Oxford Studies in Philosophy of Language Volume 3 0198892721, 9780198892724

Citation preview

Oxford Studies in Philosophy of Language

Oxford Studies in Philosophy of Language Volume 3 Edited by

ERNIE LEPORE and

DAVID SOSA

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © the several contributors 2023 The moral rights of the authors have been asserted All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2021930669 ISBN 978–0–19–889272–4 DOI: 10.1093/oso/9780198892724.001.0001 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

Contents Editors’ Preface List of Contributors

1. Scoreboards Without Scorekeepers Josh Dever

vii xiii

1

2. Singularism vs. Descriptivism? Rachel Goodman

33

3. Verbal Signaling Mitch Green

67

4. Disquotation, Translation, and Context-Dependence Richard Kimberley Heck

99

5. The Place of the Philosophy of Language in Metaphysics Thomas Hofweber

129

6. On Lying, “Strictly Speaking” Marga Reimer

156

7. À Propos de Pierre, Does He . . . or Doesn’t He? Nathan Salmón

176

8. The Schmidentity Strategy Jeff Speaks

182

9. Leverage: A Model of Cognitive Significance Stephen Yablo

205

Index

249

Editors’ Preface With its third issue, Oxford Studies in Philosophy of Language continues its role as a regular showcase for leading work in this area. Language has been a main focus of philosophical research since at least Frege’s seminal contributions at the turn of the twentieth century. Beginning with that “linguistic turn,” important work all across philosophy has often been related in some significant way to philosophy of language. This series offers a regular snapshot of state-of-the-art pieces in this central field. Published biennially, it aims to serve as a curated forum for contributions by some of the best scholars from around the world, both senior and junior. Each issue includes a small number of outstanding papers in philosophy of language, broadly construed. This third issue of our series is another good instance of the form: it includes nine new papers by a distinguished range of philosophers. Together, the papers provide a perspective on the state of the subdiscipline. In his “Scoreboards Without Scorekeepers,” Josh Dever aims to set out a “de-psychologized perspective” on a range of theoretical resources used by both dynamic theorists of language and expressivist theorists. One important such resource is the idea of a conversational “scoreboard” (see Lewis for an important early discussion), by which various conditions of the communicative situation are represented. Dever reviews “cognitivist” implementations of that idea and, while appreciating their attractions, notes their continuities with Gricean programs, on which linguistic meaning is seen as deeply integrated with systematic psychological and social phenomena; given his pessimism about whether the relevant psychological phenomena are systematic in the requisite ways, he proposes a new program, on which the key theoretical notions are mind-independent phenomena, recommending accordingly a fully “non-cognitivist” metaphysics for linguistic scoreboards. Showing that the proposed non-cognitivism has no special problem about conversational participants’ knowledge of what’s on the scoreboard, and that it can

viii

 ’ 

accommodate presupposition failure, Dever’s perspective bids fair to invert a traditional conceptual setting for popular theoretical resources. Rachel Goodman’s “Singularism vs. Descriptivism?” revisits a widely endorsed distinction between “singular” and “descriptive” thought, reviewing arguments in favor of the existence of singular thought. That sets the stage for a refined account of the debate between “singularists” who affirm that some mental states are about some thing(s) “directly” (without the mediation of description) and “descriptivists,” who oppose the idea that any of our thoughts have singular propositions (containing ordinary external particulars) as their contents. That debate connects with issues in philosophy of language concerning “direct reference,” and Goodman shows how some of the considerations deriving from the philosophy of language that might have been seen as supporting a singularist view can also be seen as challenging it. Her piece seeks to support singularism in a way that would leave it immune to the challenges deriving from philosophy of language, if also forgoing the support many saw linguistic considerations as providing. The care with which she pursues that attempt leads ultimately to an appreciation of the dialectical subtlety of the disagreement about Frege’s Puzzle. “Verbal Signaling” offers Mitch Green’s effort to find continuities between, on the one hand, paradigmatic linguistic communication, which some see as requiring complex sets of intentions as well as conformation to norms, and on the other hand “signaling,” in which “an organism conveys information to a potential recipient in such a way that both sender and receiver stand to benefit.” Distinguishing the verbal signal from the verbal “index,” and disentangling different strands wrapped up in a notion of commitment associated with assertion, Green contemplates two possible precursors to assertion: “ursertion” and “semisertion,” each with distinctive normative profiles. These in turn promise to enable a new understanding of such phenomena as expositives, grammatical evidentials, and conversational implicature. Although it may have been known for some time that contextdependence poses a problem for “disquotationalist” theories of truth and reference, Richard Kimberley Heck argues that the problem goes deeper than has yet been appreciated. According to Heck’s “Disquotation, Translation, and Context-Dependence,” the notions of truth and reference

 ’ 

ix

do not really apply to type expressions at all, giving disquotational theories—which seek to characterize truth and reference in terms of conditions on sentence types—“the wrong subject matter.” Rejecting an attempt by Field to handle the problematic context-dependence in question, Heck contends that all forms of ‘translational semantics’ rest upon use-mention confusions and so cannot help the disquotationalist. While disquotationalism must be rejected, “intentional realism” does not follow: the dependence of facts about content on pragmatic factors poses as great a difficulty for intentional realism as it does for disquotationalism. Can substantial metaphysical conclusions about what reality is like be drawn from facts about how we represent reality? In “The Place of Philosophy of Language in Metaphysics,” Thomas Hofweber argues for an affirmative answer, one that sustains a significant role for the philosophy of language in metaphysics (and he suggests that a similar structure holds for the relation between epistemology and metaphysics). For example, Hofweber points out that considerations in philosophy of language might settle that number words are non-referential: adding that “if number terms are non-referential then natural numbers do not exist,” he concludes that investigation of our representations could deliver substantial conclusions about what reality is like (in nonrepresentational respects), giving philosophy of language a constructive role in metaphysics. The piece then considers whether the questions that can be answered in that way, through investigation of our own representations, are good questions—or the right questions to be asking—and indeed whether our metaphysical questions should be given any special place in metaphysics. Marga Reimer here contributes to a revival of interest in philosophical work on lying. Her “On Lying, ‘Strictly Speaking’” stakes out a subtle combination of theses about lying: (i) deceptive intent is not necessary for lying but is a “defining feature” of prototypical lies; (ii) while one can lie in a loose sense (one can “basically” lie, or the like) through implication, any attempt to define or analyze lying would have “saying” as a necessary condition; and (iii) to be aptly described as “a liar” may require more than that one have lied—in particular, it may require deceptive intent. Reimer accordingly distinguishes lying “strictly speaking,” lying “prototypically speaking,” and lying “loosely speaking,” and suggests that

x

 ’ 

insensitivity to such distinctions may prompt overambitious interpretations of experimental studies. Ultimately, Reimer sees her work as a demonstration of how the distinctions that philosophy seeks to uncover can already be found in ordinary speech and thought. Nathan Salmón, who in Frege’s Puzzle (1986) and in a series of related papers developed and defended what can appropriately be called “textbook Millianism,” here returns—in his “À Propos de Pierre, Does He . . . or Doesn’t He?”—to the analysis of belief ascription, and to his way of confronting Kripke’s Puzzle about Belief. Salmon views Pierre as believing, but also as withholding believing, that London is pretty. His view postulates a role for guises, however—ways of taking a proposition, through which believers can bear a range of cognitive relations to propositions. In the present contribution, Salmon contends with an alternative, purely “referentialist,” approach that would oppose any involvement of guises. On this alternative approach (due to Crawford 2004), one uses the phenomenon of second-order belief in an attempt to resolve the puzzling cases: instead of defining the key condition of withholding belief, as Salmón does, in terms of guises, one views it as a matter of believing that one does not believe. While Salmon recognizes this alternative’s positive features, he gives an example—reminiscent of Mach’s shabby pedagogue—showing that one can sometimes believe that one does not believe that p even when one does not withhold belief in the way that’s crucial for resolving the puzzles. “The Schmidentity Strategy” is about an eponymous style of argumentation first bruited by Kripke in Naming and Necessity and then refined in his “Speaker’s Reference and Semantic Reference.” The general idea is deployed to resist one kind of response to Millianism—a response according to which “Cicero is Tully” does not express the relation of identity between the object that is both Cicero and Tully and itself. The schmidentity strategy proposes that, whatever “is” in English currently actually expresses, because some word could have expressed the relation of identity, holding between each thing and itself, and since the same phenomena would arise in such circumstances, the contemplated sort of response to Millianism is not ultimately adequate. Jeff Speaks here revisits the schmidentity strategy, however, characterizing it more

 ’ 

xi

precisely and considering extensions that might indicate limitations on its deployment, and even cast doubt on its dialectical role. In particular, Speaks investigates the idea of something’s being stipulated to be true that figures in the strategy and—distinguishing varieties of linguistic stipulation, and attending also to the scope of the stipulations involved— proposes that the stipulations required in some uses of the strategy may be bound to fail. Ultimately, Speaks shows how the deployment of the schmidentity strategy in defense of Millianism may involve what, to an opponent (though not to a proponent), appear to be problematic presuppositions about, for example, how different a community would have to be, from ours, for it to speak a language whose semantics was in fact Millian. Stephen Yablo is interested in “Frege Cases” too: in “Leverage: A Model of Cognitive Significance,” Yablo brings machinery that was originally devised in work on aesthetics—on metaphor, irony, hyperbole, and truth-in-a-story—to bear on the variation in cognitive content that is supposed to be exhibited by “Hesperus is Hesperus” and “Hesperus is Phosphorus.” The relevant variety of cognitive contents are incremental contents of a sort Yablo has discussed previously (2012a, 2014). In the current work, Yablo cites Hoek’s (2018) treatment of incremental content, which adverts to a new theoretical parameter: subject matter. Ultimately, Yablo argues that the cognitive contents generated in the way he proposes are adequate to the cases wanting explanation: he sees the notions of subject matter and of pivot-presuppositions as figuring in a more complex structure, a structure analogous to that of a simple lever and fulcrum. If the account he offers is successful, we would have a relatively systematic and thoroughgoing separation of semantic content from cognitive content. Together, this broad-ranging set of papers reveals the breadth and depth of work in philosophy of language today. Future issues will provide an equally diverse, rich, and valuable collection of contributions to the discipline.

List of Contributors Josh Dever Department of Philosophy, University of Texas at Austin Rachel Goodman Department of Philosophy, University of Illinois–Chicago Mitch Green Department of Philosophy, University of Connecticut Richard Kimberley Heck Department of Philosophy, Brown University Thomas Hofweber Department of Philosophy, University of North Carolina, Chapel Hill Marga Reimer Department of Philosophy, University of Arizona Nathan Salmón Department of Philosophy, University of California, Santa Barbara Jeff Speaks Department of Philosophy, University of Notre Dame Stephen Yablo Department of Linguistics and Philosophy, Massachusetts Institute of Technology

1 Scoreboards Without Scorekeepers Josh Dever

Two tempting philosophical agendas: • Philosophy of language is just a subdiscipline of the philosophy of action! Language uses are just particular strategic actions, and everything we need to know about language can be extracted from consideration of the practical rationality of such strategic actions. • Philosophy of language is just a subdiscipline of the philosophy of mind! Language uses are just expressions of certain complex mental states designed to bring about other complex mental states in other minds, and everything we need to know about language can be extract from consideration of the cognitive theory of such mental states. Both of these tempting agendas push our theorizing about language in a psychologized direction. But there are also pressures—from (e.g.) the role of truth conditions and inferential relations in language—that push in an antipsychologistic direction, and we see in both (for examples) Frege’s use of the context principle in the Grundlagen to avoid infiltration of psychological considerations into the semantics of number talk and the semantic externalists’ arguments that meanings ain’t in the head routes for respecting those pressures. Some battles must be fought by each generation anew, and the psychologistic/antipsychologistic battle in the theory of language is one of them.¹ Recently there has been a convergence of theoretical interest ¹ That historical observation is meant to be neutral on the question of who ought to win the battle. Josh Dever, Scoreboards Without Scorekeepers In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Josh Dever 2023. DOI: 10.1093/oso/9780198892724.003.0001

2

 

between dynamic theorists of language, who posit complex representational/ informational conversational scoreboards updated by language use, and expressivist theorists of language, who take the role of language to be to express and make manifest our mental states. One way to manifest the mental states is to make them be the scoreboards the dynamicists manipulate. On this picture, the goal of conversation is to create a kind of public mind—we express our mental states by imposing them on the public scoreboard. We are thus the scorekeepers—the public mind of the scoreboard is a “little us”, and we are its keepers, responsible for its growth throughout the discourse. But we can have scoreboards without scorekeepers. The aim of this paper is to set out a depsychologized perspective on the powerful semantic tools that dynamicists and expressivists have been developing.

1.1 An introduction to scoreboard technology Scoreboards have grown more elaborate and sophisticated over the years. An early scoreboard for a Harvard football game in 1893 tracked scores for both teams, quarter, down, and ball position, with the scoreboard content updated manually by a scorekeeper. As scoreboard technology improved, scoreboard updating became increasingly automated and the game features tracked increased—game clocks were added, additional player statistics were tracked, scores for other out-of-town games were displayed. The current scoreboard state-of-the-art is the Infinity Screen at SoFi Stadium—it is larger than the game field itself, and is dynamically capable of displaying more or less any information about the game, and effectively absorbs the game itself into the score of the game by incorporating instant replay and live video coverage as part of the score. Linguistic scoreboard technology is newer than sports scoreboard technology, and lags somewhat behind.² The first clear linguistic scoreboards are introduced, of course, by Lewis (1979), although the core

² Not really. Theorizing about linguistic scoreboard technology is a new activity, but the scoreboard technology itself is perhaps the oldest human technological possession. How it compares as a piece of technological flash is perhaps a matter of individual taste.

  

3

device is in place already in Lewis (1969)’s discussion of common knowledge, and especially in Stalnaker (1974)’s implementation of common knowledge as a dedicated piece of linguistic technology. Lewisian scoreboards are already expressive and flexible—among the scoreboard components that Lewis discusses are a body of information, a comparative salience relation, a standard of precision, a modal horizon, and a plan. The intervening years have seen an explosion both of new scoreboard components and of functionality of those components. Consider what you’re likely to find on a current industry-standard linguistic scoreboard: • A common ground, consisting of a set of worlds (or set of propositions, depending on your preferred fineness of grain). The common ground tracks the conversational contribution of assertions, guides belief formation by conversational participants, and determines presupposition satisfaction. • A to-do list, consisting of a list (or set?) of actions (or propositions, depending on your preferred philosophy of action). The to-do list tracks the conversational contribution of imperatives and guides domain goal formation by conversational participants.³ • A question under discussion (QUD) stack (of questions, typically conceived of as sets of question answers). The QUD stack tracks the conversational contribution of interrogatives and structures ³ Why isn’t there a presupposition analog for the to-do list? Imperatives can of course, like all speech acts, carry presuppositions—consider Stop smoking! But these presuppositions are informational presuppositions, and are satisfied (or not) in common ground, not in the to-do list. The acceptability of the imperative Stop smoking! requires that common ground contain the information that the commandee previously smoked, not that the to-do list contains the action of previously smoking (or any other action). Two possible answers. First answer: perhaps there is such an analog. (Rather than ‘presuppositions’, perhaps we should call them ‘prejussitions’. I won’t, though.) Consider the command sequence Start smoking! Then stop smoking! Perhaps the change-of-state trigger of ‘stop’ is here accommodating in the to-do list via the previous imperative Start smoking!, rather than in common ground. But it is hard to get entirely convincing cases. If there is a default expectation that commands will be obeyed, then the issuing of the first command Start smoking! may place information about future smoking in common ground, as well as on the to-do list. (Consider: A. Start smoking!, B. No, I’m not going to do that., A. Then stop smoking!) When information about prior smoking is in common ground, there is never a presuppositional barrier to the use of Stop smoking!, which makes it hard to isolate a prejussitional reading if there is one. Second answer: the structure of the scoreboard, and in particular functional distinctions between the common ground and the todo list components of the scoreboard, explain the difference. See Section 1.5 for elaboration.

4

 

•

•

•

• •

relevance relations as well as a number of relevance-downstream phenomena such as anaphora resolution and interpretation of focus-sensitive operators.⁴ A collection of discourse referents, containing discourse referents introduced by appropriate expressions (paradigmatically, indefinite descriptions). Discourse referents then control permissibility and interpretation of uses of pronouns. A modal horizon (given by a set of ‘live option’ worlds) and an (or multiple) ordering source(s) (given by an ordering relation on worlds or a set of propositions from which an ordering relation can be extracted), which jointly govern the interpretation of a number of modal and modal-adjacent (conditionals, evidentials, ‘knows’, ‘wants’) expressions. Quantifier domain restrictions, or functions from objects to sets or properties, which (unsurprisingly) control the domains over which various quantifiers range. Comparison classes or degree function ranges governing the interpretation of scalar adjectives. A distinction between foregrounded/at-issue and backgrounded/not-at-issue information, used (for example) in determination of the felicity of utterances involving evidentials.

And much much more. Fancier linguistic scoreboards may also support ‘screen within a screen’ technology, allowing for nested hierarchies of scoreboards with (for example) conditional constructions triggering the creation of temporary subordinate scoreboards inheriting some but not all features from the primary scoreboard. (The nested structure of discourse representation structures (DRSs) in Discourse Representation Theory is a paradigm example of screen-within-a-screen, but the basic idea can be implemented in many scoreboard ecosystems.) ⁴ Following on the previous footnote, we could wonder about prequisition as well as presupposition and prejussition. In fact, licensing roles of the QUD (e.g., in licensing answerform assertions or in licensing focus distributions) can be thought of as types of prequisition. Relevance implicatures more generally might profitably be reconceived as a type of prequisition accommodation. Declaratives and interrogatives both seem more deeply integrated into the mechanics of scoreboard dynamics than do imperatives; this fact emerges again when I discuss (what I take to be) the non-existence of prejussition in Section 1.5.

  

5

Sports scoreboards are typically input-only—the current content of the scoreboard is fixed by the past events of the game, but the subsequent events of the game aren’t determined by the current contents of the scoreboard.⁵ But conversational scoreboards typically point both upstream and downstream. A full linguistic story needs to tell us both (a) how the contents of the scoreboard are altered by moves made in the language game (the upstream connection) and (b) how the linguistic impact of a move in the language game is affected by the prior contents of the scoreboard (the downstream connection).⁶ The classical static story has it that the linguistic flow through the scoreboard is a flow of propositions. An utterance, made against the backdrop of a given scoreboard, launches a proposition at the scoreboard. Upon impact, the proposition alters the contents of the scoreboard.⁷ The altered scoreboard then influenced which proposition is launched by the next utterance, and so on. But it’s also tempting at this point to cut out the classical propositional middlemen. Absent propositional middlemen, our theory becomes dynamic. We can adopt a procedural semantics and associate sentences (more broadly, linguistic moves) with update functions that map from prior to posterior scoreboard contents. Static theories can be modeled dynamically—a static propositional content gives rise to update functions specifying

⁵ Typically, but not always. When the basketball player takes a low-percentage shot because the scoreboard shows the shot clock winding down, the scoreboard is influencing subsequent play. Of course, if past play influences both current scoreboard and future play, it may be possible to extract spurious non-causal correlations between current scoreboard and future play. Determining when the scoreboard itself has downstream consequences may then be empirically difficult (and may be even metaphysically underdetermined in the right situations), but there’s no principled problem with the distinction. (Did the player shoot because the scoreboard showed the shot clock winding down, or just because the shot clock was winding down? Absent scoreboard errors, it might be difficult to tell, but a full causal history should reveal which led to the player action.) ⁶ As in the previous footnote, there can be hard questions about whether downstream utterance content is fixed by the current scoreboard or by the upstream linguistic play that fixed the current scoreboard. As before, we can perhaps look to scoreboard error for insight, although it may be more difficult to say what it is for a linguistic scoreboard to be ‘in error’ than for an athletic scoreboard. (More on this below.) Again separating the epistemology from the metaphysics, we presumably (in the typical special sciences way) want to see if explanations of downstream content via scoreboard displays produce a theoretical unity that allows richer explanations. ⁷ In the simplest models, by adding that proposition to the common ground component of the scoreboard. But much more sophisticated static models are available as well.

6

 

what scoreboard change results from the integration of that proposition.⁸ But many update functions can’t be derived from propositional seeds, so we open the semantic space to a broader range of updates when we throw open the dynamic floodgates.⁹ For example, in the dynamic setting, possibility modals ◊X could perform the minimal expansion of the scoreboard to include a modal horizon supporting X, while necessity modals □X could perform a minimal contraction of the scoreboard to include a modal horizon supporting X.¹⁰ These sorts of expansion and ⁸ The principal ultrafilters of scoreboard updates, as it were. Pointing, of course, to the obvious research project. ⁹ The standard characterization move here is that of van Benthem (1986), who observes that when the scoreboard common ground is conceived as a set of worlds, propositionally-triggered updates are exactly those that are both conservative (they never expand that set) and continuous (the output of the update function given a prior scoreboard set of worlds is the union of the several outputs of the update function on the singletons of the worlds in that prior set). It’s easy to point to potential examples of linguistic phenomena lacking one or both of these features— retractions look non-conservative, as do dynamic predicative logic (DPL) uses of existential quantifiers (Groenendijk and Stokhof (1991)); modals and conditionals as treated in (e.g.) Update Semantics (Veltman (1996)) look non-continuous. But a careful examination of this standard characterization reveals a host of difficulties. Roughly, the more complicated our scoreboards become, the less obvious it is what it means for an update to be conservative (not to lose current scoreboard contents—but with a more expansive scoreboard, losses are inevitable (the game clock is always losing information about remaining time and replacing it with new information)) or what it means for a content to be continuous (not to depend on holistic features of the scoreboard—but with a more expansive scoreboard, it’s less clear what the scoreboard atoms are (pixels, when instant replay is included?)). This paper is only tangentially about the static/dynamic dispute, but it’s maybe helpful to point out that this dispute involves at least two severable elements. One question is where to ‘put’ a certain range of scoreboard updating effects—a dynamic pragmatics (e.g., Stalnaker (2018)) puts these updates into the pragmatic portion of the theory, living on top of a static semantics, whereas a dynamic semantics (see Yalcin (2014) for an overview of extensive options in this camp) treats the update effects as part of the core (‘conventional’) semantic properties of the utterance. Where to draw this line is a function of what the drawing of the line is intended to achieve; I continue (Dever (2013)) to be uncertain on that latter topic. Another question is how to characterize what Rothschild and Yalcin (2016) call the conversational systems dynamics. Rothschild and Yalcin (2015) make helpful advances on the van Benthem characterization just discussed, observing that a form of antisymmetry characterizes an interesting class of systems that fail even to be what they call weakly static. But although Rothschild and Yalcin give a usefully abstract perspective that prescinds from many of the (optional) details about scoreboard contents that drive van Benthem’s result, the significance of their formal picture still depends on unsettled questions about scoreboard structure. Suppose, for example, that scoreboards are mnemonic, in that a scoreboard encodes sequential information about all previous states of that scoreboard. Any such system will violate Rothschild and Yalcin’s principle of antisymmetry, but on the face of it even the most static system can be implemented in a mnemonic form. ¹⁰ That’s only a first gesture toward a dynamic proposal, not an actual theory. One obvious problem is that minimal expansions and contractions may not be defined. (Suppose we have the four worlds w₁ = PQ, w₂ = Pq, w₃ = pQ, and w₄ = pq, and consider an utterance of ◊p against a scoreboard whose modal horizon contains the two worlds w₃ and w₄. Is the minimally expanded modal horizon {w₁, w₃, w₄}? {w₂, w₃, w₄}? {w₁, w₂, w₃, w₄}?) Similar choice points arise when we

  

7

contraction operations aren’t easily modeled as results of adding particular propositions to the conversational record, and so reward a proposition-independent, scoreboard-oriented dynamic approach.

1.2 A first pass at scoreboard metaphysics Much needs to be done if scoreboards are going to replace propositions.¹¹ Propositions are a theoretical nexus for a collection of demands deriving from theories of truth, mind, modality, rationality, and others—if we’re going to restructure things around scoreboards, we need to show how scoreboards also can occupy that nexus, or disentangle the various threads that propositions knotted. Here I want to focus on just one aspect of the needed work. If we are going to center scoreboards in our linguistic theorizing, we need good theories about what scoreboards are—not linguistic theories about what scoreboards contain and how they are structured (although we do need those), but metaphysical theories about their nature. The literature on the metaphysics of propositions is large and active, and has provided us both with a number of metaphysical options for the natures of propositions (sets of possible worlds, certain complexes of objects and properties, certain kinds of fact, certain kinds of mental acts, certain kinds of properties, abstractions from certain mental features, or simple metaphysical quietism, to name a few) and with a number of challenges to (and responses to challenges to) formal consistency of theories of propositions, centering around the seek a minimal contraction for a necessity claim scoping over a modal, such as □◊p. Does the necessity operator contract common ground as well as the modal horizon? Or does the modal only contract the modal horizon, leaving it to internal equilibrium tendencies of the scoreboard to adjust common ground to match? (The homeostatics of linguistic scoreboards is a neglected topic.) The dynamic semanticist’s perspective is likely to be that such undefinedness is in the end a good thing—the need to give well-defined update rules provides a useful epistemic probe into scoreboard structure, by prompting us to consider what scoreboards would need to be like to allow location of (e.g.) minimal contractions. (The minimal expansion proposal, for example, invites borrowing from the extensive formal belief revision literature (see Hansson (1999) for a thorough overview) and its methods for dealing with the retraction problem (and, in particular, for structuring belief states so as to make the retraction problem tractable).) ¹¹ Some of those things, including the metaphysical things that primarily animate this paper, need to be done even if scoreboards are an adjunct to, or even an effluence of, propositions. But the projects are more theoretically pressing, and at times need to be pursued in different ways, if there’s a full-on replacement agenda.

8

 

Russell-Myhill paradox.¹² None of this work has been done, and very little has even been started, for the case of linguistic scoreboards. Lewis’s original discussion of scoreboards contains some brief remarks on scoreboard metaphysics: To the extent that conversational score is determined, given the history of the conversation and the rules that specify its kinematics, these rules can be regarded as constitutive rules akin to definitions. Again, constitutive rules could be traded in for explicit definitions: the conversational score function could be defined as that function from conversation stages to n-tuples of suitable entities that evolves in the specified way. Alternatively, conversational score might be operationally defined in terms of mental scoreboards–some suitable attitudes–of the parties to the conversation. The rules specifying the kinematics of conversational score then become empirical generalizations, subject to exceptions, about the causal dependence of what the scoreboards register on the history of the conversation. (1979, pp. 345–6)

Lewis is directly addressing the nature of scores here, but a picture of scoreboards can then easily be extracted. On one option, the scoreboard is an abstract mathematical entity (the score is the scoreboard, thought of ¹² Obvious question: is there a version of the Russell-Myhill paradox for scoreboards? If scoreboards contain propositions as constituents (for example, as part of common ground), and are individuated by contained propositions, and are propositionally permissive in allowing any proposition to occur as a constituent, then propositional Russell-Myhill arguments are automatically inherited by scoreboards. But for that particular Russell-Myhill route, solutions to the Russell-Myhill paradox are also inherited—if, for example, we think that Russell-Myhill is to be avoided via some degree of coarse-graining of the identity conditions for propositions (perhaps Booleanism, as in Dorr (2016) or Classicism, as in Bacon and Dorr (forthcoming)), then the individuation of scoreboards via propositional constituents will be automatically correspondingly coarse-grained. But the richer structure of scoreboards may allow novel routes to RussellMyhill type worries. For example, if discourse referents are individuated by their external anchors (not an obvious assumption), and scoreboards are both individuated by their discourse referents and permissive in allowing that whenever there are some discourse referents, there is a scoreboard containing all and only those discourse referents, then a new Russell-Myhill problem arises. There is a lot to be worked out here, but one suggestive fact is that the standard coarsegraining moves such as Booleanism aren’t going to be of any help in avoiding Russell-Myhill worries deriving specifically from the discourse referent component of the scoreboard. More generally, the structuring features of scoreboards differ from those of propositions that make it hard generically to port low-structure treatments from one environment to the other.

  

9

in this way); on the other, the scoreboard(s?) is(are) a mental state which then fixes what the score is.¹³ Something broadly in the tradition of Lewis’s mental state picture (but really more in a Stalnakerian tradition, developing from work on the nature of common ground) has become the industry standard for the metaphysics of linguistic scoreboards. On this industry standard, scoreboards are a (or are determined by; the exact details of whether this is a reductive analysis or something like a grounding story won’t distract us here) certain combination of psychological and epistemic facts. Since the epistemic facts are primarily routed through epistemically-constrained mental states of common knowledge, we’ll call the resulting picture a ¹³ Lewis also expresses dissatisfaction with both of these options. On the ‘mathematical’ side, he notes that the rules for scoreboard kinematics may underdetermine what the subsequent scoreboard state is. Mere kinematic underdetermination, though, isn’t enough to create a problem—so far, we could merely replace an update function with an update relation, allowing multiple scoreboards to be descendants of the current scoreboard (or, equivalently, traffic in higher-order scoreboards taken as sets of lower-order scoreboards, and then translate underdetermination of lower-order kinematics into a determinate-but-expanding higher-order kinematics). It’s only when we add the assumption that, even in the face of multiple rule-conforming candidates for the posterior scoreboard, there is one scoreboard that is the true scoreboard that these underdetermination-absorbing moves become unsatisfactory. We then need to know whence the privileged status of the one true scoreboard arises—if not within the rules, or some other suitably abstractable feature, then the mathematical approach will certainly be problematic. (There is an obvious connection here with Lewis’s picture of the actual language relation in Lewis (1975), where the need as the actual language relation as a supplement to the abstracted mathematical characterization of languages is driven by the need for some empirical tiebreaker among the underdetermined candidates.) We will return to the privileging question below. On the ‘psychological’ side, he notes that it may be difficult (‘without risk of circularity’) to specify the right mental states to constitute the scoreboard. The Cognitivist metaphysics of scoreboards discussed immediately below can be thought of as a response to Lewis’s challenge of difficulty here; we will return briefly below to circularity worries (although I am not confident that the circularity worry I focus on is the same that Lewis cautions against). Lewis’s own proposal (“a middle way”) follows his usual approach to theoretical terminology by functionalizing a second-order definition out of the theoretical role, and then taking the theoretical object itself (in this world) to be the realizer of the second-order definition. Thus Lewis suggests that “the rules specifying the kinematics of score . . . specify the role of a scoreboard; the scoreboard is whatever best fills this role”. But at the same time, he seems to presuppose that it will be mental features of conversational scoreboards that provide that “whatever”: “we assume that some or other mental representations are present that play the role of scoreboard, in the following sense: what they register depends on the history of the conversation in the way that score should according to the rules”. (So the “one true scoreboard” is fixed by what mental states there actually are that play the scoreboard role, even if more than one mental state type could play that (underdetermined) role.) When the theoretical functionalism is thus pinned to a mental white knight, the question becomes whether the mind contains the needed knight, and the Cognitivist tradition discussed below is the quest for that knight. If not thus pinned, the question becomes whether there is anything that best fits the role. (I assume that the logic of ‘best fitting’ doesn’t guarantee the existence of any candidate best-fitters.)

10   Cognitivist picture of linguistic scoreboards. The Cognitivist picture starts with a base mental attitude of acceptance. We’ll run first with the illustrative example of the common ground component of scoreboards; how (and if) to generalize to other scoreboard components will occupy us later. The acceptance picture: common ground represents a body of information shared by the conversational participants. It’s tempting to think that that information is shared, in the first instance, by way of being believed. But standard examples show that belief isn’t the right attitude. Information can enter into common ground even when not believed by all (or even any) of the participants—as when a presuppositional utterance is allowed even when some don’t believe the presupposition (we allow “John’s spouse will be at the party” even though we don’t believe that John and his partner are married, thereby contributing to common ground containing that John is married), or when a discourse is carried on under a mode that isn’t truth-oriented (we allow the utterance “The Hulk is stronger than Superman”, and thus contribute to common ground containing that the Hulk is stronger than Superman, despite not believing this, because our goal is to fictionally speculate, not to describe the world). Instead we want some other less committal attitude that we’ll stipulatively (following Stalnaker) call acceptance. But common ground is not just that which is accepted, or even that which is accepted by all conversational participants. Because linguistic interaction is meant to be a collaborative and coordinated activity, we want the linguistic scoreboard to be the kind of thing that can rationalize and explain such collaboration and coordination. But a standard picture deriving from game theory, and consideration of the Byzantine Generals problem (Lamport et al. (1982)) suggests that in order to achieve the needed coordination some form of common attitude is needed. If we all accept that p, but some of us aren’t aware that others of us accept that p, then we won’t be able to rely on acceptance of p when planning conversational contributions. (And so on for unawareness that we are all aware that we accept that p, and further iterations, for increasingly recondite conversational planning instances.) The required common attitude is then typically defined via iteration. That which is commonly φ’d is that which is φ’d by all participants, and φ’d by all participants to be φ’d by all, and φ’d by all to be φ’d by all to be

  

11

φ’d, and so on. But the iterated attitude can’t be acceptance.¹⁴ One can accept that Jones is married without believing that John is married, and that acceptance can help explain conversational behavior (willingness to use and appropriately interpret “John’s spouse”, for example). But if one merely accepts, without believing, that the other conversational participants accept that John is married, it won’t make sense to plan or interpret conversational contributions on the assumption that others accept that John is married. (One can speak on the basis of acceptance of p because, so long as we are all speaking as if p, one can predict reception of utterances (e.g.) presupposing p. But what matters is whether we really are speaking as if, not whether we are speaking as if we are speaking as if.) If our goal is to explain coordinated behavior, we need at least iterated belief; if we want to rationalize behavior and we agree with (e.g.) Hawthorne and Stanley (2008) that something is a reason to act only if we know it, then we need iterated knowledge. The final Cognitivist picture: the linguistic scoreboard is constituted by the common knowledge of acceptance by all conversational participants. X is on the scoreboard just in case everyone (with proper domain restriction)¹⁵ accepts that X, and everyone knows that everyone accepts that X, and everyone knows that everyone knows that everyone accepts that X, and so on.¹⁶ (Two versions of this picture: (i) what the score is is constituted by these Cognitive = psychological/epistemic facts, but the scoreboard itself recording that score is an abstract mathematical entity (to which the conversational participants stand in the actual scoreboard relation by virtue of that one entity reporting what the score really is), and (ii) not just the score, but also the scoreboard itself, is constituted by the Cognitive facts, so that the scoreboard is a

¹⁴ This point is made by Stalnaker (2002). ¹⁵ Any good account of scoreboard technology needs a story about the boundaries of conversations—both synchronic, fixing which people and which utterances are part of the conversation, and diachronic, fixing when the conversation begins and ends. I won’t have anything good to say about these boundary problems. (I suspect that others also don’t have anything good to say.) My rough suspicion is that the synchronic problem is harder for the Cognitivist account than for the alternative I’ll set out, and that the diachronic problem is harder for my alternative than for the Cognitivist. But I won’t attempt to back up that suspicion with actual arguments.) ¹⁶ Probably both the de re and the de dicto readings of all of those quantifiers are needed, at every level of iteration.

12   Cognitive, rather than a mathematical, entity. For the Cognitivist, it doesn’t look like there’s anything at stake in this choice, but it will loom larger for the alternative I’ll set out soon.) It’s easy to find the Cognitivist picture attractive. It’s natural to think that language is a human cognitive tool—that language means what it does because of our thoughts and our purposes in using the language, and that the use of language is a tool for achieving cognitive coordination with other speakers. Cognitivism gives us a linguistic ontology that makes sense of language as a cognitive tool, and an integration between language and mind that makes sense of why we care about linguistic meanings—the role of utterances in updating linguistic scoreboards simply tracks the role of those utterances in effecting changes in the mental states of conversational participants, as well as in our common awareness of those changes. And the shift from propositions to scoreboards gives us linguistic tools that can track and reflect a wider range of mental states—curiosities, preferences, plans, attentions; as well as beliefs. If linguistic meanings are scoreboard updates, and if scoreboards just are collective Cognitive features of us, then there’s also a clear epistemology for linguistic theorizing—to work out linguistic meanings just is to work out how we change our collective mental states upon exposure to linguistic utterances. That clear epistemology gives a sensible place, but not a uniquely privileged role, for armchair reflection on linguistic meaning. Despite this attractive integration of mind and language offered by the Cognitivist scoreboard account, I want to offer an alternative metaphysics of scoreboards that enforces a sharp separation between what the language provides and what’s going on in our heads. I say “despite”, but more honest here would be “because of”.¹⁷ The integrated Cognitivist picture shares with, for example, the Gricean program the aspiration of treating linguistic meaning as a certain well-behaved corner of human psychology (and sociology)—we get a theory of language as such because ¹⁷ I’ll retain “attractive”, though—I acknowledge the attraction. Many authors have been drawn into opening papers with claims like “Human languages are sophisticated tools for transferring mental states from one person to another” (Dever (2012)). Such it is to be in grips of a picture! (As the replacement picture offered here hopefully makes clear, I’m happy to endorse that claim with “can be used as” replacing “are”.)

  

13

we discover that the various “thought-transference” and “mind-reading” strategies that we deploy systematize around some mathematically tractable strategic clusters. (Communicative intentions, conversational maxims, and epistemic reliance on conventional regularities in the Gricean case; update functions and scoreboards in the Cognitivist case.) But I share with (e.g.) Buchanan and Schiller (2022) a pessimism about whether our psychology is even locally and heuristically systematizable in this way. I suspect any attempt to sieve a theory of language out of the soup of human psychology finds the lumps of meaning increasingly dissolving into a general cognitive slush in which there’s little more to say than, “we make more-or-less desperate use of whatever scattered sources of insight and evidence we can grab onto in the complex interpersonal situations in which we find ourselves to get ourselves more-or-less cognitively correlated with one another”. Something of a let-down from the promising start in the Begriffsschrift. Maybe we need to learn to live with disappointment, but I’d like to make a run at a depsychologized safe house for language. That’s all programmatic, which means that one can simply decline the program. So I’d also emphasize some specific difficulties that the Cognitivist inherits. As usual, they’re difficulties that can surely be addressed, but they maybe tilt the balance sheet enough to make the nonCognitivist alternatives worth more careful consideration. None of these difficulties is fully original, but I’ll try to extract a particular perspective on the Cognitivist project from them.

1.3 Three problems for cognitivist scoreboards First (and most familiar) difficulty: Cognitivism requires common knowledge of acceptance in order for a score to get recorded on the scoreboard. But common knowledge is an unrealistically demanding notion. For A and B to commonly know that p, each needs an infinite store of knowledge, and knowledge of an increasingly recondite form involving increasing iterations of epistemic operators. In actual practice, people rarely act as if they are tracking these recondite facts. Consider the Keynesian Beauty Contest, or the “Guess 2/3 the Average”

14   games—experimental results show that people are strategizing as if tracking one, or maybe two, levels of epistemic iteration. And plausible epistemic assumptions make it hard to see how they could reasonably do much better, even if we abstract from the cognitive load demands. The iterated knowledge-of-knowledge facts are logically and epistemically independent, and don’t seem to flow from any plausible background assumption, so each further level represents a new epistemic accomplishment. There’s no good story about how we come to be so (infinitely!) accomplished. Perhaps common knowledge should be treated as an idealization.¹⁸ But Lederman (2018) shows that even idealized agents under the right conditions won’t be able to achieve common knowledge. That’s not decisive—maybe we’ve just picked the wrong epistemic constraint, and we need something other than common knowledge to play the relevant role.¹⁹ ‘Playing the relevant role’ involves enabling the coordinative role of the scoreboard; it remains an open question, I think, whether there is anything plausible short of common knowledge that can do that. Second difficulty: I don’t really know what acceptance is. Stalnaker gives us a gloss on it:

¹⁸ Consequence: there typically is no linguistic scoreboard; the best we can say is that under idealized conditions there would have been a scoreboard. But that’s rather unsatisfactory—given the dynamic framework, if there is no scoreboard, there are no facts about what we mean (although there can still be facts about what we would have meant, under idealized conditions). And if there are no facts about what we mean, we’ve landed in an unexpected content nihilism. Perhaps some of the explanatory and theoretical role of content can be replaced by hypothetical idealized contents. (I am skeptical. Is it enough for holding people responsible for their utterances that they have done something that if done by (and in the company of) epistemically idealized agents would have said that p? We don’t normally hold people responsible based on such idealized analogs to their actions.) Or perhaps we simply relax the idealization and drop the common knowledge constraint. But then the scoreboard plays a much less systematic role in explaining our linguistic behavior, as agents can disagree about what the scoreboard says and must negotiate such disagreements in planning. This puts us a large step closer to what Buchanan and Schiller (2022) call the Wild West. ¹⁹ Lederman (2018) and Lederman (2014) suggest that a notion of public information can fill the explanatory role for which common knowledge is intended, but also suggest that there is no unified explanation of what makes for public information. (Perhaps for it to be public information (for us) that p is for us to believe that p, where our believing that p isn’t reducible to some host of facts about each of us believing thus-and-suchly. I think such routes are worth exploring, but it’s also worth noting that this kind of non-reductive we-attitude is a (small?) step away from the Cognitivist approach, and a step toward the alternative picture I’ll push below on which scoreboards don’t depend on mental states. Perhaps the same holds true for other implementations of public information; it’s hard to say without the idea being further developed.)

  

15

Acceptance . . . is a category of propositional attitudes and methodological stances toward a proposition, a category that includes belief, but also some attitudes (presumption, assumption, acceptance for the purposes of an argument or an inquiry) that contrast with belief, and with each other. To accept a proposition is to treat it as true for some reason. One ignores, at least temporarily, and perhaps in a limited context, the possibility that it is false. (2002, p. 716)

That’s suggestive, but getting clear on what is being suggested is hard. What is it to ‘treat as true’? Suppose we don’t believe that John is married, but we allow an utterance of “John’s spouse will be at the party”, thereby guaranteeing (given our theory of presupposition) that it’s on the scoreboard that John is married, and thus (given the Cognitivism) that we accept that John is married. There remain clear ways in which we don’t act as if John is married—we have no inclination to celebrate his anniversary, or buy him a wedding gift, or look for a wedding ring on him. Our ‘treating it as true’ seems to be limited to a linguistic treating as true. But why think that this linguistic treatment has anything to do with truth (actual or treated)? Unless we’re already supposing that the linguistic activity is fundamentally truth-oriented, this seems out of place. When we accept that the Hulk is stronger than Superman, as part of a suppositional superheroic language game, are we in any interesting sense treating it as true that the Hulk is stronger? All that really seems available here is that in accepting that John is married, we are making and allowing the linguistic moves that follow from the scoreboard’s noting that John is married. Given a dynamic theory, we can enumerate those moves—an altered QUD changes relevance relations, discourse referents for John and his spouse allow new pronominal anaphora options, and so on. But if the scoreboard is constituted by an attitude of acceptance, and the attitude of acceptance is exhausted by permitting scoreboard-licensed dynamics, we’ve traveled in a very small circle.²⁰

²⁰ The first and second difficulties interact. The more theoretically recondite the attitude of acceptance is, the more epistemically and psychologically unrealistic it is that we have knowledge of acceptance, let alone iterated knowledge of acceptance.

16   Third difficulty: the acceptance picture is built for the common ground component of the scoreboard, but it’s not clear how to extend it to other scoreboard components. Acceptance, like belief, is a propositional attitude. (Prescinding from the previous concerns) we can make sense of accepting that John is married, and thus make sense of a Cognitivist story that places (in the common ground component of the scoreboard) that John is married.²¹ But when the scoreboard includes discourse referents for John and for John’s spouse (dynamically introduced by two definites in “John’s spouse will be at the party”), what are we to be accepting to get those discourse referents on the scoreboard? How does acceptance connect us to actions, modal horizons, ordering sources, preference rankings, salience rankings, questions, and other non-propositional scoreboard components?²² There’s an obvious strategy for responding to this difficulty: ubiquitous propositionalizing. If the discourse referent isn’t the kind of thing we can stand in the acceptance relation to, we replace the discourse referent itself with a proposition about the discourse referent. The scoreboard doesn’t contain the discourse referent for John; rather, it contains the proposition that there is a discourse referent for John. And that proposition we can accept. More generally, whatever kind of entity we prima facie wanted on the scoreboard, we put instead on the scoreboard the

²¹ Well, the Cognitivism as stated gets it on the scoreboard that John is married, but it doesn’t include an account of why that information goes specifically on the common ground portion of the scoreboard. Suppose that as the result of the dynamics of a conditional utterance, the scoreboard currently contains a subsidiary local scoreboard updated with the antecedent of the conditional, as well as the global, scoreboard. Does acceptance put that John is married in common ground of the local or the global scoreboard? And whichever it does, how is the other placement achieved? Perhaps we need in addition a notion of conditional acceptance, so that that John is married goes on the local scoreboard if we conditionally accept it conditional on the antecedent of the relevant conditional. As scoreboard components proliferate, the range of acceptance-like attitudes needed will also proliferate. (This can be taken as an argument against proliferation, of course.) ²² My advice for Cognitivists here: go holistic. Take the basic unit of mental representation to be the scoreboard, so that each of us stands in a sui generis relation (‘having’) to a particular scoreboard characterizing our total mental state. To believe that p is then to have a scoreboard that supports (is unchanged after updating with) p. (Side benefit: we get, inherited from dynamic semantics for proposition attitude verbs, an account of belief that extends to belief reported by sentences that don’t have propositional content.) So we take seriously the metaphor of the scoreboard as a public mind; the public mind is that mind which bears the right relations to the several scoreboards the conversational participants have in their heads.

  

17

proposition that that entity exists, or has some relevant relation to the conversation.²³ Which does the scoreboard contain: the discourse referent, or the proposition that the discourse referent exists? The question, or the proposition that we are seeking an answer to the question? Answering this question requires a theoretical criterion for scoreboard occupancy. If the scoreboard is constitutively a record of what we commonly accept, then it will just contain the propositions. But that picture is at odds with the mechanics we in fact demand of scoreboards. Features get placed on scoreboards as a mark of, and means for achieving, their shiftability. Two sketches of examples: • Conditionals can shift the modal horizon. “The die might land 5, but if it lands even, it must not land 5” is acceptable because the conditional antecedent shifts the modal horizon, under the dynamic influence of “the die lands even”, to no longer endorse the possibility of landing 5. • Quantifiers can shift the objectual salience ranking. “He is wearing a hat, but not every man has a hat on his head” is acceptable because the “every man” quantifier shifts the salience ranking so that the pronoun “his” no longer picks out the original hatted man. So a tempting hypothesis: shiftability is criterial of scoreboard listing. The theoretical role of the scoreboard is to track a variety of broadly informational indices of evaluation that can be shifted by a variety of linguistic devices, to allow overall compositional evaluation of complex constructions. But a criterion of shiftability is at odds with a strategy of ubiquitous propositionalizing. Ubiquitous propositionalizing commits to the

²³ This propositionalizing move has a Stalnakerian pedigree. Stalnaker (1998), for example, makes this move contra Kamp in arguing that we don’t need a discourse referent component to context in addition to common ground. Stalnaker’s discussion suggests a master argument for propositionalizing—if everything about the discourse is in common ground, then it’s common ground that the particular linguistic moves of the conversation have happened. If a particular dynamic theory then dictates that those moves place various non-propositional features on the scoreboard, we can cut out the middle man and capture the same theoretical goodies by routing through the information that those moves were made.

18   scoreboard contents always being propositions, and hence to all shiftability being shifting of propositional bodies of information. But it’s not always information that we shift. That’s the lesson, for example, of Thomason conditionals (attributed in van Fraassen (1980) to Thomason). Consider the contrast between: • If it is raining, I don’t know that it is raining. • If it is raining, it might not be raining. The unacceptability of the second conditional shows that the antecedent is shifting some component of the scoreboard—namely, the modal horizon used in interpreting modals. But that shift doesn’t shift what the scoreboard represents us as knowing, as evidenced by the acceptability of the first conditional.²⁴ The phenomenon is ubiquitous in scoreboard dynamics. Similarly, quantifiers shift salience rankings without shifting knowledge of attitude rankings: • Although he is wearing a hat, each linguist at the party knows that he himself is not wearing a hat. • Although he is wearing a hat, each linguist at the party knows that the person we know to be maximally salient is not wearing a hat. So if the linguistic point of having items on the scoreboard is to allow their dynamic shiftability by linguistic operators, the presence of those items on the scoreboard needs not to be a matter of our bearing some attitude toward propositions involving those items. None of the above is a refutation of Cognitivism, but it’s enough to motivate looking for a different approach to linguistic scoreboard metaphysics. For a starkly different approach, I want to explore the possibility

²⁴ Perhaps the Thomason conditional is probing the wrong cognitive shift? But the phenomenon seems broadly insensitive to the specifics of the propositional attitude used: If it’s raining, I don’t accept for conversational purposes that it’s raining. If it’s raining, I don’t know that among the worlds taken seriously in this conversation are worlds in which it is raining.

  

19

of a mind-independent conception of linguistic scoreboards.²⁵ The core idea is that linguistic scoreboards are token mind-independent particulars of bodies of information (and other associated score components), created and manipulated by language uses—so that the theory of language is the empirical theory of the causal consequences of various linguistic expressions for the scoreboard state. We, qua linguistic agents, are demoted from scorekeepers to score viewers.²⁶

1.4 Three grades of scoreboard non-cognitivism There is a lot to do to make this alternative approach minimally plausible, and I won’t do it all here. But we’ll take at least a preliminary look at three topics. First, we’ll consider what an ontology of mind-independent scoreboards could look like—how are we to make sense of conversations involving ghostly token bodies of information? Second, we’ll consider what the connection between us and language is on this picture—if scoreboards are not reflections of our shared mental states, what theoretical role is played by having them around? And third, we’ll take an initial look at some of the detailed ways in which (e.g.) dynamic theories about scoreboard updating need to be reworked for the alternative approach. Sports scoreboards can be mind-dependent or mind-independent. In fact, they can be mind-dependent or mind-independent in two different ways. First, what the score is, or how the score is changing, is sometimes constituted by our attitudes. Part of the score in a baseball game is whether the most recent pitch was a ball or strike. On one (minddependent) view, whether the pitch was a ball or strike is simply constituted by the umpire’s decision about whether it was a ball or strike

²⁵ Mind-independence turns out to be not quite the right way to put it, but it’s a decent starting point. ²⁶ That’s a bit too strong. On the Cognitivism view, we are constitutively scorekeepers—the score is what it is more or less in virtue of our taking the score to be thus. On the alternative view, the score is what it is constitutively independently of us, but our linguistic actions can still change the score (not through us directly, via our scoring attitudes, but through our actions in using language). That doesn’t make us scorekeepers, but it can make us scorers, rather than merely spectators.

20   (so that, for example, there is no possibility of the umpire having gotten it wrong). On another (mind-independent) view, whether the pitch was a ball or strike is constituted by facts about the pitch’s location, and the umpire is just an epistemic agent for figuring out the score.²⁷ Minddependence of the score can be driven by a desire to have good epistemic access to what the score is, but the advent of (e.g.) Hawk-Eye and other automated scoring technology lets us tie the score to mind-independent but easily verified facts. Second, the recording of the score on a scoreboard can be minddependent or mind-independent. When we play bridge, we might keep track of the score by just all remembering it, and all trusting all of us to remember it. Then the bridge scoreboard is a Cognitivist entity. But of course frequently we make use of a mind-independent public object for the recording of the score—the scoreboard is the sheet of paper with the scores written on it.²⁸, ²⁹ I am looking for both forms of mind-independence, but it’s the second form that is particularly difficult to provide. First option: the mathematical scoreboard. Here we take a leaf from Lewis (1975), and treat scoreboards as mathematical entities, and languages as other ²⁷ It’s a vexed question what the actual rules of professional baseball say on this matter. The MLB rulebook contains apparently normative language about what the umpire “shall” do (5.02b(2): “The umpire shall call ‘Ball’ or ‘Strike’, as the case may be.”) But what if the umpire does not do what he shall do? Does he get it wrong, or constitutively make it the case that he is right? 8.02(a) says that “Any umpire’s decision which involves judgment . . . is final”, but what does “finality” amount to? Correctness? The rule goes on to say “No player . . . shall object to any such judgment decision” (but then contemplates penalties for objections), but impermissibility of objection is compatible both with practically irrelevant incorrectness and with constitutive correctness. ²⁸ Note that on both of these views, the scoreboard can get it wrong, misrecording the actual score. With a mind-independent scoreboard, there is another locus for error—we can get it wrong about what the scoreboard says (which we can’t do for Cognitivist-style mind-dependent scoreboards). ²⁹ We’ve seen some of the ways that scoreboards have gotten more technologically sophisticated both in determining (Hawk-Eye) and recording (video instant replay) the score. Another dimension of technological variation is scoreboard connectivity. Outbound connectivity is improved by scoreboard lighting, size, and positioning (getting the Infinity Screen’s outbound connectivity right through the right structure and placement was a significant part of its technological advance), and in more sophisticated cases by Wi-Fi-readiness that lets users receive the scoreboard state from their own devices. Inbound connectivity can let users query the scoreboard, requesting that it display particular information. Cognitivist linguistic scoreboards come with an off-the-shelf connectivity package enabled by the constitutive connection of the scoreboard to common knowledge. You get what you get with the common knowledge connectivity package—it’s a nice basic system, and it will be an issue for my non-Cognitivist alternative whether it can do as well, but there’s no real option for any connectivity not already granted by common knowledge.

  

21

mathematical entities (functions from strings to update functions on mathematical scoreboards). There is no doubt that the scoreboard, on this view, is a mind-independent entity, updated according to mindindependent rules.³⁰ But, of course, there are many mathematical scoreboards, and many mathematical scoreboard update operations. When some people are having a conversation, no one of these scoreboards is yet “present” to the conversation more than any other. So if we want some scoreboard to be this conversation’s scoreboard, another piece of the story is needed. That piece, of course, is the actual-scoreboard relation— a relation which presumably is deeply cognitively implicated, and which, prima facie, may just be the Cognitivist story repurposed from constitutive metaphysics to privileged-entity selection. The good thing about the mathematical scoreboard model is that the metaphysics is clear, and there definitely are such entities. If we’re desperate to find a psychology-free domain in which to do our linguistics, this is a safe room of last resort. But since the mathematical model can’t, without psychological involvement, privilege one scoreboard over another, the resulting style of linguistics is the linguistics of all possible languages. I’m all in favor of branches of pure math, but it does seem like a very different project from what we set out to do. So a second option: the normative scoreboard. On this model, scoreboards are constituted by clusters of norms. To say that scoreboard S is the scoreboard for conversation C is to say that participants in C, for example, may use pronouns in a way licensed by the discourse referents of S, and should make assertions that address the QUD of S, and so on. The scoreboard just is that cluster of normative facts.³¹ The normative facts here can be suitably mind-independent—what we ought to do, linguistically, need not reduce to (or otherwise connect to) anything about what we think we ought to do.³²

³⁰ No doubt, so long as your ontology of mathematics is itself a mind-independent one. ³¹ If we prefer, we can combine the first and second options: the scoreboard is a mathematical entity, and so has the internal structure of the sort that standard presentations of scoreboards in the semantics literature depicts, but the actual-scoreboard relation rests on these normative facts, rather than on psychological facts. ³² This is one spot in which the ‘mind-independence’ framing needs some qualification. Norms such as ‘One should assert so as to address QUD Q₁’ are at least indirectly about minds, and that these norms are in place is at least partially a result of speakers having intentionally performed speech acts that enshrined Q₁ on the scoreboard, so the norms are mentally

22   Normative scoreboards, unlike mathematical scoreboards, can be (on their own basis) associated with particular conversations. Given a particular conversation C with participants S, there just really are facts about what S ought to and can do in continuing the conversation, and those facts will be the normative scoreboard. And normative facts can have causal histories—that the speakers can use pronouns in certain ways was caused by earlier linguistic interventions, so we can take the dynamic semantics as a story about the genuine causal potencies of expressions. But it’s still not quite as psychology-free as we could hope for. The norms are norms about what speakers can and must do, so they are norms governing actions, and actions are at least a psychologized category.³³ A third and final option, then: informational scoreboards. Scoreboards are just real parts of our ontology. They are (for some value of ‘concrete’) concrete particulars—not so concrete as a massive video screen hanging from a stadium ceiling, but as concrete as the entities of other special sciences. Consider economics—in economics we find mention of particular entities such as recessions which can seem challenging to fit into our ontology. One response to that challenge is some form of reductionism, seeking to identify the recession with some other collection of things on better ontological footing. (The reductivist quest here might in particular be for a psychological reduction, identifying the recession with some complex of people’s attitudes—perhaps even iterated epistemic attitudes.) Maybe that reductionism will succeed,³⁴ but if it does not, another response is to accept recessions as among the furniture of the universe. We can then theorize about recessions—for example, identify their causes and effects, and investigate their epistemology by considering how and how much we know about them.

implicated in their causal history. I am assuming the relevant scope of mind-independence emerges from the details of the examples; I care more about the kinds of scoreboards bruited in the examples than I do about their exact philosophical labeling. ³³ The low-level psychological involvement here is connected to the fact that normative scoreboards have something in common with the propositionalizing strategy discussed above— since norms are all norms that φ, for some proposition φ, non-propositional scoreboard components need to be routed through propositions about how to use those scoreboard components. ³⁴ I am no metaphysician of economics, so I want to tread lightly here, but I’m skeptical.

  

23

Similarly for scoreboards. We can recognize that the science of language calls for certain entities, and take ourselves thereby to have discovered the existence of those entities. Absent a reductivist project, there may be nothing more to say about what scoreboards are than the host of theory-level claims like: they are things containing common grounds and discourse referents. On this picture, the scoreboards are just out there (in the same way that recessions are just out there), and languages are toolkits of devices for altering the contents of scoreboards. The workings of those scoreboards and those scoreboard-altering tools is in principle something that can be theorized independent of any mention of the humans (and the psychologies of the humans) using those tools.³⁵, ³⁶ Suppose now that some form of non-Cognitivist scoreboard account is viable. (Hopefully the third, informational account, but we’ll work with what we can get.) We have now decoupled the technology of language from human cognition. But then what to make of us as language users? Roughly, if scoreboards aren’t anchored in our cognitive states in the use of language, why care about scoreboards? The first step is to suggest a fundamental reorientation toward the role of scoreboards.³⁷ On the Cognitivist picture, the participants in a conversation all start with their individual bits of knowledge. Some of that knowledge is shared (and commonly shared) knowledge; that portion thereby resides on the ³⁵ Perhaps even if recessions don’t reduce to some non-economic stuff, they are at least grounded in such stuff? And thus perhaps scoreboards are also grounded in some non-linguistic stuff? (Or do we want recessions and scoreboards to be not just among the furniture of the world, but among the fundamental furniture?) Such groundings might present no real challenge to the informational scoreboard view, but we might worry that the grounding story will just take us back into the psychological soup. I suffer somewhat from grounding tone-deafness, so I’m not entirely sure how much to worry about presence or absence of grounding relations. But I would be tempted to suspect that any adequate grounding stories here will need to reject socalled “Big G” grounding (Wilson (2014)) in favor of a pluralism that defends multiple grounding relations—perhaps a multiplicity that goes beyond Fine (2012)’s metaphysical, normative, and natural to include economic and semantic grounding as well. My suspicion is that a story that appeals to specifically semantic grounding facts for linguistic scoreboards can then avoid troubling psychological contamination. ³⁶ It’s tempting to suggest that the ontology of linguistic scoreboards is just a special case of the ontology of social artifacts, and thus that familiar lessons from social ontology can be applied here. But I would be hesitant to call linguistic scoreboards social artifacts—doing so prejudges questions about the priority relations between us and language. ³⁷ It would be grandiose to call this a Copernican revolution in the theory of language, but perhaps we can lay claim to something humbler, such as a Bradleyan revolution (Dyson (1996)).

24   conversational scoreboard. Participant A knows some p that isn’t currently on the scoreboard, and so asserts some sentence S whose content is that bit of private knowledge. Everyone else, on reception of the assertion of S, adjusts their own mental state to know/believe/accept that p. Since everyone can see that everyone is doing this, the cognitive changes are suitably public, and the scoreboard now records that p. So, unsurprisingly, the scoreboard changes follow and are (systematically) induced by the cognitive changes.

