicate that proposition. (See the discussion of example (7) above.) The point about (73) is that it does not allow one to use conventional implicature (presupposition) at all in explaining why a speaker uttering (74) would typically be taken to suggest that (75) is true. Thus, (74) is an apparent counterexample to (73). 53 For example, exactly the same argument that was made to support the claim that (74) conversationally implicates (75) could be made to support the claim that (i) conversationally implicates (ii):

(i) Either baldness isn’t hereditary, or John’s son is bald. (ii) John has a son. Nevertheless, Karttunen and Peters claim that (i) conventionally implicates (iii) (iii)

Baldness isn’t hereditary ∨ John has a son

and that (iii) can be used to help explain why a speaker who utters (i) commits himself to (ii). Thus, Karttunen and Peters must admit that there are implicatures that can be explained either purely conversationally or partially conventionally.

52

•

Essay One

As with (74), a speaker who utters (76) implicates that (75) is true. Since Karttunen and Peters standardly take the preservation of an implicature in an environment like this as demonstrating conventionality, they must recognize that (75) results from a conventional implicature of (74) and (76). Consequently, (73), which fails to predict this, cannot be adequate within their system. Finally, it should be noted that (74) is not an isolated counterexample. There are additional, paradigmatically conventional implicatures that can be used to make the same point. One class of such implicatures involves the particle too. According to Karttunen and Peters, a simple sentence (77)

S too

with a focused NP, α, conventionally implicates the proposition expressed by (78): (78) ∃x(x ≠ α & S x/α)54 Thus, (79) implicates (80), whereas (81) does not. (79) (80) (81)

Susan insulted the king too. [Italics mark focused NP] Susan insulted someone other than the king. Susan insulted the king.

Nevertheless, (79) and (81) are said to express the same proposition and, hence, to have the same truth conditions. The only semantic difference between them lies in what they conventionally implicate. This account is carried over to compound sentences containing (79) and (81) as parts. (82) Susan met the king and she insulted him too. (83) Susan met the king and she insulted him. Although (82) and (83) are said to express the same proposition, only the former implicates (80).55 Since the two sentences express the same 54 Constituents other than NPs can be focused, in which case too gives rise to different implicatures. However, for our purposes, cases involving focused NPs are sufficient. For a more complete account, see Karttunen and Peters (1979, 23–33) plus the references cited there. Terminological note: In (78), ‘S x/α’ is used to indicate the result of replacing occurrences of α in S with ‘x’. 55 More precisely, a speaker who utters (82) implicates (80). According to Karttunen and Peters, this implicature results from the fact that (82) conventionally implicates (i)

(i)

Susan met the king → Susan insulted someone other than the king

plus an account of why a person implicating (i) will be taken to assume the consequent of (i) (i.e., (80)). The crucial point for the present discussion is that the conventional implicature

A Projection Problem

•

53

proposition, the implicature cannot arise from Grice’s maxims of Quantity or Quality. Nor is there any obvious way of making it come from the other maxims. Thus, it is plausible to claim that it is irreducibly conventional. Let us assume, for the moment, that this is correct. This assumption can be used to test (73). We can do this by finding examples in which (i) The disjuncts have conflicting conventional implicatures; (ii) One disjunct has an additional “unfiltered” implicature that the other disjunct does not; (iii) This “unfiltered” implicature is, by assumption, irreducibly conventional (since it arises from a disjunct analogous to (82)). In a case like this, (73) will incorrectly predict that the “unfiltered” implicature (presupposition) does not give rise to an implicature of the entire disjunction. (84) is such a disjunction.56 (84) Either Susan met the king of Slobovia and insulted him, too, or she met the president of Slobovia and lauded his democratic principles. According to Karttunen and Peters, (84) expresses the same proposition as (85): (85) Either Susan met the king of Slobovia and insulted him, or she met the president of Slobovia and lauded his democratic principles. Nevertheless, a speaker uttering (84) implicates that Susan has insulted other people, whereas a speaker uttering (85) does not. Since (84) and (85) express the same proposition, this implicature does not arise from the maxims of Quantity and Quality, and, by our earlier assumption, is not conversational at all. Thus, it should be captured by whatever principle standardly accounts for the conventional implicatures (presuppositions) of disjunctions. Since it is not captured by (73), (73) is incorrect.

(i) is needed to explain the speaker’s implicature. According to Karttunen and Peters, no purely conversational account will do. 56 To ensure that (84) satisfies (i), imagine it being uttered in a context in which it is assumed that countries with kings do not have presidents. A speaker uttering (84) in such a context assumes that Slobovia has a king or a president, but he is not sure which. To make (84) easier to process, imagine that Susan is known as a fervent democrat, and monarch-hater.

54

•

Essay One

Defeating Implicatures The results of the last section suggest that the Karttunen-Peters inheritance conditions cannot be saved by modifying them to take explicit account of cases in which the implicatures (presuppositions) of constituent sentences conflict. What prevents this strategy from working is the strictly compositional nature of their system. For example, in order to determine the conventional implicatures of (86), (86)

(P and Q) or (R and S)

one must first determine the implicatures of the disjuncts. These implicatures, Ai and Bi, are then used in computing the implicatures of (86). The problem is that Ai and Bi may contain both conflicting and nonconflicting elements. Intuitively, one wants the conflicting elements to cancel out, leaving the nonconflicting elements intact. However, the Karttunen-Peters system does not allow this kind of operation within Ai and Bi. Thus, there seems to be no way of complicating their inheritance conditions, à la (73), to account for the problematic sentences. This means that there is still no adequate treatment of sentences like (59), (64), and (65). Nor is there an explanation of examples in which disjunctions with conflicting conventional implicatures are placed in larger constructions which are holes for such implicatures. (60) If the loan company either repossessed Bill’s only car or repossessed his second car, then he will sue the company. is possible (61) It that the loan company either repossessed Bill’s may be only car or repossessed his second car. (62) No, it didn’t. [Said in response to (59)] (63) Did it really? [Said in response to (59)]

{

}

Given principle (55) (55) (A or B)i = (Ae ∨ Bi) & (Be ∨ Ai) & (Ai ∨ Bi) for disjunctions (plus their principles for conditionals, modals, negations, and questions), Karttunen and Peters predict that (60)–(63) conventionally implicate the proposition expressed by (59). (59) Either the loan company repossessed Bill’s only car, or it repossessed his second car. Why, then, don’t speakers who utter these sentences assume the truth of that proposition? Here, it is useful to distinguish two questions that might be asked about (60)–(63):

A Projection Problem

•

55

Question 1 Why is it that a speaker who utters one of them need not be assuming the truth of the proposition expressed by (59)? Question 2 Why is it that a speaker who utters one of them must not be assuming the truth of the proposition expressed by (59)? Any answer to Question 2 is also an answer to Question 1. Moreover, Question 2 is easily answered. A speaker who utters (60)–(63) conversationally implicates that he is not assuming the truth of the proposition expressed by (59).57 Nothing further is needed from a theory of presupposition to answer Question 1. The only thing required is that the theory not contradict the answer to Question 2 by incorrectly claiming that the speaker must be assuming (59). The problem is that the Karttunen-Peters theory does just that. Perhaps, then, it is time for a change of strategy. Modifying the Karttunen-Peters inheritance conditions has not helped; maybe changing the conception of the empirical predictions they make will. Let S be any sentence and let P be the proposition that the Karttunen-Peters implicature conditions (for truth-functional compounds, questions, and epistemic modals) associate with S. Until now, I have been assuming (87): (87)

A speaker who utters S indicates that he is presupposing P.

Perhaps what one ought to assume is (88): (88) A speaker who utters S indicates that he is presupposing P unless he conversationally implicates (or explicitly states) otherwise. (88) makes it impossible for the theory of presupposition to contradict the theory of conversational implicature. If it is adopted, one can retain the Karttunen-Peters algorithms without making the false prediction that a speaker uttering (60)–(63) commits himself to the truth of the proposition expressed by (59). In effect, superfluous presuppositions can be generated so long as they are always conversationally cancelled. Although the difference between (87) and (88) may appear to be slight, it signals a fundamental change in our conception of the projection problem and the status of the algorithms proposed to solve it. Karttunen and Peters think of their inheritance algorithms as semantic rules that specify the conventional implicatures of compound sentences as a function of the extensions and implicatures of their components. Since a conventional 57

See footnote 46.

56

•

Essay One

implicature of a sentence is considered to be part of its meaning, it is natural for Karttunen and Peters to regard such implicatures as noncancellable in Grice’s sense—that is, as representing invariant commitments of the speaker that cannot be withdrawn without rendering his utterance bizarre or contradictory.58 Once (88) is adopted, this picture of presuppositions as noncancellable commitments must be abandoned. If the inheritance conditions are maintained, they must be seen as generating cancellable implicatures. They can then be combined with (88) to explain why uttering (60)–(63) does not commit one to (59).59

58 See Grice (1965, 446–47; 1967, lecture 3; and 1975, 74–75) for discussions of cancellability. 59 (60)–(63) are instances of cancellation via conversational implicature. (i)–(iii) are instances of cancellation via explicit withdrawal of commitment.

(i)

a. If all of the particles released by the collision were negatively charged, then the experiment was a success. b. Some particles were released by the collision. c. If all of the particles released by the collision were negatively charged, then the experiment was a success. However, this assumes that the collision was strong enough to separate some particles from the nucleus and I am not confident that it was. (ii) a. If it’s Susan who has the documents, then we will recover them soon. b. Someone has the documents. c. If it’s Susan who has the documents, then we will recover them soon; but I am afraid that they were destroyed in the fire and will never be recovered. (iii) a. Maybe it’s Susan who has the documents. b. Someone has the documents. c. Maybe it’s Susan who has the documents; but then again, maybe no one does. The Karttunen-Peters implicature conditions associate the (a)-sentences with the propositions expressed by the (b)-sentences. Nevertheless, the (c)-discourses illustrate that the (a)-sentences can be felicitously uttered in cases in which the speaker’s commitment to (b) is withdrawn. Some readers have suggested that the (c)-discourses are all cases in which the speaker “has a change of heart” in mid-speech. I am not convinced that this is so; nor am I sure what hangs on this point. Perhaps the idea is that the sense in which the (c)-discourses illustrate cancellation is not the sense in which Karttunen and Peters deny that conventional implicatures are cancellable. If so, then these discourses do not constitute independent evidence that the Karttunen-Peters denial is untenable. Without trying to settle this issue, I wish simply to make three points. First, the existence of cancellation via conversational implicature is enough to show that the conventional implicatures of truth-functional compounds are cancellable in the sense denied by Karttunen and Peters. Second, in talking of “explicit cancellation” I shall have in mind cases like (i)–(iii). It is immaterial for my purposes whether or not such cancellation is characterized as involving “a change of heart.” Third, explicit cancellation seems much more natural with compound sentences than it does with simple sentences.

A Projection Problem

•

57

Adopting (88) is a step in the right direction. However, it does not solve all the problems. The (revised) Karttunen-Peters algorithms associate (60)–(63) with the conjunction of (89a), (89b), and (89c): (89) a. Ae ∨ Bi b. Be ∨ Ai c. Ai ∨ Bi Although (88) blocks the incorrect prediction that the speaker implicates the entire conjunction (and hence (59) itself), the individual conjuncts still cause problems. A speaker who utters (60)–(63) conversationally implicates that he is not assuming Ae, Be, Ai, or Bi.60 However, there is no reason to think that he also conversationally implicates that he is not assuming (89a), that he is not assuming (89b), or that he is not assuming (89c). This creates two difficulties. First, since the algorithms associate each conjunct of (89) with (60)–(63), and since these implicatures are not individually cancelled, one gets the bizarre result that the speaker implicates each conjunct of (89) without implicating their conjunction. Second, one wants the theory to explain why a speaker uttering (60)–(63) implicates (presupposes) (89c), but not (89a) or (89b). In cases like (74), (76), and (84), one wants the theory to explain why the speaker implicates (presupposes) more than (89c), but not (89a) or (89b). The Karttunen-Peters algorithms, plus (88), do not do this. Thus, there seems to be no way of saving their theory from being undermined by problematic counterexamples.

A Positive Proposal First Approximation So far I have tried two different solutions to the problem posed by disjunctions with conflicting implicatures. One involved changing the (revised) Karttunen-Peters algorithm to (73), while retaining the conception

(iv)

a. Nixon is guilty too. b. Someone other than Nixon is guilty. c. ?*Nixon is guilty too; but then again, maybe, no one else is.

Throughout this essay, whenever I speak of cancellation, I am talking about cancellation of the implicatures (presuppositions) of compound sentences. 60 In the next section I explain why such a speaker conversationally implicates that he is not assuming Ai or Bi. Since these are entailed by Ae and Be (in the case of (59)–(63) and (64), (65)), it follows that the speaker conversationally implicates that he is not assuming Ae or Be.

58

•

Essay One

of the output of such algorithms as non-cancellable conventional implicatures. The other involved retaining the (revised) algorithm (55), while changing the status of the output of inheritance algorithms to that given in (88). Neither solution was successful. I now turn to an alternative that I believe to be more promising. (90) A speaker who utters a truth-functional compound, question, or epistemic modal indicates that he is presupposing all of the presuppositions of its constituents, unless he conversationally implicates (or explicitly states) otherwise.61 (90) incorporates the conception of cancellable presuppositions implicit in (88), while rejecting the Karttunen-Peters inheritance algorithms in favor of the cumulative algorithm.62 It should be noted that (90) makes predictions about what speakers presuppose, not what sentences do. Of course, sentential presupposition can be defined in terms of speaker presupposition. Sentential Presupposition (20) A sentence S of a language L presupposes a proposition P iff uttering S in L indicates that one is presupposing P. Thus, it may be possible to use (90) together with the theory of conversational implicature to make derivative predictions about what sentences presuppose.63 However, this is not what (90) is directly concerned with.

61

For examples of explicit cancellation, see footnote 59. After submitting this essay to the publisher, I was informed by Lauri Karttunen and Gerald Gazdar that my (90) is very similar to a proposal explicitly advocated by Gazdar in his dissertation (1976), and now in Gazdar (1979). Since I was unaware of Gazdar’s work prior to submitting my essay, and since I have not yet had a chance to study it carefully, I cannot comment on it now. The roots of (90) in my own work can be found in Soames (1976, 374–80), where a version of (90) was briefly considered and rejected (p. 380) because of its failure to handle a specified range of “neutral” cases (with the characteristics of (105) and (106) discussed below). The strengths and weaknesses of (90) were explained in a lecture I gave on March 3, 1978, to the linguistics and philosophy programs of the Graduate Center of the City University of New York, and in a lecture I presented on January 22, 1979, to the Yale Linguistics Club, Yale University. The reader who is interested in further discussion, formalization, and advocacy of (90) should consult Gazdar (1979). 63 The difficulties involved in doing this should not be underestimated. If a constituent of a compound sentence S presupposes P, but S also carries a generalized conversational implicature cancelling the presupposition, then S will not presuppose P. (I argue below that (59) is an example of this.) However, if a constituent of S presupposes P, but S does not carry such a cancelling implicature, then the case is more complicated. In some contexts utterances of S might conversationally implicate that the presupposition P is cancelled, while in other contexts they might not. (I argue below that (65) is a sentence of this type.) What are we to say that the sentence S itself (independent of context) presupposes? 62

A Projection Problem

•

59

This approach is just the opposite of that of Karttunen and Peters. Their theory was designed to make predictions about the conventional implicatures (presuppositions) of sentences. Predictions about speakers were derived from these.64 One virtue of (90) is that it eliminates the problems with (59)–(65), (74), (76), and (84) that plague Karttunen and Peters. For example, consider again (59): (59) Either the loan company repossessed Bill’s only car, or it repossessed his second car. A: The loan company repossessed Bill’s only car. B: The loan company repossessed Bill’s second car. PA: Bill had exactly one car. PB: Bill had at least two cars. PA ∨ PB: Bill had at least one car.65 Among the presuppositions that (90) associates with the entire disjunction are PA, PB, and (PA ∨ PB). PA is incompatible with B, and PB is incompatible with A. Thus, if a speaker were assuming PA, he would also be assuming that B was untrue, and so would be in a position to assert the stronger statement A rather than the weaker one, (59). Since he, in fact, asserts the weaker statement, he conversationally implicates that he is not assuming PA. The same holds for PB. Consequently, both of these

Perhaps something of the form (i) (i)

− C→P

(where C somehow describes contexts in which the presupposition would be conversationally cancelled). I am not sure how far this program could be carried through. However, this is not a defect of (90). If (90) is correct, then the program does not need to be carried through. 64 I am indebted to Jerrold Katz for a discussion of some of the differences between an approach which centers on speaker presupposition and one which focuses on sentential presupposition. A point regarding the formulation of (90) should also be noted. Although the outputs of the cumulative algorithm implicit in (90) are speaker presuppositions, the inputs to the algorithms are sentential presuppositions of the constituents of compound sentences. Where one compound sentence is embedded in another (for example, where a disjunction is the antecedent of a conditional), a theory that does not posit sentential presuppositions of compound sentences must take the inputs to the cumulative algorithm to be presuppositions of subconstituents of the sentence uttered (for example, presuppositions of the disjuncts of the antecedent of the conditional). However, when a subconstituent of such a sentence is embedded under a genuine plug for presuppositions, its presuppositions must not be included in the inputs to the cumulative algorithm. 65 (PA ∨ PB) is presupposed by both disjuncts of (59). It is also entailed by (59), so the speaker is committed to it in any case.

60

•

Essay One

presuppositions are cancelled. However, (PA ∨ PB) is not cancelled. Thus, (90) correctly predicts that (PA ∨ PB) is implicated by a speaker who utters (59), but that PA and PB are not. The situation with (65) is analogous, except that here PA and PB are not logically incompatible. (65) Either Bill met the king of Slobovia or he met the president of Slobovia. A: Bill met the king of Slobovia. B: Bill met the president of Slobovia. PA: Slobovia had a king. PB: Slobovia had a president. If the conversational context contains the background assumption (66), then {PA, B} and {PB, A} will each be incompatible with the context. (66)

Countries with kings don’t have presidents.

On the assumption that the speaker intends to increment rather than contradict the assumptions in the context, one can now argue exactly as before that he is not assuming PA or PB. Of course, if the context does not contain something like (66), then PA and PB will not be cancelled. But that is as it should be (as Karttunen’s famous “holy underwear” example shows).66 Similar accounts can be given for (60)–(63). Thus, compound sentences whose constituents bear conflicting presuppositions can be handled by (90). This is not all. (90) can also accommodate virtually all of the data originally cited in Karttunen (1973) to discredit the cumulative algorithm that it contains. (91)

If John has children, then all of his children are bald. PB: John has children. (92) If it is true that John has children, then all of his children are bald. PB: John has children. (93) If Fred has managed to kiss Cecilia, Fred will kiss Cecilia again. PB: Fred has kissed Cecilia. (94) If Fred is married, then his wife is no longer living with him. PB: Fred has a wife.67

66

Karttunen (1973, 182–83). Karttunen (1973, 177). According to Karttunen, the antecedent of (93) expresses a proposition that is logically equivalent to the one expressed by PB. 67

A Projection Problem

•

61

Each of the consequents of these conditionals bears a presupposition PB. Nevertheless, a speaker uttering the entire conditional does not presuppose PB. As before, there are two questions one might ask about these sentences. Question 1 Why is it that a speaker uttering one of them need not be presupposing PB? Question 2 Why is it that a speaker uttering one of them must not be presupposing PB? The best one can hope for in abandoning the cumulative algorithm for more complicated conditions is an answer to Question 1: “The speaker need not be presupposing PB because PB has been filtered and so is not presupposed by the entire conditional.” However, this response leaves Question 2 unanswered. Fortunately, of course, the answer is simple. In each case, a speaker who presupposed PB and who had sufficient evidence to warrant asserting the conditional, would be in a position to make a stronger statement by uttering the consequent. Since he does not do this, he conversationally implicates that he is not presupposing PB. Once this answer has been given to Question 2, one no longer needs “filtering” conditions to answer Question 1. All one has to say is that presuppositions can be cancelled by conversational implicatures. Thus, the cumulative algorithm can be made compatible with much of the evidence that was originally used to discredit it. Further Examples of Conversational Cancelling The case for (90) can be made even stronger. For example, consider (95)–(97): (95) John has children and his sons are bald. (96) It may be that John has children and his sons are bald. (97) If John has children and his sons are bald, then baldness is probably hereditary. A speaker who utters these sentences does not presuppose (98): (98)

John has sons.

If he were presupposing this, the left conjunct of the (main clause or embedded) conjunction would be superfluous and would not be uttered. Instead, (99), (100), or (101) would be used.

62

•

Essay One

(99) (100) (101)

John’s sons are bald. It may be that John’s sons are bald. If John’s sons are bald, then baldness is hereditary.

This suggests that a speaker who does choose (95)–(97) conversationally implicates that he is not presupposing (98). In the case of (95), he must, of course, believe (98), since it is entailed by what he asserts. However, he conversationally implicates that he does not regard (98) as one of the background assumptions common to speaker and hearer (otherwise he would have used (99)). In the case of (96) and (97), he conversationally implicates that he is not sure whether (98) is true or false.68 Thus, the cumulative presupposition is conversationally cancelled without any appeal to “filtering.” In fact, these examples cause problems for the “filtering” approach. According to Karttunen and Peters, (95) conventionally implicates (102): (102)

(John has children → John has sons)

Since epistemic modals and antecedents of indicative conditionals are “holes” for conventional implicatures, Karttunen and Peters predict that (96) and (97) also conventionally implicate (102). This prediction appears to be incorrect. Although a speaker who utters (96) or (97) commits himself to the possibility that (102) is true,69 he does not commit himself to the actual truth of (102). His remark leaves it open whether he believes that (102) is true or whether he is unsure of its truth-value. Thus, an attentive audience will not conclude that he is taking it for granted. Since Karttunen and Peters predict otherwise, these examples constitute an additional problem for their theory. There are other potential problems as well; for example, (103): (103) If John doesn’t have children, then it wasn’t his child who won the fellowship. P1: John has a child. P2: Someone won the fellowship. 68 If he were sure that (98) was true and thought that his audience might regard it as controversial, he would utter (i) or (ii):

(i) John has sons and they may be bald. (ii) John has sons. If they’re bald, then baldness is probably hereditary. If he were sure that (98) was true and thought the audience also assumed it, he would utter (100) or (101). 69 The reason for this is that a speaker who utters (96) or (97) commits himself to the possibility that (95) true. Since (95) entails (102), this explains his commitment to the possibility that (102) is true. This commitment is independent of the “filtering” vs. “cancelling” approaches. To justify the “filtering” approach he must show that the speaker is committed to the truth of (102), not just to the possibility that it is true.

A Projection Problem

•

63

P1 and P2 are presuppositions of the consequent. Nevertheless, only P2 is presupposed by a speaker who utters (103). (90) handles this easily; P1 is conversationally cancelled whereas P2 is not. The Karttunen-Peters theory, on the other hand, cannot handle this example. Karttunen and Peters recognize two types of negation: “ordinary negation,” which is a hole for conventional implicatures (presuppositions), and “contradiction negation,” which is a plug.70 If the consequent of (103) is analyzed as containing “ordinary negation,” then their theory predicts that (103) conventionally implicates (presupposes) (104): (104)

(−P1 → (P1 & P2))

Since (104) entails P1, but not P2, this analysis incorrectly predicts that (103) conventionally implicates (presupposes) P1 but not P2. If the consequent is analyzed as containing “contradiction negation,” then their theory predicts that (103) does not inherit any substantive presuppositions (implicatures) from its consequent. But this leaves the speaker’s presupposition of P2 unexplained. Consequently, (103) poses a further difficulty for Karttunen and Peters.

Two Approaches to Presupposition for Compound Sentences The Karttunen-Peters “Filtering” Approach Complex conditions are associated with truth-functional connectives to determine the implicatures of truth-functionally compound sentences from the extensions and implicatures of their constituents. The Cumulative-Cancelling Approach A speaker who utters a truth-functionally compound sentence S, one of whose constituents presupposes P, indicates that he is presupposing P unless he conversationally implicates (or explicitly states) otherwise. So far I have tested these contrasting approaches against the following classes of data: A. The classical data of Karttunen (1973) ((91)–(94)) B. Data involving compound sentences whose constituents bear conflicting presuppositions (e.g., (59)–(65), (74), (76), (84)) C. Further examples of conversational cancelling ((95)–(97), (103)) 70 A “plug” is the opposite of a “hole.” Whereas a hole transmits the presuppositions of one constituent to a larger constituent, a plug prevents the presuppositions of the smaller constituent from becoming presuppositions of the larger one.

64

•

Essay One

The Karttunen-Peters approach accounts for (A), but has trouble with (B) and (C).71 The Cumulative-Cancelling approach handles each relatively easily.72 It would be nice to be able to conclude that the Cumulative-Cancelling approach solves all the problems and hence should be adopted as it stands. Unfortunately, that is not true. There is one class of examples that poses a serious problem for it (but not for Karttunen and Peters). This class consists of disjunctions and indicative conditionals with the following two characteristics: (i) They contain a constituent sentence that presupposes (implicates) some proposition P. (ii) A speaker uttering them may, but need not, be assuming P. To simplify the discussion I will concentrate only on conditionals. Relevant examples are those in which (iii) The consequent presupposes (implicates) P; and (iv) The antecedent (together with the background context) entails P, but P (together with the background context) does not entail the antecedent.73

71 Note that the problems in (C) as well as those in (B) constitute difficulties for a theory of logical presupposition based on (31) and (35). 72 There is a potential difficulty with the cancelling approach that I have not mentioned. Consider the example (i):

(i) If someone stole the jewels, then it was Sam who stole them. If the set of presuppositions of a sentence is closed under entailment, then the consequent of (i) presupposes not only (ii), but also (iii): (ii) Someone stole the jewels. (iii) Someone stole something. A speaker uttering (i) conversationally implicates that he is not assuming (ii). However, he does not conversationally implicate that he is not assuming (iii). Nevertheless, if (iii) is a presupposition at all, it should be cancelled along with (ii). Intuitively, one wants to say that once a presupposition is cancelled (e.g., (ii)), all propositions which are presuppositions solely because they are entailed by it are cancelled, too (e.g., (iii)). Although something like this seems to be right, I have not stated the CumulativeCancelling approach precisely enough (for reasons that will become clear below) to determine exactly how this proposed solution should be incorporated into it. It should be noted that this issue will arise for any approach that relies on conversational cancelling, including the approach that combines the Karttunen-Peters algorithms with (88). 73 Most of the examples in Karttunen (1973) involve cases in which P and the antecedent (plus relevant background assumptions) are mutually entailing. In such cases it is conversationally implicated that the speaker is not assuming P.

A Projection Problem

•

65

Two such sentences are: (105)

If Haldeman is guilty, then Nixon is guilty, too. P: Someone other than Nixon is guilty.74

(106) If someone at the conference solved the problem, then it was Julius who solved it. P: Someone solved the problem. What must be explained is why a speaker uttering them may, but need not, be assuming P. One part is easy. Since P does not contradict what is asserted and since assuming P is compatible with obeying conversational maxims, the speaker may assume P. According to the Karttunen-Peters approach, the other part is also easy. (105) and (106) do not conventionally implicate (presuppose) P.75 Thus, a speaker uttering one of them does not implicate P, and so need not be assuming it. This is precisely what cannot be explained by the CumulativeCancelling approach as presently stated. According to it, presuppositions of constituent sentences remain intact as assumptions of the speaker unless conversational maxims (or explicit denials) indicate that the speaker must not be assuming them. Examples like (105) and (106) show that this characterization is too narrow.76 It is not clear exactly how to modify the Cumulative-Cancelling approach so as to avoid this difficulty. What is needed is something along the lines of (107). (107) A speaker who utters a truth-functionally compound sentence S, one of whose constituents presupposes P, indicates that he is presupposing P unless a. he explicitly withdraws his commitment to P; or b. he conversationally implicates that he is not assuming P; or c. there is some readily apparent, alternative explanation of why he chose a sentence with a constituent that presupposed P rather than a logically equivalent sentence with no such constituent.

74 This example is cited in Karttunen (1974). It is to be read in such a way that Nixon is the focused element. 75 P is “filtered.” 76 If, as Gazdar informs me, the proposal he advocates is identical to my (90), then examples like (105) and (106) show that it is false as it stands and must be modified. This difficulty with the Cumulative-Cancelling approach was briefly noted in Soames (1976, 379–80).

66

•

Essay One

Clauses (107a) and (107b) are relatively unproblematic. It is clear what it is to explicitly withdraw a commitment. There is also a reasonably well-developed theory of conversational implicature to give content to (107b). The problem is clause (107c). Until it is made precise enough to give clear and unambiguous explanations of examples like (105) and (106), the Cumulative-Cancelling approach cannot be considered adequate. Nevertheless, I think that (107) is a step in the right direction. It handles much of the data that motivated the Karttunen-Peters approach and accounts for a variety of counterexamples to their theory. In addition, the approach to presupposition based on (107) has a certain intuitive plausibility. Imagine being confronted with a speaker who has uttered a truthfunctional compound, one of whose constituents presupposes P. One asks oneself why the speaker didn’t choose a logically equivalent sentence whose constituents do not presuppose P. The Cumulative-Cancelling approach says, in effect, that the null hypothesis is that the speaker used the sentence with the presupposition-bearing constituent because he was taking the presupposition for granted. This presumption remains in force unless something special about the sentence, or the context, removes it. One factor which I have argued can remove it is conversational implicature. If, as theorists, we have not isolated all such factors, I suggest we just keep looking.

A New Projection Problem for Presuppositions Whether or not the Cumulative-Cancelling approach ultimately proves to be correct, it provides a needed contrast to the approach of Karttunen and Peters. This contrast is useful in elucidating the requirements that any theory of presuppositions and truth-functional compounds must satisfy in order to be successful. Requirement 1 The theory should explain why in some cases uttering a compound sentence indicates that one is presupposing the presuppositions of its constituents. (The theory should account for the “positive” cases.) Requirement 2 The theory should explain why in other cases uttering a compound sentence indicates neither that one is, nor that one is not, presupposing the presuppositions of its constituents. (The theory should account for the “neutral” cases.)

A Projection Problem

•

67

Requirement 3 The theory should be integrated into a total account that explains cases in which uttering a compound sentence indicates that one is not presupposing the presuppositions of its constituents. (The theory should allow “negative” cases to be accounted for.) Both the Cumulative-Cancelling approach and the Karttunen-Peters approach go a long way toward satisfying the first requirement. The second requirement constitutes the chief difficulty for my proposal. The third is the chief obstacle for Karttunen and Peters. The new projection problem for presuppositions is to construct a theory that satisfies all three requirements, either by eliminating the defects in one of the approaches discussed here, or by developing an entirely different approach. Until we do this, we cannot claim to have an adequate theory of the presuppositions arising from conjunctions, disjunctions, or conditionals.

Appendix 1. Different Notions of Presupposition (19) Speaker Presupposition. A speaker X presupposes a proposition P (at time t) iff (at time t) a. X assumes P, and b. X assumes that his audience either already accepts P, or is prepared to treat it as uncontroversial. (20) Sentential Presupposition A sentence S of a language L presupposes a proposition P iff uttering S in L indicates that one is presupposing P. (21) Logical Presupposition A proposition P logically presupposes a proposition Q iff the truth of Q is a necessary condition for P to be either true or false. 2. Inheritance Conditions A. Karttunen and Peters: Alternative Formulations 1. Karttunen-Peters (1979) (12) (A and B)i = (Ai & (Ae → Bi)) (13) (If A, then B)i = (Ai & (Ae → Bi)) (14) (A or B)i = ((Ae ∨ Bi) & (Be ∨ Ai)) 2. Karttunen (1974) (22) A context X satisfies-the-presuppositions-of (A and B) iff

68

•

Essay One

a. X satisfies-the-presuppositions-of A, and b. X ∪ {A} satisfies-the-presuppositions-of B. (23) A context X satisfies-the-presuppositions-of (If A, then B) iff a. X satisfies-the-presuppositions-of A, and b. X ∪ {A} satisfies-the-presuppositions-of B. (24) A context X satisfies-the-presuppositions-of (A or B) iff a. X ∪ {− A} satisfies-the-presuppositions-of B, and b. X ∪ {− B} satisfies-the-presuppositions-of A. B. An Analogous System Based on Logical Presupposition (31) a. b. c. B B and T F * if, then T F * or T T F * T T F * T A F F F F A F T T T A F * * * * * * * * *

T T T T

B F T F *

* T * *

(35) c. A speaker who asserts P (at t) indicates that he presupposes (in the sense of (19)) the propositions that are logically presupposed by P. 3. First Revision: Presuppositions Common to Both Disjuncts A. (55) (A or B)i = (Ae ∨ Bi) & (Be ∨ Ai) & (Ai ∨ Bi) B. Crucial Data (49) Either the chairman of the department isn’t a skeptic about the external world or the chairman of the department isn’t very sensible. Ae: −(The chairman of the department is a skeptic about the external world) Be: −(The chairman of the department is very sensible) Ai = Bi = C: The department has a chairman. (50) Either the king of France isn’t wise, or the French generals aren’t under the king’s control. C: There is a king of France. (51) Either it wasn’t Harry who lost the money, or he doesn’t realize that he lost it. C: Someone lost the money. (52) Either it wasn’t Martha who tapped Bill’s phone, or he will never discover that she is the one who did it. C: Someone tapped Bill’s phone. (53) Either every radical in the house will escape, or every one of them will die. C: There is at least one radical in the house.

A Projection Problem

•

69

(54) Either all of John’s children are asleep, or none of his children are asleep. C: John has at least one child. 4. Conflicting Presuppositions A. Crucial Data (59) Either the loan company repossessed Bill’s only car, or it repossessed his second car. Ai: Bill had exactly one car. Bi: Bill had at least two cars. (60) If the loan company either repossessed Bill’s only car or repossessed his second car, then he will sue the company. is possible (61) It that the loan company either repossessed Bill’s may be only car or repossessed his second car. (62) No, it didn’t. [Said in response to (59)] (63) Did it really? [Said in response to (59)] (64) Either the lawyer located Susan’s only heir or he found the oldest of her (several) heirs. Ai: Susan had exactly one heir. Bi: Susan had more than one heir. (65) Either Bill met the king of Slobovia or he met the president of Slobovia. Ai: Slobovia had a king. Bi: Slobovia had a president.

{

}

B. First Unsuccessful Revision: Complicating the Conditions (73) (A or B)i = the conjunction of (a) (Ai ∨ Bi) and (b) −(Ai & Bi) ∨ ((Ae ∨ Bi) & (Be ∨ Ai)) 1. Examples Not Accounted for by (73) (74) Either the lawyer located Susan’s only heir or his assistant located the oldest of her several heirs. (76) If the lawyer located Susan’s only heir or his assistant located the oldest of her several heirs, then the case will be settled soon. (84) Either Susan met the king of Slobovia and insulted him, too, or she met the president of Slobovia and lauded his democratic principles. C. Second Unsuccessful Revision: Defeating the Implicatures Let S be any sentence and let P be the proposition that the Karttunen-Peters conditions associate with S.

70

•

Essay One

(88) A speaker who utters S indicates that he is presupposing P unless he conversationally implicates (or explicitly states) otherwise. 5. Positive Proposals A. First Approximation (90) A speaker who utters a truth-functional compound, question, or epistemic modal indicates that he is presupposing all of the presuppositions of its constituents, unless he conversationally implicates (or explicitly states) otherwise. B. Additional Examples (95) John has children and his sons are bald. (96) It may be that John has children and his sons are bald. (97) If John has children and his sons are bald, then baldness is probably hereditary. (103) If John doesn’t have children, then it wasn’t his child who won the fellowship. P1: John has a child. P2: Someone won the fellowship. C. Counterexamples to (90) (105) If Haldeman is guilty, then Nixon is guilty, too. P: Someone other than Nixon is guilty. (106) If someone at the conference solved the problem, then it was Julius who solved it. P: Someone solved the problem. D. Final Programmatic Proposal (107) A speaker who utters a truth-functionally compound sentence S, one of whose constituents presupposes P, indicates that he is presupposing P unless a. he explicitly withdraws his commitment to P; or b. he conversationally implicates that he is not assuming P; or c. there is some readily apparent, alternative explanation of why he chose a sentence with a constituent that presupposed P rather than a logically equivalent sentence with no such constituent.

A Projection Problem

•

71

References Gazdar, Gerald. 1976. “Formal Pragmatics for Natural Language: Implicature, Presupposition and Logical Form.” Ph.D. diss., University of Reading. ———. 1979. Pragmatics: Implicature, Presupposition, and Logical Form. New York: Academic Press. Grice, H. P. 1965. “The Causal Theory of Perception.” In Perceiving, Sensing, and Knowing: A Book of Readings from Twentieth-Century Sources in the Philosophy of Perception, ed. Robert J. Swartz, 438–72. Garden City, N.Y.: Doubleday. ———. 1967. “Logic and Conversation.” William James Lectures, Harvard University. ———. 1975. “Logic and Conversation.” In The Logic of Grammar, ed. Donald Davidson and Gilbert Harman, 64–75. Encino, Calif.: Dickenson. Hausser, R. R. 1976. “Presupposition in Montague Grammar.” Theoretical Linguistics 3:245–80. Karttunen, Lauri. 1973. “Presuppositions of Compound Sentences.” Linguistic Inquiry 4:169–93. ———. 1974. “Presupposition and Linguistic Context.” Theoretical Linguistics 1:181–94. Karttunen, Frances, and Lauri Karttunen. 1977. “Even Questions.” In Proceedings of the Seventh Annual Meeting of the North Eastern Linguistic Society, ed. J. A. Kegl, D. Nash, and A. Zaenen, 115–34. Cambridge, Mass.: North Eastern Linguistics Society. Karttunen, Lauri, and Stanley Peters. 1975. “Conventional Implicature in Montague Grammar.” In Proceedings of the First Annual Meeting of the Berkeley Linguistics Society, ed. C. Cogen, H. Thompson, G. Thurgood, K. Whistler, and J. Wright, 266–78. Berkeley, Calif.: Berkeley Linguistics Society. ———. 1976. “What Indirect Questions Conventionally Implicate.” In Papers from the Twelfth Regional Meeting of the Chicago Linguistics Society, ed. S. Mufwene, C. Walker, and S. Steever, 351–68. Chicago: Chicago Linguistics Society. ———. 1979. “Conventional Implicature.” In Syntax and Semantics, vol. 11, Presupposition, ed. Choon Kyu Oh and David A. Dineen, 1–56. New York: Academic Press. Katz, Jerrold J., and Terence Langendoen. 1976. “Pragmatics and Presupposition.” Language 52:1–17. Langendoen, D. Terence, and H. B. Savin. 1971. “The Projection Problem for Presuppositions.” In Studies in Linguistic Semantics, ed. Charles J. Fillmore and D. Terence Langendoen, 55–60. New York: Holt, Rinehart and Winston. Peters, Stanley. 1975. “Presupposition and Conversation.” Texas Linguistic Forum 2, Department of Linguistics, University of Texas at Austin, 122–33. ———. 1977. “In Consequence of Speaking.” In Basic Problems in Methodology and Linguistics, ed. Robert E. Butts and Jaakko Hintikka, 283–97. Dordrecht: Reidel.

