The Routledge Handbook of Linguistic Reference [1 ed.] 9780367630249, 9780367629724, 9781003111894

Citation preview

THE ROUTLEDGE HANDBOOK OF LINGUISTIC REFERENCE

This Handbook offers students and more advanced readers a valuable resource for understanding linguistic reference; the relation between an expression (word, phrase, sentence) and what that expression is about.The volume’s forty-​one original chapters, written by many of today’s leading philosophers of language, are organized into ten parts: I II III IV V VI VII VIII IX X

Early Descriptive Theories Causal Theories of Reference Causal Theories and Cognitive Significance Alternate Theories Two-​Dimensional Semantics Natural Kind Terms and Rigidity The Empty Case Singular (De Re) Thoughts Indexicals Epistemology of Reference

Contributions consider what kinds of expressions actually refer (names, general terms, indexicals, empty terms, sentences), what referring expressions refer to, what makes an expression refer to whatever it does, connections between meaning and reference, and how we know facts about reference. Many contributions also develop connections between linguistic reference and issues in metaphysics, epistemology, philosophy of mind, and philosophy of science. Stephen Biggs is Associate Professor of Philosophy at Iowa State University. He researches and teaches in philosophy of mind and language, epistemology, and cognitive science. Heimir Geirsson is Professor and Chair of the Department of Philosophy and Religious Studies at Iowa State University. He works primarily in philosophy of language, philosophy of mind, epistemology, and metaethics, and is the author of Philosophy of Language and Webs of Information (2013).

ROUTLEDGE HANDBOOKS IN PHILOSOPHY

Routledge Handbooks in Philosophy are state-​of-​the-​art surveys of emerging, newly refreshed, and important fields in philosophy, providing accessible yet thorough assessments of key problems, themes, thinkers, and recent developments in research. All chapters for each volume are specially commissioned, and written by leading scholars in the field. Carefully edited and organized, Routledge Handbooks in Philosophy provide indispensable reference tools for students and researchers seeking a comprehensive overview of new and exciting topics in philosophy.They are also valuable teaching resources as accompaniments to textbooks, anthologies, and research-​orientated publications. Also available: The Routledge Handbook of Phenomenology of Agency Edited by Christopher Erhard and Tobias Keiling The Routledge Handbook of Feminist Philosophy of Science Edited by Sharon Crasnow and Kristen Intemann The Routledge Handbook of Linguistic Reference Edited by Stephen Biggs and Heimir Geirsson The Routledge Handbook of Dehumanization Edited by Maria Kronfeldner The Routledge Handbook of Anarchy and Anarchist Thought Edited by Gary Chartier and Chad Van Schoelandt The Routledge Handbook of the Philosophy of Engineering Edited by Diane P. Michelfelder and Neelke Doorn For more information about this series, please visit: www.routledge.com/​Routledge-​Handbooks-​in-​Philosophy/​book-​series/​RHP

THE ROUTLEDGE HANDBOOK OF LINGUISTIC REFERENCE

Edited by Stephen Biggs and Heimir Geirsson

First published 2021 by Routledge 52 Vanderbilt Avenue, New York, NY 10017 and by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN Routledge is an imprint of the Taylor & Francis Group, an informa business © 2021 Taylor & Francis The right of Stephen Biggs and Heimir Geirsson to be identified as the authors of the editorial material, and of the authors for their individual chapters, has been asserted in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging-​in-​Publication Data A catalog record for this title has been requested ISBN: 978-​0-​367-​62972-​4  (hbk) ISBN: 978-​1-​003-​11189-​4  (ebk) Typeset in Bembo by Newgen Publishing UK

CONTENTS

Notes on Contributors

ix

Introduction Stephen Biggs and Heimir Geirsson

1

PART I

Early Descriptive Theories

15

1 The Concept of Linguistic Reference before Frege Michael Losonsky

17

2 Frege on Reference Rachel Boddy and Robert May

30

3 Fregean Descriptivism Ian H. Dunbar and Stephen K. McLeod

41

4 The Referential-​Attributive Distinction Anne Bezuidenhout

53

PART II

Causal Theories of Reference

71

5 The Case(s) against Descriptivism Michael Nelson

73

6 Fruits of the Causal Theory of Reference Scott Soames

82

v

Contents

7 The Problem of Reference Change Harold Noonan

94

PART III

Causal Theories and Cognitive Significance

105

8 Cognitive Significance Aidan Gray

107

9 Conversational Implicature in Belief Reports Jonathan Berg

123

10 Context Sensitivity and ‘Believes’ Mark Richard

132

11 A Return to Simple Sentences David Pitt

145

12 Eliciting and Conveying Information Heimir Geirsson

153

PART IV

Alternate Theories

167

13 Causal Descriptivism Olga Poller

169

14 Reference-​Fixing and Presuppositions Manuel García-​Carpintero

179

15 Names as Predicates Sarah Sawyer

198

16 Variabilism Anders J. Schoubye

212

PART V

Two-​Dimensional Semantics

225

17 Two-​Dimensional Semantics Frank Jackson

227

18 Two-​Dimensional Semantics and Identity Statements Kai-​Yee  Wong

237

vi

Contents

19 Two-​Dimensionalism and the Foundation of Linguistic Analysis Andrew Melnyk

257

20 A Puzzle about Assertion Gregory Bochner

268

PART VI

Natural Kind Terms and Rigidity

281

21 Rigidity of General Terms Genoveva Martí and José Martínez-​Fernández

283

22 The Psychology of Natural Kind Terms Emily Foster-​Hanson and Marjorie Rhodes

295

23 Pervasive Externalism Stephen Biggs and Ranpal Dosanjh

309

24 Theoretical Identities as Necessary and a priori Joseph LaPorte

324

25 The Need for Descriptivism Sören Häggqvist and Åsa Wikforss

335

26 The Accommodation Theory of Reference Richard Boyd

345

27 Science, the Vernacular and the ‘Qua’ Problem Robin Findlay Hendry

359

PART VII

The Empty Case

371

28 Mill and the Missing Referents David Braun

373

29 Fregean Theories of Names from Fiction Ben Caplan

384

PART VIII

Singular (De Re) Thoughts

397

30 Reference and Singular Thought François Recanati

399

vii

Contents

31 Singular Thoughts, Sentences and Propositions of That Which Does Not Exist Jody Azzouni

409

32 Names and Singular Thought Rachel Goodman

421

PART IX

Indexicals

437

33 How Demonstratives and Indexicals Really Work Stephen Neale and Stephen Schiffer

439

34 Demonstrative Reference to the Unreal: The Case of Hallucinations Marga Reimer

449

35 What is Special about De Se Attitudes? Stephan Torre and Clas Weber

464

36 De Se Attitudes and Action Dilip Ninan

482

37 Acting without Me: Corporate Agency and the First Person Perspective Herman Cappelen and Josh Dever

499

38 Semantic Monsters Brian Rabern

515

PART X

Epistemology of Reference

533

39 Cross-​Cultural Semantics at 15 Edouard Machery

535

40 Reference and Intuitions Daniel Cohnitz and Jussi Haukioja

551

41 The Myth of Quick and Easy Intuitions John Bengson

560

Index

577

viii

NOTES ON CONTRIBUTORS

Jody Azzouni is Professor of Philosophy at Tufts University, USA. He is the author of Talking about Nothing: Numbers, Hallucinations, and Fictions (2010) and Semantic Perception: How the Illusion of a Common Language Arises and Persists (2013). John Bengson is Associate Professor of Philosophy at the University of Wisconsin-​ Madison, USA. He has published extensively on intuition, perception, understanding, and skill. He is co-​ editor of Knowing How: Essays on Knowledge, Mind, and Action (2012) and co-​author of Philosophical Methodology: From Data to Theory (forthcoming). Jonathan Berg is Professor in the Philosophy Department of the University of Haifa, Haifa, Israel. His books include Direct Belief: An Essay on the Semantics, Pragmatics, and Metaphysics of Belief (2012) and Naming, Necessity, and More: Explorations in the Philosophical Work of Saul Kripke, Ed. (2014). His MOOC “Critical Thinking: Fundamentals of Good Reasoning” opened on edX in 2018. Anne Bezuidenhout is Professor of Philosophy and Linguistics at the University of South Carolina, USA. Her work is focused on pragmatic phenomena, such as presuppositions and conversational implicatures, which she studies using both formal and experimental methods. Her experimental work has looked at scalar implicatures and at the processing of parentheticals and non-​restrictive relative clauses. Her formal work is focused on perspective-​taking in conversation and the strategies interlocutors use to foreground and background information during conversational exchanges. Stephen Biggs is Associate Professor of Philosophy at Iowa State University, USA. He researches and teaches in philosophy of mind and language, epistemology, and cognitive science. Gregory Bochner is a Marie Skłodowska-​Curie Fellow at the Institut Jean Nicod, Paris, France (Département d’études cognitives, ENS, EHESS, CNRS, PSL University, UMR 8129). He has published on topics in the philosophy of language and mind, and especially on theories of reference and internalism-​externalism controversies about content. Rachel Boddy is a lecturer for the Department of Philosophy and Religious Studies at Utrecht University, Netherlands. Her main research interests are in the history and philosophy of logic, especially Gottlob Frege’s work. ix

Notes on Contributors

Richard Boyd is Susan Linn Sage Professor of Philosophy emeritus at Cornell University and Adjunct Professor of Philosophy at Case-​Western Reserve University, USA. He previously taught at the University of Michigan, the University of California Berkeley, Harvard University, MIT, Claremont McKenna College, and Lewis and Clark College. He has published widely on issues in the philosophy of science, philosophy of biology and metaethics, as well as on issues about natural kinds and natural kind terms. David Braun is a Professor of Philosophy, and the Patrick and Edna J. Romanell Chair in Philosophy, at the University at Buffalo, USA. He has published primarily on topics in philosophy of language, especially on empty names, indexicals and demonstratives, attitude ascriptions, modal expressions, and questions. Ben Caplan is Professor of Philosophy at the University of Kansas, USA. He has published mainly in philosophy of language and metaphysics, including the ontology of art. Herman Cappelen is Professor of Philosophy at the University of Hong Kong. He is the author or co-​author of six monographs, several textbooks and edited collections, and many articles, mostly on philosophy of language and metaphilosophy. His most recent monograph is Fixing Language (2018). Daniel Cohnitz holds the Chair for Theoretical Philosophy at Utrecht University, Netherlands. His research interests include metaphilosophy, epistemology, metaphysics, the philosophy of mind, language, mathematics, logic, and related areas of philosophy and cognitive science. His most recent book is An Introduction to the Philosophy of Logic (2019) that he co-​authored with Luis Estrada-​González. Josh Dever is Professor of Philosophy at the University of Texas at Austin, USA. He works in philosophy of language and philosophical logic, and is co-​author (with Herman Cappelen) of The Inessential Indexical (2013), Context and Content (2017), Puzzles of Reference (2018), and Bad Language (2019). Ranpal Dosanjh is Assistant Professor of Philosophy at Iowa State University, USA. He works primarily in metaphysics and philosophy of mind. Ian H. Dunbar is a PhD student in the philosophy department of the University of Liverpool, UK. Following his retirement in 2014, he undertook an MA course in philosophy at Liverpool; graduating in 2017. His MA dissertation was entitled Fregean Reference and Ontology. His PhD research follows on from that, with a working title Making Sense of Fregean Sense. Emily Foster-​Hanson is Postdoctoral Research Associate at Princeton University, USA. Her research focusing on the development of concepts has been published in Cognitive Psychology, Child Development, and Cognitive Science. Her doctoral work was funded by a predoctoral fellowship from the National Institutes of Child Health and Human Development. Manuel García-​Carpintero is University Professor at the University of Barcelona, Spain. He works in the philosophy of language and mind, and related epistemological and metaphysical issues. Currently he is completing a book on the nature of speech acts in general, and assertion in particular, entitled Tell Me What You Know. Heimir Geirsson is Professor and Chair of the Department of Philosophy and Religious Studies at Iowa State University, USA. He works primarily in philosophy of language, mind, epistemology, and metaethics, is the co-​editor of several collections, and is the author of Philosophy of Language and Webs of Information (2013). x

Notes on Contributors

Rachel Goodman is Assistant Professor of Philosophy at the University of Illinois at Chicago, USA. She works in the philosophy of mind and language, with a special focus on the theory of representation and mental reference. She is co-​editor of Singular Thought and Mental Files (2020). Aidan Gray is Associate Professor of Philosophy at the University of Illinois, Chicago, USA. His research focuses on reference, proper names, and issues surrounding Frege’s Puzzle. Sören Häggqvist is Associate Professor in the Department of Philosophy at Stockholm University, Sweden. He works in philosophy of science, philosophy of language, and philosophical methodology; in particular on thought experiments, intuitions, and natural kinds. Jussi Haukioja is Professor of Philosophy at the Norwegian University of Science and Technology in Trondheim. He specializes in philosophy of language, focusing especially on the theory of reference and its consequences for issues in metaphysics, epistemology, and philosophy of mind. Robin Findlay Hendry is Professor of Philosophy at Durham University, UK. He works on the history and philosophy of chemistry and its relationship to physics, and how this bears on general issues in metaphysics and the philosophy of science including natural kinds, emergence, and reduction. He has published many articles on these topics. With Paul Needham and Andrea Woody he edited Philosophy of Chemistry (2012), and with Sophie Gibb and Tom Lancaster The Routledge Handbook of Emergence (2019). Frank Jackson is Emeritus Professor at the Australian National University. He works in the philosophy of mind and language, and in ethics. He has held a number of visiting positions, most recently at the National University of Singapore. His books include From Metaphysics to Ethics (1998) and Language, Names, and Information (2010). Joseph LaPorte is Professor of Philosophy at Hope College in Holland, Michigan, USA. He has worked in the philosophy of language as well as the philosophy of science and metaphysics. He is the author of Natural Kinds and Conceptual Change (2004) and Rigid Designation and Theoretical Identities (2013). Michael Losonsky is Professor Emeritus of Philosophy at Colorado State University in Fort Collins, Colorado, USA. His publications include Enlightenment and Action from Descartes to Kant: Passionate Thought (2001) and Linguistic Turns in Modern Philosophy (2006). He is currently working on the reception of Leibniz’s philosophy of language. Edouard Machery is Distinguished Professor in the Department of History and Philosophy of Science at the University of Pittsburgh, USA, and the Director of the Center for Philosophy of Science at the University of Pittsburgh. He is the author of Doing without Concepts (2009) and of Philosophy within Its Proper Bounds (2017). Genoveva Martí is ICREA Research Professor at the University of Barcelona, Spain. In her research she focuses primarily on the discussion of issues in the theory of reference for singular and general terms, as well as on aspects of modality. José Martínez-​Fernández is Associate Professor at the University of Barcelona, Spain. His research interests are in logic and the philosophy of language. In particular, he has published papers on semantic paradoxes, theories of truth, many-​valued logics, and rigidity of general terms.

xi

Notes on Contributors

Robert May is Distinguished Professor of Philosophy and Linguistics at the University of California, Davis, USA. His research has been in theoretical linguistics, philosophy of language and logic, and the history of early analytic philosophy, with focus on the work of Frege. His books include Logical Form: Its Structure and Derivation (1985), Anaphora and Identity (1994), and De Lingua Belief (2006), the latter two co-​authored with Robert Fiengo. Stephen K. McLeod is Senior Lecturer in Philosophy at the University of Liverpool, UK. He is the author of Modality and Anti-​Metaphysics (2001, repr. 2017) and has published journal articles in various philosophical subdisciplines. Andrew Melnyk is Professor of Philosophy at the University of Missouri, USA. His work lies at the intersection of metaphysics, philosophy of mind, and philosophy of science, and is mostly centered on articulating and defending a physicalist view of the world. He is the author of A Physicalist Manifesto: Thoroughly Modern Materialism (2003). Stephen Neale is Distinguished Professor of Philosophy and Linguistics and holder of the Kornblith Chair in the Philosophy of Science and Value at CUNY Graduate Center, USA. He has written extensively about meaning, information, interpretation, and communication, and more generally about issues at the intersection of philosophy and linguistics. His books include Facing Facts (2001) and Descriptions (1990). Michael Nelson is Associate Professor in the Philosophy Department at the University of California at Riverside, USA. He was previously a visiting professor at the University of Arizona during 2001–​ 2002, and an assistant professor at Yale University from 2003 to 2005. He works on philosophy of language, metaphysics, and agency. Dilip Ninan is Associate Professor of Philosophy at Tufts University, USA. His research interests include the philosophy of language, epistemology, and the philosophy of mind. He has published articles on de se and de re attitudes, epistemic modals, and on the nature of semantic content. He currently serves as an Editor-​in-​Chief of the journal, Linguistics and Philosophy. Harold Noonan is Professor of Mind and Cognition at the University of Nottingham, UK. He is the author of Frege (2001) and Kripke and Naming and Necessity (2013). David Pitt is Professor and Chair in the Philosophy Department at California State University, Los Angeles, USA. He works on philosophy of mind, philosophy of language, philosophy of linguistics, and metaphysics. He is currently finishing a book manuscript, The Quality of Thought. Olga Poller is Assistant Professor of Philosophy at Pedagogical University of Kraków, Poland. Before coming to Pedagogical University in 2018, she was a Research Assistant at Jagiellonian University. Her research focuses primarily on semantics of proper names, descriptivism, and event-​free semantics of adjuncts. Brian Rabern is Reader in the School of Philosophy, Psychology, and Language Sciences at the University of Edinburgh, UK. His research focuses on issues in philosophy of language, philosophical logic, and formal semantics. He is co-​editor of The Science of Meaning (2018) and an associate editor at the journal, Linguistics and Philosophy. François Recanati is Professor at Collège de France, Paris. He has published numerous books and articles in the philosophy of language and mind, including Direct Reference (1993), Literal Meaning xii

Notes on Contributors

(2004), Perspectival Thought (2007), Mental Files (2012), and Mental Files in Flux (2016). He is a member of Academia Europaea and a foreign honorary member of the American Academy of Arts and Sciences. Marga Reimer is Professor of Philosophy at the University of Arizona in Tucson, USA. She works primarily in the philosophy of language and has published on definite descriptions, demonstratives, proper names, singular thought, rhetorical questions, metaphor, and malapropism. She also works in the philosophy of psychiatry and has published on psychopathy, delusions, and hallucinations. Recently, she has written on issues at the intersection of philosophy of language and philosophy of psychiatry, including reference to hallucinations. Marjorie Rhodes is Professor of Psychology at New York University, USA. Her research focuses on conceptual development, and has been published by the Proceedings of the National Academy of Sciences, Psychological Science, and Cognitive Psychology, among other outlets. Her scientific work is supported by the National Institutes of Child Health and Human Development and the National Science Foundation. Mark Richard is Professor of Philosophy at Harvard University, USA. He is the author of Meanings as Species (2019), Meaning in Context, Volumes I and II (2012, 2015), When Truth Gives Out (2008) and Propositional Attitudes (1990). It is unclear which is worse, his guitar playing or his singing, though it is clear that neither is something to write home about. Sarah Sawyer is Professor of Philosophy at the University of Sussex, UK. Her research includes publications on anti-​individualism, singular thought, proper names, fictional terms and fictional objects, perceptual warrant, skepticism, moral judgements and normative reasons, concepts, conceptions, and conceptual engineering. She also edited New Waves in Philosophy of Language (2009). Stephen Schiffer is Silver Professor of Philosophy at New York University, USA. He is the author of three books—​Meaning (1972), Remnants of Meaning (1987), and The Things We Mean (2003)—​and of numerous articles. His hobby is giving seminars with Stephen Neale. Anders J.  Schoubye is Associate Professor of Philosophy at Stockholm University, Sweden. His main areas of research are philosophy of language, semantics, and logic. He has published papers on a variety of topics such as the semantics of proper names and definite descriptions, the relation between literal meaning and questions, truth value intuitions and presuppositions, and the semantics of future contingents. He’s currently writing a monograph on proper names. Scott Soames is Distinguished Professor and Chair of the Philosophy Department at the University of Southern California, USA. He works on the philosophy of language, the history of analytic philosophy, and the philosophy of law. His recent books include The World Philosophy Made (2019) and Rethinking Language, Mind, and Meaning (2015). Stephan Torre is a Senior Lecturer in Philosophy at the University of Aberdeen, UK. His main research interests are in philosophy of time, de se thought, persistence, and modality. With Manuel García-​Carpintero, he edited the collected volume About Oneself (2016). Clas Weber is Lecturer of Philosophy at the University of Western Australia, Perth, Australia. His research focuses on de se content in thought and language and on foundational issues in semantics. Åsa Wikforss is Professor of Theoretical Philosophy at Stockholm University, Sweden. She has published widely in the intersection of philosophy of mind and language, including papers on the xiii

newgenprepdf

Notes on Contributors

normativity of meaning, semantic externalism, concept understanding, and natural kind terms. She is co-​editor, with Teresa Marques, of a forthcoming book on conceptual variation, Shifting Concepts. Kai-​Yee Wong is Associate Professor of Philosophy at the Chinese University of Hong Kong. He specializes in philosophy of language and metaphysics. He has also published articles on philosophical logic, epistemology, applied ethics, and Chinese philosophy.

xiv

INTRODUCTION Stephen Biggs and Heimir Geirsson

This volume explores linguistic reference. Reference is a two-​place relation. One relata refers, the other (the referent) is referred to. Candidate referring items include, among others, words, ideas, and maps. Candidate referents include, among others, objects, properties, and states of affairs. Linguistic reference is the reference relation in which the referring item is either an expression or a use of an expression—​for ease of presentation, we’ll suppress this distinction when possible in this introduction. Consider two examples. The name “Bob Dylan” and the sentence “Bob Dylan wrote Masters of War” both refer. Candidate referents for the name include, among others, an idea of Bob Dylan, a property that Bob Dylan instantiates, and the individual, Bob Dylan, himself. Candidate referents for the sentence include, among others, the thought that Bob Dylan wrote Masters of War, the state of affairs that consists of Bob Dylan instantiating the property having written Masters of War, and the truth-​value  True. Throughout much of the 20th century, explorations of linguistic reference were central to practicing philosophy. In nearly every area of philosophy, familiar discussions were recast as debates about reference. Debating whether or not numbers, moral properties, and mind-​ independent objects exist became debating whether or not we can refer to them. Debating how we know about them (supposing we do) became debating how we manage to refer to them. And so on. Sometimes this recasting did not alter the discussion in any deep way. Arguments that we can refer to non-​mental objects, for example, often parallel arguments that we can know about them. Other times, however, the recasting was quite fruitful. Consider Anthony Appiah’s (1996) argument that races do not exist: he invokes standard theories of reference for kind terms; shows that race terms fail to refer no matter which theory one adopts; and concludes, on that basis, that there are no races. Although one might try to retain the spirit of his argument without mentioning reference, the result would be attenuated, at best. Perhaps more importantly, some appeals to linguistic reference did more than recast familiar discussions. Consider two examples. First, Ludwig Wittgenstein (1953/​2009) tried to deflate the mind-​body problem by showing that it only appears to be a problem because we mistakenly treat words for mental states (such as ‘pain’) as referring expressions—​our mistake, he thinks, results from our taking the surface grammar of various sentences (e.g., “I have a pain”) too seriously. Second, Hillary Putnam (1981) argued that influential concerns about the incommensurability of scientific theories (from, e.g., Kuhn 1962) arise only when one mistakenly presumes that natural kind terms

1

Stephen Biggs and Heimir Geirsson

in science have their reference fixed by the role they play in a theory (rather than by a causal relation between the user and the world)—​and so, they refer to whatever and only whatever instantiates the properties by which one ordinarily recognizes their samples (rather than to whatever and only whatever has the same “hidden structure” as familiar samples). In each case, a philosophical problem is purportedly solved by an appeal to reference, and the result cannot be achieved (in any similar way) without appealing to reference. Even when discussions that focus on reference can be explored in other terms, moreover, focusing on reference can be fruitful. Timothy Williamson suggests that “even when language plays a purely instrumental role in philosophy, like other scientists we must study how our instruments function and sometimes malfunction, lest our naive use of them lead us into gross errors” (2013, 402). More generally, he suggests that careful attention to the meaning of our expressions is a good strategy for philosophical theorizing because, among other benefits, the focus on language invites precision in expression and theorizing. Even those who largely avoid philosophy of language can (and, we think, should) endorse Williamson’s point. What led philosophers to focus so much attention on reference? There were, of course, many factors. One cause, however, stands out: the work of Gottlob Frege, especially his seminal essay “On Sense and Reference”. Frege did not write this essay with any hope of making philosophy of language into “first philosophy” (i.e., where philosophers take the best philosophical practice to begin and perhaps end). Indeed, he did not even aim to generate interest in reference. Rather, he wrote “On Sense and Reference” to help him solve problems he had encountered while pursuing his passion:  reducing arithmetic to logic. He treated his theorizing about reference as a mere means to that end. Despite Frege’s focus, however, others saw “On Sense and Reference” as a standalone work, full of useful tools for practicing philosophy, valuable philosophical theorizing, and fascinating puzzles. Consider the distinction from which the essay gets its name. Frege distinguishes two aspects of meaning, sense and reference. Sense is tied to cognitive significance (roughly, the role an expression plays in thinking); reference is tied to truth-​conditions and truth-​values. Appeals to sense promise to explain what each expression refers to, how each refers to what it does, and how we manage to communicate with one another despite often having very different ideas of referents. Appeals to reference promise to explain what makes sentences have the truth-​conditions that they do and to identify what those truth-​conditions are, thereby setting the stage for discovering their truth-​values. Accordingly, Frege’s account of meaning promises to answer central questions in epistemology, metaphysics, and philosophy of mind, all by theorizing about language. Frege framed this discussion as a response to various puzzles. Consider a version of one such puzzle. Bob Dylan and Robert Zimmerman are the same person. The sentences “Bob Dylan is Bob Dylan” and “Bob Dylan is Robert Zimmerman”, therefore, are made true by the same fact—​the fact that the relevant individual is identical with himself. One might say, therefore, that these sentences place the same constraint on the world, say the same thing, and have the same meaning. Nonetheless, they are importantly different: the latter is potentially informative in a way that the former isn’t, and correspondingly, someone can justifiably believe “Bob Dylan wrote Masters of War” while justifiably doubting “Robert Zimmerman wrote Masters of War”. One might say, therefore, that these sentences place different constraints on the world, say different things, and have different meanings. Since the sentences cannot have the same meaning and yet have different meanings, something has gone wrong. Frege’s solution is simple: there are two aspects of meaning. One aspect, reference, tracks the fact that makes both sentences true. The other aspect, sense, tracks their distinct cognitive significances. Puzzle solved. While philosophy of language is no longer widely considered to be “first philosophy”, Frege’s legacy still looms large: many philosophers still cast traditional philosophical debates as debates about reference, appeal to facts about reference to solve philosophical problems, or, at the very least, value the clarity of theorizing that comes with a special focus on language. Moreover, and unsurprisingly given 2

Introduction

the importance of reference in recent philosophical practice, many philosophers explore reference for the sake of better understanding reference (rather than exploring reference to some other end). This volume contributes to this last kind of exploration, while recognizing that a better understanding of reference often bears fruit in various domains. Philosophers aiming to better understand reference address many issues.These issues can be explored for any particular expression (e.g., “Bob Dylan”, “wrote Masters of War”, “Bob Dylan wrote Masters of War”), though they are most often explored for kinds of expressions (e.g., names, descriptions, declarative sentences), with particular expressions serving as illustrative.These explorations are guided by questions such as the following: • Is the expression at issue a referring expression—​i.e., does it purport to refer to anything? Perhaps, for example, “Bob Dylan” and “Pegasus” are both referring expressions while “is” and “of ” are not. • What does a given referring expression refer to? Does “Madonna”, for example, refer to a property, individual, or something else? • What makes an expression refer to whatever it does? Does “David Bowie”, for example, refer to whatever individual satisfies a description that is widely associated with David Bowie, whatever individual is appropriately connected to ordinary uses of “David Bowie”, or something else? • Is the fact that an expression refers at all a fact about its meaning or a fact about something else? Put another way, is the expression’s referent part of its meaning or not? Suppose, for example, that “Bob Dylan” refers to Bob Dylan. Does it follow that “Bob Dylan” means Bob Dylan? • Supposing that reference is part of an expression’s meaning, is it the only part? Does the meaning of “Bob Dylan”, for example, include only reference or also sense? • If a referring expression has no sense, how does it manage to refer to the world? And how can we solve the puzzles Frege posed? • Whatever the facts about reference may be, how do we know them? How do we know, for example, that “Robert Zimmerman” refers to whatever it does? This volume explores these and many related questions, with a special focus on Frege and his intellectual descendants. One descendant, Saul Kripke, looms especially large. For nearly a century following the publication of “On Sense and Reference”, most philosophers accepted a broadly Fregean picture of meaning and reference. To be sure, they proposed improvements. Some hoped to shift senses from the realm of the abstract to the realm of the concrete, others hoped to complexify the descriptions that correspond to senses, and still others hoped to offer a more formal semantics. Despite such modifications, however, the spirit of Frege’s picture remained largely intact. Then, in Naming and Necessity (1980), Kripke proposed a more fundamental shift. Rather than trying to improve Frege’s account, Kripke proposed eliminating senses altogether, at least for names and natural kind terms. More generally, he offered both an alternative picture of what various expressions refer to and an alternative picture of how such expressions manage to refer to what they do. Given the deep connections between reference and other areas of philosophy, the proposed shifts had significant ramifications, including, e.g., in metaphysics, epistemology, philosophy of mind, and moral philosophy. There were precursors to Kripke’s picture, including work from Peter Strawson (1959), Ruth Barcan Marcus (1961), Putnam (1962, 1975), and Keith Donnellan (1966), but it was Kripke’s work, not its precursors, that shattered the Fregean consensus. This dispute between broadly Fregean and broadly Kripkean approaches to reference animates much of the volume. Reflecting this focus, the volume favors depth over breadth. Consequently, some important topics related to linguistic reference, including indeterminacy (e.g., Quine 1968) and vagueness (Williamson 1994), receive relatively little attention. The volume includes forty-​one original essays divided into ten sections, covering, respectively, early descriptive theories of reference and their historical roots, causal theories of reference, Frege’s puzzles, alternate theories of names, 3

Stephen Biggs and Heimir Geirsson

two-​dimensional semantics, natural kind terms, empty names, singular (de re) thoughts, indexicals, and the epistemology of reference. We now turn to a brief discussion of each section.

Early Descriptive Theories The volume begins with a historical overview of theorizing about reference and meaning. In this first chapter, Michael Losonsky traces accounts of reference from Parmenides of Elea to John Stuart Mill. After considering the role of reference in ancient and early Medieval philosophy, Losonsky explores supposition theory (of the 13th and 14th centuries). While cataloging fallacies, supposition theorists noticed that a noun can change reference depending on what is being predicated. Explaining this shift in reference led them to recognize that reference plays a crucial role in determining meaning in natural language, which then led to a more general interest in how the reference of various nouns is determined.William of Ockham suggested that nouns signify via concepts, which provide satisfaction conditions, a precursor to Frege’s view that sense fixes reference.This precursor to Frege was followed by a precursor to Kripke, with Hobbes arguing that reference is constituted by a causal relationship between our mental states and the objects in our environment. After tracing accounts of reference through Locke, Descartes, Leibniz, and von Humboldt, among others, Losonsky concludes with a discussion of Mill’s approach. With this historical background in place, the next two chapters turn to Frege’s theory of reference. As noted, Frege’s “On Sense and Reference” is often treated as a standalone work. In their chapter, Rachel Boddy and Robert May caution against that treatment.They emphasize that Frege introduces his account of reference with an eye to explaining arithmetic.They show how being mindful of this point illuminates his account, clarifying both what he says (e.g., what he claims about names, sense, and reference) and why he says what he says (e.g., why he introduces senses and why he focuses on names). In the third chapter, Ian Dunbar and Stephen McLeod argue that Frege never advances the view of reference that Kripke (and nearly everyone else) attributes to him.That view, descriptivism, consists of several theses, most centrally holding that the sense of a name is given by a description D that determines its reference via a satisfaction requirement, where any competent user of the expression can articulate D. To be sure, they say, Frege thinks that sense determines reference, but he never says that the sense of a name (or natural kind term) is akin to an articulable description. This adjustment, they suggest, allows Frege to sidestep Kripke’s critiques. Dunbar and McLeod notwithstanding, early descriptivism suggests that each name refers to the individual who satisfies the description that gives its sense, which suggests that names are (disguised) definite descriptions.This result inspired philosophers, perhaps most famously Bertrand Russell (1905), to study definite descriptions themselves. The fourth chapter, from Anne Bezuidenhout, focuses on definite descriptions, specifically, on Keith Donnellan’s (1966) distinction between their referential and attributive uses. Suppose someone says, “The woman wearing the zoot suit is wise”. An attributive use of the subject phrase refers to whoever satisfies the relevant description, regardless of what the speaker intends. A referential use refers to the speaker’s intended target, regardless of who satisfies that description. Bezuidenhout argues that the significance of this distinction is foremost epistemic, not semantic. Specifically, she claims, whereas a speaker must be directly acquainted with the referent in order to use a description referentially, she can use it attributively without such acquaintance. Bezuidenhout further argues that an attributive use is felicitous only if it terminates with an information source that is directly acquainted with the referent. Contrary to a common understanding of descriptivism, then, attributive uses of descriptions depend on a successful prior referential use.

Causal Theories of Reference The next three chapters focus on Kripke’s view of names and its counterpoint, descriptivism. In the first chapter in this group, Michael Nelson articulates motivations for descriptivism 4

Introduction

before sketching two arguments against it, Peter Strawson’s (1959) reduplication argument and Kripke’s modal argument. Strawson imagines that a distant part of the universe includes qualitative duplicates of everything around us. He suggests that a thoroughgoing descriptivism implies, implausibly, that none of our thoughts and corresponding expressions refer determinately in this scenario. Kripke’s modal argument notes that the referent of a name often has many apparently reference-​fixing properties merely contingently (e.g., Bertrand Russell is merely contingently a logician), which shows that, contrary to descriptivism, names are not modally coextensive with descriptions. Nelson then assesses versions of descriptivism that aim to avoid these pitfalls, including causal-​descriptivism (described below). In the next chapter, Scott Soames distinguishes two kinds of descriptivism about names, recounts Kripke’s examples against each, and describes Kripke’s steps toward a positive theory; namely, the historical-​chain model of reference-​fixing and transmission, according to which the referent of a name is fixed by a chain of uses connecting each use to previous uses and, ultimately, to an “initial baptism”, with each speaker in the chain intending (at least when acquiring the name) to preserve the previous speaker’s reference. Soames notes that Kripke’s work did not deter dedicated descriptivists (including, perhaps most prominently, David Lewis (1999) and Frank Jackson (e.g., 1998, this volume)), who adjusted descriptivism rather than following Kripke. Soames argues against these updated versions of descriptivism, but he also concedes that our pretheoretic conception of meaning (including for both names and natural kind terms) includes elements of what we ordinarily call understanding as well as semantic content. In the next chapter, Harold Noonan discusses a problem for any Kripkean account of reference-​ fixing, namely reference change. Consider ‘Madagascar’, which was introduced as a name for a part of the African mainland but which came to name an island—​a change that occurred despite each user intending to preserve the meaning of the previous user. Noonan discusses Evans’s attempt to solve this problem by holding that a cluster of information may dominantly be of one item even though it contains information whose causal source is different. Noonan concludes that Evans’s proposal faces unmet challenges, and more generally, that opponents of descriptivism have no obvious way to handle reference change.