1.5 Using scoreboards without scorekeepers On the non-Cognitivist picture, the scoreboard contents don’t have to have any special relation to what any conversational participant is thinking. Nevertheless, when A privately knows that p and then asserts S to express p, the standard scoreboard dynamics guarantee that the scoreboard will be updated so that its common ground now contains the proposition that p. The several conversational participants are now confronted by a p-recording scoreboard—there is a public informational entity before them reporting that p. How they then react to that confrontation is now a question of individual psychology. When the conversation is going “well”, they may be inclined to believe the information in common ground on the scoreboard, and so update their beliefs (and, given that the scoreboard is likely a reliable information source, perhaps also thereby come to know that p). But people do many things for many reasons—some participants may, upon seeing the revised scoreboard, take some more hands-off acceptance-like attitude toward p. Or they may take no attitude toward p, or some con-attitude toward p. Who knows? But the central idea is that now conversational participants are (in one way or another) training their personal psychologies on the public scoreboard. So to use language to influence someone’s beliefs, I don’t say something to change their mind and thereby change the scoreboard. Rather, I say something to change the scoreboard and thereby (hopefully) change their mind. For this inverted story about the role of the scoreboard to work out, conversational participants had better be able to know what the current

  

25

scoreboard reports. We’re not going to be able to train our psychologies on a scoreboard whose contents we can’t read.³⁸ But how do we know what the scoreboard says? Whatever a good non-Cognitivist account eventually delivers, it isn’t going to be an LED screen hanging above our heads. There’s no straightforward perceptual story about our epistemic access to the linguistic scoreboard. I don’t have any general story about our epistemic access to linguistic scoreboards—but then, I don’t have any general story about our epistemic access to much of anything. What I do have is some skepticism that there’s any special challenge faced here. Is there reason to doubt that we can learn things about linguistically-induced bodies of information? (Some reason, for example, that isn’t reason to doubt that we can learn about recessions or renaissances?) It’s not as though nothing can be said here. Some important part of how we know about scoreboard contents is that we know what has been said in the conversation, and we understand the language well enough to know how what has been said will update the scoreboard.³⁹ Of course, sometimes this method will fail us (scoreboard updates might be more complex than we expected), and sometimes we will fail it by losing track. In bad cases, everyone in the conversation may completely lose track of the scoreboard contents. In such cases, of course, the scoreboard will no longer play one usual explanatory role of accounting for cognitive changes in the conversational participants—but that the scoreboard can fail to play this role is no reason to doubt that it generally does. A final topic: I’ve suggested that we can make metaphysical sense of non-Cognitivist linguist scoreboards, and that such scoreboards can be posed to play a role in explaining how we use language to influence each other’s minds. So hopefully there is decent theoretical life in the nonCognitivist approach. However, a Cognitivist take on scoreboards has certainly dominated linguistic theorizing, and in many places that’s

³⁸ This is where the issue of connectivity technology for non-Cognitivist scoreboards arises. ³⁹ Of course, our understanding the language derives in part from our having seen in past instances what the language has done, by seeing how it has updated the scoreboard. Some of this apparent circularity will ground out (we see how it has updated the scoreboard by seeing how others have adjusted their psychology in response to the scoreboard), and some of it looks epistemically harmless.

26   resulted in the scoreboarding technology being set up in ways that are only suitable for the Cognitivist. The non-Cognitivist thus has a large task looming: to seek out areas that are designed with a Cognitivist presupposition, and figure out how to rework or replace them. As proof of concept, I will discuss one such trouble spot here. Part of the standard semantic package is that some expressions are presupposition triggers. Because of the change-of-state verb, “John hasn’t stopped smoking” presupposes that John used to smoke; because of the factive verb, “John doesn’t regret breaking the window” presupposes that John broke the window. The standard dynamic implementation of presupposition is then that presuppositions are definedness conditions on updates—the update rule for “John hasn’t stopped smoking” is only defined for scoreboards that contain the information that John used to smoke; for scoreboards that don’t report that, there is no defined update. (The scoreboard then fails to update, or is destroyed—there are theoretical choices here.) But, of course, often we can say something with a presupposition not already satisfied on the prior scoreboard, and the conversation proceeds smoothly. Consider Karttunen (1974)’s “We regret that children cannot accompany their parents to the commencement exercises”. When this sentence is uttered in a context in which the scoreboard did not already encode that children cannot accompany their parents, it does not bring the discourse to a crashing halt, or pass through the conversation with no uptake effects. Rather, somehow or other it brings it about that the common ground now does contain the information that children cannot accompany their parents. On the standard (Cognitivist) account, the mechanism here is accommodation. Accommodation is conceived as a repair mechanism. When a presuppositional sentence S, presupposing that p, is uttered in a context in which the scoreboard does not encode that p, we all see that something has been said that would crash the conversation. We then speculate that the speaker did not want to crash the conversation. Since we can see, via the presupposition, that the utterance of S can only fail to crash the conversation when p is in common ground, we assume that the speaker wants p to be in common ground. That is (given the Cognitivism), the speaker wants us all to accept that p, and be commonly aware of that

  

27

acceptance. If we are feeling cooperative, we cooperate, and do so accept. If it’s obvious enough that we are all being cooperative, we’ll be commonly aware of the acceptance, and—lo!—p is on the scoreboard, and now updating with the assertion of S can proceed apace. As Kratzer (1981) says, “this is black magic, but it works in many cases”. But as with much magic, it’s the magic of expectations—the key to achieving the effect that the magician wants is to manipulate the audience expectations. Such manipulation in turn does what’s needed only because, for the Cognitivist, our expectations about the scoreboard contents just are the scoreboard contents. But how is the nonCognitivist supposed to account for presupposition accommodation? If the dynamics of the presupposition yield an undefined update, what further mechanism can step in to get the mind-independent scoreboard into an acceptable state? If accommodation is a repair mechanism, who or what is to do the repairing? Things look just unacceptably magical here. The key for non-Cognitivism is another inversion of perspective. We should stop thinking of accommodation as a repair mechanism. Rather, accommodation is the ordinary case. In the ordinary case, a presupposition sentence such as “John doesn’t regret breaking the window” updates the scoreboard in two ways—first, by placing on the scoreboard the information that John broke the window, and second, by placing on the scoreboard the information that John doesn’t regret that. So we don’t need a repair mechanism, and hence we don’t need a psychological story about how repair happens via our rational shifting attitudes. What we do need, though, is a kind of unrepair or damage mechanism. What needs explanation on this picture is why things sometimes go wrong when presuppositional language is used.⁴⁰ The suggestion is that “John doesn’t regret breaking the window” puts two pieces of information on the scoreboard. But it doesn’t put them on in the same way. The further suggestion is that the informational/ common-ground component of the scoreboard contains two halves.

⁴⁰ It would be nice also to have an account for why presuppositions feel presuppositional, rather than assertoric. This is just to say that the non-Cognitivist story so far makes no differentiation between what is asserted and what is presupposed.

28   There is the foregrounded information on the scoreboard, and there is the backgrounded information on the scoreboard. For content to be marked as presuppositional is just for the dynamics of that content to demand updating of the backgrounded information, rather than the foregrounded information. So the more sophisticated picture is that the dynamics of “John doesn’t regret breaking the window” put in the background component that John broke the window and in the foreground component that John doesn’t regret. Two remaining questions, then: (i) what’s the difference between foregrounded and backgrounded information, and (ii) what happens when a presuppositional utterance wants to place its presupposition in the backgrounded information, but conversational participants are unhappy with the unsatisfied presupposition?⁴¹ The challenge to the non-Cognitivist is that the distinction between foregrounded and backgrounded information can’t be psychologized. We can’t, for example, say that the foregrounded information is what we are actively attending to and the backgrounded information is what we are tacitly taking for granted. But the advantage of being posed this challenge is that it leads, I think, to a more satisfactory account of the foreground/background distinction. We should look for a functional/ operational characterization of this distinction. As a first pass (probably something more sophisticated is called for, but this is enough to see the flavor of the approach), suppose that foregrounded information is that information in common ground that is open to propositional anaphora, and backgrounded information is that which is not available to propositional anaphora. So an utterance of “That’s ridiculous” updates the scoreboard in part by attaching the propositional anaphor “that” to some pre-existing information on the scoreboard—and the operational suggestion is that the scoreboard divides into attachment-available and attachment-unavailable information. This gives us a characterization of foregrounded and backgrounded scoreboard components that’s wholly non-psychological. But why, then, should conversational participants sometimes balk at updating ⁴¹ Note that on this picture an “unsatisfied presupposition” is equivalent to an informative assertion—it’s something seeking a real change in the scoreboard.

  

29

the backgrounded information? (Which is, on this picture, what happens when a presupposition dynamically demands such an update, the update is substantive, but we “refuse to accommodate”.) Here the explanation needs to be psychological/social, but that’s fine, because the phenomenon—that the conversational participants are unwilling to continue tracking the scoreboard as updated by the language—is also psychological/social. A rough proposal: we in general don’t like new information being added to the backgrounded component, because the (definitional) unavailability of backgrounded information for propositional anaphora makes it cumbersome to challenge and correct that information. (Rejection and refutation language such as “that’s false” or “what’s your evidence for that” generally relies on propositional anaphora.) So when we suspect we might want to challenge, we’d rather have the information foregrounded for dialogical convenience, and we will balk at the proposed update.⁴² Final small side benefit: I noted earlier that there’s no analog to presupposition for the to-do list.⁴³ On the current picture, that’s just to say that there is no linguistic mechanism for updating the backgrounded component of the to-do list. A possible explanation: there is no backgrounded component of the to-do list, because the to-do list, unlike the common ground, doesn’t bifurcate into foregrounded and backgrounded components. And again on the current picture, that’s because the to-do list doesn’t bifurcate into those goals/actions that are available for propositional (actional?) anaphora and those that are not. So a further possible explanation of that: we don’t in fact have a rich anaphoric ⁴² What does this talk of balking come to? The utterance really does shift the prior scoreboard S₁ to a posterior scoreboard S₂ on which the backgrounded information has been updated. But then we carry on the conversation as if we were proceeding from S₁ rather than S₂. How should we think about this divergence between what the scoreboard really is and what we are acting as if the scoreboard is? There are other instances of this general problem. Suppose a man with his pants on fire runs through the room. We say “I hope he finds a fire extinguisher.” Here we are acting as if the scoreboard contained the man in its salience ranking governing deictic pronoun use. But no bit of language has updated the scoreboard to have that salience ranking. One option is to say that our subsequent discourse happens under a fictionalist tinge— we act as if the scoreboard is one thing, even though it is another. For many purposes, the acting as if is enough. Another option is to say that although linguistic tools are the primary tools for changing the scoreboard, they are not the only tools. Sometimes a salient event (a man with burning pants) serves as a kind of utterance; perhaps sometimes a prolonged acting-as-if is another means to change what the scoreboard actually is. ⁴³ See n. 3. Read the footnotes! All the best stuff is in them.

30   structure for imperatives. Some anaphora is available here: “Put all the dishes in a single stack!”, “That’s impossible.” But it looks like these incidents of anaphora can be treated as propositional anaphora linked to states of the world made salient by the actions on the to-do list, rather than to the actions on the to-do list themselves. The lack of anaphoric structure for imperatives might in turn be linked to a general insularity of imperatives from the general discourse structure—imperatives are oneoff, goal-oriented instructions that aren’t expected to be tightly connected to the flow of the conversation. That’s why, for example, we can’t in general extract relevance implicatures from an utterance of “Pass the salt!” at dinner, no matter what the current topic is. That’s a very sketchy story, and it’s just one sketch of many that needs to be made and then filled out if the non-Cognitivist program is going to be viable. But hopefully it’s enough to give a sense for what the resulting depsychologized semantic theory might look like (and maybe, depending on your theoretical preferences, to illustrate what might be attractive about proceeding in this way).⁴⁴

References Bacon, Andrew, and Cian Dorr. ‘Classicism’. In Peter Fritz and Nicholas K. Jones, eds, Higher-Order Metaphysics. Oxford University Press, forthcoming. Buchanan Ray, and Henry Ian Schiller. ‘Pragmatic particularism’. Philosophy and Phenomenological Research, 105/1, doi: 10.1111/phpr.12801, 2022. Dever, Josh. ‘Formal semantics’. In Manuel Garcia-Carpintero and Max Kölbel, eds, The Continuum Companion to the Philosophy of Language. Continuum, 2012. Dever, Josh. ‘The revenge of the semantics-pragmatics distinction’. Philosophical Perspectives, 27(1): 104–44, 2013. Dorr, Cian. ‘To be f is to be g’. Philosophical Perspectives, 30(1): 39–134, 2016. ⁴⁴ Thanks to all those who have patiently sat through one of my Language is a beautiful tool for information structure manipulation that sadly occasionally falls into the grubby hands of language users rants over the years. Your many bits of therapeutic and argumentative guidance are much appreciated; I’m sorry to have followed them so little.

  

31

Dyson, Freeman. ‘Two revolutions in astronomy’. Proceedings of the American Philosophical Society, 140: 1–9, 1996. Fine, Kit. ‘A guide to ground’. In Fabrice Correia and Benjamin Schnieder, eds, Metaphysical Grounding. Cambridge University Press, 2012. Groenendijk, Jeroen, and Martin Stokhof. ‘Dynamic predicate logic’. Linguistics and Philosophy, 14: 39–100, 1991. Hansson, Sven Ove. A Textbook of Belief Dynamics: Theory Change and Database Updating. Springer, 1999. Hawthorne, John, and Jason Stanley. ‘Knowledge and action’. Journal of Philosophy, 105: 571–90, 2008. Karttunen, Lauri. ‘Presupposition and linguistic context’. Theoretical Linguistics, 1: 181–94, 1974. Kratzer, Angelika. ‘The notional category of modality’. In Paul Portner and Barbara Partee, eds, Formal Semantics: The Essential Readings. Blackwell, 1981. Lamport, Leslie, Robert Shostak, and Marshall Pease. ‘The Byzantine generals problem’. ACM Transactions on Programming Languages and Systems, 4(3): 382–401, 1982. Lederman, Harvey. Uncommon Knowledge. PhD thesis, Oxford University, 2014. Lederman, Harvey. ‘Uncommon knowledge’. Mind, 127: 1069–105, doi: 10.1093/mind/fzw072, 2018. Lewis, David. Convention: A Philosophical Study. Harvard University Press, 1969. Lewis, David. ‘Languages and language’. In Minnesota Studies in the Philosophy of Science, Vol. 7, 3–35. University of Minnesota Press, 1975. Lewis, David. ‘Scorekeeping in a language game’. Journal of Philosophical Logic, 8: 339–59, 1979. Rothschild, Daniel, and Seth Yalcin. ‘On the dynamics of conversation’. Noûs, 51/1: 24–48, 2015. Rothschild, Daniel, and Seth Yalcin. ‘Three notions of dynamicness in language’. Linguistics and Philosophy, 39: 333–5, 2016. Stalnaker, Robert. ‘Pragmatic presuppositions.’ In Milton K. Munitz and Peter Unger, eds, Semantics and Philosophy. New York University Press, 1974.

32   Stalnaker, Robert. ‘On the representation of context’. Journal of Logic, Language, and Information, 7: 3–19, 1998. Stalnaker, Robert. ‘Common ground’. Linguistics and Philosophy, 25: 701–21, 2002. Stalnaker, Robert. ‘Dynamic pragmatics, static semantics’. In Daniel Fogal, Daniel W. Harris, and Matt Moss, eds, New Work on Speech Acts. Oxford University Press, 2018. van Benthem, Johan. Essays in Logical Semantics. D. Reidel Publishing Co., 1986. van Fraassen, Bas. ‘Critical study: Brian Ellis, Rational Belief Systems’. Canadian Journal of Philosophy, 10: 497–511, 1980. Veltman, Frank. ‘Defaults in update semantics’. Journal of Philosophical Logic, 25: 221–61, 1996. Wilson, Jessica. ‘No work for a theory of grounding’. Inquiry, 57: 535–79, 2014. Yalcin Seth. ‘Dynamic semantics’. In Gillian Russell and Delia Graff Fara, eds, Routledge Companion to Philosophy of Language. Routledge, 2014.

2 Singularism vs. Descriptivism? Rachel Goodman

The distinction between singular and descriptive thought is widely, but not universally, endorsed in contemporary philosophy of mind and language. But a clear account of the distinction remains elusive, as do clear arguments in favor of the existence of singular thoughts. In fact, it is often disconcertingly unclear what, exactly, people who identify as singularists (who claim that there are singular thoughts about ordinary external objects) and people who identify as descriptivists (who claim that all thoughts about external objects are descriptive) disagree about. Here, I aim to set the stage for a clear account of singular thought by thinking about arguments in favor of the existence of singular thoughts. I aim to clarify the debate by examining the long-standing suggestion that singular thought plays an anchoring role with respect to thought in general. I also aim to move the debate forward by suggesting that some, but not all, of the putative disagreements between singularists and descriptivists are illusory once clarified, and that singularists can more productively argue for their view by separating it from some of the framework in which it is commonly packaged.

2.1 Singularism There is a popular idea that some thoughts—that is, mental states—are direct, and some are indirect. To think indirectly about a thing is to think about it merely by thinking about the properties it possesses. Thus, someone might think of me, Rachel Goodman, as the only Australian in the philosophy department at Chicago Circle, and their thought about Rachel Goodman, Singularism vs. Descriptivism? In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Rachel Goodman 2023. DOI: 10.1093/oso/9780198892724.003.0002

34   me would be mediated by a description that I satisfy and would exploit this relationship of satisfaction to single me out. To think directly about a thing is to think about it without this kind of descriptive mediation—that is, to think about it, but not merely by thinking about the properties it possesses. Thus, someone might sit across the seminar table from me and think of me as her, despite not knowing where I’m from, or falsely believing that I am English or South African. This distinction, between thinking of a thing merely in virtue of thinking of its properties and thinking of a thing directly, is the essence of the contemporary distinction between singular and descriptive thought. Despite the importance of tools from the philosophy of language for theorizing and understanding this distinction, it is meant to be one in the philosophy of mind, between kinds of mental state. Therefore, it is meant to be a distinction worth making because the two kinds are interestingly, psychologically different and, perhaps in virtue of this, play different roles in a larger theory of mind. Note that the distinction has been introduced without saying anything about the contents of mental states. However, on what I’ll call the traditional picture, we use a theoretical tool to understand the distinction: the tool of structured propositions. These are complexes of particulars, properties, logical operations, and so forth, which are said to be the contents of mental states. The properties and/or particulars contained in a proposition that is the content of a given mental state are the things that the agent of the state thinks about directly. If a thinker thinks about a thing only indirectly, or in an attenuated sense, the content of her thought describes that thing without containing it.¹ This leaves us with the question, ‘what kinds of things do we think about directly?’, or ‘what kinds of things are contained in the propositions that are the contents of our mental states?’ More specifically, ‘are there any particulars contained in the propositions that are the contents of our mental states?’, or ‘do we only think propositions that contain properties, relations, etc.?’ If we call propositions containing particulars singular propositions and propositions containing only properties and ¹ Russell (1905, 1910, 1912). A neo-Fregean version appealing to object-dependent concepts can be substituted if the reader prefers.

 .  ?

35

logical operations general propositions, the question is: ‘Do we have thoughts whose contents are singular propositions?’ ‘Secondly, if we do think thoughts whose contents are singular propositions, which kinds of particulars do they contain?’ Here, we can distinguish between what I’ll call the old and new versions of our traditional picture. The old traditional picture holds that some of our thoughts have singular propositions as their contents, but the particulars they contain aren’t ordinary external objects, like tables, chairs, human beings, and so forth. Russell (1910, 1912) famously held the old traditional picture: he thought that the particulars contained in the singular propositions that are the contents of our thoughts were sense-data, and perhaps ourselves. In contrast, the new traditional picture says both that some of our thoughts have singular propositions as their contents and also that, sometimes, the particulars they contain are ordinary external objects. The new traditional picture is the one held, with a caveat here and there, by those in the contemporary literature who identify as a singularists about thought.² How to (fairly) construe those in the contemporary literature who identify as descriptivists, and the nature of their disagreement with those who identify as singularists, is part of what’s at issue in what follows, and so will have to wait for now. At this stage, however, we can ask what has made the new traditional picture seem appealing and/or convincing? It is fair to say there is both a clear history of influences, which has popularized the new traditional picture, and also a good deal of unclarity about what, exactly, that history establishes. Historically, a central reason for the popularity of the new traditional picture is the fact that our central tool for theorizing the distinction between singular and descriptive thought—that is, the tool of structured propositions—has also been used in the philosophy of language, to theorize the semantic content of our linguistic utterances. This,

² The caveats are for singularists who instead appeal to Fregean structured contents (e.g. Evans (1982), McDowell (1984), Recanati (1993, 2012), Dickie (2015), and others) and for singularists who hold that empty thoughts can be ‘singular’ though they lack singular content (e.g. Burge (1977), Bach (1987, 2010), Jeshion (2002, 2010), Sainsbury (2020), Taylor (2010), and others). I take it singularists agree that referential (non-empty) singular thoughts have (either Russellian or Fregean) singular content.

36   combined with influential views about reference in natural language, and a simple, intuitive picture of the relationship between the contents of our utterances and the contents of the thoughts we express and ascribe with them—aided no doubt by influential externalist arguments that also make use of this picture³—suggests that many of our thoughts about external objects have singular propositions containing those objects as their contents.⁴ For example, Kripke (1970) is meant to have taught us that the content of a sentence containing a proper name is a singular proposition containing the referent of the name. Add to this the natural assumption that, when I sincerely assert a belief, this involves making an utterance whose content is the same as the content of my belief,⁵ and it certainly seems that, when I communicate a belief about my mother by saying ‘Luciana Goodman is Italian’, my belief has a singular proposition containing Luciana as a constituent of its content. Similarly, the simplest picture of attitude ascriptions says that a sentence with the form ‘S believes that P’ expresses a relation—in this case, the relation of believing—between a subject s, and the proposition expressed by the sentence ‘P’ in context. Add to this the fact that the proposition expressed by the sentence ‘P’ is, in many cases, a singular proposition containing an ordinary object, and, again, it seems that the tconditions for ordinary attitude ascriptions are, frequently, that the attributee has a belief with a singular proposition containing an ordinary object as its content. For example, if I report that ‘Luciana Goodman believes that Rachel Goodman is not Italian’, it looks like the truth condition on my utterance is that the content of Luciana’s belief is a singular proposition containing me. But, if we motivate the new traditional picture in this way, a problem arises. The very same (natural, intuitive!) assumptions about the

³ Burge (1977), Putnam (1975). ⁴ I hold that these influences are at least significantly responsible for the widespread philosophical popularity of the new traditional picture, not that these are the deepest motivations of singularists themselves. However, singularists have often allowed themselves to rely on these influences in supporting their view. ⁵ See, for example, Kripke’s (1979) Disquotation Principle, along with related principles like Recanati’s (1993) Congruence Principle, Jeshion’s (2001) Accessibility of Content, and Goodman’s (2018a) Semantic Content Accessibility.

 .  ?

37

relationship between the contents of thoughts and the contents of the utterances we use to express and ascribe them in fact strain the distinction between direct and indirect thought in the philosophy of mind. The issues here come fast and thick, but the general idea is simple enough. First, it seems very easy to introduce a proper name into the language, and so very easy to put oneself in a position to express one’s belief about an object using a name. Famously, we could introduce a name with an act of descriptive reference-fixing, stipulating that we’ll use the name ‘Julius’ to pick out the unique inventor of the zipper, if there is one. In principle, we could do this in any case in which we think about an object merely as the satisfier of a given descriptive condition.⁶ Second, note that this alone seems to imply that, for almost any given mental state about a particular object, there will be some context in which that mental state can be truly ascribed with an attitude ascription whose t-condition is that it relates the agent to a singular content. Indeed, even without exploiting the, admittedly, philosophically fraught possibility of names introduced by acts of descriptive reference-fixing, the way that ordinary attitude ascriptions appear to be governed by the diverse practical purposes of speakers and audiences seems to entail that putatively paradigm cases of descriptive thought can be truly ascribed with attitude ascriptions that relate their thinkers to singular contents.⁷ In short, the patterns of singular reference in natural language exert a kind of pressure on the new traditional picture: If, in any case in which an agent has an attitude to a general proposition that describes an object, her mental state can be expressed or ascribed with an utterance that has singular content, then it looks like we’ve made a distinction in the philosophy of mind, with no real difference behind it. And this makes the distinction between singular and descriptive thought look theoretically suspicious.⁸ The same influence of the philosophy of language on the philosophy of mind that popularized the new traditional picture is ⁶ Even if we think, like Jeshion (2009), that there are additional constraints on successful introduction of a name, this is enough to generate some form of the same challenge. ⁷ See Hawthorne and Manley (2012), ch. 2, for a clear and compelling presentation of the issue. See also Taylor (2002), Chalmers (2011), Recanati (2012), Goodman (2018b), Openshaw (2018), Schiller (2019). ⁸ Hawthorne and Manley (2012) take all this to provide an argument that there are no substantive non-semantic constraints on singular thought but still seem to allow that singular

38   also the source of a serious challenge, which singularists ought to take seriously. There are certainly moves to be made in response (and they have found their way into the literature in various forms), but the relationship of those moves to the status of the singular/descriptive thought distinction is underappreciated. In a nutshell, the moves involve arguing that we should give up on our very simple picture of the relationship between linguistic and mental content. For example, that one can express one’s belief with a name does not entail that the belief is singular; that one’s mental state can be ascribed by relating her to a singular proposition does not entail that the content of that state is singular.⁹ But, note this: to the extent that these moves are convincing, they also throw into question the support discussed above for the new traditional picture. Here, I begin to lay some of my own cards on the table. Singularists who are unperturbed by the challenge to their view from within the philosophy of language are now under pressure to clarify and motivate their view. And, this is a good thing! If they are unimpressed by ‘linguistic’ challenges to the new traditional picture, then they should ask themselves if one ought to have found support for the new traditional picture by appeal to the behavior of names and attitude ascriptions— and, indeed, by appeal to intuitions about what it is natural or felicitous to say about the contents of mental states—in the first place. That is, we should view the challenge from within the philosophy of language, and the kind of response this challenge elicits, as an invitation to clarify (for ourselves and for descriptivists) why we thought the new traditional picture was true in the first place. Hopefully, the answer will be that there are reasons internal to the philosophy of mind to posit singular thoughts about ordinary external objects. But here we face the uncomfortable fact that these reasons—in some cases, familiar ones we often take for granted, or often appeal to in passing—have not been as clear, or as carefully articulated and argued, thought is a ‘cognitive natural kind’. Goodman (2018b) argues, along the lines suggested here, that their ‘liberalism’ about singular thought is an unstable position. Bach (2014) also makes points along the same lines. ⁹ See Sullivan (2010), Geirssen (2013, 2018), Goodman (2018a) for the first claim, and Bach (1997, 2010), Taylor (2002), Goodman (2018b), Openshaw (2018) for the second.

 .  ?

39

as we should hope. It’s time to fill this lacuna, if we can. If we can’t, we may need to consider giving up the new traditional picture. Given this, it’s worth noting that there are several ways that this might go. For example, one could adopt some form of the old traditional picture. Alternatively, one could separate our initial singular/descriptive thought distinction—that is, the idea of singular vs. descriptive ways of thinking about a particular thing—from the structured content framework usually used to theorize it. Or, one could give up on that initial distinction entirely. Bearing these options in mind will help to clarify which commitments are doing which work in an overall picture, and which are dispensable, central, and so forth. It will also help to clarify the nature of the disagreement between those who identify as singularists and those who identify as descriptivists (indeed, it’s not always obvious which of these ways of denying the new traditional picture amounts to ‘descriptivism’).

2.2 The anchoring role Our question is whether there is motivation, internal to the philosophy of mind, for the new traditional picture. As we ask this question, our eye is on the commitments built into that picture, and the question of whether and how they earn their keep. One way to approach the question of motivation is to think about the direct/indirect thought distinction in the context of a theory of intentionality and ask if, and in what way, direct, or non-descriptive, thoughts have any role to play in that theory. Here, a common idea—albeit one, surprisingly, rarely spelled out at length—is that direct thoughts about external objects play an anchoring role: they anchor thought to the world around us.¹⁰ A common slogan among singularists is that singular thought is required to anchor thought to the world of particular things. And, a common claim is that Strawson provides an argument to this effect, in Individuals.¹¹ ¹⁰ See Russell (1910, 1912), Strawson (1959), Burge (1977), Evans (1982), Bach (2010), Dickie (2010). ¹¹ This claim is often contained in surveys of the notion of singular thought and of singular propositions. See, e.g., Fitch and Nelson (1997), Glick (2018), Nelson (2021).

40   More is at stake in discussing the anchoring role than the question of whether there are singular thoughts about ordinary external objects. Discussing it will also help us better understand what it means to reject this claim—that is, what, in addition to rejecting the new traditional picture, is involved in so-called descriptivism about thought. The claim of an anchoring role for singular thought is often accompanied by a nod to the following, famous passage, in which Strawson (1959) discusses the conditions for the success of purely qualitative descriptions that aim to pick out particular individuals: Now one may be very well informed about a particular sector of the universe. One may know beyond any doubt that there is only one particular thing or person in that sector which answers to a certain general description. But this, it might be argued, does not guarantee that the description applies uniquely. For there might be another particular, answering to the same description, in another sector of the universe. Even if one enlarges the description so that it incorporates a description of the salient features of the sector of the universe concerned, one still lacks a guarantee that the description individuates. For the other sector might reproduce these features too. However much one adds to the description of the sector one knows about—its internal detail and its external relations—this possibility of massive reduplication remains open. (Strawson 1959, 20)

We’re asked by Strawson to imagine the possibility of a scenario in which the precise arrangement of qualitative properties instantiated in some familiar corner of the universe—say, Chicago, IL—is reduplicated in some unfamiliar corner of the universe. The passage points out that we always lack a guarantee that any purely qualitative description picks out a unique object, because the possibility of this kind of ‘massive reduplication’ always remains open.¹²

¹² A ‘purely qualitative description’ is a description in terms of qualitative properties. Noting that there are certainly issues attached to properly defining this notion, I’ll allow myself to appeal to it, and will think of qualitative properties as general properties, where the contrast is with properties that incorporate relations to particulars, picked out singularly.

 .  ?

41

The passage and its surrounding discussion is commonly said to contain an argument for the existence of singular thought, but what is the argument?¹³ Strawson himself uses his thought experiment to lay out, and respond to, a challenge to the possibility of descriptive thought, which he seemingly equates with non-demonstrative thought, about ordinary external things. I’ll call this the Strawson Challenge. The starting point for the Strawson Challenge is the claim that it’s always epistemically possible—in the sense of consistent with what we know—that the arrangement of properties in some familiar corner of the universe could be reduplicated in some unfamiliar corner. If this were the case, our attempted qualitative descriptive identifications wouldn’t apply uniquely. Strawson thinks this means they’d fail of reference.¹⁴ If it is always epistemically possible that we live in a massive reduplication universe, we can never know, for any descriptive identification, that it actually refers. And this is a problem, according to the Strawson Challenge, because making reference to an object requires knowing that you have made reference to it. So, if the Strawson Challenge stands, whether or not massive reduplication is actual, descriptive identification is always thwarted, because the epistemic possibility of massive reduplication defeats the knowledge condition on successful reference. Strawson’s response to his own challenge is to save descriptive thought by claiming that all successful descriptive thoughts contain a demonstrative element that anchors them to particular objects around us. Strawson claims that it is possible to know that a descriptive thought succeeds of reference when it has a demonstrative element (because knowledge of reference would not, in this case, require knowledge that there is no massive reduplication). So the knowledge condition on reference can be satisfied through the presence of a demonstrative element in a(n otherwise) descriptive thought. ¹³ The argument itself is, surprisingly, fairly rarely discussed in detail, or outlined clearly. Though see Fitch and Nelson (1997), Campbell (2002, ch. 12, 249), Jeshion (2012), Dickie (2015, 69–72), Nelson (2021), and, especially, Brewer (1999, ch. 2), who discusses the argument in the most detail I’m aware of. ¹⁴ There are other options. E.g., if Russell (1905) was right about descriptions, these thoughts would be systematically false. I’ll set this disagreement aside for now, and allow for different options later.

42   So, for Strawson himself, the moral is that demonstrative thought— thought that refers to its object in virtue of the fact that one perceives that object—is fundamental in the sense that all descriptive thoughts require (in order to pick out an object) a demonstrative component to guard against the epistemic possibility of massive reduplication. The Strawson Challenge and the Strawson Response rely on two assumptions we might well reject, or at least prefer not to rely on. More importantly, I claim, they are unnecessary and don’t point to the real moral of the massive reduplication thought experiment. First, the Strawson Challenge relies on the assumption that massive reduplication is always epistemically possible. It’s not obvious this is true, and there is no argument offered to defend this claim. I won’t argue directly against this assumption here but it’s worth noting that Strawson himself seems to rely throughout his discussion on the suggestion that knowledge requires certainty, in order to make it seem plausible.¹⁵ Second, why assume, as Strawson does, that one can’t successfully think about an object without knowledge that one has succeeded? Again, no argument for this claim is found in Strawson, and I doubt it is intuitively plausible.¹⁶ Furthermore, the assumption may be seen to have theoretical costs, in that it pushes us to give up on a way of thinking about what’s distinctive and useful about purely descriptive thought. Here, I have in mind the idea that descriptive thoughts enable us to ‘extend the range of thought’ beyond the objects we know and interact with, precisely because they rely on the mechanism of property satisfaction. Armed with a stock of property concepts and the ability to generalize over objects, I can compose a descriptive condition. If the world is such that this condition is uniquely satisfied, I have thereby brought a

¹⁵ One might hold that the possibility of massive reduplication is merely skeptical. Brewer (1999) attempts to argue that massive reduplication, and therefore failure through multiple satisfaction, is a genuine epistemic possibility in a way that failure through emptiness is not, but I’m unconvinced. ¹⁶ Strawson’s view might be related to his claims about definite descriptions in Strawson (1950). But his claim there, that uses of definite descriptions presuppose existence and uniqueness, does not entail a knowledge condition. Relatedly, one might suggest it is infelicitous to use the definite article without knowledge of existence and uniqueness (perhaps one should make an existential claim instead in such cases). But, this is surely too strong. If there are conditions on felicitous use of the definite article, they’re more likely to include believing there is, e.g., a unique F, rather than knowing there is.

 .  ?

43

particular thing into the range of thought, merely by relying on the mechanism of satisfaction—of an object happening to match the condition I’ve specified. If there is more than one thing or nothing at all that satisfies the descriptive condition, then I have failed to pick out a particular thing. But success isn’t ruled out from the start by the lack of a guarantee—or even the lack of knowledge—of a unique match. The Strawson Response—that all descriptive thoughts have a demonstrative element—meets the Strawson Challenge on its own terms. This is not required (for reasons outlined next), and it effectively involves conceding that purely descriptive thought that is determinately about a particular thing is impossible. Whether or not purely descriptive thought is common, preserving its possibility makes sense for the reasons above. There is an alternative argument, however, also based on Strawson’s thought experiment, which better shows the existence of thoughts that play an anchoring role. I’ll call it the Revised Massive Reduplication Argument. Imagine a whole system of thoughts that is descriptive in the sense that all the thoughts in the system single out objects insofar as those objects satisfy qualitative descriptive conditions laid out in the contents of the thoughts. And now imagine that there is in fact massive reduplication in the universe. If this is the case, then all of the thoughts in the system either fail of reference, are indeterminate, or are systematically false.¹⁷ Now, ask yourself: if there were massive reduplication, would this make it the case that all of our thoughts about the particulars object around us would fail of reference, be indeterminate, or be systematically false? I think the right response is this: even if the actual world were one in which there were massive reduplication, at least some of our thoughts wouldn’t be affected. For at least some of our thoughts, how things are elsewhere is irrelevant to their intentional or referential success. Those thoughts would still be about the objects around us—those with which we interact causally, practically and informationally—even if there were duplicates in some far-distant corner of the universe, which we don’t

¹⁷ I’m now allowing for the possibility that descriptions are, e.g., Russellian—that is, allowing for different possible kinds of failure that would result from multiple satisfaction, depending on how descriptions work.

44   interact with. Which thoughts are these? Plausibly, the ones that are based on causal, practical, and informational connections, like perception, to the particular things they are about. So, the moral seems to be that some of our thoughts aren’t purely descriptive—plausibly, the ones that exploit causal information connections to the particular objects they are about to establish their aboutness (I’ll be more precise about the argument’s conclusion in Section 2.3). It’s worth noting two points about our Revised Massive Reduplication Argument. First, the Revised Massive Reduplication Argument does not assume either of Strawson’s controversial premises. The epistemic possibility of massive reduplication doesn’t play a role. Instead, what’s significant is simply the truth of the following counterfactuals. If there were massive reduplication, then purely descriptive thoughts would fail of reference, be indeterminate, or be systematically false. If there were massive reduplication, it is not the case that all my thoughts would fail of reference, be indeterminate, or be systematically false. The argument also doesn’t rely on the premise that, in order to think about an object, one must know that one has succeeded. It therefore makes space for the case in which you think a purely descriptive thought and succeed because there is only one object that answers to a certain description, even though you don’t have knowledge that this is the case. Second, my Revised Massive Reduplication Argument doesn’t purport to show that purely descriptive systems of thought are not possible. If this is true, we need another argument to show why.¹⁸ Our massive reduplication argument is more modest. It bottoms out in a basic commitment that if there were massive reduplication, this would leave the success of some of our thoughts undisturbed. Why accept this basic commitment? There is no knock-down argument that we should, but there are broader theoretical considerations that speak in its favor (and these are related to the intuitive appeal of the commitment). The broader context of our discussion is an overall theory of intentionality for thoughts. The case to be theorized is a creature with a well-functioning perceptual system and a range of perceptual ¹⁸ Burge (1977) sketches an argument of this kind.

 .  ?

45

interactions with her environment, upon which she forms thoughts, beliefs, etc. about that environment. In asking about the status of the basic commitment our argument relies on, we can step back and consider two possible theories of intentionality. According to the first, the reference of some thoughts—let us say, the perception-based ones at least—is undisturbed by massive reduplication. According to the second, massive reduplication implies crash, falsehood, or indeterminacy for all thoughts. But crash, falsehood, and indeterminacy are all ways for a thought to fail, and the case being theorized is that of a creature whose interactions are going well and whose systems are functioning properly. To theorize this case as involving widespread representational failure is strange at best. This way of motivating the basic commitment undergirding our argument takes as its starting point that, overall, our task in giving a theory of intentionality is to explain how it is we manage (succeed!) to think about the world around us, and it takes as a basic case of intentional success a thinking agent whose systems and interactions with things she interacts with are functioning well. This is not required, but is also not arbitrary.

2.3 The conclusion of Revised MR But, what conclusion can we draw from the Revised Massive Reduplication Argument? And, does the argument support the new traditional picture? I’ll state the argument’s conclusion like this: some thoughts have a certain representational feature (are tokens of a certain representational type), which I’ll put by saying that they are referentially anchored: A referentially anchored thought (type) has the representational property of being about the same object when tokened in a massive reduplication scenario, as when tokened in a scenario with no massive reduplication. Or, allowing, for example, that we might be Russellian about descriptions: A referentially anchored thought (type) has the representational property of being such that, if it would be true if tokened in a scenario

46   without massive reduplication, then it would also be true if tokened in a reduplicative scenario; and if it would be false if tokened in a scenario without massive reduplication, then it would be false and false for the same reason if tokened in a reduplicative scenario. However, this does not establish the new traditional picture, for familiar enough reasons. There could be referentially anchored thoughts that aren’t attitudes to singular propositions that contain ordinary objects. For example, thoughts about a red cube perceived by a thinker, which have contents that might be naturally expressed with (1)–(3), are all referentially anchored. But only (1) is naturally taken to express a singular content containing an ordinary external object.¹⁹ 1) This (a particular red cube) is red. 2) The cube I (a particular thinker) am perceiving is red. 3) The cube that is the cause of this (a token mental state) is red. My thought to the effect that the cube I am perceiving is red would be about the particular red cube I were perceiving (if there were one), even if there were also a qualitatively identical cube, in a qualitatively identical segment of the universe, which I were not perceiving. Seemingly, this thought would be referentially anchored because it had a content that is singular with respect to me, not the cube I perceived. So, the Revised Massive Reduplication Argument leaves it open that all our referentially anchored thoughts have contents like those naturally expressed by (2) and (3). If so, the new traditional picture is false. And indeed, those who identify as descriptivists often respond to the claim that there are singular thoughts about ordinary external objects by glossing their opposing conception of the contents of putatively singular thoughts about external objects—say, perception-based thoughts about ordinary external objects—with formulations like (2) or (3).²⁰

¹⁹ This point applies to Strawson’s version of the argument too. ²⁰ See Searle (1983), Kroon (1987), Jackson (2000, 2010, 2017), and the discussion in Chalmers (1995).

 .  ?

47

Thus, in showing that there are referentially anchored thoughts, what the Revised Massive Reduplication Argument really shows is that there is some singularity of a certain kind in thought. Firstly, the kind of singularity at stake is what I call anchoring to particulars, and may be thought of as singularity with respect to reference determination. That is, the point established by the massive reduplication argument doesn’t concern rigidity, or what are sometimes call ‘object-involving t-conditions’, but rather the failure of qualitativism about mental reference determination.²¹ In a massive reduplication scenario, for any purely qualitative description ‘the F’, there will be two actual Fs, so such descriptions will fail of reference, be indeterminate, or be systematically false. But, in a massive reduplication scenario, our thoughts would not universally fail of reference, be indeterminate, or be systematically false. This shows that a purely qualitative theory of what makes our thoughts about particular objects is false. Secondly, the argument merely shows that there is some singularity or other of this kind because, for example, what’s sometimes called ‘causal descriptivism’²² allows for non-qualitative reference determination, even though it denies singularity with respect to ordinary external objects, and instead posits singularity with respect to a privileged kind of object like an agent or a token mental state. And, indeed, whether or not they gloss their content claims in the ways represented by (2) and (3), contemporary descriptivists are what I will call anchored descriptivists, rather than pure descriptivists.²³ Their positions concerning the representational contents of mental states make space for referentially anchored thoughts, but they resist the new traditionalist claim that some thoughts have singular propositions containing ordinary objects as their contents. Both singularists and descriptivists posit referentially anchored thoughts. The disagreements seem to arise when we reach beyond this to say something about the nature of those referentially anchored thoughts.

²¹ See Recanati (2012, ch. 2) and Sainsbury (2020) for discussion of varieties of singularity, and see Chalmers (2018) for the language of ‘qualitativism’. ²² Kroon (1987). ²³ Searle (1983), Kroon (1987), Chalmers (1995), Jackson (2010). See Jeshion (2012) for the distinction between ‘internally anchored descriptivism’ and ‘pure descriptivism’, and Recanati (2012), who calls anchored descriptivism ‘relational descriptivism’.

48   In the next section, I’ll outline the way that debate (or, in some ways, lack of debate) between new traditionalists and descriptivists tends to go. But first, I want to flag a question, so we can return to it later. The fact that descriptivist opposition to the new traditional picture is sometimes glossed with formulations like those in (2) and (3) makes it natural (for singularists at least) to construe anchored descriptivists as proponents of the old traditional picture. That is, it’s natural to take descriptivists as conceding (or being obliged to concede) the need for singular propositions as the contents of some of our thoughts, but as maintaining that they are singular with respect to some privileged class of particulars, like agents, token mental states, etc. However, this might misrepresent the nature of the disagreement (and the extent of potential agreement) between new traditionalists and anchored descriptivists. This should come as no surprise, for we have surely noticed that most ‘descriptivists’ don’t make use of structured content frameworks to start off with.²⁴ So, we can’t straightforwardly classify them as proponents of the old traditional picture. This might be a merely terminological hiccup, but we should at least wonder about the extent to which this is the case.

2.4 The over-attribution objection and misattribution response So, how does the debate between new traditionalists and anchored descriptivists go? The singularist posits thoughts with contents that are naturally expressed by (1): contents that are singular propositions containing ordinary objects. Anchored descriptivists gloss their opposing content claims about the same cases with formulations like those in (2) and (3). In response, proponents of the new traditional picture appeal to what I will call the over-attribution objection.²⁵ This is, essentially, the objection that the anchored descriptivist’s content claims are psychologically implausible, because too sophisticated. ²⁴ Searle (1983) is unusual in doing so. ²⁵ Jeshion (2012) calls this ‘the richness objection’.

 .  ?

49

There are variations on this objection to be found in the literature, but here is a representative sampling: [Searle’s causal, token reflexive] theory of satisfaction conditions does not match the facts of mental ability. (Burge 1991, 206) The problem with Searle’s apparent view is not that ‘the contextual features on which perceptual and indexical beliefs depend cannot themselves be entirely represented as part of the Intentional content’ (p. 214, my emphasis), for any possible notion of Intentional content. The problem is that the notion of Intentional content that one obtains by including such features does not correspond to our cognitive states or mental abilities. (Burge 1991, 210) The complexity involved in Searle’s account of the content of experience is unmotivated for we can adequately capture the phenomenology of experience and explain our mental abilities by postulating a content that does not have the causal component and the self-referential component that Searle suggests. (Soteriou 2000, 181) . . . the [descriptive] Fregean is likely to say that the reference-fixing description in the mind of the subject must be something like ‘what I am now seeing’ or ‘what I am now touching’ (FN: Russell would say that, in such examples, the description is ‘what is causing these sense data’). But this supposes, on the part of the subject, reflective abilities the exercise of which does not seem to be required. . . . The subject need not reflect on her perceptual relation to objects in order to have thoughts about the objects she perceives. . . . (Recanati 2012, 30) . . . 2-D Relational Descriptivism entails that acquaintance relations are always represented as part of the content of singular thoughts; but this is debatable, to say the least. Kripke and many others have argued that acquaintance relations themselves need not be represented. For example, what determines the reference of the name ‘Aristotle’ in language or thought is a communication chain leading back to Aristotle, but users of the name need not have any thought regarding the communication chain, nor do they need to have the very concept of a communication chain. (Recanati 2012, 23)

50   . . . token reflexive descriptions can only be grasped by fairly sophisticated users of the language, able to reflect upon the relations between token-representations and objects in the context in which these representations occur. Indexical thinking indeed exploits these relations, but in no way presupposes the ability to reflect on them. (Recanati 2012, 34)

Allowing for variation, the over-attribution objection can be put like this. The proposals in (2) and (3) anchor an ordinary perception-based thought to a particular red cube by claiming that the thought employs property concepts that pick out, not just the properties of red-ness or cubic-ness, but also those of being a cause, being perceived, and so forth. But, surely we don’t need to use, or perhaps even have, concepts of causation or perception to think about red cubes that we perceive? Descriptivists, it seems fair to say, are almost entirely unmoved by this criticism. They frequently respond by pointing out that their content claims don’t carry the kind of psychological commitments that would make them problematically sophisticated, or psychologically implausible. For example, depending on the particular version of the overattribution objection in question, a descriptivist might respond that the claim she makes about the content of a given thought does not entail that the descriptive content is consciously entertained, or occurrently represented, or explicitly thought, or something the thinker would be in a position to specify, etc. I’ll call this the misattribution response. Again, there are variations, but here are some examples: A standard objection is that [the descriptive content claim] is ‘too complicated or sophisticated’. . . . But this objection is based on a misunderstanding. I am not claiming that the perceiver thinks to himself in words the sentence in the brackets or even that the perceiver has any consciousness of this articulation of these conditions at all. . . . The theoretical representation is a second-order characterization of a set of first-order psychological facts. . . . The complexity is a matter of the theoretical representation and not a matter of the first order psychological facts. (Searle 1991, 228)

 .  ?

51

RDC [rigidified descriptive cluster theory of mental content] is a semantic proposal about the content of terms that purport to refer, not a [proposal about] psychological mechanism. And such terms have their contents fixed by proper interpretation . . . on the basis of our dispositions to assert sentences containing such terms of various hypothetical concrete and word-mediated scenarios, and also to directly name and describe various scenarios. (Mendola 2008, 310–11) The objection is that the kinds of properties we have mentioned are fancy ones that philosophers of language and unusually alert members of the folk might think of; they are not properties ordinary, folk speakers associate with words. . . . Sometimes it is obvious which properties are associated with a word. Perhaps the speaker tells us loud and clear. But typically the association is implicit or tacit rather than explicit. It is something we can extract in principle from speakers’ patterns of word usage, not something explicitly before their mind when they use the words. I know this way of putting things—familiar though it is—will ring some alarm bells. Some will want to say that if the association is in the mind, as the description theory says, it must be explicit. They think of appeal to the implicit or tacit in this context as a kind of cheat—a way of saying something and then taking it back. (Jackson 1998, 210–11)²⁶ Sometimes philosophers are suspicious of entities such as primary intensions because they see them as reminiscent of a “description” theory of reference. But descriptions play no essential part in this framework; I use them merely to flesh out some of the character of the relevant functions from possible worlds to extensions. It is the function itself, rather than any summarizing description, that is truly central. This picture is quite compatible with the “causal” theory of reference: we need simply note that the primary intension of a concept such as “water” may require an appropriate causal connection between the referent and the subject. Indeed, we are led to believe in a causal theory of reference in the first place precisely by considering various ways the actual world might turn out, and noting what the referent of ²⁶ Jackson goes on to say more about ‘implicit association’ (see also Jackson 2017), but more on this below. Also, Jackson is talking here about linguistic reference, but he holds the same view about mental reference (personal communication).

52   the concept would turn out to be in those cases; that is, by evaluating the primary intension of a concept at those worlds. (Chalmers 1996, 58–9)²⁷

Allowing for variation across different versions, the general sentiment is that the descriptivist’s position is being misunderstood by those who make the over-attribution objection, because her content claims are not ‘psychological’ in whatever problematic way the new traditionalist is assuming them to be. Care is required to fairly represent what’s being denied. For example, Searle, in his remarks above, comes close to denying that he is making claims about psychological facts at all.²⁸ But this can’t be right by anyone’s lights, since everyone agrees that there’s an important sense in which content claims about mental states are psychological claims. A much clearer version of the response is found in Jackson and Chalmers. They certainly (and rightly) claim to be making psychological claims, but ones of a particular kind. Their content claims are said to encode the referential dependency patterns for an agent’s thoughts, which are meant to reveal certain sorts of psychological facts about the agent.²⁹ That is, their descriptive content claims are meant to capture the conditions under which an agent’s mental state (type) would refer to this or that object. Such contents encode how reference (for concepts) and truth (for whole thoughts) depend on the presence of a certain arrangement of properties in the (actual) world. Thus, for example, in the case of a perception-based concept whose content we might gloss as ‘the cube I am perceiving’, the idea is that the content encodes the fact that the thought is about whichever object has the property of being the cube I perceive. On this view, if you’ve given a (correct) specification of how the world would have to be for a thought to be true (or of how an object would

²⁷ Chalmers is responding to a concern that is slightly different (but related) to the overattribution objection here. It’s useful to include his remarks, however, since (along with Jackson’s) they articulate what I take to be the spirit of the response that interests me. ²⁸ His claim that content claims are not ‘first-order psychological facts’ is genuinely unclear to me. ²⁹ The question of what sorts of psychological facts they do and don’t actually reveal is addressed in Sections 2.5 and 2.6.

 .  ?

53

have to be for a concept to be about it), then you’ve given a specification of its content. So, the response to the charge of psychological implausibility or ‘overattribution’ is something like the following. The descriptivist’s content claims are not about whichever psychological features would make them beholden to something other than reflection (either by theorists or thinkers themselves) about how the world would have to be for the whole thought in question to be true, or how an object would have to be in order for the concept in question to refer to it. That is, their content claims are not about whichever psychological features would make them beholden to certain tests that (descriptivists think) singularists are illegitimately applying: for example, ‘Is this description something the agent consciously entertains?’, ‘. . . occurrently represents?’, ‘. . . explicitly thinks?’, ‘. . . would be in a position to specify?’, etc. My account of the misattribution response is imperfect (later sections of the chapter are partially about clarifying it), but I hope we can see that it is on the right track by noting that there is an alternative kind of response to the over-attribution objection that descriptivists could go in for, but don’t. They could respond by defending the claim that what’s being consciously entertained, or occurrently represented, or explicitly thought, etc. is captured by (2) or (3). The fact that this is not the usual response, I will suggest, should be taken as a clue as to what commitments are central to the anchored descriptivist’s view.

2.5 Anchored characters and representation* It is common for the debate between those who identify as singularists and those who identify as descriptivists to stall with some version of the misattribution response to the over-attribution objection. Singularists hold to the charge of psychological implausibility, descriptivists take the charge to underestimate their view, and the result is an impasse (and mutual irritation). I’ll aim to make progress, in this section, by illustrating two points. Firstly, something has gone wrong when singularists make the overattribution objection, at least against the best versions of anchored

54   descriptivism. There is some legitimacy to the charge of misattribution. But, secondly, in acknowledging this, we should also see that there is a question about representation that singularists are interested in, which descriptivists (by their own lights) don’t take a stand on. In the next section, I’ll claim that, given what descriptivists say about their own content claims, it’s unclear that their arguments target (what should be) the singularist’s central claim. I begin with something of a diagnosis of what has gone wrong when singularists make the over-attribution objection. In a nutshell, singularists make the charge of over-attribution because they view anchored descriptivists as, essentially, proponents of the old traditional picture. They thereby work under an assumption that the kind of fineness of grain involved in there being referentially anchored thoughts involves structured content, and then work with a common set of assumptions, according to which structured content attributions involve commitments concerning a certain sort of psychological question (which I’ll say more about below).³⁰ But, it will be a point familiar to descriptivists that accommodating referential anchoring does not require attributions of structured content. Referential anchoring involves there being tokens of representational types that would be about the same objects if tokened in a world with massive reduplication, as if tokened in a world without it. This means that referential anchoring can be captured by positing representational types characterized in terms of what objects they would be about if tokened in this or that scenario. And this can be done, without appeal to structured content, in a function-theoretic framework. One version of this approach applies something like Kaplan’s notion of character to mental states.³¹ If we take characters for mental states

³⁰ My formulation of this point is imperfect as it stands, since it might imply that singularists don’t realize that two-dimensionalist frameworks are available, which don’t employ structured content (and, of course, they do know this). The assumption is rather that the fineness of grain required for referential anchoring means that anchored descriptivist content attributions carry the sorts of commitments that would come with their attribution of a content that could be glossed with (2) or (3). ³¹ I won’t discuss the different versions of the approach, since they’re well known and the point I wish to make applies in virtue of the general approach (but see, e.g., Chalmers 1995, 1996; Jackson 2000, 2010.

 .  ?

55

to be functions from scenarios in which the state-type is tokened to the reference they would have if tokened in that scenario, then we can state the conclusion of the Revised Massive Reduplication Argument as one about the nature of the characters of some mental states.³² Namely, those characters are such that the state-types in question have the same referents when tokened in a world with massive reduplication as in a world without it. As I’ll put it, the conclusion is that some of our mental states have anchored characters. For example, we might say that the anchored character of a perceptual demonstrative thought is the function that takes us from a world considered as actual to the object perceptually attended by the agent of the thought at that world (or, from a world considered as actual to the object that is the perceptual cause of the subject’s token perceptual experience at that world, etc.). But, what commitments are, and are not, involved in the attribution of a certain anchored character? To assign an anchored character to a mental state is to claim that it is a token of a type characterized by a certain referential dependency pattern. For example, in our perceptual demonstrative example, it is to claim that, if the actual world is such that an object a is the object perceptually attended by the agent of the thought, then that thought is about a; if the actual world is such that an object b is the object perceptually attended by the agent of the thought, then the thought is about b, and so forth. But what an assignment of a particular anchored character does not do in itself is take a stance on how (psychologically, in a sense to be spelled out next) a thought with this particular referential dependency pattern is achieved in the thoughts of creatures like us. And my suggestion is that singularists are (or perhaps should be) interested in precisely this sort of psychological question. To see what sort of psychological question this is, and then to see why anchored character attributions don’t take a stance on it, we can look to the contrast between the sentences (S1) and (S2): ³² Of course, for Kaplan himself, character is a function from context to content, which is itself a function from circumstance to extension. But since, for rigid terms, character does determine extension, I’ll allow myself to talk, in describing my notion of character for a mental state, of a function from scenarios to reference. In any case, other variations of the approach (e.g. Chalmers 1995) work in the way mine does, in this respect.

56   (S1) I am tired. (S2) The agent identical to me is tired. (S1) and (S2) have the same anchored character but, I take it, there is a legitimate and familiar notion of representation, according to which they are representationally different. This is that they achieve their shared anchored character by different compositional means. (S2), but not (S1), achieves its anchored character in a way that involves representation of the relation of identity—that is, it involves a compositional component, or step, whose role it is to represent identity. It’s plausible that an analogous notion of representation can be applied to mental states. If we wish to mark the fact that it is not the only legitimate notion of ‘representation’, we can do this by dubbing it representation*. The kinds of considerations that illustrate that there is indeed a legitimate notion of representation* for thought are familiar arguments for the conclusion that thought is structured, in a particular sense. There are different ways to spell out the relevant notion of structure, which the singularist may choose between. One is in vehicular terms. According to a representational theory of mind, whole thoughts—states that have anchored characters—involve vehicles of representation, and these vehicles are themselves composed of redeployable sub-vehicle types, whose role is to isolate and represent properties, relations (and maybe objects!) in particular, repeatable ways. Another is found in Evans (1982), and involves the claim that any whole thought involves the use of distinct and redeployable abilities—call them ‘concepts’, if you want— whose role is, again, to isolate and represent (in particular, repeatable ways) the objects, properties, relations, etc. that the whole thoughts they are employed to entertain are about.³³ I won’t rehearse the arguments in favor of the existence of structure in thought, since they are, again, well known.³⁴ However, the general idea behind them is that we can best ³³ See Rescorla (2020) for the suggestion that the first way of understanding this idea is more fundamentally understood in terms of the second. And see Camp (2009) for discussion of the relation of different ways of spelling out structure in thought. ³⁴ My claim is, of course, not that these arguments are indisputable or uncontroversial. On my presentation, rejection of the claim that thought is structured in the relevant sense (either in terms of vehicles or abilities) would be reason to reject singularism. See, e.g., Evans (1982) for an illustration that this is not an arbitrary connection.

 .  ?

57

explain, for example, the productivity of our thinking abilities, and the way these abilities pattern in systematic ways, by positing whole thoughts with compositional structure, which make use of redeployable vehicle types, and/or redeployable representational abilities.³⁵ However, if thought is structured in this sense, then there is a question about the compositional structure for a mental state, which anchored character ascriptions don’t take a stance on. Thoughts structured like (S1) and (S2) have different compositional structure, but they have the same anchored character. A thought with the structure of (S2) represents* the relation of identity, whereas a thought with the compositional structure of (S1) does not.

2.6 Implications for the new traditional picture and anchored descriptivism The next point I wish to make is that being clear about the nature of anchored character attributions and representation* has implications for the way that anchored descriptivists sometimes argue for their view (and, indeed, for their opposition to the new traditional picture). Once we have introduced the notion of representation*, we have also introduced a contrast between two different roles that a property can play in a theory of reference for mental states. On the one hand, a property can play what is certainly an explanatory role with respect to the reference of a mental state, in the sense that reference supervenes or depends on possession of that property. Thus, possession of that property earns a legitimate place in our theory of reference. On the other hand, the property can be represented*.³⁶ However, the former does not entail the latter. For example, if my thought is an I-thought, then being the referent of that thought indeed supervenes on being the thing that’s

³⁵ Fodor (1975, 1987), Evans (1981, 1982), Fodor and Pylyshyn (1988). ³⁶ A related distinction, employed by diverse thinkers in different ways is between a property playing a causal vs. a computational role (Pylyshyn 2006). And see Campbell (2002, 2013) for discussion of the idea that it is often important to distinguish between different roles properties can play with respect to reference or experience.

58   identical to me. But this does not entail that the compositional structure of my thought is correctly represented by (S2). This contrast allows me to further clarify my diagnosis of what goes wrong when singularists make the over-attribution objection against anchored descriptivism. New traditionalists make this objection because they are interested in a question about the compositional structure of mental states and take the anchored descriptivist’s content claims to take a stance on this. They think some of our thoughts are referentially anchored, and they are interested in how a thought with a certain anchored character is psychologically, compositionally achieved for creatures like us (what abilities are employed, or what vehicular structure is in play?). That is, they are interested in which properties, relations, or indeed objects are represented* in a mental state with a certain anchored character. They may be wrong to lodge the claim of implausible ascriptions against certain anchored descriptivists, but their question itself seems to be a legitimate and important one. Thus, the contrast also allows me to illustrate the sense in which some anchored descriptivists’ claims and argumentative strategies simply do not touch singularists who are clear about the contrast. More sympathetically put, given the way descriptivists sometimes argue for their views about content, they are more charitably interpreted as simply not taking a stance about representation* (about the question of how anchored characters are achieved in the thoughts of creatures like us). The most compelling descriptivists sometimes claim that the very fact that we can make assessments about the referential dependency patterns of thoughts by describing the distribution of properties that make for reference is evidence that the relevant properties are part of the mechanism of reference, and thereby appropriate to mention when glossing the content of those thoughts.³⁷ A related way of arguing for descriptivism is as follows. If we start with the idea that representational features are dependent features, and note that nobody should want a magical theory of reference, then we should acknowledge that, in any given case where we have a mental state that is about an object, a, rather than an object, b, then there will be some property that a possesses and b lacks ³⁷ E.g., Chalmers (1995), Jackson (1998, 2000, 2003).