72

•

Essay One

———. 1979. “A Truth-Conditional Formulation of Karttunen’s Account of Presupposition.” Synthese 40:301–16. Soames, Scott. 1976. “A Critical Examination of Frege’s Theory of Presupposition and Contemporary Alternatives.” Ph.D. diss., MIT. Stalnaker, Robert. 1972. “Pragmatics.” In Semantics of Natural Language, ed. Donald Davidson and Gilbert Harman, 380–97. Boston: Reidel. ———. 1973. “Presuppositions.” Journal of Philosophical Logic 2:447–57. ———. 1974. “Pragmatic Presuppositions.” In Semantics and Philosophy, ed. Milton K. Munitz and Peter K. Unger, 197–214. New York: New York University Press. Wilson, Deirdre. 1975. Presuppositions and Non-Truth-Conditional Semantics. New York: Academic Press.

ESSAY TWO

Presupposition

Basic Questions To presuppose something is to take it for granted in a way that contrasts with asserting it. For example, if one assertively utters (1a)

It was Sam who broke the typewriter.

one presupposes that the typewriter was broken and asserts that Sam was the one who did it. Similarly, if one assertively utters (2a)

John is going to drop out of school again.

one presupposes that he has dropped out of school before and asserts that he will drop out in the future. In each case, the speaker commits himself both to that which he presupposes and to that which he asserts. However, there are important differences between the two. One such difference is that commitments that are presupposed are highly heritable, whereas those that are only asserted are not. If assertive utterances of a sentence S are used to assert A and presuppose P, then assertive utterances of more complicated sentences containing S often presuppose P without carrying any commitment to A. This is illustrated by the examples in (1) and (2). (1b) (1c) (1d) (1e) (P) (A) (2b)

It wasn’t Sam who broke the typewriter. Maybe it was Sam who broke the typewriter. It is unlikely that it was Sam who broke the typewriter. If it was Sam who broke the typewriter, then he will have to fix it. Someone broke the typewriter. Sam broke the typewriter. John isn’t going to drop out of school again.

This essay was written in 1983–84 while on leave from Princeton University on The Class of 1936 Bicenntenial Preceptorship, and while a guest of the Syntax Research Center at the University of California, Santa Cruz. (Minor revisions were made in 1986 prior to publication.) Portions of the chapter served as the basis for talks in the spring of 1984 at The Center for the Study of Language and Information at Stanford University. I would like to thank the Syntax Research Center at Santa Cruz for the use of their facilities; and also Joseph Almog, Saul Kripke, Julius Moravcsik, and Nathan Salmon for discussion of parts of the manuscript.

74

•

Essay Two

(2c) Maybe John is going to drop out of school again. (2d) It is unclear whether John is going to drop out of school again. (2e) Either John will pass the course or he is going to drop out of school again. (P) John has dropped out of school before. (A) John will drop out of school in the future. Heritability is such a striking feature of presuppositions that they are often identified as those commitments that are inherited in the kinds of linguistic environments just mentioned. However, presuppositions are not inherited in all environments. For example, utterances of the sentences in (3) do not presuppose P even though they contain constituents that do. (3a) (3b) (3c) (P)

If the typewriter was broken, then it was Sam who broke it. The typewriter was broken and it was Sam who broke it. Either the typewriter wasn’t broken, or it was Sam who broke it. The typewriter was broken.

The contrast between these examples and those given above raises three basic questions which, in broadest terms, define the descriptive task for linguistic theories of presupposition. Descriptive Questions (Q1) (Q2) (Q3)

What presuppositions do various constructions give rise to? Which constructions allow utterances to inherit the presuppositions of their constituents and which do not? What do utterances of arbitrary sentences presuppose?

Although these questions are important, they are not the only ones that theories of presupposition are responsible for. In addition to the descriptive task of identifying the presuppositions of various utterances, a theory of presupposition should specify the kind of phenomenon presupposition is, and how it fits into general theories of the semantic encoding of information by sentences and the pragmatic exchange of information in communicative situations. In short, an adequate theory should answer not only descriptive questions about the scope of presupposition, but also foundational questions about its nature. Foundational Questions (Q4) What is presupposition—what does it mean to say that x presupposes y?

Presupposition

•

75

(Q5) Why are there linguistically expressed presuppositions at all— what functions do presuppositions have in the representation and communication of information? (Q6) How are presuppositions of utterances affected by the semantic rules that determine the information encoded by a sentence relative to a context, and the pragmatic rules that specify the manner in which utterances increment sets of assumptions common among conversational participants? Historically, three main approaches to presupposition have been developed corresponding to three different answers to (Q4). The first approach consists of theories of logical presupposition, deriving ultimately from the work of Gottlob Frege.1 According to these theories, presupposition is, in its primary sense, a relation between propositions. A proposition P is said to logically presuppose a proposition Q iff the truth of Q is a necessary condition for P to be either true or false. Logical Presupposition: A proposition P logically presupposes a proposition Q iff for all possible circumstances w, if P is true or false in w, then Q is true in w. Sentences are said to bear logical presuppositions in a derivative sense: A sentence S logically presupposes a proposition Q (relative to a context C of utterance) iff S expresses a proposition P (in C) that logically presupposes Q. The second approach derives from the work of Peter Strawson and consists of theories of what might be called “expressive presupposition.”2 According to these theories, presupposition is a relation between a sentence, or a use of a sentence, and a proposition. A sentence (or use of a sentence) S can be said to expressively presuppose a proposition P iff the truth of P is a necessary condition for S (or a use of S) to express a proposition. On this view, an assertive utterance of a sentence S in a context C fails to semantically express a proposition if one or more of the relevant presuppositions is false. The third approach consists of theories of pragmatic presupposition in roughly the sense articulated by Robert Stalnaker.3 According to these theories, presuppositions are requirements that sentences, or utterances of sentences, place on sets of common background assumptions built up among conversational participants. Typically, the requirement is that this set of assumptions contains a specific proposition, or some proposition from a limited range of alternatives. Presuppositions in this sense are 1

See in particular, Frege (1952a, 1952b, 1952c). Strawson (1950, 1952). 3 Stalnaker (1972, 1973, 1974). 2

76

•

Essay Two

essentially things taken for granted at a given point in a conversation. The sources of these pragmatic presuppositions vary from case to case, and theory to theory. For example, logical presuppositions, expressive presuppositions, conventional implicatures, conversational implicatures, and general pragmatic strategies of context incrementation have all been held to play important roles in determining the pragmatic presuppositions of utterances. On this view, theories of presupposition are neither exclusively semantic nor exclusively pragmatic, but rather require the integration of both kinds of information. The leading ideas behind these three approaches can be made clearer by considering some paradigmatic examples in historical context.

Three Approaches to Presupposition Fregean Examples of Logical Presupposition The most widely discussed (putative) examples of logical presupposition are so-called referential presuppositions, corresponding to uses of singular terms. The classical Fregean explanation of these examples relies on his bipartite semantics of sense and reference. For example, consider the positive version of (4). (4) The queen of England is (isn’t) popular. (P) England has a (unique) queen. According to Frege, the proposition expressed by this sentence can be broken down into two parts: one part consisting of the sense (or meaning) of the subject expression, and the other part consisting of the sense (or meaning) of the predicate. Each of these is a “mode of presentation” of a referent. Senses of predicate expressions present functions from objects to truth-values (Truth and Falsity). Senses of singular terms present objects. The truth-value of the proposition expressed by a simple subjectpredicate sentence (as well as that of the sentence itself) is defined to be the value of the function referred to by the predicate at the argument referred to by the subject. An important aspect of Frege’s distinction between sense and reference is his recognition that some singular terms have senses that fail to present referents—for example, ‘the Democrat elected President of the U.S. in 1980’.4 It follows from Frege’s semantics that sentences containing such 4

For Frege, this term results from applying the definite description operator to the complex predicate ‘——is a Democrat and ——was elected President of the U.S. in 1980’. The referent of the latter is a function from objects to truth-values. The referent of the definite

Presupposition

•

77

terms (in environments in which they occur with their customary sense and reference) must lack truth-values. One such sentence is the positive version of (5). (5) The Democrat elected President of the U.S. in 1980 is (isn’t) popular. (P) A (unique) Democrat was elected President of the U.S. in 1980. By definition, the truth-value of this example is the value of the function referred to by the predicate ‘is popular’ at the argument referred to by the subject. But since the subject fails to refer, there is no such argument and hence no such truth-value. Therefore, the example is truth-valueless. Negations are treated similarly. The propositions expressed by the negative versions of (4) and (5) are taken to consist of the propositions expressed by their positive counterparts together with the sense of a sentential negation operator. This sense presents a function f as referent from Truth and Falsity to Falsity and Truth. The truth-value of each of these negative propositions is defined to be the value of the negation function f at the argument consisting of the truth-value of the corresponding positive proposition. Where this positive proposition is truthvalueless, there is no such argument and, hence, no truth-value for the negation. Thus, corresponding positive and negative examples are either jointly truth-valued or jointly truth-valueless. In both cases (4)–(5), the truth of P is a necessary condition for the positive and negative propositions to have a truth-value. As a result, these propositions (and the sentences that express them) logically presuppose P. As such, they illustrate the more general point that negations share the logical presuppositions of their positive counterparts. It is worth distinguishing those aspects of this analysis that are peculiar to Frege from those that are central to theories of logical presupposition in general. For Frege, claims about presuppositions are consequences of his compositional theory of sense and reference. His decision to take predicates to designate (total) functions from objects to truth-values has the consequence that truth-valuelessness arises from reference failure on the part of singular terms. His decision to take truth functional operators to denote functions from the truth-values of their operands to the truth-values of

description operator is a second level function which takes a function f as argument and assigns the value o iff o is the unique object that f assigns the value Truth. In the case of ‘the Democrat who was elected President of the U.S. in 1980’ there is no such object. As a result, the definite description fails to refer. Reference failure can also come about in simpler cases, e.g., ‘the king of France’. Here the expression ‘the king of ——’ could be analyzed as referring to a function from countries to their kings, which is undefined at the argument France.

78

•

Essay Two

larger, compound sentences has the consequence that these sentences will be truth-valued only if each of their truth functional constituents is. Neither of these decisions is constitutive of logical presupposition in general. What is constitutive is an analysis of (4)–(5) in terms of the following pair of assumptions: (i) The negative versions of (4)–(5), and the propositions they express, are (logically) negations of their positive counterparts.5 (ii) In each case, P is entailed (necessitated) by both the positive and negative propositions (sentences).6 It follows from these assumptions that the positive and negative examples in each case logically presuppose P; and, hence, that these examples are neither true nor false when P is untrue. It so happens that the only logical presuppositions generated by Frege’s explicit semantics are referential presuppositions (which express necessary conditions for the propositional constituents corresponding to singular terms to present referents). However, there is nothing essential in this either to theories of logical presupposition in general, or to Frege’s basic semantic framework. Suppose one dropped the Fregean requirement that the functions designated by predicates be total. For example, the function designated by ‘is forgetful’ might be defined only on animate beings; and the function designated by the factive verb ‘realize’ might be defined only over pairs consisting of individuals and true propositions. On this analysis, the deviance of the examples in (6) could be traced to the falsity of the logical presupposition arising from the sortal restriction on the predicate; and the inferences from (7a) and (7b) to (7P) could be treated on a par with those in (4)–(5). (6a) The speed of light is (isn’t) forgetful. (6b) The sum of 2 and 3 is (isn’t) forgetful. (7a) John realized that time was running out. (7b) John didn’t realize that time was running out. (P) Time was running out. A similar analysis could, in principle, be applied to the examples in (8).7

5 The relevant negation operator is sentential, and the resulting negation is true (false) iff its corresponding positive counterpart is (false) true. 6 If R and S are propositions, I say that R entails (necessitates) S iff there is no possible circumstance w such that R is true in w and S is not true in w. 7 For Frege the unrestricted universal quantifier refers to a second level function that takes a function f (denoted by a complex predicate) as argument and assigns the value Truth iff f assigns Truth to every object; otherwise it assigns Falsity. Taking this as a model, one might claim that the restricted quantifier ‘all graduate students in the class’ referred to a second

Presupposition

•

79

(8a) All graduate students in the class wrote term papers. (8b) Not all graduate students in the class wrote term papers. (P) There were graduate students in the class. Since presupposition was not one of Frege’s main concerns, he did not canvass natural language to determine the different kinds there might be. In particular, he did not consider nonreferential examples like those just illustrated. However, the ease with which his system can be extended to provide a unified account of referential and non-referential cases has made it an important model for later theories of logical presupposition. Unfortunately, there is one respect in which Frege’s theory is clearly mistaken. We have seen that Frege takes n-place truth functional operators to designate n-place truth functions; and that he defines the truthvalue of a truth functional compound to be the value of the relevant nplace truth function at the n-tuple of truth-values of its truth functional constituents. As a result, the argument used to show that a negation is truth-valueless iff its positive counterpart is truth-valueless can be generalized to yield the conclusion that a truth functional compound is truthvalueless iff one of its constituents is.8 But this conclusion is incorrect, as is shown by the fact that (9a) is true, and by the fact that the examples in (9) do not presuppose (9P).9

level function that assigned Truth to f iff there were graduate students in the class and f assigned Truth to each of them; and assigned Falsity to f iff there were graduate students in the class and f assigned Falsity to at least one of them; and otherwise was undefined. 8 When a truth functional constituent is truth-valueless, there is no such thing as the ntuple of truth-values of the constituents; and hence, no such thing as the value of the relevant truth function at that n-tuple. 9 This point was first noted by Bertrand Russell in “On Denoting.” There Russell criticizes Frege as follows: “Or again consider such a proposition as the following: ‘If u is a class which has only one member, then that one member is a member of u’, or, as we may state it, ‘If u is a unit class, the u is a u’. This proposition ought to be always true, since the conclusion is true whenever the hypothesis is true. But, ‘the u’ is a denoting phrase, and it is the denotation, not the meaning, that is said to be a u. Now if u is not a unit class, ‘the u’ seems to denote nothing; hence our proposition would seem to become nonsense as soon as u is not a unit class. Now it is plain that such propositions do not become nonsense merely because their hypotheses are false. The King in The Tempest might say, ‘If Ferdinand is not drowned, Ferdinand is my only son’. Now, ‘my only son’ is a denoting phrase, which, on the face of it, has a denotation when, and only when, I have exactly one son. But the above statement would nevertheless have remained true if Ferdinand had been in fact drowned” (Russell 1905, 484). If ‘truth-valueless’ is substituted for ‘nonsense’ in the above quotation, then the passage correctly diagnoses the central difficulty with Frege’s theory of presupposition inheritance, Further, implicit, criticism of Frege is contained in Russell’s recognition that ‘the king of France is not bald’ is sometimes understood in such a way that it is true (if there is no king.)

80

•

Essay Two

(9a) Either there is no king of France or the king of France is in hiding. (9b) If there is a king of France, then the king of France is one of the few remaining European monarchs. (9c) There is a king of France and the king of France is wise. (P) There is a (unique) king of France. A more reasonable treatment of truth functional connectives that avoids the difficulties posed by (9) is given in (10). (10a) A or B B A T F *

T T T T

(10b) If A, then B F

*

T T F * * *

(10c) A and B B A T F *

B T F * A T F *

T F * T T T * * *

(10d) Not A

T

F

*

T F *

F F *

* F *

A T F *

Not A F T *

The consequences of this treatment of the connectives will be explored further in the section “The Insufficiency of Conventional Implicature and Logical Presupposition as Sources of Pragmatic Requirements” below.

Strawsonian Examples of Expressive Presupposition The examples in (11) illustrate a different kind of presupposition. (11a) He is wealthy. (11b) This is a fine red one. (11c) That little bug is harmless. In each case, a singular term is used to refer to an entity which the rest of the sentence says something about. This entity is intimately involved in what is said by the utterance in a way that contrasts with examples like (12). (12) The president of General Motors (whoever he may be) is wealthy. For example, imagine a situation in which the pronoun in (11a) is used demonstratively to refer to a man m who, in fact, is the president of

Presupposition

•

81

General Motors. Let P be the proposition expressed by such a use of (11a) and Q be the proposition expressed by a use of (12). Clearly, P and Q are different propositions. P is true in a possible circumstance of evaluation w iff m is wealthy in w. Q is true in w iff whoever happens to be the president of General Motors in w is wealthy in w. In effect, that which a use of ‘he’ contributes to the truth conditions of what is said by an utterance is its referent in the context: whereas that which a use of ‘the president of General Motors’ contributes is its descriptive sense. Thus, an utterance of (13a) is true just in case the (present) referent m of ‘he’ was poor 20 years ago; whereas (13b) is true just in case 20 years ago General Motors had a pauper for a president. (13a) Twenty years ago he was poor. (13b) Twenty years ago the president of General Motors (whoever he may have been) was poor. The point seems to hold even in propositional attitude constructions. For example, the assertions made by utterances of the sentences in (14) seem to be the same. (14a) Mary said that he was poor. (Uttered pointing at m.) (14b) Mary said that I was poor. (Uttered by m.) (14c) Mary said that you were poor. (Uttered to m.) This suggests that the contribution of a demonstrative or indexical to the proposition expressed by an utterance of a sentence containing it is simply its referent in the context. Following David Kaplan, we may refer to this view as the thesis that demonstratives are directly referential. What happens when a use of a directly referential term fails to semantically determine a referent? In such a case, it is natural to suppose that the sentence fails to semantically express a proposition relative to the context. With this in mind, one can define a notion of expressive presupposition as follows: Expressive Presupposition: A sentence S expressively presupposes a proposition P relative to a context C iff the truth of P is necessary for S to semantically express a proposition in C. (11a–c) can then be characterized as expressively presupposing (15a–c). (15a) There is a contextually salient male under discussion. (15b) ‘This’ refers to something relative to the context. (15c) There is a contextually salient little bug under discussion. Although both logical and expressive presupposition have been motivated using examples in which the truth of a proposition is necessary for a (use of a) singular term to secure a referent, the two kinds of presupposition

82

•

Essay Two

are conceptually quite different. If S logically presupposes P relative to a context C, then the proposition expressed by S in C must entail (necessitate) P. This is not so when S expressively presupposes P relative to C (even when S succeeds in expressing a proposition). One can think of this difference as corresponding to two different stages in the semantic evaluation of a sentence. The first stage consists in associating the sentence with the proposition it expresses in the context. The semantic mechanisms responsible for this are what give rise to expressive presuppositions. The second stage consists in evaluating the truth or falsity of that proposition with respect to different (possible) circumstances of evaluation. The semantic mechanisms for determining the extensions (referents) presented by propositions and their various constituents are what give rise to logical presuppositions. It is just this two stage conception of semantics that is needed to distinguish expressive and logical presuppositions involving directly referential and non-directly-referential singular terms.10 However, there is a historical irony in this. The first presentation and discussion of expressive presuppositions is given in Peter Strawson’s influential paper “On Referring,” long before the systematic development of two stage semantic theories by David Kaplan, and others.11 As a result, Strawson’s important insights were obscured and he was unable to successfully distinguish his new notion of expressive presupposition from the Fregean 10 The two stage conception of semantics needed to distinguish logical from expressive presuppositions has important theoretical consequences for the familiar conception of semantic theory as consisting of a definition of truth relative to a context and a circumstance of evaluation. Although such a definition doesn’t mention propositions, it does associate sentence/context pairs with functions from circumstances to truth-values. If those functions are identified with propositions, then an analysis of an example like ‘The largest prime number is odd’ as logically presupposing a necessary falsehood will assign the sentence the degenerate function which is undefined on all circumstances. A problem arises when one notices that such a theory will assign the same “proposition” to ‘This is a fine red one’ in a context in which ‘this’ fails to refer. (Since ‘this’ has no referent relative to the context C, there is no circumstance E such that the referent of ‘this’ relative to C and E is a member of the extension or antiextension of ‘is a fine red one’ in E.) Thus, a semantic theory of this familiar sort will miss the distinction between expressing a proposition that lacks a truth-value in every circumstance, and failing to express a proposition at all. This distinction is captured in the semantic frameworks of Salmon (1986) and Soames (1987), where propositions are not identified with functions from circumstances to truth-values, but rather are assigned to sentence/context pairs prior to evaluation for truth-value. 11 The seminal work on the two stage conception of semantics is Kaplan (1989). Kamp (1968) is a significant precursor. The conception is developed further in Salmon (1986) and Soames (1987), where semantic theories assign sentence/context pairs structured Russellian propositions which determine, but are not determined by, functions from circumstances of evaluation to truth-values.

Presupposition

•

83

notion of logical presupposition. Since Strawson’s work has been highly influential in bringing presupposition to the attention of semantic theorists, it may be worthwhile to say a word about this. The central theses of “On Referring” are as follows: Thesis 1: Meaning is a property of expressions; referring, being true or false, and saying something are properties of uses of expressions in contexts. Thesis 2: A sentence is meaningful iff it could be used to say something true or false. Thesis 3: To give (or know) the meaning of a sentence is to give (or know) a rule for determining the contexts in which it is used to say something true and the contexts in which it is used to say something false. Thesis 4: The semantic function of a singular term (demonstrative, pronoun, name, definite description) in its primary referring use12 is to refer to an entity which the rest of the sentence is used to say something about. The meaning of such an expression is a rule for determining its referents in different contexts. Thesis 5: If a singular term b in a sentence ⎡Fb⎤ is used referringly in a context C, then this use of ⎡Fb⎤ in C says something true (false) in C iff in C, the referent of b has (doesn’t have) the property F is used to express. If the use of b fails to refer to anything, then the use of ⎡Fb⎤ in C doesn’t say anything true or false. Thesis 6 (Definition): If the truth of P is a necessary condition for a use of S in C to say something true or false, then S presupposes P relative to C.

12 In “On Referring,” Strawson says that a (uniquely) referring use of a singular term is one in which the term is used to mention some particular individual. This is intended to rule out the use of ‘the whale’ in the generic claim (i).

(i)

The whale is a mammal.

Strawson also indicates that predicative uses of singular terms are not (uniquely) referring uses. This stipulation is intended to exclude the use of ‘the greatest French soldier’ in (ii). (ii)

Napoleon was the greatest French soldier.

The idea in both cases is to exclude uses in which what is grammatically a singular term is not functioning semantically as a singular term. It is important not to confuse Strawson’s notion of a (uniquely) referring use of a singular term with Keith Donnellan’s (1966) notion of a referential use of a singular term. For Strawson, any use of a term which is genuinely singular and nonpredicative would seem to qualify as a (uniquely) referring use. There is no recognition in Strawson (1950) or (1952) that some such uses might work in an essentially Fregean fashion, while others might work demonstratively.

84

•

Essay Two

Thesis 7: Uses of ⎡G[the F]⎤, ⎡All F’s are G’s⎤, ⎡Some F’s are G’s⎤, ⎡No F’s are G’s⎤, and ⎡Some F’s are not G’s⎤, presuppose that which is expressed by ⎡There is at least one F⎤. The key thesis is Thesis 3. However, there is a problem with it. As it stands, it does not rule out, and may even be taken to suggest, that the meaning of a sentence can be represented as a function from contexts of utterance to truth-values. This is at variance with the two stage conception of semantics in which the meaning of a sentence is represented by a function from contexts to propositions, where the latter determine functions from (possible) circumstances of evaluation to truth-values. Since the circumstance of the context is one of these possible circumstances, the meaning of a sentence determines the one stage mapping from contexts to truth-values suggested in Strawson’s Thesis 3. However, the latter does not determine the former. As a result, there are important semantic distinctions that Thesis 3 does not explicitly accommodate. This is illustrated by (16a) and (16b). (16a) I exist. (16b) I am here now. Any context in which one of these would express a truth (falsehood) is a context in which the other would as well. Nevertheless, they do not have the same meaning. If I were to assertively utter both in the present context, my utterance of (16a) would express the proposition that Scott Soames exists; whereas my utterance of (16b) would express the proposition that Scott Soames is in Santa Cruz on April 6, 1984. Since these are different propositions, the semantic contents of (16a) and (16b) must be distinguished. To make this explicit, Thesis 3 should be replaced with Thesis 3′.13 Thesis 3′: The meaning of a sentence is a rule for determining the propositions it expresses in different contexts. Each such proposition determines a rule for assigning truth-values to (possible) circumstances of evaluation. This requires corresponding changes in Theses 4 and 5. To make these changes, one must decide what singular terms contribute to the propositions expressed by uses of sentences containing them. Strawson’s insistence that the semantic function of a singular term is to refer to an object, and his tendency to treat referring uses of demonstratives as prime ex-

13

The pattern of argument here follows Kaplan (1989).

Presupposition

•

85

amples of this function, suggest a reformulation in which all referring uses of singular terms are directly referential.14 Thesis 4′: The propositional constituent corresponding to a (referring) use of a singular term b in a context C is the referent of b in C. The meaning of a singular term is a rule for determining the propositional constituents corresponding to uses of the term in different contexts. Thesis 5′: If a singular term b in a sentence ⎡Fb⎤ is used to refer to an object o in a context C, and if F is used in C to express the property P, then ⎡Fb⎤ expresses a proposition in C which is true (false) in a possible circumstance w iff o has (doesn’t have) P in w. If b fails to refer to anything in C, then there is no propositional constituent corresponding to b in C, and ⎡Fb⎤ fails to semantically express a proposition in C. The theory of presupposition that emerges from this reconstruction of Strawson’s theses is a theory of expressive presupposition, as defined above. As we have seen, the combination of Theses 3′–5′ provides a plausible account of examples like those in (11) in which a pronoun, demonstrative, or demonstrative phrase is used referringly. However, it clearly produces incorrect results when extended to the range of cases mentioned in Thesis 7. This extension also conflicts with Strawson’s expressed intentions. In Chapter 6 of Introduction to Logical Theory, Strawson (1952, 175) defines presupposition as follows:15 (17) A statement (proposition) S presupposes a statement (proposition) S ′ iff the truth of S ′ is a necessary condition for S to be true or false.16 14 See also the beginning of section 4 of Strawson (1950), where he claims that to state a fact about an individual one must perform the (uniquely) referring task and the attributive task. To use an expression to perform the first of these tasks is, he says, to use it in a uniquely referring way. However, one can perform the same task without using any expression. He illustrates this by examples in which the object is presented directly, without linguistic mediation, as when one paints the words ‘unsafe for lorries’ on a bridge, or ties the label ‘first prize’ on a vegetable marrow. The suggestion here seems to be that in all these cases—linguistic and nonlinguistic—the statement made can be thought of as consisting of the object secured by the uniquely referring task, together with the property attributed to it. 15 Strawson (1952, 175). 16 Strawson distinguished statements from sentence types, sentence tokens, and acts of uttering a sentence in a context. The statement made by an utterance of a given sentence was supposed to be that which was said or asserted by the utterance. Although this made statements sound like propositions, the situation was complicated by Strawson’s insistence that there really were no such entities as statements or propositions. This policy of countenancing

86

•

Essay Two

It was apparently this definition that he had in mind when discussing presuppositions of examples of the kind mentioned in Thesis 7 (both in “On Referring” and in Introduction to Logical Theory). Since (17) defines logical presupposition, Strawson’s adoption of it belies any clear commitment to expressive presupposition, or any systematic analysis of the constructions mentioned in Thesis 7 along directly referential lines. This points up a second possible reconstruction of Strawson’s position. On this construal, his account of presupposition is basically the same as Frege’s, without the compositional semantics, but with an explicit stipulation that propositions involving restricted quantifiers are bearers of presuppositions. This theory is potentially broad in scope and has been historically influential. However, its leading ideas are not original with Strawson. As a historical point, it would be a mistake to attribute to Strawson either an account of presupposition that is systematically Fregean (logical) or an account that is systematically expressive. His major discussions include elements of both, the conflict being masked by his failure to articulate the crucial account of meaning given in Thesis 3′. Once this deficiency is corrected, Strawson’s main original contribution to the study of presupposition lies in the reconstructed account of expressive presupposition suggested by his work. This account is attractive for examples involving various kinds of indexical elements. Some have suggested that it may be possible to extend it from examples involving demonstrative phrases like ‘this little bug’ to corresponding examples involving definite descriptions like ‘the little bug’.17 What it cannot be is a comprehensive theory of presupposition in general. Pragmatic Presupposition Robert Stalnaker, and others, have argued that in order to arrive at a comprehensive theory, it is necessary to adopt a pragmatic account of what presupposition is. The account is based on the observation that

talk about statements/propositions, without taking them seriously, was an important contributing factor to his failure to distinguish between expressive and logical presupposition. Had he taken statements or propositions seriously, he might have been led to ask what sorts of things uses of various expressions contributed to them. Where the contributed elements were missing, it might have been natural to describe the situation in terms of expressive presupposition. Where the contributed elements were present, but it still seemed as if the statement or proposition failed to be true or false, logical presupposition would have been the relevant notion. 17 For discussion see Donnellan (1978), Kripke (1979), Wettstein (1981), Salmon (1982), Barwise and Perry (1983), and Soames (1986), which is essay 12 of this volume.

Presupposition

•

87

sentences are used in communication to contribute to an already existing conversational record, which contains a set of common background assumptions built up among conversational participants. Because of this it is natural for speakers to develop conventional means of indicating what assumptions they are making about the common background to which their utterances contribute. In particular, it is understandable that certain words and constructions should come to be used for this purpose. For example, the following (a)-constructions seem to be designed for use in conversations in which the information expressed by the (b)constructions is already assumed. (18a) It was NP that VPed. (18b) Something (someone) VPed. (19a) Even NP VPed. (19b) Others under consideration, besides NP, VPed. Of those under consideration, NP was among the least likely to VP. A sentence that indicates that such assumptions are being made can be thought of as putting requirements on the conversational record at the time of utterance—requirements that must be satisfied if the speaker’s communicative intentions are to be fulfilled. Presuppositions, on this view, are just such requirements. Suppose now that a speaker utters a sentence S which requires that the conversational record satisfy a certain condition—say that it contain a specific proposition P as part of the common background. Suppose further that P is not already part of the background, but that the conversational participants are ready to accept P as uncontroversial, at least for present conversational purposes. What sort of response would be reasonable on the part of hearers in such a case? The legalistic response would be to object to the speaker’s remark on the grounds that P, which was required by the remark, had not already been established prior to the utterance. The speaker could then ask whether his hearers were willing to accept P, and be told that they were. After adding P to the context, the speaker could repeat his original remark and continue along as before. But there is really no point in this. Since the hearers are ready to accept P anyway, they might as well add it to the background and let the speaker go on without objection. In other words, the most efficient and cooperative response on the part of the hearers is to accommodate the speaker by updating the conversational record so that it meets the requirements of the speaker’s utterance. The reasoning leading to this strategy of accommodation is something that conversational participants can be expected to be familiar with, or to work out for themselves. Knowing this, a speaker can exploit the strategy

88

•

Essay Two

by uttering sentences whose presuppositional requirements he knows are not already satisfied by the existing conversational record. So long as he takes the content of these requirements to be both recognizable and unlikely to provoke objections, he can rely on his hearers to accommodate him by incrementing the common background in the appropriate way. A speaker can use presuppositions in this way to introduce new information, as well as, in some cases, to subtly insinuate a point of view regarding what can be taken to be uncontroversial, and hence beyond further discussion.18 On this picture, the presuppositions of a sentence are conditions it imposes on conversational records. Often these conditions take the form of requirements that the common background assumptions present in the record contain a specific proposition. However, other types of presuppositional requirements are also possible. For example, a sentence may require that the common background contain at least one proposition from a specified range of propositions, it may require that it contain no propositions of a specified sort; or it may require that the topic of conversation at the time of the utterance be one thing rather than another. Once the presuppositional requirements of sentences have been determined, incrementation of the record occurs in accordance with the strategy of accommodation. An important feature of this approach is its eclecticism regarding the factors that give rise to different presuppositional requirements. Proponents of pragmatic presupposition have suggested that these requirements might be derived from a variety of sources—including logical presupposition, expressive presupposition, conventional implicature, and non-conventional pragmatic facts. For example, in discussing logical presupposition as a source of pragmatic presuppositional requirements, Robert Stalnaker says the following: The relation between the semantic [logical] notion of presupposition and the pragmatic notion of presupposition requirement is not, of course, just accidental. Among the reasons that a pragmatic presupposition might be required by the use of a sentence, by far the most obvious and compelling reason would be that the semantical rules for the sentence failed to determine a truth value for the sentence in possible worlds in which the required presupposition is false. Since the whole point of expressing a proposition is to divide the relevant set of alternative possible situations—the presupposition set—into two parts, to distinguish those in which the proposition is true from those in which the proposition is false, it would obviously be inappropriate to use a 18 The strategy of accommodation, implicit in Stalnaker (1973), is explicitly formulated in Lewis (1979).

Presupposition

•

89

sentence which failed to do this. Thus, that a proposition is presupposed by a sentence in the technical semantic sense provides a reason for requiring that it be presupposed in the pragmatic sense whenever the sentence is used. This explains where the semantic notion gets its name, and why linguists and philosophers have been tempted to identify presupposition in general with this semantic relation.19 According to this picture, the exchange of information in a conversation increases the number of propositions in the common background against which the conversation takes place. Taken together, these propositions determine a set of possible worlds which, at any given moment, represent the alternatives compatible with everything that has been said or assumed in the conversation up to that point. The function of an assertive utterance of a sentence S is to further constrain these alternatives by eliminating the worlds in which the proposition expressed by S is false, while retaining those in which it is true. Note that if S logically presupposes a proposition Q which is not entailed by the propositions in the common background, then the set of conversationally alternative worlds will contain some members w in which the proposition expressed by S cannot be correctly characterized as either true or false. Stalnaker’s contention is that in such a case the assertive utterance will fail in its primary purpose of determining a new set of alternative possibilities, since it will fail (barring accommodation) to determine what should be done with worlds in the old alternative set in which Q is false. Thus, Stalnaker maintains, logical presuppositions provide one kind of principled explanation of the pragmatic presuppositional requirements of sentences. A similar story might be told regarding expressive presuppositions. Suppose that the truth of P is a necessary condition for a use of S to semantically express a proposition in a context C. Suppose further that speakers standardly intend to assert, and to be recognized as asserting, the proposition semantically expressed by the sentence assertively uttered, relative to the context of utterance. Then, a speaker will assertively utter S in C only when the truth of its expressive presupposition P can be taken for granted—only when P is either already part of the conversational record, or uncontroversial enough to be added by accommodation. Thus, expressive presupposition may be another source of pragmatic presuppositional requirements.20

19

Stalnaker (1973, 452). A related phenomenon involves cases in which the speaker has an object o in mind that he wants to say something about. In order to identify the object for his hearers, he may use a descriptive or demonstrative phrase that the conversational participants presume applies to o. For example, a speaker at a party might assertively utter (a) or (b) with the intention 20

90

•

Essay Two

Finally, proponents of pragmatic presupposition maintain that some presuppositional requirements arise from more straightforwardly pragmatic sources. For example, Lauri Karttunen and Stanley Peters (1979) argue that the pragmatic presuppositions of (18a) and (19a) are Gricean conventional implicatures that are carried by these sentences independently of the propositions they express. Particularly interesting from a pragmatic point of view are constructions which, though they don’t give rise to presuppositions themselves, allow compound sentences to inherit the presuppositions of their constituents. Two examples of such constructions are conjunctions and indicative conditionals.21 The presuppositional requirements of these constructions have been characterized as follows: A set R of common background assumptions satisfies the presuppositional requirements of ⎡A and B⎤ and ⎡If A, then B⎤ iff R satisfies the requirements of A, and the result R′ of adding the proposition expressed by A to R satisfies the presuppositional requirements of B. Roughly speaking, ⎡A and B⎤ and ⎡If A, then B⎤ inherit all the presuppositional requirements of A, plus those requirements of B that are not automatically satisfied by the addition of A to the conversational record. Different theorists have proposed different explanations of these requirements. Among the simplest is one suggested by Robert Stalnaker. Stalnaker’s proposed explanation is based on two principles. First, whenever something is asserted or supposed, it is immediately added to the conversational record (in the case of supposition, the addition is often

of asserting the singular proposition that he could have expressed by (impolitely) pointing at o and uttering (c). (a) The man in the corner drinking champagne is famous. (b) That man in the corner drinking champagne is famous. (c) He is famous. Such an utterance would presuppose that o is a man in the corner drinking champagne—i.e., it would indicate that the speaker regards that proposition to be either in the conversational record already, or evident enough to be added by accommodation. The important point about this proposition is not that it be true, but that the conversational participants accept it. If they do, then the speaker may succeed in saying something true about o even though the proposition presupposed by his utterance is false—because o is in fact drinking seltzer. It should be noted that the true proposition asserted by the speaker may not be semantically expressed by his sentence relative to the context of utterance. Speakers may assert many things instead of, or in addition to, the propositions semantically expressed by their utterances. One of the factors determining what they do assert seems to be the pragmatic presuppositions of their utterances. 21 Karttunen and Peters maintain that in addition to inheriting presuppositions from their constituents, indicative conditionals carry pragmatic presuppositions that their antecedents are not known to be false. In the interest of simplicity, I am ignoring this for present purposes.