Causal Theories and Cognitive Significance The next five chapters explore Frege’s puzzles about identity and belief. Recall the puzzle about identity: “Bob Dylan is Bob Dylan” and “Bob Dylan is Robert Zimmerman” have different cognitive significance even though each is (apparently) made true by the same fact. More generally, sentences containing co-​referring names flanking an identity sign are trivial when the names are the same but informative when they are different. In the first chapter in this group, Aidan Gray notes that this puzzle is a founding problem in analytic philosophy and cuts across issues related to reference, propositional attitudes, the relation between semantics and pragmatics, among others, as well as issues related to semantic theorizing more generally. Gray presents the puzzle, explores relevant issues, and offers a taxonomy of solutions. He ultimately concludes that no single approach is likely to explain the full range of issues that the puzzle raises. In the next chapter, Jonathan Berg focuses on Frege’s puzzle about belief. The puzzle emerges in the following pair of sentences: (1) Lois believes that Clark Kent is a reporter. (2) Lois believes that Superman is a reporter. Since Clark Kent is identical with Superman, the content of the belief clause in (1) seemingly has the same truth-​conditions as the content of the belief clause in (2). Nonetheless, we are inclined to accept (1) but not (2). Berg argues that this difference in acceptability results from pragmatic content, 5

Stephen Biggs and Heimir Geirsson

not semantic ambiguity. These sentences, then, can have the same truth-​value, despite our temptation to treat them differently. In the next chapter, Mark Richard focuses on the context sensitivity of ‘believes’. Suppose that Pierre uses both ‘London’ and ‘Londres’ to refer to London, but never realizes that the names refer to the same city. When waiting for a flight to London, a fellow passenger, Barry, asks him about the gate number for the flight to London. Pierre responds that he thinks it is Gate 43. When Pierre is asked the same question in French, he replies, in French, that it is not Gate 43. So, Pierre seems to believe both that the flight to London is at Gate 43 and that it is not at Gate 43. Richard’s explanation of how Pierre can hold both beliefs involves pairing each constituent in a proposition with the constituent of the representation that determines it: if the relevant proposition is London is pretty, then it can be represented as ,with Pierre’s representational system containing both , and . After these two entries focusing on belief, David Pitt suggests that the problems that arise for sentences expressing belief, such as (1) and (2), can arise even in simple sentences—​i.e., sentences that are not embedded in attitude contexts. Consider an example: (3) Clark Kent went into the phone booth, and Superman came out (4) Clark Kent went into the phone booth, and Clark Kent came out. In order to explain the intuition that (3) can be true while (4) is false, Pitt argues that ‘Clark Kent’ and ‘Superman’ are not co-​referential. Instead, each refers to a distinct collection of time slices of the underlying four-​dimensional individual. Thus, Superman is the time slices of Kal El (Superman’s given name on Krypton) during which he is inhabiting the Superman persona, and Clark is the time slices of Kal El during which he is inhabiting the Clark persona. In the final chapter in this section, Heimir Geirsson argues both that various expressions, not just names, can generate Frege’s puzzles, and that putative solutions need to be compatible with non-​linguistic versions of it. Accordingly, solutions that focus on the meaning of names or the pragmatics of language are, at best, incomplete. More generally, Geirsson maintains, the puzzle is about how we organize, access, and elicit information. He suggests that one organizes information about what one takes to be an individual or an object in bundles or files. The relevant information, he says, can include predicates, descriptions, mental pictures or images, memory episodes, memories of scent, etc. A name can direct one to the appropriate file—​but so can scents and images. Geirsson outlines the conditions under which one files information in different files and under which even those who are in the know about the relevant identities must use the appropriate name.

Alternate Theories The next four chapters offer competing accounts of names—​accounts that sometimes extend to other expressions too. In the first chapter in this group, Olga Poller develops her version of causal-​descriptivism. As per descriptivism, causal-​descriptivism holds that any competent user of a name N associates N with a reference-​fixing description. In the spirit of causal accounts, it holds that that description implies that the referent of N is the causal source of that use of N. Poller replies to possible objections to her view, considers an alternative version of causal-​ descriptivism, and notes outstanding concerns—​e.g., it remains unclear what names contribute to corresponding propositions. In the next chapter, Manuel García-​Carpintero first notes that Kripke’s distinction between a description fixing the reference of a name and giving its meaning is often used to consign descriptive 6

Introduction

material to a metasemantic role—​i.e., to say descriptions plays a role in fixing reference but not giving meaning. García-​Carpintero counters that descriptive material is, in fact, often part of the semantics of names. He advances this view by developing an Austinian account of the semantic-​metasemantic distinction, elaborating on how the presuppositions at issue in speech acts play a reference-​fixing role for names and indexicals, and suggesting that these speech acts contribute to the meanings to which speakers intuitively commit, which implies that we cannot relegate descriptive material to metasemantics.This, he notes, undermines full-​fledged anti-​descriptivism, and in particular Millianism (advocated in this volume by, e.g., David Braun and Geirsson). In the next chapter, Sarah Sawyer presents a predicativist account of names. While names usually occur in unmodified contexts (e.g., Sophie works at the local cinema), they also occur in modified forms, taking, e.g., the definite and indefinite article, as well as occurring in both plural forms (e.g., Several Sophies work at the local cinema) and quantified forms (e.g., A Sophie joined my dance class this week). The main questions for predicativism is whether, and if so how, it can account for ordinary singular referential uses. Sawyer suggests that names are predicates in all occurrences. Singular referential uses include an implicit determiner phrase (‘the’ or ‘this’). Accordingly, “Sophie is tall” can be understood as “that Sophie is tall” or “the Sophie is tall”. Anders J.  Schoubye, in the final chapter in the group, argues for variabilism, which holds that names should be analyzed as variables. Schoubye argues that variabilism is superior to Millianism, despite their similarities, because it can explain additional data—​accounting for predicative, bound, and shifted uses of names. Bound uses occur when names function as bound variables, as in “If a child is christened ‘Bambi’, Disney will sue Bambi’s parents”. As for shifted use, we might have reasons to believe now that it is true that Del Naja might be Banksy, even though we find out later that it is false that Del Naja is Banksy. But, if names are rigid, as Millians argue, and immune to world-​shifting operators, then identity sentences are either necessarily true or necessarily false, and so, the Millian has problems accounting for us thinking that “Del Naja might be Banksy” is true. Schoubye ends on the promissory note that it is quite likely that variabilism’s explanatory powers extend beyond the issues he discusses here.

Two-​Dimensional Semantics The next four chapters explore two-​dimensional semantics. In the first, Frank Jackson, lays out his seminal version of two-​dimensionalism, a generalization of two-​dimensional semantics holding that most expressions have two intensions: the primary (or A) intension is a function from centered worlds (a subject and world considered as actual) to extension (e.g., truth-​value for a declarative sentence), while the secondary (or B) intension is a function from worlds considered as counterfactual to extension. According to Jackson, any competent user of an expression can access its primary intension (i.e., she is disposed to correctly identify the expression’s extension at any world considered as actual), and, once she has done so, can access its secondary intension (i.e., she is disposed to correctly identify its extension at any world considered as counterfactual relative to the initial world considered as actual). This view promises to restore the intuitive (Fregean) idea that we can access the meanings of our own expressions while simultaneously accounting for the Kripkean concerns that drove so many away from descriptivism. The next two entries in this section express doubts about two-​dimensionalism. In the first, Kai-​Yee Wong outlines Kripke’s critique of descriptivism, explains how Kripke understands rigid designation and the necessary a posteriori, and reviews versions of two-​dimensionalism from Jackson and David Chalmers (e.g., 2006). He then suggests that these versions fail because their explanations of necessary a posteriori truths illicitly treat metasemantic facts as semantic facts.Andrew Melnyk continues this thread in the next chapter, arguing that Jackson’s two-​dimensionalism cannot explain our knowledge of necessary truths. Melnyk first sketches a traditional account of linguistic analysis that presupposes descriptivism about general terms, noting that this traditional account cannot survive the 7

Stephen Biggs and Heimir Geirsson

considerations that led many philosophers to accept semantic externalism about natural kind terms (which holds that the facts that determine the meaning of such terms can be inaccessible to competent speakers). He then sketches Jackson’s two-​dimensionalism, which attempts to adjust descriptivism in light of externalist impulses. Finally, Melnyk argues that Jackson’s two-​dimensionalism fails because (i) it fits best with a reliablism about knowledge according to which one knows proposition P only if a reliable belief-​forming process produces P, (ii) one knows P only if one has no defeater for the belief that P, but (iii) anyone who reflects on the epistemology of reference inherent in two-​ dimensionalism acquires a defeater for the belief-​forming process at the center of that epistemology. The final paper in this section shifts gears, exploring a different use of the two-​dimensional framework. Jackson’s interpretation is broadly semantic—​he treats the two-​dimensions at issue as two intensions, which are semantic items. An alternative interpretation, from Robert Stalnaker (2003), is metasemantic—​it treats the first dimension as playing a role in determining what a use of an expression expresses and the second as playing a role in determining the extension of that use. In his chapter, Greg Bochner reviews Stalnaker’s account, arguing that, appearances notwithstanding, it contradicts the familiar brand of semantic externalism advanced by Kripke and Putnam, and so, must be revised.

Natural Kind Terms and Rigidity The next seven chapters explore natural kind terms—​roughly, terms for kinds that (at least apparently) exist naturally, rather than in virtue of, say, our classificatory practices. Standard candidate natural kind terms include “water”, “cat”, “tree”, “electron”, and “light”. Descriptivism about natural kind terms resembles descriptivism about names, holding that, a reference-​fixing description gives their meanings. Kripke rejects this descriptivism just as he rejects descriptivism about names. He then suggests extending his picture of names to natural kind terms mutatis mutandis. In the first chapter in this group, Genoveva Martí and José Martínez-​Fernández note a difficulty for this extension.Whereas Kripke holds that names have the same extension in all possible worlds in which they refer (i.e., they are rigid designators), natural kind terms clearly do not—​there is, after all, a possible world that contains more cats or less water than our own, and so, there is a possible world where “cat” and “water” have different extensions than they actually have. This suggests that such terms are not rigid designators, which would mark a deep difference between them and names. Martí and Martínez-​Fernández divide attempts to maintain that natural kind terms are rigid designators into two broad categories, essentialist and denotationalist.They then explore challenges faced by each, including the threats of trivialization (i.e., collapsing the distinction between the rigid and non-​r igid) and overgeneralization (i.e., rendering all kind terms rigid).They conclude by noting that this is a live issue, with potentially significant ramifications. In the next chapter, Emily Foster-​Hanson and Marjorie Rhodes use results from cognitive psychology to explain how we think about natural kinds. They first review properties that people attribute to natural kinds—​e.g., being obvious joints in nature, having absolute boundaries, capturing fundamental patterns of similarities and differences, and involving stable membership. They then note that this thinking is often misguided—​joints can be unclear, boundaries malleable, etc. Next, they consider where natural kind concepts come from, concluding that they result from a basic conceptual bias rather than being byproducts of scientific education—​they are, after all, used by children prior to formal schooling. Foster-​Hanson and Rhodes then consider why some kind concepts become natural kind concepts while others do not, emphasizing that introducing a label for a kind and then using that label to attribute properties to everything labeled is especially significant, though not decisive. They also note how dangerous natural kind concepts can be, since we tend to treat some social kinds as both natural and normative—​i.e., as telling us how members must be and should be, not merely how they are. 8

Introduction

Foster-​Hanson and Rhodes suggest that natural kind concepts result from a basic conceptual bias. In the next chapter, Stephen Biggs and Ranpal Dosanjh consider why we have that bias, asking why natural kind terms have what Putnam (1975) called a “natural kind sense”—​i.e., a sense such that they refer to an underlying kind if there is one but refer to a superficial kind if there is not. Biggs and Dosanjh suggest that natural kind terms have a natural kind sense because our concepts develop in a way that is connected to our explanatory practices, specifically to conform to our demand for simple, powerful explanations. This account, they suggest, has several striking consequences, suggesting, for example, that many superficial kind terms (e.g., “bully” and “voluntary”) have a natural kind sense, and that, consequently, the world plays a crucial role in determining whether any given term is a natural or a superficial kind term. In the next chapter, Joseph LaPorte explores theoretical identity statements such as “water is H2O”, offering reasons to think that, contrary to Kripke and Putnam, such statements express propositions that are knowable a priori. First, he suggests that direct reference theorists can hold that “water is water” and “water is H2O” express the same proposition because “water” and “H2O”, which have the same referent, contribute only their referents to relevant propositions, and so, both are a priori if either is. Second, he suggests that theoretical identity statements may be knowable a priori because stipulation plays a crucial role in determining the extension of constituent natural kind terms (e.g., in determining whether or not whales are in the extension of “fish”), and plausibly such stipulation yields a priori knowledge (as one who stipulates that “baloo” means bear can know a priori that a baloo is a bear). Laporte concludes by cataloging some challenges for his view. LaPorte’s challenge to the Kripke-​Putnam view of theoretical identity statements leaves their anti-​descriptivism intact. In the next chapter, Åsa Wikforss and Sören Häggqvist propose a more radical departure from the Kripke-​Putnam line. They suggest that the anti-​descriptivism of Kripke and Putnam (and even the neo-​descriptivism of Jackson’s two-​dimensionalism) fits poorly with both what sciences tell us about natural kinds themselves and how natural kind terms are used. On the former, they note that the Kripke-​Putnam view defines natural kinds by appealing to underlying essences but that such essences have proven elusive even in paradigm cases, such as water and tiger. On the latter, they cite experiments finding that substantial minorities of subjects apply natural kind terms based at least partly on descriptions. They then note that (a cluster version of) traditional descriptivism fits these facts quite well—​since it does not posit essences and can explain variation in application-​conditions as resulting from differences in which descriptions in the cluster subjects find salient. Accordingly, they conclude, a return to traditional descriptivism should replace the popular Kripke-​Putnam  view. In the next chapter, Richard Boyd presents his accommodationist account of reference. His account is animated by a commitment to scientific realism and, correspondingly, is informed by the presumption that scientific terms can refer univocally despite changes in relevant concepts that result from progressing scientific understanding of their referents. The accommodationist theory treats the reference of natural kind terms as determined by accommodation between linguistic, conceptual, inferential, and observational practices, as well as epistemically relevant causal structures. This, he thinks, preserves reference to real, mind-​independent kinds, while recognizing the roles we play in shaping which kinds our terms refer to. In the final chapter in this group, Robin Hendry argues that natural kind words are radically multivocal. Putnam held that folk and expert uses of “water” are univocal. John Dupré (1999) countered that folk uses differ from expert uses. Hendry extends the multivocality, suggesting that even within each group uses are various and unstable, guided by shifting activities and interests. Hendry develops this view by way of responding to two challenges to Putnam’s account: the qua problem (i.e., since samples of a natural kind often share many properties, one wonders what makes one rather than another definitive) and the problem of indeterminacy (i.e., the extension of natural kind terms is often indeterminate until communities of users make decisions about their application). Hendry suggests that any response to these problems should be mindful of a key goal of Putnam’s 9

Stephen Biggs and Heimir Geirsson

account—​viz., preserving convergent realism in the face of semantic challenges from Paul Feyerabend (1962) and Kuhn (1962). He concludes that, all things considered, his brand of classificatory pluralism strikes the best balance.

The Empty Case The next two chapters explore apparent reference to the non-​existent. The first, from David Braun, offers a Millian account of empty names—​i.e., names for things that don’t exist. The second, from Ben Caplan, discusses descriptivist views of empty names. Millians hold that the meaning of a name is its referent. This implies that empty names are meaningless. Braun suggests that empty names therefore contribute nothing to propositions expressed by sentences containing them, and so, they generate gappy propositions. He does not see this as a cause for concern, however, because he thinks that most allegedly empty names are not, in fact, empty. Take ‘Sherlock Holmes’. Braun says that Holmes exists as an intentional artifact (created by Arthur Conan Doyle and sustained by our thoughts of him) and that this artifact is the referent of ‘Sherlock Holmes’. Caplan first explains the multiple roles that sentences containing fictional terms can play, and then discusses two main views: one according to which fictional names refer to numbers or properties, and a second according to which fictional names do not refer but express senses given by definite descriptions—​e.g., although ‘Hermione Granger’ doesn’t refer, it expresses a sense given by a description. Finally, Caplan discusses Kripke’s objections to this brand of descriptivism about empty names.

Singular (De Re) Thoughts The next three chapters explore singular thought—​i.e., thoughts about a particular individual or object, such as John Linnell or the Campanile at Iowa State University. In the first chapter in this group, François Recanati suggests that, as per Fregeans, the individuals at issue in singular thoughts can be thought of under different modes of presentation but that, contrary to Fregeans, these modes are not akin to descriptions—​the reference of singular thoughts is determined relationally, not satisfactionally. On Recanati’s view, singular thoughts about an object require that one deploy a mental file of that object, where a mental file is not of what best satisfies the descriptions in the file (as a descriptivist would have it) but rather of whatever stands in the right relation to it—​the right relation serving as an information channel which makes it possible to gather information about the object. Mental files, then, can be understood as singular terms in the language of thought. A singular term in language, such as ‘Cicero’, inherits its reference from the mental file associated with it. In the next chapter, Jody Azzouni considers whether or not we can have singular thoughts of non-​ existent entities, such as Daffy Duck. According to many, Recanati included, we cannot—​for, singular thought requires acquaintance but we cannot be acquainted with the non-​existent. Azzouni counters that we can perceive and think about the unreal no less than the real since relevant psychological states can be directed toward either. Accordingly, he says, there is no deep semantic (or syntactic) difference between simple sentences containing “Daffy Duck” and those containing, say, “Charles Manson”, which implies that we can have singular thoughts about either. This is not because, as per Braun, apparently non-​existent individuals exist. Rather, Azzouni claims, Daffy Duck does not exist and that whatever does not exist is nothing at all. He then argues that this simple nontology is compatible with empty singularity. In the final chapter in this group, Rachel Goodman presents a more restrained view on singular thoughts. She focuses on the function of names and what that function entails about the kinds of thoughts we have when we use names to communicate. Goodman questions the common assumption that communicating with a name enables the listener to have singular thoughts about the 10

Introduction

person named.The central function of names, she says, is that in which the speaker and hearer coordinate their existing bodies of beliefs about the object at issue. As Goodman points out, understanding an utterance need not involve understanding the singular thought expressed by it.

Indexicals The next six chapters focus on indexicals—​and issues that arise when exploring them. Compare “Obama was president before Trump” to “He was president before him”. Given ordinary meanings and relevant facts about presidents, the truth-​value of the former is fixed but the truth-​value of the latter is not—​because the referents of the names are fixed but the referents of the pronouns are not. Adding appropriate context can determine the referents of “he” and “him”; for example, the speaker’s pointing at Clinton when saying “he” might make “he” refer to Clinton. When context impacts an expression in this way, that expression is context-​sensitive. A context-​sensitive word is an indexical. In the first chapter in this group, Stephen Neale and Stephen Schiffer argue that two interacting factors determine the referent of any use of an indexical: its meaning and the speaker’s intention in using it, with the former constraining the latter—​notice that context was not among the factors. For some indexicals, such as “this” and “that”, the meaning is null, and so, the speaker’s intention is unconstrained. Consider two utterances: (5) She is alive. (6) That is alive. In (5), the referent of “she” is whoever the speaker refers to but only if the intended referent is female. In (6), the referent of “that” is simply whoever the speaker refers to—​the meaning of “that” places no further constraint. Indexicals differ in meaning if they place different constraints on the referent. For example, “I” and “she” have different meanings because ordinary uses of “I” can refer to something that is not female, while ordinary uses of “she” can refer to something other than the speaker. Neale and Schiffer argue that their account is superior to familiar accounts from David Kaplan (1989) and Robert Stalnaker (1978), which assign context a reference-​constituting role.To be sure, they say, context plays an epistemic role in helping us identify the referents of indexicals, but, they insist, it never plays a metaphysical role in determining reference. Some indexicals refer (at least, in their paradigmatic uses) only if the speaker demonstrates the referent. Examples include “this” and “that” in “This goes with that”, perhaps said while pointing first to peanut butter and then to jelly. The next chapter in this group explores a special use of such demonstratives. In this chapter, Marga Reimer considers demonstrative references to the unreal, as when Macbeth, hallucinating a dagger, says, “That is sharp”. Familiar uses of demonstratives refer to objects in the speaker’s environment that are accessible to the listener(s). But what, Reimer asks, is demonstrated when one demonstrates the unreal? Using hallucinations as a paradigm case, she answers that, as in ordinary demonstrative reference, the speaker intends to demonstrate an object in the environment that is accessible to the listener(s), but, unlike in ordinary demonstrative reference, the speaker thereby demonstrates her hallucination, which is a pseudo-​perceptual experience. This kind of deferred demonstrative reference, Reimer emphasizes, is nothing new:  it happens, for example, when we refer to fictional characters by demonstrating actors playing them. Much as Oliver Sacks draws conclusions about ordinary cognition by exploring psychological abnormalities, Reimer uses this account of demonstrative reference to the unreal to draw conclusions about the nature of demonstrative reference in general, the qua problem, negative existentials, and the argument from hallucination. The next three chapters explore another kind of indexical, first-​person pronouns, which have received a great deal of attention because, some think, they can express attitudes that otherwise cannot be expressed—​i.e., they are essential indexicals. Such de se attitudes (i.e., attitudes of oneself), 11

Stephen Biggs and Heimir Geirsson

the idea goes, are distinct from any attitude that replaces the concept expressed by a first-​person pronoun with that expressed by a name. In their chapter, Stephan Torre and Clas Weber consider various ways that de se attitudes might be special, concluding that these attitudes have certain features (e.g., being un-​sharable) that potentially equivalent attitudes do not. In his chapter, Dilip Ninan considers further ways in which de se attitudes might be special, settling on the view that they play a special role in motivating action. In their chapter, Herman Cappelen and Josh Dever counter that corporations never have any of the features that supposedly make de se attitudes special even though they share any potentially special-​making features that such attitudes might have. If either of the first two chapters is right, then first-​person pronouns are essential indexicals. If the final chapter is right, then they are not. Plausibly, each indexical has a common meaning across contexts despite shifting reference. Kaplan called this constant aspect of meaning “character” and the shifting aspect of meaning “content”, holding that character and context together determine content. Meaning, on this view, is a function that takes two distinct inputs, character and context, and delivers a single output, content—​Kaplan’s view, as one might expect, is a precursor to two-​dimensional semantics. If Kaplan’s account of meaning is complete, then context cannot affect character. Semantic monsters are words that purport to violate this dictum, making context impact character. Kaplan’s semantics, accordingly, has a strict no monsters policy.The final chapter in this section, from Brian Rabern, explores semantic monsters. Rabern evaluates Kaplan’s case against monsters, finding that it is too dismissive, before explaining why the question of whether or not there are monsters is surprisingly important when assessing semantic theories.

Epistemology of Reference The final three chapters explore the epistemology of reference. An epistemology for a domain considers how we achieve (foundational) justification (or knowledge, warrant, etc.) for beliefs in that domain.The most widely explored epistemology of reference places intuition center-​stage. In the first chapter in this group, Edouard Machery explores results from experimental philosophy that challenge such intuition-​based epistemologies. He begins with a summary of the study (by Machery himself, Ron Mallon, Shaun Nichols, and Stephen Stich) that sparked interest in experimental philosophy. This study found statistically significant between-​g roup differences in Westerners’ and East Asians’ applications of names in well-​known imagined cases—​e.g., a version of Kripke’s Gödel-​Schmidt case. Since it’s implausible that members of either group are more likely than the other to apply names correctly, this variation suggests that subjects’ applications (and the intuitions that drive them) do not reveal relevant facts about reference, whatever those may be. Machery proceeds with a detailed history of the fruits of the original study, including attempts to replicate its results, often across variations in experimental design. He concludes by suggesting that fifteen years of experimental work largely confirms the assessment from the original study. Undeterred, the next two entries advance intuition-​based epistemologies of reference. In the first, Daniel Cohnitz and Jussi Haukioja aim to explain why intuitions are a reliable guide to facts about reference. While drawing several useful distinctions, they argue that intuitions track facts about reference because, for any given referring expression E, the very psychological features that produce relevant intuitions also determine E’s application conditions. In the final chapter of the volume, Jon Bengson argues that an intuition-​based epistemology of reference best explains various data, including, e.g., that we have wide-​ranging beliefs about reference, that many of these beliefs are justified, and that intuition seems to be the main source of justification for them. Bengson then considers two objections to intuition-​based accounts. One, based on the findings Machery reviews, agrees that we often rely on intuition to form beliefs about reference but denies that the resultant beliefs are justified. The other, based on reflection on philosophical practice, denies that intuition even seems to be the basis for these beliefs. Bengson’s driving idea is that these objections get grip only when one mistakenly conflates intuition with gut reaction. 12

Introduction

References Appiah, Kwame Anthony. 1996. “Race, Culture, Identity:  Misunderstood Connections”. In K. Anthony and A. Guttmann (eds.) Color Conscious: The Political Morality of Race, pp. 30–​105. Princeton University Press. Barcan Marcus, Ruth. 1961. “Modalities and Intensional Languages”. Synthese, 13: 303–​322. Chalmers, David. 2006. “The Foundations of Two-​Dimensional Semantics”. In M. García-​Carpintero and J. Maciá (eds.) Two-​Dimensional Semantics, pp. 55–​141. Oxford University Press. Donnellan, Keith. 1966. “Reference and Definite Descriptions”. Philosophical Review, 75: 281–​304. Dupré, John. 1999. “Are Whales Fish?” In D. L. Medin and S. Atran (eds.) Folkbiology, pp. 461–​476. MIT Press. Feyerabend, Paul. 1962. “Explanation, Reduction and Empiricism”. In H. Feigl and G. Maxwell (eds.) Scientific Explanation, Space and Time, pp. 28–​97. University of Minneapolis Press. Frege, Gottlob. 1892/​1997. “On Sense and Reference”. In M. Beaney (ed.) The Frege Reader, pp. 151–​180. Blackwell Publishing. Jackson, Frank. 1998. From Metaphysics to Ethics. Oxford University Press. Kaplan, David. 1989. “Demonstratives”. In J. Almog, J. Perry, and H. Wettstein (eds.) Themes from Kaplan, pp. 481–​563. Oxford University Press. Kripke, Saul. 1980. Naming and Necessity. Harvard University Press. Kuhn, Thomas. 1962. The Structure of Scientific Revolutions. University of Chicago Press. Lewis, David. 1999. “Naming the Colors”. Papers in Metaphysics and Epistemology, pp. 332–​358. Cambridge University Press. Putnam, Hillary. 1962/​1975. “The Analytic and the Synthetic”. In H. Putnam Mind, Language, and Reality: Philosophical Papers, 2. Cambridge University Press. Putnam, Hillary. 1975. “The Meaning of Meaning”. Minnesota Studies in the Philosophy of Science, 7: 131–​193. Putnam, Hillary. 1981. Reason,Truth and History. Cambridge University Press. Quine, W. V. O. 1968. “Ontological Relativity”. Journal of Philosophy, 65: 185–​212. Russell, Bertrand. 1905. “On Denoting”. Mind, 14: 479–​493. Stalnaker, Robert. 1978. “Assertion”. Syntax and Semantics, 9: 315–​332. Stalnaker, Robert. 2003. Ways a World Might Be: Metaphysical and Anti-​Metaphysical Essays. Clarendon Press. Strawson, Peter. 1959. Individuals: An Essay on Descriptive Metaphysics. Methuen. Williamson, Timothy. 1994. Vagueness. Routledge. Williamson, Timothy. 2013. Modal Logic as Metaphysics. Oxford University Press. Wittgenstein, Ludwig. 1953/​2009. Philosophical Investigations. Translated by G. E. M. Anscombe, P. M. S. Hacker, and Joachim Shulte. Blackwell Publishing.

13

PART I

Early Descriptive Theories

1 THE CONCEPT OF LINGUISTIC REFERENCE BEFORE FREGE Michael Losonsky

Ancient Greek Philosophy Reference in the philosophy of language is, at a minimum, a relationship between words and items in the world, typically individual objects, but depending on one’s views about what there is, words can also refer to events, processes, classes, or kinds. This rudimentary notion of reference is found in Pre-​ Socratic philosophical fragments. Parmenides of Elea (c. 475 BCE) refers to the custom “of naming” and that human beings have names for both “Becoming and Perishing, Being and Not-​Being”, although he believes it is a mistake to have names for Not-​Being because what is not the case is neither thinkable nor expressible, and thus these names are unintelligible (Freeman 1983, 43–​44). The 5th-​century Sophist Gorgias of Leontini, at least as it was reported by Sextus Empiricus (c. 200 BCE), distinguished human speech from things which are external to human beings and concluded from this that language cannot give any information about the external world because speech can only be about mental states. So Parmenides, on the one hand, identified the intelligibility of language with reference to existing things, making it impossible to literally say something false. On the other hand, Gorgias cannot understand how a word can refer to something else, and thus for him it is impossible even to say something true about the world. Plato (429–​347 BCE) avoids these absurd consequences by distinguishing the intelligibility of language from reference. How language represents what is independent of language and its users was a focal point for Plato (429–​347 BCE) in his dialogue Cratylus. Plato considered two positions: (a) names refer to objects on account of conventions that stipulate that the name refers to that object, and (b) names refer to objects when the name naturally has or displays the nature of the object, either as a likeness or as a description of the object. He rejected both accounts and argued that naming depends on having prior knowledge of the named objects, suggesting that the intelligibility of names rests on something distinct from either the name or object. Plato identifies the human understanding as the source of the intelligibility of names. Plato maintains that human conceptions of how things are can be incorrect while still being intelligible, and consequently, human beings can speak falsely as well as truly. In the later dialogue, Sophist, Plato returns to this topic and argues that meaningful speech can be true or false, rejecting the Parmenidean view that meaningful but false speech is impossible. True speech, Plato maintains, says about things “as they are” and false speech “says things different from those that are” (263a–​b). This is possible because meaningful speech is a “weaving [of] verbs with 17

Michael Losonsky

names” and these can be woven together to make different statements about how things are (262d). By distinguishing between how something is said to be and how things are, Plato begins to differentiate meaning and reference. Plato’s student Aristotle (384–​322 BCE) characterizes the intelligibility of human speech in terms of three components: spoken sounds, affections of the soul, and actual things. Spoken sounds signify affections of the soul, which include both thoughts and perceptions. Affections of the soul, in turn, enable human beings to think about actual objects in virtue of the human understanding having forms in common with these objects. The relationship between speech and affections of the soul is conventional, but the relationship between thought and actual things is natural because whether or not a thought or perception has forms in common with actual things is not something that can be stipulated. Language is intelligible because it signifies affections of the soul, and affections of the soul are intrinsically intelligible because forms for Aristotle, as was the case for Plato, just are the basic sources of intelligibility. The meaning of speech, then, is a function of the signified affections, but this is not a referential relationship. Reference, as a relation between words and things, holds between spoken words and actual things when the forms of the signified thought or perception are also forms of actual things. While spoken words can refer to actual things via the signified affections of the soul, the affections themselves do not refer to actual things. Thoughts or perceptions of actual things display the forms of actual things and thus actual things and affections literally have something in common, namely the same forms, but this does not mean that thoughts and perceptions of actual things are identical with actual things. For example, the perception of a stone is not identical to the stone nor is the stone itself present in the perception, but the form of the stone, without its matter, is in the perception (De Anima, 431b20–​30). Aristotle does not have a special term for reference, but he explicitly distinguishes signification from the relation to actual objects. There is no such thing as a goat-​stag, but nevertheless, Aristotle writes, “goat-​stag” signifies something (De Interpretatione, 16a16–​17). Language’s relation to actual things becomes relevant for Aristotle when considering the truth or falsity of sentences that make statements, namely sentences that are used to affirm or deny something of something. While “goat-​stag” by itself signifies a form, by itself it is neither true nor false because it does not affirm or deny anything. However, once we add a verb, for instance, “is”, “will be”, or “was”, and make a statement, whether or not there is an actual goat-​stag is relevant. The affirmation “A goat-​stag is”, where “is” is not a copula but predicates existence, is false because there are no goat-​ stags. On the other hand, the affirmation “A man is” is true because an actual thing has the form of man. The difference is that the form of goat-​stag has no matter and the thought and object are identical, but the form of man that is displayed in thought without matter is in reality combined with matter to form an actual thing. In sum, for Aristotle, to evaluate the intelligibility of speech, affections of the soul need to be considered, but when evaluating the truth or falsity of statements, the relationship of speech to actual things needs to be considered. Nevertheless, although the reference relation between words and things is recognized by Aristotle and assigned a role in his philosophy of language, this is not a central topic for him. His primary concern regarding language was meaning and the logical relationships of statements, which he understood to be independent of the question of what there is. That was the topic of metaphysics. The concern with meaning abstracted from reference also prevails in Stoic philosophy of language. The Stoics distinguished between the sound-​forms of speech (phōnē), written forms (lexis), what an utterance signifies or expresses (lekton), and the unity of sound and meaning, meaningful speech (logos). Stoic philosophy of language sharply distinguished lekta, or what is expressed with language, and the extra-​mental objects or events to which meaningful speech refers. An important property of lekta is that they are expressible in language, but they are immaterial abstractions that are not identified with occurrent psychological states or with the sentences that express them. Truth-​value is a 18

Linguistic Reference before Frege

property of lekta that form statements, not the sentences that express them (Diogenes 1925, 57–​66; Bobzien 2003, 85–​88; Law 2003, 38–​42). The concept of something expressible in language, but distinct from language, mind, and what exists plays a central role in medieval philosophy and prefigures the notion of propositional content in the philosophy of language of Gottlob Frege (1848–​1925).