 .  ?

59

that explains this fact. The property might be a relational one (like being the object perceived by the thought’s thinker) but this is nonetheless a property possessed by the object in virtue of which the thought is about that object.³⁸ For what it’s worth, I grant this last point (I don’t favor the view that the mind tracks haecceities). But, if arguments like these are given, then what they show is that the properties in question play the first role, not the second. They show, perhaps, that the properties in question have a legitimate role in a theory of reference in the sense that they make a systematic difference to which object is referred to, or that reference supervenes on possession of them. They don’t show that the properties in question are represented*. To sum up, what’s gone wrong with the over-attribution objection is that it can legitimately be claimed to misunderstand the content claims being made by sophisticated anchored descriptivists, even if they are sometimes glossed in terms like those in (2) and (3). But, I’ve also tried to show that this response—that is, that anchored character assignments (or content attributions of similar form) take a stance on referential dependency patterns, which reveal properties that reference depends on—itself comes with the implication that these content attributions do not in themselves serve to oppose singularist claims about how particular anchored characters are compositionally achieved for creatures like us. The singularist has some non-obscure reasons to think that thought is structured, and she is, on my proposed version, committed to some claims about the particular structures of certain thoughts with anchored characters. There is indeed a good question to be asked about how to justify those particular claims but, equally, the descriptivist’s content claims and forms of argument don’t defeat them, or even target them. At the end of Section 2.1, we noted different possible ways of rejecting the new traditional picture. The claims I have now outlined suggest a path for singularists, whereby they might in principle separate their initial distinction between singular and descriptive thought from the structured content framework with which it is usually packaged. Their

³⁸ Jackson (2000, 2003).

60   claim is about the structure of certain thoughts with anchored characters (it is about the nature of the abilities or vehicles at use in such thoughts). Whether there are ‘singular thoughts’ about ordinary external objects depends on how the range of referentially anchored thoughts in creatures like us achieves their referential anchoring. It depends on what properties, relations, and/or objects are represented* in those thoughts.³⁹ To be clear, I am not claiming that singularists are thereby obliged to give up on structured content, or even that unstructured content views are preferable. But I do hold that our claims about the structured contents of mental states are essentially used to captured fine-grained representational* features of those mental states.⁴⁰ I also hold that separating the central singularist view (that there are singular thoughts about ordinary, external objects) from claims about singular propositions has the benefit of clarifying the question that we are interested in, and of avoiding impasses based on equivocations on, or intractable debates about, ‘the content’ of certain mental states.⁴¹

2.7 Conclusion: agreements and disagreements I’ve suggested here that singularists might more productively argue for their view by being clear about the question that concerns them, and I’ve also suggested that this is a question about representation*. Furthermore, I’ve suggested that taking descriptivists’ arguments and

³⁹ I hope it’s therefore clear that my suggested view is not simply a variation on standard twodimensionalism. The question I say singularists are interested in is not captured with secondary intensions or content in Kaplan’s sense. Our discussion of the conclusion of the Revised Massive Reduplication Argument in Section 2.3 illustrated this. On my view, they’re interested in how anchored characters are representationally* achieved, not in patterns of reference at worlds considered as counterfactual. ⁴⁰ And I dispute that claims about vehicles/abilities are as smoothly distinguished from claims about content as some have held (e.g. Stalnaker 1984). I see the position in Rescorla (2020) as potentially aligned with my suggestions. ⁴¹ It also avoids the possible implication (which descriptivists sometimes unfairly complain about, but which singularists sometimes illegitimately lean on) that the idea of an object ‘in’ a proposition in itself explains anything. This is not to say that objects can’t be constituents of propositions, just that this is a way of modeling something (to be explained), not an explanation of anything.

 .  ?

61

responses to the over-attribution objection seriously implies that they don’t (legitimately) take a stance on this question. This may in turn suggest that I hold there to be no real disagreements between so-called ‘singularists’ and ‘descriptivists’. This is not the case, nor do I think the discussion here implies it in the end. I do think, however, that the disagreements are often mislocated or obscured, and that my suggestions here point out one way in which this is the case. I’ll end by taking stock of some agreements and disagreements. Firstly, I take my discussion of the over-attribution objection, the misattribution response, and the contrast between roles a property can play in a theory of reference to illustrate that we (new traditionalists, and anchored descriptivists) agree (or ought to, in any case) about the way that reference for anchored thoughts is determined, in the sense that we agree about what it supervenes on. For example, we agree that the class of thoughts about external objects for which massive reduplication wouldn’t make a difference is the class of thoughts for which possession by an object of some relational property (like being the object perceived) determines reference. That is, we agree that, when one has a perception-based thought about an object, a, what makes it the case that one’s thought is about a (rather than about b, or nothing at all) is the fact that a is the object one perceives. Thus, we should be careful about the sense in which we take the debate between singularists and descriptivists to be a disagreement about reference-determination (that is, whether it goes by way of properties or not). Singularism of the kind I’m proposing claims that some of our anchored thoughts (those that are singular with respect to ordinary external objects) achieve their anchoring by employing object-directed abilities (or vehicles), but they don’t thereby deny that relational properties (like being the object perceived) have a role in reference-determination. Secondly, it is open to us to agree about particular anchored character assignments. However, this implies something more controversial, which only some singularists will go in for. If we agree about particular anchored character assignments, then we agree, not only that one’s perception-based thought is about an object, a, if it is, because a is the object one is perceiving, but also that a mental state of the same type—that is, with the same character—would have been about a distinct object, b, had the actual world been one in which b were the object one perceived. Thus,

62   agreeing with anchored descriptivists about particular anchored character assignments will entail agreement that there is a legitimate content-typing such that two agents, both thinking perceptual demonstrative thoughts but about different particular objects, could be thinking thoughts of the same type. There are certainly singularists who care a lot about object-dependent thoughts, who would dispute that this kind of typing for mental states is a content-typing.⁴² However, singularists don’t need to take this stance, and many don’t. And, in fact, to allow that there is a legitimate content-typing that goes by way of object-independent anchored characters does not imply there are no purposes for which we will want to type our mental states in object-dependent terms. Whether representation* motivates this sort of typing is itself a question we can ask, for example. But, even if we agree about all of this, the way I’ve framed things does leave us with a sticking point. Our two camps start out working with different notions of ‘representation’—one concerned with the dependence or supervenience of referential facts, and one concerned with representation*. I’ve suggested that these notions can coexist, but a conflict seems to arise insofar as both camps put their preferred notion to work in solving Frege’s Puzzle. It’s at the heart of the view that thought is structured (either by redeployable vehicles, or abilities) that difference of type, in this respect, accounts for the existence of Frege cases, and sameness of type accounts for the existence of coordination or de jure coreference in thought. Similarly, those who posit character-like content for mental states (primary intensions, a-intensions, etc.) take this dimension of content to account for the same phenomena—those of cognitive significance and rational relations between mental states. Thus, our two camps assign the same role to two different kinds of sameness and difference. It is perhaps unsurprising that those who identify as singularists and those who identify as descriptivists about thought disagree about the solution to Frege’s Puzzle, but I hope the way I have framed (one aspect of) the debate between them allows us to focus on this disagreement and its significance in a new way.⁴³ ⁴² Evans (1982), McDowell (1984). ⁴³ Thanks to Dominic Alford-Duguid, Jason Bridges, Jennifer Carr, James Conant, Imogen Dickie, Aidan Gray, Riki Heck, Frank Jackson, Robin Jeshion, Michael Kremer, Mike Martin, James Openshaw, Mark Sainsbury, Alexander Sandgren, Josef Stern, Pär Sundström, and Robbie Williams for questions or discussions of the paper or related material, and to participants of the Particularity and Generality in Mind workshop at Umea University, members of the Nature of Representation project at the University of Leeds, and audiences at the University of Chicago, the University of Illinois at Chicago, and Oxford.

 .  ?

63

References Bach, K. (1987) Thought and Reference. Oxford: Oxford University Press. Bach, K. (1997) ‘Do Belief Reports Report Beliefs?’, Pacific Philosophical Quarterly 78/3: 215–41. Bach, K. (2010) ‘Getting a Thing into a Thought’, in R. Jeshion (ed.), New Essays on Singular Thought. Oxford: Oxford University Press, 39–63. Bach, K. (2014) ‘Consulting The Reference Book’, Mind & Language 29/4: 455–74. Brewer, B. (1999) Perception and Reason. Oxford: Oxford University Press. Burge, T. (1977) ‘Belief De Re’, Journal of Philosophy 74/6: 338–62. Burge, T. (1991) ‘Vision and Intentional Content’, in E. Lepore and R. Van Gulick (eds.), John Searle and His Critics. Hoboken, NJ: Wiley-Blackwell, 195–213. Camp, E. (2009) ‘Putting Thoughts to Work: Concepts, Systematicity, and Stimulus-Independence’, Philosophy and Phenomenological Research 78/2: 275–311. Campbell, J. (2002) Reference and Consciousness. Oxford: Oxford University Press. Campbell, J. (2013) ‘Susanna Siegel’s The Contents of Visual Experience’, Philosophical Studies 163/3: 819–26. Chalmers, D. J. (1995) ‘The Components of Content’, unpublished manuscript. Chalmers, D. J. (1996) The Conscious Mind: In Search of a Fundamental Theory. Oxford: Oxford University Press. Chalmers, D. J. (2011) ‘Propositions and Attitude Ascriptions: A Fregean Account’, Noûs 45/5: 595–639. Chalmers, D. J. (2018) ‘Narrow Content, by Juhani Yli-Vakkuri and John Hawthorne’, Notre Dame Philosophical Reviews, https://ndpr.nd.edu/ reviews/narrow-content/. Dickie, I. (2010) ‘We Are Acquainted with Ordinary Things’, in R. Jeshion (ed.), New Essays on Singular Thought. Oxford: Oxford University Press, 213–45. Dickie, I. (2015) Fixing Reference. Oxford: Oxford University Press. Evans, G. (1981) ‘Semantic Theory and Tacit Knowledge’, in S. Holtzman and C. Leich (eds.), Wittgenstein: To Follow a Rule. London: Routledge & Kegan Paul, 118–37.

64   Evans, G. (1982) The Varieties of Reference, ed. John McDowell. Oxford: Clarendon Press. Fitch, G., and Nelson, M. (1997/2013) ‘Singular Propositions’, Stanford Encyclopedia of Philosophy, https://plato.stanford.edu/entries/propositionssingular/. Fodor, J. A. (1975) The Language of Thought. New York: Thomas Y. Crowell. Fodor, J. A. (1987) Psychosemantics. Cambridge, MA: MIT Press. Fodor, J. A., and Pylyshyn, Z. W. (1988) ‘Connectionism and Cognitive Architecture’, Cognition 28/1–2: 3–71. Glick, E. (2018) ‘What Is a Singular Proposition?’, Mind 127/508: 1027–67. Goodman, R. (2018a) ‘On the Supposed Connection between Proper Names and Singular Thought’, Synthese 195/1: 197–223. Goodman, R. (2018b) ‘Do Acquaintance Theorists Have an Attitude Problem?’, Australasian Journal of Philosophy 96/1: 67–86. Hawthorne, J., and Manley, D. (2012) The Reference Book. Oxford: Oxford University Press. Jackson, F. (1998) ‘Reference and Description Revisited’, Philosophical Perspectives 12: 201–18. Jackson, F. (2000) From Metaphysics to Ethics: A Defense of Conceptual Analysis. Oxford: Oxford University Press. Jackson, F. (2003) ‘Narrow Content and Representation, or Twin Earth Revisited’, Proceedings and Addresses of the American Philosophical Association 77/2: 55–70. Jackson, F. (2010) Language, Names and Information. Malden, MA: WileyBlackwell. Jackson, F. (2017) ‘Armchair Metaphysics Revisited: The Three Grades of Involvement in Conceptual Analysis’, in G. d’Oro and S. Overgaard (eds.), The Cambridge Companion to Philosophical Methodology. Cambridge: Cambridge University Press, 122–40. Jeshion, R. (2001) ‘Donnellan on Neptune’, Philosophy and Phenomenological Research 63/1: 111–35. Jeshion, R. (2002) ‘Acquaintanceless De Re Belief ’, in J. K. Campbell, M. O’Rourke, and D. Shier (eds.), Meaning and Truth. New York: Seven Bridges Press, 53–78.

 .  ?

65

Jeshion, R. (2009) ‘The Significance of Names’, Mind & Language 24/4: 370–403. Jeshion, R. (2010) ‘Singular Thought: Acquaintance, Semantic Instrumentalism, and Cognitivism’, in R. Jeshion (ed.), New Essays on Singular Thought. Oxford: Oxford University Press, 105–40. Jeshion, R. (2012) ‘Descriptivism and the Representation of Spatial Location’, in W. P Kabasenche, M. O’Rourke, and M. H. Slater (eds.), Reference and Referring. Cambridge, MA: MIT Press. Kripke, S. (1970) Naming and Necessity. Cambridge, MA: Harvard University Press. Kripke, S. (1979) ‘A Puzzle about Belief ’, in A. Margalit (ed.), Meaning and Use. Dordrecht: Reidel, 239–83. Kroon, F. W. (1987) ‘Causal Descriptivism’, Australasian Journal of Philosophy 65/1: 1–17. Lewis, D. (1979) ‘Attitudes De Dicto and De Se’, Philosophical Review, 88/4: 513–43. McDowell, J. (1984) ‘De Re Senses’, in C. Wright (ed.), Frege: Tradition and Influence. Blackwell, 98–109. Mendola, J. (2008) Anti-Externalism. New York: Oxford University Press. Nelson, M. (2021) ‘The Case(s) against Descriptivism’, in S. Biggs and H. Geirsson (eds.), The Routledge Handbook of Linguistic Reference. New York: Routledge, 73–81. Openshaw, J. (2018) ‘Singular Thoughts and De Re Attitude Reports’, Mind & Language 33/4: 415–37. Putnam, H. (1975) ‘The Meaning of “Meaning” ’, Philosophical Papers 2. Pylyshyn, Z. W. (2006) Seeing and Visualizing: It’s Not What You Think. Bradford: MIT Press. Recanati, F. (1993) Direct Reference: From Language to Thought. Oxford: Blackwell. Recanati, F. (2012) Mental Files. Oxford: Oxford University Press. Rescorla, M. (2020) ‘Reifying Representations’, in J. Smortchkova, K. Dolega, and T. Schlicht (eds.), What Are Mental Representations? Oxford: Oxford University Press, 135–77. Russell, B. (1905) ‘On Denoting’, Mind 14/56: 479–93. Russell, B. (1910) ‘Knowledge by Acquaintance and Knowledge by Description’, Proceedings of the Aristotelian Society, Vol. 11: 108–28. Russell, B. (1912) The Problems of Philosophy. Oxford: Oxford University Press.

66   Sainsbury, M. (2020) ‘Varieties of Singularity’, in R. Goodman, J. Genone, and N. Kroll (eds.), Singular Thought and Mental Files. Oxford: Oxford University Press, 21–37. Schiller, H. I. (2019) ‘Acquaintance and First-Person Attitude Reports’, Analysis 79/2: 251–9. Searle, J. (1983) Intentionality. Cambridge: Cambridge University Press. Searle J. (1991) ‘Reference and Intentionality: Response to Burge and McDowell’, in E. Lepore and R. Van Gulick (eds.), John Searle and His Critics. Oxford: Basil Blackwell, 227–42. Soteriou, M. (2000) ‘The Particularity of Visual Perception’, European Journal of Philosophy 8/2: 173–89. Stalnaker, R. (1984) Inquiry. Cambridge: Cambridge University Press. Strawson, P. F. (1950) ‘On Referring’, Mind 59/235: 320–44. Strawson, P. F. (1959) Individuals. London: Methuen. Sullivan, A. (2010) ‘Millian Externalism’, in R. Jeshion (ed.), New Essays on Singular Thought. Oxford: Oxford University Press, 77–102. Taylor, K. (2002) ‘De Re and De Dicto: Against the Conventional Wisdom’, Philosophical Perspectives 36/6: 225–65. Taylor, K. (2010) ‘On Singularity’, in R. Jeshion (ed.), New Essays on Singular Thought. Oxford: Oxford University Press, 77–102.

3 Verbal Signaling Mitch Green

Overtly fictional uses apart . . . to use such an expression as “The king of France” at the beginning of a sentence is, in some sense of “imply”, to imply that there was a king of France. When a man uses such an expression, he does not assert, nor does what he says entail, a uniquely existential proposition. But one of the conventional functions of the definite article is to act as a signal that a unique reference is being made—a signal, not a disguised assertion. When we begin a sentence with “the such-and-such” the use of “the” shows, but does not state, that we are, or intend to be, referring to one particular individual of the species “suchand-such”. P. F. Strawson (1950), 331

3.1 Signals and indices Many behaviors and biological traits are designed to achieve certain ends.¹ Mammalian hearts are designed to pump blood, feathers in birds are designed for thermoregulation, and the ampullae of Lorenzini in hammerhead sharks are designed for the detection of electrical activity in prey (Fields 2007). The notion of design in question permits but does not mandate the behavior of an intelligent or sentient designer. Instead,

¹ An earlier version of this essay was presented to the Evolutionary Pragmatics Group in November 2022. I am grateful to audience members on that occasion for their questions and helpful suggestions. My thanks also to Ozcan Karabag, Magda Kaufmann, Marcin Lewinsky, William Lycan, Ruth Millikan, and Richard Moore for comments on earlier drafts of this essay. Mitch Green, Verbal Signaling In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Mitch Green 2023. DOI: 10.1093/oso/9780198892724.003.0003

68   in saying that the mammalian heart is designed to pump blood, we might just mean that the trait plays a role giving its possessor an advantage in its ecological niche. One type of job a trait or behavior can perform is to convey information. Information is conveyed to an organism when an aspect of its environment makes that information discernible to it. Such an aspect may either be a phenomenon such as a leopard’s footprints in the mud, or a biological trait such as sexual swelling in chimpanzees (Wrangham 1993); information may also be conveyed by means of conventionally meaningful language. However, just conveying information is not sufficient for communication. I convey information to a mosquito when, by emission of CO₂ through respiration, I make it discernible to that organism that there is a likely meal nearby (Gillies 1980). To follow the parlance of Maynard Smith and Harper, this is just a cue (2003, 15). Instead, it is only when the conveyance of information is the product of design that we enter the realm of communication. More precisely, when an organism conveys information to a potential recipient in such a way that both sender and receiver stand to benefit, we have a case of signaling.² An organism might signal that it is toxic, ready to mate, or that there is a food source nearby. Signaling therefore need not convey information about the signaler’s psychological state. For instance, the bright coloration on the skin of a poisonous tree frog signals its toxicity, which it could exhibit even if it had no psychological states at all (Santos et al. 2003). But signaling is not a type of natural meaning (Green 2019). A brightly colored, non-toxic tree frog might be born into a population of poisonous frogs due to a mutation. Its bright coloration still means that it is poisonous even though it is not. Accordingly, while the coloration of a genuinely toxic tree frog correctly presents it as being toxic, this mutant’s coloration incorrectly presents it as being toxic. More generally, when an organism signals that P, that signal is correct on the issue of P if P holds; and incorrect otherwise.

² Here I am departing from a conception of communication that Geurts (2022, 32) describes (but does not endorse) as the exchange of information.

 

69

Poison is expensive to produce, and takes further resources to protect one’s body against (Santos et al. 2003). For this reason, the deceptive mutant will likely have an advantage over its toxic relatives. It will therefore be more apt to survive and reproduce. Its brightly colored, non-toxic progeny may then swamp the population, but if that happens, then eventually predators will cotton on and start ignoring frog coloration in choosing their meals. By that point the signaling system has broken down in the sense that bright coloration no longer signals toxicity. Some signaling systems have evolved in such a way as to be protected against deception-induced breakdown. Funnel-web spiders, for instance, sometimes contest over suitable webs. When that happens, then rather than directly engage in combat, the spiders start to vibrate while standing on the contested web. Reichert (1978, 1984) shows that the spider whose vibration is the most vigorous tends to prevail in these contests. But vibration is a reasonably direct manifestation of the creature’s mass or, as the biologists call it, resource holding potential (RHP). As such, the funnel-web spider’s vibration not only signals but also shows its RHP. In what follows, I shall use ‘index’ to refer to signals that cannot be faked. (Thus defined, there may be no signaling systems that are indices; after all, a pebble might happen to fall on the part of a web near a comparatively puny spider causing it to come across as larger than it in fact is. Instead, we may think of indices in the present sense as an ideal to which signaling systems approximate to a greater or lesser degree.) It is not controversial, I trust, to regard certain traits and behaviors among non-human animals as well as among creatures from different kingdoms such as plants and fungi as signals. Some of these signals may also be indices, or at least approximate that ideal. Yet those who are impressed by the dramatic differences between communication systems found among non-human animals, plants, and fungi, on the one hand, and the communicative repertoire of our species, on the other, may expect the notion of a signal to have little useful application to the latter. Granted, some human nonverbal behaviors are likely cases of signaling. For instance, unintentional human facial expressions may be signals in the sense of that term used here. Most people cannot produce at will a Duchenne smile, in which not only the zygomatic major muscles around

70   the mouth, but also the orbicularis oculi muscles around the eyes, are activated. Yet it is a reasonable hypothesis that the (or at least a) function of the Duchenne smile is to convey positive affect (Green 2007). Because a Duchenne smile is not subject to voluntary control, it may be not just a signal of positive affect but also an index thereof. The Duchenne smile and like cases, however, seem far removed from the quintessential medium of human communication, namely language. In an effort to bridge signaling and certain aspects of linguistic communication, in Section 3.2 I will develop the notions of verbal signal and verbal index. After placing these concepts within the larger context of the cultural evolution of speech acts (Sections 3.3–3.4), we will then be in a position to shed light on phenomena of concern to the philosopher of language, specifically those types of self-ascription that Austin called expositives (Section 3.5a), grammatical evidentials (Section 3.5b), and certain forms of conversational implicature (Section 3.5c).

3.2 Verbal signals and verbal indices According to one influential tradition in the philosophy of language, the basic unit of human communication is the speech act. Searle says, “. . . the production or issuance of a sentence token under certain conditions is a speech act, and speech acts . . . are the basic or minimal units of human communication” (1969, 16). The conditions that Searle has in mind include the speaker’s conforming to certain norms, as well as the speaker’s making her utterance with a particular, complex set of intentions. Those intentions are more or less Gricean, in the sense of being intentions to produce an effect on an addressee in part by means of that addressee’s recognition of the speaker’s intention (1969, 49–50).³

³ I say ‘more or less’ here because Searle advocates a conception of speaker meaning in the spirit of but different from Grice’s. Also, the connection between speech acts and speaker meaning was first drawn by Strawson (1964), who argued against Austin that some but not all illocutionary acts depend on extra-semantic conventions. (A convention is extra-semantic if it depends on more than the fact that words have the meanings they do as a result of conventions. The use of ‘Check!’ in a chess match to indicate that one’s opponent is in check, for instance, would appear to depend on an extra-semantic convention.) The remainder depend on reflexive, communicative intentions. Bach and Harnish (1979) continue this tradition. See Green (2020) for fuller discussion.

 

71

Searle, as well as most other philosophers associated with the Gricean tradition concerning speaker meaning, will likely doubt that there is a phenomenon of significance for linguistic communication lacking a complex intentional structure of the sort Grice first delineated. After all, Grice (1989) provides examples seeming to show that performing an action with the mere intention of producing a psychological effect on an addressee is insufficient to count as a case of meaning, at least of the nonnatural sort. (My surreptitiously leaving Ahmad’s monogrammed handkerchief at the crime scene, with the aim of getting the police to think he is the culprit, is not a case of meaning that Ahmad is the culprit, at least according to any of our folk notions of meaning.) However, the notion of signal I aim to elucidate here is not intended to align with any of our everyday notions of meaning.⁴ Instead, it elucidates the biologist’s notion used to describe communication among non-human species, in a way analogous to the economist’s concept of a game, which may not correspond to the everyday concept thereof. Signaling may well have an explanatory role to play in helping us to understand communication among non-human species, as well as certain aspects of human behavior such as unintentional smiles and manipulated evidence. Might it shed light on aspects of language use? A first step in supporting an affirmative answer is via the notion I shall call a verbal signal, in which an agent employs a conventionally meaningful phrase for the conveyance of information. Conventionally meaningful phrases are not themselves signals, but they are designed to be so used. Further, phrases with semantic content are designed to signal in a way that is constrained by that content. Thus an indicative sentence S whose truth conditions are [[S]] is designed to be used as a signal that the actual situation is a member of [[S]]. We may put this succinctly via the idea that token indicative sentences encode information, and that

⁴ Green (2021) distinguishes four notions of meaning that are readily identifiable in everyday discourse: natural meaning, speaker meaning, linguistic meaning, and “personal” meaning (exemplified in such utterances as “My late aunt’s hunting gear means so much to me”). Green also identifies what he calls ‘organic meaning,’ which corresponds to the notion of signaling used here, but does not present that as being readily identifiable with any commonsense notion of meaning. We note, however, that public discourse such as we find in journalism does sometimes engage in signaling talk. An example is this New York Times headline from August 8, 2022: “F.B.I. Searches Trump’s Florida Home, Signaling Escalation of Inquiries.”

72   they do so in a manner constrained, although not uniquely determined, by their semantic content. Here is a sufficient condition: Sufficient Condition for Verbal Signaling: One who utters indicative sentence S with the objective of conveying the information encoded by S thereby verbally signals that information. We have defined the notion of an index as a type of signal that cannot be faked. Doing so raises the question whether there could also be a verbal index which conveys information in a way that cannot be faked. It may seem implausible that there could be verbal indices in the present understanding of that term. After all, common sense avers that talk is cheap. But many of our public undertakings of commitment are weighty. If I say (under no duress, knowing what I’m agreeing to, and so on) “You’re on!” in response to your offer to bet a sum of money on a football match we are about to witness, then I have not just verbally signaled but also shown my commitment to either gain or lose money depending on how that match goes. That is to say that those words as used in that context function as an index of a cluster of commitments. The same is true of “I bet you $1000 that Liverpool wins this match”, so long as the utterance receives the requisite uptake. I am not in a position to list all the conditions that must be met for a verbal signal to be a verbal index, but they will include such things as the speaker’s possessing the conceptual competence to understand what information her words encode, her having the authority to undertake any commitments she attempts to undertake, the securing of any requisite uptake, and her not being under duress or having made other commitments incompatible with those she is purporting to undertake. Let us encapsulate these conditions as c₁, . . . , cn. In that case we may lay down a Sufficient condition for verbal indexing: A verbal signal that meets conditions c₁, . . . , cn, is also a verbal index. Also, from our earlier characterization of an index, it follows that Verbal indexing is factive: If one verbally signals that S with a sentence-token that is also a verbal index, then S.

 

73

“You’re on!”, “Deal!”, and “I accept your bet” are plausible candidates for sentence-types whose tokenings are, when conditions c₁, . . . , cn are met, also indices of the bets they refer to. We do not normally find this feature with a change of tense (“I accepted your bet yesterday”) or of person (“He accepts your bet”), though of course both of these sentences might be used to say something true. The notion of a verbal index may help to elucidate Strawson’s characterization of a typical use of ‘the soandso’. For one, a person who verbally indexes some state of affair is not thereby illocuting: we have seen that a verbal signal is not as such an illocution, and nothing in our sufficient condition for verbal indexing contributes conditions that would trigger either the reflexive-communicative intentions characteristic of speaker meaning, or the conventional conditions whose meeting suffices for a convention-bound type of illocutionary act. In this way our account preserves Strawson’s idea that one who signals something is not thereby making a statement or an assertion. As well, recall that Strawson contends that someone who verbally signals something “. . . shows, but does not state, that we are, or intend to be, referring to one particular individual of the species ‘such-and-such’.” Prima facie, the idea that in using words one could show something distinct from the utterance itself is apt to be met with suspicion unless we can explain how this works. If, however, Strawson’s notion of a signal is elucidated as a verbal index, we are in a position to allay that suspicion. One way of characterizing the traditional notion of a performative is with the phrase ‘Saying makes it so.’ In that light, the notion of verbal index might appear like a re-packaging of the notion of a performative. In a way it is, but I hope to approach the phenomenon of performativity from a novel perspective that enables us to ask how much cognitive machinery is required to undertake various kind of commitment. Performative utterances like other kinds of illocutionary act are normally thought of as instances of speaker meaning. By contrast, given the sufficient conditions laid down above, a speaker who verbally indexes a state of affairs speaks with an objective of conveying information, but does not need to intend to produce a psychological effect on any addressee; nor therefore must she harbor any reflexive communicative intentions. In what follows, we will see how an approach in terms of

74   verbal indexing enables us to account for data in need of explanation in a more austere manner than does that of speaker meaning.

3.3 The assertive family and illocutionary commitment In the philosophy of language it is a near platitude that speakers who perform speech acts such as assertion undertake commitments.⁵ While this claim is hard to deny, we do well to disentangle different aspects of the notion of commitment in play here. One aspect of commitment flows from the fact that assertion is a type of signal. As we noted in Section 3.1, an organism that signals is correct or incorrect depending on what it signals and how the world is; this carries over to assertions in such a way that a speaker who asserts that P is right on the issue of P when P is true, and wrong on the issue of P when P is not true. Let us call this the liability dimension of assertoric commitment. It is also natural to feel that an assertor is beholden to the norm that she should believe what she says. Let’s call this the frankness norm for assertion.⁶ In addition, we also expect assertors to be prepared to respond to demands for justification of what they say. It won’t be necessary to settle well-known debates about whether the assertor needs to provide sufficient justification to establish knowledge, or instead something weaker; it will be enough for our purposes to say that the assertor must provide strong justification for what she asserts. Let us call this the fidelity constraint on assertion.⁷ Norms governing assertion, then, may be captured at least in part by adverting to the norms of liability, frankness, and fidelity (LFF). Yet assertion is not the only speech act type whose vehicle is the indicative sentence, and other members of what we might call the assertive family may be distinguished from assertion in light of the different ways in

⁵ See Shapiro (2020) for further discussion. ⁶ Frankness is a norm rather than a necessary condition since lies are still assertions. ⁷ Because strong justification need not rise to the level of knowledge, we may remain neutral on the question whether assertion is governed by a knowledge norm. See Marsili and Green (2021) and Pagin and Marsili (2021) for further discussion.

 

75

Table 3.1 Three dimensions of commitment for four members of the assertive family, and for two non-members (sheer guesses, opining, and suppositions) Speech act

Liability

Frankness

Fidelity

Assertion

Y

Y: belief

Conjecture

Y

Y: some reason to think content true; perhaps intention to investigate

Presumption

Y

Educated guess Sheer guess Opining Supposition

Y

Y: intention to treat content as true Y: some reason to think content true N Y: belief Y: intention to investigate what follows from content

Y: provide strong justification if challenged Y: provide some justification if challenged; readiness to determine truth value of content Y: treat content as true

Y Y N

Y: provide some justification if challenged N N Y: reason with content to determine what follows

which the LFF norms apply to them. These differences are set out in Table 3.1. For instance, when a researcher conjectures that P, she stands to be right or wrong on the issue of P depending on P’s truth; we expect her, if not to believe that P, then to have some reason to believe it or to investigate the matter. Further, it is inappropriate to respond to a proposition put forth as a conjecture with a challenge for strong justification such as “How do you know?” Such a response belies a misunderstanding of conjecturing. On the other hand, the speaker offering the conjecture should be able to provide some reason for thinking the content true, and if the challenger can provide strong grounds for denying P, that is normally enough to oblige the conjecturer to retract what she has said. Let us say that membership in the assertive family requires being a speech act governed by at least the LFF norms. This includes assertions, conjectures, presumptions, and educated guesses. By contrast, a sheer guess need not be backed by any particular psychological state other than a grasp of what one is guessing and of one’s liability to be right or wrong.

76   It is thus not governed by the frankness norm. Likewise, a supposition does not make an agent liable to being either right or wrong according to the truth of what she has supposed. As a result, sheer guesses, suppositions, and “opinings” will at best be cousins of assertion rather than its immediate kin. Disentangling commitment into three distinct strands enables us to reflect on the different roles they play, and in the process to speculate on how modern speech act institutions might have evolved to incorporate them. We will expand on this suggestion in Section 3.4. Until then, and as another step to prepare for what follows, it will help here to distinguish between an illocutionary act and distinct commitments that follow from it. Assertion is not deductively closed: one who asserts P has not thereby asserted Q even if P entails Q. However, she is committed to Q, and committed to that content in a way that differs from how she would have been committed to it had she instead conjectured P. Accordingly, while assertion, conjecture, etc., are not deductively closed, illocutionarilyinflected commitments are.⁸ We may accordingly distinguish between assertion and assertoric commitment; conjecture and “conjectural” commitment; and likewise for other members of the assertive family.

3.4 Cultural evolution and the genealogy of speech acts The central idea of cultural evolution, as traditionally propounded, is that behavior patterns in a social group may be accounted for in nongenetic terms as being adaptations to that group’s environment. Such patterns are transmitted by means of learning rather than through genetic means. As such, the transmission process may run through parents (vertical), but also through teachers, mentors, and other highstatus community members (oblique), as well as through peers (horizontal), as opposed to entirely vertical transmission as we have in genetic evolution (Creanza et al. 2017). Both the “teaching” and the “learning” processes emphasized in cultural evolution might be done implicitly, so ⁸ This point is further developed in Green (2020) under the rubric of illocutionary entailment.

 

77

that the “teacher” might not be intending to convey information, and the “learner” might acquire new information or skills without trying to or realizing that she is doing so. Also, although the explanation is in nongenetic terms, the phenomena to be explained might interact in interesting ways with genetic changes. That is part of the story about, for instance, the evolution of lactose tolerance among Western adults (Creanza et al. 2017, 7783). To be explained in cultural-evolutionary terms, the behaviors thus transmitted give the community in which they propagate a survival advantage over other communities that are otherwise similar including earlier versions of that same community. Cultural evolution (CE) would accordingly offer explanations of such human practices as hunting techniques and construction of tools. It could also provide explanations of patterns of behavior not essentially bound up with artifacts such as incest taboos and conversational turn-taking. Either way it might account for why early humans survived over the last 100,000 years while, say, Neanderthals and Denisovans did not. It might also account for why certain human groups have been more successful than others as measured by population, material culture, or robustness against disease or environmental disruption.⁹ The notion of behavior at issue in discussion of CE should be construed broadly so as to include institutions such as marriage, double-entry bookkeeping, and coming-of-age rites. It is not restricted to behaviors in the sense of bodily movements. Heyes (2018) argues that not just behaviors but also cognitive mechanisms should be accounted for as results of cultural evolution. We do not need to take a stand on whether Heyes is right about this, since neither behavior nor cognitive mechanism is exactly the right category for what we wish to explain. Social institutions presumably supervene both upon patterns of behavior and psychological phenomena such as cognitive and affective states.

⁹ CE is not restricted to explanation of human behaviors. It is known that non-human animals have cultures as that term is defined in the relevant literature (Whiten 2000). However, the bulk of research in CE has focused on the explanation of human behaviors and that will be our approach here as well. We also note that CE does not involve any commitment to memetics, which tries to explain a variety of phenomena in terms of the concept of memes as that notion was propounded by Dennett and Dawkins. For further discussion see Heyes (2018).

78   Cultural evolution tends to be cumulative, so that one innovation can build on others that have preceded and enabled it, and such an innovation might occur in the same generation as the one on which it builds. The result is that innovations might spread fast, leading to abrupt changes over relatively short periods of time. (Such things are also possible in genetic evolution but relatively rare.) From the point of view of CE, a lot can happen in just a few generations. In this light, we may observe that CE offers explanations of behavior patterns that are non-conventional due to being superior to all other feasible alternatives. Stone tools are outperformed by bronze, which is in turn outperformed by iron, and so forth. However, CE is also equipped to offer at least partial explanations of behaviors and institutions that are conventional. We have room for at least partial explanations of conventions in light of the fact that many behavior patterns may be seen as Nash equilibria in situations in which more than one such equilibrium is available. When we have multiple Nash equilibria, CE might explain the fact that one of these needs to be chosen. However, it won’t by itself account for how one equilibrium is in fact chosen. Instead, it will need to appeal to an extraneous posit such as an intelligent agent realizing that one of these options needs to be chosen, and settling on one for reasons not pertaining to its intrinsic superiority. That choice might be due to a natural event that by chance leads an agent to do things in one way rather than another that might have been equally viable. Let us in this spirit imagine a communicative system, Vervish, based loosely upon the well-studied system of alarm calls among vervet monkeys (Cheney and Seyfarth 1996). Creatures involved in this communicative system produce different calls for different types of predator such as leopards, snakes, and martial eagles. In response to a vervet producing one of these alarm calls, other vervets will respond appropriately, such as by scampering to high branches in a nearby tree for the “leopard” call, or hiding in the underbrush in response to a “martial eagle” call. One way of interpreting the above system of communication is to idealize it in such a way as to suppose that its users have a limited vocabulary and syntax permitting only monadic predication. Their communication system does not have the means for recursive operations that would yield the complexity that we find in the great majority of human

 

79

natural languages. We may nevertheless imagine that in addition to four noun-like terms for different kinds of predator—a (raptor), b (leopard), c (snake), and d (raptor)—Vervish also possesses four predicate-like terms for characterizing them: F (large), G (small), H (near), and K (far). Concatenations of such expressions will produce sixteen possible Vervish expressions possessing truth conditions such as the following: ‘Fa’ is true iff there is a large raptor; ‘Kc’ is true iff there is a distant snake; ‘Gb’ is true iff there is a small leopard. Vervish contains no connectives, tense markings, evidentials, or other accoutrements. It is thus an impoverished system for communication, and we may hesitate to call Vervish a language due its lack of recursive syntax.¹⁰ Also, while our experience with many modern languages might lead us to expect different lexical items such as ‘F’ and ‘c’ to be tokened sequentially when ‘Fc’ is tokened, that expectation need not be realized in Vervish. Instead, these different lexical items might appear as distinct aspects of an utterance rather than distinct components thereof. In spite of its limitations I propose Vervish as a candidate for what is sometimes called a “proto-language” (Maynard Smith and Harper 2003), and it will serve our purposes by enabling us to pose questions about the cultural evolution of speech acts. Imagine, then, an ancient primate uttering a Vervish sentence such as ‘Hb’ with an objective of conveying the information it encodes.¹¹ Because the expression ‘Hb’ conveys information that may be expressed in propositional terms, that expression stands to be true or false depending on how things are, namely depending on whether there is a nearby leopard. In addition, because the creature’s utterance was designed to convey that information, the success of that enterprise depends on the truth value of the sentence in question.

¹⁰ Green (2021) offers a definition of language in which these requirements are built in. Also, we may imagine successors of Vervish that see the introduction of grammatical complexity in one of the ways hypothesized by Heine and Kuteva (2007). ¹¹ I say ‘objective’ rather than intention because the former term carries less philosophical baggage. Sophisticated machines can have objectives, whereas it is controversial to impute intentions to them. For further discussion see Green and Michel (2022).

80   This is to say that the utterer of ‘Hb’ is liable to be right or wrong on the issue of the presence of a nearby leopard depending on whether there is indeed a leopard nearby. A speaker of Vervish would seem, then, to be beholden to the liability norm described above. She need not therefore be subject to either the frankness or fidelity norm, and so her utterances need not be assertions. More broadly, the three strands making up the commitment associated with assertion provide a starting point for speculating about what its precursors might have looked like. One such precursor would involve only the concept of liability, but not those of frankness and fidelity. Let us call this precursor ur-assertion, or ursertion for short. Perhaps the earliest utterances of Vervish are ursertions. Such a comparatively impoverished speech act may for all that play a useful role in a community. For one, ursertions may be correlated highly enough with worldly affairs to be better-than-chance sources of information. Speakers of Vervish would accordingly have good reasons to heed one another’s ursertions. Such a correlation might be supported by the fact that communities in which ursertors are reliable have better chances of survival than those in which they are not. What is more, cultural practices might grow up to support such reliability. For instance, if conspecifics track a speaker’s record of being right or wrong in what they say, that may be enough to enable them to mete out rewards for those with strong records. Such rewards may come in the form of higher position in grooming, mating, and feeding hierarchies. This would in turn incentivize speakers to get things right when they produce utterances of Vervish. Speakers who seek the rewards of being reliable ursertors do well to organize their utterances to follow their beliefs, that is, to ursert that P only when they believe that P. The reason is that ursertions are, all else being equal, more likely to track the truth if subject to a frankness norm. Thus, while ursertion is not conceptually tied to frankness, it is reasonable to suppose that until it (or a successor practice) becomes so tied, it will be unstable in the following sense: once a group of speakers have a practice of ursertion, members of that community will on the whole do better by, and thus tend toward, urserting only what they take to be the case, that is, only what they believe. This would result in a new linguistic practice, which we may call semisertion.

 

81

In light of the foregoing we may now hypothesize two evolutionary precursors to our modern practice of assertion: 1. “Ursertion”: this would be governed only by the liability norm, namely that one who urserts that P is liable to being correct or not depending on whether P is true; but the speaker is not expected to believe P or have any other psychological state we might call a sincerity condition; and the speaker does not need to be prepared to defend P if challenged. Remark: I am taking it that there could be a practice of ursertion that is not also governed by a norm of sincerity, and in particular not requiring that the speaker believe the content of what she asserts. Is this a coherent position, and in particular can one be committed to a proposition in this way without believing it? The answer seems to be that one can. For one, artifacts that are designed to convey information can be right or wrong depending on how things are, and many such artifacts lack beliefs. For another reason, our current practice of betting corresponds in some respects to ursertion. Betting is clearly a speech act, but does not require belief or any other sincerity condition on the part of the speaker.¹² 2. “Semisertion”: governed by the liability and frankness norms, but not the fidelity norm. Semisertion is more demanding than ursertion in involving two commitments rather than one; however, one who semiserts P still does not need to defend P if challenged, even if she has the expressive resources for doing so. A contemporary analogue of semisertion is opining, in which a speaker might express her point of view but not expect to respond to requests for justification for the view expressed. Similarly one might semisert without being beholden to a norm of giving reasons for what one says. This may be due to speakers treating their utterances as no more than manifestations of their states of mind, and thus not in need of justification in terms of publicly accessible reasons.¹³ ¹² This is at least true for bets against large institutions like casinos that are able to enforce the consequences of bets in ways that don’t depend on the psychological states of the bettors; perhaps when we bet against other individuals that don’t have recourse to powerful enforcement mechanisms, we are expected to be sincere. ¹³ Ursertion, semisertion, and assertion correspond to Price’s (1998) conditions 6, 4, and 5 respectively: (4) One is incorrect to assert that p if one does not believe that p; (5) One is incorrect to assert that p if, though one believes that p, one does not have adequate grounds for

82  

3.5 Applications I hope to have made it plausible that once we disentangle the different strands making up the notion of commitment as it is associated with assertion, we may hypothesize evolutionary precursors to assertion in the shape of ursertion and semisertion. Both kinds of proto-illocution are species of verbal signal, and our next question is whether aspects of language use among modern speakers may be described as either ur- or semisertions.

3.5a Expositives Speakers on occasion characterize the kind of speech act they are performing, and in so doing not only perform that speech act but also undertake a commitment pertinent to group inquiry. We see such cases in the kinds of speech acts Austin called ‘expositives’, which he described as being, “. . . used in acts of exposition involving the expounding of views, the conducting of arguments, and the clarifying of usages and references” (Austin 1962, 161). Examples of expositives include reports, conjectures, illustrations, and postulations. Yet the process by which expositives enable us to conduct arguments and the like raises questions originally posed by Cohen (1964). Consider an utterance of (1) I contend that Ukraine will drive Russia out of its territory. One who utters (1) is normally committing herself, and is heard as committing herself, to the proposition that Ukraine will drive Russia out of its territory. However, it is easy to miss the fact that this content is embedded in an attitudinative (‘I contend that . . .’). Further, we do not

believing that p; (6) One is incorrect to assert that p if, in fact, it is not the case that p (p. 248). Price also points out that what we here call semisertion (close to his concept of a “MOA”) leaves something to be desired in a community that depends on individual members being sources of information for one another. As he puts it, practices of semisertion lack the “friction” that comes of speakers being able to disagree with and on that basis argue with one another. See Shapiro (2021) for a recent discussion of Price’s work on the genealogy of assertion.

 

83

normally interpret clauses embedded within speech act verbs as being illocuted. For instance, consider: (2) Adina contends that Ukraine will drive Russia out of its territory. (3) Last year I contended that Ukraine would drive Russia out of its territory. We do not normally hear assertions of (2) or (3) as committing the speaker to the content of the complement clause. Furthermore, in the first-person, present indicative case such as (1), the force with which the complement clause is put forth is normally determined by the embedding construction. That is why (1) differs from (4) in the way in which the speaker commits herself to the complement clause: (4) I guess that Ukraine will drive Russia out of its territory. That is, while serious utterance of (1) and (4) both normally enable the speaker to undertake commitment to the complement clause, these two modes of commitment are different in a way that we have begun to explain with the help of Table 3.1 and our notion of illocutionary commitment. A contention is a type of assertion, while a guess is not. For these reasons, there is little plausibility in an approach to explaining the data pertaining to cases like (1) and (4) by holding that the attitudinatives are transparent.¹⁴ What, then, is special about (1) and similar cases that accounts for how the embedded clause is heard as being illocuted, and indeed illocuted in a way that is determined in large part by the content of the embedding clause, ‘I contend that . . .’? A popular approach to making sense of cases such as (1) is via the notion of an indirect speech act (ISA). According to this approach, a speaker of (1) illocutes its complement clause indirectly, by way of uttering the larger sentence in which it occurs. Different approaches to ISAs will take different views on the question whether this larger sentence is also illocuted. A well-known ISA ¹⁴ Let us say that an embedding clause ‘C . . .’ is transparent just in case, if ‘a’ is of the propositional type, then ‘Ca’ has the same truth conditions as ‘a’ for all values of ‘a’.

84   approach to cases similar to (1) is found in Bach and Harnish (1979). Given their reconstruction of how a performative such as ‘I order you to leave’ works (1979, 148–51), we may reconstruct their account in terms of the following inference: He is saying, “I contend that P”. He is stating that he is contending that P. If his statement is true, then he is contending that P. If he is claiming that P, it must be his utterance that constitutes his contention. (What else could it be?) 5. Presumably, he is speaking the truth. 6. Therefore, in stating that he is contending that P, he is contending that P. 1. 2. 3. 4.

Given Bach and Harnish’s earlier remarks about saying (1979, 29–33), we should interpret “say” in the foregoing argument in the thin, locutionary sense. On this interpretation, as Jary (2007) rightly notes, the move from premise (1) to premise (2) is questionable: a speaker’s utterance of an indicative sentence, even if that sentence has a truth value, need not be an assertion. (I assume that Bach and Harnish use “assertion” and “statement” to refer to the same illocutionary act-type.) What, however, if we read “stating” in (2) in the weaker, locutionary sense? Then (2) will follow from (1), but the conditional in (4) will be dubious: a mere locutionary act will not be enough to account for how a speaker manages to make a claim. After all, that locutionary act might be made under duress, or in one’s sleep. ‘I claim that P’ is a performative, and so we might look to attempts to explain performatives for progress on our question. Searle (1989) offers an account of explicit performatives as a form of “declaration”, where the notion of declaration is not reducible to a type of assertion. Instead, in a declaration an appropriately empowered speaker makes something the case by characterizing it as being the case. (If I have the authority to do so, I might declare a certain proceeding closed or even declare that my country is at war with yours.) On this basis, we may illustrate Searle’s account of performatives with the help of his description of the interpretation of an explicit performative from the hearer’s point of view. This runs as follows:

 

85

1. S uttered the sentence “I hereby order you to leave” (or he uttered “I order you to leave”, meaning “I hereby order you to leave”). 2. The literal meaning of the utterance is such that by that very utterance the speaker intends to make it the case that he orders me to leave. 3. Therefore, in making the utterance S manifested an intention to make it the case by that utterance that he ordered me to leave. 4. Therefore, in making the utterance S manifested an intention to order me to leave by that very utterance. 5. Orders are a class of actions where the manifestation of the intention to perform the action is sufficient for its performance, given that certain other conditions are satisfied. 6. We assume those other conditions are satisfied. 7. S ordered me to leave, by that utterance. 8. S both said that he ordered me to leave and made it the case that he ordered me to leave. Therefore he made a true statement. (Searle 1989, 553) Unfortunately, the above reasoning is questionable on at least two grounds. First of all, (2) is dubious. No sentence is such that its very utterance guarantees the presence of any intention. A somniloquist might utter, “I hereby order you to leave”, in his sleep, without intending to make it the case that he orders anyone to leave. Likewise for someone who speaks that sentence under duress (Green 2020). Second, consider the first sentence of (8). The word “said” it contains is notoriously ambiguous as between thin (locutionary) and thick (illocutionary, and in particular assertoric) readings. Which reading does Searle have in mind? If he intends the weaker, thin sense, then the second sentence in (8) will not follow from the first: the speaker may have uttered a true sentence, but he will not have made a statement, and thus will not have made a true statement. On the other hand, suppose that Searle has the “thick” sense of ‘say’ in mind. Then on this reading, while the first sentence of (8) will entail the second, that first sentence will not follow from the premises allegedly leading to it. Thus on either interpretation of “say” in line (8), Searle’s reasoning is fallacious. Jary (2007) also discusses performatives and offers an approach that will prove instructive for our purposes. A core idea for him is

86   performative utterances as “unavoidable context changers” (p. 212). By this he means that while a, say, felicitous assertion that P will only succeed in getting P into common ground if all interlocutors accept P and recognize one another as doing so, a felicitously uttered performative sentence serves as a manifest event: it is as sure to update CG with the information that the speaker has performed a certain act as is a rock crashing through a window in a room where we are having a conversation sure to update CG with information as to the rock’s presence. Thus, once George felicitously utters a performative such as, “I promise to meet you tomorrow”, CG now contains the proposition that George promises to meet his addressee tomorrow. By contrast, if George asserts the sentence, “It will rain tomorrow”, the only thing that is sure to enter GC is the proposition that George asserted that it will rain tomorrow. The proposition George expresses, however, only enters GC if all interlocutors accept it and each is aware that others do so. Jary also denies that explicit performatives can be true or false (2007, 211) and denies that they are used to make assertions. In the course of doing so, he attempts to support his thesis about explicit performatives as unavoidable context changers by construing them as acts of showing, building on then-unpublished work by Green (2005). Jary writes, Explicit performatives . . . are acts of showing consisting in a state of affairs and an act of pointing. What is special about them is that the utterance constitutes both the state of affairs and the act that points to it: it is both the act of showing and the event being shown. I show that I am promising by pointing to my utterance and referring to it as a promise. In showing that I am promising, I necessarily promise. Performing the illocutionary act that one refers to when showing that one performs that act is an unavoidable consequence of that act of showing. (Jary 2007, 220)

Consider Jary’s third sentence: “I show that I am promising by pointing to my utterance and referring to it as a promise”. After all, I don’t show that I am yelling, say, simply by pointing to my utterance and referring to it as a yell; I also have to be yelling. A similar observation applies to promising. Jary does remark that, to be a promise, an utterance must be

 

87

intended as a promise. How would this carry over to (1)? Presumably it would do so via the idea that for it to be a claim that P, utterance of (1) has to be so intended. But this leaves open the question, what is the force (if any) with which (1) is uttered? Jary won’t wish to say that the speaker of (1) intends to utter the entire sentence as an assertion, for that would commit Jary to the view that it is his main aim to rebut. On the other hand it will not do to think of (1) as merely being locuted, as doing so will land us back with the case of the somniloquist. To put the point in terms of a dilemma: either (1) is asserted, or it is not. In the former case, Jary relinquishes one of the main contentions of his paper. If (1) is not being asserted, however, then if the only other option is to see (1) as being merely locuted, it is difficult to see how a speaker could use a sentence like (1) to put forth P assertorically. I will next argue that the above dilemma is a false one. The reason is that a sentence such as (1) need be neither merely locuted nor illocuted for its utterer to put forth its complement P assertorically. Instead there is a third option between the locutionary and the illocutionary, found in the notion of verbal signaling. To appreciate why, observe that if I say (under no duress, knowing what I’m agreeing to, and so on) “You’re on!” in response to your offer to bet a sum of money on a race we are about to witness, then I have not just signaled but also shown my commitment to either gain or lose money depending on how that race goes. That is to say that those words as used in that context function as an index of a cluster of commitments. The same is true of “I bet you $1000 that Team Astana wins this stage”, so long as the utterance receives the requisite uptake. I am not in a position to list all the conditions that must be met for a verbal signal to be a verbal index, but they will include such things as the speaker’s possessing the conceptual competence to understand what information her words encode, her having the authority to undertake any commitments she attempts to undertake, the securing of any requisite uptake, and her not being under duress or having made other commitments incompatible with those she is purporting to undertake. Let us suppose that such conditions may be specified as c₁, . . . , cn. Then as specified in conditions c₁, . . . , cn, each agent with the requisite conceptual resources has final authority over what propositional commitments they undertake: in the absence of other, conflicting commitments,

88   we can always claim (guess, conjecture, etc.) that P and thereby commit ourselves to P. Note also that propositional commitment is not the same as belief, since I can undertake commitment to a proposition that I do not believe, by for instance insincerely asserting it. We may now see that certain avowals are when uttered felicitously also indices of the information they encode. This point of view offers theoretical underpinning, for certain cases, to the familiar thought that performative utterances have the “saying makes it so” property.¹⁵ Here is an argument: 1. S has said, “I contend that P”, with the objective of conveying the information encoded by “I contend that P”, and doing so in such a way as to meet conditions c1, . . . , cn (assumption). 2. S’s tokening of “I contend that P” is a verbal signal (1, Sufficient condition for verbal signaling). 3. S’s tokening of “I contend that P” is a verbal index (1, 2, Sufficient condition for verbal indexing). 4. S contends that P (2, 3, Verbal indexing is factive). 5. One who contends that P undertakes assertoric commitment to P. (Contentions are a type of assertion.) 6. S has undertaken assertoric commitment to P (4, 5). We have been able to deduce (6) without assuming that S’s utterance of ‘I claim that P’ is an illocutionary act. Instead, we have assumed that the utterance is a verbal signal meeting conditions c₁, . . . , cn, which do not themselves presuppose conditions sufficient for assertion or any other illocutionary act. That is to say, the speaker’s utterance of (1) is an ursertion. We were nevertheless able to conclude that in uttering “I contend that P”, S is able to contend that P, as well as to commit herself to P assertorically. Not only have we made headway on “Cohen’s problem,” we have also made a case that ursertions performed by agents with appropriate conceptual resources, and in the absence of conflicting commitments, ¹⁵ We are not here offering a general explanation of how performatives work.

 

89

can also constitute speech acts such as assertion with all the normative implications that follow. So understood, an ursertion made under the right conditions can also trigger an illocutionary act.

3.5b Grammatical evidentials Linguists have long observed, and philosophers have recently come to appreciate the fact, that, in numerous languages other than English, sentences are grammatically marked in such a way as to indicate the epistemic status of the content, or at least main content, expressed by that sentence. In English we can indicate that evidential status with lexically encoded information such as in: (7) I heard from George that it will rain tomorrow. (8) It will rain tomorrow, according to George. By contrast, in approximately 25% of the world’s languages including Turkish, Cheyenne, Korean, Quechua, Arawak, and Bulgarian, sentences contain particles that do not contribute to the main content expressed by that sentence but instead serve to indicate its epistemic status, such as being something the speaker saw for herself, heard from someone else, or inferred from something else they observed. In some languages such as Arawak (Aikhenvald 2004, 1–2), an evidential is obligatory in every utterance of an indicative, while others, such as Tibetan, allow a speaker to be non-committal about the evidence for her claim (de Villiers et al. 2009). The evidential particles are part of a closed, that is nonproductive, system, and are part of a sentence’s grammatical analysis when they occur. For this reason, we will call these expressions grammatical evidentials, or GEs. As an illustration of how some GEs operate, Murray (2017) reports that, in Cheyenne, a speaker will describe events with a further indication of the source of their information. Consider: 9. É-némene-sêstse Sandy. Sandy sang, I hear(d).

90   The evidential in this case is ‘sêstse’, which here appears as a suffix on the verb meaning ‘to sing’. About it, van Elswyck writes, “What it specifies is that the speaker has hearsay evidence for the proposition that Sandy sang” (2019, 2). Notice van Elswyck’s description of the grammatical evidential as specifying something. This leaves it undecided whether that specification is due to the fact that the content of the GE is part of what the speaker said, or is instead information carried in some other way; likewise this leaves it undecided what is the illocutionary force, if any, with which the evidential content is presented. Also, van Elswyck contends that in examples such as (9), the speaker expresses two propositions, namely that Sandy sang (what we shall call the scope proposition), and that she (the speaker) heard that Sandy sang (the evidential proposition). I take it that he would say the same thing for a case in which the evidential specifies not hearsay but instead something that the speaker witnessed. Our question is then, what is the relation between the evidential and scope propositions? After all, it would seem that the former should guide our response to the latter. Murray (2017) provides clarity on these issues by means of her tripartite analysis of evidentials as involving (a) the presentation of an at-issue content, (b) a not-at-issue restriction, and (c) what she calls an illocutionary relation, which is how the content operates on common ground as determined by the grammatical mood of the main clause.¹⁶ Further, the not-at-issue restrictions are not directly challengeable, and will be added to common ground even if the at-issue content is challenged or denied (2017, 69). That is, not-at-issue restrictions are not proposals to update common ground; they simply do update it in a way analogous to Stalnaker’s notion of a manifest event (Stalnaker 2014, 47). Thus, an utterance of (9), whether or not it updates CG with the proposition that Sandy sang, invariably updates it with the content that the speaker heard that Sandy sang. In contrast to the automatic updating of CG with the content of the evidential proposition, Murray tells us, the scope proposition is still only put forth as a proposal to update CG (2017, 70). This is the case

¹⁶ Murray in fact holds that this tripartition applies to all sentences. See (2017, 5, 59).

 

91

regardless of the content of the evidential proposition. Thus consider what Murray terms the ‘direct evidential’, an example (her 3.5) of which Murray translates as: 10. ‘Annie won, I witnessed.’ Here the main proposition, that Annie won, is understood only as a proposal to update common ground. This, however, is not a plausible analysis of the role of direct evidentials. For if the proposition that the speaker witnessed a certain event e is accepted into CG, and CG is deductively closed, then it also contains the proposition that e occurred whether or not interlocutors consciously draw that conclusion. A similar challenge arises for Murray’s analysis of what she terms inferential evidentials, one of which would be glossed as ‘Annie won, I take it.’ Murray holds that this sentence has the effect of updating GC with the proposition that the speaker infers that Annie won. She also writes, “With the inferential evidential, the speaker is not committed to the truth of the at-issue proposition, though she believes it to be a good possibility” (2017, 74). However, a few lines later, Murray writes, “. . . there is less commitment to the truth of the at-issue proposition than with the direct evidential.” Assuming that being less committed to proposition P entails that one is somewhat committed to it, Murray seems to be ambivalent about the nature of commitment associated with inferential evidentials. Our multi-dimensional approach to commitment, together with a corresponding delineation of precursors to modern speech acts that may still have a role to play in contemporary language, offer a way forward here. For in light of what we have so far observed, it does not seem likely that a speaker using a GE has performed two speech acts, one in which the scope proposition is being put forth as, say, an assertion while the evidential proposition is being put forth with some other force. But then, what other tools might we use to understand the way in which the GE is part of the overall communicative package? After all, just presenting a content (propositional or otherwise) is not to make a move in a language game. Our answer is that the evidential content is the vehicle of a verbal signal. As with expositives discussed in the last

92   section, users of GE’s are held to a liability standard in their use of those expressions, but, unlike the case of expositives, users of GE’s are most likely held to the standard of frankness. That is to say, standard uses of GEs, the evidential propositions, are semisertions rather than ursertions. In this light we may formulate an inference pattern for how CG is updated when a direct evidential is used: 1. S has said, “P*”, with the objective of conveying the information encoded by “I directly perceived that P,” and doing so in such a way as to meet conditions c₁, . . . , cn (assumption). 2. S’s tokening of “P*” is a verbal signal (1, Sufficient condition for verbal signaling). 3. S’s tokening of “P*” is a verbal index (1, 2, Sufficient condition for verbal indexing). 4. P* (2, 3, Verbal indexing is factive). 5. P (4, ‘perceive’ is factive). For the case of the inferential evidential, we may recall that fidelity comes in different forms, only one of which requires providing strong enough justification for knowledge. Conjectures when challenged should be backed up with weaker evidence, albeit strong enough to make it worth someone’s while to investigate the hypothesis being conjectured. This is to say that ‘S conjectures that P’ entails only that S is committed to P as a conjecture, rather than as something that she knows. One construal of ‘Annie won, I take it’ is that the speaker typically conjectures or hypothesizes that Annie won but does not assert that she does. We may capture this phenomenon with the following line of reasoning: 1. S has said, “P*”, with the objective of conveying the information encoded by “I take it that P,” and doing so in such a way as to meet conditions c₁, . . . , cn (assumption). 2. S’s tokening of “P*” is a verbal signal (1, Sufficient condition for verbal signaling). 3. S’s tokening of “P*” is a verbal index (1, 2, Sufficient condition for verbal indexing).