Presupposition

•

91

temporary). Second, someone who assertively utters a conjunction asserts the proposition expressed by its first conjunct prior to uttering the second conjunct; someone who assertively utters a conditional posits the proposition expressed by its antecedent as a supposition prior to uttering the consequent. Since the initial conversational record is augmented with A prior to the utterance of B, this guarantees that the only substantive presuppositional requirements arising from B will be those that remain unsatisfied after the addition of the proposition expressed by A. In this way, Stalnaker attempts to provide a non-semantic explanation of the presuppositional requirements of these compound sentences. Whether or not this explanation proves to be correct, there is something both right and important about the pragmatic approach. Presupposition is, first and foremost, a matter of what is assumed or taken for granted. As such, linguistically expressed presuppositions should be described in terms of the beliefs and assumptions of language users. Although this makes presupposition a pragmatic notion, it does not rule out semantic explanations of pragmatic facts. What it does do is avoid conflating data about the commitments carried by various utterances with highly theoretical accounts of those commitments in terms of one or another kind of semantic presupposition. This theoretical neutrality makes the pragmatic approach extremely useful in investigating the scope of presupposition in natural language. In recent years, linguists and philosophers have used this approach in pursuing the following three part strategy: First, the scope of pragmatic presupposition is described by specifying both the natural language constructions that give rise to presuppositions and those that allow larger compounds to inherit the presuppositions of their constituents. Second, an inquiry is made into the sources of pragmatic presuppositions in particular cases. Third, an attempt is made to determine whether semantic presuppositions are needed to explain pragmatic ones; or whether purely pragmatic mechanisms suffice to account for the data without appeal to special semantic assumptions.

Recent Descriptive Work The Scope of Pragmatic Presupposition As we have seen, traditional discussions of presupposition by philosophers have tended to focus on a small range of presupposition creating and inheriting constructions—the prime examples of the former being singular terms and restricted quantifiers, the prime example of the latter being negation. Recent descriptive work has greatly expanded both classes of cases.

92

•

Essay Two

Among the constructions that give rise to pragmatic presuppositions one finds a large variety of different types in addition to those discussed by Frege and Strawson; for example: (20a) (P) (21a) (P) (22a) (P) (23a) (P) (24a) (P) (25a) (P) (26a) (P) (27a) (P)

Bill regrets that he lied to Mary. (Factive) Bill lied to Mary. Ivan has stopped beating his wife. (Aspectual) Ivan has beaten his wife. Harry managed to find the book. (Implicative Verb) Finding the book required some effort. Andy met with the PLO again today. (Iterative) Andy has met with the PLO before. It was in August that we left Connecticut. (Cleft) We left Connecticut sometime. What John broke was his typewriter. (Pseudo cleft) John broke something. Pat is leaving, too. (Focus on ‘Pat’) Someone other than Pat is leaving. Even Sam passed the test. Others, besides Sam, passed the test; and of those under consideration Sam was among the least likely to do so.

One also finds a variety of constructions that typically inherit the pragmatic presuppositions of their constituent clauses. These include, in addition to negation, epistemic modals, indicative conditionals, disjunctions, conjunctions, and sentences containing certain complementizable verbs. (28a) It wasn’t Jane who solved the problem. (28b) Maybe it was Jane who solved the problem. (28c) Either it was Jane who solved the problem, or they awarded the fellowship to the wrong person. (28d) If the problem was as important as they indicated, then it was probably Jane who solved it. (28e) That it was Jane who solved the problem isn’t very likely. One way of thinking about these sentences is to see the pragmatic presuppositions of the constituent clauses as projected onto, and hence inherited by, the larger sentences. In this respect, the sentences in (28) contrast with their counterparts in (29). (29a) Jane didn’t solve the problem. (29b) Maybe Jane solved the problem. (29c) Either Jane solved the problem, or they awarded the fellowship to the wrong person.

Presupposition

•

93

(29d) If the problem was as important as they indicated, then Jane probably solved it. (29e) That Jane solved the problem isn’t very likely. The difference between these two sets of sentences is that those in (28) have the cleft sentence (30) as a sentential constituent, whereas those in (29) have its noncleft counterpart, (31). (30) (31)

It was Jane who solved the problem. Jane solved the problem.

Although both (30) and (31) entail (32), only (30) presupposes it. (32) Someone solved the problem. Utterances of (28a–e) inherit this presupposition. Utterances of (29a–e) have nothing to inherit. This sort of projection is all but ubiquitous. However, it is not universal as was illustrated by (3a–c) in the section “Basic Questions” above. The chief descriptive problem occupying presupposition theorists in the past several years—the projection problem—has been to determine which utterances inherit the presuppositions of their constituents, which do not, and why. It is useful in approaching this problem to have a characterization of the pragmatic notion of an utterance presupposition. This is given below using the notion of the conversational background to represent the background information, common among speakers and hearers, against which utterances are evaluated. The Conversational Background: The conversational background at a time t is the set of propositions P such that at t the conversational participants believe or assume P; and recognize this about each other. Utterance Presupposition: An utterance U presupposes P iff one can reasonably infer from U that the speaker S accepts P and regards it as uncontroversial, either because (a) S thinks that P is already part of the conversational background at the time of U; or because (b) S thinks that the conversational participants are prepared to add P, without objection, to the background. The projection problem is the problem of determining the presuppositions of utterances of compound sentences in terms of presuppositions associated with their clausal constituents.

94

•

Essay Two

The Projection Problem Standardly, presuppositions associated with a sentential constituent become presuppositions of an utterance of a compound sentence of which the constituent is a part. There are, however, three factors that can prevent this from happening—cancellation of the presupposition by propositions in the conversational background, cancellation by Gricean conversational implicatures, and suspension of the presupposition by “local context incrementation” utilizing other clauses in the compound sentence. The first of these factors is illustrated by a discourse in which (33b) is uttered after (33a) (where the utterances may be by the same or different speakers). (33a) There is no king of France. (33b) Therefore the king of France isn’t in hiding. (33b) is the negation of a sentence that presupposes that France has a king. Although negations typically share the pragmatic presuppositions of their positive counterparts, this utterance of (33b) does not. The reason it doesn’t is that the putative presupposition conflicts with a proposition already placed in the conversational background. When this happens, utterances of negative sentences may be felicitous, but they do not inherit the presuppositions of their positive counterparts.22 The second way in which presupposition inheritance can be blocked is illustrated by the contrast between (34) and (35). (34) If I regret later that I haven’t told the truth, I will confess it to everyone. (A) I will regret later that I haven’t told the truth. (P) I haven’t told the truth. (35) If I realize later that I haven’t told the truth, I will confess it to everyone. (A) I will realize later that I haven’t told the truth. (P) I haven’t told the truth. In each case, the antecedent A presupposes P. Since utterances of indicative conditionals normally inherit the pragmatic presuppositions of their antecedents, one would expect utterances of these sentences to presuppose P as well. However, this is true only of (34). In the case of (35) the presupposition is blocked by the conversational implicature that the speaker doesn’t know the antecedent of his statement to be true. (If he did, he could have made a stronger statement.) Since he

22

See Gazdar (1979) for further discussion.

Presupposition

•

95

doesn’t know that he will later realize that he hasn’t told the truth, one may conclude (in normal circumstances) that he doesn’t now know that he hasn’t told the truth. In this way, the normal presumption that the speaker is taking the presupposition of the antecedent for granted is defeated by a conversational implicature that indicates that he cannot be doing so.23 The third way in which presupposition inheritance can be blocked is illustrated by (36). (36) If all the Smith brothers have children, then John Smith’s children will probably inherit the family fortune. (B) John Smith’s children will probably inherit the family fortune. (P) John Smith has children. Here, P is not presupposed by an utterance of (36) even though it is presupposed by its consequent B. There is also no conversational implicature to the effect that the speaker is not assuming P. Rather the utterance is noncommittal regarding his attitude toward P—it neither indicates that he takes it to be true, nor indicates that he does not take it to be true. This can be explained using assumptions (i) and (ii). (i) The pragmatic presuppositional requirement of B arising from its grammatical subject is that a certain contextually defined set of propositions entails P. (ii) When B occurs as the consequent of an indicative conditional the contextually relevant set is the one that results from adding the proposition expressed by the antecedent to the conversational background prior to the utterance. These assumptions guarantee that the relevant presuppositional requirement of B will be satisfied by an utterance of (36) no matter whether P is in the conversational background prior to the utterance or not. (We assume that it is part of the background that John Smith is one of the Smith brothers.) Hence, the utterance provides no indication whether the speaker regards P to be true, or whether he is unsure of its truthvalue. The three factors capable of preventing presupposition inheritance have provided the basis for three different theories of presupposition projection. The first of these, presented in Karttunen (1974), is based entirely on the kind of presupposition suspension illustrated by (36). The theory takes the form of an inductive definition of a two-place relation of admittance between sets of propositions (called “contexts”) and sentences. It is assumed

23

This point is made in Stalnaker (1974).

96

•

Essay Two

that a context C will admit (20a)–(27a) only if C entails (20P)–(27P), respectively. Clauses for compound sentences include the following: (37) A context C admits a negation of a sentence A iff C admits A. (38) A context C admits ⎡Maybe A⎤, ⎡It is likely that A⎤, ⎡It is possible that A⎤, iff C admits A. (39) A context C admits ⎡A and B⎤, ⎡If A, then B⎤ iff C admits A, and C ∪ [A] admits B.24 (40) A context C admits ⎡Either A or B⎤ iff C ∪ [−A] admits B, and C ∪ [−B] admits A, and for all propositions P, if P is entailed by every context that admits both A, and B, then C entails P.25 Pragmatic presuppositional requirements are consequences of the inductive definition plus a general requirement that the conversational background prior to an utterance admit the sentence uttered. In cases in which the background does not admit the sentence, the strategy of accommodation is invoked to allow for utterance presuppositions to introduce new information. This theory accounts for presupposition suspension both in simple case like (36) and in more complicated examples like (41). (41) If Martha buys a blue dress and Susan buys a blue dress too, then Martha will regret buying a dress that is the same color as one bought by Susan. However, it cannot be accepted as it stands. The basic problem, as shown in Gazdar (1979), Soames (1979), and Soames (1982), involves a conflict between conversational implicatures and presuppositional requirements generated by the admittance conditions. The conflict arises in cases in which the latter make a prediction that something is being taken for granted, which the former deny. One example of such a case is (35). The Karttunen theory predicts that (35) requires the conversational background to entail P, and hence that utterances of (35) commit the speaker to P. The theory of conversational implicature, on the other hand, tells us that (normal) utterances of (35) conversationally implicate

24

In stating these conditions, I am using [S] to stand for the set whose only member is the proposition expressed by S. 25 (40) is not the condition for disjunction given in Karttunen (1974), but rather a modification formulated in Soames (1979). See pp. 629–30, 636–40 (30–31, 38–42 of essay 1), and footnote 24 of the latter for further discussion.

Presupposition

•

97

that the speaker is not assuming P. In every such case, the conversational implicature is genuine and the presupposition is nonexistent.26 The ability of conversational implicatures to prevent presupposition inheritance was noted independently and used to develop an alternative approach to presupposition projection in Gazdar (1979) and Soames (1979). The basic ideas of the approach are the following: (i) Presupposition creating constructions like those in (20–27) give rise to “potential presuppositions” (illustrated by 20P–27P). (ii) Compound sentences inherit all the potential presuppositions of their constituents. (iii) If Q follows from potential presuppositions P1, . . . , Pn of S, then an utterance U of S presupposes Q unless

26

Further examples of this type are given in (i) and (ii).

(ia) (b)

Either Bill regrets voting for Reagan or he regrets not voting for Reagan. If Bill regrets voting for Reagan or he regrets not voting for Reagan, then he is probably unhappy. (c) It may be that either Bill regrets voting for Reagan or he regrets not voting for Reagan. (PRE:A) Bill voted for Reagan. (PRE:B) Bill didn’t vote for Reagan. (iia) If Mary’s boss doesn’t have children, then it wasn’t his child who won the fellowship. (A) Mary’s boss doesn’t have children. (Pre:A) Mary has a boss. (Pre:B) Mary has a boss and Mary’s boss has a child and someone won the fellowship. It follows from the definition of the Karttunen admittance relation that a context C admits (ia), (ib), or (ic) only if C entails the proposition expressed by (ia). Thus, the requirement that the conversational background prior to an utterance admit the sentence uttered incorrectly predicts that (ia), (ib), and (ic) pragmatically presuppose the proposition expressed by (ia). In fact, speakers who utter these sentences conversationally implicate that they are not presupposing this. In the case of (ii), the admittance conditions require the conversational background to entail (iii), which in this case is equivalent to (iv). (iii) (Pre: A & (A → Pre: B)) (iv) Mary’s boss has a child. Thus, the theory wrongly predicts that utterances of (iia) presuppose (iv), and fails to predict that they presuppose (v). (v) Someone won the fellowship. In fact what happens in this case is that the constituent presupposition (iv) is cancelled by a conversational implicature to the effect that the speaker doesn’t know the truth-value of the antecedent of his statement. See Soames (1979, 1982) and Gazdar (1979) for further discussion of these and other examples.

98

•

Essay Two

(a) P1, . . . , Pn are jointly incompatible with the conversational background; or (b) U conversationally implicates that the speaker is not taking P1, . . . , Pn for granted. In effect, utterances presuppose all the presuppositions of their constituents except those that are incompatible with the conversational background or cancelled by conversational implicatures. This approach handles much of the data that motivated the Karttunen theory while accounting for counterexamples like (35).27 However, there is a range of examples, specified in Soames (1979, 1982), that this approach cannot accommodate. Two such examples are (36) and (42). (42) Maybe Bill proved the theorem and Mary proved it too. (P) Someone other than Mary proved the theorem. In each case, an utterance of the sentence as a whole fails to inherit the constituent presupposition P, even though there is no cancelling conversational implicature. The reason for this seems to be that the presupposition is suspended by the kind of “local context incrementation” proposed by Karttunen. The upshot of this is that a proper theory of pragmatic presupposition projection must include both mechanisms for suspending presuppositions of the sort suggested by Karttunen and mechanisms for cancelling them of the kind suggested by Gazdar and Soames. This conclusion is drawn in Soames (1982), where two different methods of incorporating these mechanisms into a single theory are explored. According to one method, Karttunen-like devices are used to generate pragmatic presuppositional requirements of sentences that are cancellable by contextual and conversational means. According to the other method, cancellation first eliminates certain potential presuppositions, with the remaining uncancelled potential presuppositions providing the input for a computation of utterance presuppositions along Karttunen-like lines. Although the second method was ultimately selected in that paper, each method has its own advantages. The end result, though not entirely free of descriptive problems, extends the reach of descriptive theories of pragmatic presupposition to a significant range of data.28

27

Plus those mentioned in the previous note. One descriptive problem with the account given in Soames (1982) involves conditionals in which the presuppositions of the consequent (together with the common background) entail the antecedent, but the antecedent (together with the common background) does not entail the presuppositions of the consequent. An example of this kind, given in Heim (1983), is (i). 28

Presupposition

•

99

Foundational Implications of Descriptive Work cancellation and accommodation

Although contextual and conversational cancellation seem to be indispensable parts of an adequate descriptive account of pragmatic presupposition, they also appear to threaten the original conception of pragmatic presupposition outlined in the section “Pragmatic Presupposition” above. According to that conception, a theory of pragmatic presupposition can be thought of as a bipartite affair. The first part specifies the presuppositional requirements that various sentences place on conversational backgrounds. These requirements are illustrated by (37)–(40), and by requirements that (20P)–(27P) be entailed by the conversational backgrounds for (20a)–(27a). The second part of the theory specifies the role of presuppositional requirements in determining how conversational backgrounds are incremented on the basis of utterances. The crucial element here is the strategy of accommodating apparent violations. According to this strategy, an assertive utterance of a sentence S that pragmatically presupposes P will result in the addition of P to the conversational background unless the hearers object. The combination of this foundational conception of pragmatic presupposition, together with the descriptive requirements just mentioned, encapsulates the Karttunen theory discussed in the last section. Since that theory is descriptively inadequate, some change in either foundational structure or descriptive requirements is needed. The introduction of contextual and conversational cancellation into this picture worked out in Soames (1982) can be thought of as a change in the former. The basic idea is to introduce a new kind of accommodation to supplement the Lewis-Stalnaker variety that we have been considering up to now. The kind of accommodation discussed by Lewis and Stalnaker might be called “de facto accommodation.” When a presuppositional requirement is not met because the conversational background does not entail the presupposed proposition, the law does not change (the requirement remains

(i) If John has children, then Mary will not like his twins. The second method of combining cancellation with Karttunen-like inheritance conditions fails to predict that utterances of (i) intuitively presuppose (ii). (ii)

John has children → John has twins.

(See, however, Soames (1982, 502–4) for complicating factors.) Further descriptive problems are noted and dealt with in the section “Varieties of Presupposition: Unresolved Issues” below.

100

•

Essay Two

in force); rather, the conversational facts are adjusted to bring the speaker’s performance into line with the law (the proposition is added). However, this is not the only kind of accommodation possible. Suppose a speaker utters a sentence that requires the conversational background to satisfy a certain condition—say, to entail the proposition P. Suppose also that something about the speaker’s utterance makes it clear that the requirement is to be waived in this case. The hearers, recognizing this, will not add P to the background, but will go ahead with the process of incrementing it on the basis of the proposition asserted. This sort of accommodation can be called “de jure accommodation.” In this case, apparent violations of presuppositional requirements are accommodated not by adjusting the existing conversational facts to fit the requirements, but by adjusting the requirements to fit the facts. In effect, presuppositional requirements become defeasible. Unless there is an indication to the contrary, they remain in force. However, there are means available to cancel them in particular cases. The addition of this kind of accommodation to the foundational conception of pragmatic presupposition increases not only its descriptive accuracy, but also its intuitive plausibility. It must be remembered that incrementation of the conversational background is a complex process involving the nature of the preceding background, the syntactic form and semantic content of the sentence uttered, the presuppositional requirements of the sentence, plus the conversational implicatures of the utterance. Without de jure accommodation, the interaction of general principles involving various aspects of this process could easily lead to communicative conflicts. For example, suppose all sentences of a certain form require preceding backgrounds to contain some proposition P. Suppose further that in certain special cases the content of the sentence, plus Gricean conversational principles, and the conversational background, generate a conversational implicature to the effect that P is not being assumed. A pragmatic strategy allowing de facto, but not de jure, accommodation would give rise to a communicative impasse, since the addition of P to the background would be both required by the rules governing presupposition and prohibited by a conflicting conversational implicature. The existence of de jure accommodation allows speakers to avoid this kind of difficulty, while adhering to simple and general rules for determining presuppositional requirements. By allowing these requirements to be defeasible, one keeps the task of computing them manageable, with exceptions to one’s general rules being clearly recognizable in virtue of other, independently needed, pragmatic principles. Thus, there is good reason why pragmatic presuppositions ought to be governed by de jure, as well as de facto, accommodation.

Presupposition

•

101

the insufficiency of conventional implicature and logical presupposition as sources of pragmatic requirements

We now have the outlines of a theory of pragmatic presupposition that combines significant descriptive content with plausible answers to a number of basic foundational questions, including the following elaborations of the original Q5 of the section “Basic Questions.” (Q5a) (Answer)

(Q5b)

(Answer)

(Q5c) (Answer)

Why should there be linguistically expressed pragmatic presuppositions at all? To provide speakers with conventional means of indicating what assumptions they are making about the conversational backgrounds to which their utterances contribute. Why, given that there are such presuppositions, should conversational participants be ready to follow a policy of de facto accommodation when the presuppositional requirements of a sentence are not satisfied by the conversational background prior to the utterance? To allow speakers to use presuppositions to introduce new, but uncontroversial, information; and to avoid pointless objections involving propositions they are ready to accept. Why should pragmatic presuppositions be defeasible? To avoid communicative conflicts with other pragmatic implicatures; and to allow the rules for computing the presuppositional requirements of sentences to be kept manageably simple.

If one were designing a language for use in communication one would presumably want it to incorporate the main elements of this account. Thus, it is not surprising that natural languages do. This points up the explanatory attractiveness of the pragmatic theory. There is, however, a serious gap in this explanatory picture. Although we have explained why there ought to be pragmatic presuppositions in general, we have not explained why various sentences carry the particular presuppositional requirements they do. In some cases, the requirements seem to be simply matters of linguistic meaning. For example, in the case of the word ‘even’ it is plausible to hold that its meaning consists in the presuppositions, illustrated by (27P), that it introduces. A similar point might be made about the cleft construction, illustrated by (24a) and (30). If this is right, then what the meaning of the cleft construction adds to the meaning of (30) (that is not contained in the

102

•

Essay Two

meaning of its noncleft counterpart (31)) is a certain presuppositional requirement. Lauri Karttunen and Stanley Peters have suggested that these presuppositional requirements are Gricean conventional implicatures that are independent of the propositions expressed by sentences bearing them.29 This suggestion is generalized to cover the constructions in (20)–(27), as well as the presuppositions arising from definite descriptions, quantified phrases, negations, conditionals, conjunctions, and disjunctions.30 For the constructions covered in Karttunen’s 1974 theory, this amounts to taking the presuppositional requirements corresponding to (37)–(40) to be conventional implicatures arising from a recursive assignment of “non-truth-conditional” content to sentences that parallels the more familiar assignment of propositional content.31 Although the resulting system is elegant, it suffers from both explanatory and descriptive problems. On the explanatory side, pragmatic presuppositional requirements are reduced to arbitrary linguistic conventions associated with lexical items and constructions. This is plausible for examples like ‘even’ and clefts; but it is implausible for other cases, particularly the connectives. Surely, there is some connection between the truth conditional content of the connectives and the pragmatic presuppositions of sentences containing them.32 We are not inclined to think that learning their meanings consists of two separate and unrelated tasks; nor do we expect to find natural languages containing connectives that share the truth conditional contents of their English counterparts while differing arbitrarily from them in their contributions to presupposition inheritance. This suggests some explanatory link between pragmatic presuppositions, truth conditional content, and general principles governing communication—a link that is missing from the Karttunen-Peters account. There are also serious descriptive problems arising from the systematic identification of pragmatic presuppositions with conventional implicatures. As Grice, Karttunen, and Peters all have stressed, such implicatures are aspects of linguistic meaning, and are therefore uncancellable. Thus, the examples of presupposition cancellation cited in Gazdar (1979), 29

Karttunen and Peters (1979). Karttunen and Peters do not explicitly mention all the constructions that their theory is meant to apply to. For example, they do not mention ‘stop’, ‘regret’ (though they do explicitly include similar factives such as ‘realize’), or ‘again’ (though they do explicitly include ‘too’, and ‘also’). Nothing in the present discussion relies crucially on any items not explicitly treated by Karttunen and Peters. 31 Karttunen and Peters adopt a version of Montague semantics in which propositions are identified with functions from circumstances of evaluation to truth-values. 32 This point is made forcefully in Heim (1983). However, it is also present, in various forms, in Stalnaker (1974), Gazdar (1979), and Soames (1982). 30

Presupposition

•

103

Soames (1979), and Soames (1982) refute the Karttunen-Peters account, along with the original theory of Karttunen (1974).33 Nor is the problem resolvable by simply changing the content of the implicatures.34 Rather, it seems that some pragmatic presuppositions are not conventional implicatures after all. A corresponding point can be made about logical presupposition. Suppose one wanted systematically to derive the pragmatic presuppositions of sentences from their logical presuppositions (including (20)–(27), negations, conjunctions, conditionals, disjunctions, and so on). To do this, one would need a nonbivalent semantics incorporating the truth tables in (10), together with the bridge principle (43) ( justified by the Stalnaker argument given in the section “Pragmatic Presupposition”). (43) If S logically presupposes P relative to a context C, then an utterance of S in C pragmatically requires the conversational background to entail P. According to this analysis, the logical presuppositions of negations, conjunctions, conditionals, and disjunctions are given in (P)—where ⎡Pre: S⎤ expresses a proposition whose truth is a necessary and sufficient condition for the proposition expressed by S to be true or false. (Pa) (Pb) (Pc) (Pd)

Pre: Not S = Pre: S Pre: (A and B) = (Pre: A & (A → Pre: B)) Pre: (If A, then B) = (Pre: A & (A → Pre: B)) Pre: (A or B) = (A ∨ Pre: B) & (B ∨ Pre: A)

Since these presuppositions parallel those predicted in Karttunen (1974) and Karttunen and Peters (1979), the examples of presupposition cancellation that falsify those theories falsify the present analysis as well.35 It is

33 For example, Karttunen and Peters claim that the conventional implicatures (pragmatic presuppositions) of conjunctions, conditionals, and disjunctions are given in (i). (Where S is a sentence, ⎡Imp: S⎤ is an expression that represents the conventional implicatures of S, and ⎡Ex: S⎤ is an expression that represents its truth conditions.)

(ia) Imp: (A and B) = Imp: A & (Ex: A → Imp: B) (ib) Imp: (If A, then B) = Imp: A & (Ex: A → Imp: B) (ic) Imp: (A or B) = (Ex: A ∨ Imp: B) & (Ex: B ∨ Imp: A) & (Imp: A ∨ Imp: B) (See essay 1, 38–42) for discussion of (ic).) Applying these conditions to (35), as well as the sentences in note 26, results in the same incorrect predictions that were shown to falsify the Karttunen (1974) account. 34 See in particular essay 1, 43–53, and Soames (1982, 499–501). 35 If anything, the problem is worse, since (43) and (Pa–Pd) lead not only to false predictions about pragmatic presuppositions, but also to false claims about the truth conditions of examples like (33), (35), and the sentences in note 26.

104

•

Essay Two

shown in Soames (1979) that this problem cannot be solved by simply changing the nonbivalent truth tables of the connectives.36 The upshot of this is that pragmatic presuppositions cannot be systematically explained as arising from either logical presuppositions or conventional implicatures. We have not shown that no pragmatic presuppositions are logically presupposed, or conventionally implicated; only that, in each case, some are not. In one respect, this conclusion is not surprising. After all, pragmatic presuppositions were introduced, in part, to allow for radically different sources of presuppositional requirements. However, in another respect, the result is disquieting. As indicated earlier, there seem to be some expressions—including truth functional connectives—whose contributions to pragmatic presuppositions are not arbitrary matters of linguistic convention or pragmatic practice, but rather are linked in some way to their propositional content, plus general pragmatic principles governing cooperative communication. Until this link is spelled out precisely, we have no explanation of why various compound sentences bear the presuppositions they do.37 This explanatory difficulty is an important unsolved problem in the field that will be taken up again in the section “Discourse Semantics and the Explanation of Pragmatic Presuppositions” below. Before doing that, however, it is worthwhile to examine the more basic question of whether logical presuppositions ever provide the explanation of pragmatic ones. Is there any sentence S and proposition Q such that the reason that S pragmatically presupposes Q is that it logically presupposes Q?

In the case of (33), the standard response by proponents of logical presupposition has been to claim that negative sentences in natural language are lexically ambiguous between a reading in which negation preserves neither-truth-nor-falsity and a reading in which neither-truth-nor-falsity is mapped onto truth. This response, under increasing attack in Wilson (1975), Kempson (1975), Atlas (1977), and Gazdar (1979), is insufficient, if one’s goal is to derive pragmatic presuppositions from logical ones. For example, if the negation in the consequent of (iia) in note 26 is claimed to be the (logical) presupposition preserving kind, then the conditional is wrongly characterized as entailing the denial of its antecedent. However, if the negation is claimed to be the (logical) presupposition blocking kind, then one gives up any hope of using logical presuppositions to derive the pragmatic presuppositions that do arise from the consequent. In effect, avoiding incorrect predictions about truth conditions requires explaining a significant range of pragmatic presuppositions independently of logical ones. 36 Essay 1, 47–50. 37 Stalnaker (1974) and Soames (1982) suggest that the link is pragmatic. However, they do not spell out mechanisms that apply to the full range of cases that need to be explained.

Presupposition

•

105

Philosophical Foundations of Nonbivalent Analyses Why Truth-value Gaps Don’t Explain Presuppositions In order to derive pragmatic presuppositions from logical ones, a nonbivalent semantics plus the bridge principle (43) are needed. In order for such a derivation to have significant explanatory force, both the semantics and the bridge principle should be independently motivated. The point of the derivation is not simply to state the pragmatic requirements of various sentences, since (in most cases) we can do that directly. Rather, the point is to explain those requirements in terms of deeper semantic and pragmatic principles that are important for the explanation of other phenomena as well. Thus, if pragmatic presuppositions are to be ascribed to logical sources, we will need both a reason for abandoning bivalence and an account of (43) according to which it reflects something more than an arbitrary and unexplained correlation. The main attempt to give the latter is the one by Robert Stalnaker discussed in the section “Pragmatic Presupposition” above. According to him, the link between the logical presuppositions of sentences and their pragmatic presuppositional requirements is provided by the function of assertions in incrementing conversational backgrounds. Stalnaker notes that the propositions in the background determine a set of conversationally alternative possible worlds compatible with everything established or assumed in the conversation at a given point. The function of an assertion is to further constrain these alternatives by eliminating some of the worlds and retaining the rest. The ones to be eliminated are those that are incompatible with the proposition asserted. These, Stalnaker maintains, are those in which the proposition expressed by the sentence uttered is false. The ones to be retained are those in which the proposition is true. Note, if S logically presupposes (relative to the context) a proposition Q which is not entailed by the propositions in the conversational background, then the set of conversationally alternative worlds will contain some members w in which the proposition P, expressed by S, cannot be correctly characterized as either true or false. Stalnaker’s point is that in such a case the assertion will fail in its primary purpose of determining a new set of alternative possibilities, since it will fail, barring accommodation, to determine what should be done with worlds in the old alternative set in which Q is false. If this is right, it provides a straightforward explanation of why a sentence that logically presupposes Q (relative to a context) should give rise to a pragmatic requirement (subject to satisfaction via accommodation) that Q be entailed by the conversational background.

106

•

Essay Two

But is it right? Certainly, if Q is false in w, then there is no basis for retaining w, since retention is possible only if the world is one in which P is true. But if P is not true in a world w in which its presupposition fails, then, surely, w will be incompatible with P, and should be eliminated. For suppose that P is definitely not true in a world in which Q is false. Conversational participants who accept P should be in a position to remove the world from the set of alternative worlds among which the conversation has not decided. Since Stalnaker’s argument assumes that w isn’t eliminated on this basis, it requires a notion of radical presupposition in which a sentence or proposition with a false presupposition cannot be correctly characterized either as true or as not true. (This is implicit in his remark that a sentence with a false presupposition will fail to divide the set of conversationally alternative worlds into two exhaustive parts. Surely it would do so if the true and the not true were jointly exhaustive.)38 The idea, apparently, is that P cannot be correctly characterized either as being true in w or as being not true in w. Because of this, it fails to determine whether w should be retained or eliminated from the set of alternative possibilities among which it is the function of the assertion to discriminate. Thus, Stalnaker’s argument for (43) depends on the view that the relevant presuppositions constitute not just necessary conditions for their bearers to be correctly characterized as true or false, but rather, necessary conditions for their bearers to be correctly characterized either as true or as not true. Let us see if we can construct an example that exhibits these stronger conditions.39 Imagine the predicate ‘smidget’ being introduced into a language by the following semantic stipulation. Smidget: Stipulative Definition (44) (i) Any adult human being under three feet in height is a smidget. (ii) Any adult human being over four feet in height is not a smidget (or is such that it is not the case that he/she is a smidget). The stipulation consists of a sufficient condition for something to be a smidget and a sufficient condition for something not to be a smidget. On the basis of these conditions, there will be clear cases in which one is justified in characterizing someone as a smidget and clear cases in which 38 Stalnaker does not explicitly call for a notion of radical presupposition according to which presupposition failure makes it impossible to correctly characterize the bearer of the presupposition either as true or as not true. However, without this notion his argument loses its force. 39 The idea that there are predicates of the type illustrated by (44) was suggested to me by Nathan Salmon in a discussion of Kripke (1975).

Presupposition

•

107

one is justified in saying of someone that he is not a smidget. Moreover, these characterizations will convey information to other members of the linguistic community. An assertive utterance of ‘Jack is a smidget’ will convey to one’s hearers the information that Jack is an adult under three feet tall and an assertive utterance of ‘Jack is not a smidget’ will convey the information that Jack is an adult over four feet tall. In short, ‘smidget’ will enter the language as a useful and meaningful predicate. The interesting thing about the predicate is, of course, that the defining conditions for something to be a smidget, and for something to fail to be a smidget, are not jointly exhaustive. Adults between three and four feet tall cannot be correctly characterized either as being smidgets or as not being smidgets. The same point can be made in certain cases in which the predicate is introduced ostensively. Rather than recite a stipulative definition, one might point to a number of adults under three feet tall and say “These people are smidgets,” and a number of adults over four feet tall and say “These people are not smidgets” (or “It is not the case that these people are smidgets”). We may suppose that adults between three and four feet in height are extremely rare, and perhaps unheard-of in the linguistic community. Thus, the occasion may never arise during the period in which the term is being introduced to specify how such individuals are to be characterized. Language, being an institution designed to meet various practical contingencies, doesn’t require linguistic conventions to be framed in terms of all logically, or metaphysically, possible circumstances. Suppose now that the word ‘smidget’ has become entrenched in the language. We now consider the question of whether a man three feet six inches tall is or is not a smidget. This question might arise either because we finally encounter a man of that height, or because we wish to evaluate a counterfactual claim. Note, the question is not “Should the concept smidget be extended so as to include or exclude the individual in the relevant circumstance?”; rather it is, “Given the concept smidget as it already exists in the language, is the individual a smidget in the circumstance or is he not?” I suggest that our linguistic conventions provide no basis for answering this question. To characterize the man as not being a smidget would be just as unjustified as to characterize him as being a smidget. The concept is simply not designed for this case. As a result, neither (45a) nor (45b) can be accepted. (45a) He is a smidget. (Said referring to the 3′ 6′ man) (45b) He is not a smidget. (Said referring to the same man) (45c) That he is (isn’t) a smidget is true. (45d) That he is (isn’t) a smidget is not true.

108

•

Essay Two

If ‘true’ and ‘apply’ satisfy (i)–(iii), it will follow that (45c–d) cannot be accepted either. (In (i)–(iii) ⎡A iff B⎤ holds whenever A and B have the same status—both true, both false, both neither true nor false, or both such that they cannot be correctly characterized as true or not true.) (i) For any predicate P and term t, ⎡Pt⎤ is true (not true) iff P applies (doesn’t apply) to the referent of t. (ii) For any sentence S, ⎡—S⎤ is true (not true) iff S is not true (true). (iii) The predicate ‘red’ applies (doesn’t apply) to an object o iff o is (isn’t) red. The predicate ‘smidget’ applies (doesn’t apply) to an object o iff o is (isn’t) a smidget. . . . Under these assumptions, the radical partiality of ‘smidget’ results in the radical partiality of ‘apply’ and ‘true’.40 This suggests that there are sentences and propositions of the kind required by Stalnaker’s argument—sentences and propositions that cannot be correctly characterized as true or as not true unless certain conditions are met. What must be noted, however, is that the examples that have traditionally been analyzed as instances of logical presupposition are not of this kind.

40

This point can be reinforced by considering a case in which the stipulative definition of ‘smidget’ is expanded to include (Siii), and the clauses for ‘true’ and ‘apply’ are formulated using (i ′)–(iii ′). (Siii) (i ′) (ii ′) (iii ′)

For all x, if x is a smidget or is not a smidget, then x is an adult human being. For any predicate P and term t, ⎡Pt⎤ is true (false) iff P positively applies (negatively applies) to the referent of t. For any sentence S, ⎡−S⎤ is true (false) iff S is false (true). Any sentence that is false is not true. The predicate ‘red’ positively applies (negatively applies) to an object iff it is (isn’t) red. The predicate ‘smidget’ positively applies (negatively applies) to an object iff it is (isn’t) a smidget.

Given this expansion, one can correctly say of a child that it is not true that he is a smidget. (Since it is not true that the child is an adult human being, we can conclude from (Siii) that it is not true that the child is a smidget or is not a smidget, from which the result follows.) However, the point about adults between three and four feet tall remains. Neither the claim that they are smidgets nor the claim that it is not true that they are smidgets can be accepted on the basis of the definition. Rather, the definition provides a reason for rejecting both claims without asserting their negations.

Presupposition

•

109

For example, consider (46). (46) The king of France is cultured (whoever he may be). Although this example expresses a proposition, the proposition it expresses isn’t true. Since there is no king of France, it can’t be true. Unlike the smidget case, we don’t feel that it would be just as wrong to say that (46) is not true as to say that it is true. Nor will one get an argument about this from traditional defenders of logical presupposition. The standard claim about such examples is that the falsity of their logical presuppositions leads to their being neither true nor false. But if (46) is neither true nor false, then it is not true, and hence distinguished from (45). This means that Stalnaker’s argument does not apply to the cases for which it was intended. In fact, the situation is worse. Examples like (45), to which Stalnaker’s argument ostensibly applies, do not give rise to pragmatic presuppositional requirements on conversational backgrounds. If they did, then those requirements would be inherited in the normal way by larger presupposition inheriting constructions. However, examples like (47) do not bear the relevant pragmatic presuppositions. (47a) Maybe Bill’s uncle is a smidget. (47b) It is unlikely that Bill’s uncle is a smidget. (47c) If Bill’s uncle is a smidget, then he is probably entitled to special benefits from the government. Although utterances of these sentences presuppose that Bill has an uncle, they do not presuppose that Bill’s uncle is over four feet or under three feet tall. This suggests that Stalnaker’s argument is not only inapplicable to standard cases of logical presupposition, but unsound as well. Since the argument is unsound, no pragmatic presupposition can be explained by appeal to it. Thus, if any pragmatic presuppositions are to be derived from logical ones, a new argument is needed, which excludes smidget type cases. Let us introduce some terminology. Following traditional discussions of logical presupposition we will reserve the term ‘logical presupposition’ for examples that satisfy the following definition. Logical Presupposition: A proposition P logically presupposes a proposition Q iff (a) for all possible circumstances w, if P is true or false in w, then Q is true in w; and (b) for all possible circumstances w, if Q is not true in w, then P is neither true nor false in w (i.e., P is not true in w and P is not false in w).