Medieval Philosophy Early medieval philosophy of language continues the trajectory set by Aristotle and the Stoics to focus on propositional content and its logical structure, abstracted from reference, as the ground of the intelligibility of language. Building on the Stoic concept of the lekton, Augustine of Hippo (354–​430) in his De Dialectica distinguished between an utterance (verbum) and the utterance used to say or express something (dictio). What is said or expressed is something sayable or expressible (dicibile), which is not something heard or seen, but is understood by and contained in the mind (Augustine 1975, v:88). What is expressible has a status independent of the actual expression because it can be understood by a mind independently of the expression. Augustine identified a fourth component of speech, namely the object that is signified by speech (res), but the intelligibility of language depends on dicta, not res. While Augustine isolated the component of what is “sayable [dicibile]”, Peter Abelard (1079–​1142) analyzes what is sayable into distinct components. Abelard distinguished the propositional content of a statement (enuntiatio, dictum propositionis) from the affirmation expressed by the statement. The content of a statement is what is understood or proposed when spoken, and Abelard classifies this as a kind of signification (significatio intellectualis), not to be confused with the signification of objects (significatio realis) (Abelard 1956, 148, 154–​156). What is understood, the propositional content, can be shared by other kinds of speech acts beside statements, such as commands, questions or wishes (Abelard 1956, 151). A statement is an assertion or affirmation of what is understood, and as such it is either true or false (Abelard 1956, 154). So truth-​value is not a property of content, but the affirmation of content. By separating propositional content from the assertion of the content, Abelard was able develop a better understanding of sentential connectives and operators, motivating the study of what came to be known as the syncategorematic terms. Abelard calls reference understood as a word-​world relationship “nominatio” or “appellatio” to distinguish that relationship from the relation of significatio. Nominatio or appellatio do not play a role in determining the propositions that language expresses. In other words, reference does not play a role in determining the semantics of natural language, and hence Abelard believes that reference can be neglected in the study of the meaning of language (Abelard 1984, 233/​308.19–​22). The primary semantic function of names is not to refer to something, but to signify concepts. The role for reference for language is to determine whether an affirmation of a proposition is true or false. While this is essential for knowing how the world is, for Abelard this is not essential for understanding natural language. Reference begins to be recognized as having a role in determining the meaning of natural language with the introduction of supposition theory in 13th-​and 14th-​century European philosophy of language. Supposition theory had two taproots: (i) the distinction between signification (significatio) and naming or calling (nominatio or appellatio), already developed by Abelard, and (ii) the concept of a suppositum, namely of something that underlies or instantiates the signified concept or nature, which already appears in Boethius’s commentaries on Aristotle. Suppositum also had a grammatical or syntactic use, namely to be the subject of predication in a statement. Combined, these concepts suggest a new area of inquiry: what objects are named by the noun that is the grammatical subject of a statement? Medieval grammarians, logicians, and philosophers turned to this issue in response to the intensified study of fallacies and sophisms that the rediscovery in Europe in the 12th century of Aristotle’s Sophistical Refutations inspired. A  widely shared diagnosis of fallacies and sophisms was that they resulted from not recognizing that while the signification of a noun can remain the same, in different 19

Michael Losonsky

contexts of statements, the objects that are named by the noun can change depending on what is being predicated in a statement. For example, Peter of Spain (d. 1277) in his widely used introduction to logic textbook Summulae Logicales maintains that while the signification of “human being” is constant, there is a difference between the statements “A human being is running” and “Human being is a species”.The failure to recognize a difference leads to fallacies.While inferring “Some human being is running” from “A human being is running” is valid, inferring “Some human being is a species” from “Human being is a species” is a fallacy (Peter of Spain 1972, 69–​72). Depending on the philosopher, what is signified by a term is a common nature, essential property, substantial form, defining concept, or a class of past, present, future, and possible individuals. While what a word signifies can change by imposing a new signification on a word—​for example, when a word is redefined—​it was assumed that given that a word has a signification, there are no changes in what it signifies. This assumption was grounded in the view that a term signifies a common nature or substantial form and these were assumed to be changeless. So the difference had to be located elsewhere, and it was located in the context of a statement. While “human being” has the same signification in different statements, the context can determine a different supposit, that is reference. For example, according to Peter of Spain, in “Human being is a species” the term “human being” supposits a universal or a human-​being-​in-​general, not an individual, while in “Some human being is a species” the term “human being” supposits individuals. Hence, inferring the latter statement from the former is invalid. The theory of supposition distinguished between three general kinds of suppositions: personal, simple, and material supposition. In the case of personal supposition (suppositio personalis), the term “human being” supposits concrete individuals, for instance in “A human being is running” or “Some human being is running”. Simple supposition (suppositio simplex) occurs when the term just supposits what it signifies, for example in “Human being is a species”. In the third kind of supposition, material supposition (suppositio materialis), the term supposits itself, as in “Human being is a term”. Material supposition is particularly interesting because it introduces the use/​mention distinction without introducing a new term, which is the case when quotation marks are used to distinguish between the use and mention of a term. The three types of suppositions were further subdivided. Peter of Spain distinguished, for example, between determinate and diffuse personal supposition. The personal supposition of “human being” in “A human being runs” or “Some human being runs” is determinate because these statements are true if there is a human being that runs. On the other hand, in “Every human being is an animal”, the supposition is diffuse because it refers to any one of its personal supposits. In all these cases “human being” has the same signification, but what changes is the supposition depending on the predicative context. Other contexts that were treated by supposition theory were the varieties of modal and temporal contexts, for instance the differences between “A human being runs” and “A human being can run”. The recognition of the semantic contribution of the context anticipated Frege’s context principle that a name is meaningful only in the context of a sentence, albeit a weaker version, namely that while a term has meaning on its own, it refers only in the context of a statement. Medieval accounts of supposition were not concerned with what determined or fixes reference for a term in different contexts, but only with distinguishing reference in different sentential contexts, primarily for the purposes of identifying fallacies and other ambiguities and puzzles gathered under the heading of Sophismata or Insolubilia. In the 14th century, most notably by William of Ockham (1285–​1349), the theory of supposition spawned an interest in the inferential relationships between generalizations involving personal suppositions—​for instance, “Every human being is an animal” and singular statements, e.g. “This human being is an animal” or “Socrates is an animal”—​thus explicitly linking reference and quantification. Relatedly, Ockham used the concept of supposition to avoid commitments to abstract entities and develop a nominalist ontology of individuals. 20

Linguistic Reference before Frege

According to Ockham, words signify individual things, not universals. Instead of universals, there are concepts that signify many things, and words signify concepts only when they are spoken of, in which case they are also individuals. Thus Ockham modifies the received Aristotelian view that words signify concepts which signify things, by maintaining that except for special contexts where someone wants to speak about concepts, words signify individual things, not concepts. However, Ockham retains the view that terms signify individual things via concepts. Words are “subordinated” to concepts, he writes, and they “signify the very things that are signified by concepts of the mind, so that a concept primarily and naturally signifies something and a spoken word signifies the same thing secondarily” (William of Ockham 1974, I:50). This holds for singular terms as well. Ockham denies that there are individual natures, such as Socrateity, but he affirms that there are individual concepts that signify only one individual. An important feature of Ockham’s notion of individual concepts for proper names is that they cannot be defined and thus there is no description that expresses the signification of a proper name. Jean Buridan (1295–​1356), Ockham’s student in Paris, explains that the individual concept for a proper name requires acquaintance with the object. According to Buridan, a person has a genuine individual concept only if they are actually observing the individual, and hence proper names signify an individual only in the actual presence of that individual. Proper names that are used later without acquaintance—​Buridan gives the example “Aristotle”—​do not signify via an individual concept, but are subordinated to a description involving universal concepts that happen to signify Aristotle (Ashworth 2004, 135–​138, 147–​148). This aspect of nominalism marks the emergence of a new conception of reference. The traditional Aristotelian conception of “reference” is a word-​world relationship determined by concepts. A term’s intelligibility as well as its reference is determined by its associated concept even in the case of singular terms. The requirement that a genuine proper name requires the presence of the named object suggests that the object is necessary for the name’s intelligibility, and that the intelligibility of proper names is not determined by a concept. While Buridan still requires that there is an individual concept that mediates reference even in the case of naming with acquaintance, he does not assign a role to the individual concept for determining reference. Moreover, the individual concept does not have a descriptive content that can be expressed. Finally, the individual concept’s signification itself is a function of the object present in acquaintance. In other words, in cases of acquaintance, the individual concept does not determine what it signifies, rather the object of acquaintance determines the individual concept. Since the genuine proper name’s referent is mediated by this sort of individual concept, its reference is also determined by the object with which the user of the name is acquainted. This suggests that a proper name is a purely referring expression whose intelligibility rests on a word-​ world relationship and not on some descriptive content.

Renaissance Philosophy The theory of reference embodied in medieval supposition theory fades away during the Renaissance. While Roman Catholic and Protestant scholastic work on language and logic persisted into the 17th century—​for instance in the works of John of St. Thomas or Franco Burgersdijk—​philosophers of language between 1450 and 1600 who expressed and exemplified the Renaissance commitment to humanist ideals rejected the increasingly technical and formal studies of medieval logical and semantic studies as irrelevant and turned to customary, literary, or civic uses of language. The complications and intricacies of the theory of supposition were particularly targeted for ridicule and understood as a consequence of a narrow focus on the formal deductive validity of syllogisms to the exclusion of informal and ordinary argumentation, especially in civic and political debate, that persuades even without deductive validity. The decline in interest in the scholastic supposition theory began with instructors of Dialectic as well as Rhetoric who had to come to terms with teaching traditional logic to a new 21

Michael Losonsky

generation of students impatient with the intricate abstractions of traditional logic. Lorenzo Valla (1407–​1457) while teaching Rhetoric at the University of Pavia launched an attack on scholastic logic in his Retrenching of Dialectic and Philosophy, later published under the more modest title Dialectical Disputations. Valla grants that supposition theory is needed for a more complete understanding of formal validity, but denies the relevance of formal validity to actual human reasoning. Formal deductive validity is relevant only if human beings, using rational tools, can attain certain knowledge, but Valla denies this. What is attainable are conclusions that are probable or persuasive, and these typically rest on non-​deductive reasoning. This requires paying attention to the conventions of speech (usus loquendi; consuetudo), which should also be the touchstone for resolving philosophical issues (Valla 2012, I:644). Accordingly,Valla does not treat the theory of supposition in his Dialectical Disputations. The influence of Valla’s humanist revision of logic was limited, but it was soon followed by more popular texts, especially Rudolph Agricola’s (1443–​1485) Dialectical Invention, first published in 1515 after his death and whose popularity eclipsed Peter of Spain’s Summulae in northern Europe. Agricola defines Dialectic as “the art of speaking with probability” and focuses on “infusing conviction” rather than apodictic certainty. Agricola turns his attention to the art of invention, developing a methodology for generating propositions that an arguer can use to complete an argument that relies on a list of twenty-​four topics or locations that were intended to organize the knowledge available for argumentation. Agricola’s focus on developing methods for organizing knowledge influenced Peter Ramus’s (1515–​1572) Divisions of Dialectic, an even more popular textbook because it sought to organize subjects in simple pairwise divisions. Both Agricola and Ramus ignore the theory of supposition. Other Renaissance writers did not ignore supposition theory; for example Jacques Lefèvre d’Étaples (c. 1460–​1536), who taught traditional logic in light of new and revised translations of Aristotle’s Organon, and published an introduction to logic, Logical Introductions (1496), which provided a summary of the traditional subjects covered by Peter of Spain’s Summulae. However, Lefèvre apologizes for presenting the “outlandish and vulgar literature” and for the “barbaric dialectics” of the theory of supposition (Lefèvre 1972, 39, 82). Similarly, Augustino Nifo (1469–​ 1538) at the University of Padua in his textbook Playful Dialectics [Dialectica Ludicra] characterizes supposition theory as a disease, but nevertheless gives a condensed and anemic version of it (Ashworth 1976, 366–​367). Since supposition theory as a theory of reference did not survive into the Renaissance, it was difficult for Renaissance philosophers of language to clearly distinguish sense and reference. For example, Valla tied signification to essential or defining concepts, and hence used “significatio”, “sensus”, and “intellectus”, and as concepts they can be subsumed by the predicaments or categories. However,Valla distinguished the signification of a term from the “thing signified [res significata]” which does not fall under a category. Valla writes, “the actual man who is signified is under a roof or the sky, however, not under a predicament” (Valla 2012, I:14, 221). Unfortunately, Valla does not address the issue of whether every term that refers requires a concept that determines reference, but his discussion of singular terms suggests that these refer when acquainted with an individual.

17th-​and 18th-​century Philosophy In the early modern period the concept of linguistic reference began to acquire clearer contours. For Francis Bacon (1561–​1626), words are “counters” or “signs of notions” of the mind, which Bacon also describes as the “souls of words”, but the notions human beings have in mind are “abstracted from things” with more or less care. Traditional logic is only concerned with relations between concepts and propositions without making any “reference to things”, and Bacon aimed to correct this lacunae with a new logic, a logic of induction, concerned with properly abstracting notions from things. Unfortunately, Bacon does not explicitly address the 22

Linguistic Reference before Frege

nature of the word-​world relationship, but his discussion strongly suggests that meaningful names track notions, and that the reference of names is determined by the associated notion. Bacon also does not discuss in detail how the mind’s notions are related to their objects, but it seems that he is taking for granted the Aristotelian conception of the relationship between language, mind, and reality. Words signify notions, which have descriptive content that is either matched or not matched by something in the world. For Bacon, human beings do not have direct acquaintance with external objects; only God and “perhaps angels and intelligences” have such direct acquaintance (Bacon 2000, 127). Thomas Hobbes (1588–​1679) explicitly endorses the view that names signify mental conceptions, not things, and language is used to keep track of mental states (as in the case of memory), guide mental states (as in the case of reasoning), and communicate what a person has in mind to others (Hobbes 1996, 24). The conceptions human beings signify—​Hobbes typically thinks of conceptions as mental images—​make language intelligible to ourselves and others. This relationship, as was the case for Aristotle, is arbitrary and voluntary. However, these conceptions can be caused by external objects, and Hobbes explicitly identifies these natural causes as the objects of our conceptions. So while a statement signifies the speaker’s conception, Hobbes identifies an additional relationship that language has to the world, namely to the object of the conception. For instance, a statement can signify a sensory image of the sun—​for example the sensed light, color, and heat—​and nevertheless the object of the statement is not the image, but the sun, because that is the object of the signified sensory image. Hobbes’s distinction between what a statement signifies and the object of what it signifies countenances a referring relationship between word and object in addition to the meaning relationship between word and mind. Linguistic reference for Hobbes is constituted by a causal relationship between our mental states and the object in our environment. This is an important departure from the Aristotelian triad of word-​mind-​object. For Hobbes and Aristotle, the word-​world relationship is mediated by mental states, but Hobbes replaces Aristotle’s shared forms with causal relationships between mind and object. This diminishes the semantic role of the signified conceptions because while they are intermediaries that make language meaningful, they do not determine reference. Instead, it is the object and how it causally interacts with the speaker that determines reference. The Hobbesian understanding of word-​world relationships is sharpened and clarified in the philosophy of language of John Locke (1632–​1704). According to Locke, the ideas of the human understanding are the only things human beings know directly, and thus, Locke concludes, “Words in their primary or immediate Signification, stand for nothing, but the Ideas in the mind of him that uses them”. Nevertheless, while human beings immediately signify only their own ideas, those ideas, however imperfectly or carelessly, “are collected from the Things, which ideas are supposed to represent” (Locke 1975, II.2.2: 405). Since the human understanding can only contemplate its own ideas, “’tis necessary that something else, as Sign or Representation of the thing it considers, should be present to it: And these are Ideas” (Locke 1975, IV.21.4: 721). So the ideas that human language immediately signifies facilitate a word-​world relationship because these ideas can represent objects in our environment, and hence words can be used “ultimately to represent Things” (Locke 1975, III.11.24: 520). In this manner Locke demarcates the divide between linguistic meaning and reference that evolved in the Middle Ages. However, Locke recognizes referring expressions in the case of words for simple ideas because what they represent is determined by their causes. Simple ideas, for example the sensations we have of color, light, temperature, or movement, are involuntary effects of objects’ causal powers, and for Locke the type identity of simple ideas is a function of their causes. Locke explicitly affirms that two simple ideas, even in two different people, represent the same object if they are caused by the same object. Consequently, passive simple ideas still intervene between words and things, what they represent is determined by their causes, not their content. This direct connection between simple ideas and the external world plays a role in Locke’s account of our knowledge of the external world, 23

Michael Losonsky

arguing that we experience simple ideas of sensation as “actually coming into our Minds” (Locke 1975, IV.2.14: 537). In the case of complex ideas, such as ideas of substances, for example gold, reference depends on how human beings have “collected from the Things” their simple ideas and tied them together in a complex idea. Reference for terms that signify complex ideas is determined by whether or not something conforms to the complex idea. While the simple ideas are determined by their causes, the complex idea into which they are collected need not have a referent. So, in the case of names of substances, Locke adheres to the traditional Aristotelian view that meaning determines reference. Continental European philosophy during the early modern period adheres closely to this traditional relationship between meaning and reference. Nevertheless, there is a place for direct reference in the philosophy of René Descartes (1596–​1650).While Descartes does not develop a philosophy of language, he assumes that what human language refers to is a function of human ideas—​he makes an exception for the Cogito, that is self-​knowledge. Human beings know objects in the external world, including God, on the basis of human ideas, but individually they have direct acquaintance with their own ideas and their own selves. Consequently, the terms a person uses for these refer directly. For example, the first-​person use of “I” refers to the person without any intervening ideas. While Descartes had little to say about language, Antoine Arnauld (1612–​1694) and Pierre Nicole (1625–​1695) in the so-​called Port-​Royal Logic and Arnauld and Claude Lancelot (1615–​1695) in the Port-​Royal Grammar developed a Cartesian philosophy of language. They preserved the Aristotelian view that words “signify”, “indicate”, or “mean” our thoughts. Thoughts consist of ideas, which for the Port-​Royalists are not images, but rather intellectual concepts that are explicated in terms of their objects or what they represent. Singular or individual ideas represent individuals, and these ideas are signified by proper nouns, for example “Socrates” or “Bucephalus”. Common or appellative nouns such as “man” or “horse”, signify universal or general ideas, which represent several things. For general ideas, the Port-​Royal Logic sharpens the scholastic distinction between signification and supposition in terms of the distinction between comprehension and extension. A general idea comprehends the attributes that necessarily constitute the idea. For instance, the comprehension of the idea of a triangle contains extension, shape, three lines, three angles, and the equality of these three angles to two right angles. The extension of the idea consists of the objects to which these attributes apply, which includes individuals as well as species or kinds.While the concept of extension is introduced for general ideas, the Port-​Royal Logic also mentions the extensions of words, assuming that common terms have the extension of the common ideas they signify. In the Port-​Royal Grammar the comprehension/​extension distinction is made in terms of a distinction between the signification of common nouns and the extension of the signification, clearly indicating that a common noun’s referent is determined by the idea it signifies. While in the Grammar, as in the Logic, proper names signify singular ideas that represent individuals, it is noteworthy that neither the Logic nor the Grammar discuss or even mention the comprehension of singular ideas of proper names. Moreover, there is no discussion of how singular ideas represent individuals, and there is no indication that the Port-​Royalists held that a singular idea comprehends attributes that uniquely apply to an individual. On the contrary, they seem to drive a wedge between signified ideas and extension in the case of singular terms. For instance, proper names are “always taken in their entire extension”, that is the extension is the whole individual with all its properties, including accidental properties and properties we do not know or do not have in mind. In fact, Arnauld and Nicole explicitly discuss cases where a singular term signifies attributes that diverge from the term’s reference. They observe that the term “true religion” comprehends a set of attributes that are satisfied by only one religion—​Roman Catholicism, they assume—​but nevertheless spoken by a Protestant or Muslim that term would not refer to Roman Catholicism, even though, on their account, it is the only thing that satisfies the concept of a true religion. The reference is determined by the “context of discourse” and “circumstances”, including what the speaker believes 24

Linguistic Reference before Frege

to be the true religion, but not by the descriptive content of the description “true religion” (Arnauld and Nicole 1996, 45; Arnauld and Lancelot 1975, 73). Gottfried Wilhelm von Leibniz (1646–​1718) assumes the traditional view that language signifies ideas and Arnauld’s distinction between the comprehension and extension of ideas, where objects fall under or instantiate concepts (Leibniz 1996, 486; Leibniz 1903, 411 and 519). However, he highlights the role of singular ideas that proper names signify. He maintains that for every individual substance, whether actual or merely possible, there is a “complete idea or concept” that includes all the attributes that apply to that individual and that together apply to that individual uniquely (Leibniz 1970, 268, 312). In other words, proper names are not different from common nouns in that they signify ideas whose comprehension determines an extension, but Leibniz combines this with the intuition in the Port-​Royal Logic that proper names are “always taken in their entire extension”. Leibniz does this by including in the individual concept all the attributes that in fact apply to the individual, past, present, and future, as well as all the attributes that could apply to it. Leibniz accepts the consequence of this view that all true statements about an individual are true in virtue of the meaning of the proper name: the attributes predicated in such statements are already contained in the concept of the subject. However, Leibniz aims to accommodate the fact that human beings speak meaningfully about individuals without knowing everything about them and that they discover new facts about individuals by rejecting the assumption that the conscious and occurrent thoughts human beings have determine the nature of their ideas. For Leibniz, ideas have a structure that is independent of human awareness and only God has full access to this structure. Leibniz believes that God’s mind is the “region” of all ideas, including ideas of everything that is possible, and that God surveys all the complete ideas of individuals when choosing who and what to create (Leibniz 1970, 647). While human knowledge is very limited, human beings nevertheless are able to refer to other individuals in virtue of their own ideas because these have a structure that “express[es] [exprime]” God’s ideas. Leibniz holds that x expresses a distinct y if the constituents of x have relations to each other that have a constant and orderly correspondence to the relations of the constituents of y (Leibniz 1970, 207; and Leibniz 1903, 15). Leibniz illustrates this with the example of an ellipse and a circle: an ellipse can express a circle because “any point whatever on the ellipse corresponds to some point on the circle according to a definite law”(1970, 208; also 1996, 131). Another example he gives of expression is the pain of a pin prick, which expresses “the motions which the pin causes in our body” because there is a “precise relationship” between the pain, which consists of insensible and minute perceptions, and the motions of the pin. Accordingly, human beings express God’s ideas in virtue of a function from human ideas to God’s ideas. Leibniz grants there is a vast difference in the perfection of divine and human ideas, but they nevertheless have enough of a structure in common to make the human mind an expression of God’s mind. But the structure of divine ideas, according to Leibniz, expresses perfectly all there is, including every existing individual (1970, 580). Hence there is a function from human ideas to the objects of the universe, and accordingly human ideas “express”, “represent”, or “correspond [...] to things” (Leibniz 1996, 109). An important feature of Leibniz’s metaphysical edifice is that the structure of ideas and the structure of reality are combinatorial structures that can be represented by a language that also has a combinatorial structure. All natural languages, according to Leibniz, have combinatorial structures and in fact, all languages have a common structure. The combinatorial structure of natural language expresses the structure of human ideas, and thus via God’s mind, natural languages express all there is, including particular individuals.The combinatorial structure of natural language includes the entailment relationships of language, and these are the “logical form [forme logique]” of language that captures its semantic content (1996, 480). Leibniz’s account of word-​world relationships outlines a new approach to the problem of reference. Rather than treating reference as a particular relation that obtains independently of the rest of language, he treats it as a function of a whole language. If the 25

Michael Losonsky

theological elements are abstracted from Leibniz’s philosophy of language, reference is a function of the form of the whole language.While Leibniz assumes that there is a unique mapping from language to reality, his notion that the reference of a term is a function of the whole language is a component of relativist and instrumentalist treatments of reference.

19th-​century Philosophy Leibniz’s insight that linguistic structure determines word-​world relationships is a keystone in the philosophy of language of Wilhelm von Humboldt (1767–​1835). Humboldt’s work on language also develops in the wake of Immanuel Kant’s (1724–​1804) Copernican Turn, namely the view that the human understanding plays an essential role in the structuring of human experience, including our perception of individual objects, so much so that the human understanding projects properties onto the perceived objects. While Kant has little to say about language, Humboldt sees linguistic structure as the source of the structure of the human understanding and hence language constitutes human perception of objects. According to Humboldt, linguistic structure or the “form of language” has three components: (i) the “sound-​form” of a language, that is, its phonology, morphology, and syntax, (ii) the “internal” or “intellectual” conceptual form that consists of the “forms of intuition and logical ordering of concepts” (Humboldt 1999, 84), and (iii) the “inner linguistic form”, namely the rules for expressing the mind’s forms of intuition and conceptual structure in the phonology, morphology, and syntax of language (Humboldt 1999, 48, 54, 84). Humboldt maintains that there is a feedback loop between these three components and this feedback accounts for the diversity of languages as well as conceptual structures. Human beings have a shared conceptual structure and inner linguistic form, but diverse environments and needs as well as spontaneous creativity produce distinct sound-​forms that in turn effect new conceptual structures. Since conceptual structures form how human beings experience the world, human beings with diverse languages have diverse “world views [Weltansichten]” (Humboldt 1999, 60). The view that language structures how human beings experience and think about the world is a hallmark of linguistic relativism and irrealism, which has been associated with critiques of the very notion of reference and a word-​world relationship. If the objects of human experience are structured by language, then arguably language does not refer to objects but constitutes them. Humboldt, however, is not an irrealist.While experience is structured by mind and language, “we do not abandon the world that really surrounds us”. Instead human beings surround themselves “with a world of sounds so as to take up and process internally the world of objects” (Humboldt 1999, 60–​61). The experience of an object is not only product of mental and linguistic activity, but it also requires “things in external nature”“to press in upon” the perceiver. Accordingly, one of the functions of the sound-​form of language is “to designate objects” (Humboldt 1999, 54–​55). Designation, for Humboldt, is not limited to objects. Words also designate concepts and syntactic structures. While occasionally Humboldt uses a special term for the designation of objects—​ Bedeutung—​this use is rare and typically he shifts easily between discussing the designation of objects and the designation of concepts (Humboldt 1999, 74 and 145). The reason for this is that for Humboldt the word-​world relationship is essentially embedded in the form of language: it is not possible to think about or refer to objects without the conceptual and linguistic structure that is used to think about and refer to objects. However, Humboldt does not affirm the traditional view, revived by Frege, that the descriptive content of concepts determines reference. While human relationships to external objects are always mediated and facilitated by mind and language, mind and language do not determine or pick out the objects that they designate. For Humboldt the word-​world relationship is a causal relationship. The objects of the external world “press in” and “operate upon” the human understanding, leaving an impression that “excites” or “animates” the mental activity that embeds it in the form of language and forms a conceptual unity that includes the sound-​form of the word that “takes the place” of the object (Humboldt 26

Linguistic Reference before Frege

1999, 55, 60). Humboldt believes that the details of this causal process are inscrutable, but he nevertheless conjectures that there is a causal connection between the linguistic sounds human beings use to refer to objects and the audial impressions caused by these objects (Humboldt 1999, 72–​73). What is certain for Humboldt is that objects bring about sensory impressions that are processed and unified into a mental and linguistic structure that includes a “designating impression” (Humboldt 1999, 55). In sum, human language has a structure of its own that includes conceptual and linguistic components, but it also refers to external objects via causal ties between these objects and the human understanding, allowing for linguistic reference that is in part determined by the object and not solely determined by the human understanding. Nevertheless, Humboldt’s primary interest is the diversity of phonological, morphological and syntactic structures of natural languages, arguably prefiguring the emphasis on syntactic structures at the expense of semantics in early 20th-​century linguistics. On the other hand, John Stuart Mill (1806–​1873), a reader of Humboldt, aimed to revive the semantic and logical perspective on natural language, which Mill associated with Scholasticism. Mill believes that the study of logic and the study of language are inseparable and accordingly Book I of his System of Logic is devoted to “Names and Propositions”. Propositions are linguistic items, namely assertions or statements that have an import or meaning which is partially constituted by inferential relationships. Names are “names of things themselves, and not merely of our ideas of things” (Mill 2011, 31). Some names are general and apply to several things, while other names are singular or individual and belong to unique individuals. Some singular names include general names, e.g. “this stone” or “the king who succeeded William the Conqueror”, while others, e.g. “John”, as used in a statement, do not include anything that belongs to something other than the named individual. Mill develops this into a distinction between “connotative” and “non-​connotative” names, which he describes as “one of the most important distinctions” that he makes and that goes “deepest into the nature of language” (Mill 2011, 37). A name has a connotation when it entails or means an attribute or set of attributes so that when the name is used in a statement, the statement implies that the subject has the attribute or attributes that the name connotes. Attributes, for Mill, are defined dispositionally as powers in objects to produce certain sensations or patterns of sensations. For example, assuming that “snow” connotes the attribute of whiteness, when “snow” is used in a statement, the statement implies that snow has the power to “excite” the sensation of whiteness. Non-​connotative names do not imply any attributes. Their only semantic function is to refer to something, typically an object, such as an individual person with that name, or an attribute, as in the case of the name “whiteness”. On Mill’s account, neither “John Stuart Mill” nor “whiteness” have a connotation.The name “whiteness” needs to be distinguished from the term “white”, which does connote the attribute of whiteness. According to Mill the object that is denoted by a non-​ connotative name is known only on the basis of the sensations brought about by the object, and the object as it is in itself—​Mill cites Kant and uses the term noumenon—​is unknown (Mill 2011, 64). However, human beings have a fundamental intuition that there is an “unknown cause of our sensations” and that this is an “external cause”, and this intuition can also be justified on the grounds that external causes best explain the law-​like patterns of our sensations (Mill 2011, 58, 65, 77). Connotative names can also have a denotation, which is fixed by the connotation, that is determined by the implied attributes. For instance, the common noun “man” denotes a class that includes “Peter, Jane, John, and an indefinite number of other individuals” and connotes some attributes, including having “corporeity, animal life, rationality, and a … human [form]”, and this name is used to denote those and only those objects with all those connoted attributes (Mill 2011, 37). Connotative singular terms, such as definite descriptions, also denote in virtue of the description’s implied attributes. So with respect to connotative names, Mill follows tradition, albeit modified by his empiricist and Kantian understanding of attributes:  human beings refer to the unknown external causes of the patterns of sensations of their connoted attributes. 27

Michael Losonsky

Mill departs from tradition by explicitly affirming that there are referring expressions, typically proper names, without any connotation and whose only semantic function is to denote. “Proper names are attached to the objects themselves and are not dependent on the continuance of any attribute of the object”, Mill writes (Mill 2011, 39). Mill has little to say about how proper names get attached to objects and in fact seems to retract this claim when he writes a little later that a proper name is “not … attached to the thing itself ” but instead is connected to our idea of that individual object (Mill 2011, 41). What Mill has in mind is that the name is not literally attached to the object, but connected to the idea of the object. When a person is acquainted with an object, the person has an idea of it, and that idea is a causal effect of that object. The name is attached to that idea to distinguish that individual object from others and to allow users of that name to make assertions about it in subsequent uses of that name without regard to its attributes. Mill compares a proper name to placing a mark on an object to distinguish it from other objects, but the mark’s only function is to distinguish that object without any implications about the object’s attributes. In the case of a proper name, however, the mark is not literally on the object, but “so to speak, upon the idea of the object” and “it enables us to distinguish it when it is spoken of, either in the records of our own experience, or in the discourse of others” (Mill 2011, 41). Thus Mill sets the stage for 20th-​century developments in the theory of reference. His distinction between connotation and denotation anticipates Frege’s distinction between sense (Sinn) and reference (Bedeutung). However, Mill drew it very differently. While Frege preserves the traditional view that a descriptive content determines reference, Mill distinguishes between names where this is the case and names that refer without any descriptive content.This line of thought that the sole semantic function of some names, for instance proper names, is to refer to an object is picked-​up by Bertrand Russell (1872–​1970) and developed in the theories of direct reference that flourish in the second half of the 20th century.

References Abelard, Peter (1956). Dialectica, ed. L. M. de Rijk, Assen: Van Gorcum. Abelard, Peter (1984). “Glosses on Peri Hermeneias” in Aristotle’s Theory of Language and Its Tradition: Texts from 500 to 1750, ed., transl. and commentary by Hans Arens, Amsterdam: John Benjamins, 231–​302. Agricola, R. (1992). De Inventione dialectica, ed. L. Mundt, Tübingen: Niemeyer. Aristotle (1982). The Complete Works of Aristotle, 2 vols., ed. J. Barnes, Princeton, NJ: Princeton University Press. Arnauld, A. and Lancelot, C. (1975). General and Rational Grammar: The Port-​Royal Grammar, trans. J. Rieux and L. Rollin, The Hague: Mouton. Arnauld,A. and Nicole, P. (1996). Logic or the Art of Thinking, trans. and ed. J.V. Buroker, Cambridge, UK: Cambridge University Press. Digital printing 2003. Ashworth, E.  J. (1976). “Agostino Nifo’s Reinterpretation of Medieval Logic”, Rivista Critica di Storia della Filosofia, 31:4, 355–​374. Ashworth, E. J. (2004). “Singular Terms and Singular Concepts: From Buridan to the Early Sixteenth Century” in John Buridan and Beyond, Historisk-​ filosofiske Meddelelser 89, ed. R. L. Friedman and S. Ebbesen, Copenhagen: The Royal Danish Academy of Sciences and Letters. Augustine (1975). De Dialectica, trans. and ed. B. D. Jackson and J. Pinborg, Dordrecht: Reidel. Bacon, F. (2000). The New Organon, ed. L. Jardine and M. Silverthorne, Cambridge, UK:  Cambridge University Press. Bobzien, S. (2003). “Logic” in The Cambridge Companion to the Stoics, ed. B. Inwood, Cambridge, UK: Cambridge University Press, 85–​123. De Rijk, L. M. (1967). The Origin and Early Development of the Theory of Supposition, vol. II, part 2, Logica Modernum, Assen: Van Gorcum. Diogenes Laertius (1925). Lives of the Eminent Philosophers, vol. II, trans. R. D. Hicks, London: William Heinemann. Freeman, K. (1983). Ancilla to the Pre-​Socratic Philosophers, Cambridge, MA: Harvard University Press. Hobbes, T. (1996). Leviathan, ed. R. Tuck. Cambridge: Cambridge University Press. Humboldt, W. v. (1999). On Language: On the Diversity of Human Language Construction and Its Influence on the Mental Development of the Human Species, ed. M. Losonsky, trans. P. Heath, Cambridge, UK:  Cambridge University Press.

28

Linguistic Reference before Frege Law,V. (2003). The History of Linguistics in Europe: From Plato to 1600, Cambridge, UK: Cambridge University Press. Lefèvre d’Étaples, J. (1972). The Prefatory Epistles of Jacques Lefèvre d’Étaples and Related Texts, ed. E. F. Rice, New York: Columbia University Press. Leibniz, G.  W. (1903). Opuscles et fragments inédits de Leibniz, ed. L. Couturat, Paris:  Alcan. Repr. 1988, Hildesheim: Olms. Leibniz, G. W. (1970). Leibniz: Philosophical Papers and Letters, 2nd ed., ed. L. Loemker, Dordrecht: Kluwer. Leibniz, G.W. (1996). New Essays Concerning Human Understanding, ed. and trans. J. Bennett and P. Remnant, New Essays on Human Understanding, New York and Cambridge, UK: Cambridge University Press. Locke, J. (1975). An Essay Concerning Human Understanding, ed. P. H. Nidditch, Oxford: Clarendon Press. Mill, J. S. (2011). System of Logic, Adelaide: eBooks@Adelaide. Digital reprint of 1882, 8th ed., New York: Harper and Brothers. Peter of Spain (1972). Summulae Logicales, ed. L. M. de Rijk, Assen: Van Gorcum. Trans. F. P. Dinneen (1990), Language in Dispute, Amsterdam: John Benjamins. Plato (1997). Plato: Complete Works, ed. J. M. Cooper and D. S. Hutchinson, Indianapolis, IN: Hackett Publishing Company. Ramus, P. (1964). Dialectique, ed. M. Dassonville, Geneva: Droz. Valla, L. (2012). Dialectical Disputations, 2 vols., ed. and trans. B. Copenhaver and L. Nauta, Cambridge, MA: Harvard University Press. William of Ockham (1974). Ockham’s Theory of Terms: Part I of the Summa logicae, trans. M. Loux, Notre Dame, IN: University of Notre Dame Press.