 

93

4. P* (2, 3, Verbal indexing is factive). 5. S undertakes ‘conjectural’ commitment to P. Conjectural commitment is, of course, still a form of commitment. We therefore see the way in which the current approach explains how, with indirect evidentials, speakers undertake commitment to the scope proposition, but in a way less demanding than if they had asserted that proposition.

3.5c Conversational implicature Philosophers concerned with conversational implicature generally categorize it as falling within the scope of speaker meaning. That is, a speaker who says that P and in so doing conversationally implicates that Q, thereby speaker means that Q.¹⁷ This understanding of implicature (hereafter I will use ‘implicature’ to refer to conversational implicature, leaving conventional implicature to one side) is plausible for those quantity implicatures—that is, implicatures whose generation depends on a violation of the Maxim of Quantity—in which the speaker possesses the information about which she is being reticent. I will here argue that it is not plausible for those cases in which a speaker’s violation of Quantity is due to her lacking the information needed to answer the question or questions that are under discussion. It is, as we shall see, also not plausible for those implicatures that depend on the Maxim of Manner for their generation. To appreciate these claims, consider Grice’s classic example of the ‘South of France’ case (1989, 32). Recall, A and B are planning a holiday in France, and it is common knowledge that A would like to see his friend C if doing so would not be too difficult. The following dialogue ensues: A. Where does C live? B. Somewhere in the South of France.

¹⁷ See Neale (1992: pp. 24ff.), for defense and development of this point, as well as Petrus (2010: pp. 4–12), Bach (2012, p. 54); Bianchi (2013, p. 110).

94   Grice glosses this example as one in which B conversationally implicates that he is not in a position to be more informative than he has been in answering as he does. Grice writes, Gloss: There is no reason to suppose that B is opting out; his answer is, as he well knows, less informative than is required to meet A’s needs. This infringement of the first maxim of Quantity can be explained only on the supposition that B is aware that to be more informative would be to say something that infringed the second maxim of Quality, “Don’t say what you lack adequate evidence for,” so B implicates that he does not know in which town C lives. (1989, 32–3)

As noted, conversational implicature is normally construed as a species of speaker-meaning, in that for an agent to implicate that P, she must harbor reflexive communicative intentions toward an addressee. On the most common construal of such intentions, in the South of France case that would require that B intend A to believe that B does not know anything more specific about C’s whereabouts than that C is in the South of France, with the further intention that A come to believe this at least in part by recognition of B’s intention that she do so. Accordingly, as Grice and many commentators who follow him, such as Levinson (1983), affirm, B not only speaker-means what he says, namely that C lives somewhere in the South of France, but also speaker-means a further content that he does not say, namely that he is not in a position to be more informative than that. In the case here imagined, the speaker has violated the Quantity maxim in not providing as much information as may reasonably have been expected. However, the conclusion that Grice draws in the abovequoted gloss—that B implicates that she does not know in which town C lives—does not follow from the considerations he cites, and is independently implausible. The reason is that we may account for B’s reticence on the hypothesis that she was aware that, had she been any more informative, she would have said something for which she lacked sufficient evidence, thus infringing Quality. More generally, just because a speaker violates a communicative norm, we cannot infer that she

 

95

speaker-means something more than she says, such as that she cannot be any more informative. Instead, her norm-violation may simply be due to her adherence to some other constraint such as her desire not to infringe a distinct, possibly communicative, norm. B’s reticence might be a manifestation of that ignorance, and thus a manifest event in the sense of that notion discussed in Section 3.4b. On the other hand, B might have spoken with an objective of conveying the information that she can be no more informative than she has been. In that case, she would be signaling her ignorance rather than merely manifesting it.¹⁸ Her implicatum, then, would be being conveyed by means of a proto-illocutionary act, either an ursertion or a semisertion. (Which of these two best describes her act depends on whether her implicatum is held to a standard of sincerity.) Similar remarks apply to many implicatures that depend on the Maxim of Manner. All else being equal, we tend to narrate events in the order in which they occurred. In so doing, we give our addressees to understand that they occurred in the order in which we have described them. As with the Quantity-based implicatures discussed above, it should now be clear that it over-describes such situations to hold that speakers speaker-mean that the events in question occurred in that order. Instead, a more plausible description of such cases is that in so speaking we signal that the events occurred in the order described. We are now in a position to appreciate the value in a strategy of construing modern-day speech acts as products of cultural evolution. Our current practice of assertion, for instance, may have evolved from more primitive practices such as ursertion and semisertion. What is more, in a way similar to Jackendoff ’s (2003) hypothesis that modern languages contain expressions that are “fossils” of more primitive forms of communication, so too our isolation of precursors to modern-day speech acts leaves open the possibility—which I’ve argued to be actualized—that such precursors are still with us and even play an important role in everyday communication.

¹⁸ She might also harbor the complex of intentions sufficient of speaker meaning, but doing so is not mandated by the communicative situation in which she finds herself.

96  

References Aikhenvald, A. Y. (2004) Evidentiality (Oxford University Press). Austin, J. L. (1962) How to Do Things with Words, 2nd ed., ed. J. Urmson and M. Sbisa (Harvard University Press). Bach, K., and R. Harnish (1979) Linguistic Communication and Speech Acts (MIT Press). Bach, K. (2012) ‘Saying, Meaning, Implicating’, in K. Allan and K. Jaszczolt (eds.), Cambridge Handbook of Pragmatics (Cambridge University Press), pp. 47–68. Bianchi, C. (2013) ‘Implicating’, in M. Sbisà and K. Turner (eds.), Handbook of Pragmatics, Volume II: Pragmatics of Speech Actions (de Gruyter), pp. 107–42. Cheney, D., and R. Seyfarth (1996) ‘Function and Intention in the Calls of Non-human Primates’, Proceedings of the British Academy 88, pp. 59–76. Cohen, L. J. (1964) ‘Do Illocutionary Forces Exist?’, Philosophical Quarterly 14, pp. 118–37. Creanza, N., O. Kolodny, and M. Feldman (2017) ‘Cultural Evolutionary Theory: How Culture Evolves and Why It Matters’, PNAS 114, pp. 7782–9. de Villiers, J., J. Garfield, H. Gernet-Girard, T. Roeper, and M. Speas (2009) ‘Evidentials in Tibetan: Acquisition, Semantics, and Cognitive Development,’ in S. A. Fitneva and T. Matsui (eds.), ‘Evidentiality: A Window into Language and Cognitive Development’, New Directions for Child and Adolescent Development 125 (Jossey-Bass), pp. 29–47. Fields, D. (2007) ‘The Shark’s Electric Sense’, Scientific American 297, pp. 74–81. Geurts, B. (2022) ‘Evolutionary Pragmatics: From Chimp-Style Communication to Human Discourse’, Journal of Pragmatics 200, pp. 24–34. Gillies, M. (1980) ‘The Role of Carbon Dioxide in Host-Finding by Mosquitos (Diptera: Culicidae): A Review’, Bulletin of Entomological Research 70, pp. 525–32. Green, M. (2005) ‘Cohen’s Problem and Dynamic Semantics’, Paper presented at the 9th International Pragmatics Conference, Riva del Garda, Italy. Green, M. (2007) Self-Expression (Oxford University Press).

 

97

Green, M. (2016) ‘Assertion’, in D. Pritchard, ed., Oxford Handbooks Online, https://www.academia.edu/28610217/Assertion_Oxford_Handbooks_ Online_2017_. Green, M. (2019) ‘Organic Meaning: An Approach to Communication with Minimal Appeal to Minds’, in A. Capone et al. (eds.), Further Advances in Pragmatics and Philosophy (Springer), pp. 211–28. Green, M. (2020) ‘Speech Acts’, in E. Zalta (ed.), Stanford Encyclopedia of Philosophy, https://plato.stanford.edu/entries/speech-acts/. Green, M. (2021) The Philosophy of Language (Oxford University Press). Green, M., and J. Michel (2022) ‘What Might Machines Mean?’, Minds and Machines 32, pp. 323–38. Grice, P. (1989) Studies in the Way of Words (Harvard University Press). Heine, B., and T. Kuteva (2007) The Genesis of Grammar (Oxford University Press). Heyes, C. (2018) Cognitive Gadgets (Harvard University Press). Jackendoff, R. (2003) Foundations of Language: Brain, Meaning, Grammar, Evolution (Oxford University Press). Jary, M. (2007) ‘Are Explicit Performatives Assertions?’, Linguistics & Philosophy 30, pp. 207–34. Marsili, N., and M. Green (2021) ‘Assertion: A (Partly) Social Account’, Journal of Pragmatics 181, pp. 17–28. Maynard Smith, J., and D. Harper (2003) Animal Signals (Oxford University Press). Murray, S. (2017) The Semantics of Evidentials (Oxford University Press). Neale, S. (1992) ‘Paul Grice and the Philosophy of Language’, Linguistics and Philosophy 15, pp. 509–59. Pagin, P., and N. Marsili (2021) ‘Assertion’, in E. Zalta (ed.), The Stanford Encyclopedia of Philosophy. Petrus, K. (2010) ‘Paul Grice: Philosopher of Language, but More Than That’, in K. Petrus (ed.), Meaning and Analysis: New Essays on Grice (Palgrave Macmillan), pp. 1–30. Price, H. (1998) ‘Three Norms of Assertibility, or How the MOA Became Extinct’, Philosophical Perspectives 12, pp. 241–54. Reichert, S. (1978) ‘Games Spiders Play: Behavioral Variability in Territorial Disputes’, Behavioral Ecology and Sociobiology 3, pp. 135–62.

98   Reichert, S. (1984) ‘Games Spiders Play III: Cues Underlying ContextAssociated Changes in Agonistic Behavior’, Animal Behaviour 32, pp. 1–15. Santos, J., L. Coloma, and D. Cannatella (2003) ‘Multiple, Recurring Origins of Aposematism and Diet Specialization in Poison Frogs’, PNAS 100, pp. 12792–7. Searle, J. (1969) Speech Acts: An Essay in the Philosophy of Language (Cambridge University Press). Searle, J. (1989) ‘How Performatives Work’, Linguistics and Philosophy 12, pp. 535–58. Shapiro, L. (2020) ‘Commitment Accounts of Assertion’, in S. Goldberg (ed.), Oxford Handbook of Assertion (Oxford University Press), pp. 75–97. Shapiro, L. (2021) ‘Truth’s Dialectical Role: From Friction to Tension’, Inquiry 64, pp. 1–21. Stalnaker, R. (2014) Context (Oxford University Press). Strawson, P. F. (1950) ‘On Referring’, Mind 59, pp. 320–44. Strawson, P. F. (1964) ‘Intention and Convention in Speech Acts’, Philosophical Review 73 (4), pp. 439–60. Van Elswyck, P. (2019) ‘Testimony and Grammatical Evidentials’, in E. Fricker, P. Graham, D. Henderson, and N. Pedersen (eds.), Routledge Handbook of Social Epistemology (Routledge), pp. 135–44. Whiten, A. (2000) ‘Primate Culture and Social Learning’, Cognitive Science 24, pp. 477–508. Wrangham, R. (1993) ‘The Evolution of Sexuality in Chimpanzees and Bonobos’, Human Nature 4, pp. 47–79.

4 Disquotation, Translation, and Context-Dependence Richard Kimberley Heck

Disquotational theories of truth and reference claim that those notions can be adequately characterized in terms of such apparent trivialities as these:¹ (1) “Every bottle is empty” is true iff every bottle is empty. (2) “David Ortiz” refers to David Ortiz. These statements concern expression types: The type sentence “Every bottle is empty” and the type name “David Ortiz.” But many sentence types—for example, “That woman is a great musician”—do not express propositions on their own. As a result, the type sentence (3) “That woman is a great musician” is true iff that woman is a great musician. does not express a proposition (since its right-hand side does not). So the disquotational model is of no obvious use where context-dependent expressions are concerned. If such cases were exceptional, then one might regard the problem as merely one of detail. But it has become clear over the past few decades that context-dependence is the norm in natural language. For example, ¹ The arguments given here apply equally to pro-sentential (Grover, Camp, and Belnap 1975) and substitutional (Hill 2002) theories of truth. I focus on disquotational approaches simply because they are more widely discussed. Richard Kimberley Heck, Disquotation, Translation, and Context-Dependence In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Richard Kimberley Heck 2023. DOI: 10.1093/oso/9780198892724.003.0004

100    actual utterances of “Every bottle is empty” almost never mean that absolutely every bottle is empty but only that every bottle in some contextually specified group is empty (see Stanley and Szabó 2000). And empty of what? Air? Or just beer? It turns out, then, that (1) suffers from the same problems as (3) does (see Carston 1988). Even old friends like “Snow is white” are context-dependent in virtue of the contextsensitivity of color adjectives (see Szabó 2001; Reimer 2002; Kennedy and McNally 2010), not to mention tense. Proper names are a particularly interesting case. There are many people named “David Ortiz,” and there are two broad ways in which this fact can be theorized. On the first, such names are ambiguous in essentially the same way that “bank” is. There is not just one word “bank” but two, which just happen to be spelled and pronounced the same way.² On this view, then, there are as many names “David Ortiz” as there are people with that name. If that were the correct view, then names would pose no problem for disquotationalism. We would simply have: (4) “David Ortiz₁” refers to David Ortiz₁. (5) “David Ortiz₂” refers to David Ortiz₂. Just as we have: (6) “bankfinancial” is true of x iff x is a bankfinancial. (7) “bankriver” is true of x iff x is a bankriver. But there are strong arguments against the ambiguity theory of names (Gray 2014; Bach 2015; Fara 2015).³ The more popular view nowadays is

² They are both homonyms and homographs, whereas “deer” and “dear” are just homonyms. (Among the evidence for this claim is the fact that the two words “bank” have distinct etymologies. It is just an accident of history that they are spelled and pronounced the same way.) ³ For a long time, I resisted those arguments. It was when I thought about the use of bare surnames, as in “Ortiz is next up to bat,” that I came around. It just doesn’t seem plausible to me that there are as many surnames “Ortiz” as there are people with that surname. If not, however, then we already need whatever resources the predicate view of name requires.

, , -

101

thus that names are predicates, a possibility suggested by such constructions as “There are many David Ortizes.” It might seem as if the predicate view of names is readily accommodated by disquotationalism, thus: (8) “David Ortiz” is true of x iff x is a David Ortiz. But this would only characterize the reference of the type “David Ortiz.” It would not help us to understand how the notion of truth should be applied to specific utterances of, say, “David Ortiz is Dominican.” On the predicate view of names, the statement (9) “David Ortiz is Dominican” is true iff David Ortiz is Dominican. is every bit as problematic as (3), and for essentially the same reason: There is nothing sensible to be said about how the notion of truth should be applied to the type sentence “David Ortiz is Dominican.” When that sentence is uttered, a specific person will (ordinarily) be referred to, and the token sentence uttered will be true just in case that person is Dominican. But the disquotational schema does not apply to utterances of (9) any more than to utterances of (3). As things now stand, therefore, it is a reasonable hunch that the notions of truth and reference do not really apply to type expressions at all. Perhaps there are some type sentences that it does make sense to call true or false (e.g., statements of pure mathematics). If so, however, not only are those rare exceptions, but one might well argue that even ‘eternal’ sentences are only derivatively called true or false: It is, in the first instance, the tokens of eternal sentences that are true or false; the sentences themselves can be called true or false only because the truthvalues of their tokens do not vary with the occasion of utterance.⁴

⁴ Indeed, the term “eternal sentence” comes from Quine, who defines it thus: “a sentence whose truth value stays fixed through time and from speaker to speaker” (Quine 1960, 177). It is the tokens whose truth-value does not vary. Even on Quine’s account, then, tokens are what is primary. Elsewhere, Quine (1970b, 13–14) notes both that disquotation only really applies to ‘eternal’ sentences and that truth is actually applied to token sentences, but he never seems to consider how those two facts can be reconciled.

102    It appears, therefore, that disquotational theories of truth and reference have the wrong subject-matter. I don’t claim that this is a knock-down argument against disquotationalism (though I do think it is quite a powerful one). Its lesson, for our purposes, is that the question how disquotationalism should handle context-dependence is of fundamental importance. It’s surprising, then, that so few efforts have been made to answer that question.⁵ Indeed, I am aware of just one such attempt, which is due to Hartry Field, and I can think of no alternatives to the view he defends. In the remainder of this paper, then, I will argue that no solution of the sort Field offers can succeed. We’ll begin that discussion in Section 4.2. First, I want to discuss another way in which context-dependence threatens to undermine one of disquotationalism’s central theses.

4.1 The generalizing role of truth Disquotationalists typically claim that the truth-predicate plays only an ‘expressive’ role: It functions, we are told, as a device of generalization (Quine 1970b, 11). Suppose, for example, that Alex says: (10)

Some of the things that Obama said were true.

The standard disquotationalist claim is that such uses of “true” are merely ‘expressive’: In making this utterance, Alex says the same thing they would have said had they instead uttered the disjunction of the various sentences that Obama uttered, and in some strong sense of “the same” (Gupta 1993, 66ff.). The truth-predicate is useful only because Alex may not know which sentences Obama uttered and so cannot utter the disjunction themselves.⁶ But this assumes that the sentences Obama uttered were not contextdependent. Since they almost certainly were, it is just false that Alex ⁵ I suspect this is because of the almost exclusive attention paid in disquotationalist writing to formal languages. ⁶ In other cases, such as “All of the axioms of Peano arithmetic are true,” we cannot, even in principle, utter all the sentences mentioned, because there are infinitely many of them.

, , -

103

could have said the same thing they said by uttering (10) by instead uttering the disjunction of the sentences Obama uttered. Moreover, for the reasons discussed above, it would make no sense for Alex to attribute truth or falsity to the type sentences that Obama uttered. So, when Alex generalizes over ‘the things Obama said,’ they are not talking about the type sentences that Obama uttered but about something else. What? Suppose you were to ask Alex, “And what things did Obama say that were true?” You would not be likely to get an answer like: (11)

Obama said “He will be elected President.”

but instead one like: (12)

Obama said that Biden would be elected President.

The point here is one made long ago by Sir Peter Strawson (1950, 130): Only very rarely, in natural language, do we attribute truth to sentences (or even to utterances). We just do not often say things like “The sentence ‘John broke the window’ is true.” We do say things like “It’s true that John broke the window, but it was an accident.”⁷ One might worry that, if we regard “true” as primarily applying to propositions, and analyze (10) as quantifying over propositions, then we will be committed, as Field (1994, 266–7) puts it, to “strange entities.” But that is just a mistake. What is being claimed is that, to evaluate statements like (10), speakers need to know things like: (13)

It is true that Biden will be elected President iff Biden will be elected President.

because the ‘things Obama said’ are things like: that Biden will be elected President. This view will commit us to ‘strange entities’ only if ‘thatclauses’ themselves refer to strange entities. Not only does that need

⁷ Compare: The sentence “John broke the window” is true, but it was an accident. That seems ungrammatical—a point emphasized by Higginbotham (2006) in a different but related context.

104    argument,⁸ but, if that-clauses do refer to propositions, then we are committed to ‘strange entities’ anyway, quite independently of how we handle generalizations involving “true.” I conclude, therefore, that there is no reason to think that actual speakers make use of a disquotational notion of truth—a deflationary notion of sentential truth—in evaluating such generalizations as (10). It is a different question, of course, whether one could do so. I doubt it, but the claim I am opposing is that a disquotational notion of truth is required if we are to make sense of such generalizations as (10). That is supposed to force opponents of disquotationalism to accept at least the legitimacy and utility of the disquotational notion, which would give disquotationalists a significant dialectical advantage (Field 1986, 59; see also Heck 2004, sec. 2; and McGee 2005, 147). But the disquotational notion is of no obvious use in this connection, and the notion of truth actually used by ordinary speakers in making and evaluating such generalizations is the propositional notion of truth at work in (13).⁹

4.2 Disquotation and translation As Field (1994, 260; 2017) emphasizes, a disquotational truth-predicate will apply, in the first instance, only to sentences of one’s own language and so not, for example, to sentences of a foreign language one does not speak.¹⁰ That might seem surprising. Even if one does not oneself know what the Polish sentence “Śnieg jest biały” means, for example, one might have thought that one could nonetheless understand what it means to say that the sentence “Śnieg jest biały” is true.¹¹ But, according to disquotationalism, that is an illusion. The reason is straightforward: If attributing ⁸ On some views, that-clauses refer to representations of some sort: sentences (Davidson 1968) or decorated trees (Larson and Ludlow 1993; Higginbotham 2006). ⁹ Which might, for all I’ve said, be deflationary, but it would be a deflationary notion of propositional truth. The arguments made here do not apply to deflationism of that sort, since propositions are not context-dependent. (That said, it’s potentially an interesting question whether deflationism is compatible with relativism about propositional truth.) ¹⁰ Shapiro (1998, 55ff.; 2003; 2005) argues that this restriction causes trouble. Field (2001b, 147–8) discusses the argument briefly. (I tend to sympathize with Shapiro but need not take a stand on the issue here.) ¹¹ Waiving, for the moment, context-dependence.

, , -

105

truth to “Śnieg jest biały” amounts simply to erasing the quotation marks, then “‘Śnieg jest biały’ is true” is just a verbose form of “Śnieg jest biały” itself. Since, by hypothesis, “Śnieg jest biały” is not a sentence one understands, one doesn’t understand “‘Śnieg jest biały’ is true” either.¹² In fact, however, all this argument immediately shows is that a disquotationalist cannot make do just with what Field calls a ‘pure’ disquotational notion of truth: one explained entirely in terms of disquotation. Rather, there is a need also for what Field calls an ‘extended’ disquotational truthpredicate, which is explained in terms of translation. A sentence I do not understand is true in the ‘extended’ sense just in case it can be translated by a sentence I do understand that is true in the ‘pure’ sense. One might worry that appealing to translation will bring in semantics through the back door. But the disquotationalist can insist, with W. V. O. Quine (1960, ch. 2), that the standards of correct translation do not have to be explained in terms of sameness of semantic content but can instead be explained in broadly pragmatic terms. There might then be no single ‘correct’ translation, but only a range of equally acceptable ones. That is not obviously a problem, however (Field 1994, 273).¹³ The extended notion of truth can also help the disquotationalist deal with context-dependence. For me to say that a specific utterance of “That woman is a great musician” is true is for me to say that it can be translated by a sentence of my language that is itself disquotationally true. Now, if what’s meant by “a sentence of my language” is a type sentence, then this does not help, since there are almost no type sentences to which it makes sense to apply the notion of truth. But there is an alternative: Take the truth-predicate to apply, in the first instance, only to utterances made by me at the present moment (Field 1994, 279–80; Heck 2004, sec. 4; David 2005, 389). Utterances of sentences like: (14)

That woman is a great musician iff the token sentence I just uttered is true.

¹² Essentially the same point applies to the idiolects of other people since, as Quine (1968, 199) famously put it, “. . . radical translation begins at home.” ¹³ Moore (2020) argues that there are more serious problems just around the corner, but I’ll not pursue that issue here.

106    will be true whenever they are uttered (assuming that, on that occasion, the utterance expresses a proposition). Such a notion has limited application, to put it mildly, but its utility can be extended by translation: If I say that some utterance U not being made by me now is true, then that means that there is some sentence of my language, a present utterance of which would adequately translate U, and that that utterance, if made by me now, would be true in the pure disquotational sense.¹⁴ That works as well for utterances of context-dependent sentences as it does for utterances of eternal ones. The ‘extended’ notion of truth that translation makes available thus offers the disquotationalist a solution to the problem posed by context-dependence. Field (1994, 280–1) makes a slightly different suggestion: that the translation be made not into natural language but into Mentalese, the language of thought. It is easy to see why such a view might seem attractive: The language of thought is arguably not context-dependent (Gross 2005), so one can apply the truth-predicate within the language of thought in a straightforwardly disquotational manner. An utterance made in natural language will then be true just in case it can be translated by some sentence of my version of Mentalese that is disquotationally true. And, again, that works as well for utterances of context-dependent sentences as it does for utterances of eternal sentences. Nothing in what follows will depend upon which of these views one prefers. My target is any view that appeals to translation to resolve the problem that context-dependence poses for disquotationalism. But these reflections do already show that another claim commonly made by disquotationalists needs to be reconsidered. Disquotationalists are fond of saying that we ‘acquire the truth-predicate’ by learning how to apply it to sentences we understand. That is supposed to be easy to do, since ⌜‘S’ is true⌝ is just equivalent to S.¹⁵ But children learning “true” understand few (if any) sentences to which they could sensibly apply “true,” so they certainly do not learn the word by applying it to (type)

¹⁴ All the “would”s here might give one pause, but let that pass. (Note, however, just how many things we have to let pass to get a view that is even worth discussing.) ¹⁵ See Kripke (1975, 701) for a discussion that’s often cited as inspiration. Whether Kripke actually held such a view is not so clear.

, , -

107

sentences. This is just the old point of Strawson’s, noted earlier, in a different guise.

4.3 Translation and context-dependence Imagine that we are in a crowded restaurant and that Maria gestures in the direction of a woman seated at the bar, saying: (15)

That woman is a great musician.

In order to focus attention squarely on the relevant issue, ignore the other sources of context-dependence besides the demonstrative, and assume that Maria is speaking the same language I do, so that there is no question what sentence I need to utter in order to translate her utterance, namely, itself. But any utterance of (15) by me will also be context-dependent, so I need to accompany my utterance by a ‘demonstration’ if I am to say anything definite.¹⁶ Indeed, given our simplifying assumptions, the only thing I need to decide, to figure out how to translate (15), is which person I should demonstrate. Suppose the right person is Yoko Miwa (who is, indeed, a great musician). Then a correct translation of (15) would be:¹⁷ (16)

That woman [said while demonstrating Yoko Miwa] is a great musician.

But why, to translate Maria’s utterance correctly, do I need to demonstrate Yoko Miwa rather than Diana Krall or Geri Allen? Because Maria was talking about Yoko Miwa. Had she been talking about Geri Allen, then I’d have needed to demonstrate her instead. But to say that Maria was talking about Yoko Miwa is just a different way of saying that she was referring to Yoko Miwa. So how I should translate Maria depends ¹⁶ In fact, demonstrations are optional (Mount 2008; Heck 2014; King 2014). But we can ignore such complications. ¹⁷ A different proposal is that (15) should be translated as “Yoko Miwa is a great musician.” The ensuing discussion would be different in detail but similar in spirit. See n. 18.

108    upon to whom she was referring. Conclusion: How we should translate a context-dependent utterance depends upon semantic facts about it; defining ‘extended’ disquotational truth in terms of correct translation thus does, after all, bring in semantics through the back door. The obvious objection is that this argument tendentiously assumes the legitimacy of such semantic notions as talking about. But the argument really just needs the following two premises: (REL) How I should translate an utterance of a sentence like (15)— in particular, whom I should demonstrate—is determined (in respect of the demonstrative) by some relation between the speaker and an object in the world: the one I need to demonstrate. (IND) The mentioned relation can be explained independently of translation. We’ll consider the status of these premises shortly. To see why they are sufficient, recall that the ‘pure’ disquotational notion of reference is supposed to be characterized by such trivialities as: (17)

“Yoko Miwa” refers to Yoko Miwa.

In fact, however, for reasons we have already seen, the ‘pure’ notion has nothing to say about utterances of demonstratives and other contextdependent expressions, such as (it would now seem) proper names. The disquotationalist thus requires an ‘extended’ notion of reference, call it Ereference. This notion, like the extended notion of truth, can be explained in terms of translation, thus:¹⁸

¹⁸ One might suggest that we should instead try: (*) Maria’s utterance of “That woman” E-refers to Yoko Miwa iff her utterance is correctly translated by “Yoko Miwa.” But, first, the left-to-right direction arguably fails, for sense–reference reasons, though that does depend upon how demanding the standards of ‘correct translation’ are. Second, if names are contextdependent, then (*) does not avoid the problems discussed in the text. Finally, the question will still arise what makes it the case that it is “Yoko Miwa” that correctly translates Maria’s utterance (rather than “Geri Allen”), and that still seems to depend upon there being some relation between Maria and Yoko Miwa herself, which is the premise we are currently discussing.

, , - (18)

109

Maria’s utterance of “that woman” E-refers to Yoko Miwa iff her utterance is correctly translated by an utterance made by me¹⁹ (while I demonstrate Yoko Miwa) of “that woman.”

The question here, to reiterate, is what facts determine which translation of Maria’s utterance is correct, which is to say: what facts determine whom I should demonstrate when translating Maria’s utterance. We obviously cannot just assume that those facts are ‘semantic.’ But—this is the first premise, (REL)—surely those facts must crucially involve some relation between Maria and Yoko Miwa: What else could possibly make it the case that I need to demonstrate Yoko Miwa when translating Maria? That claim does not beg the question against the disquotationalist by assuming that the relation in question must be semantic. There are plenty of non-semantic relations between speakers, utterances, and the rest of the world. Perhaps one of them will do the work the disquotationalist needs doing. If someone wanted to insist that the burden here is on the disquotationalist to tell us which non-semantic relation will do that work, then I wouldn’t disagree (though I’m allergic to burden-of-proof arguments). It really is very unclear what non-semantic relation might determine what the correct translation of Maria’s utterance is. (It is clear what semantic relation might determine it.) There was a time when one might have been forgiven for thinking it was pointing at (Quine 1968, 194), so that the right person for me to demonstrate is the person Maria was pointing at. But it has long been appreciated that pointing is not necessary for demonstrative reference.²⁰ Still, pointing at is a useful example, because it nicely illustrates the sort of relation the disquotationalist needs: a non-semantic relation that will do the work of a semantic one. The problem is that, if what determines how we should translate Maria is whom she was pointing at, then, by (18), Maria has E-referred ¹⁹ Better: by oneself (i.e., in your case, you), but let me speak of myself, for ease of exposition. We need to refer to an utterance of our own, recall, because we can only apply the pure disquotational notions of truth and reference to present utterances of our own. But nothing substantial would change if we spoke instead of utterances made in English. ²⁰ Indeed, pointing, in the relevant sense, probably isn’t a purely physical relation but in part an intentional one (Reimer 1991). The papers mentioned in n. 16 develop a stronger version of this point.

110    to the person she has pointed at; so E-reference reduces to pointing at. The same goes for any other non-semantic relation that might be offered instead. If we assume that this relation is explicable independently of translation—that is the second premise, (IND)—then, by the disquotationalist’s own lights, E-reference is reducible to some relation that is characterizable independently of translation. That would make E-reference a non-disquotational notion of reference (Field 1994, 281), one the disquotationalist has now been forced not just to recognize but to employ. Thus, (REL) and (IND) entail the falsity of disquotationalism, as was claimed above. There are several replies a disquotationalist might make. 1. Deny (REL). Thus, we find Field (1994, 279–81) arguing that, in explaining how to translate Maria’s utterance of “that woman”, we can make do just with facts about her “internal processing”: Whom I should demonstrate is determined by what other ‘mental files’ Maria associates with this particular utterance of “that woman.” But there are several difficulties with this view. The simplest is that there may not be any other such ‘files’: Maria may never before have seen nor heard of Yoko Miwa (Heck 2004, 339–41). 2. Concede both premises, so that E-reference is a non-disquotational notion of something, but deny that E-reference has very much to do with reference as friends of semantics would understand it (Field 1986, 89; 1994, 255). But it’s difficult to see how E-reference could come apart from reference: If Maria was talking about Yoko Miwa but E-referred to Geri Allen, of what possible significance could E-reference be?²¹ For present purposes, then, I shall set this view aside. We would need to know much more about what the relevant relation is supposed to be before we could discuss, let alone evaluate, such a view.²² ²¹ Of course, it is no help to the disquotationalist if E-reference corresponds to semantic reference and ‘talking about’ to speaker’s reference (Kripke 1977). ²² There is a suggestion that Field (2001a, 2001b) makes elsewhere, concerning ‘indication relations,’ that could probably be adapted to the present context. I discuss this proposal in other work (Heck 2022), however, and so will not discuss it here.

, , -

111

3. Deny (IND). There may be many relations between Maria and various objects in the world of which we will want to take note in deciding how to translate her words. But, this reply insists, correct translation cannot be reduced to such relations. All we can say, in the end, is that the correct translation is the one that allows us to make the best sense of Maria. Translation, on this view, is ineliminably pragmatic.²³ This last reply is the most interesting, and it needs much more discussion. Such views always seem to leave it frustratingly vague exactly what ‘making sense’ of someone involves. But waive that. What is distinctive of this reply is the central role it assigns to translation: What determines to whom Maria’s utterance of “that woman” refers is what the best way of translating her is. That, in turn, is determined by which way of translating her ‘makes best sense’ of her. A proponent of the third reply is thus defending this thesis:²⁴ (Ref Trans)

In uttering “That woman is a great musician,” Maria referred to Yoko Miwa (and said of her that she is a great musician) iff it would make the best sense of Maria for me to translate her by uttering “That woman is a great musician” while demonstrating Yoko Miwa.

In what follows, I aim to refute the claim that translation can, in this way, ground an extended disquotational notion of reference.²⁵ Given our discussion to this point, it might seem strange that the objection I am about to offer has nothing to do with context-dependence. The objection targets the role that is played in (RefTrans) by translation. It therefore also shows that disquotationalists cannot use translation to ²³ Nor does the disquotationalist have to follow Quine (1970a) in limiting the evidence for which translation is correct to Maria’s linguistic and other behavior. Making sense of her might require us to take into account how her brain works, the evolutionary history of her species, and who knows what else (see Chomsky 1969). ²⁴ To emphasize: This view denies (IND) because what follows the biconditional in (RefTrans) describes a relation between Maria and Yoko Miwa that essentially involves translation. ²⁵ As will be clear, the argument applies just as well to the alternative mentioned in n. 18.

112    explain extended notions of truth and reference that apply to sentences of other languages. But if the problem only arose when truth was attributed to sentences from other languages, it could be dismissed as marginal and deferred indefinitely. The case of context-dependence, on the other hand, is not marginal but fundamental, because contextdependence is the norm in natural language. So the case of contextdependence is the crucial one, even if the objection is more general. I should also emphasize, before we continue, that my objection targets only those views that take the standard of correct translation to be ineliminably pragmatic: ‘making best sense’ of other speakers, whatever that might mean. Views that regard correct translation as determined by preservation of independently specifiable features of utterances, or of relations in which those utterances (or their utterers) stand to the world, are immune to this objection. But those are the first and second replies above, and I have already argued that they are not viable. Only by making translation an unavoidably pragmatic enterprise can disquotationalists avoid the danger that when we describe the standards of acceptable translation for indexicals in detail we will have to bring in machinery that is powerful enough to provide a reduction of the semantic notion of reference to non-semantic terms . . . . (Field 1994, 281)

And that, of course, would be fatal to disquotationalism. So what is my objection? It is that no translation manual can ever allow one to make sense of a speaker’s words. Or, to put it more concretely: Even if I were in possession of a translation manual for Maria that was known to be correct—one that gave me, for each possible utterance she could make, a (non-context-dependent) sentence that I knew correctly translated it—that would not, by itself, help me to make sense of Maria’s words, not even a little bit. That may seem incredible. But this claim is, in fact, just a minor variant of one that Donald Davidson defends in “Radical Interpretation.” The central question that concerns Davidson in that paper is what knowledge I could have that would allow me to ‘interpret’ the utterances of another speaker. Before he introduces and defends his own answer,

, , -

113

Davidson (1973, 317) discusses one that is inspired by Quine: that it would be enough for me to know how to translate Maria’s words. Davidson argues that that is not enough. Someone might know that Maria’s utterances of “Śnieg jest biały” are correctly translated into Icelandic as “Snjór er hvítur” and yet not have the slightest idea what Maria’s words mean. The Icelandic translation will help only those who know, and make use of, information about what the Icelandic sentence means. If we put knowledge of how to translate into Icelandic together with knowledge of what the Icelandic sentence means, then that will yield the information that, when Maria utters “Śnieg jest biały,” she speaks truly iff snow is white. And that, says Davidson, is what I actually need to know to make sense of Maria. So the translation was “an unneeded intermediary” (Davidson 1973, 318). The problem with this argument is that it is not immediately clear how to extend it to the case in which the translation is made not into some language I do not understand, such as Icelandic, but into one I do understand: English. In that case, a correct translation manual would tell me that “Śnieg jest biały” should be translated as “Snow is white.” To be sure, if I did not understand English, that would not help me. But, since I do understand English, it does help me. And Davidson’s own account might seem no better off. His view is that I would be able to interpret Maria if I knew a theory of truth for her language, one that delivered such theorems as: (19)

When Maria utters “Śnieg jest biały,” she speaks truly iff snow is white.

As Davidson (1973, 317) himself says, however, “any theory is in some language.” And since this one is written in English, either I do not understand English, in which case (19) is no more helpful than (20)

Maria’s utterances of “Śnieg jest biały” are correctly translated into English as “Snow is white.”

is, or else I do understand English, in which case (19) is every bit as helpful as (20) is.

114    It is my contention that the apparent force of these considerations is due entirely to use–mention confusions. Once those have been cleared up, it will be clear why no translation manual can ever help us to ‘make sense’ of Maria. This issue has, as the mention of Davidson will have made clear, a long history. It will be worth our while, therefore, to consider it in a more general setting. We’ll do that in the next section and then return, in Section 4.5, to the question of how our discussion bears upon disquotationalism.

4.4 Translation and semantics It has often been suggested that translation might do the work that Davidson and others have thought we needed semantics to do. For example, Jerrold Katz and Paul Postal (1964) once argued that semantic theories should, in effect, translate sentences of natural language into a language composed of ‘semantic markers.’ In response, David Lewis (1970, 18–19) makes a point very similar to Davidson’s.²⁶ Jerry Fodor (1975, 119–22) then argues that Lewis must be wrong since language comprehension does not require us to know what our words mean but only how to translate them into Mentalese. In response, Ernie Lepore and Barry Loewer (1981) build on Davidson and Lewis. Later, however, Stephen Schiffer (1989, ch. 8) develops a view similar to Fodor’s. Apparently, then, more needs to be said. As said above, I contend that all forms of ‘translational semantics’ (to borrow from Lepore and Loewer) rest upon use–mention confusions.²⁷ But, since I personally find Fodor’s version of the view the most

²⁶ Davidson’s version of the argument, although published later, seems to have been independent. ²⁷ Here is Fodor (1975, 108): “A speaker is a mapping from messages onto wave forms, and a hearer is a mapping from wave forms onto messages.” Here is Lewis (1975, 3): “What is a language? Something which assigns meanings to certain strings of types of sounds or of marks.” So where is the disagreement? Lewis’s ‘meanings’ are propositions, but Fodor tells us that “messages are most plausibly construed as formulae in the language of thought” (Fodor 1975, 109, my emphasis). It’s use for Lewis and mention for Fodor. My goal in the text is to explain why it has to be use.

, , -

115

tempting, I’ll focus on it, ‘without loss of generality’. I.e., my criticisms of Fodor will plainly generalize. Here is a rough model of the ‘translational conception of comprehension,’ but one that will do for our purposes.²⁸ When someone utters a sentence, what I first do is determine how the sentence is composed of its constituent words. Perhaps this is quite complex; perhaps the ‘words’ are not at all what we’d ordinarily call “words” but are just morphemes. But once the syntactic analysis is complete, all that is left for me to do is to translate each ‘word’ that occurs in the original sentence by a corresponding expression of Mentalese, and then to put the Mentalese words together in a way that is determined by the syntactic analysis. When I’m done, I’ll have a Mentalese sentence that translates the original sentence of natural language, and then I can use the Mentalese sentence to interpret the present utterance of the natural language sentence. On the translational account, then, a central role is played in interpretation by competent speakers’ possession of a lookup table that pairs ‘words’ of natural language with corresponding ‘words’ of Mentalese. And, in many ways, that seems like a very natural idea.²⁹ But we need to be careful. There are several different ways to think of such a table and the information it contains. Consider, for example, Table 4.1. One way to read the table is as pairing nicknames of athletes with their given names, in which case the first line might be written more explicitly as: (21)

“Big Papi” is a nickname for the person whose given name is “David Ortiz.”

²⁸ This is essentially the model of semantic theory proposed by Katz and Fodor (1963, esp. sec. 7), minus various complications that exercise them a great deal but are irrelevant here (e.g., ambiguity). I believe this is also the view that Davidson (1967, 307–8) has in mind when discussing the proposal that a theory of meaning can consist just of syntax plus a dictionary. (Katz and Fodor speak in terms of a grammar and a dictionary, though Davidson does not cite them, or anyone else.) Davidson’s objection is that such a theory simply fails to address the problems that exercise semanticists (e.g., the semantics of belief sentences). He did not, I think, anticipate that someone might simply deny that those problems are of much interest, which seems to be how Field (1994, 269) would respond, and perhaps also Fodor. ²⁹ Were it not for the creativity and productivity of linguistic competence, we could make do with a phrase book: a table pairing sentences of English with sentences of Mentalese. The issues would be no different.

116   

Table 4.1 Nicknames of athletes Big Papi Queen V Baby Horse

! ! !

David Ortiz Venus Williams Alex Morgan

So, on this interpretation, the table purports to describe a relation between names (that is, between linguistic expressions). Fully to appreciate the content of the table, so understood, one does not have to understand any of the names contained in it. Even if you have never heard any of these names before, that will not affect your understanding of the table. A second construal would regard the table as pairing nicknames of athletes with those very athletes, that is, with the people who have those nicknames. In that case, the second line might be written more explicitly as: (22)

“Queen V” is a nickname for Venus Williams.

In this case, the table describes a relation between words and people. Fully to appreciate the content of the table understood this way, one does have to understand the names that occur in the right column, though not the names that occur in the left column: If one has never heard of Venus Williams, then one cannot understand (22). Yet a third construal (we’ll stop here) would regard the table as purporting to state some true identities, in which case the third line might be written more explicitly as: (23)

Baby Horse is Alex Morgan.

In this case, the relation the table describes is wholly worldly. Fully to appreciate the content of the table understood this way, one must understand the names that occur in both columns. It’s strangely puzzling, isn’t it? It’s not that it’s unclear what the three construals are. But how did you understand the table when you first you encountered it? In practice, the three construals tend to morph into one

, , -

117

another. Even if the table had been intended the second way, as pairing nicknames with people, someone who was completely unfamiliar with the names on the right could nonetheless read the table in accord with the first interpretation and thereby acquire information about co-reference. And even if the table had been intended the first way, someone who was familiar with the names on the right could nonetheless read it in accord with the second way and so extract information about the reference of the nicknames.³⁰ It just is very easy to slide back and forth between use and mention this way—as sure a sign of danger as ever there was—especially where written language is concerned: We simultaneously read the names with understanding and see them as objects in their own right. Now, how does all this bear upon the case of Fodor et al. versus Davidson et al.? What that dispute is really about is what information competent speakers have that allows them to interpret speech (Peacocke 1986):³¹ Is it information about what expressions refer to? Or is it just information about how those expressions should be translated? We are using Table 4.1 as a simple model of the sort of ‘lookup table’ that a speaker might use to interpret utterances of natural language sentences. So the question we need to ask is what information that table contains. But what we have just seen is that the answer to that question depends upon how we interpret the table. So the question becomes: On which construal of the table does it encode information possession of which would allow someone successfully to interpret utterances of the nicknames? The third construal is clearly no help. On this construal, the information contained in the table is simply a bunch of identities. Those identities have nothing to do with the names used to express them: They contain no information about those names. But information that does not even concern the nicknames cannot help one to interpret them. Any

³⁰ Note that this does depend upon the given names’ being presented (i.e., named) in a certain ‘canonical’ way. If “ ‘David Ortiz’ ” were replaced by “Bob’s favorite name,” then the conflation we are discussing would not be possible. ³¹ Strictly speaking, in Davidson’s case: What information someone could have that would allow them to interpret speech (Davidson 1984, 313).

118    construal of the table adequate for interpretation will thus have to be one on which the names in the left column are mentioned, not used.³² The first construal at least satisfies that condition, but it is no help either, for the reason already mentioned: One can know how to translate expressions from one language to another without understanding any of those expressions. The right way to state this point is: The information that a translation manual contains is insufficient to allow one to understand the language being translated.³³ It is true that, if one does understand one of the two languages, then one can parlay one’s knowledge of how to translate into knowledge sufficient to allow for interpretation.³⁴ But that is irrelevant. The question was what information enables interpretation, and what Lewis and Davidson are claiming is that information about how to translate one language into another is insufficient by itself. They are just right about that. It is when the table is interpreted the second way that it encodes information sufficient for interpretation: What you need to know is, for each nickname, who bears that name; that is what the table, so interpreted, tells you. To be sure, the table conveys that information by using people’s given names. But of course the table has to be written “in some language” (Davidson 1973, 317). That does not imply that the table contains (let alone only contains) information about the given names that are used in it (as on the first interpretation). On the contrary, on this construal, the table encodes a data structure that maps nicknames to people: the people who have those nicknames.

³² A different route to this point would begin with the observation that, when we are attempting to understand what someone else has said, we start by identifying the linguistic expression they have produced (Fodor 1975, 110–11). Certainly, this seems to be the model with which linguists operate. ³³ I owe this point to my former teacher Jim Higginbotham. He makes points in the same vicinity in Higginbotham (1988, sec. III). ³⁴ It is, actually, much less clear than is usually supposed how the transition from the mentioned sentence to the used one is to be made. In the ordinary case, one just ‘reads’ the quoted sentence. As noted above, however, this requires the mentioned sentence to be ‘presented’ in a way that makes it readable and not, again, as ‘Bob’s favorite sentence’. But, waiving that point, what is the analogue of ‘reading’ in the case of Mentalese? Even if that question can be answered, it is surprising that we are now supposing that language comprehension involves one’s forming, and then ‘reading’, names of sentences of Mentalese, since that is what a translation manual between English and Mentalese requires. (Special thanks here to Bill Warren.)

, , -

119

If one wants to respond that people cannot literally be contained in such a data structure, then one is succumbing to what appears to be an almost irresistible temptation to confuse what information is contained in a data structure with how that information is encoded.³⁵ As we are imagining it, the mapping from names to their bearers is encoded in a table-like structure that competent speakers have in their heads, one that is written in Mentalese (or whatever). So, of course, people are not contained in the table. They are mentioned in the table through the use of their names.³⁶ The same is true of the natural language names: They too are only mentioned in the table, through the use of their Mentalese names. Surely it has to be uncontroversial that we are, in some sense, able to think about other people.³⁷ Nor is a proponent of the translational conception in a position to deny that competent speakers are able to think about linguistic expressions, since such a capacity is presupposed by the ability to translate.³⁸ But then there cannot possibly be any obstacle in principle to our using whatever allows us to do those two things—think about people and think about their names—to construct a data structure that pairs expressions of natural language with people. If we can think about words and we can think about things, what possible barrier can there be to our thinking about relations between words and things? ³⁵ I find it helpful to think, in this connection, about strongly typed programming languages. If you want to write a program in C++ that will interpret decimal numerals, for example, then what you will need (at the level of digits) is a map, which is very different from a map (the first construal) or a map (the third one). Similarly, converting decimal numerals to integers is a very different programming problem than converting them to binary numerals, even if integers are represented, in the machine, in binary. ³⁶ This seems to be what Fodor overlooks. He writes: “Pretty obviously, there are computational procedures which map a representation of the acoustic properties of a speech event onto a representation of the message it encodes” (Fodor 1975, 117). But these representations are the medium in which the computation is performed. What is computed is a function from acoustic properties to ‘messages’, e.g., propositions. We’re not translating representations but mapping sounds to meanings. (Fodor’s own usage betrays him: He had previously insisted that messages are representations. See n. 27.) ³⁷ Granted, of course, the BIG issue is how that is possible, and it is an option to confine one’s disquotationalism to Mentalese and to make use of non-disquotational notions of truth and reference only in explaining how natural language utterances are mapped to ‘messages,’ as encoded in Mentalese. We’ll discuss the significance of this sort of option in Section 4.6. ³⁸ I take it to be less controversial than it might seem that language-users are able to think about expressions (cf. Soames 1992). Linguistic theory gives us good reason to suppose that this capacity is innate (see, e.g., Chomsky 1986), though, at the early stages, such thought might occur only sub-personally.

120    I suspect that part of what has made translational accounts seem attractive is a tendency to conflate two senses of “translation”. What philosophers usually have in mind when they speak of ‘translation’ is a relation between linguistic expressions, as on the first construal. That is how I have been using the term. But colloquial language is less strict. The third construal is reminiscent of what ‘translators’ at the United Nations do: They listen to speech in one language and then repeat what they have heard in another language. Understanding an utterance cannot consist in translating it in either of those senses: Translation in the philosophers’ sense does not (by itself) provide understanding; translation in the UN sense presupposes understanding. What we actually need is a process whose input is as on the first model (a sentence) and whose output is as on the third (its meaning). But that’s no longer translation, in any sense. It’s semantics.

4.5 Translation and ‘making sense’ Consider the (very partial) translation manual in Table 4.2. Someone who understands neither English nor Polish can only understand the table in accord with the first interpretation we discussed in the previous section: as mapping Polish sentences to English sentences. Such a person could not even begin to use this translation manual to ‘make sense’ of Maria’s speech. For those of us who do understand English, of course, matters are different: Now the translation manual tells us what Maria has said when she utters various Polish sentences, and we can ask whether it makes sense for Maria to have said what this phrasebook says she has said. But that, we can now see, is because our understanding of English allows us to understand the table in accord with the second interpretation discussed in the last section. On that construal, the table pairs Polish sentences not with English sentences but with ‘interpretations’ of them: It tells us what the Polish sentences mean. This is, once again, really a point about the information that is contained in a translation manual. That information, being wholly

, , -

121

Table 4.2 A translation manual Śnieg jest biały Śnieg jest zielony Trawa jest zielona

! ! !

Snow is white Snow is green Grass is green

meta-linguistic, is useless by itself if our goal is to ‘make sense’ of someone’s speech. What obscures this point is the fact that a statement like (20)

Maria’s utterances of “Śnieg jest biały” are correctly translated into English as “Snow is white.”

can convey, to English speakers,³⁹ more information than (20) actually contains, since we can combine that information with our understanding of “Snow is white” (and our knowledge of how translation works) to conclude that: (19)

When Maria utters “Śnieg jest biały,” she speaks truly iff snow is white.

But, and this is the crucial point, it is only semantic hypotheses like (19) that can be used to make sense of Maria, be it good sense or bad. Possession of a translation manual does not obviate the need to consider such semantic hypotheses. Moreover, there is simply no reason we cannot just consider (19) directly, bypassing the translation manual. That is what Davidson (1973, 318) means when he says that translation “is an unneeded intermediary”. The really crucial point, though, is that translation is not just unnecessary but also useless, except in so far as it can be used to generate such semantic hypotheses as (19). The semantic hypotheses are what do the actual work. In the specific case with which we were concerned earlier, the question was why I need to demonstrate Yoko Miwa when uttering “That woman ³⁹ Actually, since it is written, it will convey more information only to someone who can read English. Spoken versions will convey more information only to those who can understand spoken English. This illustrates, once again, the gap between the information (20) actually contains and what it conveys to certain people.

122    is a great musician” if I am to translate Maria correctly. I argued that this must be because Maria stands in some relation to Yoko Miwa herself—that’s the premise (REL). The question that remains is whether this relation is itself explicable independently of translation, as the other premise (IND) insists it must be. The objection we have been considering is that it is not: While various relations between Maria and objects in the world may well be relevant to how she should be translated, the best translation can only be chosen on pragmatic grounds, in terms of what ‘makes best sense’ of her; there is no translation-independent characterization of the relevant relation to be had. What the foregoing shows is, as promised earlier, that no translation of Maria makes any sense of her, good or bad, by itself. If you want to make sense of Maria’s utterance, then what you need is a hypothesis about who she was talking about. Since translation certainly is not needed to generate such a hypothesis, (IND) follows: The relation between Maria and an object in the world that determines how she should be translated does not have to be explained in terms of translation.⁴⁰ This does not, I should emphasize, imply that pragmatic factors do not play a crucial role in characterizing the relation between Maria and the object of her speech. Indeed, Davidson’s view, as is well known, is precisely that what makes a semantic hypothesis like (19) correct is that it (together with similar hypotheses) enables us to ‘make best sense’ of Maria. So, while Davidson would accept the letter of the argument I am giving here, he would, I suspect, have been somewhat unhappy with its spirit. On his view, the right thing to say is that there is no interpretation-independent characterization to be had of the relevant relation between Maria and an object in the world, and that the correct interpretation of Maria can only be chosen on pragmatic grounds.⁴¹ If so, then reference itself is an ineliminably pragmatic notion. We’ll consider the significance of this aspect of Davidson’s view in the final section. ⁴⁰ Of course, one could involve translation in the explanation. But one could involve translation in almost any explanation, and adding it to this one would not relieve us of the need to consider semantic hypotheses. ⁴¹ That is, on Davidson’s view, Maria has referred to Yoko Miwa iff a theory of truth incorporating that very hypothesis allows one to make best sense of her speech.

, , -

123

4.6 In closing, a caveat It’s been known for some time that context-dependence poses a problem for disquotationalism. But the problem seems to have been regarded as a fiddly technical one that will be resolved by the right kind of wizardry. I’ve argued here that the problem is both fundamental and one of principle. Moreover, attempts to invoke translation to explain how the notion of truth should be applied to utterances of context-dependent sentences founder on a dilemma. If ‘correct translation’ is determined by a relation that can be explained in terms independent of translation, then we have just grounded a non-disquotational notion of reference. If, on the other hand, we try to explain correct translation in terms of what would ‘make best sense’ of someone, then translation is the wrong sort of relation to do the job: It invokes mention where use is required. The arguments presented here cannot, however, show that semantic notions are robust enough to do serious scientific work. Since Field (1994, esp. sec. 2) often seems to think of that view—sometimes known as intentional realism (Fodor 1987)—as disquotationalism’s main opponent, that is a significant limitation. So let me explain why we must concede this point (and so why you can accept the arguments presented above even if you are not an intentional realist). Quine (1960, ch. 2) held that the way to investigate meaning is to investigate translation: If you want to know what Maria’s words mean, and what it is for them to mean that, then the right questions to ask are (i) how Maria’s utterances should be translated into your language, and (ii) what the standards of correct translation are. Quine articulates an answer to (ii) using the notion of radical translation. Davidson (1967), by contrast, thinks that we should ask (i′) what the correct theory of truth is for Maria’s language, and (ii′) what the appropriate standards are by which to adjudicate correctness. Davidson (1973) articulates an answer to (ii′) in terms of radical interpretation. So Quine and Davidson disagree about what form a ‘theory of meaning’ for Maria’s language should take. Quine thinks we can get by with translation, whereas Davidson thinks we must provide a theory of truth. That is why Davidson rejects deflationism (see, e.g., Davidson 1990, 1996). But if the question is what it is for Maria’s utterances to mean what they

124    do, then Quine and Davidson are quite close, as we see at (ii) and (ii′), and they are much closer to each other than either is to Fodor (1987, 1990) or to the early Field (1972, 1978). There are, of course, plenty of differences between Davidson and Quine here—most importantly, Davidson is no behaviorist—but Davidson’s notion of radical interpretation is not just named after but is explicitly modeled on Quine’s notion of radical translation. In particular, both have a significant pragmatic component. Now, despite Davidson’s occasional protests to the contrary, I agree with Michael Rescorla (2013, 480) that Davidson’s insistence that facts about content essentially depend upon pragmatic (and even normative) factors is inconsistent with intentional realism. And yet, so far as I can see, all of the arguments given above are compatible with Davidson’s general outlook. Indeed, I have borrowed heavily from Davidson’s arguments against Quine. So my reasons for rejecting disquotationalism are, in effect, just Davidson’s. So, unless Davidson’s views are inconsistent at this point, which I very much doubt they are, then the arguments given here cannot establish intentional realism. There is another lesson worth making explicit. It often seems to me as if disquotationalists think that, if interpretation (or translation) is essentially governed by pragmatic constraints, then that already undermines ‘inflationary’ views. And if ‘inflationism’ means ‘intentional realism’, then that may well be correct. But if it does not, then Davidson stands as a counterexample: What it is for Maria’s utterances of “Śnieg jest biały” to be true iff snow is white is for that to be the most fruitful way of ‘interpreting’ her. That may not be intentional realism, but it is not deflationism either, since non-disquotational notions of truth and reference figure crucially in the theory Davidson would have us use to make sense of Maria.

References Amour-Garb, B., and Beall, J., eds. 2005. Deflationary Truth. Chicago, Open Court. Bach, Kent. 2015. “The Predicate View of Proper Names.” Philosophy Compass 10: 772–84.

, , -

125

Carston, Robyn. 1988. “Implicature, Explicature, and Truth-Theoretic Semantics.” In Mental Representations: The Interface between Language and Reality, ed. Ruth M. Kempson, 155–81. Cambridge: Cambridge University Press. Chomsky, Noam. 1969. “Quine’s Empirical Assumptions.” In Words and Objections: Essays on the Philosophy of W. V. Quine, ed. D. Davidson and J. Hintikka, 53–68. Dordrecht: Reidel. Chomsky, Noam. 1986. Knowledge of Language: Its Origin, Nature, and Use. Westport, CT: Praeger. David, Marian. 2005. “Some T-Biconditionals.” In Amour-Garb and Beall (2005), 382–419. Davidson, Donald. 1967. “Truth and Meaning.” Synthese 17: 304–23. Davidson, Donald. 1968. “On Saying That.” Synthese 19: 130–46. Davidson, Donald. 1973. “Radical Interpretation.” Dialectica 27: 313–28. Davidson, Donald. 1984. Inquiries into Truth and Interpretation. Oxford: Clarendon Press. Davidson, Donald. 1990. “The Structure and Content of Truth.” Journal of Philosophy 87: 279–328. Davidson, Donald. 1996. “The Folly of Trying to Define Truth.” Journal of Philosophy 93: 263–78. Fara, Delia Graff. 2015. “Names Are Predicates.” Philosophical Review 124 (1): 59–117. Field, Hartry. 1972. “Tarski’s Theory of Truth.” Journal of Philosophy 69: 347–75. Field, Hartry. 1978. “Mental Representation.” Erkenntnis 13: 9–61. Field, Hartry. 1986. “The Deflationary Conception of Truth.” In Fact, Science, and Morality, ed. G. MacDonald and C. Wright, 55–117. Oxford: Blackwell. Field, Hartry. 1987. Psychosemantics. Cambridge, MA: MIT Press. Field, Hartry. 1990. A Theory of Content, and Other Essays. Cambridge, MA: MIT Press. Field, Hartry. 1994. “Deflationist Views of Meaning and Content.” Mind 103: 249–85. Field, Hartry. 2001a. “Attributions of Meaning and Content.” In Field (2001c), 157–74.