110

•

Essay Two

Smidget type cases are instances of a different relation, in which the “presupposition” constrains acceptance or assertability rather than truth-values directly. Radical Presupposition: A proposition P radically presupposes a proposition Q iff (for all circumstances w) the proposition that P is true (in w) as well as the proposition that P is not true (in w) must be rejected if Q is not true (in w). A more general relation encompassing both logical presupposition and radical presupposition is the following: R-Presupposition: A proposition P R-presupposes a proposition Q iff (for all circumstances w) both the proposition that P is true (in w) and the proposition that P is false (in w) must be rejected if Q is not true (in w). Derivative relations holding between sentences, contexts, and propositions can be defined in the usual way in each of these cases. We have used the smidget examples to show that some radical presuppositions, and hence some R-presuppositions, are not pragmatically presupposed. Since R-presupposition is not enough to yield pragmatic presupposition, no pragmatic presupposition Q of S can be explained simply by claiming that S R-presupposes Q. If there is to be an argument explaining some pragmatic presuppositions in terms of logical ones, it must apply to logical presuppositions alone. The difficulty in finding such an argument is illustrated by the following attempt. Suppose a speaker asserts a proposition P in a conversation in which one of its logical presuppositions, Q, is potentially controversial. Since Q is potentially controversial, the hearers might reject it, and thereby reject P. However, they cannot do this by assertively uttering the negation of the speaker’s sentence, since on the preferred interpretation this sentence will express a proposition that commits them to Q. Thus if they are to reject P, they must say something more complicated—e.g., “Since it is not the case that Q, the proposition P cannot be accepted.” Now, it might seem that a cooperative speaker should avoid putting his hearers in the position of having to go to such lengths. That is, it might seem that he should assert a proposition P only if he thinks that any objection to P on the part of his hearers can be expressed simply by negating his remark. If this argument were correct, it would provide a rationale for the view that logical presuppositions give rise to pragmatic presuppositions. However, the argument cannot be correct. For the same line of reasoning applies with even greater force to smidget type cases in which P radically presupposes Q; and these don’t give rise to pragmatic presuppositions.

Presupposition

•

111

Results like these present a challenge to the idea that pragmatic presuppositions can ever be explained in terms of logical presuppositions. There simply is no known reason why language use ought to be governed by the bridge principle (43) (rather than by analogous principles covering radical presuppositions). Indeed, it is hard to see how any explanation could be forthcoming. If there is no such explanation, then the fact that S pragmatically presupposes Q can never be explained simply by citing the (alleged) fact that Q is a logical presupposition of S. This argument does not show that there is no such thing as logical presupposition; nor does it show that no pragmatic presuppositions are logically presupposed. Rather, it suggests that the connection between the two notions is at best indirect. If there are logical presuppositions, they must be motivated independently of pragmatic presuppositions. If pragmatic presuppositions are to be explained, their explanation can never rest simply on logical presupposition. Paradox, Partiality, and Presupposition The importance of the smidget example to the arguments just given might lead one to wonder whether there are nonartificial smidget type predicates in natural language. I believe there are. Many vague predicates, and predicates learned ostensively, share the kind of partial definedness that characterizes ‘smidget’. So, I believe, do natural language truth predicates. Indeed, it was Kripke’s theory of partially defined truth predicates that provided the model for ‘smidget’.41 An important feature of this approach to truth lies in its avoidance of the strengthened liar paradox. (48) Sentence (48) is not true. A proper treatment of the paradox should explain the characteristics of the truth predicate that provide the basis for rejecting both the claim that (48) is true and the claim that (48) is not true. (Where to reject these claims is to refuse to accept them, without, of course, asserting their negations.)42 This is what one gets if, following Kripke, one analyzes the truth predicate as partially defined in the manner of ‘smidget’.43 41

Kripke (1975). The idea of using smidget type examples to illuminate Kripke’s conception of truth originated with Nathan Salmon. 42 See Parsons (1984) for a discussion of relevant issues. 43 See in particular p. 701, where Kripke indicates his intention of capturing the intuition that ‘true’ can be explained to someone along the lines of (i) and (ii). (i) For any sentence S, one is entitled to assert that S is true in exactly those circumstances in which one is entitled to assert S. (ii) For any sentence S, one is entitled to assert that S is not true in exactly those circumstances in which one is entitled to assert the negation of S. (In the interest of

112

•

Essay Two

What hasn’t been sufficiently appreciated is that this kind of partiality is a fundamentally different phenomenon, semantically and pragmatically, from traditional examples of presupposition studied by philosophers and linguists.44 If the argument in the previous section is correct, analyses of semantic (and other) paradoxes in terms of partially defined predicates must be sharply distinguished from theories of logical presupposition, expressive presupposition, and pragmatic presupposition generally. Three Possible Deviations from Bivalence Think of a sentence obtaining a truth-value as a result of the following process: First, the sentence is placed in a context and a proposition is determined by applying its meaning to the context. (The meaning is thought of as a function from contexts to propositions.) Next a possible circumstance is selected and the proposition is evaluated for truth-value relative to it. This gives the truth-value of the sentence relative to the context and circumstance. There are three ways in which this process might lead one to reject a strong principle of bivalence for sentences. First, the meaning of the sentence might fail to be defined in the context in question, in which case the sentence will fail to express a proposition in the context. Second, the

simplicity, we equate a sentence’s being not true with its being false, and hence with its negation being true. The basic picture could be reconstructed so as to allow sentences that are not true and not false, but nevertheless grounded in Kripke’s sense—however, there is no need to do so here.) If ‘true’ is introduced in this way, then assertability conditions for claims to the effect that something is or is not true will standardly be grounded in assertability conditions of other claims, and ultimately in assertability conditions for sentences not involving truth at all. However, in “ungrounded” cases like (iii), the directions (i) and (ii) are silent. (iii) Sentence (iii) is true. Just as adults between three and four feet tall are not covered by the instructions governing ‘smidget’, so ungrounded examples like (iii) are not covered by (i) and (ii). 44 Even Kripke likens ungrounded sentences to sentences which, for Strawsonian reasons, fail to express propositions. I believe this to be a mistake. One can, I think, construct ungrounded sentences that fail to express propositions. However, the phenomenon of ungroundedness is independent of this. Thus, I take Kripke’s comments on pp. 699–700 to be a misstatement of the central philosophical insight underlying his analysis. Parsons (1984) is, I think, guilty of a different mistake. While correctly noting the radical partiality of the truth predicate on Kripke-like analyses, he wrongly assimilates traditional examples of logical presupposition to this kind of partiality. This is just the converse of the widespread error in truth-gap solutions to the paradoxes of assimilating what are in fact instances of radical partiality to instances of logical presupposition. These issues are discussed in greater detail in chapter 6 of Soames (1999).

Presupposition

•

113

proposition expressed might contain a partially defined property that makes determinate evaluation impossible at the given circumstance. Third, the process of evaluating the proposition at the circumstance might yield a value of neither truth nor falsity. The first of these ways corresponds to a failure of what I have called “expressive presupposition” and is definitely relevant to the study of pragmatic presupposition in natural language. The second way corresponds to smidget type cases, and is not. The third type of deviation is not as clear cut. If there are such cases, they can be used to characterize a notion of logical presupposition. However, it appears that this notion has no direct connection with presupposition in the primary sense of that which is taken for granted. With this in mind, I now turn to another attempt to explain the basis for such presuppositions.

Discourse Semantics and the Explanation of Pragmatic Presuppositions A Recent Proposal Why do various sentences bear the pragmatic presuppositions that they do? In the section “The Insufficiency of Conventional Implicature and Logical Presupposition as Sources of Pragmatic Requirements,” it was suggested that pragmatic presuppositions arising from certain lexical items, like ‘even’, and certain syntactic constructions, like clefts, could be viewed as non-truth-conditional aspects of the conventional meanings of these elements. However, it was also argued that the presupposition inheriting characteristics of connectives, including ‘and’, ‘or’, and ‘if, then’, cannot be treated in this way. Since these characteristics are neither arbitrary, nor the result of special semantic stipulation, it ought to be possible to link them to the semantic contents of the connectives, plus general pragmatic principles governing communication. We have seen that this link is not provided by three valued truth tables that give rise to logical presuppositions. Thus, some other explanation is needed. Recently, Irene Heim has attempted to provide such an explanation.45 The key innovation is the development of a semantics for entire discourses, rather than individual sentences. Since pragmatic presupposition is itself a discourse phenomenon, it is natural to think that it might be linked to such a semantics.

45

Heim (1982, 1983).

114

•

Essay Two

Heim’s semantic system is designed to assign (single) propositions to discourses, where propositions are taken to be sets of possible worlds and discourses are thought of as sequences of sentences. In this framework, the meaning of a sentence is given by a rule that determines its contribution to the propositions expressed by discourses containing it. These propositions are determined by starting with the set of all possible worlds and eliminating those incompatible with each of the sentences in turn. The end result is the “proposition” that the semantics assigns to the discourse as a whole. To begin with, we can think of semantic-contexts as sets of possible worlds. The semantics consists of a recursive definition of a function, +, from sentences and semantic-contexts to semantic-contexts. For example, the clauses in the definition for negation, conjunction, and material conditionals are given in (49). (49a) C + ⎡ A and B⎤=(C + A ) + B (49b) C + ⎡ Not A⎤=C\(C + A) (49c) C + ⎡ If A, then B⎤=C\((C + A)\((C + A ) + B)) (X\Y is the intersection of X with the complement of Y.) The discourse sensitivity of the semantics shows up in an analysis of indefinite noun phrases designed to handle discourses like (50). (50) Mary met a man. He was F. The problem illustrated by this example arises from two elementary observations. The first is that the initial sentence expresses an existential rather than a singular proposition. The second observation is that the pronoun in the second sentence is anaphoric with the indefinite NP in the initial sentence. This suggests that the semantic value of the pronoun in the context should be tied to that of the indefinite NP. If the two were in the same sentence, separated by ‘and’, this would be no problem for standard semantic accounts, since the pronoun could be seen as functioning as a variable bound by an existential quantifier. However, in (50) the pronoun and noun phrase are in different sentences. This creates a problem for standard systems in which variables can be bound within, but not across, sentences. It is this difficulty which motivates Heim to expand the scope of semantics from the single sentence to the entire discourse. Her treatment of discourses containing indefinite NP’s consists of several elements: First, the semantic representation (51) is assigned to the discourse (50). (51) xi, is a man, Mary met xi, xi was F. Next, the notion of a semantic-context is enriched to make sense of the contribution of an open sentence to a discourse.

Presupposition

•

115

Enriched Notion of a Semantic-Context: A semantic-context is a set of pairs, , such that g is an infinite sequence of individuals and w is a possible world. The Proposition Determined by a Semantic-Context: The proposition determined by a semantic-context C is the set of worlds w such that for some sequence of individuals g, is a member of C. Finally, the effect of incrementing a semantic-context with an open sentence is specified (informally) as follows. Incrementation of a Semantic-Context by an Open Sentence: C + Axi = the set of all in C such that g(i) is an A in w (where g(i) is the ith element of the sequence g). Applying this rule to the representation (51) of (50), we end up with a context that determines a proposition every member of which is a world in which Mary met a man who was F. Intuitively, what is happening is this: When one uses an indefinite description one sets aside a certain variable for the rest of the discourse. One stipulates, in the case of (50), that from now on the variable xi will be used only to express further constraints on men Mary met. This permanent setting aside of the variable means that constraints on these individuals can be built up piece by piece throughout the discourse. It is like letting the scope of an existential quantifier go across sentences, except the semantics is set up so that no explicit quantifier is needed. In order to make this work, one needs to be careful about introducing variables. Whenever variables are bound by quantifiers or introduced by an indefinite description they must be distinct from all previously used free variables in the discourse. If this convention is observed, a universally quantified sentence ⎡Every A is B⎤ can be represented as ⎡Every xi, Axi, Bxi⎤, with the semantics given in (52).46

46

In order to assign the right truth conditions, the system must be set up so that the input context C to (52) satisfies (i) and (ii). (i)

If the proposition P determined by C contains a world w such that for some object o in w, o “is an A” in w but “is not a B” in w, then for some g and j, g( j) = o and is a member of C. (C contains all relevant “witnesses” from the set of worlds it determines.) (ii) For any sequences g and g' that differ at most in their ith member, and for any world w, is a member of C iff could be true in any counterfactual circumstance (structure of situations). Finally, assigning the utterance no referent in the context is unacceptable, since it leads to the incorrect conclusion that x and y say the same thing in contexts in which, unknown to them, the respective victims (Smith and Brown) have not been murdered. 16 The problem would, of course, remain even if the contextually specified situation were allowed to be factual, rather than real or actual. However, the insistence on the latter further emphasizes the difficulty, since actual situations are potentially more inclusive (bigger) than merely factual situations (see note 1). As a result, there is not much hope of excluding material extraneous to the assertions.

Incomplete Definite Descriptions

•

341

This difficulty results (primarily) from taking single, real (or actual) situations (rather than individuals or properties) to be the entities that contextually supplement the contents of sentences containing incomplete attributive definite descriptions. An equally serious difficulty arises from the other main innovation of the Barwise-Perry analysis—namely, the decision to treat contextual supplementation not as augmenting the interpretations of (occurrences of) descriptions (and other subsentential constituents), but rather as associating the utterance as a whole with a new parameter required for its evaluation.17 The key problem resulting from this decision involves cases in which unsupplemented descriptions place demands on situations that conflict with those arising from other constituents in the sentence. When this happens, the interpretation of the sentence (relative to a context) will be a type that cannot be satisfied by any real (or actual) situation. Utterances of this sort will then be characterized as false, no matter what situations are selected as their contextually determined referents. In point of fact, however, many such utterances are true. Examples of this sort include the following:18 17 Though related, these two innovations of the Barwise-Perry analysis are distinct. One way of maintaining the first without the second would be to allow contextually determined situations to supplement the contents of incomplete descriptions. On this approach, an incomplete description ⎡the F⎤ can be thought of as containing indexical reference to a situation, in the manner of ⎡the F in that situation⎤. With the situation included as a constituent of the content of the description, the sentence as a whole can be evaluated in a total circumstance of evaluation, or possible world, as in the traditional account. For the most part, the problems illustrated above using utterances of (4) carry over to this approach as well (with the exception of the one noted in the penultimate paragraph of note 14). However, the approach does not fall prey to the refutation discussed below involving the sentences in (8a)−(8c). The opposite is true of approaches that maintain the second innovation while dropping the first. One way to construct such an approach would be to let the contextually determined “referent” of an utterance be a type, R, of situation (or a set of abstract situations), rather than a single real (or actual) situation. For example, in the case of x’s attributive utterance of (4), the “referent”, R, might be the type of situation (or set of abstract situations) in which someone or other murdered the victim, Smith. The resulting proposition could then be characterized as true in a circumstance (structure) iff the set of factual situations of the circumstance (structure) which were both of type R and of type I, was nonempty and persistent on the set of factual situations of the circumstance (i.e., contained all factual situations of the circumstance of which its members were parts). This modification would, of course, change the account significantly. However, as the examples in (8a)−(8c) will show, it is still fundamentally inadequate. 18 Examples like (8a) have received considerable attention in the literature on “discourse referents.” See, for example, Lewis (1982, 348−50), and McCawley (1981, 265− 66).

342

•

Essay Twelve

(8a) The cook is more experienced than the cook who prepared the main course. (8b) The cook’s father is a cook. (8c) Everyone is asleep and is being monitored by a research assistant. Each of these examples can be used attributively to express a truth in an appropriate context. A possible context for (8a) and (8b) might be one in which two cooks prepare the food for a party (at the same time and in the same kitchen)—one cooking the main course, the other the dessert. A guest who has no idea who cooked what might truly utter (8a), while munching some dessert. Another guest, agreeing with him, might truly utter (8b), on the basis of having heard earlier that the dessert chef is the son of a famous cook (working in the same kitchen preparing food for a second party). Although these examples are true and attributive, they do not fit the Barwise-Perry analysis. According to that analysis, the interpretation of (8a) in the context is the type of situation s such that (i) there is exactly one individual, o, who is a cook (at the relevant location) in s; (ii) there is exactly one individual, o′, who is a cook who prepared the main course (at the same location) in s; and (iii) o is more experienced than o′ in s. Given that the locations of the cooks in (i) and (ii) are the same, we see that o must be identical with o′. (I here exploit the fact that a cook who prepares the main course is a cook.) But then, since there is no real (or actual) situation in which an individual is more experienced than himself, there is no real (or actual) situation of the type required by the interpretation of (8a) in the context. Thus, Barwise and Perry will wrongly predict it to be false. Analogous predictions will be made regarding (8b) and (8c), based on the facts that no one is his own father, and no one who is sleeping monitors anyone, in any real (or actual) situation. The lesson here is that contextual supplementation works at the level of constituents of sentences or utterances, rather than the level of the sentences or utterances themselves. If contexts supplement the contents of these constituents (rather than introducing a new parameter of evaluation for entire utterances), then examples like those in (8a)−(8c) can easily be accommodated. Thus, the interpretation of ‘the cook’ in the context for (8a) will be an individual concept applying to the unique individual who cooked the dessert at the party—a concept also expressed by ‘the cook of this’ in a related context in which the demonstrative is accompanied by an appropriate demonstration. With this restriction of the

Incomplete Definite Descriptions

•

343

description, the utterance no longer imposes impossible requirements on circumstances of evaluation, and can easily be true.19 The problem for Barwise and Perry is that they make no provision for this kind of contextual supplementation. This is not to say that they never allow contexts to supplement the interpretations of descriptions. As I noted earlier, they do allow supplementation of descriptions by contextually determined spatiotemporal locations. Because of this they have no more (and also no less) trouble with examples like (9) than the traditional theorist does. (9)

The cook (here) is more experienced than the cook there.

Their trouble lies with cases in which the supplementation required is not a matter of location alone. Thus, in constructing my counterexamples, I was careful to rely on contexts in which contextually determined locations were not enough to provide the interpretations needed. In so doing I was following the practice of Barwise and Perry themselves. When motivating their own account, they were careful to construct cases in which supplementation of descriptions by contextually determined spatiotemporal locations did not “complete” the descriptions, or provide them with unique referents. It was precisely in order to handle such cases that Barwise and Perry introduced their own “Austinian” theory of utterance supplementation by real (or actual) situations. What we have seen is that this theory does not handle the range of cases for which it was designed. Barwise and Perry could, of course, expand their conception of the contextual supplementation of interpretations of incomplete definite descriptions to allow more than spatiotemporal locations to be provided. However, this would undercut their analysis of the very examples used to 19 It should be noticed that the examples in (8a)−(8c) which undermine the BarwisePerry analysis are of the same sort as those originally used to motivate it. Thus, in discussing the key example used to motivate the “Austinian” approach to utterance supplementation, Barwise and Perry say the following:

First, suppose someone comes up to me and says, “The food at this party is delicious! Who is the cook?” If I say, “I am the cook” I have clearly not described things accurately. I have claimed to be the person who did the cooking for the party [when in fact two people collaborated]. But suppose instead someone comes up to me eating a piece of my famous cheesecake pastry and says, “Who made this?” Then I may truly say that I am the cook. (1983, 159) Both here and in the examples in (8) the context can be seen as containing either an implicit or an explicit demonstration. Despite this, Barwise and Perry do not use the demonstration to restrict the interpretation of the description ‘the cook’. Instead, they invoke reference to a contextually determined situation by the utterance as a whole. The examples in (8a)−(8c) show that this approach will not work. Since these examples illustrate the same phenomenon as the original example, the “Austinian” approach must be rejected, even for the cases that motivated it.

344

•

Essay Twelve

motivate their theory, as well as undermine the claimed superiority of their account over its traditional rivals. Once the situation semanticist avails himself of the kinds of contextual supplementation required by the traditional theorist, he loses any evident basis for relativizing the truth-value of nonpersistent utterances to real (or actual) situations referred to. Such an utterance may, instead, be characterized as true iff the set of real (or actual) situations in its (contextually supplemented) interpretation is both nonempty and persistent on the set of real (or actual) situations (i.e., contains all real (or actual) situations of which its members are parts). Although this treatment of descriptions can be stated within situation semantics, it makes no use of the partiality of situations—treating it more as a bother to be circumvented than an asset to be utilized. For this reason it is not likely to appeal to the situation semanticist. Central to his program is the idea that a proper semantics for a variety of constructions requires switching from total truth-supporting circumstances (of the sort provided by possible worlds) to partial circumstances (of the kind given by situations). Initially it seemed that incomplete definite descriptions would be just the sort of expressions that would provide strong support for this idea. It now looks as if attributive descriptions do not support it at all. This does not mean that there can be no substantive role for situations of any sort in the analysis of attributive definite descriptions. It is possible that a role might be salvaged, if the individual concepts expressed by (some) incomplete descriptions are analyzed as containing contextually determined situations, in roughly the manner discussed in note 17. One might try to deal with the problems noted there by taking ⎡the F⎤ as equivalent not to ⎡the F in that situation⎤, but rather to ⎡the F relevant to that situation⎤—where the demonstrative refers to a contextually given situation. For example, an attributive interpretation of an utterance of (4), made upon discovering Smith’s body, could be seen as containing the information that the murderer relevant to a particular (perceptually given) situation s, containing the victim but not the murderer, is insane. Intuitively correct truth-conditions might then be forthcoming, on the assumption that in different possible circumstances different murderers are responsible for bringing s about. Whether or not such an approach can be made to work, and extended to a broad range of cases including those in note 7, is an open question. It should be noted, however, that the approach need not involve replacing possible worlds as circumstances of evaluation in a semantic theory, but rather can be seen as adding situations (of some sort) to the entities invoked by an essentially traditional theory.20 20 Although no full-fledged theory of this sort has been developed, suggestions along these lines are made in Lewis (1982, 348−50); Salmon (1981, 18 n. 16); and Salmon (1982, 42−43).

Incomplete Definite Descriptions

•

345

Beyond Situations and Attitudes In discussing the Barwise-Perry treatment of descriptions, I have used both the general semantic framework given in Barwise and Perry (1983) and the analysis of definite descriptions presented there. Barwise and Perry (1985) changes the framework. Although they do not reanalyze definite descriptions, the thrust of their more general changes suggests the possibility of such a reanalysis. It is worth looking briefly at this possibility to determine whether it holds out any reasonable prospect of reviving their “Austinian” theory of utterance supplementation, and enhancing their ability to handle incomplete definite descriptions. There are three basic changes in theoretical perspective introduced in Barwise and Perry (1985): (i) Types of situations are no longer identified with abstract situations, or sets of such; but rather are regarded as theoretical primitives to be thought of, roughly, as properties of real situations. The meaning of a sentence is seen as a relation which takes one from a type-of-situation in which it is uttered (a context) to the type-of-situation it describes in the context (its interpretation). This eliminates the need for abstract situations as basic entities of the theory. (ii) Propositions are introduced, apparently in two varieties. The proposition expressed by a persistent utterance U is thought of as claiming that the type-of-situation T which is the interpretation of U, is realized—i.e., that some real situation or other is of type T. The proposition expressed by a nonpersistent utterance U includes both the type T and a particular real situation, s, referred to by U; the proposition is taken to claim that s is of type T. As before, utterances containing attributive definite descriptions are thought of as nonpersistent. (iii) Higher-order properties are introduced in the interpretations of certain noun phrases, including ⎡an F⎤, ⎡no F⎤, and ⎡every F⎤. It is this third change that is most important for the issues at hand. In Barwise and Perry (1983), the interpretation of indefinite descriptions closely paralleled that of definite descriptions. Thus, the interpretation of ⎡an F⎤ was taken to be a relation between abstract situations, s, and individuals, o, who “are F” in s. In Soames (1985), I argued that this account was inadequate, and suggested that the interpretations of examples like those in (10a)−(10b) involve attributions of second-order properties to first-order properties.

346

•

Essay Twelve

(10a) There are F’s (10b) An F is G. In (10a) the property of being instantiated is attributed to the property F; in (10b) the relation of co-instantiation is applied to the properties F and G. In Barwise and Perry (1985) this suggestion is incorporated into the revised version of situation semantics.21 Barwise and Perry illustrate the new analysis with the following, complex example. (11) ‘Hesperus’ referred to a heavenly bodyi and ‘Phosphorus’ referred to iti too.22 The interpretation of (11) is the type of situation s such that in s; at p: co-instantiated, [x], [y]; yes where p is a location that temporally precedes the location of utterance, and [x] is the type-of-object x such that in s; at p: heavenly body, x; yes and [y] is the type-of-object y such that in s: at p: refers to, ‘Hesperus’, y; yes in s: at p: refers to, ‘Phosphorus’, y; yes Types-of-objects are theoretical constructs that play the role of (complex) properties of objects. For two such types to be co-instantiated is for there to be an object of which both are types. Thus, (11), which is seen as saying that there is a real situation in which [x] and [y] are coinstantiated, gets assigned the correct truth-conditions. Barwise and Perry extend this “higher-order” analysis to ‘no’ and ‘every’, which are taken to stand for the disjointness and inclusion relations on types (see Barwise and Perry (1985), 146−47). Thus, (13b) and (14b) take their places alongside (12b) as interpretations of the corresponding (a) sentences—where [x] is the type-of-object x such that x is F in s at l ([x| in s: at l, F, x; yes]), and [y] is the type-of-object y such that y is G in s at l ([y| in s: at l, G, y; yes]). 21 The point in Soames (1985) was not just that examples like those in (10a) and (10b) require a higher-order analysis, but that this change motivates an alternative semantic framework in which the semantic contents of sentences are structured, Russellian propositions, rather than types of truth-supporting circumstances. Barwise and Perry adopted the higherorder analysis, without accepting the alternative semantic framework. More on this below. 22 The subscripts indicate that the indefinite description is to be understood as the antecedent of the pronoun. For more on this example see Soames (1985, n. 6); and Barwise and Perry (1985, 151−58).

Incomplete Definite Descriptions

(12a) (12b) (13a) (13b) (14a) (14b)

•

347

An F is G. [s| in s; at l, co-instantiated, [x], [y]; yes] No F is G. [s| in s: at l, disjoint, [x], [y]; yes] Every F is G. [s| in s: at l, included, [x], [y]; yes]

Moreover, the analysis cannot stop here. As I have shown (Soames 1988, 1989), the problems that force the higher-order analysis of indefinite descriptions apply with even greater force when definite descriptions are involved. Thus, the analysis of these expressions given in Barwise and Perry (1983) must be replaced with a higher-order account. Although Barwise and Perry (1985) do not say anything about this, it seems clear that something along the lines of (15a) and (15b) is needed. (15a) The F is G. (15b) [s| in s: at l, co-instantiated, [x], [y]: yes uniquely instantiated, [x]; yes] In these examples, [x] and [y] represent the properties of being an F, and being a G, in s. For example, the interpretation of (13a) is the type of situation s in which the property of being an F in s bears the disjointness relation to the property of being a G in s. A real situation, r, is of this type only if it contains these properties as constituents, and they bear the disjointness relation (at the location in question). Of course, if the two properties really are disjoint (at the location), then there can be no individual who is both an F and a G (at the location) in r—i.e., there can be no individual o such that in r: at l, F, o; yes G, o; yes. However, if r is a real situation of type (13b), there may still be a real situation, r*, of which r is a part, in which some individual is both an F and a G at the relevant location. (16) in r*: at l, disjoint, [x| in r: at l, F, x; yes], [y| in r: at l, G, y; yes]; yes in r*: at l, F, o; yes in r*: at l, G, o; yes The reason this is possible is that there is no conflict between (i) the existence of an individual that is both an F and a G in some part of reality; and (ii) the disjointness of the properties of being F-in-a-certain-smallerpart-of-reality, and of being G-in-that-part-of-reality.

348

•

Essay Twelve

Thus, a real situation may be of the type (13b) even if larger situations containing it are not.23 On this analysis, utterances of (13a), (14a), and (15a) remain nonpersistent in a straightforward sense; their interpretations are types which may hold in parts of reality while failing to hold in larger, containing parts. Thus, they pose the same problems for theories of truth and propositional content that they did under the earlier analysis of Barwise and Perry (1983). In particular, one cannot maintain that an utterance of (15a) is true iff its interpretation is realized by some real situation or other, for to do so would be to fail to capture the uniqueness claim associated with the definite article. To avoid this, Barwise and Perry appeal to the “Austinian” theory of utterance supplementation, exactly as before. Because of this, exactly the same problems arise as did earlier. For example, under the new analysis, the interpretation of an utterance of (8b) The cook’s father is a cook must be something along the lines of (17), where [x] and [y] are: [x| in s: at l, cook, x; yes] [y| in s: at l, father-of, y, x; yes cook, y; yes] (17) [s| in s: at l, uniquely instantiated, [x]; yes co-instantiated, [x], [y]; yes] As before, this type can be realized by some real situation only if someone is his own father. Thus, Barwise and Perry wrongly predict that it cannot be true. Next consider an attributive utterance of (4) made by Mary upon discovering Smith’s body. (4) The murderer is insane. Under the new analysis, the interpretation of Mary’s utterance is the type,

23 This account is an interpretation of the very sketchy remarks by Barwise and Perry (1985). It is supported in particular by their continued inclination to treat examples like those in (13)−(15) as nonpersistent, and by the remark on p. 144 that they now want to allow the possibility that a situation may be of a given type even though not all situations of which it is a part are of that type. An alternative (less likely) interpretation, which does not have these features, is discussed below.

Incomplete Definite Descriptions

•

349

(18) [s| in s: at l, uniquely instantiated, [x]; yes co-instantiated, [x], [y]; yes] where [x] and [y] are: [x| in s: at l, murderer, x; yes] [y| in s: at l, insane, y; yes] In order for the utterance to be true, it must refer to some real situation r of this type. Intuitively, this means that the property of being a murderer (at l) in r and the property of being insane (at l) in r must bear (in r) the higher-order properties and relations specified in (18). This can be the case only if r also contains a unique individual o such that for some individual o', o is a murderer of o' (at l) in r, and o is insane (at l) in r. Let us suppose that Smyth really is the (unique) murderer of Smith (though Mary does not know this), and that Mary’s utterance is true. Then the individuals o and o' in r must be Smyth and Smith, respectively. Since the proposition expressed by Mary’s utterance is, on the Barwise-Perry analysis, a complex consisting of the type (18) together with the situation r, this means that Smyth and Smith are equally constituents of Mary’s assertion. But this, as we have seen, is wrong. There is, I believe, no way to avoid these problems short of abandoning the “Austinian” theory of utterance supplementation in favor of the more traditional account. This point can be illustrated by considering a particularly simple technique for dealing with the problem posed by attributive utterances of (4). Up to now, I have assumed that Barwise and Perry intend the first-order properties of objects that are the arguments of higher-order properties in the interpretations (12b)−(15b) to be indexed for particular situations. On this assumption, r will be of type (15b) only if in r, the property of being an F (at l) in r, and the property of being a G (at l) in r, bear the appropriate higher-order properties and relations. Suppose we drop this assumption of indexing. Then what is required for r to be of the appropriate type is for the property of being an F (at l) to be uniquely instantiated in r, and for the properties of being an F (at l), and being a G (at l), to be co-instantiated in r. With this amendment, it is possible for a real situation to be of type (15b) without itself containing any individual that is F or G. Thus, in the example involving Mary’s utterance of (4), we no longer are forced to include Smyth in the proposition asserted. However, this is no victory for the Barwise-Perry treatment of incomplete descriptions. For if the unindexed property of being an F is uniquely instantiated (in any real situation), then there must be exactly one individual o for whom there exists a real situation r' such that o is

350

•

Essay Twelve

F in r′—i.e., if the property of being F really is uniquely instantiated, then there must be just one individual who is F, in reality as a whole.24 Thus, under the amended analysis, utterances of (15a) are persistent, and the original motivation for the “Austinian” theory of utterance supplementation is eliminated. (Analogous points hold for (13a) and (14a).) Moreover, any semantically significant contextual supplementation of “incomplete” NP’s, ⎡the F⎤, ⎡no F⎤, and ⎡every F⎤, will have to proceed by augmenting the operand, F, exactly as in the traditional approach. This result does not depend on abandoning indexing. What the problematic examples show is that the Barwise-Perry strategy of contextually supplementing attributive uses of incomplete definite descriptions (and other NP’s) will not work. Unlike the traditional theorist, who appeals to contextually given objects to complete the contents of incomplete descriptions (and other NP’s), Barwise and Perry leave the contents (interpretations) of subsentential constituents unsupplemented.25 In place of such supplementation, they add a contextually determined situation, referred to by the utterance as a whole, to the proposition expressed. It is precisely this that causes the problems. Thus, Barwise and Perry have no choice but to modify their account to allow for contextual supplementation of the traditional sort. For example, what is needed in the case of (8b) and (4) is for the types [x| in s: at l, cook, x; yes] [x| in s: at l, murderer, x; yes] corresponding to the operands of the descriptions to be replaced by the contextually augmented types:26 [x| in s: at l, cook, x, o′; yes] (where o′ is the contextually indicated food) [x| in s: at l, murderer-of, x, Smith; yes] (where Smith is the indicated victim) However, once this kind of contextual augmentation of interpretations is made generally available, the stated rationale for the “Austinian” theory of utterance supplementation evaporates. 24 Barwise and Perry assume that for any two real situations, s and s′, there is a real situation, s′′, of which both are parts. Thus if there were a real situation in which the property of being F had the property of being uniquely instantiated, and also real situations in which different objects were F, then there would be a real situation combining these. But such a situation is impossible, and so cannot be real. 25 Except for reference to contextually determined spatio-temporal locations. 26 Contextually augmented types incorporating the analysis of ⎡the F⎤ as ⎡the F relevant to that situation⎤ (discussed at the end of the previous section) could also be used here.

Incomplete Definite Descriptions

•

351

That theory was developed to avoid the following dilemma: If utterances of (13a), (14a), and (15a) are potentially nonpersistent, then to characterize them as true iff their interpretations are realized by some real situation or other is to assign them truth-conditions that are too weak. On the other hand, if interpretations of these utterances are not contextually augmented, in the manner of the traditional approach, then to characterize them as true iff their interpretations are realized in reality as a whole is to assign them truth-conditions that are too strong. The “Austinian” theory was designed by Barwise and Perry to chart a middle course. The failure of the theory suggests an analysis in which these utterances are true iff they have contextually augmented interpretations that are both realized and persistent. What this amounts to is simply the traditional theory, encoded in the (revised) framework of situation semantics. The central idea of the framework—namely, the partiality of the basic semantic objects—plays no role in the account. Thus, attributive descriptions, and other general NP’s, do not support the idea that a proper semantics requires truthsupporting circumstances that are partial, in the manner of situations.

Attributive/Referential This conclusion about attributive descriptions has implications for the analysis of referential uses as well. Once the interpretations of incomplete definite descriptions are allowed to be contextually augmented there is no longer such an evident need for partial “resource situations” in which to evaluate them. We have already seen that an attributive utterance of the description in (4), made upon discovering Smith’s body, should be analyzed not as expressing the concept unique murderer, but rather as expressing the concept unique murder of o (where o is Smith).27 (4) The murderer is insane. A unified treatment of descriptions will also assign this concept to referential uses of the description—for example, to utterances in which the murderer is present at the discovery of the body and the speaker wishes to assert of him that he is insane. If these different uses of (4) represent a genuine semantic ambiguity, then the proposition semantically expressed by the referential interpretation is the proposition that i is insane, where i is the unique individual who satisfies the contextually augmented description in the circumstance of the context—i.e., the unique individual who murdered Smith. 27

Or unique murderer relevant to s (where s is a perceptually given situation containing o).

352

•

Essay Twelve

The point to notice is that there is no need for the circumstance of the context to be partial in order for the description to determine an object. If the individual concept semantically associated with the description in the context were simply that of being a unique murder, then the referential interpretation would require a partial circumstance. However, the contextual augmentation of the concept eliminates this need, and allows a semantically referential interpretation to be defined (if such an interpretation is desired) using the world as a whole. Thus, semantically referential interpretations do not require partial circumstances of evaluation.28 Moreover, it is not obvious that there really are such interpretations. There are, of course, referential uses of descriptions of the sort noted by Donnellan (1966). However, these uses cannot be identified with semantic interpretations. Imagine, for example, a speaker using the description in (19) to identify a certain woman w he has in mind. (19) The woman next to Jones drinking champagne is a famous philosopher. Donnellan observes that in such a case it may be correct to characterize the speaker as having truly said of w that she is a famous philosopher— something that could also have been said by uttering ‘She is a famous philosopher’ with appropriate contextual indication of w as the referent of ‘she’. Donnellan goes on to point out that this characterization may be correct even if w is, in fact, drinking sparkling water rather than champagne. The reason it may be correct is that the description may successfully identify the person whom the speaker has in mind even when it fails to denote her semantically. The mechanism that allows this is discussed by Kripke in his commentary on Donnellan’s original paper (Kripke 1979, 14−15). Two people see Smith in the distance and mistake him for Jones. They have a brief colloquy: “What is Jones doing?” “Raking the leaves.” “Jones,” in the common language of both, is a name of Jones: it never names Smith. Yet, in some sense, on this occasion, clearly both participants in the dialogue have referred to Smith, and the second participant has said something true about the man he referred to if and only if Smith was raking the leaves (whether or not Jones was). How can we account for this? Suppose a speaker takes it that a certain object a fulfills the conditions for being the semantic referent of a designator, “d”. Then, wishing to say something about a, he uses “d” to speak about a; say he says “φ(d)”. Then, he said, of a, on that occasion, that it φ’d; in the appropriate Gricean sense (explicated above), he meant that a φ’d. 28

This example is a particularly clear case in which contextual augmentation eliminates the need for partiality. However, it is meant to illustrate the admittedly programmatic hypothesis that this sort of trade-off will hold generally.