29

2 FREGE ON REFERENCE Rachel Boddy and Robert May

1  Introduction Any historically based assessment of Frege’s contributions to contemporary thought begins, and ultimately ends, with Frege’s logicist program: the project of reducing arithmetic to logic. As conceived by Frege, to carry through this project, it was necessary to demonstrate that the truths of arithmetic could be faithfully rendered by true logical propositions; that is, by propositions expressed in purely logical terms, using no terms with any residual mathematical content. If successful, and if these propositions could be shown to be formal consequences of the logical axioms, this would be sufficient to show that arithmetic is a branch of logic. Dear to Frege’s conception of logicism, from its inception in Begriffsschrift, to its climax in Grundgesetze and the classic essays of the early 1890s, and through to its denouement following Russell’s Paradox, was the notion that logicism was at heart a scientific thesis.1 What this meant to Frege was that the truths of arithmetic, constituted as logical propositions, were to be judged in the same manner as any other scientific propositions. Frege’s crystallization of how this proceeds via the semantic method—​the idea that terms of a language are referential, and that through their composition, the truth of propositions of which they are constituents can be determined—​rates as one of his deepest and most lasting insights. Most importantly, it is at this point that the notion of reference—​Bedeutung2—​enters the Fregean picture, with the idea that the significance of terms of a language resides in their referentiality as it contributes to determining the truth of the propositions in which they occur. The Context Principle is the moniker commonly assigned to this governing principle of reference. On Frege’s perspective, scientific discourse can successfully express truths about the external world precisely because scientific languages are referential: because the expressions of the language refer to entities that exist independently of the language, the language can be used to assert propositions about those entities. From the perspective of logicism, this means that if numbers are amendable to scientific investigation, it is because they are entities external to language that are the referents of number-​terms; ipso facto, numbers exist, and arithmetical truths can be asserted. This methodology, however, immediately begs the question: what is the language that contains these terms, and whose sentences express the asserted propositions? The logicist answer is that it is the language of logic. The Begriffsschrift, as Frege called it, is expressively rich enough to have names of numbers, and express concepts under which they fall.3 30

Frege on Reference

This aspect of the logicist position coincides with another core logicist doctrine. Centrally in Grundlagen,4 but assumed throughout his work, Frege argues that numbers are abstract objects; more precisely, in that work he proposes that numbers are definitionally specified as abstract, logical objects.5 What this means is that the notion of reference must be sufficiently broad so that it encompasses not only concrete entities—​i.e., those sorts of things standardly amenable to scientific analysis—​but also abstract entities as referential values; that is, as referents also amenable to scientific analysis. But just how is this so? What justifies treating the logico-​mathematical language as referential? How can it be shown that number-​terms, logically characterized, refer to abstracta? The difficulty here is apparent, as prima facie we have no independent purchase on numbers as objects, as we do for the entities that make-​up the domain of inquiry of other sciences (for instance, via empirical observation). However, to defend his realist conception of logic and arithmetic, Frege had to explain how truths about these objects can be established, and accordingly, he had to confront these questions. This confrontation with reference—​ how Frege addresses it, and the consequences of his approach—​is the topic of this contribution. Our project is to explore how the core theoretical concepts that Frege introduces—​notably the notions of proper name and sense—​address the tensions that arise from the application of the semantic method for scientific analysis to abstracta, and the essential role that definitions play in this extension. In the story to be told, reference sits at the nexus of great themes in Frege’s logicist conception; for Frege, it is foundational for a notion of scientific meaning that is sufficient to establish arithmetic truths as scientific truths, and so procure epistemic security for mathematical knowledge.

2  Reference to Numbers At the core of Frege’s logicism is the view that logic is a science, in the sense that it is an axiomatic system, whose axioms—​the Basic Laws of logic—​express fundamental conceptual truths. On this conception of logic, which Frege develops in full in Grundgesetze der Arithmetik, the Basic Laws of logic specify the lawful properties of any conceptual universe, where a conceptual universe is a domain that conforms to the conceptual hierarchy. Such a domain is comprised of concepts and objects: the 0-​level entities are objects, the 1-​level entities are concepts under which objects fall, the 2-​level entities are concepts within which 1-​level entities fall, etc. (See Gg I, §§19, 21–​24 for discussion of the stratification into levels.) On Frege’s view, in a logically perfect language, that every term is a referential term can be established in the context of a conceptual universe so conceived.This type of completeness is required for any language appropriate for addressing properly scientific questions.6 If the truths of logic are truths about a conceptual universe, then as any conceptual universe must include 0-​level entities, the question is raised as to what they are: what are the logical objects? What sort of things fulfill this role? Frege’s gambit was the introduction of the reification of concepts, their value-​ranges, as the floor of the conceptual hierarchy. A value-​range of a function is the graph of its pairings of arguments and values; for concepts, whose values are the truth-​values, this is equivalent to their extension. Concepts agree on their value-​ranges just in case they have the same extension.7 Effectively, what Frege has done is to build the universe of concepts—​the entities of the first level and above—​into the 0-​level of the hierarchy via their value-​ranges. Accordingly, the value-​ranges are the logical objects. Given this ontology, the central mathematical claim of logicism is that cardinal numbers can be identified with particular, definable, value-​ranges.8 This is how Frege makes good on the task he sets in Grundlagen of establishing that cardinal numbers are logical objects, rather than ideas (or mental constructions), intuitive objects (e.g., points), or empirical objects (e.g., chairs).That numbers have this status does not detract, Frege insists, from their objectivity. That these objects are abstract does not in itself detract from their stature as genuine 0-​level entities of a conceptual universe. Frege’s view then is that numbers, defined as such, can be the scientific subject-​matter of logic, and that the logical laws, qua scientific laws, express truths from which the arithmetic axioms (i.e., the Peano-​Dedekind 31

Rachel Boddy and Robert May

axioms) can be derived. On this picture, arithmetic truths are scientific truths about abstracta and, consequently, arithmetical judgments are judgments about a mind-​independent realm of objects. At this juncture in the development of Frege’s account, there is a worry. Given that numbers are neither empirically nor intuitively accessible, what purchase do we have on their existence such that they may be within our cognitive grasp? What makes them objects that are accessible to thinking? Frege’s response is to cast this as the question of what justifies treating singular terms as referential, that is, as terms that can secure reference to an external realm of objects. Since reference entails the existence of a referent, if this justification is provided for number-​terms, this will be sufficient to establish the existence of numbers as Frege has characterized them, as abstract logical objects. Our cognitive grasp of numbers now becomes a matter of grasping a language in which occur referential terms whose referents are numbers. Thinking, on Frege’s view, depends on reference; not on our direct grasp of objects, but on objects as referents.9 In many contexts of use, Frege allows that we may presuppose that the justification of reference is forthcoming: “in order to justify speaking of the referent of a sign, it is enough, at first, to point out our intention in speaking or thinking. (We must then add the reservation: provided such a referent exists)”.10 This may be unproblematic for the use of ordinary names. A speaker using the name “Aristotle” to refer to Aristotle can presuppose without harm that there exists (or existed) a particular person who is the intended referent of the name; the assumption that people exist is hardly a point of controversy. But reliance on this presupposition must be closely monitored in the context of scientific languages so as to avoid begging the question. In the case of concern, simply assuming that number-​ terms refer to numbers-​objects will hardly do when the very question on the table is whether the purported referents exist; it is the existence of abstracta as referents that is the problem under consideration. In science, the promissory note comes due. It is not enough to presuppose justification of reference; it ultimately must be provided. If, as Frege develops the play, the route to the grasp of abstract objects is through language, then we are owed an account of how this occurs. Per logicism, we are owed an account of what secures that number-​terms are referential. To address this, Frege argues that the grasp of referents, inclusive of abstract referents, is to be explained by the grasp of the meaning of names for them. And this Frege argues, critically, requires the introduction of sense as a distinct component of their meaningful content.

3  Sense According to Frege, the meaning of proper names—​and of linguistic expressions in general—​is not exhausted by their reference.There is another component; the component of sense.With “Über Sinn und Bedeutung” (1892), Frege crystallized the distinction between the sense (Sinn) of an expression and its reference (Bedeutung), stating famously that “It is natural, now, to think of there being connected with a sign (name, combination of words, letter), besides that to which the sign refers, which may be called the reference of the sign, also what I should like to call the sense of the sign” (Frege 1952b, 57).The content (Inhalt) of a sign is just this pairing of a sense and a reference, such that the sign expresses its sense and refers to its referent. As Frege conceives of a language as a semiotic system, it consists of terms that are labels for senses, not labels for referents: what is symbolized by signs (via their linguistic form) are senses. In a logically perfect language, distinct signs will signify distinct senses, and no sense will be symbolized by more than one sign.11 Consequently, in a logically perfect language, of which scientific languages are a species, each name is associated with a definite sense, to which, in turn, there corresponds a definite reference.12 Senses, as Frege conceives of them, are characteristics of an object, such that as a referent, that object is made thinkable in a way comporting with those characteristics. Any such property, Frege calls a mode of determination (Bestimmungsweise) of the object, and each way of being determined 32

Frege on Reference

is a distinct objective property of the object: that Venus is determined as the morning star and as the evening star are facts about Venus. Qua sense, a mode of determination of an object is descriptive content that serves to illuminate aspects of an object in virtue of which an agent can identify the object as the referent of an expression. This is Frege’s descriptivism: each name labels descriptive content, which affords the recognition of the object that is its referent. This is what it means for a term of the language to express a sense.13 If senses are to play the role just described as mechanisms of reference, they must be cognitively accessible to agents for thought.This accessibility relation between agents and senses, Frege calls grasp. When an agent grasps a sense, they stand in a relation to a mode of determination of a referent, and in virtue of standing in that relation they can think about the referent. By grasping a sense, agents form ideations, personal mental representations that constitute modes of presentation (Art des Gegebenseins) of a referent. If grasping a mode of determination of a referent is what makes it thinkable, then a mode of presentation is the corresponding way that an agent thinks about that referent. Grasping senses empowers individuals as agents and speakers. As agents, they come to possess information that grounds their quest to establish whether thoughts about a referent are true. In this fashion, senses make referents thinkable in ways that an agent has grounds for making judgments of thoughts about that referent.14 As speakers, by grasping senses, these thoughts can be communicated. Because senses can be grasped by any speaker, senses form a common ground of shared meaning, and to the extent that they have been mutually grasped, speakers can communicate with one another.15 It is a consequence of the relation of sense and reference that no sign of a language could express a sense, qua mode of determination, without a referent. There can be no mode of determination without there being an object so determined: no object, no mode of determination. From this, however, it should not be concluded that terms without referents are necessarily nonsensical; all that can be concluded is that if they are sensible, it is not the result of grasping a sense. Language, of course, abounds with such terms; fictional terms of myth, poetry, theater, narrative literature, etc. are of the essence, and Frege is not insensitive to their existence.The senses these terms express arise differently; it is rather from acts of imagination, of imagining what would be grasped, if there were a referent. What results are mental ideations on the order of modes of presentation, but without a corresponding mode of determination. These imaginary senses may be externalized through fictional works that make them accessible for all to grasp for their imaginations. But bear in mind, sentences containing fictional terms have no truth-​value. Since they contain terms with imaginary senses, they express only imaginary thoughts; mock thoughts, as Frege calls them.16 Fictional terms are to be distinguished from certain other terms that might appear akin, but that Frege does not take to involve referential failure. That these latter terms have sense and reference, Frege argues, is critical for the development of scientific knowledge. Centrally relevant are terms of arithmetic. In this context, “the largest positive integer” is as well-​formed and meaningful a singular term as “the smallest positive integer”, witnessed by the observation that proving that there is no largest positive integer is a substantive mathematical result. Frege devotes a section of Grundgesetze (I, §11) to discussion of just this sort of case. Roughly, Frege’s account is to introduce a function that intuitively corresponds to the definite article of natural languages: its value applied to the “the smallest positive integer” is the number 1, but applied to “the largest positive integer” it is the null extension, that is, the value-​range of a concept under which nothing falls.17 We may be tempted to say that envisaging a largest positive integer is also an act of imagination. But if so, we must not overstate its similarity to expressions of the fruits of fictional imagination: for scientific languages, there are strictures on the meaningfulness of signs in force that may be sloughed over when the logical perfection of the language is not the core constraint. This section was motivated by a question: how can reference to abstract, logical objects be secured? And now we have the answer: by the grasp of senses that determine those objects as referents. To the extent that we have a language that contains terms that express those senses, it will be sufficient to secure reference. Frege’s claim is that the logical language—​the Begriffsschrift—​is such a language. 33

Rachel Boddy and Robert May

The logicist thesis is that grasp of this language is sufficient to secure reference to numbers, given that numbers are identified with logical objects, to wit, value-​ranges. In this way, by explaining their senses, we will have justified number-​terms as referential terms, with numbers as their referents, in a manner that is cognitively accessible.

4  Reference and Meaningfulness In Grundlagen, Frege poses the following question: “How, then, are numbers given to us, if we cannot have any ideas or intuitions of them?” (§62). The answer he ultimately lands on is that our cognitive purchase on numbers, qua abstract objects, resides in our grasp of senses that determine numbers as referents: our grasp of numbers is secured by our grasp of senses as expressed by number-​terms. Frege perceives, however, that this answer is not wholly satisfactory, not if the goal is to explain how knowledge of numbers can be attained, and how we can validly reason about them. For this, something more is required than what can be provided just by securing reference. Knowledge, on Frege’s view, depends on the apprehension of truth, on making judgments; accordingly, for terms to be fully meaningful, it is required to show how their reference contributes to establishing truths, in the case of numbers, arithmetic truths. It is not just that numbers are given to us by referential expressions; it is that they are given to us by meaningful referential expressions. In thinking through this problem of meaningfulness, Frege sets a core claim, that to come to have knowledge of and reason about an object, we must “see” the object as the argument of a concept. By doing so, it becomes possible to establish truths about that object, by establishing that the object falls under the concept. Through apprehending truths so composed, we come to have knowledge of the object, and can reason about it at a standard appropriate to scientific inquiry. Such judgment of a thought generates knowledge; accordingly, we have epistemic access to numbers via our grasp of arithmetical truths. On Frege’s view, the terms for concepts and objects become meaningful only as they are compositionally joined in the expression of a thought. The notion of terms being unsaturated or saturated is introduced to guide our understanding of how this proceeds, of how concept-​terms and object-​terms can be composed into meaningful sentences.18 The meaningfulness of these terms thus arises from the roles they play in determining the truth of a thought. Frege explicates these roles by mapping the composition of terms onto the semantic notion of an object functionally falling under a concept.19 It is along this line that Frege approaches the problem of explaining our epistemic access to numbers as a problem about the reference of number-​terms as they occur in assertoric sentences: “Since it is only in the context of a sentence that words have any meaning, our problem becomes this: To explain the meaning of a sentence in which a number word occurs” (Gl, §62).20 Frege here appeals to the Context Principle—​i.e., only to ask for the meaning of a word as it is used in the context of a sentence—​in focusing attention on explaining the meaning of sentences in which number-​terms occur. In Grundlagen, this takes shape by an explanation of the meaningfulness of the term “the number of F” (i.e. an expression whose reference is the (cardinal) number of a concept), in the context of numerical identities, sentences of the form “The number of F is the same as the number of G”. The choice of an identity is fortuitous, for if numbers are objects, then identity-​statements involving number-​terms, which are objectual identities, will always be meaningful; they will have determinate truth-​conditions. If these conditions are satisfied, then we will have a “criterion for deciding in all cases whether b is the same as a”, and if so, then we are justified “to use the symbol a to signify an object” (Gl, §62). The idea here is that to justify a term as referential, it is sufficient to provide an identity criterion for its referent, and to do so is nothing other than to explain the meaning of identity-​statements involving that name.The task at hand, therefore, is “to fix the sense of a numerical identity, that is, to express that sense without making use of number words or the word ‘number’ ” (Gl, §106). Failure to heed the warning is on pain of circularity.21 34

Frege on Reference

The reasoning now proceeds in the following manner. The meaningfulness of “the number of F” depends upon the meaningfulness of “The number of F is the same as the number of G”, by the Context Principle. Its meaningfulness is explained by its equivalence with the proposition that F and G are in a one-​to-​one correspondence, a relation that is logically definable. By compositionality (if a sentence is meaningful, then so too are its constituent parts22), it follows that “the number of F” is a meaningful referential term that refers to the number of a concept. The subtlety in Frege’s thought here is in distinguishing between securing that a term is referential, for which it is sufficient that it express a sense, from justifying that it is referential, which requires showing that it is meaningful. The thesis that Frege is putting forward is that justification is forthcoming if a general identity criterion that affords recognition of referents can be offered. This is a powerful thesis; among its virtues is that it allays skeptical worries that “the number of F” is not a referential term. Nevertheless, upon putting it forth in Grundlagen, Frege promptly rejects the thesis.

5  Reference and Definition The task that Frege has set out in Grundlagen is to explain the meaningfulness of numerical identities; that is, identities of the form: The number of F is the same as the number of G. That explanation, Frege tells us, arises from recognizing its equivalence with: F and G are in a one-​to-​one correspondence. This equivalence has come to be known as Hume’s Principle (HP for short). Hume’s Principle, in setting the condition under which a number can be recognized, explains “the number of F” as a meaningful referential term, and thereby legitimizes its occurrence in the language. At this juncture, a natural thought occurs: what has been accomplished has been to give a definition. Hume’s Principle explains the meaningfulness of number-​terms; but explaining meaningfulness is precisely what we intuitively expect from a definition. Accordingly, in one of the core discussions in Grundlagen, spanning §§62–​67, Frege entertains just this possibility. But he does so with a note of caution: “Admittedly”, he observes,“this seems to be a very odd kind of definition” (Gl, §63). It is not entirely novel, however. Frege points to the definition of direction in terms of parallelism as an exemplar. Nevertheless, its oddness raises a “doubt which may make us suspicious of our proposed definition” (Gl, §66), so much so that Frege firmly rejects HP as a definition of number. Frege’s reasons are that as a definition, HP would be either circular or trivial. This is argued in Grundlagen §66 and §67 respectively, as the consequences of what has come to be known as the “Julius Caesar problem”.23 The central reason to entertain a recognition condition like HP as a definition is because it settles the truth-​conditions for identities:  for HP, these are specifically numerical identities, where both terms of the identity are number-​terms, as above.This is fine as far as it goes. But if HP is to be a definition, it must settle the truth-​conditions of all identities in which number-​terms occur. But here HP fails: Frege observes that to establish the truth by HP of “the number of F is the same as q”, where q is an arbitrary singular term—​e.g. “Julius Caesar”—​there would have to be a prior definitional determination that q is a number, i.e. just what HP is supposed to be defining.“But then we have obviously come round in a circle” (Gl, §66). In a sense, what we have is a case of the definiendum being part of the definiens. If “the number of F” is being defined in the context “—​is the same as the number of G”, the circularity is plain, if it is required that the definiens only contains terms that already have an established meaning.24 If, nevertheless, we were to proceed with the proposed supplementation, then this way of introducing q must be the only way that it is introduced, since only then will it follow that if q was not introduced by the definition, then it is not a number.This is required to address the Caesar 35

Rachel Boddy and Robert May

problem. But there is collateral damage. If q must be given in only one way qua the definition, say as the number of F, then all numerical identities “would amount simply to this, that whatever is given to us in the same way is to be reckoned as the same. This, however, is a principle so obvious and so sterile as not to be worth stating” (Gl, §67).25 The definition would fail, as it would regress to triviality. The moral of the Julius Caesar problem, as Frege sees it, is that it is not possible to move from recognition to definition. A condition that is well-​equipped for the former task is not thereby equipped to fulfill the obligations of the latter. If there is a central conceptual flaw that Frege identifies in attempting this leap, it is that it turns HP into a creative definition. The definition would assert that there exist objects defined by the property of being given as numbers. But this is to misconstrue proper definition: “The definition of an object does not, as such, really assert anything about the object, but only lays down the meaning of a symbol” (Gl, §67). Definitions, Frege is telling us, do not introduce objects to be referents; rather, they introduce referential signs for those objects. Building on this, Frege sets canons for definition by which they are explicit definitions. The hallmark of explicit definitions is that they are semantically conservative over the language. Accordingly, to the extent that definitions are creative it is linguistically; it is of terms, not of entities. Definitions are creative only in their coinage of abbreviations, and are not innovative of meaning. They are stipulations that serve to introduce new terms into a language as replacements for complex terms, whose meaning has already been established. Accordingly, by a definition, that is, a sentence of the form definiendum =def definiens, what is expressed semantically is that the definiendum and definiens express the same sense, and thus have the same reference.26 On Frege’s view then, definitions are admissible only if the definiens is understood and acknowledged as a referential expression; the definiendum inherits its reference, and its status as a referential term, from the definiens.27 By this inheritance, what Frege calls the “governing principle for definitions” is met: “Correctly formed names must always refer to something” (Gg I, §28). Meeting this requirement is a hallmark of a logically perfect language:28 A logically perfect language (Begriffsschrift) should satisfy the conditions, that every expression grammatically well constructed as a proper name out of signs already introduced shall in fact designate an object, and that no new sign shall be introduced as a proper name without being secured a reference. A logically perfect language, one properly suited for science, is thus a fully referential language, i.e., a language in which every term is referential. Explicit definitions, as Frege conceives of them, introduce novel terms within such a referential milieu. This sets a basic task for Frege: he must show that the logical language is fully referential. To do so, Frege sets out to prove in Grundgesetze that every term of the Begriffsschrift is a correctly formed referential term.29 This is the so-​called “proof of referentiality”, the core purpose of which is to legitimize the use of definitions, explicitly characterized.30 The most important definition to be legitimized in the context of the logical language is the definition of number: (D#) The number of F =def the extension of the concept being equinumerous with F. To justify (D#) as a logical definition, the definiens must have a determinate meaning within logic. The definiens contains two notions, equinumerosity and extension. The first is characterized in terms of one-​to-​one correspondence. For the second, in Grundlagen, Frege presupposes “that it is known what the extension of a concept is” (Gl, §69); in Grundgesetze, he offers an explanation via the identification of extensions of concepts with their value-​ranges. With this (D#) satisfies its definitional obligations:  it introduces a term into the language with an explanation of its sense, and thereby secures that it is referential. Accordingly “the number of F” is a referential term.31

36

Frege on Reference

The core issue for Frege has thus migrated from the meaningfulness of number-​terms per se, with the central role of the Context Principle, to the meaningfulness of the base language into which such terms are definitionally introduced, where the central role is played by the “governing principle” and the proof of referentiality of the language. Frege’s core logicist claim is therefore to be understood as claiming that the referentiality of logic with respect to a conceptual universe is sufficient to establish the existence of numbers. In doing so, the definition underwrites our epistemic access to numbers, as it presents numbers to us is a way that allows us to think about them, to derive their fundamental properties and calculate with them. Note that ontologically it does not necessarily follow that numbers are value-​ranges. All that Frege requires is that we conceive of them as such, since if we do, it can be shown, to the standards of rigor required for scientific explanation, that the Basic Laws of arithmetic can be derived from the Basic Laws of logic.

6  Concluding Remark We have now reached the end of our tale about reference, but it does not have a happy ending. In setting out to show that number-​terms are referential terms whose referents are abstract, logical objects, Frege ultimately saddles himself with an inconsistent theory. That the proof of referentiality fails, and hence that the Begriffsschrift is not a fully referential language is the lesson of Russell’s Paradox. In its wake, Frege is left without an answer to the very question he posed at the commencement of the discussion in Grundlagen, as he admits in Grundgesetze:32 Even now, I do not see how arithmetic can be founded scientifically, how the numbers can be apprehended as logical objects and brought under consideration, if it is not—​at least conditionally—​permissible to pass from a concept to its extension. Left hanging after all of Frege’s efforts from Grundlagen to Grundgesetze are still the most basic questions about how the reference of the primitive vocabulary is established and, in particular, why we should think that value-​range terms, inclusive of number-​terms, are referential. Frege’s gambit was to invoke Basic Law V to govern the passage from concepts to value-​ranges. But what is shown by Russell’s Paradox is that just this move leads to inconsistency.33 For Frege, there is really no way back which would, in his view, keep logicism intact. Nevertheless, the journey has not been without great value: centrally, it has given us the sense/​reference distinction, which has played such an important role in the narrative, and which has remained at the core of Frege’s lasting contribution to both the history of philosophy and to its contemporary practice.

Acknowledgments Thanks are due to the editors, Stephen Biggs and Heimir Geirsson, and to Ed Zalta, for their helpful comments. We are especially grateful to Marcus Rossberg for detailed and valuable commentary.

Notes 1 Begriffsschrift (Frege 1879; English translation, Frege 1972); Grundgesetze der Arithmetik (Frege 1893/​1903; English translation, Frege 2013). Throughout, references are to the English translations. We shall also use the following abbreviations: Bg for Begriffsschrift, and Gg for Grundgesetze. 2 We follow the standard of translating “Bedeutung” as “reference”, adopting the usage that an expression refers to its referent. Thus, reference is a relation, whereas referent is an entity (e.g., an object or concept) that stands in that relation. We thus respect Frege’s admonition to observe “the sharp distinction between sign and its reference” (Gg II, §138). For an enlightening discussion of the significance of the translation of “Bedeutung”, see Beaney (2019).

37

Rachel Boddy and Robert May 3 “Begriffsschrift” is standardly translated as “conceptual notation” or “concept-​script”. We typically leave the term untranslated, or otherwise speak of it as the logical language. As an aid to the reader, we italicize the name of Frege’s monograph, but not the name of the logical language. 4 Die Grundlagen der Arithmetik (Frege 1884). Except as noted, references are to the English translation (Frege 1950). We abbreviate as Gl. 5 The definition is that the number of a concept F is the extension of the concept being equinumerous with F. In Grundgesetze, the notion of an extension of a concept is subsumed under the notion of the value-​range of a function, and the definition of number is adjusted accordingly. See discussion in section 5. 6 For discussion of Frege’s conception of science, see May (2018). 7 That is, concepts F and G have the same value-​range just in case exactly the things fall under F and G. This is an informal rendering of Frege’s Basic Law V. 8 Specifically, as the value-​ranges of the second-​level concept of equinumerosity, i.e., of the concept of being in a one-​to-​one relationship. See n. 5 and discussion in section 5. 9 For Frege, proper names, definite descriptions (expressions of the form “the G”), and complete assertoric sentences all refer to 0-​level entities. For assertoric sentences (those that express thoughts), their referents are truth values, the True and the False; see Grundgesetze I, §2; also Frege (1952a). In Grundgesetze I, §10, Frege poses the question whether the truth-​values are values-​ranges; arguing that there is no logical reason not to take them as such, he identifies the truth-​values with two distinct value-​ ranges. Accordingly, all these terms of the logical language refer to value-​ranges, including sentences that are names for truth-​values so understood. 10 Although our practice will be to use the Max Black translation of “Über Sinn und Bedeutung” (Frege 1952b), in this instance our preference is for Michael Beaney’s translation (Frege 1997, 156). But unlike Beaney, we translate “Bedeutung” here most appropriately as “referent”. 11 That is, no homonyms nor synonyms. Defined terms are an exception, as they have the same meaning as their definiens. See discussion in section 5. 12 In “On Sense and Reference”, in a well-​known footnote, Frege recognizes that for names used in ordinary discourse such as “Aristotle”, different speakers may associate with it different senses, with different descriptive information. In ordinary discourse, this may be unproblematic, provided, as Frege notes, that the referent remains constant: for common communicative purposes all we need know is that there is the same reference, i.e. that the same object is under discussion. In contrast, in the “theoretical structure of a demonstrative science”, of which arithmetic is a case, this kind of looseness in the sign/​sense relation that is the hallmark of a logically perfect language is unacceptable. 13 Thus far, and for the duration, our focus will be on reference to objects, i.e. 0-​level entities.We do so because of our concern with reference to numbers, and because Frege himself standardly focused the discussion this way. But language also contains terms that refer to entities of higher levels, and parallel remarks hold for reference to concepts. 14 Frege (1977a), Frege (1977b). 15 Frege prominently makes this point in both “On Sense and Reference” and “The Thought”. 16 See Frege (1979, 130). Also, see discussion in Frege (1952b), where Frege remarks “The question of truth would cause us to abandon aesthetic delight for an attitude of scientific investigation”. 17 Frege informally puts it as follows: “an expression of the kind in question must actually always be assured of reference, by means of a special stipulation, e.g. by the convention that 0 shall count as its reference, when the concept applies to no object or to more than one” (Frege 1952b, 71). More precisely, in Gg, §11, Frege introduces the function \ξ which, for a value-​range as argument, returns an object as value if that object is the unique member of the value-​range; otherwise, the argument itself is the value. Note that analyzed this way, both “the smallest positive integer” and “the largest positive integer” will refer to value-​ranges, given that numbers are definitionally introduced as values-​ranges. 18 Frege (1952a). 19 For discussion of the composition of senses into thoughts, see Heck and May (2011). As they observe, this composition is not to be explicated by functional analysis: the sense of a concept-​word is not a function, nor is the sense of an object-​word an argument; it would be inappropriate to distinguish them in terms of saturation. Rather, the engine of composition of thoughts is the composition of the sentences to which those thoughts are tied. 20 Our translation. The German reads: “Nur im Zusammenhange eines Satzes bedeuten die Wörter etwas”. 21 In the remark just cited, Frege uses “Sinn”; compare the use of “bedeuten” in the very similar remark cited above from Gl, §62; see n. 20. “Meaning”, we think, should be the preferred translation in both cases. Bear in mind that at the time of writing these words, Frege had not yet arrived at the technical sense/​reference distinction; thus, the resulting looseness of prose is not surprising.

38

Frege on Reference 22 See Grundlagen, §60: “It is enough if the proposition taken as a whole has a sense; it is this that confers on its parts also their content”. 23 In these sections of Grundlagen, Frege’s immediate target is the definition of direction. In Gl, §68, having taken the lesson, Frege rejects HP as defining number-​terms in favor of the explicit definition (D#) discussed below. Here, we transpose Frege’s arguments from directions to numbers. 24 Alternatively, the context of the definiens could be taken as “—​is the same as  —​”, which does have an established meaning. But then, substitution of the definiendum slides directly into triviality, an instance of the law of identity (a = a). 25 Note that this is not a problem for HP as a recognition principle. Recognition conditions apply over an antecedent universe, in this case a universe in which there are numbers. In this context, HP adjudicates the conditions under which concepts have the same number, that is, when the number of F is the same as the number of G. 26 Gg I, §27. In Grundgesetze, Frege re-​affirms his position stated in Begriffsschrift, §24 that for the purposes of logic, definitions are abbreviatory, and that as a consequence of their explicit form, they are eliminable and non-​creative.This assumes that other conditions are in place, including, e.g., that the definiendum is new to the language. See Gg I, §33 where Frege lays out “principles [that] govern the use of definitions”. There is, however, a difference to be noted. In Begriffsschrift, in stating definitions, Frege employs a notion of signs having the same conceptual content, (represented by the symbol “≡”). This notion is tailored by Frege to capture that the definiendum and definiens have the same meaning. In Grundgesetze, this is replaced by objectual identity; we follow this latter view in the text. For discussion of the interpretation of Frege’s view of identity in Begriffsschrift, see May (2012). 27 This inheritance principle is the first principle that governs definitions that Frege lists in Gg I, §33. 28 Frege (1952b, 70). 29 The proof is found in §§29–​32 of the first volume of Grundgesetze. 30 The sections of Grundgesetze where the proof of referentiality is located are placed by Frege in the Table of Contents under the heading “Definitions”. See Volume I, Part 1, Section 2. This encompasses §§26–​32, which are labelled “General remarks” on definitions. 31 In Grundgesetze, Frege defines not the cardinal number of a concept F, as in the Grundlagen definition (D#), but rather the number of its value-​range. By this definition, the number of a concept is the value-​range of the concept of being equinumerous with that concept’s value-​range. Our point here remains: the notions employed in the definiens are all referential, and indeed logical terms by Frege’s lights. 32 Gg II, Afterword. 33 Accordingly, the proof of referentiality fails, for if every term of the language had a reference, the language would be consistent. What is problematic is that by introducing value-​ranges, the hierarchical universe of concepts is collapsed and flattened into the 0-​level. In effect, Frege has introduced a naive set theory into the system, and so inviting inconsistency.

References Michael Beaney. “Translating ‘Bedeutung’ in Frege’s Writings: A Case Study and Cautionary Tale in the History and Philosophy of Translation”. In Philip A. Ebert and Marcus Rossberg, editors, Essays on Frege’s Basic Laws of Arithmetic. Oxford University Press, Oxford, 2019. Gottlob Frege. Begriffsschrift. Eine der Arithmetischen Nachgebildete Formelsprache des Reinen Denkens. L.  Nebert, Halle, 1879. Gottlob Frege. Die Grundlagen der Arithmetik. Eine logisch mathematische Untersuchung über den Begriff der Zahl. Wilhelm Koebner, Breslau, 1884. Gottlob Frege. Grundgesetze der Arithmetik. Begriffsschriftlich abgeleitet. Hermann Pohle, Jena, 1893/​1903. Gottlob Frege. The Foundations of Arithmetic. Translated by J. L. Austin. Basil Blackwell, Oxford, 1950. Gottlob Frege. “Function and Concept”. In Peter Geach and Max Black, editors, Translations from the Philosophical Writings of Gottlob Frege. Basil Blackwell, Oxford, 1952a. Gottlob Frege.“On Sense and Reference”. In Peter Geach and Max Black, editors, Translations from the Philosophical Writings of Gottlob Frege. Basil Blackwell, Oxford, 1952b. Gottlob Frege. Conceptual Notation and Related Articles. Translated and edited by Terrell Ward Bynum. Oxford University Press, Oxford, 1972. Gottlob Frege. “Negation”. In P. T. Geach, editor, Logical Investigations. Basil Blackwell, Oxford, 1977a. Gottlob Frege. “The Thought”. In P. T. Geach, editor, Logical Investigations. Basil Blackwell, Oxford, 1977b. Gottlob Frege.“Logic (1897)”. In Hans Hermes, Friedrich Kambartel and Friedrich Kaulbach, editors, Posthumous Writings. Basil Blackwell, Oxford, 1979. Gottlob Frege. “On Sinn and Bedeutung”. In Michael Beaney, editor, The Frege Reader. Blackwell, Oxford, 1997.

39

Rachel Boddy and Robert May Gottlob Frege. Basic Laws of Arithmetic. Derived Using Concept-​Script. Translated and edited by Philip A. Ebert and Marcus Rossberg. Oxford University Press, Oxford, 2013. Richard Heck and Robert May. “The Composition of Thoughts”. Noûs, 45:126–​166, 2011. Robert May. “What Frege’s Theory of Identity Is Not”. Thought: A Journal of Philosophy, 1:41–​48, 2012. Robert May. “Logic as Science”. In Annalisa Coliva, Paolo Leonardi, and Sebastiano Moruzzi, editors, Eva Picardi on Language, Analysis and History. Palgrave Macmillan, London, 2018.

40

3 FREGEAN DESCRIPTIVISM Ian H. Dunbar and Stephen K. McLeod

1  Introduction By ‘Fregean descriptivism’, we mean the position attributed to Frege by Kripke (1980). It is a view about proper names that is frequently contrasted with that of John Stuart Mill. ‘Proper names’, writes Mill (1882, 40), ‘are not connotative: they denote the individuals who are called by them; but they do not indicate or imply any attributes as belonging to those individuals’. While it is debatable whether he holds, by contrast with Mill, that proper names are connotative, Frege (1892, 26–​27) splits the notion of the meaning of a proper name into two components, which he calls ‘Sinn’ and ‘Bedeutung’.These terms are usually translated as ‘sense’ and ‘reference’.1 On Frege’s account, the identity statements ‘The Morning Star is the Evening Star’ and ‘The Morning Star is the Morning Star’ differ in cognitive status.The first is a posteriori and informative; the second a priori and uninformative. Frege’s distinction between sense and reference serves, he thinks, to explain such differences. The names ‘the Morning Star’ and ‘the Evening Star’ have the same referent, but, Frege believes, different senses, containing different ‘modes of presentation’ of that referent. Kripke (1980) argues against what he calls the ‘Frege-​Russell view’ of proper names. While he acknowledges that the approaches of Frege and Russell were different, Kripke (1980, 27) sees them as converging on a denial of Mill’s account, and as holding that ‘a proper name, properly used, simply was a definite description abbreviated or disguised’.2 Kripke (1980) attributes a form of meaning descriptivism to Frege: it holds that for any proper name, there is a definite description that (to use Kripke’s term, which we will shortly cash out) gives the name’s sense. The name then abbreviates or disguises the definite description, which is therefore synonymous with the name.3 Let n be a proper name and let d be a definite description that is supposed to define it. Although synonymy is a symmetric matter, we take it that meaning descriptivism holds that d is cognitively prior (i.e., prior in the order of understanding) to n. By the lights of meaning descriptivism, one might understand d while never having encountered n, or before n entered the language, but to understand the sense of n (upon an occasion of proper use of n) that d gives (on that occasion) is, thereby, to understand (i.e., grasp the sense of) d (as, on that occasion, the sense of n). Meaning descriptivism, we take it, holds that to understand (i.e., grasp the sense of) n is just to grasp the sense of d (as the sense of n), and not vice versa. Hence, it is a reductionist view of the meanings of proper names, and, we suggest, was so regarded by Kripke.