126    Field, Hartry. 2001b. “Postscript to ‘Deflationist Views of Meaning and Content’.” In Field (2001c), 141–56. Field, Hartry. 2001c. Truth and the Absence of Fact. Oxford: Clarendon Press. Field, Hartry. 2017. “Egocentric Content.” Noûs 51 (3), 521–46. Fodor, Jerry. 1975. The Language of Thought. Cambridge, MA: Harvard University Press. Fodor, Jerry. 1987. Psychosemantics. Cambridge, MA: MIT Press. Gray, Aidan. 2014. “Name-Bearing, Reference, and Circularity.” Philosophical Studies 171: 207–31. Gross, Steven. 2005. “Context-Sensitive Truth-Theoretic Accounts of Semantic Competence.” Mind and Language 20: 68–102. Grover, Dorothy, Joseph Camp, and Nuel Belnap. 1975. “The Prosentential Theory of Truth.” Philosophical Studies 27: 73–125. Gupta, Anil. 1993. “A Critique of Deflationism.” Philosophical Topics 21: 57–81. Heck, Richard Kimberly. 2004. “Truth and Disquotation.” Synthese 142: 317–52. Originally published under the name “Richard G. Heck, Jr”. Heck, Richard Kimberly. 2014. “Semantics and Context-Dependence: Towards a Strawsonian Account.” In Metasemantics: New Essays on the Foundations of Meaning, ed. A. Burgess and B. Sherman. Oxford: Oxford University Press, 327–64. Originally published under the name “Richard G. Heck, Jr”. Heck, Richard Kimberly. 2022. “Success, Failure, and the Classical Disquotational Strategy.” Manuscript. https://rkheck.frege.org/pdf/ unpublished/FailureArgument-Paper.pdf. Higginbotham, James. 1988. “Is Semantics Necessary?” Proceedings of the Aristotelian Society 87: 219–41. Higginbotham, James. 2006. “Sententialism: The Thesis That Complement Clauses Refer to Themselves.” Philosophical Issues 16: 101–19. Hill, Christopher. 2002. Thought and World: An Austere Portrayal of Truth, Reference, and Semantic Correspondence. New York: Cambridge University Press. Katz, Jerrold, and Jerry Fodor. 1963. “The Structure of a Semantic Theory.” Language 39: 170–210.

, , -

127

Katz, Jerrold J., and Paul M. Postal. 1964. An Integrated Theory of Linguistic Description. Cambridge, MA: MIT Press. Kennedy, Christopher, and Louise McNally. 2010. “Color, Context, and Compositionality.” Synthese 174: 79–98. King, Jeffrey C. 2014. “Speaker Intentions in Context.” Noûs 48: 219–37. Kripke, Saul. 1975. “Outline of a Theory of Truth.” Journal of Philosophy 72: 690–716. Kripke, Saul. 1977. “Speaker’s Reference and Semantic Reference.” Midwest Studies in Philosophy 2: 255–76. Larson, Richard, and Peter Ludlow. 1993. “Interpreted Logical Forms.” Synthese 95: 305–55. Lepore, Ernie, and Barry Loewer. 1981. “Translational Semantics.” Synthese 48: 121–33. Lewis, David. 1970. “General Semantics.” Synthese 22: 18–67. Lewis, David. 1975. “Languages and Language.” In Minnesota Studies in the Philosophy of Science Vol. 7, ed. Keith Gunderson. Minneapolis: University of Minnesota Press, 3–35. McGee, Vann. 2005. “Afterword: Trying (with Limited Success) to Demarcate the Disquotational–Correspondence Distinction.” In Amour-Garb and Beall (2005), 143–52. Moore, Graham Seth. 2020. “Theorizing about Truth Outside of One’s Own Language.” Philosophical Studies 177: 883–903. Mount, Allyson. 2008. “Intentions, Gestures, and Salience in Ordinary and Deferred Demonstrative Reference.” Mind and Language 23: 145–64. Peacocke, Christopher. 1986. “Explanation in Computational Psychology: Language, Perception, and Level 1.5.” Mind and Language 1: 101–23. Quine, W. V. O. 1960. Word and Object. Cambridge, MA: MIT Press. Quine, W. V. O. 1968. “Ontological Relativity.” Journal of Philosophy 65: 185–212. Quine, W. V. O. 1970a. “Methodological Reflections on Current Linguistic Theory.” Synthese 21: 386–98. Quine, W. V. O. 1970b. Philosophy of Logic. Englewood Cliffs, NJ: Prentice Hall. Reimer, Marga. 1991. “Demonstratives, Demonstrations, and Demonstrata.” Philosophical Studies 63: 187–202.

128    Reimer, Marga. 2002. “Do Adjectives Conform to Compositionality?” Philosophical Perspectives 16: 183–98. Rescorla, Michael. 2013. “Rationality as a Constitutive Ideal.” In A Companion to Donald Davidson, ed. Ernie Lepore and Kirk Ludwig, 472–88. Oxford: Wiley-Blackwell. Schiffer, Stephen. 1989. Remnants of Meaning. Oxford: Oxford University Press. Shapiro, Stewart. 1998. “Proof and Truth: Through Thick and Thin.” Journal of Philosophy 95: 493–521. Shapiro, Stewart. 2003. “The Guru, the Logician, and the Deflationist: Truth and Logical Consequence.” Noûs 37: 113–32. Shapiro, Stewart. 2005. “Gurus, Logical Consequence, and Truth-Bearers: What Is It That Is True?” In Amour-Garb and Beall (2005), 153–70. Soames, Scott. 1992. “Truth, Meaning, and Understanding.” Philosophical Studies 65: 17–35. Stanley, Jason, and Zoltán Gendler Szabó. 2000. “On Quantifier Domain Restriction.” Mind and Language 15: 219–61. Strawson, P. F. 1950. “Truth.” Proceedings of the Aristotelian Society. Suppl. Vol. 24: 129–56. Szabó, Zoltán Gendler. 2001. “Adjectives in Context.” In Perspectives on Semantics, Pragmatics, and Discourse, ed. R. M. Harnish and I. Kenesei, 119–46. Amsterdam: John Benjamins.

5 The Place of the Philosophy of Language in Metaphysics Thomas Hofweber

5.1 The language–metaphysics gap Although the philosophy of language had a special and distinguished place among philosophical disciplines for a good while since the rise of analytic philosophy, this central role has to a large extent been lost in the meantime. This is particularly true for the role of the philosophy of language in metaphysics. Considerations from the philosophy of language play no distinguished role in metaphysics in the contemporary debate, and apparently for good reason. The philosophy of language concerns language, naturally, and what we do with our languages. It is concerned with such problems as how we represent the world linguistically, what the basic structure of our languages is, how we use these languages in communication, and so on. Metaphysics, on the other hand, concerns what reality is like. It aims to find out about the most general features of reality. In particular, metaphysics concerns reality as it is, not merely as how we represent it to be. Thus there is little work for the philosophy of language in metaphysics. The philosophy of language is focused on us and our representations. Metaphysics is focused on reality in general instead. How we represent the world is by itself no guide to how reality is, since our representations can be mistaken or even utterly inadequate for reality. How we represent reality might thus be informative about us and our minds, but besides that it alone does not tell us anything about reality in general. The philosophy of language is thus very informative about our representations of reality, but not directly about Thomas Hofweber, The Place of the Philosophy of Language in Metaphysics In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Thomas Hofweber 2023. DOI: 10.1093/oso/9780198892724.003.0005

130   reality. Therefore the philosophy of language has no special place in metaphysics, and it is no wonder that it indeed plays no central role in contemporary metaphysics. This simple thought, which we will look at more closely shortly, has a number of consequences, both for metaphysics, but also for philosophy in general and how different parts of philosophy relate to each other. For metaphysics it means that however we are supposed to make progress in it, it won’t come from thinking about our own way of representing the world. That is in a sense too bad, since our representations of the world are reasonably accessible to us. It is no miracle how we can find out all kinds of insightful things about how we represent reality in thought or language. We have numerous sources of evidence available to us that support proposals about which sentences are grammatical, which ones are meaningful, and even which ones follow from which other ones.¹ But thinking about our representations of reality is not a direct way of determining which representations are true or false, with the possible exception of extreme cases like analytic truths or analytic falsehoods. The philosophy of language does not directly concern which sentences are true, that is to be found out elsewhere in inquiry. It only concerns other aspects of sentences: what they mean, what we do with them when we assert them, what their grammar is, and so on. None of that is a guide to what reality in general is like. To find out about that we need to pursue other avenues. But which ones? Much of contemporary metaphysics is quite divided about how such progress can happen. Among the options are moving closer to the sciences and piggybacking on their discoveries. We could then aim to derive answers to metaphysical questions from the discoveries of the empirical sciences.² Or maybe metaphysics has to make decisions on the basis of theoretical virtues alone, considering simply which metaphysical theories are simplest, most parsimonious, and so on.³ Or maybe metaphysics does not answer question of fact at all, but instead merely concerns the construction of models of what is believed to be the facts,⁴ or with the selection of useful concepts,⁵ or

¹ See Ludlow (2011) for more on evidence in linguistics. ² See, for example, Ladyman and Ross (2007). ³ See Sider (2013). ⁴ See Godfrey-Smith (2006). ⁵ See Carnap (1956) and Thomasson (2016).

       

131

the repair of flawed concepts,⁶ and so on. But all of these proposals in one way or another diminish metaphysics by either taking it to be merely derivative on the sciences, or to be much more speculative and thinly supported than other parts of inquiry, or, worst case, not to be a proper part of inquiry at all, but simply focused on our own flawed ways to represent the world. None is quite what metaphysics aspired to be in the eyes of many practitioners. All this so far assumes that reality is just there, independently of our language. But maybe that is false. Maybe there is some kind of a dependence on some or all of reality on our representations, and because of this it is possible after all to derive conclusions about reality from thinking merely about how we represent reality. Examples of such attempts include Simon Blackburn’s (1984) quasi-realism, which is intended to apply to specific subset of all the truths and all the facts, for example the moral truths. More radial is Richard Gaskin’s (2020) linguistic idealism, which takes all facts to be derivative on language. But for either one of these proposals, it is difficult to make sense of the kind of dependence that certain truths have on language. Gaskin relies on a primitive notion of ‘transcendental precipitate’ to formulate his claim that the world is “. . . a transcendental precipitate of language” (Gaskin 2020, 197). Although Gaskin says much more about the linguistic idealism he defends, the notion of a ‘transcendental precipitate’ is left as a primitive notion, and with that it is hard to see what the view comes to, not to speak of how we should evaluate the reasons given in its favor. Not unrelatedly, it is notoriously difficult to spell out the quasi-realist position and to distinguish it from straightforward realism, where there is no sense of dependence of morality on language or on what speakers do with their assertions of moral language.⁷ Thus few have found much promise in the proposal that facts depend on our language, either in general or at least in large parts. Although some facts clearly do depend on us and our language, these are paradigmatically facts about us and our language, like the fact that learning French is hard. But that there is a sense of dependence in which a large part of all the facts depend on us and our language in this sense of dependence is something that is widely ⁶ See Scharp (2018).

⁷ See Dreier (2004).

132   rejected, and for good reason. And this then seems to seal the hope for making progress in metaphysics by simply thinking about our representations of the world: which facts obtain is in general independent of us and our representations, and so thinking about our representations is informative about us and what we do, but not directly about the world in general. Since metaphysics aims at the world in general, it should look elsewhere than at language, and so the philosophy of language is not of significance for metaphysics, or so says the widely held argument. And this is reflected in contemporary metaphysics, which gets separated more and more from the philosophy of language. To put a label on it, we can call the language–metaphysics gap the alleged gap between facts about our language alone and facts about what reality in general is like. In particular, the alleged gap is supposed to block drawing conclusions about what reality in general is like from our language alone.⁸ Although officially the gap is described as concerning language, it should be understood as concerning our representations in general, be they mental or linguistic. We cannot draw conclusions about what reality is like from our representations alone, be they mental or linguistic. I will in the following sometimes talk about concepts, which are in the ballpark of mental representations, and sometimes about words, phrases, or expressions, which belong to linguistic representations. But our general topic concerns both equally. I should make clear and explicit that what is at issue is what conclusions we can draw from our representations alone. Uncontroversially, we can draw all kinds of conclusions about reality from facts about which ones of our sentences are true. The issue at hand is, instead, what conclusions about reality we can draw from our representations by themselves, with no regard to whether these representations represent accurately or truly. There is no obstacle in drawing conclusions about what reality is like from which representations are true. But there does seem to be a gap between facts about our representations alone and what reality is like. ‘Our representations alone’ here is to be understood as what our representations are like by themselves and what we aim to do with them, but leaving aside whether or not we succeed in what we aim ⁸ Heather Dyke calls a similar principle ‘the representational fallacy’ in Dyke (2008).

       

133

to do. And so understood, the gap seems to be legitimate: Simply from our attempts to represent reality one way or another we can draw no substantial conclusions about reality. The addition ‘substantial’ is important here. Since we are part of reality, there are some obvious conclusions about reality tied to us and to what we aim to do that can be drawn from our representations alone. But these would only concern a certain local part of reality, the part occupied by us and our minds. That is clearly legitimate, but also rather limited. Metaphysics is not focused on us and our minds, but on reality in general. Furthermore, ‘substantial’ is intended to demand more than merely analytic truths, if there are such truths. If some truths are analytic, then we might well be able to conclude such a truth from considerations about language alone. But even if there are such truths, few think that the debates of metaphysics are to be settled by analytic truths. Metaphysics concerns synthetic hypotheses about what reality in general is like.⁹ And so understood there is no route from the philosophy of language to metaphysics, or so it seems. Even if all this is correct, it wouldn’t mean that the philosophy of language has no role to play in metaphysics. It can certainly still be helpful, even if it doesn’t directly lead to any metaphysical insights about what reality in general is like. We should distinguish a corrective from a constructive role for the philosophy of language in metaphysics. Uncontroversially, I take it, the philosophy of language can help us avoid certain mistakes in metaphysics. For example, the philosopher of language can point out that a particular metaphysician is making a certain error in their reasoning tied to an unacknowledged scope ambiguity in one of their arguments. To point this out, and thereby to help the metaphysician avoid this error, is a real contribution to metaphysics. But it is only a corrective contribution, in the sense that it can help correct errors. The metaphysician relies on language in their theorizing and reasoning, and here insights into language can be helpful in avoiding errors. What is at issue in this paper is whether the philosophy of language can also make a constructive contribution to metaphysics: whether it can make a positive contribution to establishing what reality ⁹ Obviously, there are exceptions to this picture, among them Frege (in Frege 1884) or Frank Jackson (in Jackson 2000).

134   is like in general, not only a negative one by showing how certain metaphysicians made a mistake in their reasoning tied to language. Such a constructive, positive contribution from the philosophy of language, however, seems to be ruled out by the language–metaphysics gap. The philosophy of language here appears to be on a par with a standard view of the role of epistemology in metaphysics. Epistemology can also have a corrective role: the epistemologist can point out that a particular proposal in metaphysics is badly supported by the cited evidence. But few would hold that epistemology can make a constructive contribution to metaphysics: that we can somehow conclude what reality in general is like from considerations tied to entitlement, warrant, evidence, and so on. Epistemology and the philosophy of language are thus focused on the theorist and their reliance of representations, evidence, and entitlement. That is important, but it falls short of a constructive contribution to metaphysics. Of course, there are exceptions to this general picture. One could try to argue, for example, that a certain epistemic or linguistic fact wouldn’t be possible unless a certain metaphysical fact obtains, and via such a connection one could try to argue that this metaphysical fact has to obtain, since the epistemic or linguistic fact obtains. This would make a constructive contribution to metaphysics via a connection to epistemology and the philosophy of language. But these would be unusual arguments in contemporary metaphysics. Most metaphysical theorizing does not proceed along those lines, and many would likely take such arguments to sow more doubt about the epistemic or linguistic facts than to establish the metaphysical facts via the claimed connection. Philosophy is never completely uniform, but the general point remains that it is generally seen as hopeless to make substantial progress about metaphysical issues via a constructive contribution from epistemology or the philosophy of language. This picture thus leads to a certain separation of philosophical disciplines. Each of metaphysics, epistemology, and the philosophy of language can have a corrective role in the other ones, but none seems to have a constructive role in the other ones. The metaphysician can point out that the philosopher of language is making a certain metaphysical mistake, say by confusing ontological categories in their semantic theorizing. But

       

135

that alone doesn’t make a positive contribution to how language in fact works, only a corrective one: how a certain way of conceiving of language is flawed. This separation of disciplines in turn affects much of how philosophical theorizing proceeds, and where these various philosophical subfields can turn to for insights related to their main questions. The philosophy of language obviously can turn to linguistics for insights, but it seems to have little to gain from connecting with metaphysics or epistemology. In this paper I argue that all this is a mistake. The language– metaphysics gap is illusionary, and there indeed is a constructive role for both the philosophy of language as well as epistemology in metaphysics. Consequently these three disciplines are much more closely connected than is appreciated in the contemporary debate. In this paper I hope to make the case for this, in particular for the philosophy of language, but I will also outline how it is related to the constructive role of epistemology in metaphysics. To start, we should see how one can make a constructive contribution to metaphysics from the philosophy of language, and from thinking about our language alone. Once this has been motivated with an example we will be in a position to see how and why the language–metaphysics gap can be bridged.

5.2 Bridging the gap One could try to bridge the gap by first defending a particular philosophical theory, say a form of linguistic idealism which holds that reality is derivative of language somehow, and then rely on that theory to draw conclusions about reality from considerations about language alone. But any such attempt is only as good as the defense of the philosophical theory on which it relies. Not that it is obvious that this could not be done somehow, but I hope to make clear in this section that no such philosophical theory is required to bridge the language–metaphysics gap. The gap can be bridged straightforwardly, without relying on any substantial assumptions or any substantial philosophical theories. We can see straightforwardly that we sometimes indeed can draw significant, synthetic conclusions about reality merely by reflecting on our language

136   alone. In this section I will give an example of this and then explain why and how such an example can work in light of the considerations given above that motivated that there is such a gap. Let us consider the question whether or not natural numbers exist. This is a question about reality and what it contains. It is a question about what exists, not a question about language. And it is a question traditionally considered to be part of metaphysics, in particular of ontology. Nonetheless, I will argue, this question can be answered by considerations merely about our language alone. We can find an answer to this question about the world simply from thinking about what we do when we use language and what our language alone is like. Here is how this can go. There is an old puzzle about the occurrence of number words in natural language, one that goes back to at least Frege (1884). In languages like English number words can appear in at least two quite different syntactic positions. On the one hand, they can appear like adjectives, determiners, or modifiers, as in (1)

Jupiter has four moons.

Here ‘four’ seems to be very similar to ‘green’ in ‘Jupiter has green moons.’ and thus like an adjective. On the other hand, number words can appear like names, as in (2)

The number of moons of Jupiter is four.

Here ‘four’ appears to be a name, just like ‘Wagner’ in ‘The composer of Tannhäuser is Wagner’. These two syntactic occurrences normally come with two very different semantic functions. Names aim to refer to objects, but adjectives or determiners do not refer to objects, nor do they aim to do so; they modify nouns instead. The question is how number words can have both of these semantic functions in different occurrences, as well as why they can occupy these different syntactic positions. In essence there are two main lines to pursue, although they are not the only options. One is that number words are ultimately like adjectives, even when they sometimes appear to be names. On this line

       

137

one will need to explain why number words appear to be like names in sentences like (2). There should be some syntactic explanation of why an adjective or determiner occupies this syntactic position that seems to be contrary to its basic syntactic category. Alternatively, one could argue that number words are ultimately names, even though they appear, for some reason or other, like adjectives in sentences like (1).¹⁰ One option gives priority in some sense to number words as names, the other gives priority to number words as adjectives or determiners. Both of these options need to be spelled out more carefully, of course, including in what sense there is priority of one syntactic occurrence over the other. Maybe priority is to be understood as revealing how number words are in the lexicon, whereby the derivate occurrence of them is somehow understood as some syntactic derivation of a sentence where a number word appears in a position contrary to its lexical entry. We do not have to determine now how these options are best to be spelled out, and in principle all of them should be on the table. Suppose then, just for the moment, that a fairly strict version of the ‘adjectives first’ line is correct: Number words are primarily like adjectives or determiners, but they appear syntactically like names for some syntactic reason or other. On this line, number words are syntactically displaced when they appear in sentences like (2). Which is to say that there is some syntactic reason why the number word appears like a name. That reason is a syntactic one: it primarily comes from syntax. Suppose in addition that the syntactic reason for the displacement of number words does not give them a completely new semantic function when they are so displaced. That this is so can be derivative of what the syntactic reason for displacement is. In particular, since adjectives do not aim to refer to objects, number words, when syntactically displaced, still do not aim to refer, even though they appear to be names on a first look. They do not acquire a new semantic function simply by being syntactically displaced. Although determiners do not refer, they nonetheless have semantic values in a compositional semantic theory, and in a sense they can be seen as “denoting” those semantic values. This talk of denotation applies to any phrase, since all phrases will have semantic values, but it does not ¹⁰ See Dummett (1991, 99ff.), where Michael Dummett discusses these two options.

138   mean that all phrases are referential. Reference and “denotation” in this sense need to be distinguished. Every phrase has a semantic value, but they do not thereby refer to that semantic value. Furthermore, even referential phrases do not necessarily have their referents as their semantic values. In Montague’s treatment of proper names such names have sets of properties as their semantic value, but they refer to people, not sets of properties. And furthermore, although every phrase has a semantic value, a sentence is not simply a sequence of referring expressions, each picking out some entity. On this picture it would hard to make sense of how such a sequence of referring expressions would give rise to truth conditions and propositional content. Instead, different phrases have different semantic functions: reference, predication, modification, etc., even though every phrase has a semantic value of some kind or other. To say that adjectives or determiners do not aim to refer does not involve claiming that they do not have semantic values, which they can be said to “denote”, in this technical sense. On this option of understanding what is going on with number words, they are syntactically displaced determiners or adjectives, and they do not have the function of referring even when they occur in a position that makes them appear to be a name. Whether or not this is indeed correct is a difficult question, of course, but for anyone it should be a live option. It is something that we could discover by investigating our own natural language and why it has the features that it has. What’s more, I believe the position outlined is indeed the correct one, and I have argued for it in some more detail in Hofweber (2005), (2007), and, in particular, (2016). There I argued that there are at least two reasons why number words syntactically appear in singular term position, even though they are ultimately determiners. One of them is to achieve a focus effect, something we can see by observing how (1) and (2) interact differently with questions. The other is tied to a cognitive advantage in learning arithmetic. In the cited texts I also argued that this view generalizes to number terms more broadly understood, including expressions like ‘the number four’.¹¹ All ¹¹ One way this could be is that ‘the number’ is in apposition to the bare number word ‘four’. If ‘four’ is a non-referential displaced adjective or determiner in singular term position, then ‘the number four’ would derivatively also be non-referential, since an apposition does not turn a non-referential phrase into a referential one. For more, see Hofweber (2016, ch. 5).

       

139

this would support that number terms in general are non-referential. And this we could figure out simply by thinking about our own language. Whether or not this view is in the end correct is, of course, controversial and much beyond the scope of this paper. What matters for us here are two things: first, that this is a live option in the debate about how to understand number words in natural language, second, that we could discover that this is the correct option by thinking about our language alone. In particular, we can discover by thinking about our language alone that number terms are non-referential. To make this more explicit, let’s distinguish two ways in which a phrase can be non-referential. One is paradigmatically exhibited by an empty name, like ‘Betty Crocker’.¹² It is a name, and thus has the semantic function of reference: it aims to refer. But there is no such thing that the name refers to and so its semantic function goes unfulfilled. It aims to refer, but fails in its aim, and thus it is non-referential. We can call such names de facto nonreferential. But there is also a second sense of being non-referential, which we can call being constitutively non-referential. It applies to expressions that do something completely different than referring semantically. They do not aim to refer, since they are not even in the business of reference. On most people’s views, most expressions in natural language are like this: ‘very’, ‘many’, ‘the’, ‘green’, and so on are all constitutively non-referential. They have semantic values, and can be said to “denote” these semantic values, but they do not aim to refer to these semantic values nor to anything else. Now, if number words are primarily adjectives or determiners, and if the explanation of why they nonetheless appear in the syntactic position of a name is a purely syntactic one, then number words, too, are constitutively non-referential. And if this view generalizes to number terms more broadly, then those, too, are constitutively non-referential. And we can see that they are constitutively non-referential from considerations about language alone. All the relevant questions tied to this are just about our language, not about the world in general. They are questions about why number words have these different syntactic ¹² ‘Betty Crocker’ is the name of a brand of food products, also used for a non-existent person who answers consumer questions and suggests recipes in advertisements.

140   occurrences, what explains why number words can sometimes appear like determiners and sometimes like names, and so on. These are questions just about language alone. Similarly, whether an expression is nonreferential in the second, constitutive sense, the sense in which adjectives and determiners are non-referential, is also just about language alone. It simply concerns what the semantic function of expressions of this type is, and whether they are even in the business of reference. This contrasts with the question whether expressions are non-referential in the first, de facto, sense, which is not about language alone. An empty name is only empty because two things happen: the name tries to refer, but the world does not cooperate and it thus fails to do what it tries to do. That concerns both language and the world, not just language alone. But that ‘very’ is non-referential can be seen by only thinking about language alone and what the semantic function of intensifiers like it is: They modify other expressions, but are not in the business of reference. Taking all these points together we can now see that we can answer the worldly question about the existence of natural numbers simply from considerations about our own language alone. The key aspect of seeing this is to appreciate that if number terms are non-referential, then natural numbers do not exist. To see this, let’s assume that number terms are constitutively non-referential. Consider all the things that exist and pick one of them out at random, say object o. Could it be that o is the number 4? Now, since number terms are non-referential in the very language I use here, that means that ‘the number 4’ in the last sentence is non-referential, especially non-referential in the constitutive sense. It does something other than refer, even though it can appear, for some reason or other, in the syntactic position of a name or a singular term. But when I ask whether o is the number 4, then I ask about the identity of the number 4 and some object o. But since ‘the number 4’ is non-referential, it in particular does not refer to o. And thus the identity statement that o is the number 4 can’t be true. Thus whichever object I pick, it isn’t the number 4. Thus no thing or object is the number 4, and therefore the number 4 does not exist. And the same now applies for all the other numbers as well. Since we can find out that number terms are non-referential in the constitutive sense by thinking about our own language alone, we can thus find out that none of the natural numbers

       

141

exist: not the number 1, not the number 2, and so on.¹³ We can draw this conclusion simply by thinking about our own language. This bridges the language–metaphysics gap. It gives us the result that natural numbers do not exist, something which is fully about reality and not at all about our language, from insights that are about our own language alone. But how could this possibly work in light of what we have seen above? How could thinking about our language alone allow us to conclude something concerning the existence of numbers? How can that be, given that facts about language alone concern only us and our representations, whereas facts about numbers concern something totally different, not tied to language at all? The key to seeing how this is possible is to focus on the question we originally asked. We started our inquiry into reality, in particular the reality of numbers, by asking the question whether natural numbers exist. That question was stated in our language, naturally, and as such it involved our representations. We probe reality with those representations by asking a question about it, a question which determines the goalpost for that part of inquiry. We want to find the answer to that question, and with that we want to find the answer to something expressed with our representations. And that opens up the possibility that we can find out the answer to that question by reflecting on the representations used in the question itself. And this can be so even if the answer is not an analytic or conceptual truth, as it is not in our case. It is not a conceptual truth that natural numbers don’t exist, but nonetheless a truth that can be discovered by thinking about our own representations. That number words are constitutively non-referential has to be established via linguistic investigation. It is not itself a conceptual truth that they are constitutively non-referential, but it is something that can in principle be established simply by thinking about our own language, assuming that it is indeed the correct view. As acknowledged above, I can’t hope to defend this here, but I have done my best to do so in other work cited above. The point is that those conclusions were argued for purely by reflecting on our own language, with no regard to whether or not particular sentences in that language are true or correspond to ¹³ This argument against the existence of natural numbers is discussed in more detail in Hofweber (2016, ch. 4).

142   reality. And if these arguments are indeed successful, as I take them to be, then this would establish that numbers do not exist simply from considerations about our language alone. Thus such arguments are possible, we can bridge this gap, and by focusing on the question we originally asked we can see why and how this can be. This supports that the philosophy of language has a constructive, not just a corrective, role in metaphysics. There are some questions of fact, questions traditionally thought to be part of metaphysics, that can be answered simply by thinking about our own representations alone. Thus thinking about our own representations alone can make a constructive contribution to our project of finding out what reality in general is like. It does not merely prevent us from faulty reasoning or making mistakes in metaphysics; it can directly answer a substantial question about what reality is like—for example, the question ‘do natural numbers exist?’ I should add that all this works quite straightforwardly, without relying on a special philosophical theory like linguistic idealism or a special view of metaphysics. It works within mainstream metaphysics and mainstream philosophy of language. This, then, is my first main conclusion: we can sometimes answer questions about reality simply by reflecting on our own representations alone. And this points towards a constructive role of the philosophy of language in metaphysics. I say ‘points’ because it is not completely clear if this would truly establish such a constructive role, even if everything I said so far is perfectly correct. There are two concerns left that can result in some caution about there being a constructive role of the philosophy of language in metaphysics. The second half of this paper aims to present these concerns and at least outline some key ideas about how they can be overcome. To see how all this goes, let’s first consider the concerns that remain and then see how these concerns can be overcome. To appreciate these concerns, we can think about why what we have done so far might not be enough. We were able to bridge the language– metaphysics gap, since we focused on the question itself. We asked a question using our own language, and by thinking about the language we used we were able to answer the question we asked. But did we ask the right question? Or did we ask a bad, or even terrible, question? Maybe

       

143

we were only able to answer the question we asked because the question we asked was defective somehow, and that defect was the reason why the answer could be determined by thinking about the language used in the question. But answering a defective question is not much reason to declare victory in metaphysics. Metaphysics does not just concern any old defective question; it at least concerns proper questions. And that is the first reservation: even if we answered the question we asked, the issue remains whether we asked the right question, and with it if we made real progress in metaphysics. The second reservation is in a sense more general than the first: It is one thing to wonder whether we answered the right question, but quite another why we should give any special place to our questions at all. Why should we think that when we ask questions in our language, then this sets the proper target for what metaphysics has to achieve? Maybe our own language is so inadequate for metaphysics that none of our questions probe reality in the right way? Maybe the true target of metaphysics is not to answer one of our questions, but something quite different. We will need to discuss these two reservations in more detail before we can draw a conclusion about the place of the philosophy of language in metaphysics.

5.3 Evaluating the question Suppose that the position about number words and number terms outlined above is indeed correct for our natural language. And suppose further that this indeed allows us to answer the question whether natural numbers exist. Have we thereby made progress in metaphysics? This can be understood in an innocent way, where the answer is clearly yes, and in a more demand sense, in which the answer is unclear. Clearly we made progress in the sense that we found the answer to a question we had asked, a question that many considered to belong to metaphysics.¹⁴ But ¹⁴ There is, of course, another issue here concerning why this question should be seen as belonging to metaphysics, and not mathematics. A more detailed discussion of this issue is to be found in many places, including Carnap (1956), Schaffer (2009), and many more. My own take on this issue is in Hofweber (2016, ch. 1).

144   did we make real progress in finding out what reality is like? Here there are two worries: One is that the result achieved is purely negative— numbers do not exist—which leaves open what does exist. In particular, it leaves open whether there are other things, things that are not numbers, but that are a lot like what some philosophers have thought numbers are: certain sets, or positions in an ω-sequence, or what have you. That there are no natural numbers does not rule out that there are those other things, although it does rule out that those other things are natural numbers even if they do exist. But this should not be the real worry to be hung up about. It might well be that the non-existence of numbers has lots of positive consequences for what reality is like, and establishing that they do not exist is only a first step in a more positive story of what reality is like. For example, it could be that the reason why we can see that natural numbers don’t exist, the one coming from our own language, also gives us lots of positive insights into the philosophy of mathematics.¹⁵ The real worry should instead be this: we were able to establish, under the assumptions made, that the question ‘Do natural numbers exist?’ has a negative answer. The non-referentiality of number terms guarantees this. So we answered the question we had asked, but did we ask a good question to begin with? One concern here is that we might only have been able to answer the question as asked, since the relevant concepts or expressions involved in the question are defective in some sense. Maybe the concepts or expressions relied upon in it, in particular that of a natural number, are rather unsuitable for metaphysics. It might be that ‘natural number’ or ‘the number four’ is defective in some way, maybe tied to its being non-referential or maybe tied to something else. And it might be that because of this defect that we were able to answer a question that involved this expression. It was fine to answer the question we asked, but it would have been better to also know that we asked a good question, one properly suitable for metaphysics, involving more proper concepts. To put a label on it, we can call a metaphysically shallow result the correct answer to a question often considered to be part of ¹⁵ I have developed such a more positive philosophy of arithmetic derivative on a study of number words in natural language in Hofweber (2016).

       

145

metaphysics, with no regard to the quality of the question. We can call a metaphysically deep result the correct answer to the right question to ask in a particular part of metaphysics. Or better: the right question to ask relative to nearby alternatives. One doesn’t have to ask the best overall question, just the best question in this neighborhood. Maybe the best question overall in ontology is why there is anything at all, and the best question in the neighborhood of the question whether natural numbers exist is the question whether there are positions in an ω-sequence. If so, then only that there are or aren’t positions in ω-sequences is a deep result, but that there are no natural numbers is not. All this makes the following issue pressing: if we focus on the question we asked, and answer that question via reflection on our language alone, don’t we also have to know that we answered the right question before we can claim to have made proper progress in metaphysics? Another way to approach this issue is to note that so far we only talked about number words in English, but not how other human languages represent facts connected to numbers. Shouldn’t we move to the study of how numbers are represented in human languages more generally, and not focus too much on English? What if the non-referentiality of number terms is a particular feature of English, and it is only in light of it that we drew the conclusion we drew? If other languages are substantially different than English in this regard we might well conclude that English is somehow defective here. But it is not so clear how we can make sense of this defect. After all, the question we asked was stated in English: do numbers exist? If it is at all possible that we can answer a question by reflecting on the representations used in that question, then it seems sufficient to reflect on English representations to answer an English question. Why do we need to consider all other human languages in doing this? But the worry must remain that maybe we didn’t ask the best question we could have asked here, even if we answered the question we did ask. Maybe a different question in English, or a different question in a different language, would have been much more appropriate for metaphysics. And without answering this challenge it seems inappropriate to declare victory. It might seem impossible to make progress here within the approach of finding answers to metaphysical questions by thinking about our

146   language alone. How could we possibly know what the right question to ask is without antecedently knowing what reality is like? And how can we know what the best concepts are to employ in the question with which we probe reality, unless we already know what reality is like? In general, concepts are better if they fit reality better. To assess how good a concept is thus seems to require antecedent knowledge of what reality is like. That means in particular that it seems hopeless to try to evaluate the question we asked, and with it to evaluate the concepts we employed in that question, without first knowing what reality is like. And if that were so, then we would be doomed to at best achieve shallow results by thinking about our own representations, but never deep ones. Nonetheless, there is a line of thought that seems to give hope that there can be progress in this regard after all. It supports that it might be possible to achieve deep results simply by thinking about our own representations, with no prior knowledge of what reality otherwise is like. The main idea of that line of thought is this: What if there are some concepts such that if one were to consider replacing one of them with a different one, then one is in a position to see that such a replacement would always be irrational, no matter what the alternative concept might be. So, when I am considering the reasons for and against switching from employing my old concept C, to some new alternative one C* instead, the reasons will always favor sticking with C rather than switching to C*. In other words, what if there are some concepts that can never rationally be given up in favor of an alternative one? Let’s call a concept C inescapable just in case I can never rationally replace it with a different one, C*. It might seem dubious whether there are, or even could be, inescapable concepts. How could we rationally reject any proposed replacement of a given concept without even knowing what reality in general is like? But there are some possible candidates of such inescapable concepts nonetheless. To consider one example, take thin normative concepts like ‘ought’. We can imagine an alternative normative concept like ought*, which differs from ought in various ways, in particular that what one ought to do can be different than what one ought* to do.¹⁶ Which one of these ¹⁶ See Eklund (2017) for more on alternative normative concepts.

       

147

normative concepts should I have? When I think about this, it is arguable that it would always be irrational for me to replace my own normative concept of ought with an alternative ought*. I can simply reason as follows: if I were to make the switch, then I would reason about what to do using the concept of ought* instead of that of ought. And if that reasoning were effective, then I would do what I ought* to do. But what I ought to do and what I ought* to do might come apart. Thus if I were to make the switch I might end up doing things which I ought not to do, even if I reasoned perfectly. But if I stuck with my old normative concept, then I would not be subject to this mistake. So I should not switch, which is to say: it would be irrational for me to switch. Thin normative concepts are not the only ones that can be argued to be inescapable. Other candidates include logical concepts, the concepts of truth and fact, even the concept of natural number, and others. I have discussed inescapable concepts in much more detail in Hofweber (2023b) and (2023a), and this is not the place to get into them more. Inescapable concepts, if there are any, are relevant for our issue, because they allow us to assess whether a particular question we asked is the best one in its neighborhood, even without knowing what reality is otherwise like. If the concept of a natural number, say, is inescapable, then it is arguable that it would be irrational for me to think that asking about positions in ω-sequences is a better question to ask than asking about natural numbers. And so I should conclude that the question I did ask was the right question to ask here. And with it I should conclude that I used an appropriate language to ask the question I should have asked, even if other languages differ. This points to that it is in principle possible to make progress on whether we are asking the right question in metaphysics, without first knowing what reality is like: we can sometimes, in certain special cases, find out that we did ask the right question originally, since in these cases we can see that if we replaced one of the concepts which prominently figure in the question with an alternative one, then we would not ask a better question. If the central concepts involved in a particular question are inescapable, then it would be irrational to replace them with alternative ones. As just mentioned above, this is discussed in more detail in Hofweber (2023b), where I present several arguments in more detail for

148   particular concepts being inescapable. To be sure, this topic is only discussed here in the barest outline. Still, I hope that even in outline it points in a particular direction for how we can hope for deep results from reflecting on our own representations alone. And with it I hope it points to the fact that the philosophy of language can have a more substantial place in metaphysics than one might think. Not only can we sometimes answer questions of fact by thinking merely about the representations involved in these questions; we sometimes might even be in a position to conclude that we answered the right question, even though we don’t antecedently know what reality is like.

5.4 Assessing the status of the question Even if everything I said or outlined above is correct, the issue remains why we should give a special role to our questions in metaphysics at all. Why should we think that any of our questions have much to do with what metaphysics should in the end achieve? This might seem like a silly question at first, but I think it is well worth considering what is behind it. There are really two ways of thinking about what metaphysics is supposed to do. The first gives a special role to one or more initial questions that we ask at the outset of the project. These questions can be broad, as in ‘what is reality like in the most general ways?’, or more specific, as in ‘do natural numbers exist?’ But in either case, the relevant question sets the goalpost for metaphysical inquiry. When we ask such questions we thereby set the goalpost for metaphysics. What we need to do is answer that very question. We can modify the initial question, and thereby move the goalpost, if we determine that we should have asked a better question instead. For example, we might conclude that we should not have used the concept ‘natural number’ in this question, but instead some other concept in its place. But even then, the goalpost is set, or moved, by our questions, articulated with our representations in our language. If this is the right approach to metaphysics, then the questions we ask with our representations have a special place in metaphysics: they set the goalpost for the project, and thereby articulate what metaphysics should aim to achieve.

       

149

But that is not the only way to think of it. Alternatively, it could be that we don’t set the goalpost for metaphysics with a question, but rather that the goalpost is already there, waiting for us to meet it. The goalpost is reality itself, and whatever questions we might ask, with whatever representations or concepts, is not central to what metaphysics is supposed to do. Metaphysics is supposed to capture reality, and it needs to do so using the concepts that reality itself demands for its best description. So understood, it can be that our present concepts and representations are completely inadequate for metaphysics. We might at present be unable to reach the goalpost that reality has set for us, since the representations so far available to us are completely inadequate for doing so. And we might not even be in a position to articulate what the goalpost is. Let’s call the immanent stance towards metaphysics the one where we take the goalpost to be set by our questions, and the transcendent stance the one that takes reality to come with a goalpost for metaphysics, one that has already been set. Whichever stance is the right one will have an impact on what we should think about there being a connection between metaphysics and our own representations of reality. If we set the goalpost for metaphysics with an initial question, formulated in our own language, then there is such a connection: the goal of metaphysics itself is set by a question formulated in our language. The immanent stance thus supports a connection between our language and metaphysics, whereas the transcendent stance does not directly. It leaves open that what metaphysics ultimately should do is something that has nothing to do with anything we can state in our present language, or even more extremely, not anything that can in principle be stated in human natural languages. That seems be a big difference between the two stances. But, on the other hand, it might seem that these two stances are not all that different after all. Don’t both agree that we need to capture reality, at least in its most general way? The immanent stance takes the goal to be to answer the question ‘what is reality like?’, whereas the transcendent stance takes the goalpost to be set by reality itself. But then don’t both require that we describe reality, at least in the most general ways? Aren’t the two stances ultimately the same? I think it would be a mistake to draw this conclusion. There is a real difference between the two stances. That they seem similar is ultimately an artifact of the fact that I need to

150   use my language to describe the two stances, and when giving such a description they can seem similar. But the difference between them is a real one, one that doesn’t depend on my particular description of them. Naturally, it is hard to describe the difference without the use of language, but we can point in the right direction nonetheless. To make this clearer, we should distinguish two senses of what one might take reality to be. First, reality can be seen as the totality of facts: the totality of all that is the case. Second, reality can be seen as whatever the external thing is that determines the goalpost for metaphysics, and that demands a certain description involving particular concepts. Now, it could be that the concept of a fact is completely inadequate to capture what reality in the second sense is like. Maybe that concept is defective in some way, but it is hard for us, with our limited conceptual resources, to see this defect. Maybe it is the reliance on thatclauses to specify facts like the fact that snow is white. Maybe such that-clauses are required to be used by creatures like us, but this points to a defect in our minds and languages. If so, then aiming at reality as the totality of facts would miss the goalpost that reality as the external thing has set for us. The concept of a fact might be inadequate for metaphysics, but it might be hard for us to appreciate this, since we are bound to employ that-clauses in our language in key places. In this case the two stances would clearly come apart. And with it, on the transcendent stance, metaphysics might be a hopeless project for human beings: our minds might be inadequate to carry it out. But if the immanent stance is correct, then this won’t be so. If the goalpost is to describe reality as the totality of facts in the most general way, then we are at least in principle in the business of getting there. We can represent what the goalpost is and at least some of the facts, although it remains to be seen whether we can represent and know about enough of the facts to get a general picture of what reality is like. But even if we can’t reach the goalpost, at least we have a clear articulation of what it is, and that articulation can be given in our own language. Which of these two stances is the correct one for metaphysics? It is not easy to settle this question, and I certainly can’t hope to do so here. But there are some good reasons to favor the immanent stance. One worry about the transcendent stance is how reality as an external thing could by itself generate a demand to be described at all, and to be described using

       

151

specific terminology on top of that. Why would such a demand have any force on us, and why would it even be a demand at all, directed at us or anyone else? Reality as the external thing by itself is simply there, and it is hard to see how it by itself can generate a legitimate demand for a description. But if we ask a question, if we probe reality in a certain way with a particular question, and if we thereby engage in inquiry with such an articulated goal, then it makes sense how this gives rise to a demand on us to describe the world in a particular way. This description is the answer to the question we ourselves posed and hoped to answer. And the demand applies to us, since it was us that asks the question and hopes to answer it. The demand is tied to reality, understood as the totality of facts, since we asked what it is like. So understood, the immanent stance can make sense of there being a goalpost at all. Still, not all philosophers would agree with this way of approaching the issue. I take Ted Sider’s position in Sider (2011), for example, to adopt the transcendent stance, at least in spirit. This issue certainly deserves a more detailed discussion, but I hope highlighting the difference between the two stances, and the possible reasons for favoring the immanent stance. It at least makes sense of why it might be that asking a question, stated in our own language, has a special place in metaphysics in that it sets the goalpost for the project. This is the final part of the picture that gives the philosophy of language a central place in metaphysics.¹⁷ It is now time to look at the whole picture, all put together.

5.5 Immanent metaphysics and the status of language The philosophy of language has a constructive role in metaphysics at least in the following sense: the questions we commonly ask in metaphysics can sometimes be answered merely by thinking about our own language, in particular the language used in these questions, even though the answer is not an analytic truth. Everyone can agree with this, and that this is so can be shown, for example, by answering the question whether or not natural numbers exist by showing that number terms are ¹⁷ I defend the immanent stance as the correct one in more detail in Hofweber (2023a).

152   constitutively non-referential. But whether this by itself is enough is not so clear. To take it to the next level we should also evaluate the question we asked, and give support to the status of our questions themselves. I outlined that we can hope to achieve deep results by showing that the central concepts in the questions we asked are inescapable. And I argued that we can justify the status of our questions by showing that the immanent stance towards metaphysics is the right one: the goalpost of metaphysics is set by our initial questions, and it is adjusted from there by posing new and better questions. But it is always tied to our questions, and thus to our own language. These last two steps certainly deserve closer attention than I was able to give them here, but I hope at least the idea has come into view concerning how we might not just answer the questions we happen to have asked, but the proper questions of metaphysics, simply by reflecting on our own language. To be completely clear, no claim is made here or elsewhere that all of metaphysics can be carried out this way, nor that somehow thinking about language is the proper method for making progress in metaphysics. To the contrary, we should only expect this approach to work for a distinguished range of cases, ones where we ask questions involving certain especially central concepts, ones that we cannot rationally replace, and even then only when the situation is favorable enough so that the question we asked is indeed answerable by thinking about our own language, as is the case with the question about the existence of natural numbers. We can say that immanent metaphysics is the part of metaphysics that hopes to find all the deep results that can be gotten this way. Immanent metaphysics thus hopes to answer questions involving inescapable concepts by reflecting on the language used in the question alone. Immanent metaphysics is based on the immanent stance, which takes our questions to set the goalpost for metaphysics. Immanent metaphysics then determines which synthetic truths are guaranteed by the representations used in these goalpost-setting questions, and whether it can be rational to take other questions to be an improvement. The results of immanent metaphysics can set a framework for the rest of metaphysics, and they essentially involve a constructive role for the philosophy of language, but not just the philosophy of language. To determine whether a

       

153

particular concept can’t rationally be replaced with a different one essentially concerns questions in epistemology. Thus not only does the philosophy of language have a constructive role in metaphysics, but so does epistemology.¹⁸ I would like to stress that this approach to the role of the philosophy of language in metaphysics is quite different than what Emmon Bach (1986) called ‘natural language metaphysics’ and what Friederike Moltmann (2017) called ‘natural language ontology’. Those approaches concern uncovering the basic metaphysical categories and distinctions implicit in our representations of the world in language. These approaches do not directly concern the world, but only how the world is represented in our languages. There is much interesting work to be done concerning what kinds of central distinctions are involved in how we represent the world, what kind of things we represent the world as containing, and so on. But by itself none of this leads to insights into what the world is actually like. It only leads to insights into how we represent it to be. Bach is explicit in this limitation of his approach, for example when he memorably says, “It would be immoral of me as a linguist . . . to make claims one way or the other about whether or not these sorts of things correspond to real things in the world . . . or to nothing at all” (Bach 1986, 592). For the same sentiment, see also Pelletier (2011). Similarly, Strawson’s ‘descriptive metaphysics’ in (1959) is simply focused on our own way of representing the world, not directly on what the world is like. Not that this is not of great interest, but these approaches essentially accept the language–metaphysics gap, and they need to be distinguished from immanent metaphysics. Immanent metaphysics is not focused on our way of representing the world, but on how the world is. But immanent metaphysics is based on our ability to overcome the language–metaphysics gap. It looks at language to find out what reality is like, which is its goal. Naturally, those who are in the grip of the language–metaphysics gap will think that immanent metaphysics is impossible: we can never find out what reality is like simply by thinking about our own ways of representing reality. But ¹⁸ I have developed this view of the role of epistemology in metaphysics in more detail in Hofweber (2023b).

154   I hope to have shown that this would be a mistake: the gap can be overcome, and it is possible in at least some cases to draw conclusions about reality simply by thinking about our own representations of reality. How much can be achieved with this approach, whether it truly leads to deep results, and whether we can defend the immanent stance over the transcendent one is left to be seen. It is thus fair to conclude that a good part of this paper is somewhat programmatic. But even so, the nonprogrammatic part of this paper should be enough to defend my main conclusion: the philosophy of language does have a constructive role in metaphysics, and the disciplines of the philosophy of language, epistemology, and metaphysics are more closely connected than standard views on the relationship between these disciplines would have it.¹⁹

References Bach, E. (1986) ‘Natural language metaphysics.’ In Marcus, R. B., Dorn, G., and Weingartner, P., eds, Logic, Methodology and Philosophy of Science VII. North Holland. Blackburn, S. (1984) Spreading the Word. Oxford University Press. Carnap, R. (1956) ‘Empiricism, semantics, and ontology.’ In Carnap, R., ed., Meaning and Necessity: A Study in Semantics and Modal Logic, pp. 205–21. University of Chicago Press, 2nd ed. Dreier, J. (2004) ‘Meta-ethics and the problem of creeping minimalism.’ Philosophical Perspectives, 18(1): 23–44. Dummett, M. (1991) Frege: Philosophy of Mathematics. Harvard University Press. Dyke, H. (2008) Metaphysics and the Representational Fallacy. Routledge. Eklund, M. (2017) Choosing Normative Concepts. Oxford University Press. Frege, G. (1884) Die Grundlagen der Arithmetik: eine logisch-mathematische Untersuchung über den Begriff der Zahl. W. Koebner. Gaskin, R. (2020) Language and World: A Defense of Linguistic Idealism. Routledge. Godfrey-Smith, P. (2006) ‘Theories and Models in Metaphysics.’ Harvard Review of Philosophy, 14: 4–19. ¹⁹ My thanks to Jeff Pelletier for his comments on an earlier draft.

       

155

Hofweber, T. (2005) ‘Number determiners, numbers, and arithmetic.’ Philosophical Review, 114(2): 179–225. Hofweber, T. (2007) ‘Innocent statements and their metaphysically loaded counterparts.’ Philosophers’ Imprint, 7(1): 1–33. Hofweber, T. (2016) Ontology and the Ambitions of Metaphysics. Oxford University Press. Hofweber, T. (2023a) Idealism and the Harmony of Thought and Reality. Oxford University Press. Hofweber, T. (2023b) ‘Inescapable concepts.’ Australasian Journal of Philosophy. DOI: 10.1080/00048402.2023.2172593. Jackson, F. (2000) From Metaphysics to Ethics. Oxford University Press. Ladyman, J., and Ross, D. (2007) Everything Must Go. Oxford University Press. Ludlow, P. (2011) The Philosophy of Generative Linguistics. Oxford University Press. Moltmann, F. (2017) ‘Natural language ontology.’ In Oxford Encyclopedia of Linguistics. Oxford University Press. Pelletier, J. (2011) ‘Descriptive metaphysics, natural language metaphysics, Sapir-Whorf, and all that stuff: evidence from the mass-count distinction.’ The Baltic International Yearbook of Cognition, Logic and Communication, 6: 1–46. Schaffer, J. (2009) ‘On What Grounds What.’ In Chalmers, D., Manley, D., and Wasserman, R., eds, Metametaphysics: New Essays on the Foundations of Ontology, pp. 347–83. Oxford University Press. Scharp, K. (2018) ‘Philosophy as the study of defective concepts.’ In Burgess, A., Cappelen, H., and Plunkett, D., eds, Conceptual Engineering and Conceptual Ethics, pp. 396–416. Oxford University Press. Sider, T. (2011) Writing the Book of the World. Oxford University Press. Sider, T. (2013) ‘Against parthood.’ In Bennett, K. and Zimmerman, D., eds, Oxford Studies in Metaphysics, Volume 8, pp. 237–93. Oxford University Press. Strawson, P. (1959) Individuals: An Essay in Descriptive Metaphysics. Methuen. Thomasson, A. L. (2016) ‘What can we do when we do metaphysics?’ In D’Oro, G. G. and Overgaard, S., eds, The Cambridge Companion to Philosophical Methodology, pp. 435–58. Cambridge University Press.

6 On Lying, “Strictly Speaking” Marga Reimer

6.1 Lying “strictly speaking” Remarks such as the following might occur naturally in ordinary everyday conversations: A: Strictly speaking, I may have lied, but since I wasn’t trying to deceive anyone I can hardly be called a “liar.” B: You may have said that you were going to move to Phoenix today, but your tone was clearly sarcastic, so you basically lied. You’re such a liar! The naturalness of such utterances is easily brought out by situating them in the relevant sorts of contexts. Suppose that A is making reference to a forced false confession elicited after having been arrested for, and mercilessly grilled about, their alleged role in a recent homicide. Suppose further that A has no knowledge of the crime in question and falsely confesses only to put an end to the relentless and hours-long questioning. Suppose that B is responding to the addressee’s claim that they had “just said” that they would be moving to Phoenix that day. Suppose further that what the addressee (C) had actually said (“Yeah, I will be moving to Phoenix today—you know how much I hate it here in Tucson!”) was, in fact, patently sarcastic. In what follows, I will reflect on the significance of remarks like those of A and B to philosophical accounts of lying.¹ I will suggest that such remarks

¹ By “philosophical” accounts of lying I simply mean definitions or (in other words) “conceptual analyses” of lying that analytic philosophers have proposed over the years. Marga Reimer, On Lying, “Strictly Speaking” In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Marga Reimer 2023. DOI: 10.1093/oso/9780198892724.003.0006

 , “   ”

157

(in conjunction with various other factors) lend credence to three core ideas. First, the intention to deceive is not a necessary condition for lying generally, even though deceptive intent is a defining (and thus necessary) feature of prototypical lies.² Second, while ordinary folk, including analytic philosophers when engaged in ordinary conversation, might regard (non-logical) implying as a potential vehicle for lying in a loose sense (“basically” lying), saying is required for lying in a strict sense: in the sense relevant to philosophical “definitions” (or “conceptual analyses”) of that familiar concept. Third, aptness³ of the descriptor “liar” depends, in part, on whether deceptive intent was present during a presumptive lie. The format of this paper is as follows. After a brief and tentative characterization of some “ordinary varieties” of lying (strict, prototypical, and loose), I develop and defend the proposed characterization of lies in the strict sense. I do so in a way that respects (i) the intuitive distinction between lies in the strict sense and prototypical lies, and (ii) the equally intuitive distinction between lies in the strict sense and lies in the loose sense (“basically” lies). I then go on to suggest that sensitivity to these ordinary varieties of lying and the distinctions between them effectively imposes two constraints on philosophical definitions of lying. First, reference to deceptive intent has no place in such definitions, even though it unquestionably does in characterizations of prototypical lies. Second, while lying through Gricean implicature (a kind of lying, “loosely speaking”) has no place in philosophical definitions of lying, Gricean saying (or something close to it) is required for lying in the strict sense, in the sense relevant to philosophical (vs. dictionary) definitions of that notion.⁴ I then discuss the relevance of the second constraint to recent (and future) experimental studies regarding lying, studies designed to elicit Perhaps the least controversial definition of such accounts is an extensional one: an account of the sort featured in Mahon’s (2015) “Definitions of Lying and Deception” entry in Stanford Encyclopedia of Philosophy. ² By “prototypical” I simply mean exemplary or paradigmatic. Thus, “prototypical lies” can be thought of as uncontroversial or (in other words) clear-cut cases of lies. ³ Such “aptness” is to be understood pragmatically rather than semantically so as to allow for the (conceptual) possibility of liars who do not tell lies and truth-tellers who do. There is thus a sense in which “liar” is not synonymous with “lier”: that is, with one who lies. See Section 6.6 below for details. ⁴ Dictionary definitions of lying invariably make mention of deceptive intent and, in this respect, characterize prototypical cases of lying. As argued below, lying in the strict sense does not require deceptive intent.

158   folk views as to what “counts” as a lie, or as having lied. In the penultimate section of the paper, I connect the views and arguments of the preceding sections to our natural inclination (or disinclination) to call someone a “liar.” Finally, I conclude by reflecting briefly on the potential philosophical importance of how we (as ordinary speakers vs. philosophical theorists) think and speak about lying and about those who lie: about those whom we think of as “liars” (and those whom we don’t).

6.2 Lies, lies, and more lies! Below I introduce and define some terminology, all of which is grounded in ordinary usage. Importantly, the proposed “definitions” are best understood as “characterizations” rather than as “necessary and sufficient conditions” or (in other words) “criteria.” Indeed, I do not think (non-stipulative) criterial definitions of the notions in question are even possible, given the plausible existence of “borderline” cases.⁵ Let us begin, then, with the notion that figures most prominently in the present paper: the notion of lying “strictly speaking” or (in other words) lying in a “strict” sense.

6.2 (i) Lying “strictly speaking” (lying in a “strict” sense) This sort of lying, the sort explicitly invoked in A’s remarks, involves saying something, in a roughly Gricean sense of that notion,⁶ believed by the speaker to be false. Thus, saying (as employed herein) is to be distinguished from implicating including, and in particular, implicating ⁵ To make the point vivid, just consider cases where a speaker says something that they mistakenly believe to be false and do so with the intention to deceive the addressee into believing it true. While some theorists would call this a lie, others would not. A third option, which I favor, would be to claim that the concept of lying is indeterminate in the sense that there is no fact of the matter as to whether such utterances are “lies” in the strict sense. In that case, philosophers in search of a “criterion” for lying are not simply “barking up the wrong tree”; their confusion is deeper than that, for there is no cat in any tree (no criterion to be “captured”) to begin with. What, then, are the proposed criteria for? Perhaps they are criteria for a particular philosopher’s preferred understanding of prototypical lies. ⁶ Grice (1975).

 , “   ”

159

via Gricean conversational implicature.⁷ Saying is also to be distinguished from any philosophical (or otherwise theoretical) notion of saying (stating or asserting) that goes beyond anything implicit in the Gricean notion of that familiar linguistic phenomenon. Importantly, Grice’s original distinction between “saying” and “implicating” was introduced by Grice as one that approximates the ordinary everyday distinction between saying and implying. Such proximity to ordinary usage (embraced throughout the present paper) is important insofar as it helps minimize unnecessary and potentially distorting intrusion of theory into conceptualization of the relevant data.

6.2 (ii) Lying prototypically speaking (lying in a “prototypical” sense) Lies of this sort occur when one says something they believe to be false and does so with the intention to deceive the addressee into believing what they say (“what is said”) to be true. These are paradigm lies—the kinds of lies one naturally thinks of when trying to come up with good examples (exemplary cases) of lies. Thus, although A admits to the possibility of having lied “strictly speaking,” in denying any deceptive intent they effectively deny having lied prototypically. Because those who lie often know that what they are saying is false,⁸ it is often (although arguably not always) the case that lying involves saying what is in fact false, not merely what is believed to be false. As a classic example of this, consider the child who lies about having taken a cookie from the cookie jar just as they are wiping the crumbs from their face. Does prototypical lying involve (that is, require) such factual falsity? For purposes of the present paper, I will not build factual falsity into the proposed characterization of prototypical lying. This should not, however, be read as implying that such falsity is not a “characteristic feature” of lies—perhaps it is; it is certainly generally true of lies—at least if we

⁷ Grice (1975). ⁸ More generally, speakers (uttering declarative sentences) generally know whether what they are saying is true or false.

160   assume that persons generally know whether what they are saying is true or false. But whether factual falsity is, or is not, a characteristic feature of lies, and thus a defining feature of prototypical lies, is (again) irrelevant to the purposes of the present paper. (For more on the issue of factual falsity, see Section 6.3(ii) below.) I take it as a given, as something too obvious to require an argument, that not all lies are prototypical lies⁹—any more than all birds are prototypical birds, than all chairs are prototypical chairs, or than all games are prototypical games. Psychologically (if not logically) compelling arguments are, however, readily available in the form of presumptive lies that appear not to involve deceptive intent, such as coerced lies or bald-faced lies, both of which have been discussed at length in the recent literature.¹⁰ (For more on such cases, see Section 6.4(i) below.)