Incomplete Definite Descriptions

•

353

This is true even if a is not really the semantic referent of “d”. If it is not, then that a φ’s is included in what he meant (on that occasion), but not in the meaning of his words (on that occasion). There are three significant features of this account. First, it shows that the phenomenon of referential use is not limited to descriptions, but extends to other expressions, such as proper names, for which no relevant semantic ambiguity exists.29 Second, it explains referential use in terms of a general process in which semantic information, the speaker’s intentions, and background information are combined. Third, it accounts for how a person using the description in (19) might say something true about someone, w, even if the description fails to denote w. The reason this is possible is not that the description has a special semantics invoked for the occasion, but that there is a contextually recognized presumption that w fits the standard semantics of the description,30 and that the speaker intends to exploit this presumption.31 29 Although Barwise and Perry posit a referential/attributive ambiguity in the semantics of definite descriptions, they do not do so for proper names. According to them, each proper name n has the property expressed by ⎡being an n⎤ semantically associated with it. Although this property is not part of the interpretation of the name, it is supposed to constrain its reference. Barwise and Perry indicate that the property is to be understood metalinguistically, as something like having the name n, or being named n (1983, 166−67). Exactly what it is to have the name n, or to be named n, is not explained. However, they seem to have in mind some substantial, sociolinguistic condition. For one thing, the property of having the name n seems, on their account, not to be indexical—I do not have the name n in one context simply in virtue of the fact you happen to call me n in that context, while failing to have it in other contexts. For another thing, Barwise and Perry regard an utterance of ⎡He is n⎤ as providing substantive metalinguistic information to the hearer, in virtue of the fact that n is semantically associated with the property of having the name n (which constrains its reference). But this would be impossible if the property of having the name n relevant to a particular utterance of n were just the property of being the referent of that utterance. Finally, the examples given by Barwise and Perry involve an individual’s having his given name, as well as certain special cases like ‘Wednesday’ and ‘Christmas’. All of this seems to point to a view in which having the name n is a matter of being associated with the name by some established convention of the linguistic community. One question that arises for this view is how to account for private nicknames and temporary names—“Let’s call him ‘Bozo’,” or “Get lost Bozo.” But whatever is said about this, the analysis does seem to have the right consequences for Kripke’s Smith/Jones example. When the speaker sees Smith in the distance and mistakes him for Jones, his use of ‘Jones’ semantically refers to Jones, not Smith. Moreover, although the speaker may succeed in saying something true about Smith, his words say something false about Jones. Thus, whatever may be the case for definite descriptions, Barwise and Perry need something like Kripke’s pragmatic mechanism to account for referential uses of proper names. 30 As Donnellan points out, the speaker does not actually have to believe the presumption, so long as he recognizes that the conversational participants are willing to accept it, if only “for the sake of argument.” 31 If Kripke’s pragmatic account is accurate, then the speaker may have asserted more than one thing—the (false) proposition semantically expressed by his sentence in the context, and the true proposition determined by the pragmatic mechanism.

354

•

Essay Twelve

This pragmatic account poses a dilemma for those who see referential uses of definite descriptions as reflexes of semantically referential interpretations. To insist that the speaker’s assertion that w is a famous philosopher reflects the semantic content of (19), even in contexts in which w does not satisfy the description, is to ignore the distinction between semantic and pragmatic information, and to invite the multiplication of unmotivated semantic ambiguities, not only for descriptions, but also for names and other expressions.32 However, to posit semantically referential interpretations only for cases in which the object fits the description is to complicate the semantics in a way that may be unnecessary. Surely, any pragmatic account capable of explaining the speaker’s referential assertion in cases in which the description (or name) does not fit the object talked about will apply equally well to cases in which it does. Thus, if semantically referential interpretations are to be defended, more evidence than has so far been presented will have to be given. Barwise and Perry do not provide it. In Barwise and Perry (1983), they are more concerned with articulating a framework for stating referential interpretations than with marshalling evidence for them. The switch from abstract to real situations in Barwise and Perry (1985) seems to entail a restriction of referential interpretations of descriptions to cases in which the description really applies to the object talked about—for if the resource situation is real, the constituent of it that satisfies the description must really have the properties required by the description. The question that remains unanswered is why such interpretations are needed, given that referential uses of names and descriptions that do not semantically apply to the intended individual must be treated pragmatically.33 32 Kripke maintains that even quantified statements in a purely Russellian language can have referential uses (1979, 16−18). 33 One piece of evidence for semantically referential interpretations given in Barwise and Perry (1983, 147), involves the following example:

3.

The dog growled at the rabbit that sneezed. Suppose Jim has been telling us about a situation. He has mentioned only one dog, Clarissa, and two rabbits: Hugh, who sneezed while eating Clarissa’s food, and Fang. We would naturally take Jim’s utterance of (3) to describe the event e: in e: at l: growling at, Clarissa, Hugh; yes where l is the location to which Jim was referring, a location temporally preceding that of the utterance. The definite descriptions contribute Clarissa and Hugh to the event described. Two factors enable them to do this: 1. the properties various parts of the expression designate; 2. the unique possession of those properties by Clarissa and Hugh in the situation built up by Jim’s discourse.

In this example, Jim’s discourse is taken to provide a resource situation for semantically referential interpretations of the descriptions in (3). Suppose, however, that Jim’s story has

Incomplete Definite Descriptions

•

355

Conclusion In light of all this, it is reasonable to conclude that the semantics of definite descriptions do not call for partial circumstances of evaluation. This does not mean, of course, that they cannot be treated in a revised framework of situation semantics. However, it does mean that they fail to provide support for the central tenet of the program—namely, that a proper account of semantic information requires total circumstances of evaluation to be replaced by partial situations. It is, of course, true that partial circumstances can be used to construct fine-grained semantic contents that are better candidates for objects of propositional attitudes than those provided by standard possible worlds accounts. However, this is also true of the semantic contents provided by other theories—in particular theories that identify the contents of sentences (in contexts), with structured, Russellian propositions (see Salmon 1986a, 1989; Soames 1985, 1988, 1989). Such propositions are complexes in which the contents of subsentential constituents are combined in a structure closely related to that of the sentences that express them. These propositions are not themselves circumstances of evaluation, but rather determine, without being determined by, sets of truth-supporting circumstances. In earlier work (Soames 1985, 1988, 1989) I argued that this conception of semantic content is superior to those developed using either partial or total truth-supporting circumstances. For example, I argued (using auxiliary assumptions common to the Russellian and situational frameworks) that the semantic contents of the following (a) and (b) examples

been systematically false. In particular, suppose that Clarissa is really a wolf and that Hugh did not sneeze (also suppose that no other dog or sneezing rabbit is talked about). Then the discourse will not provide any real resource situation in which the descriptions are semantically defined. Thus, (3) will not semantically express the proposition that Clarissa growled at Hugh. However, the speaker may have asserted this. Moreover, all of this could occur without the speaker or his audience realizing that the story was false. But then, since pragmatic mechanisms explain the speaker’s referential assertion in this case, why is a special semantic interpretation needed when the story happens to turn out true? In fact, the problem goes much deeper. For many cases of this sort, in which a description is used in a discourse, no singular proposition is either asserted or expressed. For example, suppose I say: “A student came to my office before I arrived this morning and removed a book. I noticed something was wrong because when I got there the door was ajar, and the janitor said that he had not been in. What really bothers me is that the student must have a key to the building.” I might say this having no idea about the identity of the intruder (but just thinking that whoever it is must be a student). In such a case there is no individual about whom I am talking, no assertion that he (or she) came to my office, removed a book, and has a key to the building, and no singular proposition toward which I bear a propositional attitude (the propositional attitude of being bothered that) if my remark turns out to be true.

356

•

Essay Twelve

must be distinguished, despite the fact that, intuitively, they are true in the same circumstances.34 (20a) (20b) (21a) (21b) (22a) (22b)

t is F and t is G t is F and t is G and something is such that it is F and it is G t is an F and t is G t is an F and t is G and an F is G t = the F and t = the G t = the F and t = the G and the F = the G

As Salmon (1986b) points out, the arguments involving these examples can be extended to show that the (a) and (b) sentences in (23) and (24) must also have different contents. (23a) (23b) (24a) (24b)

tRt t R itself (i.e., t self-R’s, or [λx(xRx)]t) t is F and t is G t is F and G (i.e., [λx(F(x) & G(x))]t)

In each case, syntactic differences in sentences are paralleled by semantic differences in their contents. Accordingly, I see these results as confirming the basic intuition behind the structured propositions approach: Semantic contents of sentences are not circumstances of evaluation, but rather are “syntactically” structured complexes which themselves undergo evaluation. However, situation semanticists see things differently. While conceding that these examples must be assigned different contents, Barwise and Perry will take them as indicating that the initial conception of situations must be modified to allow the necessary distinctions. In particular, the principles in (25) will be given up in favor of their counterparts in (26).35 34

In these examples, ‘t’ stands in for a proper name, demonstrative, pronoun, or variable. Barwise and Perry (1985) explicitly give up (25a, b) and adopt (26a, b), on the basis of arguments involving (20) and (21). Since the arguments involving (22)−(24) are exactly analogous, I am confident that they would give up (25c−e) and adopt (26c−e) as well. (The arguments involving (20)−(22) utilize the auxiliary assumption that a situation (circumstance) that supports the truth of each conjunct supports the truth of a conjunction, and vice versa.) In fact, Perry has accepted the argument involving definite descriptions in his comment on Soames (1989), given at the Pacific Division Meetings of the American Philosophical Association, March 1985. That argument, it turns out, is based on an example slightly more complex than (22). 35

(i) t = the x:Fx & t = the x:Gx & the x:Fx = the x:Fx & t = the x:t = x (ii) (i) & the x:Fx = the x:Gx (Any two-place relation can be used in these examples in place of identity.)

Incomplete Definite Descriptions

•

357

(25a) A situation in which o is F is one that supports the truth of the claim that something is F. (25b) A situation in which o is F and o is G is one that supports the truth of the claim that an F is G. (25c) A situation in which exactly one object o is F, and, moreover, o is G supports the truth of the claim that the F is G. (25d) A situation in which o bears R to o is one that supports the truth of the claim that o bears R to itself, i.e., that o self-R’s. (25e) A situation in which o is F and o is G is one that supports the truth of the claim that o is both F and G. (26a) In order for a situation to support the truth of the claim that something is F it must be one in which the property of being F has the property of being instantiated; for this it is not sufficient for it to be a situation in which o is F. (26b) In order for a situation to support the truth of the claim that an F is G it must be one in which the properties of being F, and being G, bear the co-instantiation relation; for this it is not sufficient for it to be a situation in which o is F and o is G. (26c) In order for a situation to support the truth of the claim that the F is G it must be one in which the properties of being F, and being G, bear the co-instantiation relation, and the property of being F has the property of being uniquely instantiated; for this it is not sufficient for it to be a situation in which exactly one object o is F, and, moreover, o is G. (26d) In order for a situation to support the truth of the claim that o bears R to itself (i.e., self-R’s) it must be one in which o has the property of self-Ring; for this it is not sufficient for it to be a situation in which o bears R to o. (26e) In order for a situation to support the truth of the claim that o is F and G it must be one in which o has the compound property of being F-and-G; for this it is not sufficient for it to be a situation in which o is F and o is a G.36 These modifications in the notion of a situation raise the question of whether we ought to continue to regard them as a species of truthsupporting circumstance. They might even lead one to wonder whether 36 In each of these cases (26a−e), the denial that situations of a certain type support the truth of a sentence, does not carry with it the claim that it is (metaphysically) possible for the sentence to be untrue when situations of the relevant type are real (or actual). For it may be that although r does not itself support the truth of S, it is connected by what Barwise and Perry call a “necessary constraint” to a situation r' which does—in which case the existence of r will guarantee that S is true, without, in the technical sense, supporting the truth of S.

358

•

Essay Twelve

the modified situational approach and the structured proposition approach are really notational variants of the same underlying view. Although important similarities exist, I do not think they are. It must be remembered that the appeal of partial circumstances of evaluation in situation semantics is not limited to the construction of fine-grained objects of the attitudes. Rather, partiality is supposed to be systematically significant in the analysis of a variety of constructions. Thus, it is of considerable theoretical importance to determine whether it really is. Initially it was thought that definite descriptions provided strong evidence for partiality. I have argued that, in fact, they do not; the analysis of definite descriptions is not facilitated by the kind of partiality that situation semantics provides.

References Barwise, Jon, and John Perry. 1983. Situations and Attitudes. Cambridge: MIT Press. ———. 1985. “Shifting Situations and Shaken Attitudes.” Linguistics and Philosophy 8:105−61. Donnellan, Keith S. 1966. “Reference and Definite Descriptions.” Philosophical Review 75:281−304. Kaplan, David. 1978. “Dthat.” In Syntax and Semantics, vol. 9, Pragmatics, ed. Peter Cole, 221−43. New York: Academic Press. ———. 1979. “On the Logic of Demonstratives.” In Contemporary Perspectives in the Philosophy of Language, ed. Peter A. French, Theodore E. Uehling, and Howard K. Wettstein, 401−12. Minneapolis: University of Minnesota Press. ———. 1989. “Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and Epistemology of Demonstratives and Other Indexicals.” In Themes from Kaplan, ed. Joseph Almog, John Perry, and Howard Wettstein with the assistance of Ingrid Deiwiks and Edward N. Zalta, 481−563. New York: Oxford University Press. Kripke, Saul A. 1979. “Speaker’s Reference and Semantic Reference.” In Contemporary Perspectives in the Philosophy of Language, ed. Peter A. French, Theodore E. Uehling, and Howard K. Wettstein, 6−27. Minneapolis: University of Minnesota Press. Lewis, David K. 1982. “Scorekeeping in a Language Game.” Journal of Philosophical Logic 8:339−59. McCawley, James D. 1981. Everything That Linguists Have Always Wanted to Know about Logic. Chicago: University of Chicago Press. Salmon, Nathan. 1981. Reference and Essence. Princeton: Princeton University Press. ———. 1982. “Assertion and Incomplete Definite Descriptions.” Philosophical Studies 42:37−45. ———. 1986a. Frege’s Puzzle. Cambridge: MIT Press.

Incomplete Definite Descriptions

•

359

———. 1986b. “Reflexivity.” Notre Dame Journal of Formal Logic 27:401−29. ———. 1989. “Tense and Singular Propositions.” In Themes from Kaplan, ed. Joseph Almog, John Perry, and Howard Wettstein with the assistance of Ingrid Deiwiks and Edward N. Zalta, 331−92. New York: Oxford University Press. Soames, Scott. 1985. “Lost Innocence.” Linguistics and Philosophy 8:59−72. ———. 1988. “Direct Reference, Propositional Attitudes, and Semantic Content.” In Propositions and Attitudes, ed. Nathan Salmon and Scott Soames, 197−239. New York: Oxford University Press. ———. 1989. “Direct Reference and Propositional Attitudes.” In Themes from Kaplan, ed. Joseph Almog, John Perry, and Howard Wettstein with the assistance of Ingrid Deiwiks and Edward N. Zalta, 393−419. New York: Oxford University Press. Wettstein, Howard K. 1981. “Demonstrative Reference and Definite Descriptions.” Philosophical Studies 40:241−57.

ESSAY THIRTEEN

Donnellan’s Referential/Attributive Distinction

What is keith donnellan’s distinction between referential and attributive uses of definite descriptions, and what significance does it have for the philosophy of language? In my opinion the answers to these questions lie in Donnellan’s discussion of how a speaker may use a sentence containing a description to assert a singular proposition about an individual corresponding to the description. This is the so-called “referential use” of a description. A referential use of a description ⎡the F⎤ in a sentence ⎡The F is G⎤ is one in which a speaker has an individual o in mind about whom he wishes to make an assertion; the speaker uses the description to identify or pick out o, and the speaker intends his remark to be taken as asserting a singular, Russellian proposition about o to the effect that o “is G.” Since the speaker’s primary purpose is to say of o that o “is G,” this purpose need not be defeated even if the description, ⎡ the F⎤, does not correctly describe o. Even if F turns out not to apply to o, the speaker’s use of the description may still be successful in identifying for his audience the person he wishes to make an assertion about, in which case the speaker may succeed in asserting the singular proposition that o “is G.” This appeal to the notion of a singular, Russellian proposition is, admittedly, a somewhat theory-laden way of bringing out what Donnellan attempted to describe in more intuitive terms. Still, I think it accurately represents what he was getting at. For example, in his 1966 article, “Reference and Definite Descriptions,” he says the following: Using a definite description referentially, a speaker may say something true even though the description correctly applies to nothing. The sense in which he says something true is the sense in which he may say something true about someone or something. This sense is, I think, an interesting one that needs investigation. Isolating it is one of the byproducts of the distinction between the attributive and referential uses of definite descriptions.1 About this sense in which a speaker uses a description referentially to assert something true of an individual, Donnellan says the following: 1

Donnellan (1985, 244).

Donnellan’s Distinction

•

361

When a speaker says, “The Φ is ψ,” where “the Φ” is used attributively, if there is no Φ, we cannot correctly report the speaker as having said of this or that person or thing that it is ψ. But if the definite description is used referentially we can report the speaker as having attributed ψ to something. And we may refer to what the speaker referred to, using whatever description or name suits our purpose. Thus, if a speaker says, “Her husband is kind to her,” referring to the man he was just talking to, and if that man is Jones, we may report him as having said of Jones, that he is kind to her. If Jones is also the president of the college, we may report the speaker as having said of the president of the college that he is kind to her. And finally, if we are talking to Jones, we may say, referring to the original speaker, “He said of you that you are kind to her.” It does not matter here whether or not the woman has a husband or whether, if she does, Jones is her husband. If the original speaker referred to Jones, he said of him that he is kind to her. Thus where the definite description is used referentially, but does not fit what was referred to, we can report what a speaker said and agree with him by using a description or name which does fit.2 In this passage Donnellan tells us that a referential use of a description to refer to an individual o is a use in which the speaker says of o that o is such and such. What, we might ask, is it to say of an individual that it is such and such? The answer, it seems to me, is that to say of an individual that it is such and such is to assert the singular proposition that predicates such and such of that individual. However, this point may bear some reviewing. The locus classicus of the propositional attitude locutions, say of, assert of, believe of, and so on, in the philosophy of language is W. V. Quine’s 1956 paper, “Quantifiers and Propositional Attitudes.”3 There, these locutions are used to express Quine’s so-called “relational” senses of propositional attitude verbs. These senses, you will recall, were invented to give the effect of quantifying into ordinary propositional attitude constructions without violating Quine’s infamous metatheorem proclaiming the unintelligibility of quantifying into opaque constructions. Quine’s idea went something like this. When we say ‘There is someone such that Ralph believes that she is a spy’, it may seem as if we are binding from the outside the pronoun/variable she, which occurs inside an opaque construction created by the ordinary two-place belief predicate. However, according to Quine, this appearance is deceptive. Really, when we say ‘There is someone such that Ralph believes that she is a spy’, we are saying that 2 3

Donnellan (1985, 246). Quine (1956).

362

•

Essay Thirteen

Ralph believes of someone that she is a spy; and when we say that Ralph believes of someone that she is a spy, we are saying that a three-place belief relation holds between the believer (Ralph), the subject of the belief (in this case some person), and the attribute that the subject of the belief is supposed to have (spy-hood). All this simply to avoid the dogma that quantifying into opaque constructions is unintelligible. Fortunately, these days we no longer have to resort to such indirection. Following David Kaplan’s4 masterful deconstruction of Quine’s mistaken metatheorem we can, in Kaplan’s words, unite subject and attribute in the object of belief, eliminate superfluous relational senses of propositional attitude verbs, and give a straightforward account of quantification into the content clauses of attitude ascriptions. An ascription, ‘Someone x is such that Ralph believed, or asserted, that x is a spy’, is evaluated in the same way as any other quantified sentence.5 It is true iff the open formula ‘Ralph believed, or asserted, that x is a spy’ is true relative to some assignment of a person o to the variable x. This, in turn, is the case if Ralph believed, or asserted, the proposition expressed by the formula ‘x is a spy’ relative to an assignment of the individual o to ‘x.’ What proposition is this? It is the singular, Russellian proposition that predicates spyhood of the individual o. In Quinean terms, it is just the fusion of subject and attribute corresponding to the second and third argument places of the invented relational sense of the belief predicate. The application of this thoroughly standard treatment of quantification to Donnellan’s account of the referential use of a definite description is easy to see. As we have noted, according to Donnellan, a speaker who uses the description in ⎡The F is G⎤ referentially to refer to an individual o, says of o that o “is G.” Taking the say of locution to indicate quantifying-in we may paraphrase Donnellan’s claim as follows: In cases in which there is an object o such that the speaker uses the description in ⎡ The F is G⎤ referentially to refer to o, the speaker says that o “is G.” Applying the straightforward semantics of quantifying-in to this claim, we get the result that in these cases the speaker asserts the singular, Russellian proposition that predicates G-hood of o. In short, referential uses of definite descriptions are cases in which the speaker asserts a singular proposition about the individual the description is used to refer to. This, I believe, is the content of Donnellan’s claim, which I take to be both important and correct. As an illustration of the importance of the 4

Kaplan (1986). I assume that pronouns in English are capable of performing the semantic function of variables, so that ‘There is someone such that Ralph believes that she is a spy’ and ‘Someone x is such that Ralph believes that x is a spy’ may be taken as expressing the same proposition. 5

Donnellan’s Distinction

•

363

notion of a singular proposition for our understanding of language and language use, Donnellan’s discussion of referential uses of definite descriptions complements the standard, no-frills account of quantification into attitude ascriptions just discussed in a striking and interesting way. In both cases there are strong reasons for viewing agents as bearing propositional attitudes toward genuine, nondescriptive, singular propositions. Although the reasons are different in the two cases, they fit together remarkably well. In the quantificational case the commitment to singular propositions arises from four extremely plausible, and independently motivated ideas; first is the standard view that a quantified sentence ⎡For some x Fx⎤ is true iff the open formula ⎡Fx⎤ is true relative to some assignment of a value to the variable x; second is the standard selection of ordinary objects as values (and full semantic contents) of objectual variables; third is the treatment of attitude ascriptions like ⎡α believes, asserts, that S⎤ as reporting relations between agents and the proposition expressed by S; and fourth is the conception of the proposition expressed by a sentence as being determined by the semantic contents of its subsentential constituents. It follows from these ideas that an ascription, ⎡ There is some x such that Ralph asserted that Sx⎤, reports that Ralph asserted a singular, nondescriptive proposition, expressed by the formula ⎡Sx⎤ relative to an assignment of an object to ‘x’. The reason the proposition must be nondescriptive is that there is no descriptive information associated with the variable relative to an assignment; the only semantic content available to be contributed by the variable to the proposition expressed by the formula is the object assigned to it. Hence the commitment to singular propositions in the account of quantification into attitude ascriptions. This means that the truth of such ascriptions requires agents to have asserted singular propositions. But how, one might ask, do they do this? Although we may use a variable in reporting that an agent has asserted a singular proposition, the agent himself certainly didn’t use a bare variable in asserting the proposition. So how did he do it? Mustn’t he have used some closed term, which itself incorporates descriptive information? And if so, mustn’t the proposition asserted by the agent be descriptive, and hence not a genuine singular, Russellian proposition? To give complete and illuminating answers to these questions would be a long and complicated task. Following Keith Donnellan, David Kaplan, and Nathan Salmon, I would argue that part of the story involves an account of the use of proper names and indexicals to assert, and express belief in, singular propositions. However, for present purposes we may put that aside. Donnellan has already shown us how a speaker can use a definite description referentially to assert, and express belief in, a singular proposition.

364

•

Essay Thirteen

Here, although descriptive information plays an important role in specifying the proposition asserted, and believed, it does so solely by helping to identify the object the proposition is about. Once we have the object, the descriptive information falls away, and the object itself becomes a constituent of the proposition. The reason we know that the descriptive information does not get into the proposition is that the proposition may be true even if the description used referentially in asserting it does not apply to the object. Hence, the referential use of a description complements our account of quantification into attitude ascriptions by explaining how descriptive devices may be used by agents to assert, and express belief in, the genuine nondescriptive singular propositions that must be asserted, and believed, if attitude ascriptions involving quantifying-in are to be true. Throughout this discussion, I have not said anything about whether the singular proposition asserted by a speaker who uses a description referentially should be regarded as semantically expressed by the sentence used by the speaker, relative to the context of utterance. Much has been written about this, often under the guise of another question—namely, whether the semantic referent of a description relative to a context of utterance in which it is used referentially is just the object that the speaker uses it to refer to.6 My own view is that Donnellan’s distinction between attributive and referential uses of definite descriptions is not a distinction within semantics, but rather is a pragmatic distinction about the uses speakers make of certain pieces of language. Thus, I would maintain that when a speaker uses a definite description referentially to assert a singular proposition about an object corresponding to his use of the description, that proposition is not semantically expressed by the speaker’s sentence relative to the context of utterance. 6 In order to determine whether the singular proposition asserted by a speaker who uses a description referentially is semantically expressed by the speaker’s sentence, relative to the context, one needs to know whether the semantic content of the description relative to the context—i.e., that which it contributes to the proposition semantically expressed by the sentence in the context—is the individual whom the speaker refers to in using the description. If the semantic content of the description in the context is the individual the speaker refers to, then that individual will also be the semantic referent of the description in the context. However, since the semantic referent of a term (relative to a context) is not always identical with its semantic content (relative to the context), one cannot immediately move from the claim that the semantic referent of a description, in a context in which it is used referentially, is a particular individual to the claim that its semantic content in the context is also that individual. Unfortunately, this distinction is often blurred, or neglected, in discussions of the semantic referent of a referential use of a description. Nevertheless, what is typically at issue in these discussions is whether the semantic content of such a use is the object that the speaker refers to in using the description. This is the issue with which I will be most concerned.

Donnellan’s Distinction

•

365

There are several reasons for this. First, as Saul Kripke has shown, given a standard quantificational treatment of descriptions, we can typically explain referential uses of definite descriptions by appealing to that semantics together with independently needed Gricean conversational principles. Thus, there is no need to complicate our theory to accommodate referential uses. Second, as Kripke has also pointed out, the referential/attributive distinction applies not just to descriptions, but to other parts of language as well, for example proper names.7 It would seem to be a strange and unexplained coincidence, as well as a needless complication, if we had to posit parallel semantic ambiguities for all these cases. Third, in situations in which a description ⎡The F⎤ is used referentially to pick out an object o to which G applies but F does not, the natural thing to say about a speaker who has uttered ⎡The F is G⎤ is not just that he has said something true. Rather, the natural thing to say (assuming that G doesn’t apply to the unique thing that F does) is that although the speaker has said something that is literally false, namely the preposition expressed by ⎡the F is G⎤, he has also said something true, namely the proposition expressed by ⎡X is G⎤ relative to an assignment of o to ‘x’. This natural description of the case is precisely what one would expect if the proposition semantically associated with the speaker’s sentence is the familiar general one, while the additional assertion of the singular proposition arises from general conversational principles plus special facts about the context. If, on the other hand, the singular proposition were also the proposition semantically expressed by the speaker’s sentence in the context, we would have no explanation of the natural intuition that even though the speaker may have succeeded in saying something true, he also said something false.8 Finally, it seems to me that the whole phenomenon of true assertion despite misdescription invites a pragmatic account. It is part of the very nature of semantic rules that they apply generally to expressions whenever and wherever they occur. The idea that these rules might routinely be inapplicable, and that the interpretation of particular occurrences of expressions in particular contexts must depend on factors idiosyncratic to those contexts, is foreign to what semantics is, and is an indication that we have stepped outside its proper domain. I should emphasize that, in my view, the fact that the referential/attributive distinction is a pragmatic, rather than a semantic, one in no way diminishes its significance. On the contrary, part of the beauty and importance of the distinction depends on the essentially pragmatic character of a speaker’s true assertion of a singular proposition about an individual, even in cases in which the description used referentially is a misdescription of that individual. Indeed, I have argued that it is precisely such cases 7 8

Both of these points are made in Kripke (1979). Essentially this point is emphasized by Stephen Neale (1990, 91–92).

366

•

Essay Thirteen

that most graphically show that the true proposition asserted by the speaker must be a genuine, nondescriptive, singular proposition. I confess, I don’t know the extent to which Donnellan would, today, dispute this view of the theoretical nature of his distinction. However, it is not the view he presented in his 1978 paper, “Speaker Reference, Descriptions, and Anaphora.”9 The main task of that paper was to investigate the theoretical significance of the referential/attributive distinction. Donnellan put the problem this way: While the referential/attributive distinction proves, I believe, to appeal to our intuitions, vagueness about the role of speaker reference threatens its significance. Are there two uses of definite descriptions in the sense of two semantic functions in one of which the description conveys speaker reference, and in the other not? Or is it rather that definite descriptions are used in two kinds of circumstances, in one of which there is an accompanying phenomenon of speaker reference though it has no effect on the semantic reference of the description?10 The latter option spelled out in this remark is, I take it, essentially the pragmatic view of the distinction I have endorsed. Donnellan immediately goes on to say about this view that if it is correct, then “it is not clear what importance we should attach to the distinction in the philosophy of language.”11 If nothing else, I hope the discussion up to now has dispelled the idea that the importance of Donnellan’s distinction depends on it being taken as semantic. But putting the issue of importance aside, we still may ask whether there are reasons for viewing it as semantic. The burden of Donnellan’s 1978 paper was to show that there are. As he said, the task of the paper was to “offer certain arguments derived from a consideration of the phenomenon of anaphora to show that speaker reference cannot be divorced from semantic reference.”12 The sorts of cases involving anaphora that Donnellan had in mind are illustrated by the examples in (1). (1) a. A man came to the office this morning. He tried to sell me an encyclopedia. b. A man came to the office this morning. The man tried to sell me an encyclopedia. c. A man came to the office this morning. The man who came to the office this morning tried to sell me an encyclopedia. 9

Donnellan (1979b). Donnellan (1979b, 28). 11 Donnellan (1979b, 28). 12 Donnellan (1979b, 28). 10

Donnellan’s Distinction

•

367

According to Donnellan, the initial clause in each of these examples is not a simple existential generalization to the effect that at least one man came to the office this morning. Instead, it serves to introduce an individual about whom the speaker wishes to make further assertions; and it expresses a proposition that is true only if that individual came to the office this morning. On Donnellan’s view, the anaphoric elements in subsequent sentences, the pronouns or definite descriptions, have as their semantic referents the individual introduced by the speaker’s use of the initial clause. It is important to realize that this individual, the one the speaker has in mind, need not be one that uniquely satisfies the descriptive information contained in the clause. This is illustrated by the fact that the person introduced by the speaker’s use of the clause, ‘A man came to the office this morning’, need not be the only man that came to the office on the given morning. In fact, it is apparently not required that the individual referred to by the speaker satisfy the initial clause at all; for example the man may have come to the office in the afternoon rather than the morning. In such a case, even though the initial clause is false, Donnellan thinks that the speaker’s use of it may still succeed in introducing an individual about whom later sentences in the discourse may make true assertions. Such, at any rate, was his view in 1978. What is the bearing of this discussion on the theoretical status of the referential/attributive distinction? There are several points which, I believe, Donnellan wanted to make about these examples. First, he thought that the individual introduced by the initial clause in these examples is an individual the speaker refers to in very much the same sense as does a speaker who uses a definite description referentially. In both cases the speaker has an individual in mind about whom he wishes to make an assertion—an individual who may, but need not, satisfy the descriptive information used by the speaker in referring to him. Second, Donnellan maintained that the referent of the anaphoric pronoun in the second sentence of (1a) is determined semantically to be the individual the speaker referred to in uttering the initial clause. Thus, the notion of speaker reference is not strictly pragmatic, but has semantic import for the interpretation of anaphoric pronouns. Third, Donnellan believed that the referents of the “anaphoric descriptions,” as we might call them, in the second sentences of (1b) and (1c) are also determined semantically to be the individual the speaker referred to in uttering the initial clause. If this is right, then it would seem that an individual might be the semantic referent of an “anaphoric description” even though the individual does not uniquely satisfy the description, and may not satisfy it at all. From all this it appears that Donnellan regarded the uses of the “anaphoric descriptions” as semantically referential. If they do have this status, then the second sentences in (1a, b, c) should each semantically

368

•

Essay Thirteen

express the singular proposition that attributes to the man whom the speaker is referring to the property of trying to sell the speaker an encyclopedia. Although Donnellan did not explicitly describe the situation in this way, he did say that a sentence containing one of these anaphoric elements expresses a proposition whose truth conditions depend on the individual semantically referred to by that element.13 When the anaphoric element is a pronoun, he gave examples intended to show that the proposition expressed does not require anything to satisfy the descriptive information used in the initial clause to introduce the speaker’s referent.14 Finally, he characterized sentences containing anaphoric descriptions as equivalent to corresponding sentences containing anaphoric pronouns15 and explicitly said that even when a definite description is used, the proposition expressed does not imply that anything uniquely satisfies the description.16 All of this supports the interpretation according to which Donnellan viewed the second sentences of (1a, b, c) as expressing singular, rather than general, propositions. There is, of course, a difference that Donnellan recognized between uses of anaphoric elements, like those in (1a, b, c), and the original examples of referential uses of definite descriptions. In the original examples, a definite description is used referentially, without any antecedent with which it is anaphorically linked. According to Donnellan the difference between the two constructions—the one in which the definite description is used on its own, and the one in which it is anaphorically linked to an antecedent—is as follows: In referential contexts, those where speaker reference is present, the choice of which construction to use is, I believe, a matter of the speaker’s expectations and intentions toward his audience: does he expect and intend that they will recognize who or what he has in mind? If he does, then he will use a definite description with no further introduction. If not, he will begin with an introduction via an indefinite description. What the latter does, so to speak, is to announce that the speaker intends to speak about a particular thing or particular things following under a certain description—for example he intends to speak of a particular king or a particular man or particular men who came to the office. There is no implication in most cases that he will speak of everything falling under the description. Having done this, he can then go on to use a definite description or a pronoun to refer to 13

Donnellan (1979b, 38−39). See, for example, Donnellan’s (1979b) discussion of his examples (6−8), pp. 33−34, and also his discussion of examples (23−25), p. 41. 15 Donnellan (1979b, 35−36) . 16 Donnellan (1979b, 36−38) . 14

Donnellan’s Distinction

•

369

what he wants to talk about. Where the speaker intends and expects his audience to be able to recognize what he speaks about from the description used (plus attendant circumstances), such an introduction is otiose. . . . This account does not imply that the one construction is grammatically derived from the other, although it is consistent with that possibility. But if it does represent the difference between the two as far as why one might be used in some circumstances and the other in other circumstances, it also suggests that there should be no real difference in truth conditions or semantic references.17 From this, we may conclude that, on Donnellan’s view, the semantic function of the two kinds of occurrences of referentially used definite descriptions is the same. So, if one is used in a sentence to semantically express a singular proposition about an individual the speaker has in mind, then the other one functions in the same way. I have already indicated why I do not think that referential uses of nonanaphoric definite descriptions lead to the semantic expression of singular propositions by sentences containing them. I hold the same view about uses of anaphoric definite descriptions in the sorts of contexts discussed by Donnellan. One reason for this is that the same sorts of features involving anaphoric linking emphasized by Donnellan are found in contexts in which the speaker clearly doesn’t have any particular individual in mind, and so cannot be asserting a singular proposition about such an individual. This is illustrated by the discourse in (2). (2) a. A student came to my office before I arrived this morning and removed a book. I noticed something was wrong because when I got there the door was ajar, and the janitor had not been in. What really bothers me is that the student must have had a key, and I am afraid he will be back. (Said by A) b. I share your worry that he had a key, and may be back, but I don’t think that he was a student. It seems more likely that he was an ex-employee. (Said by B) In such a case, neither party in the discussion need have any definite idea about the identity of the intruder. (A may simply assume that it was a student, whereas B regards the class of ex-employees as a more likely source of suspects.) Thus, there is no individual that the speakers have in mind, no individual they are talking about, no assertion about anyone that he (or she) came to A’s office, removed a book, or has a key, and no singular proposition toward which either speaker need bear a propositional 17

Donnellan (1979b, 40).

370

•

Essay Thirteen

attitude (such as being bothered that, being afraid that, or being worried that) in order for the speaker’s remark to be true. Consequently, the relevant clauses in this discourse do not semantically express singular propositions. By parity of reasoning, I see no reason to believe that similar clauses in Donnellan’s examples do either. The existence of examples like (2) shows that there must be a semantics for anaphoric linking which assigns general, rather than singular, propositions to clauses containing anaphoric pronouns and descriptions.18 This semantics is, of course, also available in contexts, like those emphasized by Donnellan, in which the speaker happens to have a particular individual in mind. In Donnellan’s examples involving anaphoric linking, a speaker uses an initial clause containing an indefinite description to introduce a referent that he, the speaker, has in mind, but whom his audience is unable to recognize. Having announced his intention to speak of a particular individual in this way, the speaker goes on to use sentences containing pronouns or definite descriptions anaphorically linked to the initial indefinite description. The question at issue in these cases is whether the semantic referent of these anaphoric elements is just the speaker’s referent, and whether the truth conditions of the propositions semantically expressed by sentences containing these elements involve the speaker’s referent directly (rather than via a set of associated descriptions). In addition to questioning the need to complicate the semantics by adding a second, nondescriptive treatment of anaphoric linking, we can, I think, draw on some plausible intuitions to show that even in the cases that Donnellan provides, the speaker’s sentences do not semantically express singular propositions. Let us begin by imagining a speaker A assertively uttering the sentences in (1a) or (1b) to a hearer B, who, A realizes, does not recognize the individual that A has in mind. This is represented in (3).

18 A version of (2a) was used to make this point in of Soames (1986, n. 33) (essay 12 in this volume, n. 33). Example (2b) extends the point to show that anaphoric pronouns in discourses like these must have some descriptive content, even when that content is not a straightforward repetition of the content in the clause containing the antecedent. Stephen Neale (following Martin Davies) emphasizes this point (1990, 202−3). He also uses the following example to indicate that an anaphoric pronoun may have a descriptive content different from its antecedent in cases involving a variety of quantified antecedents.

A: Three people who ate at Luigi’s last Monday died from food poisoning. B: No one ate at Luigi’s last Monday; it’s closed on Monday. They must have eaten there on Tuesday. Examples of this sort reinforce the point that anaphora in these constructions do not require the antecedent to be a singular term, or to denote, or introduce, a single individual.