41

Ian H. Dunbar and Stephen K. McLeod

Kripke (1980, 32–​34, 53–​60), distinguishes between meaning descriptivism and reference-​fixing descriptivism, both of which he attributes to Frege,4 and both of which he argues against. According to reference-​fixing descriptivism, for any proper name, n, there is a definite description, d, such that n and d are co-​referential and either d or d’s sense fixes n’s referent.5 Kripke (1980, 57) is illustrative of the distinction between these two forms of descriptivism: [S]‌uppose we say, ‘Aristotle is the greatest man who studied with Plato’. If we used that as a definition, the name ‘Aristotle’ is to mean [i.e., to be synonymous with] ‘the greatest man who studied with Plato’. Then of course in some other possible world that man might not have studied with Plato and some other man would have been Aristotle. If, on the other hand, we merely use the description to fix the referent then the man will be the referent of ‘Aristotle’ in all possible worlds. The only use of the description will have been to pick out to which man we mean to refer. In section 2, we describe in more detail the descriptivist view that Kripke attributes to Frege. Within meaning descriptivism, we distinguish between two constituent theses. These are name descriptivism, which holds that proper names are abbreviations of definite descriptions, and sense descriptivism, which holds that the senses of proper names are given by (i.e., as we understand this piece of Kripkean terminology, reduce to the senses of) definite descriptions. In section 3, we argue that there is little evidence, other than the evidence of Frege’s supposed commitment to sense descriptivism, that Frege was a name descriptivist. In section 4, we question the good standing of Kripke’s depiction of Frege as a sense descriptivist. In section 5, we argue that Frege did not think that the reference of a proper name was fixed by a definite description or by the definite description’s sense.

2  Descriptivism in Kripke’s Frege Kripke (1980) attributes various descriptivist theses to Frege. Frege … thought … [that] really a proper name, properly used, simply was a definite description abbreviated or disguised. Frege specifically said that such a description gave the sense of the name. Kripke 1980, 27 Frege and Russell certainly seem to have the full blown theory according to which a proper name is not a rigid designator and is synonymous with the description which replaced it. Kripke 1980, 586 Frege should be criticized for using the term ‘sense’ in two senses. For he takes the sense of a designator to be its meaning; and he also takes it to be the way its reference is determined. Identifying the two, he supposes that both are given by definite descriptions. Ultimately, I will reject this second supposition too; but even were it right, I reject the first. A description may be used as synonymous with a designator, or it may be used to fix its reference.The two Fregean senses of ‘sense’ correspond to two senses of ‘definition’ in ordinary parlance. They should carefully be distinguished. Kripke 1980, 59 Alongside reference-​fixing descriptivism, Kripke here attributes the following descriptivist theses to Frege. 42

Fregean Descriptivism

Name descriptivism: for any proper name, it is an abbreviated or disguised definite description. Sense descriptivism: names have senses and, for any proper name, and any sense that the name has, there is some definite description that gives that sense [see section 1 above]. Synonymy: for any proper name, there is a definite description with which it is synonymous. Denial of rigidity: ordinary proper names are not rigid designators. Name descriptivism and sense descriptivism are both elements of meaning descriptivism, but name descriptivism, which neither mentions nor alludes to the notion of sense, does not entail sense descriptivism. So, name descriptivism alone does not entail meaning descriptivism. Sense descriptivism, as we have formulated it, leaves unsettled the question of whether, for any proper name, it has just one sense (be that on any given occasion of, or common to all occasions of, its proper use). Thus, it leaves open the question of whether a definite description gives the (rather than just a) sense of the name (on a given occasion of its proper use, or common to all such occasions).

3  Frege and Name Descriptivism In section 1, we suggested, on Kripke’s behalf, that when one expression, d, gives the sense of another, n, this is to be understood as an asymmetrical and reductive relationship: one cannot understand n without understanding d but one can understand d without understanding n, and there is nothing more to understanding n but grasping the sense of d (as the sense of n). If giving sense means what we suggest Kripke takes it to mean, then sense descriptivism indeed entails name descriptivism. This interpretation of what Kripke intends by the notion of giving sense enables us to read Kripke as having provided a valid argument (in the quotation from Kripke (1980, 27) that opens section 2 above) for his interpretation of Frege as a name descriptivist. There appears to be little evidence, other than the evidence that Frege was a sense descriptivist, for taking him to have been a name descriptivist. Kripke’s main piece of evidence for attributing a position that is logically stronger than name descriptivism, namely meaning descriptivism, to Frege appears to be the ‘Aristotle footnote’ in Frege’s ‘On Sense and Reference’ (which we quote later, at the beginning of section 4). In that footnote, Frege does not say that the name abbreviates or disguises any of the definite descriptions with which it is associated. If the footnote is a statement of descriptivism, then it is of sense descriptivism rather than explicitly one of name descriptivism. So, again, Kripke’s attribution of name descriptivism to Frege would seem to rest upon the view, accordant with our explanation of the content of sense descriptivism (and, more specifically, its notion of giving sense) that sense descriptivism entails name descriptivism. The Aristotle footnote can, we suggest, support Kripke’s assertion that for Frege ‘a proper name, properly used, simply was a definite description abbreviated or disguised’ (1980, 27) only if the footnote supports sense descriptivism: for the footnote contains no more direct commitment on Frege’s part to name descriptivism. Let us proceed to the question of whether the footnote does support sense descriptivism.

4  Frege and Sense Descriptivism In the Aristotle footnote, Frege (1892, 27) writes: In the case of an actual proper name such as ‘Aristotle’ opinions as to the sense may differ. It might, for instance, be taken to be the following: the pupil of Plato and teacher of Alexander the Great. Anybody who does this will attach another sense to the sentence ‘Aristotle was born in Stagira’ than will a man who takes as the sense of the name: the teacher of Alexander the Great who was born in Stagira. So long as the reference remains the same,

43

Ian H. Dunbar and Stephen K. McLeod

such variations of sense may be tolerated, although they are to be avoided in the theoretical structure of a demonstrative science and ought not to occur in a perfect language. From around this point in his career onwards, Frege is normally fastidious in observing his distinction between expressions and their senses. Here, however, Frege does not carefully distinguish between definite descriptions themselves and their senses. The differing opinions are apt to be construed, we suggest, not as about which definite description is the sense of the name, but about which gives it. (On this point, we think, Kripke interprets Frege charitably.) The Aristotle footnote raises two separate questions about what Frege is saying about the senses of proper names. Does Frege support sense descriptivism: i.e., the thesis that these senses are given by definite descriptions? Does Frege believe that a proper name can have multiple senses? Before we return to the main issue of whether Frege was a sense descriptivist, let us briefly consider the second question. We cannot attempt decisively to answer it here; instead, we merely offer some relevant considerations.7 May (2006, 126–​131) argues that despite the appearance of two definite descriptions in the footnote, Frege believed that a proper name can only have one sense. For Frege, May argues, the definite descriptions that feature in the footnote do not give different senses of ‘Aristotle’, but rather, as Frege writes in the footnote’s first sentence, varying (presumably false) opinions as to what the sense of ‘Aristotle’ is. It is true that Frege then talks about that which can be tolerated in natural languages as ‘variations of sense’. May’s interpretation would require that phrase to be replaced by ‘variations of opinions about sense’: accordingly, it would be these that Frege would hold to be debarred when the language involved is a logically perfect language.8 The main text, both in the sentence to which the footnote attaches and later, suggests that Frege indeed believes that any given proper name has but one sense: The sense of a proper name is grasped by everybody who is sufficiently familiar with the language or totality of designations to which it belongs. Frege 1892, 27 [O]‌ne can hardly deny that mankind has a common store of thoughts which is transmitted from one generation to another. In the light of this, one need have no scruples in speaking simply of the sense [translator’s italics], whereas in the case of an idea one must, strictly speaking, add whom it belongs to and at what time. It might perhaps be said: just as one man connects this idea, and another that idea, with the same word, so also one man can associate this sense and another that sense. But there still remains a difference in the mode of connection.They are not prevented from grasping the same sense; but they cannot have the same idea. Frege 1892, 29–​30 In the second quotation, Frege is entertaining for the purposes of argument, rather than endorsing, the suggestion that different users of a name associate distinct senses with it. His point is that even if that suggestion is true, senses are public in that any sense that belongs to a proper name is graspable by any two users of the name. In the first quotation, to which the Aristotle footnote is appended, Frege is saying that rather than a word’s sense merely being graspable in this way, it is actually grasped by everyone who is sufficiently au fait with the language.9 To return to the main business, does the Aristotle footnote really show, as Kripke interprets it, that Frege was a sense descriptivist? Maunu (2015, 110–​111) criticizes as ‘uncharitable’ the use of the footnote as decisive evidence that Frege was a descriptivist, and he quotes two passages that appear to sit ill with the attribution of sense descriptivism to Frege: 44

Fregean Descriptivism

The sentence ‘Mont Blanc is over 4000 m high’ does not express the same thought as the sentence ‘The highest mountain in Europe is over 4000 m high’, although the proper name ‘Mont Blanc’ designates the same mountain as the expression ‘the highest mountain in Europe’. Frege 1906, 208 [192] ‘Copernicus’ and ‘the author of the heliocentric view of the planetary system’ designate the same man, but have different senses; for the sentence[s]‌‘Copernicus is Copernicus’ and ‘Copernicus is the author of the heliocentric view of the planetary system’ do not express the same thought. Frege 1914, 243 [225] In each example a proper name and a co-​referential definite description are given, and sameness of sense is denied, for those specific examples. These examples do not rule out the possibility that Frege might think that there can be, or even that there must be, for each example, another definite description that does give the sense of the proper name. Nevertheless, Frege never gives us a clear indication that he does think this. Indeed, judging by these later works, this looks very unlikely to have been his considered view, whatever his view was in 1892. The upshot of this incursion into the history of Frege’s discussion of specific proper name/​definite description co-​referential pairings is as follows. Frege (1892) provides the main evidence that Frege was a sense descriptivist and even there it is in a footnote the credentials of which, as a genuine endorsement of this kind of descriptivism, are dubious. In other works Frege makes remarks that, even if not strictly inconsistent with sense descriptivism, make it look very unlikely that Frege was a sense descriptivist. So, even if, in the footnote, Frege did, whether inadvertently or otherwise, endorse this kind of descriptivism, there appears to be quite a strong case for taking this to have been a temporary lapse.10 In order to assess whether Frege was a sense descriptivist, it is pertinent to look not only at Frege’s discussions of specific co-​referential proper name/​definite description pairings, but also at general statements he makes about names and their senses. Frege writes: It may perhaps be granted that every grammatically well-​formed expression figuring as a proper name always has a sense. But this is not to say that to the sense there also corresponds a reference. Frege 1892, 28 The examples that Frege then gives are of a probably empty and a certainly empty definite description: respectively, ‘the celestial body most distant from the earth’ and ‘the least rapidly convergent series’. It seems that by ‘grammatically well-​formed expression’, Frege did not intend to include proper names, like ‘Paris’,‘New York’, or ‘Robert Louis Stevenson’, that (even when they are composed from more than one word, and are thus syntactic compounds) are not compound expressions in the sense, for which we now reserve the term, that they possess a function-​argument structure. For Frege, compound names have senses that derive from the senses of their constituent expressions. By contrast, the sense of a simple name has to be conferred upon it (e.g., by baptism or by definition). Frege (1892, 31) sets out some of the terminology pertinent to his distinction between sense and reference: A proper name (Eigenname) (word, sign, sign combination, expression) expresses its sense, stands for (bedeutet) or designates its reference. By employing a sign we express its sense and designate its reference.

45

Ian H. Dunbar and Stephen K. McLeod

For Frege it is of secondary importance whether an Eigenname is (in his sense) simple, like ‘Paris’ and ‘Robert Louis Stevenson’, or compound, like ‘the capital of France’. What really matters is that both sorts of expression designate objects (e.g., Frege 1892, 40–​44). While Frege includes definite descriptions among the different sorts of syntactic units that can be Eigennamen, he does not privilege definite descriptions over simple names. He is content to think that Eigennamen exhibit this syntactic diversity, without ever suggesting that simple names are always abbreviations of, or disguised, compound names. Frege (1892, 41) remarks: A logically perfect language (Begriffsschrift) should satisfy the conditions, that every expression grammatically well constructed as a proper name out of signs already introduced shall in fact designate an object, and that no new sign shall be introduced as a proper name without being secured a reference. Those proper names that are ‘constructed … out of signs already introduced’ are compound names. As we already mentioned, these inherit their senses from those of the expressions from which they are composed. What about the case of those proper names that are neither compound nor empty? We suggest that, for Frege, there are two ways in which their senses are acquired. The first way is by functioning as an abbreviation for a compound name the sense of which derives from that of its constituent expressions. Accordingly, a simple name that is an abbreviation for a compound name is synonymous with, and also derives its sense from the constituent expressions of, the compound name. The second way is simply by being introduced to refer to an object, without functioning as a form of shorthand for a compound name. Any symbol that is used, in this way, to refer thereby has a sense, simply in virtue of that use. It becomes an expression that neither admits of a definition nor is synonymous with a definite description. For Frege, we suggest, the two ways in which proper names in logically perfect languages come to designate, and thus to have their senses, are the same two ways in which non-​empty proper names in natural languages come to designate, and thus to have their senses. Some of Frege’s remarks about definitions in mathematics might help to clarify his views about how proper names acquire senses. We have … to distinguish two quite different cases: (1) We construct a sense out of its constituents and introduce an entirely new sign to express this sense. This may be called a ‘constructive definition’, but we prefer to call it a ‘definition’ tout court. (2) We have a simple sign with a long established use. We believe that we can give a logical analysis of its sense, obtaining a complex expression which in our opinion has the same sense. We can only allow something as a constituent of a complex expression if it has a sense we recognise. The sense of the complex expression must be yielded by the way in which it is put together. That it agrees with the sense of the long established simple sign is not a matter for arbitrary stipulation, but can only be recognised by an immediate insight. No doubt we speak of a definition in this case too. It might be called an ‘analytic definition’ to distinguish it from the first case. But it is better to eschew the word ‘definition’ altogether in this case, because what we should here like to call a definition is really to be regarded as an axiom. In this second case there remains no room for an arbitrary stipulation, because the simple sign already has a sense. Only a sign which as yet has no sense can have a sense arbitrarily assigned to it. So we shall stick to our original way of speaking and call only a constructive definition a definition. According to that a definition is an arbitrary stipulation which confers a sense on a simple sign which previously had none.

46

Fregean Descriptivism

This sense has, of course, to be expressed by a complex sign whose sense results from the way it is put together. Frege 1914, 227 [210] While the passage concerns the use of definitions in mathematics, it might indicate what Frege thought about how compound names, including definite descriptions, can give the senses of some simple names. A  proper name is a special case of a ‘simple sign’ (einfaches Zeichen) and a definite description is a special case of a ‘complex expression’ (zusammengesetzer Ausdruck). When the relationship between a simple proper name and an associated definite description is one of case (1), then we have a restricted, but full-​blown, version of meaning descriptivism. The simple name is an abbreviation for the definite description. The simple name’s sense is the same as that which, by virtue of the latter’s composition, the definite description expresses. Moreover, the definite description gives the sense of the simple name. In case (2), however, the simple name is not given its sense by the definite description, for the simple name already had its sense, and that sense was grasped by competent users of the name, before the simple name came to be related, in cognition, to the definite description.The simple name’s sense had already been conferred upon it in some way, independently of the use of the definite description. The definite description, then, does not give, but merely articulates, that sense. We deploy the compound name as a means of analysing what that pre-​existing sense is. We do this, thinks Frege (1914, 226 [209]), so that we can produce proofs that were not possible while the sense of the name was not explicitly articulated. The correctness of such an analysis ‘can only be recognised by immediate insight’ (Frege 1914, 227 [210]), rather than proved, because the analysis is precisely what is needed if we want to carry out proofs. The sense of the simple name is grasped, because the name has already been in proper use, prior to the attempt having been made to articulate that sense by appeal to the sense of the definite description. In judging the value of this attempt at articulation, we must exercise our prior, unarticulated grasp of the simple name’s sense. Frege has here identified two types of proper name (as special cases of simple expressions). There are those that start off without a sense and are given one by stipulation, using a compound name that inherits its sense from its components. For these proper names, both name descriptivism and sense descriptivism are true. A proper name that is of the other variety already has a sense, and descriptions are used only when we try to articulate what this sense might be. In describing this second type, Frege does, however, assume that our grasp of the prior, tacit, sense is accurate enough for us to know when we have achieved an accurate analysis. When we succeed in giving the prior sense a post hoc definition, we obtain the result that the definite description expresses the sense of, and is therefore synonymous with, the original name. The asymmetry that characterizes sense descriptivism, however, is not present. In distinguishing between two forms of ‘definition’ in mathematics, Frege does not intend to suggest, either for mathematical or for more general language, that, for each simple expression, it is pertinent to one of the two cases. On Frege’s view there are simple signs that are indefinable. Indeed, we suggest that Frege takes those simple names that are ordinary proper names in natural languages, and concerning which we think he has been mischaracterized as a meaning descriptivist, to be indefinable.We have already encountered some of his discussion of the following examples: ‘Aristotle’, ‘Mont Blanc’, ‘Copernicus’ and ‘Columbus’. In the last three cases, he produces ‘famous deeds’-​style definite descriptions that he states do not give the senses of the corresponding names. In the Aristotle footnote he gives two definite descriptions that reflect what he calls differing opinions as to the sense of the name. Frege then says in the footnote that this might be good enough in normal circumstances, but that it would be better if it could be avoided.11 Frege wrote relatively little about the senses of those names that feature in everyday discourse. What he does write about this topic mostly concerns the question of whether each proper name 47

Ian H. Dunbar and Stephen K. McLeod

does, or should, have only one sense. There is only meagre and indecisive evidence that he thought that the sense of each such proper name is given, or ought to be given, by a definite description.

5  Frege and Reference-​Fixing Descriptivism Kripke (1980) attributes to Frege the view that the referent of any given proper name is fixed by its sense and that the sense, in turn, is given by a definite description. We will shortly argue that it is because Kripke attributes sense descriptivism, along with the view that sense fixes reference, to Frege that Kripke thinks that Frege is a descriptivist about reference fixing. Our remaining purpose in this section, other than to give the aforementioned argument, is to clarify, in so far as it bears on Frege, the dialectical territory in which Kripke’s arguments against reference-​fixing descriptivism work. The slogan ‘sense determines reference’ is now, apparently thanks to Dummett’s coinage and influence, part of orthodox Frege exegesis. Dummett (1981b, 249; 1991, 87) distinguishes between ‘weak’ and ‘strong’ interpretations of it.12 Weakly interpreted, it just means that, for any two referring expressions, sameness of sense entails sameness of reference. Thus interpreted, it is logically independent of sense descriptivism: i.e., neither entails the other. The weak interpretation can be distinguished from the strong one via the thesis that an expression’s sense is a way in which its referent is determined.13 While the strong interpretation entails this thesis, the weak interpretation does not. On the strong interpretation, the sense of an expression constitutes a manner in which something is determined as being its referent, or, to put it in terms of our understanding of the expression, that is, of what is involved in grasping its sense, to grasp the sense is to apprehend the condition for something to be its referent. Dummett 1981b, 249 This quotation is ambiguous: it does not make it clear whether it is the question of how an expression comes to refer that is of concern, or rather that of how it comes to have the referent that it has. Despite this, and that the quotation leaves it quite obscure as to what it is to determine something as being an expression’s referent, we take it that it at least illustrates the crux of the difference between the weak and the strong interpretations. Interpreted weakly, ‘sense determines reference’ is, we think, a misleading phrase: for, unlike on the strong interpretation, it is not an allusion to the questions of how it is that a referring expression comes to refer, or comes to have the referent that it has. Interpreted weakly, the slogan is a truism about Frege’s philosophy. When interpreted strongly (as putative Frege exegesis) it suffers, in our opinion, from the flaw that there is little in Frege’s works to suggest that he had a general interest in these questions. In the quotation from Kripke (1980, 59) near the start of section 2 above, Kripke uses ‘fix its reference’ in a way that suggests that he intends talk of sense determining reference and talk of sense fixing reference to be two ways of saying the same thing. If the sense is, as Kripke understands reference fixing, to fix the reference then, presumably, the sense must, in accordance with the strong interpretation of ‘sense determines references’, be identical to, or perform the role of being, a way in which the reference is determined. Kripke (1980, 53–​60) takes reference-​fixing descriptivism to be distinct from, consistent with and weaker than meaning descriptivism. Kripke (1980, 57) characterizes reference-​fixing descriptions in the following terms. A reference-​fixing description serves to pick out the object to which a speaker intends, via the speaker’s use of a particular proper name, to refer. This relates reference fixing specifically to our referential intentions, rather than to the enterprise of linguistic understanding. In so 48

Fregean Descriptivism

doing, it accords with Dummett’s first characterization of the strong interpretation, which appeals to reference determination without appealing to understanding, but not with Dummett’s (apparently inequivalent, but see Dummett (1981b, 549)) second characterization, which does the reverse. Indeed, in order to understand a referring expression, rather than to introduce it anew into the language, the expression’s reference must already have been fixed (independently of speakers’ subsequent referential intentions). In order to assist with the understanding of some of the material that follows, we mark out, further, the following four theses as distinct. Reference fixing: for any proper name, something other than a speaker’s mere use of it plays the role of picking out the object to which the speaker intends, in using the name, to refer. Sense fixes reference: for any proper name, its sense plays the reference-​fixing role. Reference-​fixing descriptivism: for any proper name, either an associated definite description or the sense of that definite description plays the reference-​fixing role. Reference-​fixing pure descriptivism: for any proper name, it is an associated definite description itself (rather than that description’s sense) that plays the reference-​fixing role. Reference-​fixing sense descriptivism: reference fixing, sense fixes reference and sense descriptivism all hold; for any proper name, the sense of an associated definite description gives the name’s sense and plays the reference-​fixing role. Reference-​fixing sense descriptivism and reference-​fixing pure descriptivism are contrary views. The form of reference-​fixing descriptivism that Kripke (1980, 59) attributes to Frege specifically involves the notion of sense, and it consists of the following two theses (i) the reference of a proper name is fixed by its sense; (ii) the sense is given by a definite description (i.e., sense descriptivism). These two theses together entail, but are not entailed by, reference-​fixing descriptivism as we have formulated that thesis (largely to the letter of Kripke’s account). According to Kripke’s Frege, it is thanks to a definite description with which a proper name is co-​referential that we are able, on an occasion of use of that proper name, thereby to pick out the object that is the name’s referent. This is because the definite description gives the name’s sense and the sense is that which fixes the name’s reference. It is not because the definite description itself, rather than its sense, plays the reference-​fixing role. In relation to reference fixing, descriptivism and Frege, the dialectic of Kripke (1980) can, we suggest, charitably be reconstructed as follows. Frege was a reference-​fixing sense descriptivist and a meaning descriptivist. Meaning descriptivism falls to the following objection (Kripke 1980, 3–​ 15, 48–​49, 74–​76). While any given ordinary proper name is a rigid designator, referring to the same individual under all counterfactual suppositions, this is not true of any co-​referential definite description that is associated with the name. Accordingly, substitution, in a sentence, of the name by a co-​referential definite description can result in a sentence with a different modal status to that of the original sentence. Since this could not happen if the name and the description were genuinely synonymous, meaning descriptivism is false. That concludes the objection. Now in holding that senses are both the meanings of, and fix the references of, proper names, Frege attributed to senses, construed as being given by definite descriptions, two roles between which a sharp distinction must be observed. Decoupling reference-​fixing descriptivism from meaning descriptivism enables the independent evaluation of reference-​fixing descriptivism. So as to observe the distinction that is violated by Frege’s notion of sense, let us now not think of Fregean senses, construed as the elements of linguistic meaning that explain synonymy and the lack thereof, as the putative reference fixers. Rather, let us take the definite descriptions (semantically, rather than merely syntactically, individuated) that Frege thinks give those senses as themselves being the putative reference fixers. That is, let us consider the thesis, which does not invoke Frege’s notion of sense, that for any proper name, an associated definite description itself plays the reference-​fixing role. For various reasons, even this thesis must 49

Ian H. Dunbar and Stephen K. McLeod

be rejected (Kripke 1980, Lecture II). These reasons include that a speaker who associates a definite description that is false of the name’s bearer can still succeed, when using the name, in referring to the name’s bearer (Kripke 1980, 82–​89). Thus interpreted, Kripke moves from arguing against a position that we think Frege did not hold (namely meaning descriptivism) to one that Kripke rightly thinks that Frege did not hold (namely reference-​fixing pure descriptivism). Whether Frege held that sense fixes reference, at least as understood by Kripke, is another question and one that we do not have room to address. We end merely with a reminder that Kripke’s interpretation of Frege on this point aligns with Dummett’s reading of Frege as a supporter of ‘sense determines reference’, strongly interpreted.

6  Conclusion We have taken Fregean descriptivism to be a view, attributed to Frege by Kripke, that incorporates both meaning descriptivism and reference-​fixing descriptivism. For each of these views, in turn, we have broken it down into constituent theses between which Kripke himself did not clearly distinguish. In the case of meaning descriptivism, among its constituent theses are name descriptivism, according to which, for any proper name, it is an abbreviated or disguised form of a definite description, and sense descriptivism, according to which the senses of names are given by definite descriptions. In Frege’s writings, the Aristotle footnote is the main evidence that Frege was a sense descriptivist. If one supposes that the footnote commits Frege to sense descriptivism, then one must take it that Frege denies that a proper name is synonymous with any definite description that gives its sense, for the footnote is meant to suggest that one name can (presumably correctly) have associated with it many definite descriptions that differ in sense. Alternatively, one might follow May (2006) in holding that the Aristotle footnote is not about sense itself, but rather about opinions about sense. Elsewhere Frege appears to express the belief that each name has but one sense. Frege (1914), in discussing definitions, distinguishes between cases where a compound name assigns a sense to a proper name that had no prior sense (which is sense descriptivism for a limited class of expressions), and those where the name already (somehow) had a sense, and definite descriptions are used in attempts to articulate what that sense is. In these latter cases, Frege leaves it open as to how this original sense was established. We have argued that Kripke’s interpretation of Frege as a reference-​fixing descriptivist stems from his ascription of two other views, each logically weaker than reference-​fixing descriptivism itself, to Frege. These are sense descriptivism and the view that sense fixes reference. The meaning descriptivism and the reference-​fixing descriptivism of Kripke’s Frege have sense descriptivism as their common, logically weaker, core. Since we do not concur with Kripke’s view that Frege was a sense descriptivist, we do not share his reasons for thinking that Frege was a reference-​fixing descriptivist.

Acknowledgements We thank Michael Hauskeller, Daniel Hill and the editors for their comments.

Notes 1 We follow this convention, including in quoted passages from Frege’s works. For works published in Frege’s lifetime, references are to the pages of the original publications. For posthumously published works, references are to the German edition followed, in square brackets, by the English edition. 2 Kripke uses ‘proper name’ in a manner intended to exclude definite descriptions themselves. Frege’s ‘Eigenname’ (‘proper name’), on the other hand, includes them.

50

Fregean Descriptivism 3 What Kripke (1980, 31) calls ‘the cluster concept theory’ involves a version of meaning descriptivism that requires a formulation different to that, upon which we will shortly expand, just given. 4 Kripke does not name these two types of descriptivism. We adopt shortened forms of the names used by Macià (1998, 447). 5 We take it, and we think that we are following Kripke in this, that reference-​fixing descriptivism does not necessarily involve the notion of sense. It is one thing to claim that a definite description fixes a name’s reference, another to say that it is the sense of the definite description that does so, and yet another to say that it is the name’s sense that does so. In section 5, we explain the relevance of these nuances. 6 Kripke (1980, 48–​49) introduces the notion of rigid designation and the view that proper names are rigid designators. 7 To pursue the matter in greater depth, see May (2006) and the works cited therein. 8 Logically perfect languages are artificial, and they are to be constructed so that each symbol within them has a clear and determinate sense and so that there are no cases in which two token symbols (individuated purely syntactically) of the same type symbol are token expressions (i.e., symbols with senses) of different expression types (in that their senses differ). 9 On May’s exegesis, Frege holds that this is so even when two competent users of a name have differing opinions as to what the name’s sense is. See May (2006, 128). May therefore sees Frege as having held that one might grasp a sense without being able properly to articulate, or to specify, exactly what it is that one has grasped. 10 We agree with Dummett (1981a, 97–​98, 110–​111) that Frege’s sense theory does not entail sense descriptivism. Beyond his brief comments on specific co-​referential name/​definite description pairings, Frege frequently invokes the notion of sense, including when he writes that empty names have senses. Such passages do not show, and we do not think that they strongly suggest, that he was a sense descriptivist. 11 There is a related discussion, controversial among Frege scholars, about Dr Gustav Lauben and expressions that refer to him, in Frege (1918, 65–​66). See May (2006, 121–​126). 12 If Dummett was indeed the slogan’s originator then it is odd that he distinguishes between two interpretations of it. From Dummett (1967, 105) onwards the slogan repeatedly arises in Dummett’s writings. (In 1967, Dummett uses ‘meaning’, rather than ‘reference’, where Frege would use ‘Bedeutung’.) ‘Sense, Frege says, determines reference’ writes Dummett (1991, 87). Dummett thereby attributes to Frege a slogan that, thus formulated, Frege never enunciates and an ambiguity that Frege himself does not bequeath to us.The slogan also occurs in numerous other sources (that we do not have space to cite) that have appeared since the 1970s. 13 The chances of the strong interpretation’s being a truism about Frege’s philosophy would be greater were it to be Frege’s view that the senses of non-​empty proper names are modes of presentation of their referents. In fact, Frege (1892, 26–​27) writes, albeit more obscurely, that these senses contain (‘enthalten’) modes of presentation of their referents. May (2006, 131) makes play of this distinction.

References Dummett, M. A. E. (1967) ‘Frege, Gottlob’ in P. Edwards (ed.), Encyclopedia of Philosophy, vol. III, pp. 225–​237 (New  York:  Macmillan), repr. (as ‘Frege’s Philosophy’) in M. A.  E. Dummett, Truth and Other Enigmas (London: Duckworth, 1978), pp. 87–​115. Page references are to the latter. Dummett, M. A. E. (1981a) Frege: Philosophy of Language (2nd ed., London: Duckworth). Dummett, M. A. E. (1981b) The Interpretation of Frege’s Philosophy (London: Duckworth). Dummett, M. A. E. (1991) Frege and Other Philosophers (Oxford: Oxford University Press). Frege, G. (1892) ‘Über Sinn und Bedeutung’, Zeitschrift für Philosophie und philosophische Kritik 100: 25–​50, trans. M. Black as ‘On Sense and Reference’ in P. T. Geach and M. Black (eds), Translations from the Philosophical Writings of Gottlob Frege (3rd ed., Oxford: Blackwell, 1980), pp. 56–​78. Frege, G. (1906) ‘Einleitung in die Logik’, unpublished; reproduced in H. Hermes, F. Kambartel and F. Kaulbach (eds), Nachgelassene Schriften (Hamburg:  Felix Meiner, 1969), pp. 201–​212, trans. P. Long and R. White as ‘Introduction to Logic’ in H. Hermes, F. Kambartel and F. Kaulbach (eds.), Posthumous Writings (Chicago, IL: University of Chicago Press, 1979), pp. 185–​196. Frege, G. (1914) ‘Logik in der Mathematik’, unpublished; reproduced in H. Hermes, F. Kambartel and F. Kaulbach (eds.), Nachgelassene Schriften (Hamburg: Felix Meiner, 1969), pp. 219–​270, trans. P. Long and R. White as ‘Logic in Mathematics’ in H. Hermes, F. Kambartel and F. Kaulbach (eds.), Posthumous Writings (Chicago, IL: University of Chicago Press, 1979), pp. 203–​250. Frege, G. (1918) ‘Der Gedanke. Eine logische Untersuchung’, Beiträge zur Philosophie des deutschen Idealismus I (1918–​1919): 58–​77, trans. A. Quinton and M. Quinton as ‘The Thought: A Logical Inquiry’ in E. D. Klemke (ed.), Essays on Frege (Urbana, IL: University of Illinois Press, 1968), pp. 507–​535.

51

Ian H. Dunbar and Stephen K. McLeod Kripke, S. A. (1980) Naming and Necessity (Oxford: Blackwell). Macià, J. (1998) ‘Does Naming and Necessity Refute Descriptivism?’, Theoria (Spain) 13: 445–​476. Maunu, A. (2015) ‘Frege and the Description Theory:  An Attempt at Rehabilitation’, Grazer Philosophische Studien 92: 109–​116. May, R. (2006) ‘The Invariance of Sense’, Journal of Philosophy 103: 111–​144. Mill, J. S. (1882) A System of Logic (8th ed., New  York:  Harper and Brothers). Project Gutenberg eBook 27942 (2009).

52

4 THE REFERENTIAL–​ATTRIBUTIVE DISTINCTION Anne Bezuidenhout

1  Introduction The main goal of this chapter is to argue against the idea that the referential-​attributive distinction as applied to definite descriptions is a semantic distinction regarding the type of proposition expressed by sentences containing such descriptions. In section 2, I argue that Donnellan’s referential-​attributive distinction can be accommodated within a variety of semantic frameworks, suggesting that the existence of this distinction by itself does not tell us anything about the correct semantics for definite descriptions. Hence, the significance of this distinction, if any, must lie elsewhere. In section 3, I argue that the significance is in fact an epistemic one, with the difference between the uses representing differences in epistemic access to the (contextually unique) entity denoted by the description. In pure referential uses, the speaker is directly acquainted with the denoted entity. In attributive uses, speakers denote an entity but they additionally convey a general proposition to the effect that they do not have, or that it is not relevant for hearers to have, epistemic access to a situation within which a unique entity can be directly identified. Furthermore, such attributive uses will be felicitous only if the interlocutors are hooked into an appropriate communicative network that terminates with an information source who is directly acquainted with the denoted entity. If there is no such terminus, reference will fail and no truth-​evaluable claim is made (whether the interlocutors are aware of this or not). This has two corollaries, namely that the referential-​attributive distinction is in fact a continuum and that there will be uses towards the referential end of the spectrum that tolerate the possibility of misdescription. In section 4, building on this reconstrual of the referential-​attributive distinction as having merely epistemic rather than semantic import, I ask whether (in the absence of uses of phrases such a ‘whoever/​whatever that is/​may be’) there is any grammatical reflex of this distinction. I conclude that there is not, at least in English. This can be taken as support for the claim in section 3 that the distinction has epistemic rather than sematic significance (given that semantic contrasts are often, even if not invariably, grammatically marked). Finally, in sections 5 and 6, I survey some recent work in linguistics and experimental pragmatics that touches on Donnellan’s distinction. Section 5 briefly surveys the semantic accounts of argumental (as opposed to predicative) uses of definite descriptions offered by Coppock and Beaver (2015) and Elbourne (2013). In both these frameworks, the referential-​attributive distinction is one that is made within the class of uses that Coppock and Beaver class as definite uses; that is, uses that denote entities, if they denote at all. My own reconstrual of the referential-​attributive distinction as having merely epistemic import is at least 53

Anne Bezuidenhout

compatible with the idea that the distinction is one drawn within the class of definite uses that denote a (contextually unique) entity. However, my claim is that what is expressed in both cases is a singular (or object-​dependent) proposition. What is different about the attributive use is that the speaker also conveys a general proposition quantifying over situations in which a unique entity is identifiable. In section 6, I briefly describe some experimental work by Rostworowski and Pietrulewicz (2019) that supports my claim that descriptions can be felicitous even in cases where the entity denoted by the description fails to satisfy the description. Section 7 offers some conclusions.