6.2 (iii) Lying “loosely” speaking (lying in a “loose” sense) Lying in this sense, in the sense invoked in B’s remarks, would include cases where one “basically lies” indirectly through implying (e.g., through conversationally implicating) rather than directly through saying. Such “more or less” or “for all intents and purposes” lying can thus be viewed as a variety of misleading—distinguished, in part, not only by its indirectness but also by the fact that it is naturally (although not inevitably) characterized as “lying” in ordinary speech—as in B’s remarks.¹¹

⁹ At the same time, it seems to me that philosophical definitions of lying often derive their initial plausibility from their successfully capturing not unreasonable characterizations of prototypical lies, rather than of lies generally. Perhaps this is clearest in the case of definitions that require an intention to deceive the addressee into believing true the content of what is said— something that the speaker believes to be false. Part of what makes this definition so appealing (at least initially) is that the “best examples” of lies involve such intent. The difficulty is that a definition of lying, a philosophical definition of lying, must not exclude cases of lies that are not paradigmatic. ¹⁰ For a discussion of these sorts of lies, see Mahon’s (2015) Stanford Encyclopedia of Philosophy entry, “The Definition of Lying and Deception.” ¹¹ Might it then be possible to lie using irony or metaphor? B’s remarks, given their prior linguistic context, would appear to suggest that it is indeed possible to “basically lie” using irony—that is, to lie in a loose sense. I suspect the same to be true of metaphor. Moreover, if the metaphors in question are dead or dying (resulting in secondary literal meanings), then lying in

 , “   ”

161

Of these three ordinary varieties of lying, it is lying strictly speaking that is primary, with lying prototypically and lying loosely speaking being derivative: that is, understood in terms of the (logically) prior notion of lying in a strict sense. Thus, prototypical lies are defined in terms of the characteristic features of lies in the strict sense; lies loosely speaking are defined in terms of implicature which is defined in terms of the more basic notion of saying, which is involved in lies in the strict sense. Importantly, for present purposes, lying strictly speaking is the notion of lying according to relevant philosophical definitions of lying— definitions which aspire to precision and exactitude, at least insofar as the data allows for this. Let us therefore have closer look at that notion.

6.3 Lying in a “strict” sense: In defense of a philosophical “definition” As indicated above, I am sympathetic to the idea that to lie, in a strict sense, is to say (in a roughly Gricean sense) something one believes to be false. That’s it. Although I hesitate to construe this definition as a “criterion” for lying, as a statement of “necessary and sufficient conditions,” I do view it as a characterization (a construal vs. a definition proper) that can be seen as comporting with, or conflicting with, the various criterial definitions of lying that have been proposed in the philosophical literature. After all, all such definitions presumably aim to capture what lying, strictly speaking, involves—even if that phenomenon is arguably misconstrued as something amenable to criterial elucidation.¹² But regardless of whether a criterial definition of lying is possible, the question to which the philosopher seeks an answer is simple

a strict sense would presumably be possible. Consider, for example, Mahon’s (2015) case of the farmer who utters “We have tomatoes coming out of our ears” with the intent to deceive the addressee into believing they had a bumper crop that year. ¹² Here, I am construing an “exact meaning” as a “particular meaning” where “particular” meanings may in fact be indeterminate and so not amenable to specification in terms of necessary and sufficient conditions. This should not be controversial as many plausible dictionary definitions, despite their particularity (distinguishability from the meanings of other words), are “exact” in this sense. Indeed, their exactness arguably essentially involves indeterminacy.

162   (or at least simple to state): What exactly is involved in lying?¹³ For those who prefer the language of meaning to the language of doing, the operative question might be: What exactly does it mean to lie? My tentative defense of the proposed characterization of lying, of lying in the “strict” sense, involves exploring and expanding, however briefly, upon the notions of saying, believing, factual falsity, and deceptive intent. Let us therefore consider these notions in turn, beginning with the notion of saying.

6.3 (i) Saying Why construe lying in terms of saying rather than more broadly in terms of (say) communicating? In other words, is saying (vs. communicating) really a necessary condition for lying? Saying is considerably narrower than communicating; requiring saying for lying effectively precludes conflation of lying and patent misleading, the latter of which is naturally understood in terms of (Gricean) implicature. However, saying, as I use that notion here, is not to be conflated with anything stronger (narrower) than Gricean saying. Insofar as the Gricean notion of saying (adopted herein) mirrors the ordinary notion of saying, its requirement precludes intrusion of question-begging philosophical theory as might occur (for example) with a more restrictive sense of saying that amounts to full-blown assertion and thus to speaker commitment to the content of what is said.¹⁴ A further (and related) advantage of construing lying in terms of saying, rather than more generally in terms of communicating or more narrowly in terms of asserting, is that doing so comports with the common understanding of lying which arguably involves saying something, something the speaker believes to be false. As Wiegmann and ¹³ Ironically, it may turn out (as suggested below) that there is nothing “exact” about what lying involves, given the possibility of genuine “borderline” cases: utterances about which there is no “fact of the matter” as to whether they are lies or not lies. That there are indeed such cases is part of the motivation behind prototype analyses of lying such as that proposed by Coleman and Kay (1981). ¹⁴ There would be a different, though parallel, danger involved in weakening the ordinary Gricean sense of saying so that certain cases of implicating might count as saying.

 , “   ”

163

Meibauer (2019, p. 1) note, “virtually all philosophers agree that an important—or probably the most important—desideratum of a definition of lying is to capture people’s understanding and use of this concept.”¹⁵

6.3 (ii) Factual falsity of what is believed to be false Is saying something merely believed to be false sufficient for lying? Isn’t it also necessary that what is said actually be false? There are at least four reasons for questioning this. First, consider how much about which we might all be wrong—deeply wrong. Reflect briefly on the history of science which, over time, has prompted revision, if not outright rejection, of what was formerly regarded as “common sense.” Suppose that 600 years ago someone said to someone else, “the earth rotates around the sun.” Suppose further that they (like most of their contemporaries) believed such a statement to be false—indeed, patently false. Intuitively, any such person would have lied, notwithstanding the statement’s factual—and happenstantial—truth. (This is not to say the happenstantial truth-teller might not try to escape the charge of having lied “strictly speaking” on the grounds that what they said was in fact true.)¹⁶ Second, any intuition that factual falsity is necessary for lying in a strict sense can, to some extent, be explained (and indeed “explained away”) by the fact, alluded to earlier, that most liars know, in an ordinary (vs. hyperbolic philosophical) sense of “know,” that what they are saying is in fact false. And, if something is known, in an ordinary sense of that notion, to be false, then of course it is false—as when one student says to another that an upcoming exam has been postponed while knowing full well that it has not. However, this suggests only (or at most) that factual ¹⁵ However, in contrast to Wiegmann and Meibauer, I believe that the common “understanding and use” of the concept of lying crucially involves a distinction between lying strictly speaking and lying loosely speaking, as evidenced by the remarks of ordinary speakers A and B. Additionally, I do not believe (in contrast to Wiegmann and Meibauer) that the strict sense of lying requires deceptive intent—although I believe this is true, indeed obviously true, of prototypical lies. ¹⁶ On my view, such a speaker would be exploiting the indeterminacy inherent in the notion of lying, a notion that arguably admits of borderline cases, such as cases where the presumptive liar mistakenly believes what they are saying to be false.

164   falsity is a characteristic feature of lies—not a necessary one. In that case, it is perhaps a defining feature of prototypical lies. Third, perhaps it is possible to regard cases of presumptive lying in the absence of factual falsity as “borderline” cases: as cases about which there is no “fact of the matter” as to whether or not they are lies. This would certainly be consistent with the fact that theorists are more or less split over whether utterances of the sort in question are lies or not. Viewing these presumptive lies as on the border between utterances that are lies and utterances that are not lies would not, however, undermine the proposed definition of lying (strictly speaking), as that definition is not proposed as a criterion, as a set of necessary and sufficient conditions, for lying. Fourth, it is natural to think of those who say (and with deceptive intent) what they believe to be false as “liars”—regardless of whether what they say is in fact false. These folks are liars who tell the truth “by accident,” much like those who mistakenly believe what they say to be true spread falsehoods “by accident.” Yet we have little or no tendency to call the latter persons “liars.” We say simply that they “happen to be” wrong— just as those who say what they mistakenly believe to be false “happen to” to be right. However, as suggested by the remarks of A and B (and as argued in the penultimate section of the present paper), the aptness of the descriptor “liar” is no guarantee of a person’s having lied “strictly speaking.” It may simply reflect the presumed presence of deceptive intent.

6.3 (iii) Deceptive intent Why is there no reference to deceptive intent in the proposed characterization of lying? While deceptive intent is indeed a defining (and so necessary) feature of prototypical lies, to suppose that it is a necessary condition for lying generally leaves awkwardly unexplained (or inadequately explained) certain cases of prima facie lies. These would include the coerced lies and bald-faced lies that have gotten so much attention in the recent literature.¹⁷ Given the proposed construal of lies in the “strict” sense, such prima facie lies might, in at least some cases, be just what they ¹⁷ See the seminal papers of Carson (2006) and Sorenson (2007) for details.

 , “   ”

165

appear to be: genuine, if atypical (that is, non-prototypical) lies. There would accordingly be no ad hoc “explaining away” to be done here as things would be more or less as they appear to be insofar as utterances of the sort in question are naturally (if somewhat misleadingly)¹⁸ called “lies.” What, now, of the philosophical significance of the proposed construal (and defense thereof) of lying in the strict sense? As argued below in Sections 6.4 and 6.5, that construal arguably helps make sense of the absence of anything approaching a consensus with regard to certain key issues in the philosophical literature on lying. Part of the problem driving this failure of consensus is (I believe) insufficient attention to (or ignorance of) the two commonplace distinctions reflected in the remarks of A and B: the distinction between lying strictly speaking and prototypical lying and the distinction between lying strictly speaking and lying loosely speaking. Furthermore, insensitivity to the latter distinction raises important methodological issues insofar as it has implications for the philosophical relevance of recent experimental studies that appear to lend credence to the idea that conversational implicature is a potential vehicle for lying. (See Section 6.5 below for details.)

6.4 Ongoing debates, overlooked distinctions Let us now turn to a couple of debates in the recent philosophical literature on lying and consider whether attention to the distinctions discussed above might help facilitate their resolution.

6.4 (i) Is the intention to deceive a necessary condition for lying? Despite the enormous amount of attention that has been given to the question as to whether deceptive intent is necessary for lying, there is nothing approaching a consensus on this issue. However, I suspect that ¹⁸ Calling such presumptive lies “lies” would be misleading only insofar as it might suggest that such utterances, like “typical” lies, involve deceptive intent.

166   cases of lying (strictly speaking) without the intention to deceive are both genuine and not uncommon enough to be considered anomalies. They arguably occur in at least some cases of bald-faced lies and in at least some cases of coerced lies. Importantly, all that matters (with regard to theory or analysis) is that a significant number of these cases, more than an isolated few, involve the absence of deceptive intent: specifically, the absence of an intent to deceive the addressee into believing what is said to be true—something the speaker themselves believe (if not know) to be false. But if this is right, then why is there still so much controversy surrounding the issue? I suspect that the answer lies (at least partly) in the failure to recognize and appreciate an ordinary distinction made at least implicitly by ordinary folk (such as A) in ordinary contexts: the distinction between lies in the strict sense and prototypical (or, more colloquially, typical) lies. The literature is full of examples of what appear (prima facie) to involve genuine lies in the absence of deceptive intent. While some of these cases involve coerced lies, others involve bald-faced lies. This is not of course to say that such cases never involve deceptive intent, only that they do not always appear to—and that is enough to cast doubt on philosophical definitions of lying that make deceptive intent a requirement for lying. Such intent is, however, a defining feature of prototypical lies. It is, in other words, characteristic of lies in general—which is arguably why it is mentioned in all “dictionary definitions” of lying and of lies. This is hardly a surprise, given that such definitions often aim (when appropriate) to capture prototypical cases. Think, for example, of dictionary definitions of kind terms like bird, chair, or game—many of which make reference to what is “typically” the case so as to allow for such things as penguins, beanbag chairs, and solitaire. The more general, and perfectly ordinary, distinction operative here is that between types and prototypes. That this is indeed an ordinary distinction, and thus one that a philosophical definition of an ordinary phenomenon should aim to capture or at least respect, is reflected in the widespread ability to recognize (e.g.) that penguins are genuine birds, albeit unusual ones, that beanbag chairs are genuine chairs, albeit unusual ones, or that solitaire is a genuine game, albeit an unusual one. That this sort of type/prototype distinction is commonly made in the

 , “   ”

167

case of lying is reflected in A’s remarks: although A recognizes that, strictly speaking, they did in fact lie, they also recognize that the lie they told was an unusual (non-prototypical) one—something reflected in the inappropriateness of conceptualizing or labelling A as a “liar” on account of their saying something they believed to be false. (For more on the significance of this latter point, see Section 6.6 below.) Let us now turn to another controversy, newer (though no less important) in the philosophical literature on lying. The issue here concerns whether lying is possible in the absence of (roughly Gricean) saying; whether (more specifically) lying though Gricean conversational implicature is possible. That this is indeed possible is lent credence by B’s remarks. B refers to C as a “liar” on the grounds that C’s ironic statement, though literally true, was used to implicate something C believed to be false: that C would not be moving to Phoenix that day.¹⁹ Part of what makes this issue particularly topical is that it is tied in with emerging methodological issues in the philosophical literature on lying: issues concerning the relevance (or irrelevance) of experimental studies designed to elicit folk views regarding the conditions for lying.²⁰

6.4 (ii) Is it possible to lie through Gricean conversational implicature? I believe (for reasons indicated above) that lying in the strict sense requires Gricean saying; it is thus not possible to lie in this sense through conversational implicature. Resistance to this idea would appear to reflect a different, if equally natural, conflation: a conflation of lying strictly speaking and “basically” (or “more or less”²¹) lying. This conflation, which is both pervasive and unobjectionable in ordinary conversation, is evident in B’s remark. Indeed, not only is such a ¹⁹ As suggested below in Section 6.6, B may have an additional reason for referring to C as “such a liar.” ²⁰ See, for example, Wiegmann et al.’s (2021) paper, “Lying, Deceptive Implicatures, and Commitment.” ²¹ This expression, applied to the phenomenon of lying, seems particularly revealing insofar as it suggests a spectrum notion of lying, something that would render misguided the search for a “criterion” for lying.

168   conflation unobjectionable in ordinary conversation, it arguably reflects an important truth regarding the wrongness of intentional deception, namely: it is the intentional deception itself that is wrong, not the means by which it is conveyed. This would explain why, while we are not inclined to think of A as a “liar,” despite their having lied (as even A themselves acknowledge), we are inclined to think this of B’s addressee, whose ironic utterance is intentionally deceptive, though not “strictly” a lie. Interestingly, some recent experimental studies (discussed in the immediately following section) appear to lend credence to the idea that it is in fact possible to lie through Gricean conversational implicature. However, as I argue below, appearances here are potentially misleading. It is possible that they are rooted in a conflation between lying “strictly speaking,” as A might put it, and lying “loosely speaking”—or (in other words) “basically lying,” as B might put it. In sum, while sensitivity to the distinction between lies strictly speaking and prototypical lies suggests that deceptive intent is not necessary for lying in the strict sense, sensitivity to the distinction between lies strictly speaking and lies loosely speaking suggests that conversational implicature is not a possible vehicle for lying in the strict sense. Because philosophical definitions of lying are presumably concerned with lying in the strict (and indeed strictest) sense, such definitions should not incorporate the deceptive intent definitive of prototypical lies; nor should they countenance, as a potential vehicle for lying, conversational implicature of the sort characteristic of lying in the loose sense.

6.5 Can experimental studies inform philosophical definitions? When theorists (vs. ordinary folk) conflate lying in a strict sense with lying in a loose sense, the likely result is false or misguided theories rather than enhanced communicative efficiency or apt moral judgement. Thus, consider recent experimental studies which, intentionally or not, elicit judgements concerning lying in a less than strict sense: intentional

 , “   ”

169

deception carried out through Gricean conversational implicature rather than Gricean saying.²² Study participants evidently do sometimes characterize such deception as “lying”—much like B does. While the studies themselves may be unobjectionable and the results quite interesting, the interpretation of those results is potentially problematic. A theorist might, for example, claim that their study’s results undermine philosophical definitions of lying that require Gricean saying for lying. But the validity of such a claim is potentially undermined by an insensitivity to the distinction between lying in a strict sense and lying in a loose sense. Perhaps what the studies actually support is the considerably more modest claim that lying “loosely speaking” is possible through conversational implicature. We ordinarily, and as ordinary folk rather than as theorists, conceptualize lying in terms of “saying something known”²³ (or believed) to be false. We do not conceptualize lying disjunctively in terms of saying, implying, or otherwise communicating something believed to be false. Indeed, such an expansive characterization would encompass patent misleading no less than prototypical lying. Even B, who accuses C of “basically lying,” might concede that C didn’t lie “strictly speaking.” Otherwise, what is the significance of the “basically” proviso here? But that C might not have lied “strictly speaking” is of no real consequence to B, whose point is that C’s behavior is as reprehensible as it would have been had they instead lied in the strictest of senses (e.g., “I mean it when I say I will be moving to Phoenix today”). Thus, experimental studies that suggest that lying through Gricean implicature is possible do not necessarily undermine philosophical analyses of lying according to which saying (vs. saying or otherwise communicating, including conversationally implicating) is necessary for lying.

²² Although this section is concerned with experimental studies that bear on the question of whether it is possible to lie through Gricean implicature, it would be interesting to consider experimental studies that bear on the question as to whether lying in the strict sense is possible in the absence of deceptive intent. My suspicion is that such studies would suggest that this is indeed possible—at least insofar as the studies’ participants have an understanding of lying that is as nuanced as A’s appears to be. ²³ Philosophers, with their hyperbolic sense of know, might naturally construe lying more cautiously in terms of belief.

170   I suspect that lying “strictly speaking” in the absence of saying simply does not exist; it is a theoretical artifact insofar as the assumption that it does exist has no real explanatory power. It would explain the results of experimental studies of the sort in question only if the studies’ participants were able (during or prior to those studies) to demonstrate a sensitivity to the distinction between lying in a strict sense and lying in a loose sense. However, were such sensitivity demonstrated, participants would, I suspect, identify cases involving intentional deception through conversational implicature as cases of lying “loosely speaking” rather than as cases of lying “strictly speaking.” This certainly seems likely if we regard B, who talks of “basically lying” despite the truth of what was “actually said,” to be representative of study participants in terms of having a nuanced (if commonplace) conceptualization of the kind of duplicity B’s addressee engages in. The fact that B thinks of C as “such a liar” even though they did not (on the proposed view) lie strictly speaking, helps make sense of experimental studies that suggest lying in the absence of saying is possible. Perhaps the study’s participants (ordinary speakers like B), recognizing that the deceptive speaker in a hypothetical scenario (perhaps someone like C) is aptly called a “liar,” naturally regards that speaker as having in fact lied. After all, is it not an analytic truth that liars are those who lie? However, as suggested above and argued below (in Section 6.6), the descriptor “liar” in such cases is apt for capturing a moral flaw characteristic of persons who intentionally deceive others, whether through lying in the “strict” sense or in the “loose” sense. That descriptor does not, however, seem applicable to those who, like A, are coerced into lying yet who nevertheless lie strictly speaking. To see this, just imagine that A’s addressee had responded with, “You’re such a liar—even if that false confession was basically forced out of you.” There would be something decidedly odd, and patently unfair, about such a statement. Thus, my general take on experimental studies of the sort in question is that both their construction and their interpretation could be undercut by an insensitivity to the distinction between lies in a strict sense and lies in a loose sense—the latter of which include cases where the communicated content is not said but merely implicated. Importantly, this is a distinction that ordinary folk make, as evidenced by B’s remarks. Thus,

 , “   ”

171

the cases in question should ideally be followed by a prompt asking whether the speaker, who implicated something they believed to be false and did so with deceptive intent, was lying in a “strict” sense or in a “loose” sense. My suspicion is that the latter would be the preferred choice in cases of the sort in question—something of relevance to lying in a loose sense (“basically” lying) but presumably of no relevance to philosophical definitions (or “conceptual analyses”) of lying, which are concerned with lying in a strict sense. That lying “loosely speaking” would indeed be B’s preferred choice is at least suggested by the fact that B accuses C of having “basically lied,” notwithstanding the acknowledged truth of what they “actually said” via their ironic utterance. That it might be possible to lie loosely speaking through Gricean conversational implicature is certainly interesting, especially to pragmatics (vs. semantics), but it is arguably of no relevance to philosophical definitions of lying. This does not, however, mean that experimental studies in support of such a possibility are of no potential relevance to philosophical definitions of lying. Such relevance would simply require (as just suggested) that the studies’ prompts both confirm participant sensitivity and reflect author/study sensitivity to the distinction between lying in a “strict” sense and lying in a “loose” sense. Importantly, such author/ study sensitivity would not involve any kind of “outside imposition” of theory on data, as the distinction in question is an ordinary one. Indeed, as suggested above, such ordinariness is abundantly evident from colloquial speech. Just consider the ordinariness of utterances like: I guess I kinda lied, She pretty much lied, You more or less lied, He didn’t exactly lie, For all intents and purposes, they lied, etc. Such utterances, which would appear to make reference to lying in a loose sense, are meaningful only against the backdrop of a stricter sense of lying.

6.6 When those who lie are not liars; when those who do not, are Before concluding, let us return once again to the remarks that open the present paper. Together, these ordinary everyday remarks suggest a twofold irony. While they suggest that those who lie in a strict sense

172   are not always aptly called “liars,” they also suggest that those who lie in only a loose sense, while saying what is literally true, sometimes are aptly so called. Although ironic, this is not especially surprising. What is “bad” about lying, prototypically understood, is not that it involves saying something believed to be false; it is that it involves the intention to deceive the addressee into believing what is said (and believed by the speaker to be false) to be true. This would, after all, put the trusting addressee at a clear epistemic disadvantage vis-à-vis the speaker.²⁴ This twofold irony, even if easily explained, leads to a puzzle. As cases of coerced and bald-faced lies suggest, not all lies (in the strict vs. prototypical sense) involve deception. Yet while we don’t typically call someone who was coerced into lying a “liar,” we often do call bald-faced liars “liars”—perhaps especially if they are prone to such “in your face” lies. We tend to condemn their behavior, while not condemning the behavior of those who have been coerced into lying. Why might that be? It cannot be that while the one involves deceptive intent, the other does not, as neither typically appears to.²⁵ Yet deceptive intent would appear to be, as just noted, a significant part of what makes lying morally wrong—and thus deserving of the pejorative descriptor “liar.” I suspect the puzzle arises because we think of someone who tells baldfaced (vs. coerced) lies as morally blameworthy for their verbal behavior and we want to somehow give expression to this thought and we may choose to do this by calling bald-faced liars “liars.” However, this is not because of any intent to deceive—which is again, in at least some cases of bald-faced lies, effectively ruled out. It is, rather, because of the flagrancy and nonchalance with which the bald-faced liar insults the addressee. Not only do they say, willingly say, something that is obviously false, thereby insulting the intelligence of the addressee or at least disrespecting

²⁴ This might suggest that debates over whether lying is “worse than” misleading (or vice versa) are, to some degree, misguided. What makes one act of intentional deception worse than another such act is arguably the content of the deception; the means by which that content is conveyed would seem to be of little, if any, significance. While a straightforward lie might seem more bold, an act of “mere” misleading might seem more sly. Either way, it is arguably a wash when it comes to which form of deception, generally speaking, is “worse.” ²⁵ Insofar as “bald-faced” is interpreted as meaning, or entailing, obvious to all, a bald-faced liar, who knows they are a bald-faced liar, cannot possibly (or at lease practically) intend to deceive the addressee into believing what they say to be true.

 , “   ”

173

them, they do so without any shame or embarrassment. In this way, they disrespect the addressee twice over—a kind of “adding insult to injury.” By calling such persons “liars” we express our moral disdain for their verbal duplicity—even if we cannot accuse them of any genuine deception. In contrast to the person who, like A, lies under coercion, the baldfaced liar who claims that, “since I wasn’t trying to deceive anyone I can hardly be called a liar,” is likely to be met with eye-rolling and perhaps an alternative, and more colorful, epithet.²⁶ A small but interesting point is suggested by the remark made by C after they have lied loosely speaking (and with deceptive intent) through conversational implicature. C anticipates, and tries unsuccessfully to escape, the accusation of being a liar by insisting that they had only “just said” they were moving to Phoenix that day. This particular remark, in contrast to the previous deception it seeks to justify, is arguably a bald-faced lie—one that trades on an ambiguity in what it means to “say something.” Thus, B’s accusing C of being “such a liar” might be justified twice over. This will take some explaining. There is a roughly Gricean sense of saying (adopted in the proposed construal of lying strictly speaking) that is tied closely to literal meaning, and then there is another “looser” sense of saying, one that approximates the kind of saying captured in indirect speech reports, such as (e.g.) “C said that they weren’t moving to Phoenix, that they were happy here in Tucson.” B might have provided just such a report, arguably an accurate report, after C’s patently sarcastic remark about moving to Phoenix that day and hating it in Tucson. Thus, when B responds to C’s remark about what they had “just said” by calling C “such a liar,” it is not implausible to suppose that they would have had a twofold rationale for that response.²⁷ First, C’s initial implicature was carried out with deceptive intent. Second, C’s patently equivocal follow-up amounted to a bald-faced lie, given the Gricean sense of “saying” that C was clearly intending.

²⁶ This would be the likely response to Carson’s (2006) case of the student who denies to the Dean that they cheated, while knowing full well that the Dean knows, and knows that they know that she knows, that they are lying. ²⁷ This is not to suggest that B was necessarily conscious of this twofold rationale.

174  

6.7 Taking cues from ordinary ways of speaking If there is a general moral to be drawn from the observations, reflections, and arguments of the present paper, it concerns the importance of ordinary ways of thinking and speaking to philosophical “analyses” of ordinary everyday concepts like lying (knowledge, truth meaning, etc.). The point, however, is not the simple (though important) one of not straying too far from ordinary usage in one’s analysis. This is, of course, essential insofar as the theorist wishes to avoid charges of stipulation or ad hocery. The point, much in the spirit of the ordinary language philosophy of J. L. Austin, is rather that the distinctions, complexities, and nuances that analytic philosophers seek to uncover and thence to articulate are already present, often explicitly so, in ordinary everyday ways of thinking and speaking. Philosophical “definitions” of lying are a case in point. While we (as analytic philosophers) would not want our definitions (our “analyses”) of lying and related phenomena to run roughshod over theoretically important distinctions, we shouldn’t pretend that sensitivity to such distinctions is simply the result of philosophical “analysis” or especially nuanced reflection. Indeed, such distinctions are often explicit in the ordinary conversations of ordinary folk, as the remarks that open the present paper are designed to show. Those remarks, when situated in the contexts that give rise to them, can be seen as invoking, whether implicitly or explicitly, distinctions between (e.g.) strict and prototypical senses of lying, between strict and loose senses of lying, between saying and implying, and between saying in a roughly Gricean sense and saying in the looser sense operative in indirect speech reports.²⁸ If ordinary speech involves subtle distinctions and other complexities, then obviously ordinary thinking does as well. Hence, the significance of the informal yet revealing remarks that open the present paper—remarks that suggest that ordinary folk, even in their unreflective moments, are well poised to make the distinctions that inform philosophical analyses— analyses which are intended to articulate strictly (literally, precisely, ²⁸ This last distinction, as suggested in Section 6.6, is relevant to C’s equivocal claim that they had “just said” they were moving to Phoenix that day.

 , “   ”

175

exactly) what a given concept means. This should hardly come as a surprise given that the philosophers constructing those analyses are ordinary language speakers themselves (both chronologically and logically²⁹) long before they are analytic philosophers.

References Carson, T. L. (2006) “The Definition of Lying.” Noûs 40(2), 284–306. Coleman, L., and P. Kay (1981) “Prototype Semantics: The English Word Lie.” Language 57(1), 26–44. Grice, H. P. (1975) “Logic and Conversation.” In The Logic of Grammar, D. Davidson and G. Harman (eds.). Encino, CA: Dickenson, 64–75. Mahon, E. (2015) “The Definition of Lying and Deception.” In Stanford Encyclopedia of Philosophy, https://plato.stanford.edu/entries/lyingdefinition/. Sorensen, R. (2007) “Bald-Faced Lies! Lying without the Intent to Deceive.” Pacific Philosophical Quarterly 88(2), 251–64. Wiegmann, A., and J. Meibauer (2019) “The Folk Concept of Lying.” Philosophy Compass 14(8). John Wiley & Sons. Wiegmann. A., P. Willemsen, and J. Meibauer (2021) “Lying, Deceptive Implicatures, and Commitment.” Ergo 8(50). https://doi.org/10.3998/ ergo.2251.

²⁹ One cannot possibly “do” analytic philosophy without knowing a natural language in which to do it.

7 À Propos de Pierre, Does He . . . or Doesn’t He? Nathan Salmón

Six years after his landmark Princeton lectures on “Naming and Necessity”, Saul Kripke filled a much needed-to-be-filled gap with his landmark 1976 lecture on “A Puzzle about Belief ”.¹ Kripke there defended Millianism against an oft-made but misplaced objection. Contrary to a common misconception, Kripke does not endorse Millianism—indeed, he adamantly asserted the denial of one of its consequences—but his defense is brilliantly forceful. Kripke hypothesized a normal French speaker, Pierre, who comes to speak English through immersion—and, as it happens, who scrupulously avoids contradicting himself. Unaware that the cities he calls ‘London’ and ‘Londres’ are one and the same, Pierre assents, reflectively and non-reticently, to ‘Londres est jolie’ but dissents from ‘London is pretty’, assenting instead to ‘London is not pretty’. Kripke asks, ‘Does Pierre believe that London is pretty?’ The Millian, such as myself, answers that Pierre unfortunately believes both that London is pretty and that it is not. The astute Millian concludes that Pierre even knows he believes both. How then to accommodate the fact that Pierre would never knowingly believe contradictory things? In Frege’s Puzzle (Salmón 1986) and in “Being of Two Minds: Belief with Doubt” (Salmón 1995), I analyze the notion, x believes p, where p is a proposition, as

¹ Kripke (1979). See https://quoteinvestigator.com/2010/05/03/fills-a-much-needed-gap-pt2/.

Nathan Salmón, À Propos de Pierre, Does He . . . or Doesn’t He? In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Nathan Salmón 2023. DOI: 10.1093/oso/9780198892724.003.0007

̀   ,   . . .   ’   ? B:

177

∃g[x grasps p by means of g & BEL(x, p, g)],

where the variable ‘g’ ranges over something like ways of taking a proposition, or proposition guises, and BEL is a ternary relation of cognitive inward assent that a believer bears to a proposition by means of a guise. Analogous ternary relations are posited for certain other propositional attitudes, such as hope, wishing, and fear. I also analyze a notion, x withholds believing p: WB:

∃g[x grasps p by means of g & ~BEL(x, p, g)].

One can withhold belief, in my sense, by disbelieving (believing the denial). Pierre both believes and withholds believing that London is pretty. Taking the proposition that London is pretty one way, g, Pierre believes it. Taking it another way, g 0 , he withholds belief—precisely by believing that London is not pretty, taking the negated proposition in a particular way, Neg(g 0 ). Pierre is acquitted of irrationality because the guises through which he believes and withholds believing the same thing are distinct. We are acquitted of self-contradiction because WB is consistent with B. Alas, the problem is not yet solved. Kripke suggested a way to fortify his puzzle. Suppose Pierre reflectively, non-reticently assents to ‘Londres est jolie’, just as before, but now he neither assents to nor dissents from ‘London is pretty’, pleading ignorance. Asked ‘Is London pretty?’ he says he has no opinion, that he neither believes nor disbelieves. The grounds for saying that Pierre believes that London is pretty now seem exactly counterbalanced by grounds for saying he does not. Although one can both believe and withhold believing the same thing—as Pierre does—in the name of all that is decent, no one can both believe and not believe the same thing. If we say Pierre does not believe because he suspends judgment, how shall we express that he also believes? If we say he believes, how shall we express that he suspends judgment?²

² David Kaplan was first to urge the importance to belief attribution of suspension of judgment. See Kaplan (1971), at section XI, pp. 141–2. Kaplan’s concern here is with suspension of judgment of a, de re, that it/they F (specifically, of Ortcutt that he is a spy); Kripke’s concern is

178  ´  Withheld belief, in my sense, is not to be confused with suspension of judgment, which is strictly a stronger notion. Yet it is still not so strong as to entail failure to believe: SJ: ∃g[x grasps p by means of g & x grasps ~p by means of Neg(g) & ~BEL(x, p, g) & ~BEL(x, ~p, Neg(g))]. Crucially, SJ is—unlike failure to believe, ~B—perfectly compatible with B. One can suspend judgment concerning p, in my sense, while also believing p. Pierre does both.³ Belief attributions do not designate any particular way of taking the proposition—any particular guise—nor do guises figure in the semantic content. Guises do not play the semantic role of a Fregean Sinn or a Millian connotation. Nor do they play the semantic role of a Fregean Bedeutung or a Millian denotation. But belief attributions do traffic in guises, by existentially generalizing over them. Some philosophers of cognition oppose any involvement in guises, however honorably intended. They seek instead a pure “referentialist” picture, one that says yes to denotations but, unlike Mill, just says no to connotations, and even to utterly non-semantic guises. Pierre knows that he believes that London is pretty, but he also sincerely says in English that he does not. Does the latter fact perhaps provide a way to evade guises? In Crawford (2004), Sean Crawford proposes that my notion WB be replaced by a notion of second-order belief, which is also compatible with believing but which evidently avoids guises and BEL altogether, at least if believing does: B~B: x believes that ~(x believes p). Crawford claims to reject my notion of withheld belief. It might be more accurate, however, to say instead that Crawford proposes an alternative with de dicto suspension of judgment concerning whether Fa (specifically, whether London is pretty). From the perspective of Millianism, these are two sides of the same coin, a distinction without a difference. ³ Cf. Salmón (1986), pp. 111, 172n1.

̀   ,   . . .   ’   ?

179

analysis of the relevant notion in terms of second-order belief in lieu of BEL. Suspension of judgment concerning p could then be analyzed thus: x believes that ~(x believes p) & x believes that ~(x believes ~p).⁴ In this way, ‘Ralph suspends judgment concerning whether Ortcutt is a spy’ can be true despite Ralph’s also believing of Ortcutt as the man in the brown hat, that he is a spy,⁵ and ‘Pierre suspends judgment concerning whether London is pretty’ can be true despite Pierre’s remaining steadfast that Londres is pretty.⁶ One positive feature of Crawford’s clever proposal is that it easily extends to notions of withheld wishing, withheld hoping, withheld fearing, and so on, by assigning pride of place to meta-belief. One can say, for example, that x withholds hope from p iff x believes that ~(x hopes p). One can also say that x suspends hope concerning p iff: (x withholds hope from p) & (x withholds hope from ~p). On the other side, I contend against Crawford that withheld belief is no more a matter of one’s beliefs about one’s beliefs than ordinary belief is. Typically, if x believes p then x believes that x believes p, but this is not invariably true. In any case, as will become clear presently, it is possible for x to believe mistakenly that x believes p. The slippage between firstorder and second-order belief points to a fundamental flaw in Crawford’s attempt to avoid guises. I can accept that Crawford’s preferred notion B~B is extensionally equivalent to WB if, but only if, the following biconditional obtains: WB $ B~B. There’s the rub. I flatly reject the right-to-left conditional, B~B ! WB. For B~B will be true by my lights if x mistakenly believes that person

⁴ This is significantly different from the analysis Crawford actually provides, and also from the one I proposed in “Being of Two Minds”. Arguably, these differences are merely a matter of terminology. ⁵ See Quine (1971), and n. 3 above. ⁶ Kripke (1979), at p. 258. See also my Salmón (1998), especially at pp. 108–10. Both this article and “Being of Two Minds” are reprinted in Salmón (2007).

180  ´  [pointing to an image in a large, wall-size mirror] is not themself but someone else altogether, while (for whatever reason) x also mistakenly believes that that person [same mirror image] does not believe p. This can occur even when x does not withhold belief from p, in the relevant sense. This consideration also shows that B~B is in fact inadequate to Crawford’s objective of capturing the cognitive state of the believer who also withholds belief. Furthermore, it is difficult to see how B~B could be strengthened sufficiently without in the end appealing to something along the lines of a guise. In particular, Héctor-Neri Castañeda’s construction x believes that they themself do not believe p (or ‘x believes that they* do not believe p’) evidently attributes a secondorder belief by means of a first-person, de se guise.⁷ Since the relevant notion of withheld belief evidently unavoidably invokes guises, the sort of consideration raised here is not merely a peripheral point but in fact central to the main issue at hand. Guises are not semantic values of belief attributions, but they are something.⁸, ⁹

References Atkins, P. (2017) “A Russellian Account of Suspended Judgment,” Synthese, 194, pp. 3021–46. Crawford, S. (2004) “A Solution for Russellians to a Puzzle about Belief,” Analysis, 64, 3, pp. 223–9. Kaplan, D. (1971) “Quantifying In,” in Linsky, ed., pp. 112–44. Kripke, S. (1979) “A Puzzle about Belief,” in A. Margalit, ed., Meaning and Use (Dordrecht: D. Reidel, 1979), pp. 239–83. Linsky, L., ed. (1971) Reference and Modality (Oxford: Oxford University Press, 1971).

⁷ Castañeda’s original “quasi-indexical” constructions are ‘he himself ’ and ‘he*’. ⁸ The criticism of Crawford presented here is different from those presented in a couple of insightful essays: Philip Atkins (2017), at pp. 3032–5; and Chris Tillman (2005). ⁹ I am grateful to Teresa Robertson Ishii for discussion, and to Clairol for their catchphrase.

̀   ,   . . .   ’   ?

181

Quine, W. V. O. (1971) “Quantifiers and Propositional Attitudes,” in L. Linsky, ed., pp. 101–11. Salmón, N. (1986) Frege’s Puzzle (Atascadero, CA: Ridgeview, 1986, 1991). Salmón, N. (1995) “Being of Two Minds: Belief with Doubt,” Noûs, 29, 1, pp. 1–20. Salmón, N. (1998) “Is De Re Belief Reducible to De Dicto?” in A. Kazmi, ed., Meaning and Reference (University of Calgary Press, 1998), pp. 85–110. Salmón, N. (2007) Content, Cognition, and Communication: Philosophical Papers II (Oxford University Press, 2007). Tillman, C. (2005) “A Millian Propositional Guise for One Puzzling English Gal,” Analysis, 65, 3, pp. 251–8.

8 The Schmidentity Strategy Jeff Speaks

8.1 The schmidentity strategy in action The schmidentity strategy derives its name from the following wellknown passage in Naming and Necessity: It is, for example, thought that if you have two names like ‘Cicero’ and ‘Tully’ and say that Cicero is Tully, you can’t really be saying of the object which is both Cicero and Tully that it is identical with itself. On the contrary, ‘Cicero is Tully’ can express an empirical discovery. . . . And so some philosophers, even Frege at one early stage of his writing, have taken identity to be a relation between names. Identity, so they say, is not the relation between an object and itself, but is the relation which holds between two names when they designate the same object. . . . If anyone ever inclines to this particular account of identity, let’s suppose we gave him his account. Suppose identity were a relation in English between the names. I shall introduce an artificial relation called ‘schmidentity’ (not a word of English) which I now stipulate to hold only between an object and itself. Now then the question whether Cicero is schmidentical with Tully can arise, and if it does arise the same problems will hold for this statement as were thought in the case of the original identity statement to give the belief that this was a relation between the names. If anyone thinks about this seriously, I think he will see that therefore probably his original account of identity was not necessary, and probably not possible, for the problems it was originally meant to solve, and that therefore it should be Jeff Speaks, The Schmidentity Strategy In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Jeff Speaks 2023. DOI: 10.1093/oso/9780198892724.003.0008

  

183

dropped, and identity should just be taken to be the relation between a thing and itself. This sort of device can be used for a number of philosophical problems.¹

In this passage, Saul Kripke considers an objection to the semantic hypothesis that ‘is’ in sentences like ‘Cicero is Tully’ expresses the identity relation. The objection is that sentences of this form ‘can express an empirical discovery.’ Roughly, the idea is that in some cases speakers may need to resort to empirical evidence to discover whether sentences of this kind are true. Why is this thought to be an objection to semantic hypothesis in question? The thought is that if the hypothesis were true, sentences of this form would be trivial if true, and hence would never, in the above sense, ‘express an empirical discovery.’ So, if the hypothesis were true, it would not be the case that speakers would sometimes need to resort to empirical evidence to discover whether such a sentence is true. But they do. So the hypothesis is false. Kripke’s response is an argument against the conditional claim about what would be the case were the hypothesis true. We begin by stipulating that ‘is schmidentical’ expresses the relation of identity. We then notice that, if we introduced this predicate into our language, ‘the question whether Cicero is schmidentical with Tully can arise.’ That is, speakers would sometimes need to resort to empirical evidence to discover whether this sentence is true. But ‘is schmidentical’ was stipulated to express the identity relation. So if ‘is’ in English sentences like ‘Cicero is Tully’ expressed identity, presumably the same would be true of ‘Cicero is Tully.’ But then the conditional claim on which the objection relies is false, and the objection fails. Kripke lays out this form of argument more explicitly in ‘Speaker’s Reference and Semantic Reference’: I propose the following test for any alleged counterexample to a linguistic proposal: If someone alleges that a certain linguistic phenomenon in English is a counterexample to a given analysis, consider a ¹ Kripke (1972, 108).

184   hypothetical language which (as much as possible) is like English except that the analysis is stipulated to be correct. Imagine such a hypothetical language introduced into a community and spoken by it. If the phenomenon in question would still arise in a community that spoke such a language (which may not be English), then the fact that it arises in English cannot disprove the hypothesis that the analysis is correct for English.²

In each of the cases we’ll consider, we have a semantic hypothesis H and a piece of evidence E which is proposed as an objection to H. The schmidentity strategy is a way of defending H against the objection. A face-value reading of the passage just quoted would suggest that an instance of the strategy would consist of a defense of the relevant instance of the following claim: P In the nearest world in which H is stipulated to be true, E would be true. The strategy also seems to assume that all instances of the following schema are true: N E If in the nearest world in which H is stipulated to be true, E would be true, then E is not evidence against H. Given that N E seems, at a first glance, plausible,³ a natural thought is that the question of when the schmidentity strategy can be employed boils down to the question of which instances of P are true.⁴

² Kripke (1977, 265). Emphasis in original. ³ It is worth noting that, even if it is plausible, N E is certainly not trivial. On one reading it is an instance of the more general claim that if P □! Q then it is not the case that Q is evidence against P. On a probability-raising conception of evidence this amounts to the claim that P □! Q entails that it is not the case that P(Q|P) < P(Q). But some would reject this sort of connection between probabilities and the truth conditions of counterfactuals. I set worries of this general sort to the side for purposes of this paper. ⁴ To be sure, there is a reading of Kripke’s discussion of these topics which makes the intended argument more ambitious than the one just outlined. In the passage quoted from Naming and Necessity, Kripke says, ‘If anyone thinks about this seriously, I think he will see that therefore probably his original account of identity was not necessary, and probably not possible’

  

185

In the original case, we have: H:

A predicate R expresses classical identity.

E: Sentences of the form ⌐ N R M ¬ are sometimes both true and such that some speakers who understand the sentence can’t know whether the sentence is true without empirical evidence. For the reasons Kripke gives, the instance of P for these values of H and E seems true. This is, intuitively, an extremely compelling form of argument. And it would appear to have wide application. Let’s briefly look at a few other well-known applications of the strategy. The first is Kripke’s defense of a Russellian theory of descriptions against an objection based on the existence of referential uses of definite descriptions. In response to this objection, Kripke has us consider a language in which sentences containing definite descriptions are taken to be abbreviations or paraphrases of their Russellian analyses: for example, ‘The present king of France is bald’ means . . . ‘Exactly one person is at present king of France, and he is bald,’ or the like. Would the phenomenon Donnellan adduces arise in communities that spoke these languages? Surely speakers of these languages are no more infallible than we. They too will find themselves at a party and mistakenly think someone is drinking champagne even though he is actually drinking sparkling water. . . . they will say, ‘The man in the corner drinking champagne is happy tonight.’ They will say this precisely because they think, though erroneously, that the Russellian truth conditions are satisfied. Wouldn’t we say of these speakers that they are referring to the teetotaler, under the misimpression that he is drinking

(emphasis added). What does Kripke mean when he says that the original account is, in addition to being unnecessary, ‘not possible’? One reading is that he thinks that the schmidentity example is not just a defense of the hypothesis that ‘is’ expresses identity against the objection that identity sentences can be informative but also a positive argument against the competing hypothesis that it expresses a relation between names. I set this idea to the side here. For discussion, see the criticism of the ‘schmidentity attack’ in Ramachandran (1995, §§III–IV).

186   champagne? And, if he is happy, are they not saying of him, truly, that he is happy? Both answers seem obviously affirmative.⁵

Here the relevant hypothesis and evidence seem to be as follows: H:

Russell’s theory of descriptions is true.

E: On some occasions in which a speaker utters a sentence ⌐ The F is G ¬ we will take the speaker to have referred to o, and said something true about o, even though o is not in the extension of F. Again, for the reasons Kripke gives, it seems plausible that the relevant instance of P is true.⁶ But then if, as seems plausible, the relevant instance of N E is true, it follows that the existence of referential uses of descriptions is, not just weak evidence against Russell’s theory, but no evidence at all. Next, let’s consider Nathan Salmón’s use of the schmidentity strategy to defend the Millian theory of names against the objection that substitution of co-referring names in the complements of attitude ascriptions can seem to change the truth-value of the ascription. Salmón claims that this phenomenon

⁵ Kripke (1977, 265–6). This is what Kripke calls the ‘intermediate Russell language.’ The differences between this language, and the weak and strong Russell languages, won’t matter much for present purposes. ⁶ Devitt (1981, §6) objects that this instance of P is false: ‘Russell English differs from English in that definite descriptions are not devices in it for the designational mode of identifying reference. However, other devices for that mode remain: demonstratives, pronouns, even names. So if the speaker has a particular object in mind (the teetotaler) and means to refer to him, he will use one of those devices, not a definite description like “the man in the corner drinking champagne”. Speakers of Russell English will behave differently from speakers of English’ (520). (Here ‘Russell English’ is a language like ours but for the fact that Russell’s theory of descriptions is stipulated to be true.) But this argument, I think, neglects the fact that there may be pragmatic reasons for speakers of Russell English to use definite descriptions rather than these other referential devices. Suppose, for instance, that we know of no name for the teetotaler, and that our conversation is taking place after the party, when he’s no longer around to be demonstrated. Then it will be very natural for a speaker of Russell English to use the definite description as a way of picking out the teetotaler, and it will be very natural to describe the speaker as referring to the teetotaler even if he is not drinking champagne. It is not hard to come up with other cases of this kind. For more discussion of Devitt’s criticism, see Ramachandran (1995 and 1996).

  

187

would arise even in a language for which it was stipulated—say, by an authoritative linguistic committee that legislates the grammar and semantics of the language, and to which all speakers of the language give their cooperation and consent—that [Millianism] is correct. Suppose, for example, that such a committee decreed that there are to be two new individual constants, ‘Schmesperus’ and ‘Schmosphorus’. (I am deliberately following the genius as closely as possible.) It is decreed that these two words are to function exactly like the mathematician’s variables ‘x’, ‘y’, and ‘z’ as regards information value, except that they are to remain constant . . . the constant value of the first being the first heavenly body visible at dusk and the constant value of the second being the last heavenly body visible at dawn. Suppose further that some English speakers—for example, the astronomers—are aware that these two new constants are co-referential, and hence synonymous. Nevertheless, even if our character Jones were fully aware of the legislative decree in connection with ‘Schmesperus’ and ‘Schmosphorus’, he would remain ignorant of their co-reference. Jones would dissent from such queries as ‘Is Schmesperus the same heavenly body as Schmosphorus?’ Would those who are in the know—the astronomers—automatically regard the new constants as completely interchangeable, even in propositional-attitude attributions? Almost certainly not. . . . speakers who agree to abide by the legislative committee’s decree about ‘Schmesperus’ and ‘Schmosphorus’ and who recognize that these two terms are co-referential—especially if these speakers do not reflect philosophically on the implications of the decree—might for independent pragmatic reasons be led to utter or to assent to such sentences as . . . ‘Jones believes that Schmesperus is Schmesperus, but he does not believe that Schmesperus is Schmosphorus.’ The astronomers may be led to utter the latter sentence, for example, in order to convey (without knowing it) the complex fact about Jones that he agrees to the proposition about Venus that it is it, taking it in the way he would were it presented to him by the sentence ‘Schmesperus is Schmesperus’ but not taking it in the way he would were it presented to him by the sentence ‘Schmesperus is Schmosphorus’. The astronomers would thus unknowingly speak in a way that conflicts with the usage to which they

188   have agreed. This, in turn, would lead to their judging such belief attributions as ‘Jones believes that Schmesperus is Schmosphorus’ not only inappropriate but literally false, and to the unmistakable feeling that substitution of ‘Schmosphorus’ for (some occurrences of ) ‘Schmesperus’ in such attributions as ‘Jones believes that Schmesperus is Schmesperus’ is logically invalid. Insofar as the same phenomena that give rise to Frege’s puzzle about identity sentences and to the appearance of substitutivity failure would arise even in a language for which [Millianism] was true by fiat and unanimous consent . . . these phenomena cannot be taken to refute the theory.⁷

The pattern is (by design, of course) just the same. The relevant instances of H and E are as follows: H:

Millianism about names, along with some compositionality assumptions, is true.

E: Competent speakers of the language sometimes judge that attitude ascriptions have different truth-values despite those ascriptions differing only in the substitution of names which those speakers know to be coreferential. Again, it seems quite plausible that the relevant instance of P is true: as Salmón says, even speakers who were aware of the relevant linguistic decrees would exhibit much of the linguistic behavior used as evidence against Millianism. This linguistic behavior is typically regarded as the most significant argument against a Millian treatment of names. Nonetheless, if this instance of the schmidentity strategy is sound, it follows that this linguistic behavior is no evidence at all against Millianism. This is a striking result. Could it really be true that speaker intuitions about substitution failures are no evidence at all against the Millian theory of names?⁸ ⁷ Salmón (1990, 14–15). See also Salmón (1989, §II). ⁸ To be sure, Salmón only makes the weaker claim that intuitions about substitution failures ‘cannot be taken to refute’ Millianism. But, for reasons already given, it looks like the considerations Salmón gives in favor of his conclusion also support the stronger conclusion that speaker intuitions about substitution failures are no evidence at all against Millianism.

  

189

More recently, Salmón has deployed the same strategy to undercut the motivations for the view, which Kit Fine calls ‘semantic relationism,’ that sentences like ‘Cicero is Cicero’ and ‘Cicero is Tully’ differ in meaning despite ‘Cicero’ and ‘Tully’ not differing in meaning.⁹ On this view, in the former type of sentence the pair of names represent the object denoted as the same. Using Fine’s terminology, let’s say that sentences of this type express coordinated propositions whereas the latter do not. The fact that sentences like ‘Cicero is Cicero’ and ‘Cicero is Tully’ seem to differ in meaning is taken to be strong evidence against nonrelationist versions of Millianism. So here we have the following values of H and E: H:

Non-relationist Millianism is true.

E: Competent speakers of the language judge that pairs of identity sentences of the form ⌐ N is N ¬ and ⌐ N is M ¬ systematically differ in meaning. To argue that these intuitions about differences in meaning are not in fact evidence against non-relationist Millian views of names, Salmón imagines some semantically sophisticated speakers stipulating that their language should express no coordinated propositions. So, in the new language, ‘Cicero is Cicero’ and ‘Cicero is Tully’ are stipulated to express the same uncoordinated proposition. But as Salmón points out, these linguistic stipulations will neither remove their knowledge that every name refers to the same thing as itself nor remedy their ignorance of the fact that certain pairs of distinct names refer to the same thing. So, reasoning similar to that given in the passage just quoted suggests that, even after the stipulations, there will be a felt difference in meaning between sentences like ‘Cicero is Cicero’ and sentences like ‘Cicero is Tully.’¹⁰ But that suggests that the relevant instance of P is true. And that, given the relevant instance of N E, implies that the central piece of evidence used to support ⁹ Fine (2007). Related views are defended in Putnam (1953); Richard (1983); Taschek (1995). For criticism, see Salmón (2012), Salmón (2015), and Soames (2015, ch. 6). ¹⁰ Salmón works out the example in considerably more detail in Salmón (2018, §1).

190   semantic relationism against its non-relationist rivals is no evidence at all against the latter.

8.2 The strategy extended We have already seen that the schmidentity strategy has wide application. But a plausible case can be made that the strategy can be extended yet further, in ways which might seem to cast some doubt on the strategy itself. Let’s consider first the linguistic hypothesis that the semantic contents of sentences, and the denotations of that-clauses, are sets of possible worlds. A very well-known objection to this view is that it makes semantic contents too coarse-grained to be the objects of propositional attitudes. One way to press the objection is to point out that speakers are often inclined to regard a pair of attitude ascriptions as differing in truthvalue despite the fact that the complements of the ascriptions are true in just the same possible worlds.¹¹ On its face, the fact that speakers are strongly inclined to regard pairs of ascriptions of this sort as sometimes differing in truth-value is evidence against the idea that the semantic contents of sentences, and denotations of that-clauses, are sets of worlds. Might the schmidentity strategy come to the aid of the proponent of possible worlds semantics? It seems that it can. Here we have the following values for H and E: H:

The semantic contents of sentences, and the denotations of thatclauses, are sets of possible worlds.

E:

Competent speakers of the language sometimes judge that attitude ascriptions have different truth-values despite those ascriptions differing only in the substitution of sentences which those speakers know to be true in just the same possible worlds.

It seems that here we can follow Salmón’s lead, and imagine that the speakers of a language give their assent to a governing linguistic

¹¹ For a well-known presentation of the objection, see Soames (1988).

  

191

committee. We can further imagine that the committee, being very impressed with the beauty and simplicity of intensional semantics, decrees that the semantic contents of sentences are to be sets of possible worlds. (We can imagine this to have been accompanied by other decrees with respect to subsentential expressions of various sorts.) We may imagine that speakers of the language agree that, from this point forward, there are to be no differences in meaning without a difference in truth conditions. This would then be a world in which H is stipulated to be true. We can now ask whether, in this world, E would still be the case. And it seems that it would be. For consider the sort of stock example often used in arguments against the possible worlds view of propositions. Let Violet be a mathematically advanced middle school student. She will, of course, assent to such sentences as ‘2 + 2=4’ and ‘9  7=63.’ Now suppose that Violet is asked, ‘Are there infinitely many prime numbers?’ Violet might well understand what a prime number is, and so understand the question perfectly well. But she might also think that, because prime numbers become less and less common as numbers get larger, they must run out at some point; so she might say ‘No.’ Now, those who are both familiar with the intensionalist linguistic decree and know more about prime numbers than Violet will be well aware that ‘2 + 2=4’ and ‘There are infinitely many prime numbers’ mean the same thing in the language which they and Violet speak. But they might, for ‘independent pragmatic reasons,’ be inclined to say things like ‘Violet knows that 2 + 2=4, but doesn’t know that there are infinitely many primes.’ They might do this to convey facts about (e.g.) the mathematical sentences which Violet is disposed to accept. They would thus, like Salmón’s astronomers, ‘unknowingly speak in a way that conflicts with the usage to which they have agreed.’ And we can expect that this habit, in time, might lead them to judge sentences like ‘Violet knows that there are infinitely many primes’ to be not just misleading or inappropriate, but literally false. And so we find that, even in a language in which the possible worlds view of content is stipulated to be correct, speakers will be inclined to judge that some attitude ascriptions differ in truth-value despite differing only in the substitution of sentences which those speakers know to be true in just the same possible worlds. But that, it seems, is enough for the

192   relevant instances of P and N E to be true. And that allows us to derive, via the schmidentity strategy, the highly surprising result that speaker intuitions about the truth conditions of attitude ascriptions are no evidence at all against the thesis that the semantic contents of sentences are sets of worlds. One may have one two responses to this example. On the one hand, one may take this to be unexpectedly good news for the possible worlds semanticist. But, on the other hand, one may take this very unexpected piece of good news to be an indication that something has gone awry with our application of the schmidentity strategy. I want now to consider another example, which I think suggests that the second response is more likely to be correct. Impressed with the success of their intensionalist decree, we can imagine that our friends on the linguistic committee from the previous example decide to simplify the semantics of their language yet further. They might therefore decree that the semantic contents of sentences, and the denotations of thatclauses, be not sets of possible worlds but truth-values. (We can further imagine that they make corresponding decrees with respect to subsentential expressions; they decree, for instance, that the semantic contents of monadic predicates will be, not functions from worlds to functions from individuals to truth-values, but simply functions from individuals to truth-values.) We can further suppose that the speakers of the language consent to the decree. Despite the assent which this linguistic stipulation receives, would speakers of the language still sometimes judge that attitude ascriptions have different truth-values despite those ascriptions differing only in the substitution of sentences which those speakers know to have the same truth-value? Despite the fact that these ascriptions would, given the linguistic stipulation, have just the same semantic content, the line of reasoning present in our various instances of the schmidentity strategy seems to show that they would. And that seems to be enough to show that the relevant instance of P is true. The problem is that the result to which this leads—that speaker intuitions about the truth-values of various attitude ascriptions are no evidence against our now extremely coarse-grained semantics—seems quite implausible. The schmidentity strategy seems to generalize so easily

  

193

that it can be used to defend virtually any semantic hypothesis which makes semantic values comparatively coarse-grained against speaker intuitions which would seem to cut in the opposite direction. But surely speaker intuitions of this kind can sometimes be evidence against semantic hypotheses. So the examples discussed in this section leave us with a kind of dilemma. On the one hand, the relevant instances of both P and N E look quite plausible. On the other hand, they seem to deliver incorrect results when applied to the hypotheses and objections just considered.

8.3 ‘Stipulated to be true’ Both P and N E include the phrase ‘stipulated to be true.’ One diagnosis of our current dilemma is that it is based on a conflation of two interpretations of this phrase.

8.3.1 Stipulations within worlds On the first interpretation, to consider the nearest world in which a certain linguistic hypothesis is stipulated to be true is to consider the nearest world in which speakers of a language make a certain sort of overt linguistic stipulation. We might thus imagine, as Salmón does, some sort of committee issuing decrees of certain sorts. 8.3.1.1 Efficacious vs. inefficacious stipulations To go this route is to imagine the stipulation taking place within the relevant possible world. The problem with this route is that linguistic stipulations are not always successful; the fact that someone, or some committee, stipulates that a class of expressions has some linguistic property is no guarantee that the expressions have that property.¹² Let’s say that a linguistic stipulation is efficacious iff the relevant ¹² A related worry is raised in passing in Ramachandran (1989, 468).