Donnellan’s Distinction

•

371

(3) A says to B: A man came to the office this morning. He (the man) tried to sell me an encyclopedia. The first observation about this case is that, as a result of hearing A’s remark, B should be able, truly, to report what A said. In particular, (4a), as said by B, should be true, and (4b), which reports B’s assertions, should also be true. (4) a. A said that a man came to the office this morning. A also said that he (the man) tried to sell her an encyclopedia. (Said truly, by B.) b. B asserted that A said that a man came to the office this morning. B also asserted that A said that he (the man) tried to sell her an encyclopedia. It is useful to contrast this observation with a case in which the speaker, A, uses the pronoun, ‘he’, not as an anaphoric term, but as a genuine demonstrative, accompanied by a gesture of pointing to a certain person m whom she has in mind, but whom her audience cannot see, locate, or identify in any way (except as the individual, whomever he may be, that A has in mind). In such a situation, if A assertively utters the sentence ‘He was obnoxious’, without any anaphoric linking, attempting to assert the singular proposition that m was obnoxious, A’s audience will not be in a position to report her as having asserted that proposition. It is noteworthy that we don’t feel this way about the case involving A’s use of (3), where we take it for granted that the hearer, B, can successfully report A’s assertions. Since the audience is, in each case, equally in the dark about any singular proposition that A might have in mind, this suggests that in the case of (3), A’s utterances express general propositions available to both speaker and hearer. The second observation about this case is that, having reported what A said, B should be in a position to believe that which he, namely B, has just reported. Thus, the report (5), said by B, should be true. (5) I believe that A said that a man came to the office this morning. I also believe that A said that he (the man) tried to sell her an encyclopedia. (Truly said by B) A similar point is that if B trusts A, he should be in a position to believe what A said, in which case his report (6) should be true. (6) I believe that a man came to the office this morning. I also believe that he (the man) tried to sell A an encyclopedia. (Truly said by B) The crucial question here is whether the clauses containing the anaphoric pronouns or descriptions in (4), (5), and (6) express singular

372

•

Essay Thirteen

propositions about the individual A had in mind. We can test this by checking our intuitions about certain related ascriptions. For example, think again about (4b). (‘B asserted that A said that a man came to the office this morning. B also asserted that A said that he (the man) tried to sell her an encyclopedia’.) If the clause in (4b) containing the anaphoric pronoun or description expressed a singular proposition about the individual that A had in mind, then truth of (4b) would guarantee the truth of the sentences in (7).19 (7) a. There is some man such that B asserted (of that man) that A said that he tried to sell her an encyclopedia. b. B asserted that A said that you tried to sell her an encyclopedia. (Said addressing the individual A had in mind) Now it seems to me that these claims are false. Since B had no idea who A was talking about, B would sincerely deny, when confronted with any man, that he had asserted of that man that A said anything about him. Indeed, it seems likely that B would maintain that there is no particular man about whom he, namely B, had asserted of that man that A said something about him. This, it seems to me, is evidence that the claims in (7) are false, and hence that the relevant clause in (4b) does not express a singular proposition about the person A had in mind. These points are even clearer in the cases of (5) and (6). If the clauses in (5) and (6) containing the anaphoric elements expressed singular propositions about the individual A had in mind, then the truth of (5) and (6) would guarantee the truth of certain other ascriptions. First consider (5)—where B truly reports ‘I believe that A said that a man came to the office this morning. I also believe that A said that he (the man) tried to sell her an encyclopedia’. If the clause containing the anaphoric pronoun or description expressed a singular proposition about the person A had in mind, then it would be true that (8a) There is some man such that B believes (of that man) that A said that he tried to sell her an encyclopedia. If we were addressing the person that A had in mind, it would also be true to say that 19 If the relevant clause in (4b) expressed a singular proposition, the move from (4b) to (7a) would scarcely be controversial, and would, in any case, be justified by the role of singular propositions in understanding quantifications into attitude ascriptions discussed above. The corresponding move to (7b) is potentially more contentious. Although I accept an account of attitude ascriptions involving demonstratives that would sanction it, I will not argue for that account here. As we will see below, Donnellan himself is implicitly committed to sanctioning the move (on the assumption that the relevant clause in (4b) expresses a singular proposition).

Donnellan’s Distinction

•

373

(8b) B believes that A said that you tried to sell her an encyclopedia. But since, in fact, B doesn’t have anyone in mind, these claims are not true. Thus, the clause in (5) containing the anaphoric pronoun or description does not express a singular proposition about the person A had in mind. The examples in (9) can be used to make the same point about (6). (9) a. There is some man such that B believes (of that man) that he tried to sell A an encyclopedia. b. B believes that you tried to sell A an encyclopedia. (Said addressing the individual A had in mind) The argument I have just sketched may be summarized as follows: P1. If A assertively utters the sentences in (3) to B (in the sort of case that Donnellan has in mind), then B is in a position to report what A said, to believe that report, and, if B trusts A, to believe what A said. (Shown by the truth of (4)−(6)) P2. In doing these things B does not assert or believe singular propositions about the individual A had in mind. (Shown by the falsity of (7)−(9)) P3. If A’s use of the sentence in (3) containing the anaphoric pronoun or description semantically expressed a singular proposition about the person A had in mind, then A’s utterance would be an assertion of that proposition; if that were so, then B’s report of what A said would also be an assertion of a singular proposition about the individual A had in mind, and B’s believing that report would involve B’s believing a singular proposition about that individual, as would B’s believing what A said. C. A’s use of the sentence in (3) containing the anaphoric pronoun or description does not semantically express a singular proposition about the person A has in mind. In short, Donnellan’s anaphoric examples do not provide a basis for interpreting the referential/attributive distinction semantically. In closing, I would like to connect the material in this discussion with some important insights in Donnellan’s paper, “The Contingent A Priori and Rigid Designators.”20 There Donnellan considers cases of using a definite description D to introduce, and rigidly fix the referent of, a name N. He argues, persuasively, that although relying on such a procedure may be sufficient for an otherwise competent speaker to come to know, a priori, that the sentence (10) (10) If there is someone (something) that is D, then N is D 20

Donnellan (1979a).

374

•

Essay Thirteen

expresses a truth, it cannot, by itself, be sufficient for the speaker to come to know, a priori, the contingent truth that (10) expresses. For example, if I now use the description, ‘the next Republican to be elected President of the United States’, to introduce and rigidly fix the referent of the name ‘ReRon’, I may be sure, in the absence of any further empirical investigation, that the sentence (11) If there is someone who will be the next Republican to be elected President of the United States, then ReRon will be the next Republican to be elected President of the United States expresses a truth, even though I have no idea who the next Republican President will be. However, in such a situation I do not know the truth that (11) expresses—I do not know that if there is someone who will be the next Republican to be elected President of the United States, then ReRon will be the next Republican to be elected President of the United States. The reason I do not know this is that knowledge of that which is expressed by (11) would be, in Donnellan’s words, de re knowledge of an individual to the effect that that individual will be the next Republican to be elected President of the United States, if anyone is. In other words, according to Donnellan, knowledge of the proposition expressed by (11) is knowledge of a certain singular proposition. Thus, on Donnellan’s view, (12) can be true only if the examples in (13) are also true.21 (12) Soames knows that ReRon will be the next Republican to be elected President of the United States, if anyone is. (13) a. There is someone such that Soames knows (of that person) that that person will be the next Republican to be elected President of the United States, if anyone is. b. Soames knew in 1993 that you would be the next Republican to be elected President of the United States, if anyone would. (Said sometime in the future addressing the next Republican President) But, as Donnellan would agree, the sentences in (13) are clearly false. Consequently, (12) is too. Donnellan does not limit himself to examples in which the attitude in question is knowledge, but rather extends his point to other propositional attitudes, including believing and asserting as well.22 Thus, on his view, if I were to assertively utter (11), I would not succeed in asserting,

21 22

Donnellan (1979a, 54−55). Donnellan (1979a, 56−57).

Donnellan’s Distinction

•

375

or expressing my belief in, the singular proposition it expresses. The argument for this should by now be familiar. Since the sentences in (14) can be true only if I assert and believe a singular proposition about the person who will be the next Republican to be elected President of the United States, the sentences in (14) are true only if those in (15) are. (14) a. Soames asserted that ReRon will be the next Republican to be elected President of the United States, if anyone is. b. Soames believes that ReRon will be the next Republican to be elected President of the United States, if anyone is. (15) a. There is someone such that Soames asserted/believes (of that person) that that person will be the next Republican to be elected President of the United States, if anyone is. b. Soames asserted/believed in 1993 that you would be the next Republican to be elected President of the United States, if anyone would. (Said sometime in the future addressing the next Republican President) But then, since the examples in (15) are clearly false, those in (14) are too. In my view, these observations, drawn from Donnellan’s discussion in “The Contingent A Priori and Rigid Designators,” are just right. What I have tried to do in the argument based on examples (3)−(9) is to extend these observations to his cases of anaphoric linking. If the argument is successful, it shows that Donnellan’s examples of anaphoric linking in “Speaker Reference, Descriptions, and Anaphora” are not cases in which a speaker uses a sentence containing an anaphoric pronoun or description to semantically express a singular proposition. Hence, such cases do not provide a reason for construing the referential/attributive distinction semantically.

References Donnellan, Keith S. 1979a. “The Contingent A Priori and Rigid Designators.” In Contemporary Perspectives in the Philosophy of Language, ed. Peter A. French, Theodore E. Uehling, and Howard K. Wettstein, 12−27. Minneapolis: University of Minnesota Press. ———. 1979b. “Speaker Reference, Descriptions, and Anaphora.” In Contemporary Perspectives in the Philosophy of Language, ed. Peter A. French, Theodore E. Uehling, and Howard K. Wettstein, 28−44. Minneapolis: University of Minnesota Press. First published in Syntax and Semantics, vol. 9, ed. Peter Cole (New York: Academic Press, 1978). ———. 1985. “Reference and Definite Descriptions.” In The Philosophy of Language, ed. A. P. Martinich, 265−77. New York: Oxford University Press. First published in Philosophical Review 75 (1966): 281−304.

376

•

Essay Thirteen

Kaplan, David. 1986. “Opacity.” In The Philosophy of W. V. Quine, ed. Lewis E. Hahn and Paul A. Schilpp, 229−89. La Salle, Ill.: Open Court. Kripke, Saul A. 1979. “Speaker’s Reference and Semantic Reference.” In Contemporary Perspectives in the Philosophy of Language, ed. Peter A. French, Theodore E. Uehling, and Howard K. Wettstein, 6−27. Minneapolis: University of Minnesota Press. Neale, Stephen. 1990. Descriptions. Cambridge: MIT Press. Quine, W. V. 1956. “Quantifiers and Propositional Attitudes.” Journal of Philosophy 53, no. 5: 177−87. Soames, Scott. 1986. “Incomplete Definite Descriptions.” Notre Dame Journal of Formal Logic 27, no. 3: 349−75.

ESSAY FOURTEEN

Why Incomplete Definite Descriptions Do Not Defeat Russell’s Theory of Descriptions

Introduction A central lesson of “On Denoting” is that singular definite descriptions do not belong to the same category of expressions as names and demonstratives; they are not singular terms (Russell 1905). Instead, sentences containing them are quantificational. As a result, the meanings, or semantic contents, of these sentences are not singular propositions about objects denoted by the descriptions they contain. Instead, they are general propositions in which higher-order properties corresponding to quantifiers are ascribed to lower-order properties, or propositional functions, expressed by the formulas on which the quantifiers operate. Thus, the official Russellian analysis of (1a) is (1b), the content of which is the proposition informally given in (1c). (1a) The F is G (1b) ∃x [∀y (Fy ↔ y = x) & Gx] (1c) The property of being both G and uniquely F is instantiated. (or, the propositional function which assigns to any object o the proposition that o is both G and uniquely F “is sometimes true”) On Russell’s original view, definite descriptions are syncategorematical. Although there is a rule determining the contribution made by a description, ⎡the F⎤, to propositions expressed by sentences containing it, this grammatical constituent is not assigned any independent semantic content on its own, and the proposition semantically expressed by (1a) contains no constituent corresponding to it. For our purposes, this feature of the view can be dropped, without significant loss. The most interesting and important features of Russell’s analysis can be retained, along with a welcome gain in generality, by including ⎡the F⎤ in the category of generalized quantifiers, such as ⎡every F⎤, ⎡some F⎤, and ⎡most F’s⎤. On this, modified, Russellian analysis, ⎡the F⎤ can be thought of as expressing the property of being instantiated by whatever uniquely instantiates the property expressed by F (or of being a propositional function that assigns a true proposition to whatever object o uniquely satisfies F). The analysis given in (1) is then replaced by that given in (2).

378

•

Essay Fourteen

(2a) The F is G (2b) [the x: Fx] Gx (2c) Being G is instantiated by whatever uniquely instantiates the property of being F (or, the propositional function which assigns to any object o the proposition that o is G assigns a truth to whatever object uniquely satisfies F) It is this modified and improved Russellian analysis that I will defend against the problem posed by occurrences of incomplete definite descriptions like ‘the car’ in sentences such as (3a). (3a) I parked the car just behind some cars across the street. (3b) [the x: x is a car] I parked x just behind some cars across the street Intuitively, the problem is obvious. Although the apparent Russellian reading, (3b), of (3a), couldn’t possibly be true—since it requires both that there be one and only one car x (in the universe of discourse) and also that I parked x behind some other cars across the street—my utterance of (3a) might, nevertheless, be completely unproblematic, and result in the assertion of something true, and nothing false. Examples like this are extremely common. If they cannot be accommodated, then Russell’s theory of descriptions must be regarded as incapable of providing adequate semantic analyses of many sentences of ordinary English, and other natural languages. The seriousness of this challenge to Russell’s theory can be measured by the reaction to it of one of the theory’s most prominent recent defenders. Writing in defense of the theory against objections originating in Keith Donnellan’s distinction between attributive and referential uses of descriptions, Saul Kripke says the following: If I were to be asked for a tentative stab about Russell, I would say that although his theory does a far better job of handling ordinary discourse than many have thought, and although many popular arguments against it are inconclusive, probably it ultimately fails. The considerations I have in mind have to do with the existence of “improper” definite descriptions, such as “the table,” where uniquely specifying conditions are not contained in the description itself. Contrary to the Russellian picture, I doubt that such descriptions can always be regarded as elliptical with some uniquely specifying conditions added. (Kripke 1977, 6) My burden will be to show that Kripke’s worry can be laid to rest; the existence of incomplete, or “improper” definite, descriptions does not defeat Russell’s theory, when taken as a semantic account of ordinary English.

Russell’s Theory of Descriptions

•

379

The Nature of the Problem (4a) The book is on the table. (4b) I drove the car to work. (4c) The student got into an argument with a student from another school. (4d) The murderer must be insane. The problem posed by sentences containing incomplete descriptions, like those in (4), is that speakers who assertively utter them often succeed in saying something that is both germane and true, while avoiding the assertion of anything false—despite the fact that the literal meanings, or semantic contents, assigned to these sentences by a Russellian semantic theory appear, quite clearly, to be false (due to the fact that the description fails to pick out an object uniquely). If these contents are, indeed, false, then the Russellian needs to explain both how it is that something other than the semantic content of the sentence gets asserted, and why, even if something else is asserted, the semantic content of the sentence uttered is not asserted as well. One of the things that makes this challenge difficult is a plausible, and widely presupposed, conception of the relationship between meaning and semantic content, on the one hand, and assertion, on the other. As a first approximation, we may express this principle as follows: The Traditional Connection between Meaning, Semantic Content, and Assertion A sincere, reflective, competent speaker who assertively utters S (speaking literally, nonironically, and nonmetaphorically) in a context C says (or asserts), perhaps among other things, the proposition semantically expressed by S in C (also known as the semantic content of S in C).1 Suppose that this principle is correct, and that a speaker who assertively utters ⎡The F is G⎤ in a context C, in which its Russellian analysis is false, nevertheless succeeds in saying something both germane and true, while 1 Indexical sentences, like ‘I am hungry’, semantically express different propositions in different contexts of utterance. Nonindexical sentences, like ‘2 + 2 = 4’, semantically express the same proposition in all contexts. In both cases, the semantic content of S in C is the proposition semantically expressed by S in C. However, when S is nonindexical its semantic content (which remains invariant from context to context) can be identified with its meaning, whereas when S is indexical, its meaning is a function from contexts to semantic contents. Although this distinction should always be kept in mind, in cases in which little hinges on it I will skip over it lightly.

380

•

Essay Fourteen

avoiding the assertion of anything false. If such cases are genuine, they constitute a powerful argument against a purely Russellian treatment of descriptions. To rebut the argument one would have to show that the cases aren’t really counterexamples to Russell—either because (i) the principle connecting semantic content and assertion is incorrect, or because (ii) the proposition expressed by the Russellian analysis of ⎡The F is G⎤ is actually true, not false, in C, or because (iii), the speaker in C has, in fact, said something false, even though he may also have said something true. Since, heretofore, none of these options, or any combination of them, has seemed sufficient to handle all relevant cases, the idea has gained currency that sentences containing incomplete definite descriptions are semantically ambiguous between attributive, Russellian readings, in which the descriptions are generalized quantifiers, and demonstrative, or referential, readings, in which ⎡the F⎤ (semantically) refers to the contextually salient object that satisfies F—in a manner analogous to the way in which ⎡this F⎤ does.2 On this story, when I use (3a) to talk about the particular car c that I have been driving, the singular proposition that I parked c just behind some cars across the street is both the semantic content of the sentence uttered, in the context, and also the proposition asserted. No wonder that I both assert something true, and say nothing false. On this approach, we posit special non-Russellian, semantic readings of sentences containing incomplete descriptions in order to preserve both the traditional connection between meaning, semantic content, and assertion, and our intuitions about what speakers do, and do not, assert when they utter sentences containing such descriptions. This, I shall argue, is a mistake.

Why Incomplete Definite Descriptions Are Not Semantically Referential There are four main considerations that tell against the view that sentences containing incomplete definite descriptions have semantically referential readings. The first arises from the fact that the distinction between complete and incomplete definite descriptions is nonlinguistic. Since, in most cases, the question of whether there is more than one thing that satisfies F is a nonlinguistic one about what exists in the world, there is no way, within linguistic theory, to identify which descriptions are 2

Variants of this approach can be found in Wettstein (1981) and Barwise and Perry (1983). For a critique of Barwise and Perry, see essay 12 in this volume. Nathan Salmon critiques Wettstein in Salmon (1982).

Russell’s Theory of Descriptions

•

381

complete, and which incomplete. Thus, if linguistic theory is to assign semantically referential meanings to incomplete definite descriptions, it must assign such meanings to all definite descriptions—which is defensible only if all such descriptions are semantically ambiguous. But there is no case to be made for a universal ambiguity of this sort. The second difficulty is that the semantically referential account mischaracterizes the truth-conditional content of speakers’ assertions. For example, if a speaker assertively utters (4c), ‘The student got into an argument with a student from another school’, in a context in which the incomplete description ‘the student’ is used to pick out a particular student s, then the semantic proposal identifies the singular proposition expressed by (SA4c) as both the semantic content of the sentence uttered, and the proposition asserted by the utterance. (SA4c) s got in a fight with a student from another school. (where the semantic content of ‘s’ is the individual it refers to) However, this proposition does not exhaust the assertive content of the speaker’s remark, since it leaves out the characterization of s as a student.3 That this is part of what the speaker asserts is indicated by the obvious truth of the reports, (R14c) and (R24c). (R14c) The speaker said the following: that a student got into an argument with a student from another school. (R24c) The speaker said that two students got into an argument. Thus, the semantic proposal fails to correctly characterize the data about assertion that motivates it. The third problem is that the approach doesn’t generalize properly to instances of what have been called referential uses of descriptions, and other relevant terms.4 In the case of descriptions, the most significant examples, for our purposes, are those in which a speaker uses what turns out to be an inaccurate description to pick out a certain individual, and say something about him. When I say, “The man in the corner drinking champagne is a famous philosopher,” intending to single out what I take to be one particular, contextually salient, man m from a group including several men in the corner drinking champagne, I am using an incomplete description. If, in fact, m happens to be drinking Riesling, then we have a standard, Donnellan-type case of misdescription. As Donnellan famously pointed out, this does not prevent me from truly asserting, of m, that he 3

Nathan Salmon (1982) makes a version of this point. The seminal article in the development of the distinction between attributive and referential uses of descriptions is Donnellan (1966). For a discussion of this distinction in that, and related articles, see essay 13. 4

382

•

Essay Fourteen

is a famous philosopher. However, my ability to do this is not captured by the postulated semantically referential reading—since, according to that reading, m is the semantic content of the description in this context only if m is both a man in the corner drinking champagne, and the denotation intended by the speaker. This is problematic. We don’t want one explanation for the assertion of the contextually determined singular proposition when the incomplete description, ⎡the F⎤, is a misdescription, and a different explanation for the case in which it is accurate. If one insists that the explanation must be semantic in both cases, then one must change the posited semantically referential interpretation of the description, so that F makes no contribution to the determination of the referent of the description in the context. This is implausible. It is hard to see why a normally compositional semantic theory should simply ignore the content clause of a description, on one of its readings. Moreover, we have already seen that F standardly does make a contribution to what is asserted by a speaker who uses the description in ⎡The F is G⎤ referentially. The defender of semantically referential readings of these sentences would have no hope of explaining this, if he were to take F as making no contribution to these readings, whatsoever. Thus, the proposed modification of the semantically referential interpretation must be rejected. If there are such interpretations, F contributes to them. This means that the proponent of such interpretations is committed to giving a semantic explanation for a speaker’s ability to use ⎡The F is G⎤ to assert, of o, that it has the property expressed by G, in cases in which o satisfies F, while offering a different, pragmatic, explanation of the speaker’s ability to do the same thing in cases in which o turns out not to satisfy F. But surely, if a pragmatic explanation is going to be required anyway, it should be extended to both cases—thereby removing the central reason for positing semantically referential readings in the first place. This point is reinforced by Kripke’s observation that cases of correct assertion, despite referential misdescription or misapplication, occur not only with definite descriptions, but also with names (Kripke 1979; see in particular pp. 14-15). In Kripke’s example two people see Smith in the distance and mistake him for Jones. The speaker says “Look, Jones is raking leaves,” thereby saying something true about Smith, (and also something false about Jones). This fact is not explained by any semantic ambiguity, according to which, on one interpretation, names refer to their bearers, whereas on a second interpretation they refer to contextually salient individuals mistaken for their bearers. There is no ambiguity, and the required explanation is purely pragmatic. What is needed is one explanation of how, in all these cases (those involving misapplications of

Russell’s Theory of Descriptions

•

383

names, those involving misdescriptions of contextually salient individuals, and those involving correct, but partial or incomplete, descriptions of such individuals) speakers are able to assert truths that are not semantic contents of the sentences they utter. The postulation of semantically referential readings of descriptions cannot do this on its own. Nor, as we shall see, is it needed as part of a theory that does. The final nail in the coffin of the semantically referential theory of incomplete descriptions comes from utterances of (4d)—‘The murderer must be insane’—in which the incomplete description ‘the murderer’ is used attributively to refer to whoever turns out to be the one who committed a certain murder, in the absence of any de re belief, of a particular person p, that p is the murderer. Suppose, for example, that (4d) is uttered upon coming across the victim of an extraordinarily cruel and senseless murder. In such a case, the incomplete description ‘the murderer’ might well be elliptical for ‘the murderer of him’ said, demonstrating the body. In this case, the actual murderer of the man in question may be unknown, and the speaker need not be taken as asserting any singular proposition about that person. Since incomplete definite descriptions are not always used referentially, no semantically referential interpretation is capable of explaining all the data.

The Simple Russellian Treatment of Definite Descriptions Putting aside the failed view that sentences containing incomplete descriptions have semantically referential interpretations, we return to the problem posed by these examples. The problem is the apparent mismatch between the Russellian semantic contents of these sentences and the propositions speakers assert by uttering them. Since this problem involves the way in which the information semantically encoded by sentences interacts with salient information in the contexts in which these sentences are used, the tools available for solving the problem come from both semantics and pragmatics. Three in particular stand out—the ability to vary the domain of quantification from one context to the next, the ability to use pragmatic maxims to force reinterpretation of a speaker’s remarks, and the ability to use contextual information to supplement the semantic information encoded by expressions in a particular context. The first of these is illustrated by an utterance of (4b)—‘I drove the car to work’—in a context in which there is only one car under consideration (and I drove it to work). Whether one regards the contextual determination of the domain of quantification to be a matter of context-sensitive semantics (so that the semantic contents of quantifiers vary from one

384

•

Essay Fourteen

context to the next), or a matter of pragmatic supplementation of a constant semantic content (which determines what speakers in different contexts assert), the phenomenon applies to quantifiers generally, and has nothing special to do with definite descriptions (Neale 1990, chap. 3, pp. 94−98). Either way, the Russellian is free to invoke contextual determination of the domain of the quantifier to explain why, in the case imagined, I succeed in saying something true, and nothing false, when I utter (4b). I succeed because there is one and only one car in the contextually determined domain—even though the world as a whole contains many cars. This sort of case is relatively easy. However, not all cases are so straightforward—as is evidenced by the fact that I may say something true, and nothing false, by assertively uttering (3a), ‘I parked the car just behind some cars across the street’, even though, in this case, the domain of discourse (and hence of quantification) clearly includes more than one car.5 Here, something more than contextual determination of the domain seems to be at work.6 It is natural to think that pragmatic reinterpretation and contextual supplementation may be involved. The idea goes something like this: The (Russellian) semantic content of (3a) in a context in which many cars are under discussion is the obvious falsehood that there is one and only one car in the domain of discourse, and I parked it just behind some cars across the street (which are also in the domain). Since it is evident to everyone in the conversation that this proposition is false, it is evident that it can’t be what I, the speaker, am inviting my hearers to believe. Given the presumption that I am obeying the conversational maxim not to say anything known to be false, one must, therefore, reinterpret my utterance. Instead of asserting the semantic content of the sentence uttered, I should be taken as having asserted some contextually obvious enrichment of 5

The importance of examples like this is stressed by Lewis (1979). In saying this, I am implicitly rejecting an analysis according to which (i) each occurrence of a quantifier phrase in a sentence is taken to contain a hidden domain variable and (ii) contexts are expanded to include new parameters (essentially just a list) providing interpretations for each such variable. On this analysis, each quantifier occurrence is assigned a contextually determined interpretation that allows it to range over a specially designated subset of the entire domain. With this freedom, incomplete definite descriptions can just about always be made complete by combining them with the right context, in this sense of ‘context’. However, contexts, in this sense, are highly abstract objects artificially designed to always give us the propositions we want. Although for some theoretical purposes this sort of analysis may be perfectly acceptable, it is not, in my opinion, what we want from an analysis of how, in natural language, understanding the meaning of a sentence and knowing the circumstances in which it is uttered allow one to interpret the utterance. Since this is what we are after, the abstract analysis that does away with incomplete definite descriptions by fiat can be put aside. These matters are discussed in more detail in essay 10. See also pp. 12–15 of the introduction to this volume. 6

Russell’s Theory of Descriptions

•

385

it—such as the proposition that I parked the car belonging to me just behind some cars across the street. On this story, the semantic content of the sentence I utter is false in the context of utterance. However, the normal presumption that I should be taken as asserting its semantic content is defeated by the obvious conversational implicature that I am not committing myself to, or inviting my hearers to believe, what we all recognize to be false. To determine the proposition I did assert, one appeals to salient background information about me and my car which is shared by the conversational participants. This information is used to pragmatically enrich the content of the description ‘the car’, so that its contribution to the proposition I asserted is the information explicitly carried by ‘the car belonging to me’. The simple Russellian treatment of incomplete descriptions claims that all relevant cases can be handled in this way. This account requires a small, but plausible and independently motivated, modification of our earlier principle stating the traditional connection between meaning, semantic content, and assertion. The needed modification goes roughly as follows. The Traditional Connection between Meaning, Semantic Content, and Assertion: 2 A sincere, reflective, competent speaker who assertively utters S (speaking literally, nonironically, and nonmetaphorically) in a context C says (or asserts), perhaps among other things, the proposition p that is semantically expressed by S in C (also known as the semantic content of S in C)—provided that the presumption that the speaker intends to commit himself or herself to p (and to invite his or her audience to believe p) is not defeated by a conversational implicature to the contrary. In addition to this modification, the simple Russellian treatment of incomplete descriptions also requires the ability to pragmatically enrich the semantic content of an expression to arrive at the proposition asserted by a speaker who utters a sentence containing it. However, this, too, is independently motivated, and not unique to definite descriptions. The generality of the proposed Russellian explanation of our use of incomplete definite descriptions is illustrated by the fact that its central features apply to sentences containing quantifiers of all sorts. (5a) No one knows that you and I are here. (5b) Everyone looks up to Larry. It is clear (i) that I can use (5a) without asserting the obvious falsehood that absolutely no one (including me and you) knows that we are here,

386

•

Essay Fourteen

and (ii) that I can use (5b) without asserting the incoherent proposition that everyone (in a domain of discourse that includes Larry) looks up to Larry. These facts parallel those involving incomplete definite descriptions. As in the case of (3a), the semantic contents of (5a) and (5b) in the relevant contexts are obviously false. Since this is evident to everyone in the conversation, it is also evident that they are not what I, the speaker, am inviting my hearers to believe. Given the presumption that I am obeying the conversational maxim not to say anything known to be false, I know that my hearers will reinterpret my utterance. Instead of asserting the semantic content of the sentence uttered, I will be taken to have asserted some contextually obvious enrichment of it—such as the propositions expressed by (6a) and (6b), relative to an assignment of a contextually obvious group to ‘us’ or ‘them’. (6a) No one but us (i.e., no one else) knows that you and I are here. (where ‘us’ designates me and my audience) (6b) Every one of us/them looks up to Larry. (where ‘us’ or ‘them’ designates a group that doesn’t include Larry) Similar remarks apply to other examples. Thus, the proponent of the simple Russellian treatment of incomplete definite descriptions can claim for it the virtue of requiring nothing in our semantic and pragmatic theories not already required by our use of other quantifiers.

Worries about the Simple Russellian Treatment Despite the attractiveness of this theory, there are worries suggesting that it is not quite right. The first inkling that something may be wrong is a worry about pragmatic reinterpretation. Unlike the examples given above, clear cases of pragmatic reinterpretation tend to be quite obvious; speaker-hearers recognize that the sentences uttered are literally false, and they know that they are being invited to come up with contextually obvious reinterpretations. For example, imagine that you and your friend Mary are listening to the remarks of a well-known campus orator, when Mary turns to you and says, “Norman really is God’s fountain pen, isn’t he?” On hearing this, you immediately realize that she can’t be asserting the obvious falsehood semantically expressed by the sentence she utters—one which predicates of Norman the property of being a certain kind of artifact (used to write by depositing ink on paper) possessed by God. Since that proposition is transparently false, you realize that Mary is not asserting it, and you look for an alternative interpretation of her remark. Depending on what else is taken for granted in the conversation, you may take her to have asserted (i) that God really is using Norman to

Russell’s Theory of Descriptions

•

387

communicate his thoughts and desires, (ii) that (God aside) Norman really does have the truth on the matters about which he is speaking, or (iii) that Norman is a windbag who takes himself to be an authority, even though he really isn’t. The salient point is that the need to reinterpret is transparent. This is doesn’t seem to be so with the utterances of (3a), (5a), and (5b) previously considered. On hearing remarks like these in conversation, we don’t, standardly, take ourselves to be invited to reinterpret them, and when making these remarks ourselves, we don’t, normally, take ourselves to be relying on hearers’ reinterpretations. Perhaps there is some innocent explanation of this linguistic phenomenology that preserves the explanation given in the previous section. But perhaps not. Perhaps the phenomenology is a hint that something more serious is wrong. That this is so is suggested by the following example. The context is a math competition in which a large group of students, located in different rooms, are given five math problems, on which they will be evaluated. Of these problems, the last is the most difficult, and any student solving it is guaranteed a place in the finals—though one can also reach the finals by solving all the other four. The teacher in room 1 has corrected the results from her room, and has determined that one, and only one, of her students solved problem 5. In this setting, she assertively utters (7a), and (7b). (7a) I am passing back your papers. The good news is: (7b) The student who solved problem 5 will compete in the finals with other high-scoring students in the competition. It is obvious from the context that what she has really asserted, in uttering (7b), is something along the lines of (7c). (7c) The student in this group that solved problem 5 will compete in the finals with other high-scoring students in the competition. This is not explained by the simple Russellian account we have given. In the context of utterance, the domain of discourse for the teacher’s remarks includes all students in the math competition, including those taking the exam in other rooms—as is indicated by her reference to other high-scoring students. Thus, unless there is reason to posit some further contextual supplementation of the description, ‘the student who solved problem 5’, the quantification it involves will range over a group including all students taking the test. On the simple Russellian treatment, such contextual supplementation will come only after the semantic content of the sentence uttered has been found wanting, and a reinterpretation of the speaker’s remarks is undertaken. But there is nothing in the context to force such a reinterpretation. According to the present account, the Russellian semantic content of (7b) (in

388

•

Essay Fourteen

the context) entails that one and only one student (in the math competition as a whole) solved problem 5. It is not assumed in the context that no student taking the exam in any of the other rooms—or more than one— solved that problem (or that the teacher could have no knowledge of such a student if there were exactly one). Since there is no such assumption, there will be no conversational implicature defeating the presumption that the semantic content of the sentence uttered is asserted. This creates two problems for the simple Russellian treatment of incomplete descriptions. First, the modified principle connecting meaning, semantic content, and assertion that it contains wrongly predicts that the teacher did assert the overly expansive (and potentially false) proposition semantically expressed by the sentence she uttered. Second, since no reinterpretation is forced, nothing in the theory will explain that what the teacher really asserted was a more restricted (true) proposition, along the lines of (7c). For the Russellian, the moral of the story is that the second version of our principle, the modified traditional connection between meaning, semantic content, and assertion—appealed to in the simple Russellian treatment of incomplete descriptions—is incorrect. Fortunately, however, what has failed is not something specific, or intrinsic, to Russellianism. If there are other, independent, reasons to reject the principle, and to adopt an alternative on which there is no general (defeasible) presumption that the semantic content of a sentence is asserted by normal assertive utterances of it, and no need to restrict contextual supplementation to cases of reinterpretation, then a Russellian semantics of sentences containing incomplete definite descriptions might be preserved. Of course, finding these reasons, and formulating such an alternative, may seem like a tall order. Since no account that completely divorces semantic content from assertion could possibly be correct, the challenge is to articulate an appropriately intimate connection between the two that does not carry with it even a defeasible general presumption that semantic contents of sentences uttered must be asserted by normal utterances of them. It is this to which I now turn.

A New Conception of the Connection between Meaning, Semantic Content, and Assertion The origin of the new conception in my own work is located in thoughts about the semantics of names, natural kind terms, and propositional attitude ascriptions. I have come to believe that the apparent intractability of some of the most important and long-standing problems in this area can be traced to a conception of the connection between semantic content and assertion very much like the one we have just found wanting.

Russell’s Theory of Descriptions

•

389

For that reason, I have begun to flesh out an alternative conception of the border between semantics and pragmatics in which the relationship between semantic content and assertion is substantially looser, and more indirect, than it has often been taken to be. This is the picture I will apply to the problem of incomplete definite descriptions.7 Central to the picture is a conception of meaning, or semantic content, as least common denominator. In explaining this here, I will simplify matters by restricting attention, in the first instance, to indexicalfree sentences, the semantic contents of which do not vary from context to context. If S is such a sentence, its meaning and semantic content can be identified. This meaning is what is common to what is asserted by utterances of S in all normal contexts in which it is used literally, without conversational implicatures canceling its normal force. Although the meaning, or content, of S is often a complete proposition, and, hence, a proper candidate for being asserted and believed, in certain cases—for example, those containing genitive constructions, like ‘John’s car’—it may only be a skeleton, or partial specification, of such a proposition. In many contexts, the semantic content of S— whether it is a complete proposition or not—interacts with an expanded conception of pragmatics to generate a pragmatically enriched proposition that it is the speaker’s primary intention to assert. Other propositions count as asserted only when they are relevant, obvious, necessary and a priori consequences of the speaker’s primary assertions, together with salient presuppositions of the conversational background. On this picture, our previous principle, the modified traditional connection between meaning and assertion, is replaced by the following principle. Meaning, Assertion, and Pragmatic Enrichment If M is the meaning (or semantic content) of an indexical-free sentence S, then normal, literal uses of S result in assertions of propositions that are proper pragmatic enrichments of M (provided that conversational implicatures do not cancel the normal force of the utterance, or require it to be reinterpreted). When M is a complete

7 Although the conception I will present differs in significant ways from those of others, it bears important similarities to some of ideas in the expanding literature on new kinds of pragmatic enrichment. Notable among these are those going by such names as ‘impliciture’ and ‘explicature’—which stand somewhere between the traditional semantic notion of what is said (by a sentence) and the familiar Gricean notion of implicature. A representative sample of this literature includes Bach (1994) and (2001), Bezuidenhout (1997), Carston (2002), Chierchia (2004), Horn (2005), Recanati (1993), Sperber and Wilson (1986) and Taylor (2001).

390

•

Essay Fourteen

proposition, it counts as asserted only if M is an obvious, relevant, necessary, and a priori consequence of pragmatically enriched propositions asserted in uttering S, together with salient, shared assumptions in the conversational background.8 The key notion is that of a proper pragmatic enrichment of a proposition semantically expressed by a sentence. The types of enrichment most relevant to resolving the issues in this essay involve uses of names and descriptions. First, consider the linguistically simple name ‘Carl Hempel’, which refers to a now deceased philosopher of science. When we use a name like this in a particular context, it is often pragmatically enriched. When you ask, “Were any of your neighbors in Princeton philosophers?”, and I answer, “Yes, Carl Hempel was my neighbor,” what I assert is that the philosopher, Carl Hempel, was my neighbor. What I mean by this use of the name is roughly the same as what the phrase ‘the philosopher, Carl Hempel’ means. However, this is not what the name itself means. Different speakers who use the name to refer to the same man may, and often will, associate it with widely different descriptive information. Because of this, different uses of ⎡Carl Hempel was F⎤ in different contexts will result in somewhat different assertions, due to different pragmatic enrichments. Since there is little or no substantial descriptive information common to all these enrichments, the meaning of the name (the common assertive content it carries in all relevant contexts) is simply its referent. The same is true of ‘Peter Hempel’, which is what Mr. Hempel’s friends and colleagues used to call him. Because of this, the two names mean the same thing. Hence, (8a) and (8b) have the same linguistic meaning, even though utterances of the two will nearly always assert and convey different information (Soames 2002, chap. 3). (8a) Peter Hempel was Carl Hempel. (8b) Carl Hempel was Carl Hempel. If this seems counterintuitive, it is, I think, because one is not clear about what one’s intuitions are tracking. Properly understood, the claim that S means, or semantically expresses, p is a theoretical claim about the common informational content contained in what is asserted and conveyed by utterances of S in different contexts. When ordinary speakers are asked whether two sentences mean the same thing, they standardly do not focus on the question of whether what is common to that which is asserted and conveyed in all contexts involving competent speakers by 8 It is not ruled out that sometimes there is no pragmatic enrichment. For these cases, we let semantic contents count as proper pragmatic enrichments of themselves.