2  Classic Treatments of the Referential-​Attributive Distinction Keith Donnellan (1966) introduced the referential-​attributive distinction in the context of the debate between Russellians and Strawsonians as to the correct semantics for definite descriptions. Russell (1905), in opposition to Meinong (1904/​1960) and Frege (1892/​1960), proposed to analyze sentences such as (1) below as expressing existential generalizations, along the lines indicated in (2) (in which a proper treatment of tense and aspect is suppressed): (1) The father of Charles II was executed. (2) ∃x [FatheredCharles II (x) ∧ ∀y (FatheredCharles II (y) → y=x) ∧ WasExecuted  (x)]

Russell’s point was that definite descriptions such as ‘the father of Charles II’ are not singular referring expressions whose function is to pick out an individual that the predicate then says something about. In fact, Russell thought that descriptions do not have meaning in isolation. As the analysis in (2) shows, there is no expression purporting to refer to Charles II’s father. There are just quantifiers, variables, predicates, and clausal connectives. (We could of course have broken down the one-​place predicate ‘FatheredCharles(x)’ into a two-​place relation ‘Fathered(x,y)’, and then we would have needed to use the referring expression ‘c’, which is a name for Charles II, in the second place of this relation. Nevertheless, it would still be true that there is no term/​name referring to Charles II’s father). Two further points about this Russellian analysis should be noted. Firstly, (2) logically entails that someone fathered Charles II (that is, a father of Charles II existed) and that Charles II had only one father. Secondly, given the analysis in (2), it should be clear that there are several different sites for attachment of a negation operator, meaning that it will be ambiguous as to what is asserted by either of the negative variants of (1) below: (3) It is not the case that the father of Charles II was executed. (4) The father of Charles II was not executed. The negation could be construed as taking widest scope, meaning that there was no uniquely existing father of Charles II who was executed (either because Charles II had no father or he had more than one father or his unique father was not executed). Alternatively, the negation could take narrow scope and attach to the predicative clause, meaning that there was a uniquely existing father of Charles II but he was not executed. Strawson (1950, 1952), on the other hand, embraced the analysis of (1) according to which it is a subject-​predicate sentence. He argued that a speaker could use the definite description ‘the father of Charles II’ as an expression to refer to Charles II’s father and could use the predicate ‘was executed’ to say something about that individual.The net result would be an assertion that the father of Charles II was executed. The felicitous use of a definite description, however, requires the satisfaction of certain presuppositions. In the case of (1), the presuppositions are that Charles II had one and only one father. If these presuppositions are not satisfied, the assertion of (1) would fail and there would be no 54

The Referential-Attributive Distinction

truth-​evaluable proposition expressed.The attempted assertion would result in a truth-​value gap.This is a prediction different from the one made by Russell’s analysis, which predicts that if Charles II had no uniquely existing father, then (1) would be false. Elbourne (2013, 46) quotes Strawson (1950, 332) in a passage in which Strawson raises the issue of incomplete descriptions by noting that normal uses of ‘the table’, in utterances such as ‘The table is covered in books’, do not assume that there is one and only one table in existence. Elbourne takes this as evidence that, although Strawson thinks definite descriptions have existence presuppositions, they do not have uniqueness presuppositions.The attribution of this view to Strawson is unwarranted.The fact that Strawson thinks that ‘the table’ can be felicitously used despite the myriad of tables in the universe does not mean that he thinks it can be felicitously used in a local context/​situation in which two or more tables are all equally salient or “topical”.Thus, I will assume that Strawson holds that ‘the F’ presupposes the unique existence of an F in the appropriate local conversational context. Strawson (1950, 330) suggests that presupposition failure will bring conversational proceedings to a halt, in the sense that hearers would be inclined to challenge false presuppositions rather than focusing on speakers’ attempted assertions. As he puts it in Strawson (1964, 106), the assertive enterprise is “wrecked” by a false presupposition. However, subsequently, Strawson confronts the issue as to whether the failure of the existence and uniqueness presuppositions of a definite description always leads to catastrophe, e.g., when someone uses a description of the form ‘the F’ in a situation in which there is no uniquely existing F. He discusses cases such as (5) below (see Strawson 1964, 112): (5) The Exhibition was visited by the King of France today. He agrees that intuitively we think that (5) is false rather than truth-​valueless. His first pass at an answer to this problem is to appeal to the topic-​comment structure of (5) and note that the definite description is not the topic of the sentence. Rather it has been “absorbed” (1964, 112) into the predicate expression and is part of a larger expression that is used to comment on the topic, which in this case is the Exhibition. As I said, this was Strawson’s first pass at a solution. His views are in fact more nuanced than this, as I argue in Bezuidenhout (2016). I will say no more about this particular issue of whether or not there can be cases of non-​catastrophic presupposition failure. It has been discussed in some detail by Reinhart (1981), Lasersohn (1993), Atlas (2004), and von Fintel (2004). When Donnellan (1966) entered the fray, the debate he engaged with was whether sentences of the form of (1) are subject-​predicate sentences that express singular propositions or whether they express general propositions, as the analysis in (2) suggests. Donnellan was not much concerned with whether existence and uniqueness are entailments or presuppositions, nor with other issues that Strawson had raised, such as the problem of incomplete descriptions. What Donnellan claimed was that sentence (6) below has two distinct kinds of uses. (Donnellan actually used the example ‘Smith’s murderer is insane’. However, Coppock and Beaver (2015, 416–​423) have shown that possessives pattern differently from definite descriptions, so in order to steer clear of this issue, I will use (6) to illustrate the referential-​attributive distinction): (6) The murderer is insane. In one situation, we are to imagine that a detective has just stumbled upon the savagely mutilated body of Smith, and on the basis of the state of the body, the detective surmises that only a lunatic could have inflicted such wounds, and thus he utters (6) to his associates, meaning to convey that Smith’s murderer, whoever that turns out to be, is insane. This is the so-​called attributive use. In another situation, Anne and Bob enter a courtroom where Jones is on trial for Smith’s murder. Jones, who is standing in the dock, is behaving oddly; perhaps he is twitching and drooling and rolling his eyes around. Anne doesn’t know Jones by name, but knows that the man in the dock is the person accused of murdering Smith. Anne turns to Bob and utters (6), gesturing towards Jones as she utters 55

Anne Bezuidenhout

this. Donnellan’s intuition is that in this situation Anne has referred to Jones and said of him that he is insane. This is the so-​called referential use. Donnellan concludes that neither Russellians nor Strawsonians can account for both types of use of ‘the murderer’. Russellians can handle the attributive reading and Strawsonians the referential one, but not vice versa. Donnellan’s alternative account was never spelled out to the satisfaction of his immediate contemporaries. He made it clear that he was not arguing for a semantic ambiguity account, according to which ‘the murderer’ is both a singular referring expression and a quantificational expression and context must determine which one the speaker intends to be using. Donnellan did suggest that we could perhaps talk of a “pragmatic ambiguity” (1966, 297). However, others at the time did not understand what this would amount to and were thus not inclined to develop this line of thinking. For many—​Neale (1990) would be a representative example—​the issue was decisively settled in favor of the Russellian analysis when Kripke (1977) argued that, in a language in which it is stipulated that ‘the F’ is to be analyzed à la Russell, Donnellan’s two uses could still arise. The referential use, according to which a singular proposition is expressed, would be construed as conversationally implicated, via Gricean reasoning, by the saying of something general in a context where a more determinate interpretation would be relevant. In other words, Kripke was arguing that one can maintain a univocal Russellian semantics for ‘the F’ (i.e., regard ‘the F’ as an incomplete symbol that only has meaning in the context of a sentence, which in turn has the logical form of an existential generalization), and wed this semantic account to a Gricean pragmatics to get the desired result that ‘The F is G’ can be used to communicate both a general and a singular proposition. The account that treats referential interpretations as Gricean conversational implicatures requires that referential interpretations be cancelable, since cancelability is a mark of a conversational implicature. That is, one ought to be able to (implicitly or explicitly) deny the implicated meaning without semantic anomaly. For example, it should be possible for Anne in the courtroom scenario imagined above to continue her utterance of (6) with (7): (7) I do not mean to suggest he [pointing again to Jones] is insane. I know he is innocent of the murder charges. The actual murderer, whoever that turns out to be, is insane, given what he did to Smith’s body. In the conversational context as described, Anne’s utterance of (7) would be at odds with her prior utterance, as all signs available to the hearer, Bob, would have been that Anne had Jones and his erratic behavior in mind when she uttered (6). Inasmuch as (7) is a legitimate follow-​up to (6), it could be treated as a correction of a misleading utterance rather than as an implicature cancelation. Whether (7) really is a case of implicature cancelation or simply a case of utterance clarification is something I will not pursue here. See also Neale (2004), who, in response to criticisms of the Gricean implicature account, develops an alternative Russellian account to explain referential uses. Referential uses turn out, he thinks, to be special cases of “improper” descriptions, which require contextual quantifier restriction. The result of such contextual restrictions is that an object-​dependent proposition is expressed in the case of referential uses of descriptions, since the restriction clause will contain a singular referring expression (such as a name or a demonstrative expression). More importantly for present purposes, note that by using Kripke-​style reasoning, one could as well argue that ‘the F’ has a univocal referential meaning and that the attributive interpretations are the ones that are pragmatically derived. One might attempt to defend this univocal referentialist semantic account by appealing to the fact that referring expressions more generally, including pronouns and indexicals, can have attributive readings and yet this does not incline us to say that pronouns and indexicals are not, after all, referring expressions. For example, coming across a large footprint in the sand, one could utter (8) intending to convey something along 56

The Referential-Attributive Distinction

the lines of (9). Similarly, in a context in which you are hiding in a warehouse from some armed insurgents and have just opened the door to let one of your comrades in, the comrade could utter (10) intending to convey something like (11): (8) (9) (10) (11)

He [gesturing at a large footprint in the sand] must be a giant. Whatever male entity left the footprint in the sand must be a giant. You shouldn’t have just opened the door like that. I might have been an insurgent. You shouldn’t have just opened the door like that. It is possible that whoever was at the door was an insurgent.

Arguably, definite descriptions can be used in some cases to focus on a salient property of an entity rather than the entity itself, just as in the attributive uses of pronouns and indexicals above. Someone wishing to argue for a univocal referential semantics could argue that the referential use of (6) is basic and that the attributive use is pragmatically determined in the same way that the attributive readings of (8) and (10) are. In the attributive use of (6), it is the property of being Smith’s murderer that the speaker is focusing on rather than on the entity who has the property, just as in (8) it is the property of being the male who left the footprint that matters and in (10) it is the property of being the person at the door that is important. The notion of attributive uses of referring expressions is discussed by Nunberg (1993), Rouchota (1992), and Bezuidenhout (1997). See also Nunberg (2004) for arguments against grouping definite descriptions with other referring expressions in this way. Devitt (2004) and Reimer (1998) have responded to Kripke’s arguments in favor of a univocal Russellian semantic analysis of definite descriptions by pointing to the fact that there are robust conversational conventions that favor referential interpretations of definite descriptions. Devitt concludes that ‘the F’ is genuinely semantically ambiguous. Reimer has instead argued that there may be a univocal semantics for ‘the F’ which is underspecified for either a referential or an attributive reading. This appeal to an underspecified meaning is also defended by Recanati (1993) and Bezuidenhout (1997). The important take-​ away from the discussion above is that it is possible to accommodate Donnellan’s referential-​ attributive distinction within a framework that endorses either a univocal Russellian semantics, a univocal referentialist semantics, a semantic ambiguity account, or an underspecified semantic meaning account. This shows that the existence of this distinction by itself does not tell us anything about the correct semantics for definite descriptions. Thus, the significance of this distinction, if any, must lie elsewhere. In the following section, I suggest that the referential-​ attributive distinction does not, after all, amount to the difference between expressing a singular (or object-​dependent) proposition versus a general proposition.1 Rather, it concerns differences in a speaker’s epistemic access to the referent of ‘the F’, where ‘the F’ is always to be regarded as a singular referring expression (or so I will argue).

3  Epistemic Access and the Referential-​Attributive Contrast In this section I question the significance of the test that is frequently invoked to identify the presence of an attributive reading of ‘The F is G’. In spelling out the attributive use of ‘the murderer’ in (6) above, I used the phrase ‘whoever that turns out to be’. Adding this phrase is widely taken as a means of signaling that the speaker intended to convey a general proposition rather than a singular or object-​dependent one. However, one could argue that a speaker who uses ‘the murderer’ attributively is not expressing the general proposition represented in (12) below, but is instead expressing the singular proposition represented in (13), whose subject is a determinate individual. This individual is not currently known (by acquaintance) to the speaker, although the individual is in principle discoverable via following a trail of evidence (possibly one that would require the resources of multiple investigators and one that stretches both forward and backwards in time): 57

Anne Bezuidenhout

(12) Some unique person is a murderer of Smith and is insane. (13) x {the unique individual at the end of an evidential trail who fits the murderer-​of-​Smith role} is insane. In other words, speakers who explicitly use phrases like ‘whoever that turns out to be’ or ‘whoever that is/​might be’ are signaling either (i)  that their current epistemic state does not allow for the identification of a determinate individual or, (ii) if the speaker happens to be acquainted with the individual, that it is irrelevant for current conversational purposes for the hearer to be so acquainted. Nevertheless, the speaker is committed to there being a determinate (and unique) individual who bears the role property in question, which in the case of (6) is the property of being a murderer of Smith. This “whoever that turns out to be” test does not, therefore, indicate the expression of a general proposition quantifying over individuals. The speaker who uses (6) attributively is expressing an object-​dependent proposition about an entity and is presupposing that there is an evidential pathway that can (in principle) be followed and that terminates with (the presentation of) a unique entity. If such an evidential chain does not exist, the description fails to refer and the utterance is neither true nor false.2 Furthermore, by using the phrase ‘whoever that might be’, the speaker also communicates that a situation that would ground an acquaintance relation with the entity denoted by the description is not currently available to the speaker or not currently relevant for the hearer’s purposes. Thus, a general proposition is communicated, but it is one that quantifies over situations rather than entities. What the speaker does not have (or believes that it is not relevant to provide) is a way of accessing the precise situation within which the uniquely existing entity referred to by ‘the murderer’ can be directly identified. If (13) is the correct representation of the attributive use of (6), this would mean that both referential and attributive uses of ‘the murderer’ serve the function of picking out an individual that the predicate ‘is insane’ goes on to say something about.The difference lies in the epistemic availability of the entity in question to the speakers and hearers in the relevant conversational context. A referential use of (6) could be represented as follows: (14) x {the salient individual in this courtroom on trial for Smith’s murder that the speaker is gesturing to} is insane. It should be obvious by a comparison of (13) and (14) that the major difference between the cases has to do with the nature of the shared knowledge that the speaker is relying on for the reference to an individual x. In the attributive case, the shared knowledge is minimal and depends crucially on the stated role property of being a murderer of Smith, although it also requires an awareness (not necessarily conscious) of the possibility of tracing an evidential pathway (either backwards or forward or both) to x.3 In the referential case, the shared knowledge is much richer and depends to a large extent on the shared perceptual environment of the speaker and hearer, as well as on mutual encyclopedic knowledge of the trial and what it means to be standing in the dock, etc. That is to say, when (6) is used referentially, a situation that grounds a relation to the entity is immediately and mutually cognitively available. It is also worth noting that, because of the rich shared cognitive environment, the role property of being a murderer of Smith need not be the decisive factor in securing reference to a unique entity in the case of (6) uttered in the courtroom scenario imagined above. Suppose that Bob knows that Jones has been framed for Smith’s murder. Bob will nevertheless know that Anne has referred to Jones by her utterance of (6). He might go on to point out to her that the trial is a sham, but this would not defeat her successful reference to Jones. The idea that the descriptive property F could be false of x and yet ‘the F’ might nevertheless successfully refer to x has been much disputed. Donnellan (1966, 287) first suggested this as a possibility. 58

The Referential-Attributive Distinction

For example, at a party, Anne might try to tell Bob something about Jones, who is standing across the room holding a Martini glass, by uttering (15) below: (15) The man drinking a Martini is a history professor. Donnellan’s intuition is that Anne can successfully refer to Jones by means of the description ‘the man drinking a Martini’ (and can say something true if Jones is indeed a history professor) even if it turns out that Jones has water in his Martini glass. Moreover, if there is a man at the party, Brown, who is actually drinking a Martini and who is not a history professor but who is hidden behind a screen so that Anne and Bob cannot see him and are thus totally unaware of his presence, Anne’s utterance of (15) does not denote Brown and say falsely of him that he is a history professor. It is implausible in this situation to insist that the description in (15) semantically denotes Brown and that Jones is merely the speaker’s implicated reference, as defenders of the Russell-​plus-​Grice account advocated by Kripke (1977) have claimed. For one thing, the conversational context for the purposes of the exchange between Anne and Bob does not include unobserved spaces at the party but only the immediate space of which Anne and Bob are mutually perceptually aware. Just as the discovery that Jones has been falsely accused of Smith’s murder does not undermine Anne’s courtroom reference to non-​murderer Jones (because the interlocutors share a rich perceptual environment), so too the accidental existence of a hidden Martini-​drinking Brown does not undermine Anne’s party reference to non-​Martini-​drinking  Jones. In opposition to Donnellan, Reimer (1998) and Recanati (2013) argue for the view that the referent of ‘the F’ must be an F for successful (semantic) reference to occur. Recanati (2013) calls the descriptive content explicitly encoded by a definite description a “linguistic” mode of presentation of the referent and notes that it “may not be determinate enough to fix the reference”; such linguistic modes of presentation often only ‘constrain’ the reference (Recanati 2013, 162). He goes on to argue that it is a psychological mode of presentation that fixes reference. He conceives of such psychological modes of presentation as mental files. Moreover, mental files fix reference in a relational, as opposed to a “satisfactional”, manner. That is, reference is fixed in virtue of the causal (e.g., perceptual or testimonial) relations that hold between an entity and a particular mental file. It is not fixed because the entity satisfies the descriptive information recorded inside the mental file. However, Recanati does think that in order to count as the semantic referent, as opposed to being merely the speaker’s intended referent, the entity must satisfy this descriptive information. In cases of “improper” uses of referring expressions, the speaker’s reference, which is fixed by the mental file, fails to satisfy the encoded character of the expression and hence there is no semantic reference in such cases. However, Donnellan’s view of the matter seems more acceptable once the referential-​attributive distinction is recast as one about the type of epistemic access that a speaker has to the situation that grounds an acquaintance relation to the entity denoted by the description. The necessity for the referent of ‘the F’ to be an F seems primarily to constrain attributive uses. In the attributive case—​as I am currently understanding this case, namely as expressing a singular proposition but also communicating a general proposition that quantifies over situations—​the descriptive information is crucial as it is an essential part of the thin information available to connect the interlocutors to the referent, along with the network of socially enabled cognition that the interlocutors are connected to that constitutes their evidential link to the referent. This means that successful reference in the attributive use of (6) requires the referent to satisfy the descriptive property of being a murderer. So, if it turns out that Smith committed suicide and that the wounds to his body were inflicted after his death by wild animals, then the attributive use of (6) will fail and no proposition will be expressed at all (just as Strawson maintained). I suggest that we can treat the descriptive information contained in a definite description of the form ‘the F’ as background information relative to the at-​issue contents of utterances of sentences of the form ‘The F is G’ or ‘x stands in relation R to the F’—​in other words, utterances where ‘the F’ is 59

Anne Bezuidenhout

playing a subject or object argument role. (The issue of predicative uses of ‘the F’ as in sentences of the form ‘x is the F’ will be mentioned briefly in section 5 below). The at-​issue content is the singular proposition that x is G (or that x stands in R to y), where x (or y) is the unique referent, if any, of ‘the F’ in the conversational context. The background descriptive information helps the hearer to understand which entity is the speaker’s intended referent, but it need not always be true of the entity in question. However, this background information can be richer or poorer, and more or less necessary for communicative purposes, depending on the speaker’s own resources and on what resources the speaker can assume are mutually manifest (in the sense of Sperber and Wilson 1995, 38–​46) to the interlocutors in the conversational setting. Thus, there is a continuum of cases where the descriptive information is needed to a greater or lesser degree. Cases such as the following, imagined to be discourse initial, are ones where the encoded descriptive material is sufficiently rich to narrow down the background within which the unique (‘first’ or ‘sole’) referent is to be located (although I will problematize this claim in the next section with respect to (16)): (16) The first human to step on the surface of Mars will be a woman. (17) The sole winner of the Oregon $6.5 million lottery in August 2019 was a 94-​year old air force veteran who almost threw away his Game Megabuck winning ticket. On the other end of the continuum are the cases mentioned above (of Anne’s courtroom or party references to Jones), where the mutual perceptual and cognitive environments of the interlocutors are so rich that the descriptive information that is invoked by the speaker can afford to be degraded in various ways without impeding the hearer’s ability to figure out what unique entity the speaker is referring to. It is important to stress that there can be cases where the mutual cognitive environment of the interlocutors is sufficient for the hearer to understand what the speaker is talking about, despite the fact that neither speaker nor hearer is personally acquainted with the referent, as in the case of the detective and his associates who stumble upon Smith’s body and start talking about ‘the murderer’. Theirs is a situation very similar to the one which Kripke (1980) argues we find ourselves in when we use an ordinary proper name such as ‘Plato’. We are able to talk about Plato, even though we are not personally acquainted with the philosopher in question. We can do this because we have picked up the name ‘Plato’ from various trusted sources, although we may not remember exactly who or what those sources were. Those sources in turn conditioned their use of the name ‘Plato’ on their sources, and so on via a network of sources that stretches back in time and place to those who were directly acquainted with Plato. The difference in the imagined detective case is that this would be a dubbing event in which a new term, ‘the murderer’, is introduced to stand for the unique referent who is presupposed to be at the end of an evidential path that may extend both forwards and backwards in time. Actually, we can suppose that the detective is a bit more imaginative than this and dubs the referent ‘the slasher’. His associates then pick up this term and they begin talking about the slasher as they go about their investigations, assembling evidence and trying to figure out who the slasher is. Something like this actually happened in the 1970s in Britain when someone was murdering prostitutes and the police dubbed him ‘The Yorkshire Ripper’. Subsequently, a man by the name of Peter Sutcliffe was discovered to be the murderer. Most uses of definite descriptions likely fall somewhere in between these two extremes (i.e., very rich shared knowledge versus very thin shared knowledge). For instance, there will be cases in which speakers are acquainted with or have identifying knowledge of their intended referents. They may then attempt to provide their hearers with the epistemic resources to themselves identify these referents or, more minimally, to provide hearers with sufficient resources to be able to refer to the referents in the same socially enabled way that I can refer to Plato by using the name ‘Plato’ 60

The Referential-Attributive Distinction

and that the detectives in Britain in the late 1970s could refer to Peter Sutcliffe via the description ‘The Yorkshire Ripper’ even before they nabbed him for his crimes. Some examples of attested uses of definite descriptions cited by Abbott (2019, 120) are instructive to examine. These are examples taken at random from Abbott’s local newspaper (the Travers City Record Eagle). Abbott cites these examples to demonstrate the ease with which one can find naturally occurring examples in which definite descriptions introduce new information into the discourse context. These cases are a challenge to Heim’s (1982) familiarity criterion for the use of ‘the F’: (18) “Police said Thursday that they have caught the Baseline Killer, the gunman responsible for nine slayings that spread terror across the Phoenix area for nearly a year and a half” (Travers City Record Eagle, December 8, 2006, p. 5A). (19) “Property owners who live along Duck Lake’s shores could soon be on the hook for a new special property tax. The proposed tax would fund the maintenance and operation of a dam used to control the lake’s level” (Travers City Record Eagle, October 15, 2015, p. 1A). (20) “Jason Anthony Ryan was caught in a lie when he initially told investigators he didn’t know Geraldine Montgomery. That’s part of the argument Kalkaska County Prosecutor Mike Perrault made Wednesday in his opening statements in Ryan’s trial for the 1996 rape and murder of a Kalkaska resident, Geraldine Montgomery” (Travers City Record Eagle, October 15, 2015, p. 3A). Abbott’s point is that none of these examples assume that readers are already familiar with the entities in question (the gunman, the dam maintenance plan, or the prosecutor’s argument). However, these examples also illustrate something about the level of descriptive detail that speakers use when talking about a new entity, which in turn shows something about the level of epistemic access that these authors have to the referents of the highlighted definite descriptions and whether and to what degree this epistemic access is transmitted to their readers. I think it would be fair to say that neither authors nor readers are (or need to be) intimately acquainted with these entities.Yet this does not compromise their ability to communicate successfully about these entities via the use of these descriptions. Much of the communicative burden here rests on a complex social network of information sources. There is no expectation that readers will acquire identifying knowledge of an entity simply by grasping the descriptive properties explicitly mentioned by the authors. Even though the authors of these remarks have added a fair amount of descriptive detail to help their readers understand what is being said, readers will still have to put some level of trust in the writers as reliable information sources, and writers in turn will have to trust their own information sources, and so on across a complicated and extended social web of information transmission. For example, the snippet (18) about the Phoenix murderer was likely taken from a newswire service for the purposes of the local Travers City newspaper report. Thus, readers are at least two levels of epistemic access removed from those actually acquainted with the murderer. In the case of the Kalkaska County trial, it is possible that a Travers City reporter traveled to Kalkaska, which is only about 25 miles away from Travers City, in order to attend the trial and report on it firsthand. Even so, newspaper readers will need to take it on trust that the reporter has captured the prosecutor’s argument accurately enough (or that it is sloppy in ways that do not matter; if we really cared, we have enough information to go back and check the trial transcript, or if that is still too sloppy we could ask the prosecutor for his own trial notes). These examples together demonstrate that references via definite descriptions frequently depend on complex social networks, and require trust in the testimony of others. These sorts of cases are representative of the large band of cases that lie between the two extremes mentioned above—​of speaker and hearer sharing rich perceptual representations on the one extreme and of speaker and hearer sharing very thin conceptual representations on the other. In the vast majority of cases, there will not be a need for rich shared cognitive and perceptual representations of entities if instead the interlocutors are plugged into an appropriately rich social information network 61

Anne Bezuidenhout

and can defer to others who do have cognitive and perceptual representations that amount to identifying knowledge of an entity. To conclude this section, if the idea of a continuum of cases, which depends on different degrees of shared background information and different degrees to which interlocutors defer to their information sources, is accepted, we would have to replace the referential-​attributive distinction by a referential-​attributive continuum.

4  Signaling Referential versus Attributive Uses In the previous section I argued that referential and attributive uses of ‘the F’ in sentences of the form ‘The F is G’ or ‘x stands in R to the F’ are always to be understood as uses of a singular referring term being used to pick out some contextually unique entity. The difference between the two cases concerns the level of epistemic access that the speaker (and hence the audience) has to the intended referent. In attributive uses of ‘the F’, the speaker typically has thin epistemic access to the referent and is primarily interested in the referent qua bearer of the role of being F rather than the referent as a thing in itself.4 To illustrate this, consider the following example. Suppose Anne and Bob are driving through a small town whose main road is full of potholes. Anne expresses her frustration by uttering (21) below: (21) The mayor, whoever that turns out to be, should be impeached. Let us suppose that the town is small enough so that it is the mayor’s job to maintain the roads, rather than, say, the job of a city works manager.This level of detail is irrelevant to my example, which could be changed to ‘The city works manager should be impeached’.With respect to (21), I claim that Anne is more interested in the entity qua bearer of the role of being mayor of the town than in the actual occupier of the role. This could be revealed if Anne subsequently discovers that the role of mayor in this town is not occupied by an individual but by a committee whose joint decisions determine official actions and policy in this town. It seems that Anne should be willing to affirm that it is the committee that should be impeached. Similar remarks could be made about example (16), ‘The first human to step on the surface of Mars will be a woman’. This has a natural attributive reading, according to which the speaker is interested in the entity qua role bearer rather than in the entity per se. In this case there may be questions as to whether someone is or is not the role bearer. For example, if the one who first steps on the surface of Mars has had electronic circuits implanted in her brain that significantly enhance the speed of her motor responses or has had genetic enhancements that make her capable of detecting infrared radiation, one might say that she is not a human. Or it might be that the first human on the surface of Mars will be wearing a robotic suit whose leg movements do not require its occupant to actually move her own legs. Will this count as stepping on the surface of Mars? We can leave this to the Guinness Book of Records to decide, when and if this ever happens. One might wonder how we are able to tell whether a description is being used referentially or attributively in the absence of explicit uses of phrases such as ‘whoever that turns out to be’. Are there any other grammatical markers of the distinction? Bezuidenhout (2013) concluded that there were not, at least not in English. Subsequently, Hedberg, Gundel and Borthen (2019, 112–​113) drew my attention to Borthen’s (2007) data involving Norwegian bare superlatives, which Borthen argues are marked for an attributive use. An example cited by Hedberg, Gundel and Borthen (2019, 112) is as follows: (22) Beste skiløper får en premie. best-​Def skier gets an award. ‘Whoever turns out to be the best skier will get an award’. 62

The Referential-Attributive Distinction

One might be tempted on this basis to suggest that when superlatives are used with definite descriptions (as in ‘the best skier’), they too are marked for attributive uses. However, this is not supported by the evidence. Borthen notes that (23) below, the definite description counterpart of (22), can have both referential and attributive readings: (23) Den beste skiløperen  får en premie. the best-​Def  skier-​Def gets an award. ‘The best skier will get an award’. A similar distinction between bare superlatives and their definite description counterparts applies in English, although English does not mark definiteness on nouns and adjectives, as does Norwegian. Consider the following pairs: (24) (25) (26) (27)

Winner takes all. The winner takes all. Last one at the table does the dishes. The last one at the table does the dishes.

The cases with bare superlatives, (24) and (26), do indeed seem to have only attributive readings where it is the entity qua bearer of the role property, whoever that turns out to be, that is the intended referent, whereas the versions with definite descriptions, (25) and (27), arguably have both types of reading, although admittedly their referential readings are a bit harder to get. (They also have habitual readings, but those are not ones relevant to the current point.) To get a referential reading for (25), for instance, imagine that the finalists for a competition are lined up on stage and we in the audience don’t yet know who the winner is. The emcee’s helper scoops up all the prizes from the table on which they were displayed and walks towards Jones, one of the finalists on stage. As she places all the prizes in Jones’s arms, the emcee says: ‘Aaand … the winner takes all!’ Here ‘the winner’ refers determinately to Jones, which is plain for all of us in the audience to see (even if we don’t know Jones by name—​perhaps we think of him as that contestant who did an amazing juggling act, keeping twenty balls in the air while riding around the stage on a unicycle—​and even if someone else was actually the winner, and the emcee’s helper was accidentally awarding the prize to the wrong person). One reason that referential readings are harder to get for (25) and (27) is that these readings construe the verbs as being used in the “historic” or “narrative” present tense, in which the speaker is narrating some unfolding series of events. This is perhaps a slightly marked form of speech. It would be easier to get both readings for the definite description with, say, future tense marking: (28) Winner will take all. (29) The winner will take all. It is still the case that (28) has only an attributive reading, while (29) allows both referential and attributive readings. Thus, bare superlatives in English (and Norwegian) are constructions marked for an attributive reading. However, as tempting as it may be to assume that superlatives attached to definite descriptions are likewise marked for an attributive use, in fact both referential and attributive uses of such constructions are possible. Thus, in the absence of phrases such as ‘whoever/​whatever that is’, there do not appear to be any grammatical markers of the referential-​attributive distinction, as argued in Bezuidenhout (2013).

63

Anne Bezuidenhout

5  Contemporary Linguistic Treatments of the Referential-​Attributive Distinction In this section I raise the question as to whether the account of the referential-​attributive distinction outlined in sections 3 and 4 is compatible with some recent accounts of definite descriptions offered by linguists. I  mention briefly the accounts given by Coppock and Beaver (2015) and Elbourne (2013), since both address the issue of Donnellan’s referential-​attributive distinction, the former in passing and the latter in some detail. Coppock and Beaver (2015) argue for a contrast between determinate and indeterminate readings of definite descriptions. They note that while the readings that Donnellan would count as referential count as determinate in their sense, they would also count some uses of descriptions that are clearly non-​referential (such as fictional uses or anaphorically bound uses) as determinate. Coppock and Beaver write: although the referential/​attributive distinction is similar to our determinate/​indeterminate distinction insofar as referential uses carry an existence presupposition and attributive uses do not (Donnellan 1966, p. 289), it is not exactly the distinction we are making. Indeed, on Elbourne’s (201[3]‌) analysis, which is compatible with ours, both referential and attributive uses are determinate (type e). 2015, 3795 Coppock and Beaver’s main line of argument in their paper is that predicative uses of definites, such as (30a) below, are basic and “argumental” descriptions, such as (31a) and (32a), are derived: (30) a. b. (31) a. b. (32) a. b.

Scott is the (only) author of Waverley. Scott is not the only author of Waverley. Scott saw the (only) keynote address. Scott did not see the only keynote address. Scott gave the (only) keynote address. Scott did not give the only keynote address.