194   expressions have, after the stipulation, the semantic properties they are stipulated to have. Once the possibility of inefficacious stipulations is noticed, it is fairly clear that what the proponent of the schmidentity strategy needs to consider is not the nearest world in which some linguistic stipulation is made, but the nearest world in which that stipulation is made and is efficacious. Otherwise the relevant instance of N E would lose all plausibility. The fact that some bit of linguistic behavior would occur in a world in which a semantic hypothesis is false but claimed by some speakers to be true would not, on its face, imply much about whether that linguistic behavior is evidence against the hypothesis. The proponent of the schmidentity strategy might reasonably concede this point without thinking that it matters much for the success of the strategy. For surely there is a presumption that linguistic stipulations are successful unless the world fails to cooperate in some unexpected way. Sure, it can happen that my stipulation that ‘Vulcan’ refer to the planet causing perturbations in the orbit of Mercury is inefficacious because there is no such planet. But it is not as though we have attempts to name children and pets constantly going south; in the ordinary case, linguistic stipulations are efficacious. This is, I think, an entirely fair point about the eponymous use of the schmidentity strategy. There is little doubt that the identity relation exists, and little doubt that we are capable of having thoughts about this relation. Given these facts, it is entirely obscure why we should be incapable of introducing a term to stand for this relation. Identity is more abstract than my dog; but it is hard to see why that fact should make introducing terms for these things all that different. But our later uses of the schmidentity strategy differ both from this initial use and from ordinary linguistic stipulations in two respects. The first is a difference in the linguistic property the stipulation is about. While the distinction is marked using different vocabulary in different approaches to natural language semantics, virtually all such approaches recognize a distinction between the semantic contents of expressions and what those expressions designate or refer to. I’ll use the term ‘semantic value’ for the latter. Of course, it is true that some theories—like Millianism—hold that some expressions have their semantic values as

  

195

their semantic contents. But proponents of these views still endorse the distinction between semantic content and semantic value, even if they think that occasionally one and the same entity plays both roles. Garden-variety linguistic stipulations are stipulations of semantic values. When I introduce a name, I stipulate what the name is to stand for; I don’t, in the ordinary case, make any stipulations with respect to its semantic content. Of course, if my stipulation is efficacious, I will have succeeded in giving the name a semantic content as well as a semantic value. But this is, as regards the stipulation, a somewhat secondary effect. The name introduced comes to have whatever semantic content is needed to secure the wanted semantic value. But consider, by contrast, the linguistic stipulation employed as part of Salmón’s defense of Millianism. There the stipulation is not that a certain expression designate a certain entity; rather, the stipulation is that certain expressions have a certain sort of semantic content. The same is true of Kripke’s defense of the Russellian analysis of definite descriptions, and the imagined defense of possible worlds semantics in the previous section. The second difference between ordinary linguistic stipulations and the stipulations imagined in instances of the schmidentity strategy is a difference in scope. Familiar linguistic stipulations concern only one expression, like ‘Rufus’ (or ‘schmidentity’). By contrast, many of the examples discussed above concern more general linguistic stipulations, which are directed at, not just individual linguistic expressions, but broad categories of linguistic expressions (for example, definite descriptions). So the stipulations in question are stipulations with respect to the types of semantic contents certain broad categories of linguistic expressions have. The problem is that there is reason to think that at least some linguistic stipulations of this kind are bound to fail. 8.3.1.2 Modally stable semantic properties The reason is that certain facts about the semantic properties of expressions in a language seem to be, in the following sense, modally stable: F is a modally stable property of natural language iff if one natural language has F, then all actual and nearby possible languages also have F

196   This definition is obviously vague. But it will be enough to make clear the main points that follow. If a certain property of a natural language is modally stable in this sense, then it is plausible that it is not sensitive to relatively local facts about linguistic usage. Among the relatively local facts about linguistic usage are facts about the linguistic stipulations speakers of the language have made. Given this, it seems plausible that if a language has a modally stable property F, and we imagine speakers of the language making a stipulation which, if efficacious, would make it the case that the language lacks F, the nearest world(s) in which that stipulation is made is/are world(s) in which the stipulation is inefficacious. Let’s say that stipulations which are inconsistent with some modally stable property of a language conflict with that property. Which properties of languages are modally stable? Consider the view of a structured propositions theorist who thinks that the semantic contents of sentences must be more fine-grained than possible worlds. It would be very odd for this theorist to adopt this view of English, but not of (say) Bengali. Surely if the semantic contents of sentences of English are more fine-grained than sets of possible worlds, the same goes for the semantic contents of sentences of other natural languages. And if this is true of other actual natural languages, it is presumably also true of languages spoken by communities in nearby possible worlds. After all, just the same considerations which count in favor of the view that sentences of English express structured propositions would seem to count in favor of the view that the sentences of languages spoken by merely possible agents who are similar to us do the same. So, it seems plausible that the property expressed by the open sentence sentences of L have structured propositions as their semantic content is modally stable. The same, of course, goes for the opposed view that sentences have sets of worlds as their contents. If this view is true of one natural language, it would seem to hold for all. Note that the proponent of the view that properties of this sort are modally stable need not think that it is metaphysically impossible that languages lack the relevant property. They need only think that speakers

  

197

of a language which lacked the property would have to be quite different than, and therefore relatively modally remote from, actual speakers. Nor is this only true of sentences. It seems plausible that this holds for expressions of any syntactic category. It would, for instance, be bizarre if the subsentential expressions of English had Russellian contents whereas the expressions of German had Fregean contents. Certain facts about the form of the correct semantic theory for a language seem to be modally stable. 8.3.1.3 Modal stability, names, and descriptions The discussion in the preceding subsection is enough to show what is wrong with the imagined uses of the schmidentity strategy discussed in Section 8.2: if the semantic contents of sentences of our language are in fact more fine-grained than sets of worlds, then this property of our language seems to be modally stable, and hence not the sort of property which can be undone by linguistic stipulation. So the defender of the idea that propositions are more fine-grained than sets of worlds would seem to have a principled reason for rejecting the relevant stipulation as inefficacious: it conflicts with what that theorist regards as a modally stable property of our language. This gives the structured propositions theorist a principled reason to think that the relevant linguistic stipulation is inefficacious, and hence to reject the relevant instance of N E. Because the dialectical situation here is a bit complex, it is worth pausing a moment to think about how this line of argument works. We can imagine a theorist undecided as between the hypotheses (H) that the denotations of that-clauses are structured propositions and (H’) that they are sets of worlds. (Suppose for simplicity that these are the only options.) The theorist then encounters some evidence E—in this case, judgements about the truth conditions of certain attitude ascriptions—which seem to favor the former hypothesis over the latter. On the basis of E, they come to believe that H rather than H’ is correct. In reply, we imagine the sets of worlds theorist employing the schmidentity strategy, and arguing, on the basis of the relevant instances of P and N E, that E does not favor H after all, since if H’ were true, E would be true. But the argument for this conclusion relies on the assumption that a stipulation that H’ is true would be efficacious.

198   But because H and H’ are both modally stable properties of languages, this relies on the assumption that H’ is in fact true. Because our theorist believes on the basis of E that H’ is false, they have reason to reject this assumption, and hence reject the relevant instance of N E. This gives an outline of the reasons why a particular instance of the schmidentity strategy might fail. With this in mind, let’s turn to the applications of the schmidentity strategy discussed in Section 8.1. As already mentioned, the considerations just laid out would seem to pose no problem for the original use of the schmidentity strategy. Containing or failing to contain an expression with a certain semantic value is not, in general, a modally stable property of a language. So there would appear to be no modally stable property of our language which conflicts with the stipulation that ‘schmidentity’ have the identity relation as its semantic value. How about Kripke’s defense of the Russellian analysis of definite descriptions? Here, unlike the ‘schmidentity’ case, we have a stipulation with respect to the semantic contents, rather than semantic values, of a class of expressions. This might lead one to assimilate this case to the case of the possible worlds view of propositions. But in this case the stipulation is that definite descriptions have the same meaning as another expression type in our language; a sentence ⌐ The F is G ¬ is stipulated to mean the same thing as ⌐ Exactly one thing is F and that thing is G ¬. Stipulations of this kind—in which one expression is stipulated to abbreviate, and so have the same semantic content of, another—are a somewhat special case. Here the language is already agreed to have an expression with the relevant semantic content. So there is no reason to think that the stipulation that another expression have that content should conflict with any modally stable property of the language. So our present discussion seems to give us no reason to doubt that the relevant instance of N E is false. Let’s consider next Salmón’s defense of Millianism against objections based on apparent substitution failures of coreferential names in the complements attitude ascriptions. Is there a principled reason for the anti-Millian to view the linguistic stipulation involved in this instance of the schmidentity strategy as in conflict with some modally stable property of our language?

  

199

Here there is, I think, no one answer. It depends on the sort of antiMillian view one has. Consider first a Fregean of a fairly traditional stripe, who takes the semantic contents of expressions to be abstracta which are individuated by Frege’s criterion of difference for sense based on distinctions of cognitive significance. This sort of traditional Fregean will regard objects like ‘Mont Blanc with its snowfields’ to be altogether the wrong sort of thing to be the semantic content of a linguistic expression.¹³ To be sure, one might give all sorts of arguments against this kind of view. But the proponent of this view would seem to have a principled reason for thinking, not just that our language contains no Millian terms, but also that this absence of Millian terms is a modally stable property of our language. Salmón’s use of the schmidentity strategy would thus—on the ‘stipulation within worlds’ interpretation—seem to have little force in blocking the use of Frege’s Puzzle as a motivation for this sort of view. The orthodox Fregean has reason to think that the would-be Millian linguistic stipulation is inefficacious, and thus that the relevant instance of N E is false. There are other less wildly anti-Millian views which would seem to have the resources to make the same move. Consider, for example, the Russell of ‘Knowledge by Acquaintance and Knowledge by Description,’ who agrees with the Millian that some singular terms ‘merely and solely name’ objects.¹⁴ But because Russell both held that we can introduce these Millian terms only for the objects of acquaintance and had a restrictive view of the possible objects of acquaintance, he thought that ‘I’ and ‘this’ were the only Millian names for particulars. It is fairly clear that Russell thought that these views about the relationship between naming and acquaintance and about the scope of acquaintance are not simply true due to accidental features of our language. Someone with views like Russell’s would, like the orthodox Fregean, therefore seem to have principled reasons for thinking that Salmón’s imagined linguistic stipulations with respect to ‘Schmesperus’ and ‘Schmosphorus’

¹³ The example, as is well known, is from Frege’s correspondence with Russell. See Gabriel et al. (1980). ¹⁴ Russell (1911, 121).

200   conflict with a modally stable property of our language, and hence are inefficacious. But these two examples don’t mean that Salmón’s argument is without force; this sort of reply to Salmón’s use of the schmidentity strategy would not seem to be open to all non-Millians. Consider a theorist who endorses the (plausible) view that the semantic content of a variable relative to an assignment is the value assigned, but denies Millianism about ordinary proper names. Presuming that the theorist has no idiosyncratic views about the sort of things which can be the values of variables, it is hard to see what principled reason they could have for resisting the idea that stipulations like those in Salmón’s example could be efficacious. It is thus hard to see how—presuming that they agree with Salmón that the relevant instance of P is true—they could take apparent substitution failures in attitude ascriptions to be evidence against Millianism. Consider now Salmón’s use of the schmidentity strategy to undercut the motivations for semantic relationism. Would the stipulations in this case be efficacious? Salmón thinks that it is clear that they would be. After all, classical semantics makes no distinction between coordinated and uncoordinated propositions. But surely, Salmón argues, it is possible to speak a language which fits classical semantics in this respect; as Salmón says, ‘There are no semantic gods to forbid such a language, no semantic police who will hit the back of our hands with a ruler if we set up such a language.’¹⁵ This is not so obvious to me. As above, we can consider the hypothesis that semantic relationism is true of English, and ask whether, given this supposition, it is plausible that semantic relationism is true of every other natural language. It seems overwhelmingly plausible to me that this is so. Just as it would be bizarre to think that English sentences have structured propositions as their semantic contents whereas Bengali sentences have sets of worlds as their contents, it would be bizarre to think that semantic relationism was true of English but not of Bengali. Given this, it seems plausible that being a language of which semantic relationism is true is a modally stable property.

¹⁵ Salmón (2018, 221).

  

201

What, according to the proponent of semantic relationism about English, does it take for semantic relationism to be true of a language? A plausible answer would seem to be that it is true of any language in which it is a linguistic convention that multiple uses of a name all have the same semantic value, and in which it is possible to understand distinct coreferential names without knowing that the names co-refer. But those conditions are satisfied by the post-stipulation language that Salmón imagines.¹⁶ This suggests that there are principled reasons for the semantic relationist to resist Salmón’s use of the schmidentity strategy by denying the relevant instance of N E. They should concede that post-stipulation speakers would exhibit the relevant linguistic behavior, but should deny that the stipulation is efficacious, and for that reason should deny that the post-stipulation existence of the linguistic behavior implies anything about whether that behavior is evidence against non-relationist Millian views of names. In this section, I’ve offered a number of preliminary thoughts about the success of various instances of the schmidentity strategy. But the primary moral of this discussion is methodological. The moral is that, if we think of the relevant linguistic stipulations as happening within the relevant possible worlds, the key question is whether these stipulations would be efficacious. And that, I’ve suggested, depends largely on the question of whether there are any modally stable properties that our language actually has but would lack if the stipulation were efficacious. If one has reason to believe that there are such modally stable properties, that provides reason to deny the relevant instance of N E.

8.3.2 Stipulations about worlds A proponent of the schmidentity strategy might reply to the preceding line of thought by saying that the version of the schmidentity strategy discussed in Section 8.3.1 takes talk about speakers making linguistic ¹⁶ As Salmón notes, ‘It should be noted here that their agreement to speak Schmenglish does not eradicate their knowledge or ignorance, as the case may be, concerning whom the names “Cicero” and “Tully” name. Those who knew that the two names are co-designative in English know that they are equally co-designative in Schmenglish. In particular, their agreement to speak Schmenglish does not obliterate their knowledge that every proper name that designates at all in Schmenglish is exactly synonymous with itself, and hence co-designative with itself ’ (Salmón 2018, 218).

202   stipulations much too seriously. Adapting another well-known lesson from Naming and Necessity, they might say, ‘“Possible worlds” are stipulated, not discovered by powerful telescopes.’¹⁷ There’s thus no reason to consider possible worlds in which speakers make various overt linguistic stipulations and examine those worlds with a semantics-scope to see whether the stipulations are efficacious. Instead, we can simply stipulate ‘from the outside’ that a given world is one in which the semantic hypothesis in question is true. Wouldn’t that be a simple way around the difficulties canvassed above?¹⁸ Perhaps. But there is a worry that the move simply shifts the locus of our difficulties from N E to P. If the linguistic property which conflicts with the relevant stipulation is modally stable, then the nearest world in which the stipulation is true may be one in which users of the language are quite different than us. And in that case, intuitions about what those speakers would think or do are bound to be much less reliable. To see the force of this, consider again Salmón’s defense of Millianism against arguments based on apparent substitution failures in attitude ascriptions, and the stipulation that ‘Schmesperus’ and ‘Schmosphorus’ are Millian names. What would a linguistic community have to be like for this stipulation to be true of it? From the point of view of a Millian, the answer is clear: it would be just like our actual linguistic community. But from the point of view of a Fregean who accepts Frege’s criterion for difference of sense based on distinctions of cognitive significance, it seems that the speakers of a Millian language would have to be quite a bit different than us. On this view, for a language to contain distinct co-referential Millian names n and m it would have to be impossible that (roughly) a speaker understands the names and, on reflection, is unsure whether a pair of sentences differing only with respect to substitution of n and m have the same truth-value.¹⁹ But presumably speakers who satisfy this condition would not ever be inclined to regard attitude ascriptions which differ only in the ¹⁷ Kripke (1972, 44). Emphasis in original. ¹⁸ This second interpretation of the schmidentity strategy arguably fits better with Kripke’s presentation of the argument. ¹⁹ This is just a first pass. A better formulation would, for reasons related to cases like Kripke’s (1979) example of ‘Paderewski,’ have to be more careful about the distinction between expression types and expression tokens. It is not obvious whether this can be done. For some of the difficulties, see Speaks (2013).

  

203

substitution of coreferential Millian names as differing in truth-value. And that would make the relevant instance of E false at that world, thus falsifying the relevant instance of P. This is, of course, just the difficulty raised in Section 8.3.1 in a new form. When a stipulation conflicts with a modally stable property of language, we can either consider the nearest world in which the stipulation is made, or the nearest world in which it is true. If we take the former route, there will often be principled reasons for thinking that the stipulation is inefficacious, which will falsify N E. If we take the latter route, there will often be principled reasons for thinking that E is false at that world, which will falsify P. *

* *

The schmidentity strategy is a powerful form of argument. Given the significance of the conclusions it has been used to defend, it is surprising that it has not received more explicit discussion. My aim has been less to argue for or against the strategy as such, but rather to advance the discussion of what conditions must be satisfied for the strategy to be successful.²⁰

References Devitt, Michael. 1981. “Donnellan’s Distinction.” Midwest Studies in Philosophy 6 (1): 511–26, https://doi.org/10.1111/j.1475-4975.1981. tb00456.x. Fine, Kit. 2007. Semantic Relationism. Malden, MA: Blackwell. Gabriel, Gottfried, Hans Hermes, Friedrich Kambartel, Christian Thiel, Albert Veraart, Brian McGuinness, and Hans Kaal. 1980. Gottlob Frege: Philosophical and Mathematical Correspondence. Oxford: Blackwell. Kripke, Saul. 1972. Naming and Necessity. Cambridge, MA: Harvard University Press. Kripke, Saul. 1979. “A Puzzle about Belief.” In Meaning and Use, ed. Avishai Margalit, 239–83. Boston: D. Reidel Publishing Co. Kripke, Saul A. 1977. “Speaker’s Reference and Semantic Reference.” In Studies in the Philosophy of Language, ed. Peter A. French, Theodore

²⁰ Thanks to Nathan Salmón for helpful discussion of the issues discussed in this paper.

204   E. Uehling, Jr, and Howard K. Wettstein, 2: 255–96. 1. Minneapolis, MN: University of Minnesota Press. Putnam, Hilary. 1953. “Synonymity and the Analysis of Belief Sentences.” Analysis 14 (5): 114–22, https://doi.org/10.1093/analys/14.5.114. Ramachandran, Murali. 1989. “Sense and Schmidentity.” Philosophical Quarterly 39 (157): 463–71, https://doi.org/10.2307/2219830. Ramachandran, Murali. 1995. “Methodological Reflections on Two Kripkean Strategies.” Proceedings of the Aristotelian Society 95: 67–81, https://doi.org/10.1093/aristotelian/95.1.67. Ramachandran, Murali. 1996. “The Ambiguity Thesis versus Kripke’s Defence of Russell.” Mind and Language 11 (4): 371–87, https://doi.org/ 10.1111/j.1468-0017.1996.tb00052.x. Richard, Mark. 1983. “Direct Reference and Ascriptions of Belief.” Journal of Philosophical Logic 12 (4): 425–52. Russell, Bertrand. 1911. “Knowledge by Acquaintance and Knowledge by Description.” Proceedings of the Aristotelian Society 11: 108–28, https:// doi.org/10.1093/aristotelian/11.1.108. Salmón, Nathan. 1989. “How to Become a Millian Heir.” Noûs 23: 211–20. Salmón, Nathan. 1990. “A Millian Heir Rejects the Wages of Sinn.” In Propositional Attitudes: The Role of Content in Logic, Language, and Mind, ed. C. Anthony Anderson and Joseph Owens, 117–39. Stanford, CA: CSLI Press. Salmón, Nathan. 2012. “Recurrence.” Philosophical Studies 159 (3): 407–41. Salmón, Nathan. 2015. “Recurrence Again.” Philosophical Studies 172 (2): 445–57. Salmón, Nathan. 2018. “Cognition and Recognition.” Intercultural Pragmatics 15 (2): 213–35. Soames, Scott. 1988. “Direct Reference, Propositional Attitudes, and Semantic Content.” In Propositions and Attitudes, ed. Nathan Salmón and Scott Soames, 197–239. Oxford: Oxford University Press. Soames, Scott. 2015. Rethinking Language, Mind, and Meaning. Princeton, NJ: Princeton University Press. Speaks, Jeff. 2013. “Individuating Fregean Sense.” Canadian Journal of Philosophy 43 (5): 634–54, https://doi.org/10.1080/00455091.2014.925678. Taschek, William W. 1995. “Belief, Substitution, and Logical Structure.” Noûs 29 (1): 71–95, https://doi.org/10.2307/2215727.

9 Leverage: A Model of Cognitive Significance Stephen Yablo

Analytic semantics got its start when Frege, first in Begriffschrift and then “Uber Sinn und Bedeutung,” began to worry about differences in cognitive content between sentences that in some good sense say the same— sentences attributing the same properties to the same objects, or putting the same objects in the same relations. He was particularly struck by claims like Hesperus = Phosphorus, Hesperus = Hesperus.

(1) (2)

How is it that (1) is more informative than (2), when they both say of the same object (Venus) that it is identical to the same object (Venus)? Similar problems are raised by Hesperus is never far from Phosphorus, Hesperus is never far from Hesperus,

(3) (4)

or, Phosphorus being the Morning (not the Evening) Star, Phosphorus is often seen in the evening, Hesperus is often seen in the evening.

(5) (6)

How can (3) be more informative than (4), or (5) more informative than (6), when they attribute the same features to the same things? Stephen Yablo, Leverage: A Model of Cognitive Significance In: Oxford Studies in Philosophy of Language Volume 3. Edited by: Ernie Lepore and David Sosa, Oxford University Press. © Stephen Yablo 2023. DOI: 10.1093/oso/9780198892724.003.0009

206   Cognitive content is a label for whatever explains/underlies the felt difference in information value; (1), (3) and (5) are more informative than their counterparts by virtue of possessing non-trivial cognitive contents. We’ll start as Frege did with claims of the form a = b.

9.1 Coreferring Names Frege’s response was in two stages. He conjectured in the Begriffsschrift that the sought after contents would be metalinguistic, to the effect that the names involved co-referred in the relevant language. Since the sentences are in English, the informational cash value of (1) is that «Hesperus» and «Phosphorus» co-refer in English,

(7)

while (2)’s cash value is that «Hesperus» and «Hesperus» co-refer in English.

(8)

This runs into a problem, he decided. It’s true that (7) is more informative. But it’s informative about the wrong thing(s). (7) misrepresents what someone learning (1) comes to know: the sentence a = b would no longer refer to the subject matter, but only to its mode of designation; we would express no proper knowledge by its means. But in many cases, this is just what we want to do ([Frege, 1948], 209)

By “proper” knowledge, Frege means knowledge of how matters stand non-linguistically. (1) is informative, he is saying, not primarily about language, but the astronomical facts we use language to state. But then the information we seem to be getting from (1) is not really captured by a claim like (7) which speaks mainly of English words. I stress English words because some have sought to drive the point home using the “Church-Langford translation test” ([Linsky, 1983]).

:     

207

Translation ought presumably to preserve cognitive content. So a contentextraction method that delivers different results when applied to ϕ and its translation cannot be right. If indeed (1) shares a cognitive content with its Italian translation Espero = Fosforo,

(9)

then the analogue of (7) for Espero = Fosforo should be (7) again, or its Italian translation: «Hesperus» e «Phosphorus» co-riferiscono in inglese.

(10)

Is this what the Frege scheme applied (in Italian) to Espero = Fosforo gives us? No, we get instead «Espero» e «Fosforo» co-riferiscono in italiano.

(11)

This is in no way equivalent to (10)! It concerns itself with the referential properties, in a language (10) never considers, of names (10) never mentions.¹

9.2 Coincident Senses Problems like these are what led Frege to his better-known later proposal that (1) owes its non-trivial cognitive content to the fact that «Hesperus»

¹ Here is how Linsky puts it (adjusted for consonance with the main text): In translating (7), it is necessary to leave the quoted material intact, for it is the names «Hesperus» and «Phosphorus» that are said to denote the same thing (in English). Any translation of (7) must be about these very names and not about their Italian translations, «Espero» and «Fosforo». Suppose now that you are an Italian who does not understand any English at all. Then it is clear that (9) and (10) convey different information to you. (9) tells you an important astronomical fact, but (10) tells you nothing about the heavens at all. It tells you, rather, a fact about the English language. It is clear that (9) and (10) have different cognitive content. Then (1) and (7) must also convey entirely different information, and for the same reason. One is about the heavens and the other is about English words. Thus Frege is forced to abandon his earlier view ([Linsky, 1983], p7)

208   and «Phosphorus» differ in sense—in how they present their (identical) referents. Suppose that «Hesperus» has H as its sense while «Phosphorus»’s sense is P. How is the fact that H ≠ P supposed to explain the non-triviality? The idea seems at first to be that Venus-presented-H -ly = Venus-presented-P -ly

(12)

is more informative than Venus-presented-H -ly = Venus-presented-H -ly.

(13)

But the truth is that neither provides a whole lot of information; Venus presented thus-ly is trivially the same object as Venus presented such-ly, whether it’s the same mode of presentation twice over or not. (As a pound of salt weighs the same as a pound of sodium chloride, whether salt = sodium chloride or not.) An informational difference does exist between H and P present the same object,

(14)

and H and H present the same object.

(15)

But this way of formulating the contents runs into the same sort of objection as the Begriffsschrift account. What (1) tells us is no more about modes of presentation than names; perhaps it is less about modes of presentation, since no one pre-Frege had heard of such things, and few have heard of them yet. (1) gives information about astronomy, (14) about abstract objects. The “news” encoded in (1) is not captured therefore by (14), for the same sort of subject-matterish reason that (7) failed to capture that news.²

² (14) does do better with the Church-Langford objection, assuming «Espero» and «Fosforo» agree in sense with their English counterparts.

:     

209

9.3 Definite Descriptions That (14) concerns modes of presentation stems from its containing names «H» and «P» of the senses H and P. This looks at first like an insuperable difficulty, since it is precisely H’s distinctness from P that confers on (14) its greater informativeness (relative to (15)). But, the fact that H and P are named suggests a way out, for it is not clear that their contrasting effects on cognitive content require them to be named, or hence that H and P have to figure in the contrasting contents’ subject matters. Why do I say this? H and P differ not merely as abstract objects, but in the features they predicate of the referent, by virtue of which it is the referent. H picks the referent out as the ϕ—the first celestial body visible in the evening sky, say—while P picks it out as the ψ—the last celestial body visible in the morning sky. Suppose that analogues of (14) and (15) could be found that incorporated the predicative material (ϕ, ψ) directly, rather than naming objects encoding that material. These analogues would not contain «H» and «P», and would not be about modes of presentation; they’d be about whatever celestial bodies were in fact ϕ and ψ. This in essence was Russell’s approach. (1) and (2)’s cognitive contents for him were The first evening-visible celestial body = the last morning-visible celestial body, (16) and The first evening-visible celestial body = the first evening-visible celestial body. (17) Russell appears to accomplish by this maneuver what Frege could not. The contents he gives us are (i) astronomical as to topic, and (ii) unequally informative about that topic. This might solve the problem, if informativeness and aboutness were the only desiderata in play. But they aren’t: a certain kind of fidelity is required to the target sentences (1) and (2). Otherwise one could use any old (1*) and (2*) that were about astronomy, informative in the case of

210   (1*), and trivial in the case of (2*)—no matter how much violence they did to the meanings of (1) and (2). Nice as it sounds in principle to say that X* and Y* cannot explain the cognitive difference between X and Y unless they are faithful to X and Y – that paraphrases should be in some appropriate sense “true” to the sentences whose cognitive contents they purport to capture – what this comes to in practice remains to be seen. A good place to start is with standard objections to Russell’s approach. These turn out to revolve in many cases around fidelity worries.

9.4 Fidelity as Synonymy Perhaps X* is faithful to X only if they mean the same—albeit X* might render the meaning more perpicuously, or in a way that better brings out its news value. The description account is in trouble if fidelity is synonymy, for reasons developed by Kripke in Naming and Necessity. (16) agrees in meaning with (1) only if «Hesperus» means the same as «the first evening-visible celestial body» and «Phosphorus» the same as «the last morning-visible celestial body». But there are reasons to doubt that «Hesperus» does mean this, and to doubt more generally that «n» means the same as some associated description «the χ». These reasons tend in the literature to be filed under three headings: modal, semantic, and epistemic. modal names are rigid; descriptions usually pick out different objects at different worlds semantic one can grasp «n»’s meaning despite being unaware that it refers to the χ³ epistemic it is a priori that the χ, if it exists, is a χ—but not that n, if it exists, is a χ.

³ This comes in two flavors: either «n» does refer to the χ, but competent speakers needn’t be aware of this (the “argument from ignorance”); or competent speakers take «n» to refer to the χ, but they’re wrong (the “argument from error”). The first possibility is illustrated by «Feynman»— people tend to know only that he’s a physicist—the second by «Peano»—it was supposedly Dedekind who discovered Peano’s Axioms.

:     

211

A number of other objections have been raised. E.g., «n» and «the χ» differ metasemantically; names require in most cases a causal link to the referent. They are of different grammatical categories. «The χ» is a determiner, a kind of quantifier phrase; a quantifier phrase can’t mean the same as a name. Some would respond to the first worry that we can switch to rigidified definition descriptions («the actually first evening-visible . . . »); to the second, that grasping a meaning doesn’t require one to recognize it however presented; to the third, that analyticity does not entail apriority (and/or that defeasibility does not preclude apriority); to the fourth, that causal relations can play a role in the semantics of descriptions too, since «the χ» denotes the one and only χ in a certain contextually indicated situation, singled out in part causally; to the fifth, that grammatical category potentially crosscuts meaning (e.g., names retain their meaning through type-shifting operations that force them sometimes into determiner mode). A different sort of response concedes the non-synonymy of X* with X. It asks why fidelity should require meaning agreement in the first place. All we know so far about cognitive content is that it’s the I-know-not-what that explains differences in felt information value. Whether it takes an out and out synonym to accomplish this is not obvious. Sentences’ semantic contents—what they say – are often distinguished from their assertive contents – what speakers are apt to say with them. X* might line up less with the first than the second. Views along these lines will be considered shortly. First let’s look at an option that Kripke himself considers as he wrestles (in [Kripke, 1980]) with the phenomenon of contingent truths that appear as a cognitive matter to hold unconditionally.

9.5 Reference-Fixers If ordinary proper names are disguised descriptions, then they share a meaning, presumably, with the descriptions they stand in for. Kripke rejects Russellianism (as a semantic theory) for this reason. But he allows

212   that descriptions can function metasemantically as specifiers of the item a name «n» refers to—the item that Kripkeans generally take to exhaust «n»’s meaning. Now, the referent can hardly fail to be picked out by some description or other. The suggestion to be interesting has got to be that «the F» carries some kind of authority. This is NOT merely to require that «n is F» be a priori. Those with enough mathematical training know a priori that «π = 4
(1  1/3 + 1/5  1/7 + 1/9 . . . )». Still, the series is not what fixes the reference: π is supposed to be the ratio of the circumference of a circle to its diameter. It seems to me that here this Greek letter is not being used as short for the phrase ‘the ratio of the circumference of a circle to its diameter,’ nor is it even used as short for a cluster of alternative definitions of π, whatever that might mean. It is used as a name for a real number. ([Kripke, 1980], 60)

A reference-fixing description spells out what qualifies x as the referent; «n» stands for x if and because x is F and nothing else is F. A priority is conferred on «n is F», or «n is F, if anything is (uniquely) F», by the fact that a non-F has no hope⁴ of counting as the item in question. How is this not just Russell again? The point about reference-fixing descriptions is that while they guide us to the meaning (= the referent), they do not make it into the meaning.⁵ If the cognitive content of «A meter is the length of this stick» is given, despite the difference in meaning, by «This stick’s length is the length of this stick», perhaps their non-synonymy is not an obstacle either to «The first evening-visible celestial body = the last morning-visible celestial body» giving the cognitive content of «Hesperus = Phosphorus». This explains in principle how coreferential substitutions can affect cognitive content—news value—while leaving meaning unchanged. ⁴ For certain speakers, at least, at certain times. ⁵ As Moses supposedly led the Israelites to the promised land, without making it in himself.

:     

213

Say «n» has its reference fixed by «the F», while «m», which corefers with it, has its reference fixed by some other description, or by no description at all. Then n is F, if anything is (uniquely) F

(18)

will be uninformative, while m is F, if anything is (uniquely) F,

(19)

packs an informational punch. The contrast is non-semantical since the two sentences mean the same. You may or may not find this convincing. (Is it the truth (18) expresses that’s trivial, or the fact that it expresses some truth or other?) But never mind that for now. Our problem is much more basic. Cognitive significance issues arise for all kinds of name. But for a name to have its reference fixed (in the sense under discussion) by a description is quite uncommon.⁶ Kripke does not even in the meter stick case want to put the referent entirely at the mercy of the associated description.⁷ The stipulator is said to have in advance “a certain length he wants to mark out.” The standard meter stick S is chosen with that end in mind. But the choice might be so ill-advised as to call the exercise into doubt: What happens if he is wrong, and it is a different length than intended? It might be, for instance, that the stick is a millionth of an inch long, but emitting magnification rays that delude us into seeing it as longer. Or maybe the stick is a mile long, but much farther away than anyone had realized. I take it that it is no part of the reference-fixer’s understanding of ‘meter’ that it continues to stand for the length of S even if S is much shorter or longer than it appears. ([Yablo, 2008]: 184)

⁶ “I also think, contrary to most recent theorists, that the reference of names is rarely or almost never fixed by means of description” ([Kripke, 2011]: 21). ⁷ Confirmed in conversation (June 2022).

214   This is part of the reason that Kripke quickly moves on to the model of initial baptism followed by chains of reference-preserving intentions. The new model is sometimes thought to grant its own sort of authority to descriptive material. The baptismal act stands no chance of succeeding, on this view, unless a sortal comes in to disambiguate. One knows a priori (to begin with, anyway) that the sortal attaches if the referent exists.⁸ Kripke considers this and rejects it: Even if a sortal is used to disambiguate an ostensive reference, surely it need not be held a priori to be true of the object designated. Couldn’t Dobbin tum out to belong to a species other than horses (though superficially he looked like a horse), Hesperus to be a planet rather than a star, or Lot’s guests, even if he names them, to be angels rather than men? ([Kripke, 1980]: 116, note 58)

Reference-fixers as we find them in nature are not—neither in the purely descriptive case («Neptune») or when a demonstrative is involved («meter», «Dobbin»)—a tool for tying ourselves to the semantic mast.⁹ They rest almost always on empirical presuppositions whose failure could and should prompt a rethinking of the stated descriptive conditions χ. In practice χ is recommended by x, the referent in waiting, as much as x is recommended by χ.

9.6 Diagonal Propositions When reference-fixing descriptions function as advertised, they associate with «n is G» a proposition that, although not literally expressed by that sentence, has some claim to be regarded as «n is G»’s cognitive content. I am speaking of

⁸ [Thomasson, 2014]. ⁹ If Dobbin could turn out not to be a horse, it is hard to see why “mythical” animals could not surprise us both on the score of existence and biological type. Unicorns, if there are any, could turn out to be a previously unknown sort of horse. This is one of the likelier scenarios, in fact, on which they turn out to exist ([Yablo, 2020]).

:     

215

The proposition that is true at w iff «n is G» as uttered in w expresses a singular truth there. (20) What the reference-fixing description, if we have one, gives us is a tidy way of identifying the singular truth—the one that is expressed in w by a (true) utterance of «n is G». It’s The proposition attributing G-ness to whichever object ow uniquely satisfies in w the description fixing «n»’s reference—whichever object is uniquely F there. (21) The handy mode of identification set out in (21) is rarely available if reference-fixing descriptions rarely exist. But maybe we can live with that. It is (20) after all that gives the cognitive content. And it is still available for that purpose even in the absence of a simple articulation of what the other proposition is that has to be true in w for the cognitive content to hold there. This in a nutshell is Stalnaker’s diagonalization strategy. S’s customary or horizontal content is the proposition that holds at w iff S as uttered here in @ portrays w accurately. The proposition that holds at w iff S as uttered in w offers a correct description of w is S’s diagonal content. A name «n» does not need its reference fixed by description to make a well-defined contribution to the diagonal contents of «n»-sentences. It’s enough that «n» picks out different objects on different hypotheses w about the world of utterance. The reasons the reference varies are metasemantic, as in Kripke. But speakers are not expected to be in possession of a formula picking out the referent in each w, or even each w that might for all they know really obtain. A decent analogy is with demonstratives like «that». A demonstrative picks out different objects on different hypotheses about speaker intentions, where their hand is pointing, the surrounding scene, etc. An explicit rule is too much to hope for, but we have a good idea in practice.¹⁰ The issue of which object «n» picks out on different

¹⁰ [Stojnić et al., 2017, Stojnić, 2021].

216   hypotheses about our intentions, the referential traditions we’re party to, etc., is similar. An explicit rule is not to be expected, but neither is it particularly needed.

9.7 Pragmatic Repair An asserted S is presumed to express the associated horizontal proposition, often called what is said after Kaplan. Explicitly it’s the set of counterfactual worlds w that are correctly described by S as uttered in @. S is assigned the diagonal proposition only as part of a pragmatic repair strategy. We start as a rule with the horizontal proposition; it’s the one delivered by the semantics. Sometimes though trouble arises— principles crucial to communication are violated—if the horizontal interpretation is allowed to stand. The repair strategy then kicks in to substitute a different proposition that respects the principles. To appreciate the kind of trouble involved, we need to look briefly at Stalnaker’s account of assertion as a move in the cooperative activity of conversation. The point of this activity is to whittle down bit by bit the set of worlds still in play—the so-called context set Γ. “Still in play” here is shorthand for worlds that might, for all we know at a particular point in the conversation, be actual. A proposition is presupposed if it holds in all Γ–worlds. To make an assertion is to reduce the context set in a particular way . . . . all of the possible situations incompatible with what is said are eliminated. To put it a slightly different way, the essential effect of an assertion is to change the presuppositions of the participants . . . by adding the content of what is asserted to what is presupposed. ([Stalnaker, 1999], 88)

The horizontal proposition is contraindicated if adding it to what’s presupposed is not, for reasons we’re coming to, a sensible move given the goals just outlined. We reach for the diagonal content in hopes that adding it is more in keeping with the project that we are jointly engaged in.

:     

217

How does the horizontal content reveal itself as unsuitable? Stalnaker has a fascinating story to tell. The following “can be defended,” he says, “as essential conditions of rational communication, as principles to which any rational agent would conform if he were engaged in a practice that fits the kind of very abstract and schematic sketch of communication that I have given” ([Stalnaker, 1999], 86). CONTINGENCY A proposition asserted is always true in some, but not all, of the possible worlds in the context set. TOTALITY Any assertive utterance should express a proposition, relative to each . . . world in the context set, and that proposition should have a truth value in each . . . world in the context set. UNIVOCALITY The same proposition is expressed relative to each possible world in the context set. ([Stalnaker, 1999], 88, labels mine) When any of these is in danger of being violated, the search begins for an alternative content. There are at least two reasons, given this background, why an assertive utterance of «Hesperus = Phosphorus» cannot be taken at face value—two reasons why it cannot be the horizontal content that’s asserted. First, the horizontal content is a necessary truth, hence necessary within Γ. To assert a proposition true in all Γ-worlds would be selfdefeating; intersecting Γ with an always-true proposition does not cut the context set down any, which was the point of the exercise.¹¹ Well, but what if we could arrange for «Hesperus = Phosphorus» to express either the necessary truth that Venus = Venus, or the necessary falsehood that Venus = Mars, depending on whether it is Venus or Mars that’s visible in the morning? Then our assertion could perhaps cut down the context set. Worlds w where the necessary falsehood is expressed would be thrown out as misdescribed by what «Hesperus = Phosphorus» says in w. Worlds where the necessary truth is expressed would be retained.

¹¹ “[T]o assert something which is already presupposed is to attempt to do something that is already done.” “If the speaker says something that seems prima facie to be trivial, . . . one may look for another interpretation of what he said” (ibid., 89).

218   A different problem now arises, to do with the third principle. If «Hesperus = Phosphorus» might express either that Venus = Venus or that Venus = Mars, then the audience can’t tell what the content is that the speaker is proposing to intersect with the context set. It is hard to see how in that case they’re to judge the advisability of such a move. The diagonal proposition appears to help with both problems. CONTINGENCY is respected since «Hesperus = Phosphorus» has a true horizontal content in some w ∈ Γ and a false one in others. As for UNIVOCALITY, the diagonal content is uniquely singled out, since we know more or less how the referents vary as we move through the context set. Consider: Hesperus = Phosphorus, it is now three o’clock, an ophthalmologist is an eye doctor. In each case, to construct a context which conforms to the first principle, a context in which the proposition expressed is neither trivial nor assumed false, one must include possible worlds in which the sentence, interpreted in the standard way, expresses different propositions. But in any plausible context in which one of these sentences might reasonably be used, it is clear that the diagonal proposition is the one that the speaker means to communicate (ibid., 92) Let’s grant for now that the diagonal proposition gives the intuitive cognitive content in cases where CONTINGENCY and UNIVOCALITY force us to look for an alternative to the horizontal proposition. Does this solve the cognitive content problem? That depends on whether it is only in such cases that the cognitive content comes apart from the horizontal content.

9.8 Trigger Problems Stalnaker’s strategy generates distinct cognitive contents just when diagonalization is triggered—that is, just when asserted sentences need to be reinterpreted on pain of violating principles like CONTINGENCY and

:     

219

UNIVOCALITY.¹² It is natural to wonder whether this gets us all the cognitive distinctions we need. There could in principle be cases where 1. two sentences differ in cognitive content, though 2. they share a horizontal content, and 3. diagonalization isn’t triggered since the principles are not at risk of being disrespected. Stanley raises this kind of worry in ([Stanley, 2010]). Imagine it is common ground in a certain community that Hesperus and Phosphorus are distinct objects h and p. Speakers see a clear cognitive difference between Hesperus is never far from Phosphorus,

(22)

which is viewed with suspicion, and Hesperus is never far from Hesperus.

(23)

which is trivially correct. (22) and (23) agree in semantic/horizontal content, so (22) at least will have to be reinterpreted. But it is not clear reinterpretation is licensed: Diagonalization does not enter the picture, because there is no room for uncertainty about meaning. Everyone is certain that «Hesperus» has a certain denotation and «Phosphorus» has a certain denotation, and that they are distinct (ibid., 110).¹³

Stalnaker responds that it’s the theorist, not the community, who decides when diagonalization is called for.¹⁴ From the theorist’s perspective, there is a kind of equivocation occurring: between the sentences’ “real”

¹² TOTALITY can be ignored for now; it comes up later in connection with existence-claims. ¹³ [Stalnaker, 2010] holds that a “derived” context set should be used relative to which (22) and (23) do differ in horizontal content. This raises large issues that cannot be discussed here. Another possible response is that  is threatened; but this calls at most for a change of example. ¹⁴ Stanley remarks on this as well in raising a different problem.

220   horizontal contents as delivered by the semantics, and the “fake” horizontals wrongly imputed by the community. But how does this violate the principles as stated? If there is no “uncertainty about what is meant,” rooted in uncertainty about which world is actual, then UNIVOCALITY is not genuinely under threat; rather a threat is drummed up to gain us access to diagonalization. Williamson beating some neighboring bushes has Stalnaker arguing as follows: If the sentence «Hesperus = Phosphorus» is true, it expresses a necessarily true proposition. If «Hesperus = Phosphorus» is false, it expresses a necessarily false proposition. But any necessarily true proposition is distinct from any necessarily false proposition. Thus the proposition «Hesperus = Phosphorus» expresses if it is true is distinct from the proposition it expresses if it is false. So what proposition «Hesperus = Phosphorus» expresses depends on whether it is true. Therefore not knowing whether «Hesperus = Phosphorus» is true involves not knowing what proposition «Hesperus = Phosphorus» expresses. ([Williamson, 2011], 519)

This line of thought proves too much, he thinks. One could argue with (almost) equal justice that uncertainty about whether «Venus is hot» is true involves, since true propositions are distinct from false ones, not knowing what proposition it expresses. Stalnaker might say that this misrepresents his position. If someone wondering whether Hesperus = Phosphorus can’t pick out the sentence’s horizontal content, the reason is not that you have to know a content’s modal status, or truth-value, to know its identity. It’s that to know a content’s identity, you should not be in the dark about what it says. And you are in the dark, if it might for all you know be a proposition about Venus, or Mars, or the two together. The real problem it seems to me is as follows. Haziness about horizontal contents plays a crucial role in the mechanism pushing us onto diagonal contents. But when we look at Frege’s original data point, the haziness isn’t there. He was not in any doubt about Hesperus’s identity with Phosphorus when he wrote, of «Hesperus = Hesperus» and «Hesperus = Phosphorus», that they are

:     

221

obviously statements of differing cognitive value; [one] holds a priori and, according to Kant, is to be labeled analytic, while statements of the [second] form . . . often contain very valuable extensions of our knowledge and cannot always be established a priori ([Frege, 1948])

If knowing the horizontal contents did not blind Frege (or his audience) to the difference in cognitive value, then it’s hard to see how it could have been a difference in diagonal content he was picking up on. Diagonalization wasn’t triggered.¹⁵ Now «Hesperus = Phosphorus» clearly was in one sense uninformative for Frege. It did not when he wrote “On sense and reference” tell him anything he didn’t already know. This just shows, however, that the notion of informativeness at work in that paper had not much to do with actual surprise value. «Hesperus = Phosphorus» packs an informational punch that does not greatly depend on whether the information was already in one’s possession.¹⁶

9.9 Pragmatic Enrichment Diagonalization is one sort of pragmatic process that might be called on to generate distinctive cognitive contents. But it is not the only one. If diagonalization kicks in in only some of the cases where cognitive contents are needed, we ought to be on the lookout for alternative mechanisms that kick in more easily. Here we look at two enrichment mechanisms, one familiar and the other not, that seem to fit the bill. The first is implicature. «S₁» might convey more than «S₂» because it implicates things that «S₂» doesn’t—because, to put it another way, «S₁»’s “enhanced” content, obtained by folding in implicatures, is distinct ¹⁵ You might say it was triggered by CONTINGENCY. But we can change the example. «Hesperus tends to be visible in the evening» is of lower cognitive value than «Phosphorus tends to be visible in the evening». But neither has as its horizontal content a proposition true throughout the context set. ¹⁶ Perhaps «Hesperus = Phosphorus» only seems to pack an informational punch, because one remembers what it was like to learn it for the first time? (Compare Bayesians on the problem of old evidence.) Or perhaps a kind of reverse accommodation occurs in which a closed question is reopened. These are interesting possibilities, not explored further here.

222   from the enhanced content of «S₂». Why would this get us additional cognitive contents? Well, an implicature is generated when imputing a wish to convey that P helps to bring an utterance into line with Grice’s conversational maxims: Quantity Make your contribution as informative as, and no more informative than, required for current purposes. Quality Try to make your contribution one that is true; do not say that for which you lack adequate evidence. Relevance Be relevant. Omit any information irrelevant to the current exchange. Manner Be perspicuous. Avoid obscurity, ambiguity, prolixity. Be orderly. The classic example is damning with faint praise. When someone writes in a reference letter that Smith has good handwriting and is highly punctual, this conveys that Smith is not such a great philosopher. One would not be dwelling on handwriting unless flattery more to the point would be false. Assuming that the same proposition (that Smith has good handwriting) is expressed throughout the context set, and that it draws a nontrivial line through that set, diagonalization gets no grip. A problem may have occurred to you. Grice took implicatures to be on the whole nondetachable—if something is implicated by my assertive utterance of «S», then it will not generally be “possible to find another way of saying the same thing, which . . . lacks the implicature in question” ([Grice, 1975], 39). Nondetachability tells us that «Smith has excellent penmanship» ought to share the bad-philosopher implicature, a prediction that’s borne out. Remember though that «S₁» and «S₂» were to differ in their implicatures despite saying the same; that’s how the distinctive cognitive contents are meant to be generated. The outcome we’re after here is what nondetachability says we can’t have. This would be bad if nondetachability were an ironclad rule. But Grice allows as well for implicatures turning on how a thing is said; these are the manner implicatures. «P or Q» tends for instance to implicate «Not both», hence our temptation to read «or» as exclusive.

:     

223

«P or Q or both» obviously lacks this implicature. The contrast is not because of what «(P ∨ Q) ∨ (P ∧ Q)» strictly says—it’s logically equivalent to «P ∨ Q»— but the way it is said, with «Both» explicitly listed as a possibility. Implicature has long been a mainstay of Millian apologetics about cognitive significance. The implicature gambit is especially prominent in the work of Salmon and Soames.¹⁷ The Stanford Encyclopedia introduces it as follows: the theory that the meaning of a name is its referent entails that coreferential names are synonymous . . . [this] seems untenable in light of names like «Superman» and «Clark Kent». Millians have proposed that the source of these linguistic intuitions is a difference in implicature ([Salmon, 1986], [Soames, 1989], [Berg, 2012]). One hypothesis invokes metalinguistic implicature, noting that «Clark Kent can fly» implicates that a man called «Clark Kent» can fly ([McKay, 1981], [Berg, 1988], [Berg, 1998]) ([Davis, 2019]).

The specifically metalinguistic version sits ill with the cognitive equivalence, already noted, of «Hesperus = Phosphorus» with «Espero = Fosforo». Another worry is that coreferential substitutions seem sometimes to turn truths into falsehoods, whereas statements with false implicatures tend to strike us as misleading rather than false.¹⁸ A final, seemingly fatal, problem is that «S»’s cognitive content should often be weaker than, or independent of, its semantic content, rather than an extension of the semantic content. Otherwise «Hesperus isn’t Phosphorus» would be cognitively absurd, by virtue of conveying in part that Venus isn’t Venus.¹⁹ ¹⁷ [Salmon, 1986], [Soames, 2002]. Soames lays less weight on implicature in later papers ([Soames, 2005], [Soames, 2008]), partly for reasons we’re getting to. ¹⁸ The falsity of so-called “implicitures” may be another matter ([Bach, 1994]). See also the literature on intrusive implicature. ¹⁹ Grice doesn’t want the speaker’s message to include the semantic content either, e.g., in cases of irony or hyperbole. He avoids it by postulating that the speaker only makes as if to say S. But, someone wrong about the identity facts is really saying a ≠ b, not only making as if to say it. It is for reasons such as this that Soames abandons his one-time view that “the semantic content of a sentence is always a complete proposition that is asserted and conveyed by normal utterances in every normal context” ([Soames, 2005], 256).

224   Diagonalization and implicature are the mechanisms that have attracted most of the attention. But they are not the only imaginable mechanisms. Here is one that’s never been discussed to my knowledge, though it has an obvious intuitive appeal.²⁰ «A»’s information value lines up on the face of it with the effects of «A»-testimony on the hearer’s belief state. Some of those effects will be “direct,” that is, we come to believe the statement’s semantic content. Others will be mediated by collateral beliefs, say of the form «A!C». It stands to reason that «A»’s information value would be a function in part of the conclusions falling out of it by arguments of the form A A!C ∴C

(24)

This is not going to get us distinctive cognitive contents, unless different conclusions flow in this way from semantically indistinguishable premises «A₁» and «A₂». But if «A!C» is a standard issue indicative conditional, different conclusions will flow, for semantic equivalents are known not to be freely substitutable in indicative antecedents. Stalnaker made the point in his original paper on the topic ([Stalnaker, 1975]). His example involved empty names, but it holds for coreferential names too. Take these in order. «Vulcan exists ! Babinet was onto something» seems hard to argue with; it was Babinet who conjectured that there was such a planet as Vulcan. Certainly it’s more plausible than «Sasquatch exists ! Babinet was onto something». Yet the antecedents, to the extent they have a semantic content at all, have the same one given that the names are both empty.²¹ Babinet being onto something figures in the inferential upshot of only one of the sentences, their semantic equivalence notwithstanding. Now a coreference case. Paderewski₁, the Polish pianist, is one and the same individual as Paderewski₂, the statesman, but Al doesn’t ²⁰ True when I wrote it, but no longer. See Berto and Hornischer (2023). ²¹ [Yablo, 2020], [Yablo, 2021], [Yalcin, 2021].

:     

225

know this. Al finds «Paderewski₁ died in 1941! Polish pianists have lived to over 80» plausible, Paderewski₁ being for him a pianist. «Paderewski₂ died in 1941 ! Polish pianists have lived to over 80» does not share in this plausibility, since Paderewski₂ is not to his knowledge a pianist. The Paderewski-sentences differ in their inferential upshot for Al, though semantically speaking they say the same thing. (They differ in their immediate inferential upshot even if Al knows of the identity.) None of this is surprising on the view that indicatives are evaluated suppositionally—by imagining we have just been informed of the antecedent, and asking ourselves what we are now to think of the consequent. Thus Ramsey: If two people are arguing If P, will Q? and are both in doubt as to P, they are adding P hypothetically to their stock of knowledge and arguing on that basis about Q.²²

The question is whether adding «a is thus and such» to one’s knowledge base always prompts the same cognitive adjustments as adding «b is thus and such», given only that «a» and «b» corefer. Frege cases resolve this pretty decisively. We find different «Q»s plausible depending on the name employed in «P». Should the oracle say, «Hesperus ≠ Phosphorus», we would likely conclude that one planet appears in the evening, a different planet in the morning. We wouldn’t know what to think if it said, «Hesperus ≠ Hesperus», but certainly not that different planets appear early and late. «Hesperus ≠ Phosphorus ! different planets appear early and late» strikes us accordingly as much more plausible than «Hesperus ≠ Hesperus ! different planets appear early and late». «Different planets appear early and late» belongs to the indicative upshot of «Hesperus ≠ Phosphorus», but not to that (if we can even make sense of it) of «Hesperus ≠ Hesperus». This on the view we’re considering is why the

²² Ramsey, “General Propositions and Causality,” in [Ramsey and Mellor (ed.), 1990]. (Quine in the same spirit calls indicative if/then “a dramatic idiom,” assessed by “feign[ing] belief in the antecedent and see[ing] how convincing we then find the consequent” ([Quine, 1960], 222).)

226   two differ in cognitive content. How it all plays out at a semantic level is not obvious, but there are proposals out there.²³ Stalnaker comes across as a fellow traveler, when he says that indicative suppositions are likelier to be diagonalized than counterfactual: To interpret the statement If Aristotle hadn’t existed, the history of philosophy would have been very different from the way it was, we do not need to diagonalize, since in any possible context appropriate to THAT statement, it will be presupposed that Aristotle does exist. . . . It is interesting to note that if the conditional were in the indicative mood, the result would have been different. This is because an indicative conditional is appropriate only in a context where it is an open question whether the antecedent is true. So to say If Aristotle didn’t exist is to suppose just what is asserted when one asserts Aristotle didn’t exist ([Stalnaker, 1999], 94)

It’s to suppose in other words the diagonal proposition. If indeed diagonalization more easily triggered when «A» is supposed than asserted, this could give the indicative-inferential upshot strategy an advantage over the norms-of-assertion strategy. Either way, we are still saddled it seems with a problem raised above for implicature. «A»’s cognitive content should sometimes be weaker than, or independent of, its semantic content. Adding «A»’s indicative upshot to its semantic content yields something stronger. There may be ways of getting around this, but it’s a complication we would rather avoid. I want to move on to a mechanism that generates weakened/ independent contents “naturally,” without special tinkering.

9.10 Presupposition Subtraction Something is presupposed if it’s to be treated as true for conversational purposes, where this is understood all around. Presuppositions P are not

²³ [Weatherson, 2001], [Santorio, 2012], [Glick, 2012], [Santorio, 2018], See also [Hintikka, 1970], [Lewis, 1983], [Stalnaker, 1986].

:     

227

known a priori; they may turn out to be false, and may even be (on the side, as it were) known to be false. Still they are a priori-like in the sense of being “already settled.” They lay down parameters for discussion rather than being up for discussion themselves. Recall that reference-fixing descriptions were once thought to generate cognitive contents via associated a priori conditionals: «n is thus and so» conveys that the F (if anything is uniquely F) is thus and so, where «n» has its reference fixed by «the F». The idea has fallen on hard times, we saw in Section 9.5, not least because ab initio referential stipulations are rare. A lot of the objections fall away, however, if reference-fixers are conceived instead as presuppositions about the kind of thing a word refers to. Garcia-Carpintero takes this line in [Carpintero, 2000]. Kripke explores it too in unpublished lectures,²⁴ though he speaks rather of preconceptions or prejudices.²⁵ How would prejudices/presuppositions “about Hesperus” make their way into the cognitive content of «Hesperus»-sentences, as opposed to «Phosphorus»-sentences? Since we are trying to explain to the Millian, and to philosophers at large in a Millian-ly acceptable way, how « . . . Hesperus . . . » and « . . . Phosphorus . . . » can differ in cognitive content (to the point sometimes of disagreeing in intuitive truth value) despite saying the same thing at a semantic level, certain answers are unavailable. We can’t easily call «Evening-appearing-ness is a preconception about Hesperus» true, and «Evening-appearing-ness is a preconception about Phosphorus» false, because this is precisely the sort of puzzling pair that Millians officially disallow, and that we are trying on their behalf to make room for.²⁶ Pending a theory of cognitive content, preconceptions (for the Millian) about Hesperus = preconceptions about Phosphorus = preconceptions about Venus. There is I think a way around this problem. But it requires a subtler notion of presupposition than whatever is common ground at a certain point in the conversation. A distinction will be needed between active ²⁴ The lectures were on secondary properties. [Gomez-Torrente, 2011] is an excellent discussion. (I heard them at Michigan in 1989.) ²⁵ See [Swanson, 2006], [Swanson, 2012], for a metalinguistically flavored presuppositionalism. ²⁶ Compare Kripke’s admission in [Kripke, 2013] that he is giving “proposition about Holmes” a special quasi-intensional use that he has no clear right to qua Millian.

228   presuppositions—those we’re currently leaning on, and expecting our audience to lean on as well—and the rest. Regular presuppositions hold their ground on this view. But not all of them are drawn on by the machinery that wrings «S»’s cognitive content out of its semantic content. Let’s say, to have a word for this, that speakers pivot, when uttering «S», on certain presuppositions but not others. An example of Donnellan’s gives the flavor, though we’ll be interpreting it in our own way. Why do we hear «The guy drinking a martini is a philosopher» as carrying the information that that guy is a philosopher? The story we’re working up to points to the fact that we’re pivoting on the (shared) assumption that that guy is a martini-drinker (the only one around). An example Donnellan does not give, but could have, is the following. This time we are wondering, not about people’s professions, but which sorts of cocktail go with which areas of academic endeavor. Someone says, pointing at the martini-drinker, «Well, that guy is a philosopher». With the right sort of stage-setting, we can hear this as saying that the martini-drinker is a philosopher. We are pivoting, again, on the assumption that that guy is the martini-drinker—except now with a view to working martinis into the content of «He’s a philosopher». (Earlier we were pulling martinis out of the content of «The martini guy’s a philosopher».) No doubt one would like to know more about how pivot-presuppositions are singled out on particular occasions. Certainly though how one says a thing—the sentence «S₁» one hits on to express the semantic content |S₁| (= |S₂|)—has a crucial role to play. «S₁» acquires a cognitive content by raising certain presuppositions to prominence, which then reshape its semantic content according to principles we are still getting to. This is part of the story of how «Hesperus is thus and so» and «Phosphorus is thus and so» associate themselves with different cognitive contents. They do it by feeding different presuppositions into the content-generation machine. «Hesperus is thus and so» feeds in what I will unhelpfully call presumptions “about Hesperus” while «Phosphorus is thus and so» feeds in presumptions “about Phosphorus.” What could it mean for a presumption P to be “about Hesperus” or “about Phosphorus”? More is required than for P to attribute a property to Hesperus (Phosphorus). Property-attributions don’t divide up along the contemplated lines, as the properties P attributes to Hesperus are

:     

229

attributed also to Phosphorus, and vice versa. How are presumptions pertaining specifically to Hesperus marked off from those pertaining specifically to Phosphorus? One clue is the contexts in which P is activated, that is, called up for duty from the reserve army of possible pivots. P is to a first approximation “about Hesperus” if it is activated by the use of «Hesperus» as opposed to «Phosphorus». Timidity is a presumption “about Clark Kent,” because he presents himself as timid when the name «Clark» is used. X-ray vision is a presumption “about Superman,” because it leaps (noninferentially) to mind when that’s how he’s picked out. That Superman can be timid is not news exactly, but it takes some remembering when he’s going under that name. That Clark can fly is not news either; we already knew it, or were in a position to figure it out. What makes it a presumption not specifically “about Clark” is that we have to reason our way to it from the corresponding presumption “about Superman.” More needs to be said, obviously, about how we tell that it is P (rather that Q) that is activated by a particular choice of words. I don’t have a proper test to offer; we will have to get by with the following. P is activated to the extent that a certain whispered, off-stage prompt is in order when someone is processing S: «P, don’t forget». The point of the prompt is to ward off misunderstanding, as, e.g., when Hesperus is “confused” with the Morning Planet. A related test is run from the processor’s perspective. Say I’ve received S as testimony from a reliable informant. Unsure how to factor it into my thinking, I might say, «P, right?». My informant should welcome this query and answer in the affirmative. By contrast «Q, right?» would be cause for alarm («Hesperus is the Morning Planet, correct?»)—all the more so, if Q is inconsistent with P («Hesperus is seen in the morning, yes?»). Supposing that P is indeed activated, how does it make its mark on S’s cognitive content ⟦S⟧? Cognitive contents are found by asking what S adds to P.²⁷ ⟦S⟧ in other words is the Δ that “completes the enthymeme”

²⁷ [Yablo, 2012a, Yablo, 2014].

230   P Δ ∴S

(25)

S’s cognitive content becomes then a kind of incremental content S  P; it is what remains of S when P is subtracted.²⁸ Definitions are given elsewhere; the idea is that S  P is the proposition Δ that is (i) true in w iff P ⊃ S has a gap-bridging truthmaker in w (and P ⊃ ¬S doesn’t) (ii) false in w iff P ⊃ ¬S has a gap-bridging truthmaker in w (and P ⊃ S doesn’t) (26) — where A gap-bridging truthmaker for B ⊃ A is a fact τ compatible with B that combines with B, using as much of it as possible, to imply A. (27) This begins to explain how it can figure in «Hesperus is thus and so»’s cognitive content that a certain evening-visible planet is thus and so, whereas it’s the thus-and-so-ness of a morning-visible planet that’s conveyed when «Phosphorus» is plugged in. Hesperus is the first evening-visible planet Δ ∴ Hesperus is thus and so.

(28)

²⁸ The label “that which remains” works best if P is implied by S, which is not always the case. The king of France is in that chair implies France has a king; what remains is, for one thing, that the chair is non-empty, which is itself implied. But suppose S and P are The guy with the martini is a philosopher and That guy is drinking a martini. S implies neither P nor S  P = That guy is a philosopher. That that guy is a philosopher is not (as) happily described as “what remains of The guy with the martini is a philosopher” given that it is logically independent of The guy with the martini is a philosopher. (Note that we can obtain something in the vicinity of Stalnaker’s diagonal proposition by letting P be the T-biconditional ‘S’ is true iff S. The present approach can be seen as extending the diagonalization strategy to presuppositions not of that precise form.)