Russell’s Theory of Descriptions

•

391

utterances of one of the sentences is the same as what is common to that which is asserted and conveyed by utterances of the other sentence. Instead they often focus on what they typically would use the sentences to assert and convey in various contexts, or what information they typically would gather from assertive utterances of them. In short, they often focus on whether they would typically mean the same thing by the two sentences in particular cases, rather than on whether the two sentences mean the same thing in the common language of their community. The latter is, I suggest, the sense of meaning relevant to semantic theories of natural language. This is the sense in which (8a) and (8b) have the same meaning; the information invariantly contributed by one to what is asserted and conveyed in different normal contexts by uses of it is the same as the information invariantly contributed to what is asserted and conveyed by uses of the other. This is compatible with the fact that in virtually all contexts in which the sentences might be assertively uttered, speakers in those contexts would use them to assert and convey different information. How shall we think about semantic content and pragmatic enrichment in cases like this? In general, we may take the semantic content of a sentence S to be a structured complex, the constituents of which are the semantic contents of the grammatically significant constituents of S. Thus, the proposition semantically expressed by (8a) is a complex the constituents of which are the identity relation, the pair consisting of Mr. Hempel and Mr. Hempel, and the past tense operator. Now suppose that I assertively utter (8a) in a context in which what I assert is that my neighbor Peter Hempel was the philosopher Carl Hempel, represented by (A8a). (A8a) [the x: x was my neighbor & x = m] was [the x: x was a philosopher & x = m] (where the content of ‘m’ is the man, Mr. Hempel) This asserted proposition is a proper pragmatic enrichment of the proposition semantically expressed by (8a). It arises by enriching the semantic content of each name to form a description, the content of which includes the semantic content of the name. In what follows I will look at a complementary class of cases—those in which the semantic content of a description is pragmatically enriched, often by the inclusion of a contextually salient individual. In some of these cases, I will draw attention to an important feature of the new principle of meaning, assertion, and pragmatic enrichment first explored in “Naming and Asserting” (essay 9) and extended in “The Gap between Meaning and Assertion” (essay 10). It is a consequence of this principle that when one assertively utters a sentence S, the semantic content of S is

392

•

Essay Fourteen

often not itself asserted by the speaker’s utterance—even if that content is itself a complete proposition, and the agent is speaking literally and unmetaphorically. These types of cases will be important to our final, Russellian, account of incomplete definite descriptions, and Donnellanstyle referential uses of descriptions.

Referential Uses of Names and Descriptions In Donnellan’s famous example, a speaker assertively uttering (9) uses the description it contains to pick out, and focus attention on, a certain man m, of whom the speaker predicates the property of being a famous philosopher. (9)

The man in the corner drinking champagne is a famous philosopher.

Although the (Russellian) semantic content of the sentence uttered is given by (S9), the speaker asserts something else—namely, the singular proposition expressed by (A9a). (S9)

[The x: x is a man & x is in the corner & x is drinking champagne] x is a famous philosopher (A9a) m is a famous philosopher (where the semantic content of ‘m’ is the man m himself) With the new conception of the relationship between semantic content and assertion in place, this is no threat to the Russellian analysis of definite descriptions. In this context, the fact that it is evident to all that m is the intended denotation of the description results in the pragmatic enrichment of the speaker’s utterance represented by (PE9). (PE9) [the x: x is a man & x is in the corner & x is drinking champagne & x = m] x is a famous philosopher (where the content of ‘m’ is as before) This is the primary proposition asserted by the speaker. Since the propositions expressed by (A9a) and (A9b) are obvious necessary and a priori consequences of it, they, too, count as asserted. (A9b) m is a man & m is in the corner & m is drinking champagne (with ‘m’ as before) What about (S9), the proposition semantically expressed by the sentence uttered? Although it is not a necessary consequence of the enriched proposition (PE9), it is an obvious necessary and a priori consequence of (PE9), plus the proposition expressed by (BP9).

Russell’s Theory of Descriptions

•

393

(BP9) [the x: x is a man & x is in the corner & x is drinking champagne] x = m Since in most Donnellan-type cases this proposition will be a background presupposition of the conversation, the semantic content of the sentence uttered will itself usually count as having been asserted by the speaker’s utterance. In these cases, the speaker’s referential use of the description in assertively uttering (9) results in the assertion of the propositions represented by (PE9), (A9a), (A9b), and (S9). In situations in which m is the man in the corner drinking champagne, all of these will be true, provided that m is also a famous philosopher. When there are two men in the corner drinking champagne, but it is, nevertheless, contextually obvious that the speaker is talking only about m (because the other men in the corner are already well known and m is the only new guy), the propositions— (PE9), (A9a), and (A9b)—arising from pragmatic enrichment will all be true, but the semantic content (S9) will not be. However, this need not count as a black mark against the speaker, since in cases like this (BP9) will standardly not be a background presupposition, and so (S9) will not count as asserted.9 The speaker is open to the charge of error in cases of misdescription— in which m is not a man in the corner drinking champagne. In these cases (PE9) and (A9b), and sometimes (S9), will be false. However, these culpable errors are mitigated by the fact that the asserted proposition (A9a) remains true. Since in many cases this proposition will be more important to the conversation than whether m is in fact drinking champagne, or really in the corner, the fact that the speaker has, strictly speaking, asserted one or more falsehoods, will matter less than his having asserted an important truth. These seem to be the right results. A further virtue of the account is the way it generalizes to Kripke’s case of referential misdescription involving proper names. In Kripke’s example two people see Smith in the distance and mistake him for Jones. The speaker says, “Look, Jones is raking leaves,” thereby saying something true about Smith, and also something false about Jones. These results fall out of the present approach. The speaker’s utterance of (10), in a conversation in which the proposition represented by (BP10) is a background 9

As will become clear, this idea will be extended to typical uses of sentences containing incomplete descriptions that succeed in saying something true, and nothing false. In such cases, pragmatic enrichment completes the description, allowing the speaker to assert truths. However, when the description is recognized to be incomplete, there will be no presupposition corresponding to (BP9), and the (false) semantic content of the sentence uttered standardly won’t count as asserted—because it won’t be a consequence of the pragmatically enriched propositions asserted, together with the background presuppositions of the conversation.

394

•

Essay Fourteen

presupposition, results in the assertion of several pragmatically enriched propositions, including (PE10). (10) (BP10) (PE10)

Jones is raking leaves. The man, s, seen in the distance = Jones (where the semantic content of ‘s’ is the man Smith, seen in the distance) [the x: x = Jones & x = s] x is raking leaves (where ‘s’ is as before)

Since both (S10) and (PE10a) are obvious necessary and a priori consequences of these propositions, both the semantic content of the sentence uttered, and a singular proposition resulting from pragmatic enrichment, are asserted. (S10)

j is raking leaves (where the semantic content of ‘j’ is the man Jones) (PE10a) s is raking leaves (where the semantic content of ‘s’ is the man Smith) In Kripke’s example, one of these propositions is true, and the other false, giving us the correct result that the speaker has said something true, and also something false. Thus, the new conception of the relationship between semantics and pragmatics allows us to account for referential uses of both names and definite descriptions, without departing from Russell’s theory of descriptions.

Incomplete Definite Descriptions and Other Quantifiers Next, consider the incomplete description in (4d), ‘The murderer must be insane’—which can be used either attributively, or referentially. Imagine an attributive use made after coming across the body of a victim v. Whereas the semantic content of (4d) has the truth conditions of (S4d), the speaker’s utterance of it results in the assertion of the pragmatically enriched proposition represented by (PE4dA). (S4d) (PE4dA)

[the x: x is a murderer] x must be insane [the x: x is a murderer of v] x must be insane (where the content of ‘v’ is the victim)

Since the semantic content represented by (S4d) is not a necessary, or an a priori, consequence of the pragmatically enriched proposition asserted (together with the presuppositions of the conversation), it does not count as asserted. Hence, if the murderer of v really is insane, then the speaker counts as asserting something true, and nothing false—exactly as it should be.

Russell’s Theory of Descriptions

•

395

Next consider a referential use of the description in which (4d) is assertively uttered in a context in which a certain man m has correctly been identified as the murderer of v, and the speaker intends to say of m that he must be insane. As before, the semantic content of (4d) is (S4d). However, the pragmatically enriched proposition asserted is either (PE4dR1) or (PE4dR2), depending on whether the victim also has been identified. (PE4dR1) [the x: x is a murderer & x = m] x must be insane (where the content of ‘m’ is the man identified as the murderer) (PE4dR2) [the x: x is a murderer of v & x = m] x must be insane (where ‘m’ is as before, and the content of ‘v’ is the victim) Since the semantic content of the sentence uttered is not a consequence of these assertions (plus the presuppositions of the conversation), it does not count as asserted—just as before. By contrast, the simple singular proposition represented by (PE4dR+) does count as asserted in both cases, since in each case it is an obvious necessary and a priori consequence of the primary, pragmatically enriched, proposition asserted. (PE4dR+)

m must be insane

Thus, when the murderer m really is insane, the speaker is again correctly characterized as asserting truths, while saying nothing false. The same points are illustrated by a case in which (4c) is used to say something about a particular student s. Here the semantic content has the truth conditions of (S4c). (S4c) [the x: x is a student] x got into an argument with a student from another school Since the truth of (S4c) requires (i) that there be exactly one student (in the domain of discourse), and (ii) that that student got into an argument with a (different) student from another school (also in the domain), (S4c) must be false. However, this is of no concern to the speaker. Since (S4c) is not a consequence of the pragmatically enriched proposition (PE4c) that is asserted, the false semantic content is not something the speaker asserts. (PE4c) [the x: x is a student & x = s] x got into an argument with a student from another school (where the content of ‘s’ is the student the speaker is talking about) However, the propositions (A4c1), (A4c2), and (A4c3) are consequences of (PE4c), and so are asserted. (A4c1) s got into an argument with a student from another school (with ‘s’ as before)

396

•

Essay Fourteen

(A4c2) A student got into an argument with a student from another school. (A4c3) Two students got into an argument. In this way, we not only get the correct result that the speaker has used the description ‘the student’ to say of a contextually salient individual that he or she got into an argument with a student from another school, we also capture the two further assertions that proved problematic to the semantically referential theory. We also handle the crucial example (7b)—‘The student who solved problem 5 will compete in the finals with other high-scoring students in the competition’—which undermined the simple Russellian treatment of incomplete descriptions, formulated earlier. Since the semantic content of this sentence is not a necessary or a priori consequence of the pragmatically enriched proposition—that the student in this group who solved problem 5 will compete in the finals with other high-scoring students in the competition—that is asserted, the semantic content is not asserted by the speaker’s utterance. The crucial point is that the new principle of meaning, assertion, and pragmatic enrichment includes no general presumption that the semantic content of the sentence uttered is asserted, unless there is a conversational implicature to the contrary. Hence, the fact that there is no such implicature in this case is no obstacle to getting the right result. The account also generalizes to examples like (5a) and (5b)—‘No one knows that you and I are here’ and ‘Everyone looks up to Larry’—which contain other quantifiers. Although the semantic contents of these sentences are false in the contexts in which they are uttered, these contents are not asserted because they are neither necessary nor a priori consequences of the pragmatically enriched propositions—that no one but us knows that you and I are here, and that every one of us or of them look up to Larry—that the sentences are used to assert. Thus, our explanation generalizes, as it should.

Is This Solution to the Problem of Incomplete Descriptions Truly Russellian? This completes my defense of the Russellian analysis of definite descriptions as generalized quantifiers. It is, admittedly, somewhat programmatic, relying, as it does, on a new and not yet fully developed conception of the relationship between semantic content and assertion. For that reason, any final verdict on it must await further work on several fronts—including the articulation of precise theories of how pragmatic

Russell’s Theory of Descriptions

•

397

enrichment takes place, and the formulation of substantive constraints on what counts as a proper pragmatic enrichment of the semantic content of a sentence. Nevertheless, I hope to have made the general approach plausible, and, in so doing, to have illustrated how the seminal ideas presented a century ago by Russell in “On Denoting” continue to guide and inform the contemporary study of language. With this heritage in mind, I close by asking how Russellian my defense of Russell really is. Central to the defense is my conception of the meaning of a (nonindexical) sentence in a language as the least common denominator abstracted from that which speakers mean and assert by utterances of it in all normal contexts. When meaning is understood in this way, a gap is created between what a sentence means and what it is used to assert that allows pragmatic features of contexts of utterance to interact with semantic contents of sentences uttered in systematic ways to determine what is, and what is not, asserted. Because incomplete descriptions are subject to routine pragmatic enrichment without affecting their austere Russellian semantics, the problems they have traditionally been thought to pose for Russell can, I think, be solved. That said, I cannot claim my defense of Russell to be thoroughly Russellian. My conception of the meaning, or semantic content, of a sentence S as the least common denominator of what speakers in different contexts use S to assert cannot, I am afraid, be found anywhere in Russell. Worse, it seems to be quite different from his standard way of looking at things. As I read him, the project of giving a theory of the meaning of a sentence in the common language of a linguistic community, along the lines that I have suggested, was not one of his primary concerns. When he talked of meaning, he mostly had in mind what an individual means by his or her use of a sentence, or an expression, at a given time. From this perspective, the problems posed by so-called incomplete descriptions are not very serious. Since it is routine for speakers to pragmatically enrich their utterances, what they mean by ⎡the F⎤ is typically not incomplete, when applied to the contexts in which they utter sentences containing it. There is, I think, an interesting lesson here. The supposed problem of incomplete definite descriptions is, in fact, not a serious problem for broadly Russellian analyses of definite descriptions. It seems to be a problem only if one is confused about, and runs together, two different candidates for the objects of analysis—roughly, what speakers mean by utterances of sentences containing such descriptions, and what those sentences mean in the common language. Given either choice, Russell’s theory works very well. But, whichever choice one makes, one must carry it through consistently. If one is analyzing what speakers mean and assert, one must recognize that the assertive content of ⎡the F is G⎤—though broadly Russellian—differs widely from one context to the next, and

398

•

Essay Fourteen

often is considerably richer than the semantic content (in the context) of the sentence in the common language.10 Something like this was, I think, Russell’s preferred perspective. However, one can also take the perspective of a semantic theory of sentences in the common language. As I have argued, Russell’s theory of descriptions works well from this perspective, too. However, if one adopts this perspective, one must not assume that when speakers assertively utter S, they mean, and assert, the proposition that is the semantic content of S (in the context). If this is right, then the problem for Russell posed by incomplete descriptions is really a pseudoproblem that arises primarily from muddying these two perspectives by indiscriminately mixing them together.

References Bach, Kent. 1994. “Conversational Impliciture.” Mind and Language 9:124−62. ———. 2001. “You Don’t Say?” Synthese 128:15−44. Barwise, Jon, and John Perry. 1983. Situations and Attitudes. Cambridge: MIT Press. Bezuidenhout, Anne. 1997. “Pragmatics, Semantic Underdetermination and the Referential/Attributive Distinction.” Mind 106:375−410. Carston, Robyn. 2002. Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell. Chierchia, Gennaro. 2004. “Scalar Implicatures, Polarity Phenomena, and the Syntax/Pragmatics Interface.” In Structures and Beyond, ed. Adriana Belletti, 39−103. Oxford: Oxford University Press. Donnellan, Keith S. 1966. “Reference and Definite Descriptions.” Philosophical Review 75:281−304. Horn, Laurence R. 2005. “The Border Wars: A Neo-Gricean Perspective.” In Where Semantics Meets Pragmatics, ed. Ken Turner and Klaus von Heusinger, 21−48. Amsterdam: Elsevier. Kripke, Saul A. 1977. “Speaker’s Reference and Semantic Reference.” In Contemporary Perspectives in the Philosophy of Language, ed. Peter A. French, Theodore E. Uehling, and Howard K. Wettstein, 6−27. Minneapolis: University of Minnesota Press.

10 In characterizing these assertive contents as broadly Russellian, I am here ignoring Russell’s problematic, epistemologically driven, views that drastically restrict the sorts of things with which we can be acquainted, and which can enter into propositions we are capable of entertaining. See Soames (2003, vol. 1, chap. 5) for discussion. Among the other problems posed by Russell’s extreme epistemology is that it would not allow contextual supplementation of the content of incomplete descriptions by ordinary objects—people, places, and things. Since this is needed to account for many ordinary uses of descriptions, my defense of Russellianism is a defense of his semantics, divorced from his epistemological doctrines of acquaintance.

Russell’s Theory of Descriptions

•

399

Lewis, David K. 1979. “Scorekeeping in a Language Game.” Journal of Philosophical Logic 8:339−59. Neale, Stephen. 1990. Descriptions. Cambridge: MIT Press. Recanati, François. 1993. Direct Reference. Oxford: Blackwell. Russell, Bertrand. 1905. “On Denoting.” Mind 14:479−93. Salmon, Nathan. 1982. “Assertion and Incomplete Descriptions.” Philosophical Studies 42:37−45. Soames, Scott. 2002. Beyond Rigidity: The Unfinished Semantic Agenda of “Naming and Necessity.” New York: Oxford University Press. ———. 2003. Philosophical Analysis in the Twentieth Century. 2 vols. Princeton: Princeton University Press. Sperber, Dan, and Deirdre Wilson. 1986. Relevance: Communication and Cognition. Cambridge: Harvard University Press. Taylor, Ken. 2001. “Sex, Breakfast, and Descriptus Interruptus.” Synthese 128:45−61. Wettstein, Howard K. 1981. “The Semantic Significance of the ReferentialAttributive Distinction.” Philosophical Studies 44:241−57.

This page intentionally left blank

PA RT F I V E

Meaning and Use: Lessons for Legal Interpretation

This page intentionally left blank

ESSAY FIFTEEN

Interpreting Legal Texts: What Is, and What Is Not, Special about the Law

How is the content of positive law related to its authoritative sources—including written constitutions, statutes, and administrative rules, understood in light of the beliefs, intentions, and presuppositions of those who produced them, and those to whom they are addressed? Progress can, I think, be made in answering this question by seeing it as an instance of the more general question of what determines the contents of ordinary linguistic texts. I will, therefore, look at recent advances in semantics and pragmatics, and extract lessons for legal interpretation by tracking implications for different kinds of “hard cases” in the law. I will call a legal case genuinely hard iff its (legally correct) outcome is not determined by all nonlegal and nonmoral facts plus the linguistically based content of the relevant legal texts—including everything asserted and conveyed therein. These cases divide into three types: (i) those in which the texts say too little to produce any result, (ii) those in which the texts are inconsistent, and thus generate contradictory results, and (iii) those in which the texts yield a single result, which is, paradoxically, legally incorrect. In a genuinely hard case, a correct outcome can be reached only by an innovative judicial decision—which, effectively, creates new law. In contrast, a case is semantically hard iff the meanings of the relevant legal texts, plus all nonlegal and nonmoral facts, fail to determine its (legally correct) outcome. The distinction between semantically, and genuinely, hard cases turns on the distinction between the semantic contents of legal texts and their complete, linguistically based contents. Appreciating this distinction involves grasping several important lessons. First, literal meaning is more austere, abstract, and less transparent than it is often taken to be. Thus, even if fidelity to the text is the touchstone of interpretation, a text’s meaning is often not readily identifiable. Second, meaning is not what faithful interpreters should be looking for anyway—since even when it is identified, it may fail to determine the text’s content. That content, which encompasses everything conveyed or asserted by the text, often includes information that goes beyond the semantic contents of the sentences involved.

404

•

Essay Fifteen

Typically, an agent produces a sentence in a context with a communicative goal and topic, a record of what has been supposed or established up to then, and assumptions about the beliefs and intentions of participants. This pragmatic information interacts with the semantic content of the sentence to add content to the discourse. In recent years, we have learned that the pragmatic determinants of this content are not minor add-ons to semantic content. Semantic content is often merely a vehicle for getting to pragmatically enriched content, and sometimes the semantic content of a sentence is not itself asserted, or even included in what the speaker is committed to. The semantic-cum-pragmatic information-generating process governing the routine interpretation of linguistic texts and performances may start with literal meaning, but it doesn’t end there. The import of this for judicial interpretation is that the slide from semantically hard cases to genuinely hard ones should be resisted. It’s true that the meanings of legal texts, plus the facts of a case, often fail to determine its outcome. But this shouldn’t be taken to show that the content of the law embodied in those texts doesn’t determine the outcome, and mustn’t be used to invite judicial legislation. Just as what I say, and commit myself to, by uttering a sentence, is often a function of more than its semantic content, so “what the law says,” and is committed to, is often a function of more than the semantic contents of relevant legal texts. Just as you have no standing to reinterpret my remark to conform to your moral and political views, simply because the meaning of my sentence doesn’t fully determine the content of my remark, so judges applying the law have no standing to reinterpret it, simply because the linguistic meanings of the relevant legal texts don’t fully determine the content of the law. There are other principles at work filling the gap between sentence meanings and the contents of texts, legal or otherwise. Finally, even when a case is genuinely hard—because the full linguistic content of authoritative texts fails to determine a correct outcome—there are further principles to consult that routinely guide the interpretation of incomplete, inconsistent, or otherwise defective linguistic materials. Applicable in both legal and nonlegal contexts, these principles constrain the resolution of hard cases, and limit the moral and political discretion of judges in making new law. In what follows I will illustrate these points.

The Austerity and Nontransparency of Meaning One of the chief results of recent years is the nondescriptionality and indefinability of many terms once thought to be definable—ordinary

Interpreting Legal Texts

•

405

proper names and natural kind terms, among them. Although these expressions may carry descriptive information, that information doesn’t define them. For example, virtually everyone who uses the words ‘Bill Clinton’ and ‘Washington, D.C.’ knows that the former names a past president of the United States, and the latter names the capital of the country. Similarly for ‘water’ and ‘gold’. One would be hard pressed to find speakers who didn’t know that water is a clear, drinkable liquid that falls from the sky, or that gold is a valuable yellow metal. However, this information doesn’t define the terms. A childhood friend of Clinton, marooned on a desert island for forty years, would, on returning, still understand the name, use it to refer to the same man we do, and successfully communicate with us in so doing—despite not knowing what most people regard as the most important things about Clinton. Although users of a name standardly associate some information with it, this information varies from speaker to speaker, and virtually none of it is required for mastering the name. Thus, the only semantic content left for it is its referent—which explains why it is neither an a priori nor a necessary truth that Bill Clinton is (if he exists) a former president of the United States, as it would be if the name were definable in terms of common knowledge about Clinton. These points carry over to natural kind terms, which are really just names of kinds, rather than individuals. The “definition” of ‘water’ is not, ‘The clear, drinkable, liquid that falls from the sky, etc.’, but ‘That stuff’, said pointing at water. We can, of course, be pretty sure that a man in the desert asking for water, wants a drink, and that a farmer who checks for rain knows that it will water his crop. However, truths like ‘Water is a clear, drinkable liquid that falls from the sky’ are neither necessary nor a priori, as they would be if ‘water’ were descriptively definable. Nor must such truths be known in order to understand the word. An unfortunate, whose only knowledge of water was of the cloudy stuff draining from a nearby laundry, could use and understand the word, based on his limited acquaintance with water, just as our modern-day Robinson Crusoe could use, and understand, the name ‘Bill Clinton’ based on his acquaintance with Clinton. To be acquainted with a natural kind is to be acquainted with its instances. What makes a kind natural is that observable properties of its instances share a common explanation. In introducing a general term for such a kind, we stipulate that it applies to certain encountered particulars, and that other particulars will count as instances iff they share whatever features causally explain the properties observed in the original sample. For ‘water’, ‘gold’, and ‘green’, these features turn out (roughly) to be having the molecular structure H2O, having atomic number 79, and having a certain spectral reflectance property. These causal-explanatory

406

•

Essay Fifteen

features of the original samples determine the extension of the terms, even though they were initially unknown, and needn’t be known now in order to understand the terms. Thus the reference of such a term isn’t determined by satisfaction of descriptive conditions encoded in its meaning. It has no such meaning. Rather, its meaning is the kind it names, and its reference is determined by agreed upon instances, plus a special similarity relation, holding between instances of the kind, the specific content of which needn’t be known, even by competent speakers. A modified version of this picture extends beyond natural kind terms to other indefinable terms, including Wittgenstein’s example of the term ‘game’. Although there is no one thing that all and only games share, there are paradigmatic examples of games, plus a similarity relation such that anything that bears it to something that clearly is a game itself counts as a game. In this case, however, the similarity relation is not a causal-explanatory one, but something more closely tied to speakers’ interests and priorities. The relevance of these points to the law is illustrated by two legal cases. The first is PGA v. Martin, in which the question is whether the Americans with Disabilities Act requires the Professional Golfers Association to waive its rule requiring all tournament competitors to walk the course. The Supreme Court ruled (i) that the act does apply to the PGA, and (ii) that the PGA is so required because the rule change needed to accommodate the disabled doesn’t affect golf’s essential, or fundamental, nature. Justice Scalia dissents, correctly, in my opinion, on (i), but incorrectly on (ii), which is the only part of the case I will discuss. His position seems to be that the question of what is essential to golf—beyond the rules codified by its authoritative rule-making body—is nonsensical. This, I believe, is an error. The crucial points, as I see them, are these: First, the word ‘golf’, like ‘game’, can’t be defined by specifying necessary and sufficient conditions for its application. Rather, its meaning is the activity it designates, which is a kind, or type ( just as the World Series is a kind or type), instances of which are sequences of events of a certain sort. Second, although the activity golf is “defined” or “constituted” by its rules, the sense in which this is so isn’t that of defining a word. You can know what the word ‘golf’ means, as well as what the activity golf is, without knowing all the rules. Also, the meaning and the activity can both survive rule changes. The word ‘baseball’ didn’t change meaning, or reference, when the American League adopted the designated hitter rule—and the Red Sox didn’t cease playing the game and start playing something else, when that happened. Similarly, ordinary golfers don’t cease playing golf when the seasons change, and they switch from so-called summer to winter rules.

Interpreting Legal Texts

•

407

In what sense, then, is golf constituted by its rules? In the sense that each round of golf is made up of actions guided by the rules. Just as Admiral Nelson’s ship Victory was constituted by the planks of wood, pieces of iron, sheets of canvas, and strands of rope that made it up, so a round of golf is constituted by the rule-guided actions that make it up. The Victory was constituted by its parts, even though an inventory of its parts doesn’t give the meaning of the name, and having precisely those parts isn’t essential to the ship (since it could have existed with slightly different ones). The activity golf is similarly constituted by its rules, even though they don’t give the meaning of the word, and having precisely those rules isn’t essential to golf (since the rules can vary a bit, without the activity being lost). Thus, there was nothing absurd about the Court’s finding that one of the rules—namely that tournament players must walk the course—isn’t essential to golf. Scalia’s mistake was, I think, due to one or both of two errors. The metaphysical sense in which the rules of golf constitute the activity (without being essential to it) must not be confused with the linguistic sense in which certain conditions define a word (and hence are essential to its meaning). If one conflates these two, one may wrongly think that since the rules constitute golf, they are essential to what we mean by ‘golf’ and so cannot be changed without altering our conception of the game. The second error is to run together the correct Wittgensteinian observation—that when questions of proper application arise for words like ‘golf,’ there is no higher authority than its use by competent speakers—with the incorrect idea that the PGA’s authority over its tournaments makes its rules committee the arbiter of competence in determining the application of the word. These errors illustrate how confusion about meaning can adversely affect legal reasoning. Another legal example illustrating the austerity and nontransparency of meaning is Nix v. Hedden. The case turns on whether tomatoes count as vegetables rather than fruits, and so are subject to a tariff on vegetables that excludes fruits. The issue is whether the meaning of the word ‘fruit’ excludes all vegetables, while including the edible “ripened seedbearing ovary of the plant,” as my dictionary puts it—thereby precluding tomatoes, peas, beans, eggplants, cucumbers, squash, and peppers from being vegetables. The court found otherwise, ruling that ‘vegetable’ and ‘fruit’ have no specialized meaning in commerce, and so must be understood in their ordinary meaning, according to which tomatoes, along with the others, are vegetables. Botanically speaking, tomatoes are fruit of a vine, just as are cucumbers, squashes, beans and peas. But in the common language of the people . . . all these are vegetables, which are grown in kitchen gardens, and which, whether eaten cooked or raw, are, like potatoes, carrots,

408

•

Essay Fifteen

parsnips, turnips, beets, cauliflower, cabbage, celery and lettuce, usually served at dinner, with, or after the soup, fish or meats which constitute the principal part of the repast, and not, like fruits generally, as desert. (My emphasis) In my view, the court’s decision was correct, and its reasoning defensible. It correctly notes that the term ‘fruit’ has a botanical meaning encompassing any edible “fruit of the vine,” containing the seed-producing part of the plant. On this meaning, it is a natural kind term the extension of which includes tomatoes, beans, and peas. Less obviously, but still plausibly, the court suggests that ‘fruit’ also has an ordinary meaning in which it contrasts with ‘vegetable’, both of which being “defined” by paradigmatic instances—including, one imagines, oranges, lemons, plums, and melons, in one case, and carrots, cucumbers, lettuce, and tomatoes, in the other. The way to think of these ordinary meanings is, of course, not in terms of verbal definitions, but along the lines suggested by Wittgenstein’s example of games. The end result is a plausible analysis of the statutory language, from which the court’s decision follows, given the reasonable assumption that the words are to be taken in their ordinary meaning. Suppose however—for the sake of argument—that this view of language is incorrect, because ‘fruit’, like other natural kind terms, has only a single “scientific” meaning—the “fruit of the vine” one. Suppose further that vegetables are, by definition, edible parts of plants that aren’t fruits. On this view, the common classification of beans, peas, peppers, and tomatoes as vegetables is not evidence of linguistic ambiguity, but a mistake based on ignorance (on a par with ignorance of the molecular structure of water). It would, therefore, follow that a literal interpretation of the statutory language excludes tomatoes from the tax. Would it also follow that the court’s decision was legally incorrect—or, if not incorrect, correct only by virtue of a justifiable substitution of the court’s superior wisdom for that of the legislature? I don’t think so. However, to explain why, I must turn to my second linguistic lesson for legal interpretation.

The Gap between the Content of a Text and the Meaning of Its Sentences The main point is that the semantic content of a sentence doesn’t always determine what is asserted and conveyed by literal uses of it. Sometimes more than the semantic content is asserted or conveyed, and sometimes the semantic content isn’t asserted at all. Since the content of the law in-

Interpreting Legal Texts

•

409

cludes everything asserted and conveyed in adopting the relevant legal texts, meaning is sometimes merely a guide to interpretation, to be supplemented by other things. Consider, for example, definite descriptions, ⎡ the F⎤. The basic semantic fact about these expressions is that a sentence ⎡ The F is G⎤ expresses a truth iff there is a single thing satisfying F (in the relevant circumstance), and that thing also satisfies G. The basic pragmatic fact is that these descriptions are often used referentially, to focus attention on a particular individual, about whom the speaker makes a statement. These points are illustrated by (1a), the literal meaning of which is given by (1b).1 (1) a. The man in the corner drinking champagne is a famous philosopher. b. [the x: x is a man & x is in the corner & x is drinking champagne] x is a famous philosopher When the description in (1a) is used referentially, the speaker focuses attention on a particular man m. Since it is clear to all that m is the intended referent, this information combines with the meaning of (1a) to produce (1c), which is the primary proposition asserted by the utterance. (1) c. [the x: x is a man & x is in the corner & x is drinking champagne & x = m] x is a famous philosopher (where the content of ‘m’ is the man it designates) Since (1d) is an obvious consequence of (1c), it, too, counts as asserted. (1) d. m is a famous philosopher What about the semantic content (1b) of the sentence uttered? Although it isn’t a necessary consequence of the primary assertion (1c), it is an obvious consequence of (1c), plus the proposition (Pre), identifying m as the man in the corner drinking champagne. (Pre) [the x: x is a man & x is in the corner & x is drinking champagne] x = m In many contexts, this will be presupposed by speaker-hearers. When it is, the semantic content of (1a) will also count as asserted. When it isn’t, the semantic content won’t be asserted, but rather will serve merely as raw material from which the real assertion is constructed. Thus, when there are two men in the corner drinking champagne, but it is obvious that the speaker is talking only about (the new guy) m, (Pre) won’t be 1 See Donnellan (1966) for the original discussion. Also see essays 13 and 14 in this volume.

410

•

Essay Fifteen

presupposed, and (1b), won’t be asserted—though (1c) and (1d) will. In this case the speaker says something true, and nothing false, even though the semantic content of her sentence is false. She is, of course, open to the charge of error in cases of misdescription—in which m is not a man in the corner drinking champagne. In these cases (1c) will be false. However, this culpable error may be mitigated, if (1d) is true. Since in many cases that assertion will be more important to the conversation than whether m is in fact drinking champagne, or really in the corner, the fact that the speaker has, strictly speaking, said something false will matter less than her having stated an important truth. With this in mind, let’s return to Nix v. Hedden, and the linguistic controversy about the terms ‘fruit’ and ‘vegetable’. When we left off, we were considering the hypothesis that ‘fruit’ has only the single, botanical meaning, “fruit of the vine,” and that vegetables are, by definition, edible parts of plants that aren’t fruits. On this hypothesis, the literal meaning of the statutory language excludes tomatoes from the tax. Nevertheless, I maintained, even if this were so, it still wouldn’t follow that the court had either decided the case incorrectly, or correctly substituted its own amended statute for the one passed by the legislature. It should now be clear why. The content of the statute is what the lawmakers asserted and committed themselves to, in adopting the statutory language. If, as seems plausible, they were using the terms ‘fruit’ and ‘vegetable’ referentially—on analogy with referential uses of descriptions—then they thereby asserted, among other things, that tomatoes, cucumbers, and peas are subject to the tax, even if they misdescribed them. If, in addition, this assertion was their central legislative point, the court’s decision was both justified and deferential. This is just one example of how the meaning of a sentence may differ from what is asserted by uttering it. There are many ways to create a gap between meaning and assertion. Here are some obvious ones: (a) Driving an out-of-towner to a destination in Los Angeles, I say, “This is going to take some time,” or “It’s some distance from here.” In so doing, I assert that this is going to take a substantial amount of time, or that it is a considerable distance from here— not that this is going to take an amount of time greater than zero, or that the distance is greater than zero, which are the semantic contents of the sentences uttered.2 (b) A doctor examining a gunshot wound says to the patient, “You aren’t going to die,” thereby asserting that the patient isn’t going to die yet, or from this, not that death will never come.3 2 3

Examples like these are given in Carston (1988). The example comes from Bach (1994).