They give an analysis of predicative uses of ‘the F’ according to which such uses carry a uniqueness presupposition but not an existence presupposition.The lack of an existence presupposition is shown by what Coppock and Beaver (2015, 383–​386) call anti-​uniqueness cases, such as (30b), which entails that there is more than one author of Waverley. It is clear that we cannot in this case be presupposing that there exists an ‘only author’. On the other hand, argumental uses of ‘the’ do typically carry an existence presupposition; this is the case in (31a) and its negative variant (31b), in which the description is an argument of the verb ‘see’. However, not all argumental uses carry existence presuppositions. For example, in (32b), where the description is an argument of the verb ‘give’, we get the same sort of anti-​uniqueness effects that we get with predicative uses. Statement (32b) has a reading according to which there was more than one keynote address and Scott gave one of them. Such a reading does not presuppose the existence of an ‘only keynote address’. (There is, of course, also a reading of (32b) according to which there was only one keynote address but Scott did not give that sole keynote address.) The exact details of Coppock and Beaver’s account—​of how they derive argumental uses from predicative ones and exactly how they defend their claim that descriptions have both determinate and indeterminate uses—​is not important for the purposes of the present discussion. It is enough to note that they agree with Elbourne (2013) that the referential-​attributive distinction is one that applies within the class of determinate uses of definite descriptions that denote entities (if they denote at all). Thus, I  turn now to a brief examination of Elbourne’s (2013) 64

The Referential-Attributive Distinction

account of definite descriptions and the referential-​attributive distinction and of the ways in which his account differs from my own. Elbourne (2013, 104–​119) claims that his univocal semantics for the definite article ‘the’ can account for Donnellan’s two uses. According to Elbourne, a definite description such as ‘the murderer’ hosts a (covert) situation variable s. (Elbourne refers to this as a situation pronoun.) That is, the English definite article ‘the’ takes two arguments, a noun phrase and a situation variable s. The denotation of the noun phrase ‘murderer’ is a function from an entity to a proposition that is truth-​evaluable with respect to a situation s, and ‘the’ imposes a condition on s, namely that there be exactly one thing in the situation s that has the property of being a murderer. If this condition/​ presupposition is met, the denotation of the whole description is an entity, namely, the unique murderer in s. If it is not met, the description will lack a denotation. Furthermore, the situation variable can either be assigned a contextual value or it can be bound. These two possibilities would correspond to the referential and attributive readings of (6) and amount to two ways in which one can introduce the presupposition that there is exactly one murderer in the relevant situation. See Elbourne (2013, 50). Assigning a contextual value to the covert situation variable in ‘the murderer’ results in a proposition which specifies the restricted situation s* within which the unique murderer is to be located. An attributive reading, on the other hand, results from binding the situation variable and it yields a purely general proposition that does not pick out any particular individual, since there is no particular restrictor situation given in which an entity fitting the description is to be found. However, if there is a unique murderer in the (wider) topic situation, that entity will be the one denoted by ‘the murderer’. There are two significant ways in which Elbourne’s account differs from the one I am proposing. Firstly, on Elbourne’s account, there can be no misdescription in the referential case. The entity denoted by ‘the F’ must be an F or else the expression is not defined and no proposition will be expressed by its use, because “the whole semantic computation will grind to a halt” (Elbourne 2013, 49). Moreover, Elbourne represents the attributive use as a “purely general proposition” in which “no individual murderer (real or imagined) features” (2013, 115), whereas my account supposes we are always referring to a contextually unique entity when we make a felicitous use of a definite description, although we may be referring to this entity in a more or less socially enabled way. On my account, a felicitous attributive use of ‘the murderer’ in the situation imagined for (6) is one in which there is an evidential pathway terminating with a unique entity who is a murderer of Smith. If there is no such path, there is no denotation and thus no truth-​evaluable claim is made. I noted above that Elbourne thinks of the referential-​attributive distinction for uses of ‘the F’ as amounting to “two ways, depending on whether or not the relevant situation pronoun is bound, in which a presupposition that there is exactly one [F] in a situation being talked about can be introduced” (Elbourne 2013, 50). That there is a unique F can either be a part of the restrictor situation s* in which the proposition expressed by ‘The F is G’ is fixed or it can itself become a part of the proposition that is expressed (and so strictly speaking will no longer be a presupposition, since what is asserted is not presupposed). This suggestion echoes the way in which Grice (1989, 269–​282) treats presuppositions in his essay titled ‘Presupposition and Conversational Implicature’. Grice’s main argument in this essay is that presuppositions can be reanalyzed as Gricean conversational implicatures. However, with respect to utterances of the form ‘The F is G’, Grice assumes that they communicate the unique existence of an F (here being neutral as to whether the fact of this unique existence counts as an implicature or a presupposition). Grice then suggests that there are two options as to the form of expression one could use to convey this information. He suggests we use ‘The F is G’ when our main point concerns the attribution of the property of being G to the unique F and we assume that we will not be challenged by our interlocutors as to the existence of this unique F. On the other hand, if there is some question

65

Anne Bezuidenhout

as to whether or not there is a unique F, we would use the form ‘There is a unique F and it is G’. Grice suggests that we make this choice by following a Principle of Conversational Tailoring that in effect allows us to background information that is not controversial. Conversational Tailoring Principle: “Frame whatever you say in the form most suitable for any reply that would be regarded as appropriate” or “Facilitate in your form of expression the appropriate reply”. Grice 1989, 273, 276 Elbourne (2013) does not say what sorts of conversational motivations might drive one to use a definite description either referentially or attributively. It would make sense for him to invoke something like Grice’s Principle of Conversational Tailoring, which suggests that the presuppositions of existence and uniqueness will be backgrounded when this unique entity’s having the property G is what is at issue, whereas the presuppositions will be foregrounded (become part of the at-​issue or asserted content) when the existence of a unique F is what is at issue. If he adopted this Gricean suggestion, this would make for a third difference between my account and Elbourne’s account, as I  am saying that the motivations for using the description either referentially or attributively are epistemic (rather than being driven by concerns of what is at issue versus what is backgrounded) and depend on the nature of the shared epistemic resources of the conversational participants.

6  Experimental Work on Cases of Misdescription In this section I briefly describe some experimental work by Rostworowski and Pietrulewicz (2019), testing ordinary people’s judgments about cases of misdescription of the sorts discussed above (e.g., Donnellan’s Martini glass example or the case in which (6) is used to refer to Jones in the courtroom, despite the fact that Jones is not the murderer of Smith). Rostworowski and Pietrulewicz (2019) designed a series of experiments to probe the extent to which misdescription is tolerated by ordinary speakers. They were interested in whether or not an utterance containing a definite description will be judged true, even if the speaker’s intended reference fails to satisfy the description. Participants in this study were presented with five scenarios. In one, a description was used with a clear reference to an entity satisfying the description but the claim made about the entity was false.The other four scenarios involved misdescription.The misdescription scenarios used the description either referentially or attributively and either there was no entity in the scenario that satisfied the descriptive material or there was something that satisfied the description but that was clearly irrelevant to the speaker’s intended communication. Each scenario ended with a character (e.g., Mary) uttering a sentence containing a definite description (e.g., ‘The owner of this apartment is an art expert’). Participants read such scenarios and then were asked to agree or disagree with the claim that what the speaker (e.g., Mary) said was true. Rostworowski and Pietrulewicz (2019) found that there was significantly more agreement with the claim that what the speaker said was true in the misdescription conditions than in the clear reference condition. The highest level of agreement was for the cases (both referential and attributive) in which there was nothing in the context that actually did satisfy the description. There was slightly less agreement (but still significantly more than in the clear reference condition) when there was an irrelevant entity that did in fact satisfy the description (whether the description was used referentially or attributively). Rostworowski and Pietrulewicz then go on to argue that their results cannot be explained away by appeal to a distinction between semantic versus speaker reference of the sort invoked by Kripke (1977) and they offer their own salience-​based semantic account of descriptions that they claim can account for the data.

66

The Referential-Attributive Distinction

I will not pursue their arguments here. The main interest of Rostworowski and Pietrulewicz’s (2019) experimental findings for my purposes is that they support my earlier claim that misdescription can be tolerated, especially in contexts in which the conversational participants share a rich epistemic environment. Another interesting feature of their study is that they found these effects for both referentially and attributively used descriptions. Although I argued in sections 2 and 3 that attributive uses are more likely than referential ones to require that the denoted entity satisfy the descriptive material, this applies to those cases in which interlocutors have very thin epistemic access to an entity or to cases in which the speaker is interested in the entity qua bearer of the role property specified by the description. However, I also made the point that there is likely to be a continuum of cases, where misdescription becomes more tolerated the richer the shared epistemic environment becomes. Thus, it is not surprising to find that attributively used descriptions can in certain cases tolerate misdescription.

7  Conclusions In section 2, I showed that it is possible to accommodate Donnellan’s referential-​attributive distinction within a variety of different semantic frameworks, including ones that endorse a univocal Russellian semantics, a univocal referentialist semantics, a semantic ambiguity account, or an underspecified semantic meaning account for definite descriptions. This shows that the existence of this distinction by itself does not tell us anything about the correct semantics for definite descriptions. Hence, the significance of this distinction, if any, must lie elsewhere. In section 3, I argued that the significance is in fact an epistemic one, with the difference between the uses representing differences in epistemic access to the (contextually unique) entity denoted by the description. In pure referential uses, the speaker is directly acquainted with the denoted entity. In attributive uses of ‘the F’ on the other hand, speakers denote a unique entity but in addition convey a general proposition to the effect that they do not have, or that it is not relevant for hearers to have, epistemic access to a situation within which a unique F can be directly identified. Furthermore, such attributive uses will be felicitous only if the interlocutors are hooked into an appropriate communicative network that terminates with an information source who is directly acquainted with the denoted entity. If there is no such terminus, reference will fail and no truth-​evaluable claim is made (whether the interlocutors are aware of this or not). I also argued that there was in fact a continuum of cases here. On the referential end of the continuum, speakers and hearers share rich perceptual and conceptual environments which enable them to jointly pick out the (contextually unique) entity denoted by the description. Moreover, because their mutual cognitive environments are so rich, such uses tolerate the possibility of misdescription. On the attributive end, the shared cognitive environment of the interlocutors is perceptually and/​or conceptually very thin and may amount to no more than a mutual grasp of the explicitly encoded descriptive content. Finally, I argued that most ordinary uses of definite descriptions probably fall somewhere between these two extremes, where neither speaker nor hearer is directly acquainted with the denoted entity, but they are appropriately hooked into an informational network that supports the speaker’s use and ties the interlocutors indirectly to the denoted entity. I supported this claim by citing some ordinary speaker uses from Abbott (2019), drawn from Abbott’s local newspaper. In section 4, building on this reconstrual of the referential-​attributive distinction as having merely epistemic rather than semantic import, I  asked whether there was any grammatical reflex of this distinction. That is, when phrases such as ‘whoever/​whatever that may be’ or ‘whoever/​whatever that turns out to be’ are not explicitly used by speakers to indicate an attributive use, is there some other grammatical means by which speakers can signal an attributive use? After a brief discussion of Borthen’s (2007) Norwegian data contrasting bare superlatives (which have only an attributive use) with their definite description counterparts (which have both referential and attributive uses),

67

Anne Bezuidenhout

I conclude that there is not any grammatical marking of the referential-​attributive distinction (at least in English). This can be taken as support for the claim in section 3 that the distinction has epistemic rather than sematic significance (since semantic contrasts are often, even if not invariably, indicated by grammatical ones). In section 5, I surveyed some recent work in linguistics that touches on Donnellan’s distinction. I briefly mentioned the semantic accounts of argumental (as opposed to predicative) uses of definite descriptions offered by Coppock and Beaver (2015) and Elbourne (2013). In both these frameworks, the referential-​attributive distinction is one that is made within the class of uses that Coppock and Beaver class as determinate uses; that is, uses that denote entities, if they denote at all. As Elbourne spells out the distinction, it amounts to two ways that the presupposition of unique existence of an F can come into play when a speaker utters a sentence where ‘the F’ is either a subject or direct or indirect object of a verb. The presupposition can either be part of the background or it can be made at-​issue and be directly expressed (which means, in effect, that the presupposition has been canceled—​since something that is asserted is no longer a presupposition). These two options in turn correspond to two ways of treating the situation variable that Elbourne claims is an argument of ‘the’ in languages like English; this situation variable can either be assigned a value in context (the referential case) or it can be bound (the attributive case). My own reconstrual of the referential-​attributive distinction as having merely epistemic import is at least compatible with the idea that the distinction is one drawn within the class of determinate uses that denote a (contextually unique) entity. However, my claim is that what is expressed in both cases is a singular (or object-​dependent) proposition. In attributive uses of ‘the F’, speakers in addition convey a general proposition to the effect that they do not have, or that it is not relevant for hearers to have, epistemic access to a situation within which a unique F can be directly identified. With such (epistemically impoverished) uses, however, speakers must be connected to an informational network of an appropriate sort, as such uses of descriptions are socially enabled ones.6 In section 6, I briefly mentioned some experimental work by Rostworowski and Pietrulewicz (2019) that supports my claim that descriptions can be felicitous even in cases where the entity denoted by the description fails to satisfy the description.Their data shows that misdescription is possible both for referentially and attributively used descriptions. This offers some support for my claim that the referential-​attributive distinction in fact represents a continuum of cases, depending to the degree of epistemic access that the interlocutors have to the denoted entity. In principle, the richer the epistemic environment and the more overlap there is between the interlocutors’ cognitive environments, the more forgiving the hearers will be for misdescriptions, because there will be alternative “epistemic pathways” to the denoted entity.

Notes 1 In this chapter I use the terms ‘singular proposition’ and ‘object-​dependent proposition’ interchangeably, even though they are not quite equivalent.The former term is used by those who support the Russellian notion of a proposition as a structured entity, which, in the circumstance that the sentence expressing that proposition contains a singular referring expression, has an object as one of its constituents.The latter term is used by those who support the neo-​Fregean idea that propositions have modes of presentation as constituents but who also support the possibility of de re or object-​dependent modes of presentation—​modes of presentation that are individuated partly in terms of their (e.g., causal) relations to the objects to which they refer. It is not necessary to settle this debate between Russellians and Fregeans for the purposes of addressing the referential-​attributive distinction. 2 Note that this evidential path may extend both backwards and forwards in time, as detectives will follow clues that allow them both to reconstruct the past as well as make predictions about the future.This means that I am committed to the claim that we can in principle refer (in the socially enabled way characterized in the body of this paper) to entities that do not yet exist, such as the first human to step on the surface of Mars, who may very well turn out to be a person who, as of 2019, is not yet born. I do not think this is implausible, given that I am not claiming that we have acquaintance knowledge of any unborn person. Of course, this also means

68

The Referential-Attributive Distinction that we may not know now whether we are successfully referring or not. This also seems acceptable to me as we frequently pick up referring expressions, including proper names and definite descriptions, and use them in communication despite having only weak epistemic access to their referents (e.g., when I implicitly trust you to be appropriately epistemically connected to the referent of your expressions when you talk about your cousin Freddy’s friend’s dog Plato). 3 A reviewer wondered whether it is always possible to trace an evidential pathway to some entity and suggests that we would not be able to do this for an attributive use of ‘the largest fish in the ocean’. Firstly, I assume that we are not talking about the largest species of ocean-​dwelling fish but rather an individual fish (of some species or other) and that there is some (implicit) spatio-​temporal restriction, say to the largest fish currently living in one of Earth’s oceans (cf. ‘the oldest currently living human being’). Obviously, the oceans are too vast for us to conduct a search for such an individual fish and no one has an interest in conducting such a search, as the scientific resources needed for the search would be prohibitively expensive and the scientific payoff would be vanishingly small. However, we do know that there must be such a fish, as we know there are finitely many fish in the ocean at any point in time and that they can be ranked according to their size. So potentially we could locate that fish. Similar remarks apply to descriptions such as ‘the smallest ant currently living on Earth’, ‘the hottest sun currently burning in the Milky Way’, etc. 4 It is also possible that the speaker has direct acquaintance with the entity in question but deems that the audience doesn’t need to have such epistemic access in order to understand the speaker’s conversational contribution. This is likely to occur in cases in which the speaker is focusing on the entity qua bearer of a role property and where the conversational point concerns the functionary rather than the entity who just so happens to currently occupy that role. The example in the text about the mayor could be adapted to illustrate the point. I might say to you that the mayor should do thus-​and-​so intending you to understand that it is the role-​occupant with his/​her mayoral hat on, so to speak, who should do thus-​and-​so. The fact that I happen also to be acquainted with the current mayor because she lives next door to me will be irrelevant to the point I want you to take away. I would make the same point even if someone different were the current mayor. It is also possible that the speaker is acquainted with the referent of the description but under a different guise and so is unaware of this fact. 5 To say that these uses of definite descriptions are of (semantic) type e is to say that they denote entities. 6 Although I cannot here delve too deeply into the issue of definite descriptions (DDs) in modal contexts, some explanation has to be given for the (apparent) non-​r igidity of DDs. For example, consider (i) ‘The murderer might have been insane’. Suppose we’re talking about Smith’s murder and that Jones has been convicted of this murder. There are supposed to be two readings for (i), depending on whether the DD takes wide or narrow scope with respect to the modal. On the wide-​scope reading, which I claim is the most natural reading, we are saying that the actual murderer, Jones, might have been insane (although actually he is perfectly sane and is just pleading insanity to escape punishment, say). On the narrow-​scope reading, we are saying that it is possible that Smith’s murder was committed by an insane person, Jeffery Dahmer, say. The problem is that if events are individuated partly by their agents, then we can’t switch out the murderer without changing the event. So, on the narrow-​scope reading of (i), we won’t be talking about the actual murder of Smith anymore. We will in effect be imagining alternative histories for Smith, each of which terminates in him being murdered by someone and saying that one of these alternate murders was committed by an insane person. I do not deny such a reading is possible. However, I think it is a reading in which we are focusing on the role properties of the constituents of the murdering event and have abstracted away from all but the role of the victim, Smith. Smith anchors the utterance of (i) to the actual situation whose modal properties are being talked about. Such narrow-​scope readings are similar to the attributive readings of indexicals and pronouns discussed in section 1 (see ­examples 8–​11).

References Abbott, B. (2019). Definiteness and familiarity. In J. Gundel and B. Abbott (Eds.), The Oxford Handbook of Reference. (pp. 117–​129). Oxford: Oxford University Press. Atlas, J. (2004). Descriptions, linguistic topic/​ comment and negative existentials. In M. Reimer and A. Bezuidenhout (Eds.), Descriptions and Beyond. (pp. 342–​360). Oxford: Oxford University Press. Bezuidenhout, A. (1997). Pragmatics, semantic underdetermination, and the referential/​attributive distinction. Mind 106: 375–​410. Bezuidenhout, A. (2013).The (in)significance of the referential-​attributive distinction. In A. Capone, F. Lo Piparo, and M. Carapezza (Eds.), Perspectives on Pragmatics and Philosophy. (pp. 351–​366). Berlin: Springer. Bezuidenhout, A. (2016). Presupposition failure and the assertive enterprise. Special issue editor F. Domaneschi, Topoi 35(1): 23–​35.

69

Anne Bezuidenhout Borthen, K. (2007). The distribution and interpretation of Norwegian bare superlatives. Working papers ISK 4, Department of language and communication studies, Norwegian University of Science and Technology. Coppock, E. and Beaver, D. (2015). Definiteness and determinacy. Linguistics and Philosophy 38: 377–​435. Devitt, M. (2004). The case for referential descriptions. In M. Reimer and A. Bezuidenhout (Eds.), Descriptions and Beyond. (pp. 280–​305). Oxford: Oxford University Press. Donnellan, K. (1966). Reference and definite descriptions. Philosophical Review 75(3): 281–​304. Elbourne, P. (2013). Definite Descriptions. Oxford: Oxford University Press. von Fintel, K. (2004). Would you believe it? The king of France is back! (Presuppositions and truth-​value intuitions). In M. Reimer and A. Bezuidenhout (Eds.), Descriptions and Beyond. (pp. 315–​341). Oxford: Oxford University Press. Frege, G. (1892/​1960). On sense and reference. In P.T. Geach and M. Black (Eds.), Translations from the Philosophical Writings of Gottlob Frege, 2nd edition. (pp. 56–​78). Oxford: Blackwell. Grice, H. P. (1989). Studies in the Way of Words. Cambridge, MA: Harvard University Press. Hedberg, N. Gundel, J. and Borthen, K. (2019). Different senses of ‘referential’. In J. Gundel and B. Abbott (Eds.), The Oxford Handbook of Reference. (pp. 100–​116). Oxford: Oxford University Press. Heim, I. (1982).The Semantics of Definite and Indefinite Noun Phrases. PhD thesis, University of Massachusetts, Amherst, MA. Kripke, S. (1977). Speaker reference and semantic reference. Midwest Studies in Philosophy 2(1): 255–​276. Kripke, S. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press. Lasersohn, P. (1993). Existence presuppositions and background knowledge. Journal of Semantics 10: 113–​122. Meinong, A. (1904/​1960). Über Gegenstandstheorie.Translated as ‘The theory of objects’, in R. Chisholm (Ed.), Realism and the Background of Phenomenology. (pp. 76–​117). Glencoe, IL: Free Press.. Neale, S. (1990). Descriptions. Cambridge, MA: MIT Press. Neale, S. (2004). This, that, and the other. In M. Reimer and A. Bezuidenhout (Eds.), Descriptions and Beyond. (pp. 68–​182). Oxford: Oxford University Press. Nunberg, G. (1993). Indexicality and deixis. Linguistics and Philosophy 16: 1–​43. Nunberg, G. (2004). Descriptive indexicals and indexical descriptions. In M. Reimer and A. Bezuidenhout (Eds.), Descriptions and Beyond. (pp. 261–​279). Oxford: Oxford University Press. Recanati, F. (1993). Direct Reference. Oxford: Blackwell. Recanati, F. (2013). Reference through mental files:  Indexicals and definite descriptions. In C. Penco and F. Domaneschi (Eds.), What is Said and What is Not: The Semantics/​Pragmatics Interface. (pp.159–​173). Stanford, CA: CSLI Press. Reimer, M. (1998). Donnellan’s distinction/​Kripke’s test. Analysis 58(2): 89–​100. Reinhart, T. (1981) Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27: 53–​94. Rostworowski, W. and Pietrulewicz, N. (2019). Are descriptions really descriptive? An experimental study on misdescription and reference. Review of Philosophy and Psychology 10: 609–​630. Rouchota,V. (1992). On the referential/​attributive distinction. Lingua 87: 137–​167. Russell, B. (1905). On denoting. Mind 14: 479–​493. Sperber, D. and Wilson, D. (1995). Relevance: Communication and Cognition. 2nd edition. Oxford: Blackwell. Strawson, P. (1950). On referring. Mind 59: 320–​344. Strawson, P. (1952). Introduction to Logical Theory. London: Methuen. Strawson, P. (1964). Identifying reference and truth-​values. Theoria 30: 96–​118.

70

PART II

Causal Theories of Reference

5 THE CASE(S) AGAINST DESCRIPTIVISM Michael Nelson

Descriptivism is the thesis that, for any proper name n, there is some definite description ‘the F’ such that n and ‘the F’ are semantically equivalent. Descriptivism is the thesis that the semantic content of a proper name (i.e., the term’s contribution toward what sentences containing used occurrences of that term express) is more fully and perspicuously expressed by some definite description. Descriptivism entails, then, that for any sentence of the form ‘n is G’, there is some sentence of the form ‘The F is G’ such that the two sentences express the same proposition. Descriptivism is related to what we can call an anti-​singularity ambition in both philosophy of language and mind. Genuine singular terms, a term whose sole semantic purpose is to refer to an individual, and in particular, a bit of extra-​mental, physical, objective reality, and so whose semantic content is exhausted by its referent—​what, following David Kaplan (see Kaplan 1989), are devices of direct reference—​have been viewed with suspicion by some (most famously, Gottlob Frege (see Frege 1948), who thought that all reference was mediated, if not by description then at least by a mode of presentation, where, for any object, there are multiple, cognitively nonequivalent modes of presentation of that object1). The anti-​singularity ambition, then, aims to rid language of genuine singular terms and to account for all thought about a bit of extra-​mental, physical, objective reality as indirect thought about that object.There is, on this view, no thought that is directly about extra-​mental, physical, objective reality, where a thought about an object o is a direct thought about o just in case it is about o in virtue of its containing o as a constituent. In this essay, I  present a case against descriptivism that calls into question this ambition, arguing for the indispensability of direct thought about concrete, particulars in the world. I also present the more traditional case against descriptivism from Saul Kripke’s Naming and Necessity (Kripke 1980). Before we focus on the main theme of this essay, namely, problems with descriptivism, we will first briefly rehearse some of the primary reasons in favor of the thesis. (The classic source of these motivations is Bertrand Russell’s ‘On Denoting’ (Russell 1905).) The first, and most famous, is Frege’s argument for the sense/​reference distinction: the puzzle of propositional attitudes. Intuitively, co-​referring proper names, i.e., a pair of proper names that have a common referent, cannot be intersubstituted salva veritate in propositional attitude ascribing environments. For example, intuitively, the sentence ‘Duke Orsino believes that Cesario is his favorite servant’ is true while the sentence ‘Duke Orsino believes that Viola is his favorite servant’ is false. (The example derives from Shakespeare’s Twelfth

73

Michael Nelson

Night, which we shall pretend to be actual fact.) Grant the data and suppose, for reductio, that all proper names are devices of direct reference and that a sentence of the form ‘A believes that p’ is true (in a context C) just in case the object designated by a stands in the believing relation to the proposition expressed by p (in C). Then, as ‘Cesario’ and ‘Viola’ both refer to the same person, the proposition expressed by ‘Cesario is [Duke Orsino’s] favorite servant’ is the same as the proposition expressed by ‘Viola is [Duke Orsino’s] favorite servant’ and so Duke Orsino both believes and does not believe one and the same proposition, which is a contradiction. So, at least one of our assumptions is false. The assumption of direct reference seems the likely candidate. A second motivation is the problem of nonexistence.The sentence ‘Othello does not exist’ is both meaningful and true. After all, line up all of the people that exist, ever have existed, or ever will exist and Othello will not be found in the line. (Some people, myself included, think that Othello does in fact exist, although as a fictional character, as they adopt a form of fictional realism, according to which the creatures of fiction are existent, albeit abstract, individuals. Proponents of this view are best to claim that our target sentence is in fact false, perhaps being used to communicate the thought that Othello is not a real, concrete person, which ordinary people, and even some philosophers, naturally conflate with not existing. So, the true thought behind an ordinary, sincere assertion of the sentence ‘Othello does not exist’ is the thought that Othello is not a real, concrete person.) But syntactically, at least, ‘Othello’ is a proper name. Suppose, for reductio, that all proper names are devices of direct reference. Then the meaningfulness of our target sentence requires that ‘Othello’ has a referent (as, given that assumption, the only candidate meaning for a proper name is its referent). If we insist that absolutely everything there is exists (i.e., if we reject the Meinongian doctrine that there are entities with being, entities that are, that do not exist), then that referent exists. So, the meaningfulness of our target sentence, given our two assumptions, guarantees its falsity. So, there are no true negative existentials with proper names, as their meaningfulness, which is a prerequisite for their truth, suffices for their falsity. Descriptivism promises to avoid all of these problems. Let n and m be distinct, cognitively nonequivalent, co-​designating proper names. If those names are semantically equivalent to distinct definite descriptions, then the puzzle of belief seems to be resolved. Distinct definite descriptions can co-​designate the same object but still express distinct contents, insofar as the descriptive conditions they encapsulate are distinct. For example, the description ‘the square root of 16’ and ‘the sum of 2 and 2’ both designate the same number: 4. But the thought that the sum of 2 and 2 is the sum of 2 and 2 is not the same as the thought that the square root of 16 is the sum of 2 and 2, as most of the age of 8 know the first but not the second. So, if the meaning of a proper name is given by a definite description, the sentences ‘n is Y’ and ‘m is Y’ can express distinct propositions as long as the definite description associated with n expresses a distinct descriptive content from the definite description associated with m. Return to our earlier example involving ‘Cesario’ and ‘Viola’. Suppose that ‘Cesario’ is associated with the definite description ‘the young man professing Duke Orsino’s love to Olivia’ and ‘Viola’ is associated with the definite description ‘the twin sister from Messaline’. (We are still pretending that Twelfth Night describes actual fact and so that Messaline is an actual location.) Then Duke Orsino believing that Cesario is his favorite servant does not conflict with his not believing that Viola is his favorite servant, as the proposition that the young man professing his love to Olivia is distinct from the proposition that the twin sister from Messaline is his favorite servant and so Duke Orsino can be in a position of believing true the first without believing true the second. The problem of true singular negative existentials also has an elegant solution if the descriptivist thesis is true. Suppose that the descriptive equivalent of ‘Othello’ is ‘the general of Venice who is married to Desdemona’. (Forgive the fact that the name ‘Desdemona’ is also short for some definite description, if descriptivism is true.) The meaningfulness of ‘The general of Venice who is married to Desdemona exists’ does not presuppose the existence of a designation for the apparent subject 74

The Case(s) against Descriptivism

expression ‘the general of Venice who is married to Desdemona’. Indeed, if we accept Russell’s theory of definite descriptions, then the genuine subject of the sentence is the variable ‘x’, which is then bound by a definite description operator the x[x is a general of Venice and x is married to Desdemona]. The sentence has content because the descriptive contents exist, even if they fail to describe anything. Russell himself claimed that we should conceive of ‘exists’ not as expressing a first-​order property of individuals but instead as a second-​order property. Crudely, ‘Othello does not exist’ expresses the proposition that nothing is the general of Venice who is married to Desdemona. However, one can insist that ‘exists’ expresses a first-​order property of individuals and still avail oneself to the heart of the Russellian explanation of true negative singular existentials. On this view, ‘Othello does not exist’ expresses the proposition [It is not the case that: the x[x is a general of Venice who is married to Desdemona](x exists)]. This negative proposition is true and it does not require that there is some object o such that, when o is the value of x, the open sentence ‘x exists’ is false (and hence o has some shadowy reality of not existing). Rather, the restricted quantifier simply never delivers a value to be assigned to ‘x’ as nothing in the domain satisfies the descriptive condition. The puzzle of belief and the problem of meaningful and true negative singular existentials are the primary motivations for the thesis of descriptivism. They provide abductive evidence for the thesis’s truth by offering solutions to problems that seem hard to solve otherwise. One way to undermine this motivation is to provide solutions to these problems that do not require descriptivism and argue that those solutions are better than the solutions the descriptivist provides. Another way to respond is to argue that descriptivism faces problems of its own, even if we grant that supposing its truth provides solutions to the problems discussed above, and simply leave unsolved the motivating considerations in favor of descriptivism. In the remainder of this entry, I will present two sets of arguments that descriptivism is false. (I also think that the first strategy is a promising strategy and that there are nondescriptivist solutions to both the problems of belief and negative existentials that are superior to the descriptivist solutions. However, that would take us too far afield for this short entry against descriptivism.) A thorough-​going form of descriptivism, the thesis that there are absolutely no devices of direct reference and that all thought about any individual o is a descriptive thought about o, that is, a thought about o in virtue of containing a descriptive condition that does not presuppose o for its identity and that o uniquely satisfies and not a thought about o in virtue of containing o as a direct constituent, faces I think fatal problems developed by Peter Strawson in his reduplication argument in ­chapter 1 of his Individuals (Strawson 1959), which he used to establish the necessity of what he called demonstrative identification as being fundamentally distinct from descriptive identification. This argument, if successful, attacks not only the thesis of descriptivism, that, for any proper name, there is a definite description that is semantically equivalent to it, but also more broadly conceived descriptivist theses that reference is always mediated through satisfaction conditions. (Very crudely, identifying an object is a precondition on having a thought that is determinately about that object in particular and not just objects of a given kind.) We will recast the reduplication as an argument that there are some terms that are devices of direct reference. I should say at the outset that the argument does not show that proper names are devices of direct reference; at best, it shows that some terms are devices of direct reference. That is, at best it shows the necessity of direct reference in order for there to be thought about the external world. I will return after presenting the argument to the step from this conclusion to the conclusion that proper names are devices of direct reference. Suppose, for reductio, that all thought about extra-​mental reality is descriptive. So, let o be some extra-​mental particular. Then this supposition entails that any thought about o is about o only indirectly, in virtue of accurately describing o. So, when I think about Russell, thinking, say, that he is a great philosopher, my thought is something of the form [The person who discovered the inconsistency of Basic Law V is a great philosopher]. (Strictly speaking, ‘Basic Law V’ is a name for a proposition, crudely, the proposition that the extension of the concept F is identical to the extension of the concept G just in case, for every object x, x falls under F if and only if x falls under G. We will 75

Michael Nelson

continue to abbreviate this with the name ‘Basic Law V’.) Suppose now that we live in a universe in which there is massive reduplication. So, for everything that is or has happened around us, in a far distant part of the universe, there is something qualitatively identical that is or was happening there. Then, insofar as my thought that Russell is a great philosopher has as its content the proposition that the person who discovered the inconsistency of Basic Law V is a great philosopher, then my thought is not determinately about anything, as there are, under the reduplication hypothesis, two distinct individuals who (independently) discovered the inconsistency of Basic Law V. But that seems counterintuitive. Matters are worse. Right now I am sitting in front of a lovely black cat, thinking how cute he is. If all my thought about extra-​mental reality is descriptive, then the content of my thinking is purely qualitative. But if we suppose again that we live in a massive reduplication universe, then my qualitative thoughts, whatever the properties they involve, are not determinately about the object I am perceiving and intuitively thinking about, as they equally well describe both Luca, the cat I am sitting in front of, and his doppelgänger on the other side of the universe. If all thought is purely descriptive, then, assuming the reduplication hypothesis, no thought is determinately about any extra-​mental particular. But surely my thought is about the object causing my perceptions and not some other object perhaps even outside my light cone altogether and regardless not causing my perceptions. A descriptivist may retort as follows: “Ours is not, as a matter of fact, a reduplication universe and so our thought is, as a matter of fact, determinately about particulars, at least as far as the reduplication argument is concerned. Why concern ourselves with an implausible possibility?” Insofar as the argument assumes the truth of the reduplication hypothesis, this response seems apt. We shouldn’t be moved by what would be true were something that we have absolutely no reason to think to actually be true were true. But I think that this misses the force of the reduplication argument. The reduplication argument does not rest on the plausibility of the reduplication hypothesis. The reduplication hypothesis need only be coherent and its truth independent of whether or not we have determinate thoughts about objects in our immediate environment to cause problems for the pure descriptivist. I take it that the reduplication hypothesis is coherent. (Those who accept the Identity of Indiscernibles and thereby reject the possibility of there being two numerically distinct particulars that are qualitatively identical will argue that the reduplication hypothesis is not coherent. While I reject the Identity of Indiscernibles, this is a topic beyond the scope of this essay. Also, this response fails to adequately respond to Strawson’s argument, as we can recast the reduplication argument so that the qualitative indiscernibility is gauged to the discernment of the thinker, thereby not violating the Identity of Indiscernibles.) We can now argue as follows. Compare two possibilities. The first is what we take to be actuality, where I am sitting in front of my cat thinking of his cuteness and there is not a qualitatively identical situation happening anywhere else in the universe, although, of course, there are plenty of situations of the same general kind. The second is one in which the reduplication hypothesis is true. In particular, suppose that there is a portion of the universe outside any of our light cones that qualitatively duplicates our surroundings and so there is a qualitatively identical situation to the one in which I  am involved with numerically distinct objects. Intuitively, whether my thought is determinately about Luca, the cat I am perceiving, is indifferent to which possibility we consider. That is, supposing there to be a qualitatively indiscernible situation to mine somewhere else in the universe should not affect the content and truth of my thought. However, if the thesis of pure descriptivism is true, then it does, as my thought is only about a particular if my thought uniquely qualitatively individuates a particular, which it would not do in the second possibility. The important point of this reparsing of the reduplication argument is to see that the argument does not just rest on the premise that, if ours is a reduplication world, then none of our thoughts are determinately about particulars in our environment. Instead, the argument relies on the 76