:     

231

What goes in for Δ is clear: the first evening-visible planet is thus and so. You might wonder why evening-visibility doesn’t also worm its way into the cognitive content of «Phosphorus is thus and so», given that this enthymeme Phosphorus is the first evening-visible planet Δ ∴ Phosphorus is thus and so

(29)

is “solved” by the same Δ. The answer is that (29) is the wrong sort of enthymeme. The right sort, for any given S, will be of the form P, Δ ∴ S, where P is activated in context by S. The evening-visibility prejudice pertains to Hesperus, not Phosphorus. It is activated by «Hesperus»-talk; S in (29) (Phosphorus is thus and so) is a piece of «Phosphorus»-talk. Here then is a first-pass account of how «Hesperus is thus and so» and «Phosphorus is thus and so» can differ in cognitive content despite saying the same. Part of the story is that swapping one name out for another in S changes the sort of Δ needed to make P, Δ ∴ S a valid argument (first-order valid, for definiteness). Δ can be anything if S is «Hesperus is identical to Hesperus». But suppose it is «Hesperus is identical to Phosphorus»; then something more like «The first eveningvisible planet is identical to Phosphorus» is called for. Another thing coreferential substitutions can change in P, Δ ∴ S, beyond the argument’s first-order validity, is its relevance in the first place to S’s cognitive content. P, Δ ∴ S bears on ⟦S⟧ only if S makes P a suitable thing to pivot on. «Hesperus is thus and so» brings evening-visibility assumptions to the fore, while morning-visibility assumptions are activated by «Phosphorus is thus and so».

9.11 Subject Matter The definition of S  P has been sketched, but not fully spelled out. One reason for not going into detail is that you can find that elsewhere.²⁹ A more important reason is that there is another way of conceiving ²⁹ [Yablo, 2012a, Yablo, 2014]. See also [Yablo, 2012b], [Yablo, 2016], [Yablo, 2017]. For other views of subtraction see [Fuhrmann, 1996], [Humberstone, 2011]: 677–707, [Fine, 2017], [Fine, 2020].

232   incremental contents, developed by Daniel Hoek, that makes those details less relevant ([Hoek, 2018]). The new approach has moreover a number of advantages over the old: explicitness, compositionality,³⁰ and less reliance on newfangled machinery, to mention three. Hoek’s idea is to shift some of the content-determination work onto a new parameter: subject matter m. The parameter is called for anyway if S can convey different things as we vary the m under discussion.³¹ An example from Soames to get us started; subject matter will not be involved at first, but it soon works its way in: I am in an auditorium, attending a lecture. Two university officials enter the room, interrupt the lecturer, and announce, “There is an emergency. We are looking for Professor Scott Soames. Is Professor Soames here?” I stand up, saying, as I do, “I am Scott Soames.” My intention in saying this is to indicate that I am the person they are looking for. Although this is not the semantic content of the sentence I uttered, they immediately grasp this, and the three of us leave the auditorium. Later, another member of the audience reports what happened to a third party. He says . . . “Professor Soames said [told them] he was the person they were looking for, and the three of them left.” ([Soames, 2002], 74–75)

This can be explained by the subtraction method without subject matter getting involved. The pivot-presupposition P is They are looking for Scott Soames. I am Scott Soames’s incremental content relative to this P is the Δ that completes the enthymeme You are looking for Scott Soames Δ ∴ I (the man calling attention to himself) am Scott Soames

(30)

The thing to put in for Δ here is what Soames would express by saying I (the man calling attention to himself) am the one you’re looking for. ³⁰ A ∧ B’s new-style content will for instance be the conjunction of A’s with B’s. ³¹ While holding P fixed.

:     

233

Imagine now a variant of the case where Soames on account of his rare blood type is used to being sought out as a donor. The officials are wearing Red Cross hats so it’s clear what they want him for. It is not too much of a stretch, against this background, to hear him as saying, I (the man calling attention to himself ) am the one you’re looking for to donate blood. If on the other hand the visitors are in fire gear, while an app on his phone is telling him that smoke alarms are going off at home, we might hear him as saying, I (the man calling attention to himself ) am the one you’re looking for re the fire. A tempting account of these variant cases is that Soames is speaking in them to different issues: where is the individual with the special blood? in the first, where is the individual with the burning house? in the second.³² I called the m parameter “new,” but subject matter has made several appearances in this paper already. We started with Frege’s complaint about his Begriffschrift theory that it turns «Hesperus = Phosphorus» into a claim about words rather than planets. (Begriffschrift fans like to point out that identity-claims are sometimes about words, as when I ask you what «barrister» means, and you tell me, wrongly, that barristers are just lawyers.) S’s implicatures are sensitive in some cases to what we were talking about when S was produced. I was taking a lot of Oxy back then normally implicates that I’m not taking much now.³³ But the implicature goes away when the issue is a class-action suit against Purdue Pharma based on marketing practices at the time in question. Martinis were worked into the cognitive content of «That guy’s a philosopher» by letting the question under discussion be how cocktail preferences line up with field of expertise. Even the proposed account (26) of subtraction, notionally subject-matter-free, gets into this neighborhood if a sentence’s aboutness-features bubble up from its truthmakers. (A truthmaker τ for B ⊃ A uses “as much of B as possible” iff τ minimizes the extent to which C ⊃ A is also verified, C ranging over parts of B. C is part of B only if its subject matter is included in B’s

³² Hoek gives the example of a myth-invoking sentence conveying that the sun has set in Tripoli, if we’re talking about Tripoli, or that it has set in Alexandria, if it’s Alexandria we’re talking about. ³³ As in the Mitch Hedberg joke: “I used to do a lot of drugs . . . I still do, but I used to as well.”

234   subject matter—which involves at least that each C-verifying γ is contained in a B-verifying β.)

9.12 Conversational Exculpature Next is to explain how A  B gets relativized to a choice of m, so that m n A B becomes a distinct proposition from A  B. Let a subject matter m, following Lewis, be an equivalence relation on worlds, or what comes to the same, a partition of logical space into m-cells. m S P is the set-of-worlds proposition that is

(i) uniformly true in m-cells containing P ∧ S-worlds, but not P ∧ ¬S-worlds (ii) uniformly false in m-cells containing P ∧ ¬S-worlds, but not P ∧ S-worlds (iii) uniformly gappy in m-cells containing both sorts of world. (31) m P so defined is wholly about m; its truth-status never varies Note that S  within an m-cell. P we’d prefer to be as little about m as possible, to the point ideally that

P has no bearing at all on how things stand m-wise; the P-region overlaps every m-cell. (32) Call that the Independence condition. To assume, for instance, that it’s a martini the guy is drinking leaves it entirely open what his academic field may be. Assuming that he’s a metaphysician, by contrast, fixes his field as m philosophy—with the result that S  P is necessarily true, making it useless as a source of information about actuality.³⁴ The talk of “pivoting” on a presupposition is to indicate that P stakes no claim; its job is purely expressive. All we normally ask of it is to set the stage for S’s depiction of actual m-conditions as the ones obtaining in ³⁴ Hence the strangeness of The metaphysician over there is a philosopher, even on a referential use of the definite description.

:     

235

S ∧ P-worlds as opposed to ¬S ∧ P-worlds. That job is made more difficult if P takes a stand of its own on m: The fact that contextual presuppositions P are often irrelevant [to how matters stand m-wise] is essentially connected to their role as a bridge from the speaker’s claim S to the question m under discussion ([Simons, 2005]). Where P is compatible with a certain answer to m, there is a question as to whether adding S to P still leaves the answer open. But if P rules out a certain answer to m already, there is nothing to test. Thus contextual presuppositions P can play their bridging role only because they have little or no bearing on m. They owe their ability to make the speaker’s utterance relevant in part to their own irrelevance. ([Hoek, 2018]: 172-3, with notational tweaks)

Independence has a lot to be said for it.³⁵ The attraction for us is that (31) takes an especially simple form when Independence and one other condition (Aboutness, coming next) are satisfied. A total proposition is wholly about m iff its truth-value in m-equivalent worlds is the same; it is true in both, or false in both. For a partial proposition to be wholly about m is similar, except we focus on worlds where it has a truth-value.³⁶ Writing S↾P for the partial proposition that is true/false in the same P-worlds as S, and otherwise undefined, Aboutness says that S↾P is wholly about m; if true (false) in w, it’s not false (true) in any m-equivalent of w. (33) The promised simplification, premised on these two conditions’ holding, is this: m S P = the propn true (false) in worlds m-equivalent to worlds where S↾P is true (false). (34)

³⁵ Full independence may be too much to hope for in some cases, as Hoek discusses. ³⁶ Thus a partial proposition is wholly about m if it never flips from true to false between m-equivalent worlds.

236   m S  P comes into clearer focus when developed graphically, stage by stage. This is attempted in the next section, and applied to examples in the sections after that.

9.13 Pictures Take first the subject matter m that the speaker means to be addressing when S is uttered. m like any subject matter is a partition of logical space—the large, light rectangle in Figure 9.1—into pairwise disjoint and mutually exhaustive cells—the nine columns. Also depicted in Figure 9.1 (as a grey wave) is the S-region. Clearly S is not itself about m; its truthvalue varies as we move around in a given m-cell. We will need to restrict S to obtain a basis for extrapolating to a proposition wholly about m. This is where the pivot-presupposition P comes in, represented in Figure 9.2 as a dark bar. S is better-behaved within the P-region, so much so that S’s m behavior there can serve as a model for S  P’s behavior in the rest of logical space. The grey wave (S) is restricted in Figure 9.3 to the P-region. S↾P is true in the grey p-bricks (the first, fifth, and ninth) and false in the others

Figure 9.1 S (grey wave) over m (the nine light columns)

Figure 9.2 P (dark bar) superimposed

:     

237

Figure 9.3 S↾P = S restricted to the P-region

m Figure 9.4 S  P = S↾P extended along m lines

(the dark P-bricks). Figure 9.4 extrapolates S↾P to the rest of logical space m along lines laid down by m. We get as desired a proposition S  P that is (i) wholly about m, while (ii) taking its cue in each m-cell from S’s truth-value in that cell’s P-worlds.

9.14 Leverage This gives us something like a proof of concept for the P/m cognitivecontent-generation machine. Next is to try it out on examples. Take «The Earth will explode tomorrow» and «Terra will explode tomorrow». Though semantically equivalent, they are not equally terrifying; the Earth-ian formulation has an apocalyptic feel that its Terra–ian counterpart can’t match.³⁷ Why would this be? It is not as if we’re in any real doubt about Terra’s identity with the Earth. The theory advises us to look first for a difference in topic. Someone uttering «The Earth will explode tomorrow» is telling us what will happen to us tomorrow. «Terra will explode tomorrow» sounds more ³⁷ Likewise «The Sun will burn out tomorrow» vs «Sol will burn out tomorrow».

238   like something a theoretical geologist would say, on being presented with data about the third planet from Sol. The sentences differ too in the assumptions they raise to prominence. It’s a preconception about the Earth (but not Terra) that it is this very planet. Which means (we said) that speaking of a planet as the Earth raises to pivot status the fact of its being the planet beneath our feet. But, why would these topical and presuppositional differences endow one of the sentences with a more alarming cognitive content than the other? If one is looking for a proposition about our immediate future that’s equivalent in The Earth is this very planet-worlds to The Earth is about to explode, the obvious candidate is that this very planet is about to explode. Whereas if one is looking for a theoretical-geological proposition that’s equivalent in Terra is the third planet from Sol-worlds to Terra is about to explode, the obvious candidate is that the third planet from Sol is about to explode.³⁸ Consider next an example of Stalnaker’s: «My cousin is not a boy anymore». This would normally be heard to say that my cousin (Cuz, for short) is now a man, albeit its literal content allows for various other possibilities. How do we get Cuz now being a man out of «Cuz is not a boy anymore»? Well, the sentence semantically presupposes that Cuz was a boy: a young, male, human being. Cuz is pragmatically presupposed, in the absence of indications to the contrary, to be still a male human being. Independence would fail if the matter under discussion were my cousin’s gender; the presuppositions already answer that. Probably then we are talking about another condition of boy-ness, being of a certain age. The one and only proposition about my cousin’s age that agrees in He’s still male-worlds with My cousin is not a boy is that my cousin is now grown up. Now let’s vary the case a little. We are watching your still visibly 10year-old cousin doing skateboard tricks in the park. The issue can’t be his/her age, because we know it, well enough anyway to see that Cuz is young enough to be a boy. Presumably then we are talking about another condition on boy-ness, being of a certain gender. The one and only

³⁸ The de se aspects are not of the essence. They serve mainly to make the example more gripping. (That the sole inhabited planet is about to explode is bad news too.)

:     

239

proposition about my cousin’s gender that agrees in My cousin is 10-worlds with My cousin is not a boy any more is that my cousin is now non-male, presumably a girl. This hardly scratches the surface of what My cousin is not a boy anymore (henceforth C) can be made to convey. With enough ingenuity m regarding P and m, C  P can represent Cuz as leaving the boys’ room through any number of doors: surgery, deification, evolution into a higher species, and so on. The sentence can, in a context where enough other doors are blocked, “say” that my cousin was transformed like Lot’s wife into a pillar of salt, or by a malfunctioning Parfitian teletransportation device into five boys.

9.15 Implication and Validity The big enchilada of course is Hesperus = Phosphorus. What astronomical information does it convey, relative to assumptions like Ph = Hesperus is the first evening-visible planet and Pp = Phosphorus is the last morning-visible planet? The answer we’re hoping for is: the first evening-visible planet is the last morning-visible planet. Whether this is the answer we get depends on (i) the matter m under discussion, (ii) what precisely is presupposed, and (iii) the value the theory assigns to m Hesperus = Phosphorus  P. A natural choice of m, for Ph and Pp to be m-independent, is which planets thus and so specified are identical to which otherwise specified. (Independence would be at risk if m were which planets thus and so specified are identical to which named planets: Venus, Mars, etc.) The φ-planet might for all Ph (Pp) tells us be either identical to the ψ-planet or distinct from it. Various m-conditions are also possible compatibly with the conjunction of Ph with Pp (Php, call it). Php takes no stand, for instance, on whether the currently fastest-moving planet is the currently brightest one. Yes, but Independence requires Php to be compatible with all m-states. And here we seem to run into trouble. Assuming with Kripke that Hesperus = Phosphorus holds necessarily, Php (Hesperus = the first EVP ∧ Phosphorus = the last MVP) implies, it would seem, that the first EVP = the last MVP. But now let w be a world where the first EVP is Venus and the last MVP is Mars; and let m(w) be its m-cell. Php cannot

240   as a First EVP = last MVP-implier be consistent with every m-cell, for in m(w) the Morning and Evening Planets are distinct. You can guess our response from the stress laid earlier on logical necessity, as opposed to metaphysical. Php does imply The first EVP = the last MVP in a sense. But it doesn’t imply it logically. Students who reckon a = the φ b = the ψ So, the φ = the ψ a valid form of argument will be asked to repeat the course. A further premise is needed: a = b. The argument would be valid, if a = b were a logical truth. But it isn’t; there are countermodels. Which gives us a way out. Php is NOT in the relevant sense inconsistent with m-states where the first EVP ≠ the last MVP, since it does not in the relevant (logical) sense imply that the two are identical. This may seem at odds with the logical space diagrams in Section 9.13. It all depends, though, on how the diagrams are interpreted. We are back in business if logical space is made up of logically possible worlds, a proper subset of which are metaphysically possible. (Worlds may be identified for these purposes with maximal consistent sets of interpreted sentences.³⁹) Php is again m-independent, and the construction can proceed as before. m Let’s try it. Hesperus = Phosphorus  Php is w-true, according to (34), just when w is m-equivalent to a Hesperus = Phosphorus ↾ Php—world. The characteristic feature of these latter worlds where m is concerned is that the planet, be it Venus, Mars, or whatever, appearing last in the morning turns up again first in the evening. (It’s always Venus, in Phpm worlds, but m takes no notice of that.) Hesperus = Phosphorus  Php is true, then, just where the first EVP = the last MVP. It is false in worlds m-equivalent to Hesperus ≠ Phosphorus ↾ Php—worlds. What these latter have in common m-wise is that the first EVP (be it Venus, Mars, or ³⁹ There might be advantages to using instead the world-surrogates developed in [Lederman, 2001]. Thanks here to Daniel Hoek.

:     

241

whatever) is distinct from the last MVP; so we’re talking about the worlds where different planets appear early and late. Whereupon m Hesperus = Phosphorus  Php comes out just as we’d hoped. It’s true where the Evening Planet is the Morning Planet, and false where they’re distinct. You might still be unsatisfied. The whole idea was that «Hesperus = Phosphorus» should differ cognitively from «Hesperus = Hesperus», despite agreeing with it semantically. They don’t agree even semantically, however, if worlds are maximal consistent sets of sentences (maxisets). «Hesperus = Hesperus» is consistent with «Hesperus is φ but Phosphorus is not-φ»; «Hesperus = Phosphorus» by the indiscernibility scheme for identity is not consistent with «Hesperus is φ but Phosphorus is not-φ». But then the two belong to different maxisets. Insofar as «S»’s semantic content is the set of «S»-containing maxisets, the semantic contents are going to be different. Reply: This forgets that we have two versions of logical space up and running. There’s the set Λ of logical possibilities, and the subset Ω of Λ that throws out worlds that aren’t metaphysically possible. We have accordingly two candidates for the role of «S»’s semantic content: |S|Λ and |S|Ω (= |S|Λ intersected with Ω). You’re right that |Hesperus = Phosphorus|Λ ≠ |Hesperus = Hesperus|Λ. But the intensions again coincide when Ω is substituted. Ω is exhausted just as much by its «Hesperus = Phosphorus»-worlds as its «Hesperus = Hesperus»-worlds. If semantic contents are |S|Ωs, they’re coincident in much the same cases as the traditional problematic assumes. A second objection now suggests itself, this one on the cognitive side. There may indeed be worlds ruled out by the extended cognitive content of «Hesperus = Phosphorus», made up of logically possible ws. This hardly explains why the sentence should strike us as informative, for the ruled-out worlds might not have had any genuine (metaphysical) chance of obtaining. If «Hesperus = Phosphorus» is to provide real, actionable, information, a cognitive content is needed that fails in metaphysically possible worlds. Reply: Much as «S»’s semantic content can be conceived as the set |S|Ω of metaphysically possible S-worlds, its cognitive content can be m conceived as the set ⟦S⟧Ω of metaphysically possible S  P-worlds.

242   ⟦Hesperus = Phosphorus⟧Ω is the set of metaphysically possible worlds whose Morning and Evening Planets are identical. This set falls well short of Ω, by virtue of excluding (metaphysically possible) worlds where, e.g., the last morning-visible planet is Venus and the first eveningvisible planet is Mars. «Hesperus = Phosphorus» is cognitively non-trivial, then, even in its import for metaphysically possible worlds.

9.16 More Pictures The mechanism at work in these examples might remind us of something from high school physics (Figure 9.5)—the sort of arrangement that Archimedes was talking about when he said, “Give me a lever long enough and a fulcrum on which to place it, and I shall move the world.” The fulcrum role is played in our system by the pivot-presupposition P; see Figure 9.6.⁴⁰ Where the left-hand ball in Figure 9.5 uses the fulcrum as a force modifier—a force multiplier in the usual case (“multiply your output,” the sign says), S in Figure 9.6 uses P as a content modifier. This in two ways. Pivoting on a fixed P gets S to express a modified content, influenced by but distinct from S’s semantic content. Pivoting also multiplies, if we allow P to vary, the range of contents S can be used to express. As for the right-hand ball (the world, for Archimedes), this becomes an arc made up of the ways worlds can be, or be said to be, where a given LEVERAGE MULTIPLY YOUR OUTPUT

Lever

Fulcrum

Figure 9.5 Lever/fulcrum arrangement

⁴⁰ Fulcrums in physics are sometimes called pivot-points.

:     

243

LEVERAGE MULTIPLY YOUR OUTPUT m Pointer s P

Figure 9.6 Pointer/pivot arrangement

Pivot

subject matter is concerned. The arc if m has k cells is made up of 2k-2 propositions, one for each non-trivial union of m-cells. The lever, finally, becomes in Figure 9.6 a pointer indicating which such union—which wholly-about- m proposition – is expressed in the relevant context by S.

9.17 Summing Up A lot of the above machinery was developed by fictionalists, instrumentalists, if-thenists, and figuralists of various stripes. Some of it traces back to work in aesthetics on metaphor, irony, hyperbole, and truth in a story.⁴¹ These are not mainstream analytic topics, obviously, though the reasons aren’t always clear. Hyperbole has a lot in common with loose speech, which is on everyone’s radar lately. Be all that as it may, cognitive significance is as mainstream as topics get. You can find it in some of the field’s founding documents (“The Thought,” “On Sense and Reference,” “On Denoting”). Should it bother us if tools made for metaphor are brought into the inner sanctum, where the serious philosophy is supposed to be done?⁴² Frege did not think of cognitive content as additional to first meaning; he thought it was part of first meaning. Add to this his determination to

⁴¹ See for instance [Walton, 2005], and the last few chapters of [Yablo, 2014]. ⁴² [Crimmins, 1998] attempts to bring Waltonian make-believe to bear on content-attribution; it anticipates in some ways what we are trying here. [Richard, 2000] responds from an orthodox perspective.

244   keep color, tone, register, secondary sense, and the like out of scientific semantics, and one suspects he would have hated the present project. A lot has changed since 1892. Most of us want nowadays to link cognitive content to the information imparted by a sentence, and to resist any simple reduction of information imparted to literal semantic content. Thus Soames: the relationship between the semantic content of a sentence and what the sentence is used to say, or assert, is looser than commonly thought. ([Soames, 2005], 260)

The difference is that where he wants to get this result by making semantic contents schematic—capable of being filled out in more than one way—we have sketched a more flexible, and more productive, arrangement. Semantic contents generate cognitive contents by operating levers in a larger piece of machinery, involving also the presupposition P being leveraged, and the subject matter m being addressed. m Not all of the contents S  P obtainable in this way are the kind of thing Frege would have wanted to count into S’s sense, or the thought expressed. This could be seen as a problem. It could be, however, that the time has come for cognitive content to take its place alongside other forms of non-semantic content that have traditionally been accorded less respect. What sets it apart from these other forms is a fair question. It’s a question the leverage model invites but does not take a stand on.⁴³

References Bach, K. (1994). Conversational impliciture. Mind & Language, 9:124–62. Berg, J. (1988). The pragmatics of substitutivity. Linguistics and Philosophy, pages 355–370. [Berg, 1998]. ⁴³ These ideas have been bouncing around for a long time. I’m not sure at this point who all needs thanking, but at least the following should be mentioned: Hüseyin Güngör, Sally Haslanger, Daniel Hoek, Kai von Fintel, Jack Spencer, Manuel García-Carpintero, Katharina Felka, Agustín Rayo, Ken Walton, David Hills, Seth Yalcin, Daniel Rothschild, Andy Egan, Martín Abreu Zavaleta, Thony Gillies, Dilip Ninan, Sarah Moss, Dan Greco, Robert Stalnaker, Mark Richard, Justin Khoo, Eric Swanson, Ofra Magidor, Mark Crimmins, Matteo Plebani, and Kit Fine. Special thanks to Hüseyin for catching a major error.

:     

245

Berg, J. (1998). In defense of direct belief: Substitutivity, availability and iterability. Lingua e Stile, 33(3):461–470. Berg, J. (2012). Direct Belief. De Gruyter Mouton. Philosophical Studies, 180: 2727–52. Berto, F. and L. Hornischer. (2023) Cognitive Synonymy: A dead parrot? Philosophical Studies, 1: 1–26. Carpintero, M. G. (2000). A presuppositional account of reference fixing. Journal of Philosophy, 97(3):109–147. Crimmins, M. (1998). Hesperus and Phosphorus: Sense, pretense, and reference. Philosophical Review, 107(1):1–47. Davis, W. (2019). Implicature. In Zalta, E. N., editor, The Stanford Encyclopedia of Philosophy. Metaphysics Research Lab, Stanford University, Fall 2019 edition. Fine, K. (2017). A theory of truthmaker content II: Subject-matter, common content, remainder and ground. Journal of Philosophical Logic, 46(6):675–702. Fine, K. (2020). Yablo on subject-matter. Philosophical studies, 177(1): 129–171. Frege, G. (1948). Sense and reference. Philosophical Review, 57(3): 209–230. [Fuhrmann, 1996] Fuhrmann, A. (1996). An essay on contraction. University of Chicago Press. [Glick, 2012] Glick, E. (2012). A modal approach to intentional identity. Noûs, 46(3):386–399. Gomez-Torrente, M. (2011). Kripke on color words and the primary/ secondary quality distinction. In Berger, A., editor, Kripke, pages 290–323. Cambridge University Press. Grice, H. P. (1975). Logic and conversation. In Cole, P. and Morgan, J. L., editors, Syntax and Semantics 3: Speech Acts, pages 41–58. Academic Press, New York. Hintikka, J. (1970). Objects of knowledge and belief. Journal of Philosophy, 67(21):869–883. Hoek, D. (2018). Conversational exculpature. Philosophical Review, 127(2): 151–196. Humberstone, L. (2011). The Connectives. MIT Press. Kripke, S. (1980). Naming and Necessity. Harvard University Press, Cambridge, Mass. Kripke, S. (2011). Identity & necessity. In Philosophical Troubles, pages 1–26. Oxford University Press.

246   Kripke, S. (2013). Reference and Existence: The 1973 John Locke Lectures. Oxford University Press. Lederman. H. (2021). Fine-grained semantics for attitude reports. Semantics and Pragmatics, 14(1). Lewis, D. (1983). Individuation by acquaintance and by stipulation. Philosophical Review. 92(1):3-32. Linsky, L. (1983). Oblique Contexts. University of Chicago Press. McKay, T. (1981). On proper names in belief ascriptions. Philosophical Studies, 39(3):287–303. Quine, W. V. O. (1960). Word and Object. MIT Press, Cambridge, MA. Ramsey, F. P., and Mellor D. (ed.), (1990). Philosophical Papers. Cambridge University Press. Richard, M. (2000). Semantic Pretense. In Everett, A. J. and Hofweber, T., editors, Empty Names, Fiction and the Puzzle of Non-Existence, pages 205–232. CSLI. Salmon, N. (1986). Frege’s Puzzle. MIT Press, Cambridge. English. Santorio, P. (2012). Reference and monstrosity. Philosophical Review, 121(3):359–406. [Santorio, 2018] Santorio, P. (2018). Alternatives and truthmakers in conditional semantics. Journal of Philosophy, 115(10):513–549. Simons, M. (2005). Presupposition and Relevance, pages 329–355. Semantics versus Pragmatics, Szabo, Z.G. Clarendon Press, Oxford. Soames, S. (1989). Direct Reference and Propositional Attitudes, pages 393–420. Themes from Kaplan. Oxford University Press, New York. Soames, S. (2002). Beyond Rigidity. Oxford University Press, Oxford. Soames, S. (2005). Naming and asserting. In Szabó, Z. G., editor, Semantics vs. Pragmatics, pages 356–382. Oxford University Press, USA. Soames, S. (2008). Drawing the line between meaning and implicature–and relating both to assertion. Noûs, 42(3):440–465. Stalnaker, R. (1975). Indicative conditionals. Philosophia, 5(3):269–286. Stalnaker, R. (1986). Counterparts and identity. Midwest Studies in Philosophy, 11(1):121– 140. Stalnaker, R. (1999). Assertion, pages 78–95. Oxford University Press. Stalnaker, R. (2010). Responses to Stanley and Schlenker. Philosophical Studies, 151(1):143–157.

:     

247

Stanley, J. (2010). “Assertion” and intentionality. Philosophical Studies, 151(1):87–113. [Stojnić, 2021] Stojnić, U. (2021). Context and coherence: The logic and grammar of prominence. Oxford University Press. Stojnić, U., Stone, M., and Lepore, E. (2017). Discourse and logical form: Pronouns, attention, and coherence. Linguistics and Philosophy, 40(5): 519–547. Swanson, E. (2006). Frege’s puzzle and pragmatic presupposition (ms). Swanson (ms.) says “unpublished”; I prefer Yalcin (ms.) which does not. [Swanson, 2012] Swanson, E. (2012). Propositional attitudes. In Claudia Maienborn, K. v. H. and Portner, P., editors, Semantics: A Handbook of Natural Language Meaning. De Gruyter. Thomasson, A. L. (2014). Ontology Made Easy. Oxford University Press. Walton, K. (2005). Metaphor and prop-oriented make-believe. In Kalderon, M. E., editor, Fictionalism in Metaphysics. Clarendon Press. Weatherson, B. (2001). Indicative and subjunctive conditionals. Philosophical Quarterly, 51(203):200–216. Williamson, T. (2011). Reply to Stalnaker. Philosophy and Phenomenological Research, 82(2):515–523. Yablo, S. (2008). Beyond rigidification: The importance of being really actual. In Things. Oxford University Press. Yablo, S. (2012a). Aboutness Theory. https://www.academia.edu/2604116/ Aboutness_Theory. Yablo, S. (2012b). Explanation, extrapolation, and existence. Mind, 121 (484):1007–1029. [Yablo, 2014] Yablo, S. (2014). Aboutness. Princeton University Press. Yablo, S. (2016). Ifs, ands, and buts: An incremental truthmaker semantics for indicative conditionals. Analytic Philosophy, 57(3):175–213. Yablo, S. (2017). If-thenism. Australasian Philosophical Review, 1 (2):115–132. Yablo, S. (2020). Nonexistence and Aboutness: The Bandersnatches of Dubuque. Crítica (Mexico, DF), 52(154):77–100. Yablo, S. (2021). Say Holmes exists; then what? In Art, Representation, and Make-Believe, pages 345–366. Routledge. Yalcin, S. (2021). Iffy knowledge and iffy existence. (ms.).

Index Note: Tables are indicated by an italic “t ”, following the page number. For the benefit of digital users, indexed terms that span two pages (e.g., 52–53) may, on occasion, appear on only one of those pages. acceptance 10–11, 16–17, 26–7 common knowledge of 11–14 conditional 16n.21 defining 14 mental attitude 9–10, 15–16 propositional attitude 15–16 accommodation, presupposition as a repair mechanism 26–7 action, philosophy of 1, 3 adjectives 136–9 color 99–100 modification of nouns 136–7 non-referential 139–40 number words, and 136–7, 139–40 scalar 4 anaphora pronominal 15 propositional 28–30 resolution of 3–4 anchoring see also descriptions; descriptivism; mental states; singularism; thought anchored character attribution and representation 53–8, 61–2 anchored descriptivists 47–9, 57–62 of direct thoughts about external objects 39–40 over-attribution objection against anchored descriptivism 48–54, 58 role of 39–45

singularity anchoring to particulars 47 antisymmetry principle 6n.9 arguments 123–4 ambiguous 133–4 based on apparent substitution failures in attitude ascriptions 202 burden-of-proof 109 of descriptivists 60–1 expositives 70, 82–9 externalist 35–6 presumptive lies 160, 163–4 propositional Russell-Myhill 8n.12 schmidentity strategy see schmidentity strategy (in Kripke’s writings) semantic externalists 1 singularism 33, 53–4 and thought 56–9 unusual 134 Aristotle 49, 226 assertions 74–6 vs. assertive family 74–5, 75t and commitment 74, 75t, 80 in conversation 216 expositives 83 fidelity constraint on 74 frankness norm 74 liability dimension of assertoric commitment 74 norms governing 74–5 performatives 85–6

250  assertions (cont.) pragmatic repair 216 as signals 74 assertive family 74–6 vs. assertions 74–5, 75t illocutionary acts and commitments 76, 83 attitude ascriptions 36, 38 schmidentity strategy 186, 188, 190, 198, 202–3 substitution failures 200, 202 truth-values 190–3, 197 Austin, J. L. 70, 82, 174 Bach, E. 153–4 Bach, K. 83–4 bald-faced lies 165–6, 172–3 baptism model 214 beliefs/belief attributions 16, 36, 38, 159–60, 176–80 see also Kripke, Saul; Salmón, Nathan both believing and withholding believing the same thing 176–9 first-order 179 guises 177–80 language, using to influence 24 and propositions 176–8 relation of believing 36 second-order 178–9 what is believed to be false 159–60, 163–4 Blackburn, S. 131–2 Buchanan, R. 12–13 Byzantine Generals problem 10 Castañeda, H.-N. 180 Chalmers, D. J. 51–3 characters anchored characters and representation 53–8, 61–2 mental states 54–5, 62 “Church–Langford translation test” 206–7 coerced vs. bald-faced lies 165–6, 172–3

cognitive contents 205–41 see also descriptivism; mental states; reference-fixing, descriptive; singularism; thought cognitive significance issues 213 coincident senses 207–8 CONTINGENCY 218–19 conversational exculpature 234–6 conversational implicature see conversational implicature co-referring names 206–7 definite descriptions 209–10 diagonalization 214–16, 218–20, 223 distinctive 224 fidelity as synonymy 210–11 Hesperus = Phosphorus utterance 212, 214n.8, 217–18, 221, 223, 225, 229–31, 233–4, 239–40 incremental 230, 232 leverage 237–40 non-semantic content 240–1 pictures 236–7 pragmatic enrichment 221–6 pragmatic repair 216–20 presupposition subtraction 226–32 reference-fixers 211–14 subject-matter 231–4 trigger problems 219–21 UNIVOCALITY 218–20 Cognitivist scoreboards 9–14, 18–19 non-Cognitivist 24 grades of non-Cognitivism 19–24 viability of 30 problems for 13–19 Cohen, L. J. 82–3, 88–9 coincident senses 207–8 commitments anchored character attribution and representation 53, 55 and assertions 74, 75t, 80, 82 assertoric 74, 76, 88 basic 44–5 conflicting 88–9 conjectural 76, 93

 descriptive/singular thought distinction 38–9 expositives 82 illocutionary 74–6, 83 indices as a cluster of 72, 87–8 massive reduplication 44–5 misattribution response 53 modes of 83 multi-dimensional approach to 91–2 nature of 91 propositional 87–8 psychological 50, 54 public undertakings of 72 semisertion 81 speaker 162 strands of 76 to strange entities 103–4 undertaking 72–4, 83, 87–8, 93 common knowledge 2–3, 9–14, 93 communication 69–70, 129–30, 216 chain of 49 everyday 95 as exchange of information 68 linguistic 70–1 in non-human species 69–72, 78–9 primitive forms 95 rational 217 and speech act 70 Vervish expressions 78–80 concepts 70, 130–2, 144–6, 149–50 of causation 50 central 147–8, 151–2 flawed 130–1 inescapable 146–8, 152–3 logical 147 normative 146–7 object-dependent 36n.4 property 42–3, 50 references 52–3 replacing 147–8 representation 56–7 of truth and fact 147 conditionals 17 first 18 a priori 226

251

second 18 Thomason 17–18 conjectures 74–6, 82, 92 context-dependence disquotationalism 102, 105–7, 123 natural language 99–100 norm of 99–100 sources 107 and translation 107–14 and truth 102–3 context principle, in the Grundlagen 1 conversation assertions in 216 context set 216 cooperative activity of 216 flow of 29–30 Gricean maxims 221–2 imperatives 3–4 ordinary 156–7, 167–8, 174 scoreboard metaphysics 10 turn-taking 77 conversational exculpature 234–6 conversational implicature see also cognitive contents and diagonalization strategy 223 Gricean 93–4, 167–8, 222 and lying 167–8, 173 Manner, Maxim of 95 nondetachable 222 Quantity, Maxim of 93–5 signaling 70, 93–5 conversational scoreboards 1–2, 5, 8–9 Crawford, S. 178–80 cultural evolution (CE) behavior patterns 78 Nash equilibria 78 of speech acts 70, 76–81, 95 Davidson, D. 112–14, 117–18, 121 on theory of meaning 123–4 deceptive intent 156–7, 164–5, 173 see also lying absence of 165–6 coerced vs. bald-faced lies 165–6, 172–3

252  deceptive intent (cont.) whether necessary condition for lying 165–7 declarations 84 demonstratives 108, 216 descriptions 197–201 see also descriptivism; referencefixing, descriptive definite 42n.16, 195, 198, 209–10 “description” theory of reference 51–2 disguised 211–12 epistemic 210 and fidelity 210 general 40 indefinite 4 mediation of thought by 33–4 metasemantic differences from names 211 misattribution response 51–3 and names 210–12 purely qualitative 40, 47 reference-fixing 49, 212, 214–16, 226 referential uses 186 Revised Massive Reduplication Argument 44–6 Russellian theory of 45–6, 185–6, 195, 211–12 schmidentity strategy 197–201 semantic 210–11 stipulations 195 token reflexive 50 descriptivism 60–1 see also descriptions; descriptivism; singularism; traditional picture anchored descriptivists 47–9, 57–62 causal 47 content claims 47–8, 50, 52–4, 58–9 descriptive mediation 33–4 descriptive/singular thought distinction 33–9, 59–60, 60n.41 agreements and disagreements 60–2 opposition to new traditional picture 38–40, 48

over-attribution objection against anchored descriptivism 48–54 criticism of 58–9 misattribution response 50–3 paradigm cases of thought 37 possibility of descriptive thought 41 pure descriptivists 47 purely descriptive thought 43 reference-fixing 37, 49, 211–15, 226 relational 49 determiners 136 modification of nouns 136–7 non-referential 139–40 number words, and 137, 139–40 semantic values 137–8 syntactically displaced 138–9 diagonalization 216, 218–20, 223 indicative suppositions 225–6 propositions 214–16 discourse referents 4, 15–17 discourse representation structures (DRSs) 4 disquotationalism 99–102 context-dependence 102, 105–12, 123 names 100 propositions 99 ‘pure’ disquotational notion of reference 108 theories 99, 101–2 translation 104–7 and context-dependence 108–12 truth, role of 102–4 extended 105 truth-predicate 102, 104–6 epistemology linguistic scoreboards 12, 22 and metaphysics 134–5 questions in 152–3 evidentials 4, 79 direct 91–2 grammatical 70, 89–93 illocutionary relation 90 indirect 93 inferential 91–2

 not-at-issue restriction 90 presentation of at-issue content 90 tripartite analysis 90 expositives 70, 82–9 see also illocutionary acts and commitments attitudinative, content embedded within 82–4 complement clauses 83 contentions 83 examples 82 illocuted clauses 82–4 performatives 84–6 sentences 85 extended references (E-references) 108–10 external objects (ordinary), thoughts about 35, 47, 61 direct, anchoring role of 39–40 perception-based 46 singular 33, 35–6, 38, 40, 59–61 facts astronomical 206 belief as to 130–1 Cognitive 11–12 content 124 epistemic 9–12, 134 internal processing 110 knowledge-of-knowledge 13–14 and language 131–2, 141–2 linguistic 134 local, about linguistic usage 196 mathematic sentences 191 mental ability 49 metaphysical 134 mind-independent but easily verified 19–20 normative 21–2 numbers, about 141–2, 145 presumed 229 psychological 9–12, 52–3 first-order 50 quasi-realism 131–2 reality, about 132–3, 150

253

referential 62 schmidentity strategy 191, 194 semantic 107–9, 195, 197 that-clauses 150 totality of 150–1 fidelity constraint on assertion 74 definite descriptions 209–10 forms 92 liability, frankness and fidelity (LFF), norms of 74–6 norm of 80–1 as synonymy 210–11 Field, H. 102–6, 123 Fine, K. 189 Fodor, J. 114–15, 117 frankness liability, frankness and fidelity (LFF), norms of 74–6 norm for assertion 74, 80–1 standard of 91–2 Frege, F. L. G. 49, 136, 202–3, 207–8, 220 Begriffsschrift 205–6, 208, 233–4 context principle, in the Grundlagen 1 Frege’s Puzzle see Salmón, Nathan “On Sense and Reference” 221 “Uber Sinn und Bedeutung” 205 game theory 10 García-Carpintero, M. 226 Gaskin, R. 131–2 GEs see grammatical evidentials (GEs) Goodman, R. 33–4 grammatical evidentials (GEs) 70, 89–93 see also evidentials content 90 illustrations 89–90 speech acts 91–2 updating 91 Green, M. 86 Grice, P. conversational implicature 93–4 and lying 157–9, 162, 167–9

254  Grice, P. (cont.) conversational maxims 221–2 Gricean program 12–13 on intentions 70–1 guesses 75–6 Harnish, R. 83–4 Hawthorne, J. 10–11 Hesperus = Phosphorus utterance 212, 214n.8, 217–18, 221, 223, 225 Heyes, C. 77 Hoek, D. 231–2, 235 Hofweber, T. 138–9, 147–8 homographs 100n.2 homonyms 100n.2 illocutionary acts and commitments 76, 83 proto-illocution 82, 95 illustrations 82 imperatives 3, 29–30 implicature see conversational implicature; Grice, P. independence 235 indexicals 49–50, 112 indicatives 225 indices as a cluster of commitments 72, 87–8 and signals 67–70 verbal 72–4 indirect speech acts (ISAs) 83–4 information 26–8, 85–6, 90, 93–5, 112–13, 115–18, 121, 206, 221, 240 attachment-available and unavailable 28 backgrounded 4, 27–9 bodies of 2–3, 18–19, 25 causal 43–4 common ground 9–10, 16n.21, 24, 28 conveying 68, 71–4, 76–7, 79–81, 92, 95, 118, 121 encoding 71–2, 79–80, 87–9, 117–19, 239–40 foregrounded 4, 27–9

informational indices of evaluation 17 interpretation, enabling 118 names, use of 118 new 28–9, 76–7 possession 117 propositional 17–18 public 14n.19 relevance 222 reliable 24 sequential 6n.9 shared 9–10 shifting 17–18 sources 80, 89, 234–5 speaker 117 sports-related 2 in a translational manual 118, 120–1 value of 187, 205–6, 211, 223 informational scoreboards 22 intentionality see also deceptive intent Gricean 70 intentional content, notion of 49 intention of speaker see meaning reflexive-communicative intentions 73 theory of 39, 44–5 verbal signals 70 intentional realism 123–4 interrogatives 3–4 Jackendorff, R. 95 Jackson, F. 52–3 Jary, M. 84–7 Kaplan, D. 54–5, 216 Karttunen, L. 26 Katz, J. 114 knowledge antecedent 145–6 common 2–3, 9–14, 93 condition 41 establishing 74 infinite store of 13–14 iterated 10–11, 13–14

 justification for 92 knowledge-of-knowledge facts 13–14 lack of 42–3 private 23–4 proper 206 shared 23–4 Kratzer, A. 27 Kripke, S. see also schmidentity strategy (in Kripke’s writings) on acquaintance relations 49 baptism model 214 both believing and withholding believing the same thing (example of Pierre) 176–9 on contingent truths 211 defense of Russellian theory of descriptions 185–6, 195 meter stick case 213 and Millianism 176 and presupposition subtraction 226 rejection of Russellianism 211–12 writings Naming and Necessity 176, 182–90, 201–2, 210 “A Puzzle about Belief ” (1976 lecture) 176 “Speaker’s Reference and Semantic Reference” 183–4 language see also language–metaphysics gap colloquial 120 conventionally meaningful 68, 71–2 facts about 131–2, 141–2 hypothetical 183–4 Mentalese (language of thought) 106, 114–15, 119 and mental states 1–2 and metaphysics 129–54 natural see natural language proto-language 79–80 psychologistic/antipsychologistic battle in theory of 1–2 science of 23

255

transcendental precipitate of 131–2 using to influence beliefs 24 language–metaphysics gap 129–35 see also metaphysics bridging 135–43, 153–4 illusionary 135 Lederman, H. 14 Lepore, E. 114–15 leverage 237–40 Lewis, D. 2–3, 8–9, 20–1, 114, 118 liability, frankness and fidelity (LFF), norms of 74–6 linguistic idealism 142 Loewer, B. 114–15 lying “basically lying” 160, 167–8, 170–1 borderline cases 164 calling someone a liar 170, 172–3 constraints on philosophical definitions 157–8 debates 165–8 deceptive intent 164–5, 173 absence of 165–6 coerced vs. bald-faced lies 165–6, 172–3 whether necessary condition for lying 165–7 defining 161–2 dictionary definitions 166 experimental studies informing philosophical definitions, whether 168–71 Gricean conversational implicature, through 157–9, 162, 167–9, 171, 173 in a “loose” sense 157–8, 160–1, 165, 168–74 moral blameworthiness 172–3 ordinary varieties of lying 157–8 ordinary ways of speaking, taking cues from 174–5 paradigm lies 159 philosophical “definitions” defense of 161–5

256  lying (cont.) experimental studies informing, whether 168–71 and ordinary ways of speaking 174 philosophical literature 167 presumptive lies 160, 163–4 in a “prototypical” sense 156–61, 163–5, 168, 171–2 in a “strict” sense 156–9, 161, 168–70, 174 deceptive intent see deceptive intent defense of a philosophical “definition” 161–5 factual falsity of what is believed to be false 163–4 philosophical significance 165 saying 159–60, 162–3, 173 twofold irony in ordinary everyday remarks, puzzle of 171–3 types and prototypes 166–7 what is believed to be false 159–60, 163–4, 169 Manley, J. 10–11 massive reduplication see reduplication, massive; Revised Massive Reduplication Argument meaning 1, 211 conventional 68, 71–2 definite descriptions 198 differing in 189–91 exact 161n.12 linguistic 12–13, 71 literal 85, 173 and lying 161–2 ‘making sense’ and translation 120–2 natural 71n.4 organic 71n.4 personal 71n.4 and signaling 68, 71–2 speaker 70–1, 73–4, 93–4, 95n.18 theory of 115n.28, 123–4 and translation 120–3 truth 174 words 70n.3, 112–14, 123

Meibauer, J. 162–3 Mentalese (language of thought) 106, 114–15, 119 mental states 9–10, 23–4, 34–5, 52–3, 57–8 see also thought ascription 37–8 characters for 54–5, 62 and Cognitivism 12 collective 12 compositional structure 57–8 contents 34–5, 38, 47, 52–3, 60, 62 content-typing 61–2 explanatory role of property for 57–8 and language use 1–2, 12n.17 property, roles played by 57–8 relation between 62 representation of 56–60 rigidified descriptive cluster theory of mental content 51 scoreboards as 1–2, 8–10 shared 19 token 46–8 types 34, 61–2 metaphysics see also descriptions; descriptivism; reality contemporary 130–1 corrective vs. constructive role for philosophy of language in 133–4 defining 129–30 descriptive 153–4 epistemology, role of 134–5 evaluating the metaphysical question 143–8 immanent stance towards 149–51, and status of language 151–4 language–metaphysics gap 129–35 bridging 135–43, 153–4 illusionary 135 linguistic scoreboards 7–13 metaphysically deep or shallow results 144–5 natural language 153–4

 natural numbers, questioning existence of 143–8 ontology 136, 144–5 place of philosophy of language in 129–54 representations, place in 148 and separation of philosophical disciplines 134–5 status of metaphysical question, assessing 143–4, 148–51 transcendent stance towards 149–51 meter stick case 213 Millianism defense of 176, 188, 192, 195 Kripke on 176 names, Millian theory of 186, 188–9, 199–203 non-relationist versions 189 presupposition subtraction 226–7 mind, philosophy of 1, 33 see also mental states; thought anchoring role 39 influence of philosophy of language on 1, 37–8 representational theory of mind 56–7 singularism 34, 36–9 misattribution response 50–3 descriptions 51–3 modal horizon 4 modal stability names and descriptions 197–201 semantic properties 195–7 Moltmann, F. 153–4 morphemes 115 Murray, S. 89–91 names 38, 49, 116–18 ambiguous/ambiguity theory of 100–1 cognitive significance 213 constitutively non-referential 139–42 coreferential 186, 198, 201–3, 206–7, 223–4 co-referring 186, 206–7 de facto non-referential 139

257

and descriptions 210–12 disquotationalism 100 empty 224 given 115, 118 metasemantic differences from descriptions 211 Millian theory 186, 188–9, 199–203 modal stability 197–201 Naming and Necessity see under Kripke, Saul natural language 119 nicknames 115–18 number words as 136–7 pairs of 189–90 predicate view of 100–1 proper 36–7, 100, 108, 137–8, 200, 211–12 relation between 116, 182–3 retention of meaning 211 schmidentity strategy 197–201 surnames 100n.3 natural language 35–6, 139, 143–4 disquotationalism 99–100, 103, 106 metaphysics 153–4 number words in 136, 139, 143–4 ontology 153–4 translation 119 normative concepts 146–7 nouns 136–7 numbers facts about 141–2, 145 natural 148 questioning existence of 143–8 prime 191 number words and adjectives 136–7, 139–40 and determiners 137, 139–40 in English 145 as names 136–7 in natural language 136, 139, 143–4 non-referential 139–42 syntactically displaced 137–9

258  ontology linguistic 12 metaphysics 136, 144–5 natural language 153–4 scoreboards 19, 22 operators, focus-sensitive 3–4 over-attribution objection against anchored descriptivism 48–54 criticism of 58–9 misattribution response 50–3 performatives explicit 84, 86 traditional notion of 73–4 as unavoidable context changers 85–6 philosophical “definitions” defense of 161–5 experimental studies informing, whether 168–71 philosophy of action see action, philosophy of philosophy of mind see mind, philosophy of pictures 236–7 pivot-presuppositions 228–9, 232, 240 possible worlds view 51–2, 190–1 of content 191–2 of propositions 7–8, 191, 198, 201, 218 semantics 190, 192, 195 and sentences, semantic contents 190, 196 sets of 190–2 stipulations 201–2 Postal, P. 114 postulations 82 pragmatic enrichment 221–6 pragmatic repair 216–20 assertions 216 context set 216 horizontal content 217 preconceptions 226 prejudices 226 presumptions 75–6, 228–9

presuppositions 26 pivot-presuppositions 228–9, 232, 240 presupposition subtraction 226–32 sentences 26–7 pronouns 4, 21–2 proper names 36–7, 100, 108, 137–8, 200, 211–12 propositions see also diagonalization; schmidentity strategy (in Kripke’s writings); scoreboards; truth acceptance as a propositional attitude 16 and belief attributions 176–8 contained on scoreboards 17 co-ordinated 189 defining 7–8 diagonal 214–16 disquotational theories 99 evidential 90–2 flow of 5 general 34–5 horizontal 216–17 leverage 240 linguistic scoreboard contents 3–7, 12, 17–18 mental states 34–5 natures of 7–8 partial 235 possible worlds view of 7–8, 191, 198, 201, 218 presupposed 216 schmidentity strategy 191, 196–7 scope 90–2 scoreboards replacing, whether 7–8 sentences expressing 218 singular 34–6, 46–8, 60 static propositional content 5–7 structured 34–6, 196–7 that-clauses 103–4 theories of 7–8 and truth 103–4, 104n.9, 220 ubiquitous propositionalizing 16–18

 proto-illocution 82, 95 proto-language 79–80 quantifier domain restrictions 4 Quine, W. V. O. 101n.4, 105, 112–13 on theory of meaning 123–4 Ramsey, F. P. 225 realism intentional see intentional realism quasi-realism 131–2 reality see also metaphysics antecedent knowledge 145–6 capturing 149 derivative of language 135–6 facts about 132–3, 150 in general 129–30, 132–4, 142, 146 language–metaphysics gap 129–34 and numbers 143–4 probing 141–3, 145–6, 150–1 representations of 129–33, 141–2, 153–4 substantial conclusions about 132–3 Recanati, F. 50 reductionism 22 reduplication, massive 41, 43–7, 54–5, 61 see also Revised Massive Reduplication Argument; Strawson, Peter F. epistemic possibility of 41–2, 44 and object-involving t-conditions 47 remaining open 40 and rigidity 47 thought experiment 42 reference-fixing 37, 49, 211–15 descriptions 49, 212, 214–16, 226 diagonal propositions 215 presupposition subtraction 226 Reichert, Paula 69 relational descriptivism 49 reports 82 representation and anchored character attribution 53–8, 61–2 content-typing 61–2

259

descriptive/singular thought distinction 53–4, 60–2 discourse representation structures (DRSs) 4 of identity 56 of mental states 56–60 misattribution response 50 notions of 56–7, 62 of reality 129–33, 141–2, 153–4 theoretical 50 theory of mind 56–7 resource holding potential (RHP) 69 Revised Massive Reduplication Argument 43–5 conclusion of 45–8, 54–5 rigidified descriptive cluster theory (RDC) 51 Russell, B. 35–6, 209 see also Russell-Myhill paradox Kripke’s rejection of Russellianism 211–12 on reference-fixing 211–12 theory of descriptions 45–6, 185–6, 195, 211–12 Russell-Myhill paradox 7–8 Salmón, N. 189, 222–3 see also schmidentity strategy (in Kripke’s writings) “Being of Two Minds: Belief with Doubt” 176 Frege’s Puzzle 62, 176 referentialist approach of 178 schmidentity strategy, use of 186–91 satisfaction conditions, theory of 49 Schiffer, S. 114 Schiller, H. I. 12–13 schmidentity strategy (in Kripke’s writings) 176, 182–203 see also Kripke, S. attitude ascriptions 186, 188, 190, 198, 202–3 “Cicero is Tully” example 182–3, 189–90 defense of Millianism 188, 192

260  schmidentity strategy (in Kripke’s writings) (cont.) extension of 190–3 NO EVIDENCE 184, 186, 189–93, 202 objection to semantic hypothesis 183 origin of name in Naming and Necessity 182–3 possible worlds semantics 190 PREDICTION 184–6, 189–90, 192–3, 202 and Salmón’s use of strategy 186–91 stipulated to be true 184, 193–203 efficacious vs. inefficacious stipulations 193–5 ‘is schmidentical’ expressing identity relation 193–203 modal stability 195–201 names and descriptions 197–201 vs. ordinary linguistic stipulations 195 semantic contents 195 semantic properties, modally stable 195–7 stipulations about worlds 201–3 within worlds 193–201 scoreboards 4–30 alteration of contents 5, 23 bridge 20 Cognitivist 9–14, 18–19 problems for 13–19 combinations of psychological and epistemic facts 9–10 common ground 3, 5n.7, 9–10, 16, 27–8 components 3–4, 18 backgrounded 4, 27–30 discourse referent 8n.12 foregrounded 4, 27–30 conversational 1–2, 5 scores 8–9 to-do list 3, 29–30 downstream and upstream connections 5 epistemic access 24–5

epistemology 12, 22 expansion and contraction of 5–7 informational 22 Lewisian 2–3 as mental states 1–2, 8–10 metaphysics 7–13, 25–6 mind-independent conception of 18–21 nested hierarchies of 4 non-Cognitivist 24–6 grades of non-Cognitivism 19–24 viability of 30 normative 21–2 ontology 19, 22 perceived as abstract mathematical entities 8–9, 20–2 propositions on 3–7, 12, 17–18 quantifiers 4, 17–18 question under discussion (QUD) stack 3–4, 15, 21 recording of scores on 11–12, 20 representational/informational conversational 1–2 ‘screen within a screen’ technology 4 shiftability 17–18 sport 2–3, 5, 19–20 technology 2–7 temporary subordinate 4 theories on nature of 7–8 updating 2, 5–7, 19 using without scorekeepers 2, 24–30 Searle, J. R. 49, 52–3, 70–1, 84–5 semantics analytic 205 compositional theory 137–8 descriptions 210–11 descriptivism 211 facts 107–9, 195, 197 hypotheses 121–2, 183 modally stable semantic properties 195–7 non-semantic content 240–1 possible worlds view 190, 192, 195 procedural 5–7

 semantic relationism 189 sentences, semantic contents 190, 196 singular and descriptive thought 35–6 theories 114, 194–5 translational 114–20 semisertion 80–1, 95 sentences associate 5–7 cognitive content differences between, stating the same 205–6, 208 ‘eternal’ 101–2 expositives 85 expressing different propositions 218 Icelandic 112–13 indicative 71–2 locutionary and illocutionary 85, 87–8 Paderewski-sentences 224–5 presuppositional 26–7 proper name, containing 36 semantic contents 190, 196 token 105 type sentences 105–7 Sider, T. 150–1 signals assertions as 74 breakdown of system 69 defining signaling 68 Duchenne smile 69–70 and indices 67–70 non-human species 68–70 verbal see verbal signals singularism 33–4 see also descriptivism; traditional picture anchored character attribution and representation 53–8, 61–2 anchoring role 39–45 descriptive/singular thought distinction 33–9, 59–60, 60n.41 agreements and disagreements 60–2 mind, philosophy of 34, 36–9

261

ordinary 37 Revised Massive Reduplication Argument 43–8 conclusion of 45–8 singularity of certain kind in thought 47 singular propositions 34–6, 46–8, 60 Soames, S. 222–3, 232–3, 240 speech acts assertions 74 clauses embedded within 82–3 cultural evolution 70, 76–81, 95 expositives 70, 82–9 genealogy of 76–81 grammatical evidentials (GEs) 91–2 indirect speech acts (ISAs) 83–4 intentionally performed 21n.32 modern, delineation of precursors to 91–2 semisertion 80–1, 95 and speaker meaning 70n.3 speech acts 70 ursertions 80–1, 95 verbs 82–3 Vervish (imagined communication system), among vervet monkeys 78–80 sport scoreboards 5, 19–20 technology 2–3 Stalnaker, R. C. 2–3, 9–10, 14, 17n.23, 90, 219, 226, 238–9 and diagonalization strategy 215–16, 219–20, 225–6 and pragmatic repair 216–17, 220 Stanley, J. 219 stipulations see schmidentity strategy Strawson, P. F. 40–4, 67, 70n.3, 73, 103, 106–7, 153–4 Individuals 39 Strawson Challenge 41–3 Strawson Response 41–3 on a typical use of ‘the soandso’ 73 subject-matter 231–4

262  syntax 78–9 number words syntactically displaced 137–9 recursive 79 that-clauses 103–4, 150, 192 denotations of 190, 197 theories disquotational 99, 101–2 dynamic 19 false/misguided 168–9 intentionality 44–5 metaphysical 130–1 modelling dynamically 5–7 philosophical 135–6 of propositions 5–8 scoreboards 7–8 semantic 114, 194–5 static 5–7 of truth 99n.1, 101–2 Thomason conditionals 17–18 thought see also mental states and arguments 56–9 contents 34, 36–7 descriptive/singular thought distinction see descriptivism; singularism direct and indirect 33–5, 39 linguistic and mental content 37–8 object-dependent 61–2 referential dependency patterns 52–3, 58–9 referentially anchored 45–6 regarding ordinary external objects see external objects (ordinary), thoughts about structured 56–7 traditional picture see also descriptivism; singularism and anchored descriptivism 57–60 defining 34 new 35–9, 45–6 denying, as descriptivism 38–40, 48 linguistic challenges to 38

over-attribution objection 48 pressure on 37–8 old 35, 38–9, 48, 54 transcendental precipitate of language 131–2 translation “Church–Langford translation test” 206–7 context-dependence 107–14 and disquotation 104–7 and ‘making sense’/meaning 120–3 in philosophers’ sense 120 and semantics 114–20 standards 105 at the United Nations 120 trigger problems 219–21 truth see also disquotationalism analytic 141–2 conceptual 141–2 contingent 211 disquotational notion of 104 extended disquotational notion of 105, 108 generalizing role of 102–4 and propositions 103–4, 104n.9, 220 role of 102–4 sentential 104 and strange entities 103–4 theories of 99–102, 99n.1 truth-predicate 102, 104–6 truth-values 190–3, 197, 235 ubiquitous propositionalizing 16–18 ursertions 80–1, 95 use-mention confusions 114–15 Van Elswyck, P. 90 verbal indexing 72–4 defining 72 factive 72 notion of a verbal index 73–4 and performatives 73–4 sufficient condition for 72

 verbal signals applications 82–95 assertive family and illocutionary commitment 74–6 conversational implicature 70, 93–5 cultural evolution 76–81 defining 71–2 expositives 70, 82–9 grammatical evidentials 70, 89–93 speech acts 70 sufficient condition for 72 verbs change-of-state 26 speech act 82–3 Vervish (imagined communication system), among vervet monkeys 78–80 Wiegmann, A. 162–3 Williamson, T. 220

263

words 50–1, 73, 187 ambiguous 85 associated properties 51 claims about 233–4 context function 72, 87–8 encoding of information 72, 87–8 English 206–7 learning 106–7 meanings 70n.3, 112–14, 123 Mentalese see Mentalese (language of thought) as morphemes 115 number words 136–45 and people 116 presuppositions 226 and proper names 100 in sentences 115 and things 119 translation 111–13, 115 word-mediated scenarios 51