Interpreting Legal Texts

•

411

(c)

Calling my friend from the airport I say, “I have arrived,” thereby telling him that I have arrived in Boston, not that I have arrived somewhere. When he asks, “Have you eaten?”, I take him to be asking whether I have eaten dinner this evening, not whether I have ever engaged in the activity of eating.4 (d) When I say, “I have two children,” in response to “How many children do you have?” what I assert is that I have exactly two. When I say, “Yes, I have two beers,” in response to your suggestion that we have a drink together, expressed by your question, “Do you have two bears in the fridge?” what I assert is that I have at least two. When I say, “Students are required to take three classes and allowed to take five” in response to, “How many classes are students allowed to take?”, what I assert is that they are required to take at least three, and allowed to take from three up to five. When I say, “I live four hundred miles from San Francisco” in response to, “How far do you live from San Francisco?”, what I assert is that I live approximately four hundred miles from San Francisco. ‘Exactly’, ‘at least’, ‘up to’, ‘approximately’—these are additions to what is asserted, generated by special features of the context of utterance, over and above semantic content.5 In all these cases, what is communicated isn’t the semantic content of the sentence uttered, but something richer, to which meaning and obvious background assumptions have both contributed.6 Another linguistic construction—involving sentences that report the use of something—brings us to a well-known legal case about which Stephen Neale (2007) has illuminatingly written. When one uses something, one always uses it in some way, to do some thing. But in reporting that use, how it has been used, and even what it has been used for, isn’t always explicitly mentioned. For example, what is explicitly said using the sentences in (2), can be implicitly said using (3)—provided that the context makes clear what’s being taken for granted. (2) a. I used a hammer as a tool to pound in the nails. b. I used a hammer as a weapon to fight off the burglar. c. I used a hammer as a doorstop to keep to the door from closing. (3) I used a hammer. In each case, the phrase ‘used a hammer’ carries its ordinary meaning. In (2), this meaning is supplemented by the ordinary meaning of the phrase 4

Again, see Bach (1994). For systematic discussion, see essay 11 in this volume. 6 For a more comprehensive overview of this conception of the relationship between meaning, semantic content, and assertion see essay 10 in this volume. 5

412

•

Essay Fifteen

following it. In (3), the extra information comes from context. Nevertheless, what the speaker says may be the same in both cases. Of course, it’s also possible to report that one used a hammer, while giving no information about how it was used—in which case one might either utter (4), or utter (3) in a context in which it is clear that (4) is all it is being used to say. (4) I used a hammer to do something. With this in mind we turn to Smith v. the United States. The question is whether petitioner’s attempt to trade a gun for drugs constituted a use of a firearm during (what was acknowledged to be a) drug-trafficking crime. The Supreme Court ruled that it did. Here is the statutory language: Whoever, during and in relation to any crime of violence or drug trafficking . . . uses or carries a firearm, shall, in addition to the punishment provided for such crime of violence or drug trafficking crime, be sentenced to imprisonment for five years . . . The majority opinion reads, in part (italics in all quotations are mine): Petitioner argues that exchanging a firearm for drugs does not constitute “use” of the firearm within the meaning of the statute. . . . In essence, petitioner argues that he cannot be said to have “used” a firearm unless he used it as a weapon, since that is how firearms most often are used. . . . We confine our discussion to what the parties view as the dispositive issue in this case: whether trading a firearm for drugs can constitute “use” of the firearm within the meaning of [the statute]. When a word is not defined by statute, we normally construe it in accord with its ordinary or natural meaning. . . . Surely petitioner’s treatment of his [gun] can be described as “use” of the firearm within the every day meaning of that term. Petitioner “used” his [gun] in an attempt to obtain drugs by offering to trade it for cocaine. The Court here tacitly assumes that the content of the statute—what it says about additional penalties—is given by its meaning, which is identified as the ordinary or natural meaning of its words. Petitioner makes the same assumption, while claiming that the ordinary meaning of ‘use a gun’ is to use it for the purpose for which it is normally intended—as a weapon. As the majority puts it: In petitioner’s view, [the text] should require proof not only that the defendant used the firearm but also that he used it as a weapon. . . . Petitioner and the dissent . . . contend that the average person on the street would not think immediately of a guns-for-drugs trade as an

Interpreting Legal Texts

•

413

example of “using a firearm.” Rather, that phrase normally evokes an image of the most familiar use to which a firearm is put—use as a weapon. Petitioner and the dissent therefore argue that the statute excludes uses where the weapon is not . . . employed for its destructive capacity. . . . The dissent announces its own, restrictive definition of “use.” “To use an instrumentality,” the dissent argues, “ordinarily means to use it for its intended purpose” . . . Having framed the issue as that of determining the ordinary meaning of the phrase ‘uses a firearm’, the majority holds that those who use a firearm as a weapon don’t exhaust those to which the phrase truly applies, which also include those who merely trade it for something. Thus, the Court concludes, the ordinary meaning is controlling, and the petitioner is subject to the additional penalty. Against this, Justice Scalia argues in dissent: In the search for statutory meaning, we give nontechnical words and phrases their ordinary meaning. . . . To use an instrumentality ordinarily means to use it for its intended purpose. When someone asks “Do you use a cane?” he is not inquiring whether you have your grandfather’s silver-handled walking-stick on display in the hall; he wants to know whether you walk with a cane. Similarly, to speak of “using a firearm” is to speak of using it for its distinctive purpose, i.e., as a weapon. To be sure, “one can use a firearm in a number of ways,” . . . including as an article of exchange . . . but that is not the ordinary meaning of ‘using’ the one or the other. The Court asserts that the “significant flaw” in this argument is that “to say that the ordinary meaning of ‘uses a firearm’ includes using a firearm as a weapon” is quite different from saying the ordinary meaning “also excludes any other use.” The two are indeed different—but it is precisely the latter that I assert to be true. The ordinary meaning of “uses a firearm” does not include using it as an article of commerce. I think it perfectly obvious, for example, that the objective falsity requirement for a perjury conviction would not be satisfied if a witness answered “no” to a prosecutor’s inquiry whether he had ever “used a firearm,” even though he had once sold his grandfather’s Enfield rifle to a collector. The striking thing about this case is how a shared conflation of the meaning of the statutory language with the content of the resulting statute leads to an incoherent debate in which the dissent and majority talk past each other. The majority is right to maintain (i) that the phrase ‘uses a firearm’ occurs in the text with its ordinary, literal meaning, and (ii) that this meaning doesn’t preclude use other than as a weapon. However, the

414

•

Essay Fifteen

majority is wrong to think that this shows that the content of the statute—i.e., the law that the statutory language was used to enact— covers uses of firearms other than as weapons. By contrast, Scalia is wrong in claiming that the ordinary meaning of the phrase excludes uses of firearms for sale or trade. The ordinary meaning is silent about the manner of use. Thus, when the phrase occurs in a sentence, the resulting assertion must be completed—either by the content provided by an explicit phrase (in the manner of (2)), or by pragmatically supplied content from the context of utterance (in the manner of (3)). Since the latter option was employed by Congress, the job of the Court was to infer what Congress asserted from the incomplete semantic content provided by the statutory language. For this, Scalia’s examples are telling. What would one be querying if one asked, “Do you use a cane?” Unless something special about the context indicated otherwise, one would be asking whether you use a cane for walking. What would one be asserting, on the witness stand, if one said “No,” to the question, “Have you ever used a firearm?” Not that one had never sold a firearm, but—unless something special in the context signaled something different—that one had never fired, or used a firearm as a weapon. Similarly, unless there is something in the congressional record not explicitly mentioned by the majority, what Congress should be taken to have asserted is that an additional penalty will be added to crimes in which a firearm is used as a weapon. The central confusion in the Supreme Court’s reasoning—between the meaning of the statutory language and the content of law enacted using that language—also marred the decision of the Circuit Court of the District of Columbia, when the case was heard there—as is evident from the following passage, quoted with approval by the Supreme Court majority. It may well be that Congress, when it drafted the language of [the statute] had in mind a more obvious use of guns in connection with a drug crime, but the language [of the statute] is not so limited; nor can we imagine any reason why Congress would not have wished its language to cover this situation. Whether guns are used as a medium of exchange for drugs sold illegally or as a means to protect the transaction or dealers, their introduction into the scene of drug transactions dramatically heightens the danger to our society. The Circuit Court seems to be saying (i) that Congress may well have understood itself to be prohibiting only a narrow range of uses of a firearm, but (ii) that even if that was its understanding, that isn’t what it achieved, because the language it used doesn’t explicitly rule out a broader class of cases, and (iii) that the Circuit Court itself thinks there

Interpreting Legal Texts

•

415

is good reason for wanting the statute to apply more broadly. However, (ii) is confused. If the Congress did understand itself to be asserting that an additional penalty is attached to a narrow range of cases, then the fact that the language used to do this could have been used (in another context) to assert that the penalty attached to a broader range of cases is irrelevant. The content of the statute is what Congress did say, using that language, not what it could, and in the opinion of the Circuit Court should, have said.

Constraints on the Resolution of Genuinely Hard Cases So much for my first two lessons—the austerity of meaning and the gap between the content of a text and the meaning of its sentences. Given these, one can distinguish genuinely hard cases in the law from those that are merely semantically hard. In the latter, legally correct outcomes aren’t determined by the meanings of the authoritative texts, plus the relevant facts. In the former, they aren’t determined by the facts plus the full linguistic contents of the texts, encompassing everything asserted or stipulated in adopting them. Roughly put, a genuinely hard case is one in which the totality of facts plus pre-existing law doesn’t determine a correct outcome. One subtype of this sort occurs when an outcome is determined, but it is legally incorrect. At first, this may sound incoherent. If facts plus the controlling legal texts really determine a single outcome, what could it mean to say that it is incorrect? The outcome could, of course, be morally or politically suboptimal, but that’s beside the point. Courts do not have a general mandate to correct moral and political imperfections, and an imperfect law is still the law. No, the subtype is one in which a single outcome is determined by the full linguistic contents of the authoritative legal texts, but that outcome is legally incorrect—and not because it conflicts with other, equally authoritative texts. For if that were so, then the law as a whole wouldn’t determine a single outcome—which, by hypothesis, it does. A toy example is provided by Lon Fuller’s fable about sleeping in the train station. Let us suppose that in leafing through the statutes, we come upon the following enactment: “It shall be a misdemeanor, punishable by a fine of five dollars, to sleep in any railway station.” We have no trouble in perceiving the general nature of the target toward which the statute is aimed. Indeed, we are likely at once to call to mind the picture of a disheveled tramp, spread out in an ungainly fashion on one of the

416

•

Essay Fifteen

benches of the station, keeping weary passengers on their feet and filling their ears with raucous and alcoholic snores. . . . Now let us see how this example bears on the ideal of fidelity to law. Suppose I am a judge, and that two men are brought before me for violating this statute. The first is a passenger who was waiting at 3 A. M. for a delayed train. When he was arrested he was sitting upright in an orderly fashion, but was heard by the arresting officer to be gently snoring. The second is a man who had brought a blanket and pillow to the station and had obviously settled himself down for the night. He was arrested, however, before he had a chance to go to sleep. (1957, 664; my emphasis) Here, the correct outcomes seem to be that the second defendant is guilty, while the first isn’t—even though the innocent party did what the law explicitly prohibited, while the guilty party didn’t. How can that be? One possible answer is that the law isn’t quite what it seems to be. Even though the sentence used to enact it mentions sleeping and only sleeping, one might maintain that the law itself—namely, what was said or stipulated in enacting it—prohibits one from using, or attempting to use, a railway station as a place to sleep. Since the first defendant, though dozing off, wasn’t doing that, while the second defendant was, the second would properly be judged guilty, while the first wouldn’t. The virtue of this account is that it gets what, intuitively, seem to be the right results, while capturing what the lawmakers would have taken pains to say, had they thought about these cases, and wished to remove doubt. Nevertheless, the analysis is a stretch. Fuller’s example is not a reallife case, but a partially specified fantasy that could be completed in various ways. Although one can imagine completions of the story in which the lawmakers really did understand themselves to be asserting the contextually enriched content that gives the desired results, one can also imagine completions in which they didn’t give the matter much thought, and were content with the first formulation that sprang to mind. In these latter versions of Fuller’s fable, it is hard to justify the idea that what the lawmakers asserted was sufficiently nuanced to dictate the desired results in his examples. There is, after all, a distinction between what one actually says in a given context, and what one would say, if one considered things more carefully. Are there grounds for thinking that what lawmakers actually assert in all Fuller-type cases must be nuanced enough to determine the correct outcome in every problematic future application? That doesn’t seem very likely. But then, we can’t maintain both that the content of the law is always what

Interpreting Legal Texts

•

417

was said or stipulated in enacting it, and that it determines the correct outcomes in all relevant examples.7 In my opinion, the most realistic analysis maintains the basic identification of a law with what was said in enacting it—while adding the proviso that judges (and other officials) have the authority to make minor adjustments in its content to avoid transparently undesirable results in cases not previously contemplated. Although explicitness in writing the laws is desirable, everyone knows that it is impossible to anticipate every eventuality, and that, sooner or later, one reaches a point of diminishing returns. Recognizing this, we implicitly give judges a limited prerogative to make changes in the law they interpret—a prerogative not found in all genres of interpretation. This is an important respect in which legal interpretation is special, and fundamentally different from some other forms of linguistic interpretation. Adding this special interpretive prerogative to the general linguistic model for determining the content of arbitrary texts isn’t difficult. Given a context in which lawmakers adopt statutory language, we apply the model to extract what is said in the usual way. Next, we prefix an operator—‘roughly’ or ‘approximately’—to that output, as the contribution to what is said made by the genre of lawmaking itself. The effect of this contribution is to make what is said—which we continue to identify with the content of the law enacted—somewhat indefinite, open-ended, 7 Similar remarks apply to the problematic case Fuller poses for H.L.A. Hart’s example of a local ordinance excluding vehicles from the park. Hart’s point was that because ‘vehicle’ is vague, its “core meaning” will provide clear cases to which it applies and clear cases to which it doesn’t apply, while leaving a range of borderline cases “in the penumbra” for which there is no definite answer to the question of whether or not the term applies. As a result, Hart thought, there is a certain open-endedness and room for discretion in adjudicating the penumbral cases. Fuller’s example is intended to extend this open-endedness and discretion even to clear cases. He writes, “What would Professor Hart say if some local patriots wanted to mount on a pedestal in the park a truck used in World War II, while other citizens, regarding the proposed memorial as an eyesore, support their stand by the ‘no vehicle’ rule? Does this truck, in perfect working order, fall within the core or the penumbra?” (1957, 663). Fuller’s implicit contention is that the outcome of the case is unclear, and not fully determined by the local ordinance, despite the fact that the truck is not in the penumbra of ‘vehicle’, but rather is a clear example of something to which the term definitely applies. Perhaps. However, the incompletely specified example is not entirely clear. Although there may be some ways of filling out the case for which Fuller’s contention is correct, one can easily imagine other ways of filling it out for which the contention fails because the content of the ordinance is not identical with the semantic content of the sentence expressing it—e.g., ‘No vehicle is allowed in the park’—but rather with a modest contextual enrichment of that content—No use of a vehicle (as a vehicle) is allowed in the park. In versions of the example in which this is what the rule-makers understand themselves to be prohibiting in passing the ordinance, the content of the ordinance fully determines that it does not apply to the vehicle-cum-statue, and further judicial discretion is not warranted.

418

•

Essay Fifteen

and subject to authorized precisifications when required by previously unanticipated cases. Though Fuller himself seemed to think otherwise, I believe these authorized revisions must be severely restricted. His examples don’t show that judges have the authority to rewrite laws the literal application of which would violate their own views of what should, or shouldn’t, be legal. What is established is much more limited. When the literal application of a law to a case violates the clear intention guiding the legislators in adopting it, judges are authorized to make minimal modifications to bring it into line with what the legislators were trying to accomplish. The second type of genuinely hard case is one in which facts plus the full, linguistic contents of the controlling legal texts fail to yield any determinate outcome, because that content is vague. Understanding such cases requires understanding how vague language works. In illustrating this, I will draw on a theory of vague predicates according to which they are both context-sensitive and partially defined.8 Although competing treatments are available, the main legal lessons I will draw can, I think, be translated into other theoretical frameworks. Vague predicates are, as I said, partially defined. To say that P is partially defined is to say that it is governed by linguistic rules that provide sufficient conditions for P to apply to an item, and sufficient conditions for P not to apply, but no conditions that are both individually sufficient and disjunctively necessary for it to apply, or not to apply. Because the conditions are mutually exclusive, but not jointly exhaustive, there will be items not covered by the rules for which there are no possible grounds for accepting either the claim that P applies to them, or the claim that it doesn’t. P is undefined for these items. In addition to being partially defined, vague predicates are also context-sensitive. Given such a predicate P, one begins with a pair of sets. One of them, the default determinate-extension, is the set of things to which the rules of the language of the community, plus nonlinguistic facts, determine that P applies. The other, the default determinateantiextension, is the set of things to which the rules of the language plus the facts determine that P doesn’t apply. P is undefined for o just in case o is in neither of these sets. Since the sets don’t exhaust all cases, speakers have the discretion of adjusting the extension and antiextension of P so as to include initially undefined cases to suit their conversational purposes. Often they do this by explicitly predicating P of an object o, or by denying such a predication. When one does this, and one’s conversa-

8

The account is spelled out in Soames (1999, chap. 7), plus essays 13 and 14 of volume 2.

Interpreting Legal Texts

•

419

tional partners go along, the extension (or antiextension) of P is expanded to include o, plus all objects that bear a certain relation of similarity to o. Consider, for example, the contextual expansion of the extension of the predicate ‘blue’ that occurs when I say of an object for which it is initially undefined, “That’s blue,” and my conversational partners agree. Complications aside, the result is the adoption of a new contextual standard that counts o, all objects perceptually indistinguishable in color from o, and all objects discriminately bluer than o as blue. Note how my discretion in applying vague terms to borderline cases is constrained. If I dictate that o is to count as blue, I can’t deny that objects bluer than, or indistinguishable from, o are too. Content-changing precisifications are possible, but only if they are principled changes. This lesson applies to legal texts. Suppose that a legal text stipulates that things in category C are to be treated in a certain way, while things not in C are to be treated differently. Suppose further that C is a vague predicate, and so undefined for a range of items—including the item x on which the case turns. Since x is neither in the default determinateextension, nor the default determinate-antiextension, of C, the rules of the linguistic community, plus relevant facts, don’t determine that C applies to x, or that it doesn’t. Still, application or nonapplication of C to x may be determined by discretionary standards adopted by authors of the text. Thus, the first step in adjudicating the case is to examine the legislative history to discover whether the lawmakers implicitly encoded such a standard into the law. If they did, then the content of the law determinates the outcome of the case, and the job of interpretation is simply to discern that content from the linguistic meaning of the text, plus the context in which it was produced. If they didn’t adopt such a standard, then the job of interpretation is to formulate one—thereby creating new law by precisification. This generation of new content by filling gaps in the original text is another respect in which legal interpretation is importantly different from some other kinds of linguistic interpretation. This special gapfilling role is justified by the fact that the court has no alternative, if it is to decide the case at all, and by the fact that legislators are well aware of this aspect of adjudication, and often deliberately employ vague language with future judicial precisification in mind. In so doing, they implicitly cede limited legislative authority to courts. What considerations should be employed in exercising this authority is a normative question to which broad philosophical principles—including Fullerian appeal to broad social purposes, Dworkinian balancing of the morally best outcome with deference to legal tradition and precedent, and Scalian passion for minimizing judicial intrusion into the democratic process—provide

420

•

Essay Fifteen

different answers. The philosopy of language, by contrast, provides none.9 Legal lessons about vagueness are illustrated by two well-known cases: Curran v. Mount Diablo Boy Scouts, and Randall v. Orange County Council, Boy Scouts of America. In Curran, the Boy Scouts were held to be a business organization and thereby subject to a 1959 California antidiscrimination law—which was understood to require treating heterosexuals and homosexuals alike. In Randall, which was based in large part on the Curran finding, the Scouts were held not to be a religious organization, and therefore to violate the same antidiscrimination law by requiring belief in God. The cases turned on the vague terms ‘business establishment’ and ‘religious organization’. If the latter applies, or neither do, then both cases were wrongly decided. Only if the former applies, but the latter doesn’t, were the initial outcomes legally correct.10 These outcomes cannot be justified by claiming that the rules of the language determined them. Although the Boy Scouts do have assets, sell a few things, and employ people, commerce is only distantly related to their main activity, and the organization certainly is not a clear case of a business establishment. Either it is definitely not a business, or it is a borderline case—and hence in the undefined range for the expression. As for being a religious organization, it has some aspects one would expect— such as requiring belief in God and adherence to a moral code—while lacking others—such as prayers, worship services, and sectarian doctrines. Thus, a plausible case can be made that it is a genuine borderline case, and so in the undefined range of the predicate ‘religious organization’. Suppose, for the sake of argument, that this is true of both predicates. On this supposition, it is a matter of discretion whether or not to count 9 There is a third type of vagueness case that is similar to the second, but with a more limited range of indeterminacy. In these cases, it is clear that rule-makers adopted some discretionary standard, but indeterminate which standard, within a certain range, they adopted. It is thus determinate that the content of the law falls somewhere in a range of possible contents, but indeterminate where. In such cases, the job of interpretation is to discern the range of indeterminacy, and to create new, precisified, legal content, by making a principled selection from within that range. Complexity aside, there is nothing essentially new here. The nature of the interpretive task in such cases is a combination of the tasks involved in the other two types of vagueness cases. However, there is something new worth noting. Once it is granted that what speakers say, or otherwise commit themselves to, may be to some extent indeterminate, it is evident that not all vagueness arises from vague language. Sometimes the move from the linguistic meaning of a legal text to that which the rule-makers were saying, or committing themselves to, in adopting it is itself an independent source of vagueness. In such cases, it is determinate that the content of the law falls somewhere within a range of possible contents, but indeterminate where. 10 The decisions were later overturned by the California Supreme Court.

Interpreting Legal Texts

•

421

them as applying to the Scouts. Did the legislators who enacted California’s antidiscrimination statute exercise this discretion by implicitly adopting standards counting the Scouts as a business, but not a religious, organization? There is nothing to indicate that they did, which isn’t surprising—since, as the dissent in Randall points out, had it been understood in 1959 that passage of the act would have required the Scouts to condone homosexuality and atheism, “a fire storm of public protest would have descended on the Legislature.” No, the legislators didn’t themselves adopt standards dictating the outcomes later reached. Thus, the only justification for those outcomes can come from judicial resolution of the supposed vagueness. The crucial question is whether the judges articulated principled criteria for narrowing the range of the terms ‘business organization’ and ‘religious organization’ so as to exclude the Scouts from the extension of the latter, while including it in the extension of the former. It is arguable that they didn’t. As the dissent in Randall maintained, and the California Supreme Court later agreed, the limited range of commercial and financial activities cited provide an insufficient criterion to render the Boy Scouts a business organization, since—if it were applied consistently— the criterion would lead to the unwanted result that nearly all organizations of any appreciable size are businesses—including most churches, charities, civic, and other private organizations. Thus, the initial rulings failed an elementary test. When the vagueness of terms is judicially narrowed, the result must not be a naked assertion that the term applies to a certain borderline case, or doesn’t, but a principled explanation that narrows indeterminacy in a comprehensible way, and generalizes acceptably to similar cases. The final type of genuinely hard case is one I will simply mention. These are cases that fall under two or more legal provisions which, when combined with the facts at hand, determine inconsistent outcomes. If neither provision is clearly subordinate to the other, no unique outcome is determined, and the job of the court is to resolve the conflict by revising existing law in some justifiable way. The problem is one of subtraction. As with other kinds of inconsistent language use—including inconsistent definitions and paradoxes of various sorts—what one wants is a purified, consistent content gotten by eliminating, or weakening, parts of the original content. Of course, not just any purification will do. The aim is to find the consistent content that is maximally faithful to the original—or, if more than one such content is maximally faithful, the intersection of all maximally faithful purifications. This, of course, is easier said than done. How one comes up with the relevant weakenings, and how, once one has them, one compares their relative fidelity to the original texts are problems for which no algorithmic solution seems possible. Still, the

422

•

Essay Fifteen

guiding principle is clear. In striving for resolution, one tries to preserve as much of the pre-existing law as possible. Although there is room for judicial discretion in this, it isn’t essentially expansive in nature.

Conclusion This result fits my overall theme of the genuine, but limited, legislative role of legal interpretation. The meanings of legal texts are too austere to determine the content of the law. As a result interpretation is often needed to reach proper outcomes. However, a judge’s view of the legitimate social, moral, and political purposes of the law is not the crucial ingredient added by interpretation. The first and most important task of interpretation is to discern what a text says, or has been used to say. As I have argued, its meaning is only one of the determinants of this. What speakers assert or commit themselves to is, quite standardly, a function not only of the meanings of their words, but also of other information present in contexts of utterance. Linguistic contexts in which laws are passed are no exception. There are, to be sure, special features of such contexts that can affect how the gap between meaning and content is filled. The legislative process is governed by purposes that transcend, and sometimes conflict with, the conversational ideal of the efficient and cooperative exchange of information. Consequently, the way in which conversational maxims based on this ideal contribute to filling the gap between the meaning and content of legal texts may, in some cases, differ from their contribution to filling similar gaps in ordinary conversations. In addition, the issue of who constitutes the audience for language use in the legislative context is up for grabs in a way it isn’t in ordinary contexts. Other lawmakers are certainly part of the audience—but so, perhaps, are the law-enforcement officials who will implement the law, the lawyers who will explain it, the judges who will interpret it, and the population at which it is aimed. This complexity can, of course, affect the precise way in which the gap between the meaning of legal language and its content is filled, but it doesn’t change the overall picture. The first, and most important, step in legal interpretation is not moral, social, or political, but linguistic. Presented with a legal text, one must first identify what the relevant actors said, stipulated, or otherwise committed themselves to in adopting it. For this, one needs to know how semantics and pragmatics interact to generate content. A similar lesson applies to adjudicating genuinely hard cases, in which the full, linguistically determined content of the law is inadequate. To resolve these cases, judges need to understand the semantics and pragmat-

Interpreting Legal Texts

•

423

ics of vague language, as well as the ways in which consistent contents can be extracted from inconsistent ones. After all this is taken into account, there may be a residue of remaining judicial discretion for which Fullerian or Dworkinian views of the broad social, political, and moral matters are relevant, but it isn’t very large.11

References Bach, Kent. 1994. “Conversational Impliciture.” Mind and Language 9:124−62. Carston, Robyn. 1988. “Implicature, Explicature, and Truth-Theoretic Semantics.” In Mental Representations: The Interface between Language and Reality, ed. Ruth M. Kempson, 375−409. Cambridge: Cambridge University Press. Donnellan, Keith. 1966. “Reference and Definite Descriptions.” Philosophical Review 75:281−304. Fuller, Lon. 1957. “Positivism and Fidelity to Law: A Reply to Professor Hart.” Harvard Law Review 71:630−72. Neale, Stephen. 2007. “On Location.” In Situating Semantics: Essays on the Philosophy of John Perry, ed. Michael O’Rourke and Corey Washington, 251− 393. Cambridge: MIT Press. Soames, Scott. 1999. Understanding Truth. New York: Oxford University Press.

11 Thanks to my colleague Andrei Marmor for many helpful comments and discussions, as well as for teaching me most of the (all too limited) legal philosophy I know.

This page intentionally left blank

Index

Almog, Joseph, 73 Atlas, Jay David, 104n35, 308n15, 316n16, 317n18 assertion, 1–2, 251–76; and the content of positive law, 16 403–4, 408–9, 412–15, 417–18, 422; of multiple propositions, 7, 252, 252, 255, 261–62, 266–67, 281, 285–88, 293–94; primary, 7–8, 251, 258, 260n15, 264, 266, 267, 268, 270, 272, 280, 284, 286, 294; and semantic content, 7, 251–60, 266–73, 275–76, 278–86, 316–21, 379–80, 385–86, 388–91, 408–15. See also conversational implicature attributive/referential distinction, 11–12, 286–88, 291–93, 351–54, 331n2, 332, 333–38, 360–375, 380–83, 392–94, 409–10. See also incomplete descriptions Austin, J. L., 337 Bach, Kent, 6, 260n15, 275n26, 279n1, 317n18, 320–1n22, 324n24, 389n7, 410n3, 411n4 Barker, Chris, 276n27, 324n24 Barwise, Jon, 10, 86n17, 167n12, 176n19, 179n21, 291n12, 329–358, 380n2 Bever, T. G., 134n2 Bezuidenhout, Anne, 279n1, 288n10, 389n7 Bigelow, John, 179n21 Boolos, George, 167n12 Bowman, Brian, 308n6, 324,24 Bresnan, Joan, 153n24 Bromberger, Sylvain, 136n4 Carnap, Rudolf, 226 Carston, Robyn, 7, 260n15, 279n1, 309n8, 313n13, 316n16, 317n18, 322n22, 324n24, 389n7, 410n2 Cartwright, Richard, 27n11 Chierchia, Gennaro, 279n1, 389n7 Chomsky, Noam, 6, 133n, 134n2, 141, 141–42n9, 143nn10 and 11, 147n17, 148n18, 151n22, 159n, 168, 176, 203, 206, 226, 227

conventional implicature, 26n8; as presupposition, 3, 25, 101–2. See also projection problem for presupposition conversational background, 93–94, 128, 261, 266, 275, 276, 280, 287 conversational implicature, 2, 8–9, 26, 29n16, 279; and assertion, 252, 255, 258, 259, 265n20, 267, 268, 269, 300–303, 306, 307–9, 310; 278–81, 296, 298–306, 320, 384–88. See also presupposition and conversational implicature conversational maxims, 2, 29n16, 41n35, 46n47, 53, 298–306; role in assertion, 8, 306, 309–10, 316–19 Cooper, David, 134n1 Cooper, Robin, 133n, 159n, 176n19 Cresswell, Max, 179n21 Curran v. Mount Diablo Boy Scouts, 420–21 Davidson, Donald, 6, 183, 185n2, 208–10, 218, 220, 225–33, 234, 237– 44 Davies, Martin, 12, 370n18 Davis, Stephen, 134n1 definite descriptions, Russell’s theory of, 12, 377–98. See also attributive/referential distinction; incomplete descriptions Dennett, Daniel, 165n9 direct reference, 81–82, 85, 86, 188, 192, 360–64, 368–375 Donnellan, Keith, 10, 83n12, 86n17, 162n4, 206, 292, 293, 331n2, 352, 353n30, 360–75, 378, 381, 381n4, 392, 393, 409n1 Dworkin, Ronald, 419, 423 Evans, Gareth, 12 Fodor, Jerry, 133n, 134n2, 140n8, 150n21, 159n Foster, John, 178n20, 185n2, 210–12, 214, 218, 220, 225, 230–32, 240, 241

426

•

Index

Frege, Gottlob, 3n2, 75–80, 86, 92, 229, 242, 283 Frege’s Puzzle, 7, 251 Fuller, Lon, 415–18, 419, 423 Gabbay, Dov, 167n12 Gadd, Nate, 309n7, 324n24 Garrett, M. F., 134n2 Gazdar, Gerald, 58n62, 65n76, 94n22, 96–98, 102, 104n35, 119 Goodman, Nelson, 142n9 Grice, Paul, 9, 26n8, 102, 298–306 Grice’s Paradox, 9, 305–6 hard cases in law, 403; genuinely, 403–4, 415–22; semantically, 403–4, 415 Harman, Gilbert, 134n1 Harnish, Robert, 133n, 159n Hart, H.L.A., 417n7 Hausser, R. R., 35n30, 49n50 Heim, Irene, 4, 98n28, 102n32, 113–24 Higginbotham, James, 133n, 208n1, 215, 216n7, 225, 233–38 Hintikka, Jaakko, 167n12 Horn, Larry, 124, 279n1, 309n8, 313n13, 316n16, 319n20, 324n24, 389n7 incomplete descriptions, 9–12, 288–94, 329–58, 394–96; and discourse anaphora, 11–12, 354–5n33, 394–96, 366–73; pragmatic enrichment of, 10–11, 293–96, 394–96 Jeffrey, Richard, 167n1 Johnson-Laird, P. N., 174n17, 175n18 Kamp, Hans, 82n11 Kaplan, David, 81–82, 83n12, 161n1, 192n10, 202n1, 274n24, 330, 331, 362–63 Karttunen, Lauri, 3, 25–67, 90, 95–99, 102–3, 116, 119, 120, 124, 125n52, 126n53 Katz, Jerrold, 6, 59n64, 134n2, 140n8, 143n10, 144n13, 151n22, 159n, 162n4, 202–6 Kempson, Ruth, 104n35 King, Jeff, 244n38, 296n15, 324n24 Kripke, Saul, 4n5, 8, 73, 86n17, 106n39, 111, 112nn43 and 44, 126–27n54, 161, 162n4, 197n13, 253, 283, 284, 287,

292, 331, 352, 353nn29 and 31, 354n32, 365, 378, 382, 393, 394 Langendoen, Terrance, 23–25 Larson, Richard, 232n23, 240, 241n35 Lewis, David, 88n18, 99, 125, 128n55, 179n21, 341n18, 344n20 liar paradox, 111–12 linguistic competence, 4; and performance, 147–51; psychological theories of, 5–6. See also semantic competence linguistic intuitions, 139; fallibility of, 1, 322 linguistic theories, 133–56; as conceptually distinct from psychological theories 134–36, 144–51, 322–24; data for, 139–40, 146–47, 152–53; as empirically divergent from psychological theories, 134–36, 152–55; importance of for psychology, 135, 156; leading questions for, 36–37; methodology for, 141–43, 322–24; subject matter of, 137–40 linguistic universals, 5, 143n10, 175–77 Ludlow, Peter, 240, 241n35 MacFarlane, 312n11, 324n24 McCawley, James, 341n18 McGlone, Mike, 258n11, 259n13 meaning, 1; austerity of, 272, 281, 323, 397, 403, 407, 422; and definition, 404–7; knowledge of, 186–97, 208–23, 322; as least common denominator, 7–8, 254–56, 259–60 274–76, 280–81, 296, 323–24, 389–91; and truth conditions, 6, 159–61, 183–94, 208–21, 278–79; and use, 1, 6, 197. See also semantics Moravcsik, Julius, 73, 159n, 167n12, 208n1 names, 2, 161–62, 195; linguistically simple, 7, 253, 254, 256, 260, 263, 266, 267, 275, 282; nondescriptionality of, 15, 404–5; partially descriptive, 253, 254, 256, 261–67, 281–282; substitution of, 7, 251 natural kind terms, 15, 162n4, 405–6, 407–8, 410 Neale, Stephen, 7, 12, 365n8, 370n18, 384, 411

Index necessary truths, 6, 202–26 Nix v. Hedden, 407–8, 410 numerical quantifiers, 306–19, 322 Parsons, Terrence, 111n42, 112n44, 188n7 Partee, Barbara, 162n6, partial definition, 111–113, 418–19 Perry, John, 10, 86n17, 179n21, 291n12, 329–58, 380n2 Peters, Stanley, 3, 25–67, 90, 102–3, 116, 120, 124, 125n52 PGA v. Martin, 406–7 Poland, Jeff, 133n, 159n, 164n8 possible world (state), 10, 37, 88, 89, 105, 114, 115, 146, 162, 178, 179, 185n2, 187, 188n6, 203, 222, 254, 278, 330, 331, 332, 336, 341, 344, 355 Postal, Paul, 6, 202–6 pragmatics. See also pragmatic presupposition —pragmatic enrichment, 7–8, 251–60, 266–76, 280–88, 293–96, 316–19, 321, 389–96, 403–4, 408–15; of incomplete descriptions 10–11, 293–96; optional vs. required, 316–17; and presupposition, 8, 287, 288, 281, 285, 287, 288; vs. semantic specification of hidden variables, 13–15, 280n3, 289–90, 384n6 —pragmatic properties, 1 —relationship to semantics, 6–9, 279–81, 283, 296, 322–24, 379–80, 385–86, 388–91, 404 presupposition —accommodation of presuppositions, 4, 87–88; and cancellation, 99–100; de facto, 99, 125–26; de jure, 100 —cancellation of the presuppositions of compound sentences, 58–67, 70, 94–95, 98, 125; as limiting the sources of pragmatic presuppositions, 101–4 —and conversational implicatures, 3, 41n35, 45–46, 51nn52 and 53, 55–66, 70, 94–98 —and deviations from bivalence, 112–13 —expressive, 3, 75, 80–86, 89, 113 —logical, 3, 33, 35–37, 47–49, 67, 76–80, 82, 105–13 —and negation, 39–40, 77–78 —pragmatic, 3, 75, 86–91; and conventional implicature, 101–3; foundations

•

427

of, 99–102; and logical presupposition, 103–11; and R-presupposition, 110; and radical presupposition, 110; scope of, 91–93 —presupposition inheritance, 26–31; 33–35, 67–68, 95–96; attempted explanations of, 90–91; Frege’s theory of, 79–80; and the identification of presuppositions, 73–74; and the problem of conflicting presuppositions, 43–53, 69; and symmetric conditions for disjunction, 38–42, 68–69 —projection problem for presupposition, 2–3, 94–98; cumulative hypothesis for, 24–25, 27–28; Karttunen and Peters proposed solution to, 25–67; the new projection problem, 66–67 —R-presupposition, 110 —radical, 110 —and reference failure, 76–78 —semantic discourse, 116; and accommodation, 118–19, 121–22; and pragmatic presupposition, 116–17, 124; and quantification, 117–18, 121–22; and truth functional compounds, 116, 120, 122–23; —sentential, 33, 38, 67 —speaker, 3, 31–33, 49, 67 —suspension (filtering) of presuppositions of compound sentences, 63–66, 95–6 —utterance, 93 propositions, 187, 278–79; and truthsupporting circumstances, 187–88, 222, 278, 329–58; structured (Russellian), 5, 11, 14, 82n11, 146, 187–92, 222–3, 243, 244, 278, 285, 356, 338n21, 355, 243–46, 278–79, 286, 346n21, 355–58, 360–64, 368–375 Putnam, Hilary, 161, 162, 164 Quine, Willard V., 134n1, 142n9, 172n15, 225–29, 361–62 Randall v. Orange County Council, Boy Scouts of America, 420–21 Recanati, François, 7, 260n15, 279n1, 288n10, 321n22, 389n7 Richard, Mark, 5n7, 208n1, 238n30 Russell, Bertrand, 12, 79n9, 226, 242, 243, 377–80, 396–98

428

•

Index

Salmon, Nathan, 14n12, 73n, 82n10 and 11, 86n17, 106n39, 111n41, 188nn5 and 7, 189n9, 285n7, 292n13, 329n, 331n2, 334n7, 344n20, 355, 356, 363, 380n2, 381n3 Savin, Harris, 23–25, Scalia, Antonin, 406–7, 413–14, 419 Segal, Gabriel, 232n23 semantics —discourse semantics, 113–24 —psychological relevance of, 174–77, 197–200, 322–24 —relationship to pragmatics, 6–9, 251–76, 279–81, 283, 296, 322–24, 379–80, 385–86, 388–91 —semantic competence, 5–6, 163, 165–73, 177–79, 182–94; and the Augustinian picture, 194–97 —semantic content, 2; as constraining assertion, 7, 254, 256, 259–60, 274–76 279–81, 296, 316–22, 388–91; incomplete, 8, 252, 260, 274–76, 280, 295–96, 316–22; limited knowledge of, 9, 322, 406; and legal interpretation, 16; pragmatic enrichment of, 7, 257–60, 266–76, 279–88, 293–96, 316–22, 389–96, 404, 408–15. See also propositions —semantic properties, 1, 138–39, 144–45, 153, 164–65; knowledge of, 5–6, 166–69, 183–86, 205–6 —semantic representations, 169–71 —semantic theories, 1, 5, 159–61, 186–94; as contingent, 202–6, 243–46, 322–24. See also truth theories —situation semantics, 10, 329–58; the attributive/referential distinction in, 351–54; nonpersistent truth in, 335, 336, 337, 338, 339n14, 344, 345, 348, 351; persistent truth in, 335, 336, 339n14, 341n17, 344, 345, 350, 351; utterance supplementation in, 337, 340–44; 345, 348, 349–50; vs. structured propositions, 355–58; what-is-said in, 338–40 situations: abstract, 329–38, 341, 345, 354; actual, 330n1, 335–44, 357n36;

factual, 330, 331, 334–37, 339–41; real, 330, 338–51, 54n33, 57n36; resource, 331, 332, 351, 354, 355n33 smidget, 106–13 Sober, Elliot, 154n25 Smith v. the United States, 412–15 Sperber, Dan, 279n1, 320n22, 389n7 Stanley, Jason, 13–15 Stalnaker, Robert, 25n6, 28n13, 37n32, 50n50, 75, 86–91, 95n23, 99, 102n32, 103, 105–6, 108–9, 124, 125 Stitch, Stephen, 134n1, 143n10 Strawson, Peter, 75, 82–86 Tarski, Alfred, 226–29, 235, 243 Taylor, Ken, 275,n26, 279n1, 389n7 Thomason, Richmond, 128n55, 162n5 truth conditions, 2, 25, 278–79; knowledge of, 5, 182–86, 208–21; and meaning, 6, 159–61, 186–87, 208–21, 225–46; of propositions, 190–91, 221–23; and psychology, 144–46, 161–79, 232–33 truth theories, 6, 11, 160, 161, 169, 175, 184, 185n2, 208–22, 225–42; translational, 211, 213–21, 230–42 truth-value gaps, 3, 35–37, 47–49, 76–80, 105–111; and paradox 111–13 T-sentences, 169, 170, 171, 209–21, 230–32, 234–41; translational, 211, 213–18, 230–32, 234–41 vagueness in the law, 17, 417n7, 418–21 von Eckardt, Barbara, 133n, 159n Wasow, Thomas, 133n, 150n21, 159n, 176 Wettstein, Howard, 86n17, 291n12, 292n13, 331n2, 334n7, 380n2 what-is-said, 2, 9, 124, 278, 299–300, 303, 305–6, 416–17. See also assertion Wilson, Deirdre, 31n19, 43n41, 104n35, 279n1, 320n22, 389n7 Wittgenstein, Ludwig, 194n11, 245, 406