The Case(s) against Descriptivism

claim that, intuitively, the mere presence of a causally unconnected part of the universe that qualitatively duplicates my surrounding does not affect the aboutness of my thoughts about my immediate environment. The reduplication targets a pure form of descriptivism, according to which all thought about particulars in the external world is indirect, descriptive thought and so there is no device of direct reference. Rejecting pure descriptivism is compatible with the thesis that, for any public language proper name, there is some definite description to which it is semantically equivalent, and hence the thesis of descriptivism with which we began. Undercutting the anti-​singularity ambition shows the need for a duality in our account of thought, akin to Kant’s duality between concept and intuition, Russell’s distinction between thought by description and thought by acquaintance, and Strawson’s distinction between descriptive and demonstrative identification. Of the great philosophers with a deep theory of thought and language, Frege claimed that all thought was cut from a single cloth: all thought, it is plausible to read Frege as maintaining, is conceptual and all aboutness is descriptive. The reduplication argument, as I interpret it, casts doubt on that thesis. But other descriptivists, like Russell and Strawson, in their different ways, reject pure descriptivism but think that proper names are nonetheless “abbreviated definite descriptions”. So, how do we move from the conclusion of the reduplication argument to a rejection of the thesis of descriptivism as described in the opening of this essay? For a pure descriptivist, the only form of aboutness is descriptive aboutness; there is a single mold for how any proposition whatsoever gets to be about a particular and that is via descriptive identification. A hybrid view like Russell’s, by contrast, thinks that there are multiple forms of aboutness. Russell argued for an extremely limited view of acquaintance-​based, direct thought about particulars that was grounded on a strict epistemic condition: one has acquaintance-​based, direct knowledge about a particular o only if misidentification of o is impossible and knowledge of o’s existence is certain. He concluded that the ordinary objects we perceive and interact with are not known by acquaintance because identity confusions and doubt are coherent. Sense-​data (and perhaps the self) are the only particulars that we think about directly. Everything else we think about descriptively. On Russell’s view, the canonical form of the thought that I have about my cat, for example, as I am sitting in front of him is as follows: [the cause of THIS (internally demonstrating my current sense-​ data had while perceiving my cat) is cute]. My thought, then, is singular with respect to my sense-​data and descriptive with respect to the object the thought is ultimately about, my cat. I can’t coherently doubt my sense-​data’s existence (while I am having it, which is the only time I can think about it directly) and cannot misidentify it. Strawson was more liberal, thinking that we can think directly (demonstratively identify, as he put it) external world objects. But he thought that perception (and memory relations) were necessary for direct thought. So, as I have had perceptual experiences of my cat but not Russell, I can think directly about my cat but not Russell. On an even more liberal view, I can think directly of Russell without having ever had perceptual contact with him in virtue of being part of a communication chain that originates in perceptual states caused by Russell; crudely, those who baptized Russell with his name perceived him and had perceptually based direct thoughts about him and then passed those direct thoughts on through communicative acts that ultimately led to my thoughts about Russell. The reduplication argument shows the necessity of direct thought and proves false a pure form of descriptivism. We now need to establish a set of criteria for when a thought is direct to decide between the conservative views like Russell’s and the more liberal views that accept communication-​ based direct thought. The ultimate plausibility of a descriptivist thesis about proper names turns on this broader set of topics. The standard anti-​descriptivism arguments are not due to Strawson but instead Saul Kripke, presented in his lectures of 1970 at Princeton University, later published as Naming and Necessity. Kripke developed three arguments against descriptivism: the modal argument, the semantic argument, and the epistemic argument. These three arguments are distinct from the line of argument developed above from the reduplication argument. 77

Michael Nelson

The first is the most famous and has proven the most influential. Kripke observed that the “great deeds” descriptions that are standardly associated with famous proper names—​‘Richard Nixon’ names the US president caught up in the Watergate scandal, for example—​typically involve contingent properties of their bearers—​Richard Nixon might not have been caught up in the Watergate scandal, either because he might not have been aggressively spying on his political adversaries or because he might not have been caught. But suppose, for reductio, that, for every proper name n, there is some “great deed”, contingent property F uniquely had by the bearer of n such that n and ‘the F’ are semantically equivalent.Then, ‘Necessarily, n is identical with n’ and ‘Necessarily, the F is identical with n’ are both true, as the first is true, given the necessity of identity, and the second differs from the first only in the substitution of one term with another that is semantically equivalent. But, the first is true and the second is false, as we have just seen that the bearer of n is only contingently F. So, for example, let’s suppose that the descriptive equivalent of the name ‘Bertrand Russell’ is ‘the most famous British anti-​war logician’. Then, as ‘Necessarily, Bertrand Russell is identical with Bertrand Russell’, the sentence ‘Necessarily, the most famous British anti-​war logician’ is also true, as it results from the first by substituting the synonymous, by hypothesis, ‘the most famous British anti-​war logician’ with ‘Bertrand Russell’ in the first. However, that is counterintuitive; surely it is possible that Russell instead supported the war, never went into mathematical logic or never earned much fame for his efforts, or that some other British logician both protested the war and was more famous than Russell. The semantic and epistemic arguments bleed into one another. However, they are different. The semantic argument turns on the observation that many speakers are competent with a name even though they do not know any of the bearer of that name’s great, individuating deeds and perhaps even wrongly think that the person did things that as a matter of fact s/​he did not do at all. If such speakers are nonetheless competent with the name and use it to refer to its bearer, then it is very implausible that the name is semantically equivalent with a great deeds description, as then there would be no slack between one’s understanding of the name and one’s knowledge of the bearer’s deed. The epistemic argument is similar to the modal argument in that it relies on substitution principles that claims of semantic equivalence seem to license. So, if n and ‘the F’ are semantically equivalent, then ‘It is knowable a priori that, if anything uniquely F-​ed, then the F is F’ and ‘It is knowable a priori that, if anything uniquely F-​ed, then n is F’ are both true, as they only differ in the substitution of one term with a semantically equivalent term. However, the first is true but the second is false. The first is true because the semantics of the definite description guarantees that the entity, if any, designated by a definite description satisfies the descriptive condition the definite article operates on. The second is false, at least if the property F is a “great deeds” property, as it is only empirical experience that justifies one in knowing, say, that Russell authored ‘On Denoting’, that Frege began the modern logicist tradition, etc. So, I know a priori that, if anyone uniquely authored ‘On Denoting’, then the author of ‘On Denoting’ authored ‘On Denoting’, but I only know through experience that it is Russell that authored ‘On Denoting’ … . And what’s more (and this is where the semantic and epistemic arguments bleed into one another) my understanding of and competence with the name ‘Bertrand Russell’ does not provide me with knowledge that Russell authored ‘On Denoting’. And the same considerations hold for any proper name and any great deed. The most famous response to the semantic and epistemic arguments is the causal descriptivist’s response. (Proponents of this view include Lewis (1984) and Kroon (1987).) The causal descriptivist takes a que from Kripke’s own causal “picture” of the semantics of proper names, according to which the reference of a use of a name is a function of the causal chains connecting that use with other uses and ultimately back to the “baptism” of the referent with that name (setting to the side cases in which the name is introduced by description). So, while my semantic competence with the name ‘Bertrand Russell’ by itself may not enable me to know that Russell authored ‘On Denoting’, it does, assuming the truth of the causal picture of names, put me in a position to know that Russell stands at the end 78

The Case(s) against Descriptivism

of the chain of uses of the name ‘Bertrand Russell’ of which this use is a part. So, as far as the semantic and epistemic arguments are concerned, ‘Bertrand Russell’ is semantically equivalent with the definite description ‘the person at the end of the chain of this use of “Bertrand Russell” ’. More generally, abstracting away from a reliance on the causal theory of reference, the metalinguistic descriptivist claims that, for every name n, there is a definite description of the form ‘the bearer of n’ that is its semantic equivalent. Such a version of descriptivism, as it eschews “great deed” descriptions, seems immune to the considerations brought to light by Kripke’s semantic and epistemic arguments. One objection to the metalinguistic descriptivist is that it threatens circularity. Part of the appeal of descriptivism is that it promises to explain reference: a name n refers to its bearer in virtue of the fact that that object satisfies the descriptive condition the name expresses.That is a substantive explanation if the descriptive condition is a great deeds condition, or at least is independent of semantic properties and does not simply piggy-​back off the name having a reference. (There are still things to explain, like how the name came to be associated with that descriptive condition. But the mechanism of reference, as it were, is explained.) But the metalinguistic descriptivist drains that explanation, as the descriptive condition associated with a name is simply the condition of being a unique bearer of the name. The causal descriptivist does better on this score, as it piggy-​backs off a substantive theory of reference determination. A second objection to both versions of descriptivism is that it seems that a person, say, a toddler, may be semantically competent with a name but not possess metalinguistic concepts and certainly may not apprehend the causal theory of reference and so not be in a position to understand the metalinguistic and causal descriptions claimed to be semantically equivalent with the name. But, if two terms a and b are semantically equivalent, then they have the same semantic content. Competence with a term involves in some way knowledge of its meaning and two terms with the same semantic content have the same meaning. So, if causal descriptivism is true, toddlers are in fact not competent with names or they in fact do believe, if only tacitly, the causal theory of reference. Neither option is terribly plausible. There have been three main descriptivist lines of response to the modal argument. The first two stem from Michael Dummett’s early engagement with Kripke’s work. (See Dummett 1973.) The first is the so-​called widescope response, according to which proper names are synonymous with definite descriptions that take mandatory widescope over any modal operators in sentences in which they occur. So, while the sentence ‘Possibly, something sitting is not sitting’, which we can semi-​formalize as follows: Possibly [something x (x is sitting and not: x is sitting)] is false (and incoherent), the sentence ‘Something sitting is possibly not sitting’, which we can semi-​ formalize as follows: Something x (x is sitting and possibly[not: x is sitting]) is not. The difference concerns the relative scope of the modal operator ‘possibly’ and the quantifier ‘something’. In the second, but not the first, the ‘x is sitting’ is evaluated at a world distinct from the world, the shift of which is induced by the possibility operator that occurs between, at which ‘not: x is sitting’ is evaluated. So, assuming that the descriptive equivalent of ‘Nixon’ is ‘the US president caught up in the Watergate scandal’, while the sentence ‘It is possible that the US president caught up in the Watergate scandal was not involved in the Watergate scandal’ is false, the sentence ‘The US president caught up in the Watergate scandal might not have been involved in the Watergate scandal’ is not. Kripke responded by pointing out that the phenomena supporting the modal argument can be formulated in terms of simple sentences, where scope does not meaningfully arise, at least if the scope of an operator occurrence is limited to the sentence in which it occurs. For example, the following dialogue seems true: 79

Michael Nelson

Nixon wasn’t caught up in the Watergate scandal. What I  just said is false but could have been true. The second response to the modal argument, also found in Dummett, rests on the claim that modal operators are not propositional operators but instead operate on what Dummett labels ingredient sense. So, two semantically equivalent expressions can embed differently under modal operators because modal operators are sensitive to something other than semantic content. On this version of Dummett’s response, then, ‘Nixon’ and ‘the US president caught up in the Watergate scandal’ are semantically equivalent but have different ingredient senses and so ‘Possibly Nixon wasn’t involved in the Watergate scandal’ is true while ‘Possibly the US president caught up in the Watergate scandal wasn’t involved in the Watergate scandal’ is false. To fully make sense of this response, we would need to provide an account of ingredient sense and explain how modal operators function on these values. That is beyond the scope of this short article. We can, however, illustrate the kind of idea by briefly considering the functioning of the term ‘so-​called’. Even though, to use Quine’s example, ‘Barbarelli’ and ‘Giorgione’ are, we will suppose, synonymous and even though the context is not overtly quotational, ‘Giorgione was so-​called because of his size’ is true while ‘Barbarelli was so-​called because of his size’ is false. The explanation is that ‘so-​called’ in part operates not on the semantic content of the subject expression but on the expression itself and so substituting synonyms affects the value of ‘so-​called’. Dummett’s second response rests on the idea that modals operate in similar ways. (Graeme Forbes (1990) used similar machinery, although without distinguishing between semantic content and ingredient sense, in his Fregean account of propositional attitude verbs. (See also Stanley 1997).) The third response to the modal argument, first suggested by Alvin Plantinga (see Plantinga 1978), is the rigidifier response, which has proved the most popular. Kripke’s most basic observation is that the descriptions most readily associated with proper names involve contingent properties of the bearers of those names. In Kripke’s terminology, an ordinary definite description is a nonrigid designator, designating different objects with respect to different possible worlds, while a proper name is a rigid designator, designating the same object with respect to every possible world (with respect to which it designates anything at all). Plantinga noted that it is quite trivial to turn, however, a nonrigid definite description into a rigid designator by “rigidifying” it with the actuality operator. So, while the (nonrigid) definite description ‘the US president caught up in the Watergate scandal’ designates different people with respect to different possible worlds, even designating, say, Kripke with respect to a possible world in which Kripke wins the presidency and then breaks into his political rival’s offices and gets caught, the rigidified definite description ‘the actual US president caught up in the Watergate scandal’ designates Nixon with respect to any possible world where Nixon exists regardless of what contingent doings Nixon does at that world and regardless of whether there is someone else caught up in the Watergate scandal at that world. So, the difference in modal embedding that Kripke highlights with his modal argument disappears when the descriptivist moves from ordinary definite descriptions to their rigidified counterparts. (For further discussion of Kripke’s critique of descriptivism, see Soames’s entry in this volume.)

Further Reading Bach, K., 1981. ‘What’s in a Name’, Australasian Journal of Philosophy, 59(4): 371–​386. Evans, G., 1973. ‘The Causal Theory of Names’, Proceedings of the Aristotelian Society, Supplementary Volume 47: 187–​208. García-​Carpintero, M., 2017. ‘The Mill-​Frege Theory of Proper Names’, Mind, 127: 1107–​1168. Nelson, M., 2002. ‘Descriptivism Defended’, Noûs, 36: 408–​435. Russell, B., 1911. ‘Knowledge by Acquaintance and Knowledge by Description’, Proceedings of the Aristotelian Society, 11: 108–​128. Searle, J., 1958. ‘Proper Names’, Mind, 67: 166–​173.

80

The Case(s) against Descriptivism Searle, J., 1983. Intentionality, Cambridge: Cambridge University Press. Soames, S., 2001. Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity, New York: Oxford University Press.

Note 1 The interpretation of Frege’s philosophy of sense is complex and disputed along many dimensions. Some, for example Gareth Evans in Varieties of Reference (Evans 1982), argue that Frege’s theory banned sense without reference, despite the many occasions where Frege apparently talks about meaningful senses that express thoughts but lack reference precisely because their subject term lacks a reference. Evans claims that Frege claimed that such sentences have mock thoughts as their contents; however, the example of the sentence ‘Odysseus was set ashore at Ithaca while sound asleep’ from ‘On Sense and Reference’ does not contain mention of mock thoughts or pretense, where he instead writes that the sentence “obviously has a sense” (p. 215).Tyler Burge in ‘Sinning against Frege’ (Burge 1979), argues that Frege is not a descriptivist, as he does not claim that ordinary proper names are synonymous with definite descriptions. Even if he can be interpreted as a descriptivist, he seems to have thought that ordinary proper names do not have unique descriptive equivalents, as his discussion in footnote 2 on p. 210 of ‘Sense and Reference’ shows. For a more detailed discussion of the relationship between Frege’s views on the sense of proper names and Kripke’s target in Naming and Necessity, see the entry in this volume by Dunbar and McLeod.

References Burge, T., 1979. ‘Sinning against Frege’, Philosophical Review, 88(3): 398–​432. Dummett, M., 1973. Frege: Philosophy of Language, London: Duckworth. Evans, G., 1982. The Varieties of Reference, Oxford: Oxford University Press. Forbes, G., 1990. ‘The Indispensability of Sinn’, Philosophical Review, 99(4): 535–​563. Frege, G., 1948 [originally published in German in 1892]. ‘Sense and Reference’, Philosophical Review, 57(3): 209–​230. Kaplan, D., 1989.‘Demonstratives’, in Almog, J., Perry, J. and Wettstein, H., eds., Themes from Kaplan, Oxford: Oxford University Press, pp. 481–​563. Kripke, S., 1980 [based on transcripts of a series of lectures presented in 1970]. Naming and Necessity, Cambridge, MA: Harvard University Press. Kroon, F., 1987. ‘Causal Descriptivism’, Australasian Journal of Philosophy, 65: 1–​17. Lewis, D., 1984. ‘Putnam’s Paradox’, Australasian Journal of Philosophy, 62(3): 221–​236. Plantinga, A., 1978. ‘The Boethian Compromise’, American Philosophical Quarterly, 15(2): 129–​138. Russell, B., 1905. ‘On Denoting’, Mind, 14(56): 479–​493. Stanley, J., 1997. ‘Names and Rigid Designation’, in Hale, B. and Wright, C., eds., A Companion to the Philosophy of Language, Oxford: Blackwell Press, pp. 555–​585. Strawson, P. F., 1959. Individuals: An Essay on Descriptive Metaphysics, London: Methuen.

81

6 FRUITS OF THE CAUSAL THEORY OF REFERENCE Scott Soames

The causal theory of reference arose from Saul Kripke’s attack in Naming and Necessity on descriptivist analyses of proper names. His target was the view that, like most meaningful expressions, proper names express concepts that determine their extensions. In the case of names, these were thought to be individual concepts the unique instantiators of which were their referents. Since singular definite descriptions also express such concepts, it seemed obvious that names must be synonymous with descriptions associated with them by users. Kripke’s attack on this view was an earthquake that shook the foundations of natural-​language semantics—​despite the fact that proper names play a very small role in the overall enterprise. It did so because it challenged a fundamental tenet of the then dominant conception of semantic theorizing, summarized as follows: The Meaning of a Non-​Indexical Expression E is, (a) what is understood by speakers who use E competently and so know what E means, (b) what E contributes to the compositionally determined meanings (semantic contents) of all sentential clauses in which E occurs, (c)  what determines the extension of E at world-​states, thereby contributing to modal truth conditions of sentences containing E, and (d) what E standardly contributes to illocutionary contents of uses of sentences containing E. When Kripke concluded that the meanings of names are never those of descriptions, while refusing to speculate about what names do mean, he rejected descriptive theories of names as being inconsistent with (c) above, arguing that the referent of a name at any world-​state is fixed by the causal-​ historical chain associated with it at the actual world-​state. When he extended this view to natural kind terms, the repudiation of standard semantic architecture became more serious and was resisted by causal descriptivists, who sought to preserve it by incorporating the causal-​historical theory of reference into their analysis names and natural kind terms.The story of the resulting dispute and both the positive insights and the further challenges to traditional semantic architecture to which it led is the story of the fruits of the causal theory of reference. The chief strategy for defending descriptivist theories of the meaning of names and natural kind terms was to rigidify alleged reference-​fixing descriptions, thereby avoiding Kripke’s modal argument. The chief strategy for defending descriptivist accounts of reference was to invoke descriptions other than those Kripke had shown to be untenable.These defenses were motivated by a striking idea; 82

Fruits of the Causal Theory of Reference

since facts, not magic, determine what a term refers to, it must be possible to formulate and verify a theory giving those facts.1 Since verification requires showing the referents of a term t at different world-​states match what we, ordinary users, would take t to refer to (as used there), a description extractable from the theory we implicitly follow, typically without being able to explicitly formulate it, must correctly fix t’s referent. In short, descriptivism, properly understood, can’t possibly fail. Although influential, this argument provoked suspicion. How could what is, in effect, an a priori argument tell us what form an empirical theory must take? Before addressing this perplexity, we must note another factor motivating resistance to Kripke. Millianism, which was the theory of meaning most congenial to his antidescriptivism, seems incomplete, problematic, or both. Surely, one is inclined to think, the referents of ‘Hesperus’, ‘Phosphorus’, and ‘Earth’, don’t exhaust their meaning; somehow, more content must be included. Moreover, the assertions made, and beliefs expressed, using the names don’t seem to survive when different coreferential names are used. Doesn’t this show that the meanings of names are often not their referents, and may sometimes be descriptive? Isn’t the case even stronger for natural kind terms? Surely, there is more to “knowing what they mean”, than simply being able to use them to designate the relevant kinds. Understanding the causal theory of reference, and its significance, requires addressing these concerns.

The Anti-​Descriptivist Treatment of Names in Naming and Necessity Kripke distinguished two versions of descriptivism—​one purporting to give the meaning and referent of a proper name for a speaker and one purporting only to do the latter.

Strong Descriptivism about Names The meaning of a name n for a speaker x is a description D associated with n by x; its referent at w is whatever uniquely satisfies D at w. Corollary: if D gives the meaning of n, then the propositions semantically expressed by n is F and D is F are the same, as are those expressed by if n exists, then n is D, and, if D exists, D is D. Since the latter is both necessary and knowable a priori, so is the former. In general, substitution of D for n in a sentence S preserves modal and epistemic profile.2

Weak Descriptivism about Names Descriptions associated with a name by speakers semantically fix its referent without giving its meaning. Once fixed at the actual world-​state, it remains the same at other world-​sates. Corollaries:  (i) speakers believe they can uniquely describe what they use names to refer to; (ii) when D semantically fixes the referent of n, then n refers to o iff o uniquely satisfies D; (iii) when D fixes n’s referent for A, A knows on the basis of semantic knowledge alone that if n exists (or existed) expresses a truth, then n is (or was) D also expresses a truth. If D fixes the referent of n, n is F and D is F express different propositions. n is F expresses a singular proposition analogous to that expressed by a use of that is F, said demonstrating o.3

Against Strong Descriptivism Having distinguished these versions of descriptivism, Kripke observed that proper names are rigid designators, even though the descriptions speakers typically associate with them—​which, he seemed to assume, are those they would most readily volunteer if asked “To whom/​what are you using n to refer?”—​aren’t.This is important because rigid and nonrigid designators make different contributions to the modal truth conditions of sentences containing them. If I were to say “Aristotle could have been so-​and-​so”, there is a single individual whose being so-​and-​so at any possible world-​state is 83

Scott Soames

sufficient for the truth of my remark and whose being so-​and-​so at some possible state is necessary for its truth. Since ‘Aristotle’ is rigid, the needed verifier doesn’t change from world-​state to world-​state. Kripke puts this by saying that a use of ‘Aristotle’ (at the actual world-​state) designates the same man at any world-​state at which he exists, and never designates anything else. This isn’t true of most well-​known descriptions we associate with the name—​e.g., ‘the greatest student of Plato’, ‘the father for formal logic’, or ‘the teacher of Alexander the Great’. Because they aren’t rigid, we can say that had the world been different, Aristotle would have existed without being the greatest student of Plato, the father of formal logic, or the teacher of Alexander. Since we can’t correctly say that had the world been different Aristotle would have existed without being Aristotle, none of those descriptions, no conjunction or disjunction of them and no nonrigid description generally, is synonymous with ‘Aristotle’. That, in brief, is Kripke’s modal argument. Can strong descriptivism be saved by rigidifying the putative meaning-​giving descriptions? One way of rigidifying uses the indexical operator actually of two-​dimensional modal logic. If S expresses proposition p at a context C (and an assignment of values to variables) Actually S expresses the proposition that p is true at the world-​state @ of C. So, when S is true at @, Actually S is necessary. When a nonrigid description the x: Gx uniquely designates o at @, the description the x: Actually Gx rigidly designates o at all world-​states possible from @ at which o exists and never designates anything else. Thus, if Kripke’s modal argument were the only problem for strong descriptivism, rigidifying descriptions in this way would save it. But there are other problems. Just as substituting nonrigid descriptions for names can change the modal profile of a sentence S, substituting ‘actually’-​r igidified descriptions can change the epistemic profile of S. This, in turn, leads to different modal-​profiles of belief reports differing only in such substitution. This is illustrated in (1), where ‘the x: Gx’ nonrigidly designates the bearer of name ‘n’. (1)  a.  Bill believes that n is F. b. Bill believes that the x: Gx is F. c. Bill believes that the x: actually Gx is F. Since ‘n’ rigidly designates o, the belief (1a) reports Bill as having will be true at w only if o is F at w. Because the description in (1b) isn’t rigid, the belief it reports doesn’t have this property. Since (1a,b) report different beliefs,‘n’ doesn’t mean what ‘the x: Gx’ does. (1c) also differs from (1a). For the belief reported by our use of (1c) to be true at w, Bill must believe (at w) that the unique individual who, at @, is G is F (at w), which is not required by (1a). Since substitution changes the epistemic profile of n is F, the x: actually Gx is F doesn’t give the meaning of n. Thus, names don’t mean the same as ‘actually’-​r igidified descriptions. Suppose, however, that a potentially meaning-​giving description D is rigidified using the indexical dthat operator of Kaplan (1989), which, when prefixed to D produces a singular term the semantic content of which (at the context) is the object denoted by D. In one way, this is just what the doctor ordered, since if genuine reference-​fixing descriptions of the kind sought by weak descriptivism can be found, then rigidifying with ‘dthat’ will avoid the modal argument, without succumbing to this objection.4 Still the price of this move is steep, since, without other potentially problematic semantic changes, rigidifying with ‘dthat’ reinstates Frege’s Puzzle, the avoidance of which has, historically, been the chief force motivating descriptivism.

Against Weak Descriptivism Unlike strong descriptivism, the purportedly reference-​fixing descriptions associated with names by weak descriptivism needn’t also determine their meanings. Hence, it is not defeated by the fact that substituting them for names may change the modal or epistemic profile of a sentence. Nevertheless, Kripke argued, it is refuted by arguments showing that, for most names (and natural kind terms), no 84

Fruits of the Causal Theory of Reference

semantic rule determines their reference to be whatever, if anything, uniquely satisfies descriptions speakers associate with them. Rather, we often successfully use a name to refer to its conventional bearer, even though the information we associate with it is incomplete or erroneous. For example, speakers who know of Cicero only that he was a famous Roman, or of Einstein only that he was a brilliant scientist, can use the names to refer to the men, despite realizing that they can’t describe them uniquely. Thus, it seems, the referent of their use of the names can’t be determined by any (uniquely identifying) description they associate with them. In other cases, one’s use of n refers to its bearer despite being guided by a description that denotes something else. Imagine a student whose use of ‘Thales’ is guided by the description “the pre-​Socratic who held that all is water”. Suppose some pre-​Socratic hermit unknown to anyone uniquely held that view, while a famous pre-​Socratic philosopher—​called by a name that has come down to us as ‘Thales’—​held a different view, which was misunderstood by his contemporaries and transformed in the passage through time into the view that all is water. Then, it seems, the student’s use of ‘Thales’ would refer to the famous philosopher, despite being guided by a description of the hermit, contrary to weak descriptivism. A different argument involves speakers who can uniquely describe the referent of their use of a name n, but only by relying on other names the referents of which they can uniquely describe only by using n. Consider x, whose only knowledge of Cicero is that he was the Roman orator who denounced the senator Catiline, and whose only knowledge of Catiline is that he was the senator denounced by the Roman orator Cicero. If x were given the referent of either name, x could descriptively fix the referent of the other. But x can’t do either, since all x knows is that some Roman orator denounced some Roman senator, which, we may assume, doesn’t determine the referent of either name. Nevertheless x uses ‘Cicero’ to designate Cicero and ‘Catiline’ to designate Catiline. Kripke mentions that some are in similar shape with ‘Einstein’ and ‘Relativity Theory’. Such circularity isn’t unusual. Most descriptions we might volunteer when asked about the reference of names (or natural kind terms) include other names (or natural kind terms), often producing widespread, interconnected dependencies, which, when considered together, don’t uniquely determine referents of any of the terms.The challenge to producing any unique, purely qualitative grounding of names and kind terms is, therefore, daunting.The idea that our entire network of such terms can be given a global reference-​determining descriptive grounding is an unsupported article of faith. Kripke did believe that in rare cases names are introduced by stipulating that they are to rigidly designate whatever is denoted by the description that introduces them—​as when one says “Let ‘Jack the Ripper’ name the person who committed such-​and-​such crimes”. In such cases, Kripke seemed to suggest, merely understanding n puts one in a position to know that n is/​was D (if anything was) is true—​where D was used in the stipulation. In such cases associating n with a particular descriptive content is taken (temporarily) as a common achievement of all who understand n. But nothing like this is true of ordinary names. For them, there are no privileged (reference-​determining) descriptions and no special semantic knowledge apart from that expressed by ‘n’ designates n. With an ordinary name, to understand it is just to use it to refer to the same thing other users do—​no matter how this is accomplished. Competence with such a name n involves having a referential intention determining something o as referent, and awareness that to use n is to say something of o. Although one can acquire the needed intention simply by speaking a language in which n already names o, and intending to use n as others do, this is a general fact about language applying to many kinds of words, not a special semantic fact about names (or natural kind terms).

The Historical Chain Model of Reference-​Fixing and Transmission After criticizing strong and weak descriptivism, Kripke offered commonsense platitudes pointing toward a positive theory of reference-​fixing. First, he observed, a name is introduced, either by 85

Scott Soames

stipulation—​“I name this ship ‘The Queen Mary’ ”, or “Let’s call him ‘Greg’ ”—​or by a recurring pattern of uses of a new phrase, e.g. ‘Green Lake’, to refer to some body of water. Once introduced, the name is used in conversation to refer to its bearer. New people pick it up intending to refer to whatever their sources did. The process continues, producing a chain of reference transmission. Sometimes descriptive content accompanies passing the name from one user to another. But this content needn’t be accurate or reference determining. Normally, reference is determined by the chain itself. If one who acquires a name intends to use it to refer to whatever one’s sources used it to designate, it may not matter which other beliefs one may have. Instead of those beliefs determining reference, often the chain itself partially determines the contents of one’s beliefs. In this way, referring and believing become, in part, community efforts. Kripke didn’t offer this commonsense sketch as a precise or comprehensive theory. For example, no attempt was made to answer Q. Q What, if anything, does x’s use of n designate if (a) x stands in a chain governing n grounded in o, (b) x intends to use n to refer to whatever grounds the chain, (c) x also intends to use n to refer to o* with which x is acquainted or to which x is connected by a significant description, (d) o ≠ o•, and (e) x wrongly believes that o* grounds the chain. The Madagascar example discussed in Evans (1973) and Berger (2002) is a case of this type. Aside from disputed historical details, the basic idea is that newcomers pick up the name ‘Madagascar’ from locals, intending to use it to preserve its previous reference, while wrongly assuming it had designated the big island off the east coast of Africa (rather than a portion of the continent). Because the island is what the newcomers wished to communicate about, it ultimately became the referent of the name. This suggests that although historical chains of reference transmission are weighty factors in determining the referents of uses of names, the dynamics of communicative situations also play a role. A genuine theory of reference determination, which Kripke didn’t offer, would make proposals about this, while marshaling evidence to verify it. There are various ways of coming to have a referential intention regarding a name that may conflict with the intention of preserving its previous reference, thus making reference assignments not determined by chains of reference transmission possible. Although Kripke didn’t tell us how to determine reference in these cases, he did provide an illuminating example on which one might build. Kripke (1979) discusses a case in which Mary and Janet see Smith in the distance, mistaking him for Jones. Mary says, “Look, Jones is raking leaves”, thereby saying something true about Smith and something false about Jones. Because of her conflicting intentions, her use of ‘Jones’ can plausibly be understood as referring to both. If Mary were to immediately add, “Run over and warn him to get inside; a tornado is about to hit”, her reference to Smith might be paramount. Depending on how the scenario develops, this could lead to Smith becoming the temporary, or permanent, referent of ‘Jones’ in the language of the community.5 Although I haven’t yet touched on parasitic descriptions that identify the referent of one’s use of n as being whomever or whatever certain others use n to designate, they don’t change the basic picture. Consider Kripke’s ‘Peano’ example. Most who have heard of Peano believe he discovered the now standard Peano Axioms of arithmetic. But he didn’t. Although he published them, he credited Richard Dedekind with the formalization. Since many who use ‘Peano’ to refer to Peano don’t know this, the description they most strongly associate with the name—​the one who first formalized arithmetic—​doesn’t designate their referent. Nor does the person to whom most people refer when they use ‘Peano’ unless most ‘Peano’ users have other non-​descriptive ways of picking him out. Even the person to whom most experts, i.e., mathematicians, refer when they use ‘Peano’ may not suffice. Since mathematics is a big field, there is no guarantee that most mathematicians are aware of the provenance of the Peano Axioms. What about the person to whom most users of the name attribute such-​and-​such axioms? That’s no 86

Fruits of the Causal Theory of Reference

good either, since it requires independently identifying to whom most people refer when they say, “Peano discovered those axioms”. Thus, we still have no way of saving weak descriptivism.

The Architecture of Causal Descriptivism about Names Despite these problems, dedicated descriptivists like David Lewis thought that descriptivism’s salvation was lying in plain sight. Did not Kripke and his allies refute the description theory of reference, at least for names of people and places? … I disagree. What was well and truly refuted was a version of descriptivism in which the descriptive senses were supposed to be a matter of famous deeds and other distinctive peculiarities. A better version survives the attack: causal descriptivism. The descriptive sense associated with a name might for instance be ‘the place I have heard of under the name “Toromeo” ’ or maybe ‘the causal source of this token: Toromeo’, and for the account of the relation being invoked here, just consult the writings of causal theorists of reference.6 Lewis’s thought builds on Kripke’s picture of reference transmission in which speakers pick up proper names (and other kinds of words) and use them to designate what others do. Because speakers routinely use language to communicate, they know this. They realize that many of their words have somehow inherited their reference from that of others, without knowing precisely how this occurs.7 A sophisticated speaker might even associate n with the description D—​the referent of those uses of n from which my present use of n somehow inherits its reference. But D doesn’t semantically fix the referent of her use of n. For D to do that, satisfaction of D must be the mechanism by which her reference to o is established, in which case the fact that o satisfies D must be the reason she refers to o when using n. It isn’t. On the contrary, the pattern of explanation is the opposite. The only reason D picks out o as her referent is because some other, unidentified, process has already determined that her use of n refers to o. When circularity-​generating descriptions like D are avoided, the best we can normally provide are descriptive approximations that pick out the right referent in some but not all cases.Thus parasitic descriptions don’t save weak descriptivism.8 Is there another way of insuring the existence of descriptions required by weak descriptivism? Recall the argument from Jackson (1998b), summarized in the first paragraph of this chapter. He says: If speakers can say what refers to what when various possible worlds are described to them [in Kripkean thought experiments used to refute proposals about which descriptions fix referents of which names], description theorists can identify the property associated in their minds with, for example, the word ‘water’ [or with the name ‘Thales’]: it is the disjunction of the properties that guide the speakers in each particular possible world when they say which stuff [or which individual], if any, counts as water [or Thales].The disjunction is in their minds in the sense that they can deliver the answer for each possible world when it is described to them in sufficient detail, but it is implicit in the sense that the pattern that brings the various disjuncts together as part of the possibly highly complex, disjunction may be one they cannot state.9 This argument, which is meant to apply to names and natural kind terms, purports to show that each has its reference fixed by a description semantically associated with it by users, even though no particular descriptions have been shown to do so. Surprisingly, the argument is based on Kripkean demonstrations that familiar candidate descriptions don’t determine reference. Because, Jackson thinks, we have reliable intuitions about what terms we understand designate in the imaginable scenarios that Kripke relies on in his anti-​descriptivist demonstrations, we must be guided by internalized descriptions determining their referent in all scenarios. 87

Scott Soames

There are three main problems with this argument. First, the ability to refute particular descriptivist analyses by finding, for each proposed description D, at least one world-​state w at which the referent of n (as used by speakers in w in the way we use n at @) differs from the denotation of D at w, does not presuppose that we have the ability, required by Jackson’s argument, to correctly identify, for every possible world-​state, what, if anything, n designates there. It is perfectly possible, I think likely, that some scenarios involving conflicting referential intentions (of the sort illustrated above) would not result in uniform judgments by competent users of the name, even when the name does designate something. It is a familiar fact about theorizing that the best account of clear instances of an important linguistic phenomenon often generalize to initially unclear cases in ways that ordinary competent speakers don’t uniformly recognize. Although our theories rest on a base of clear pretheoretic judgments, no defensible methodology limits correct extensions of central linguistic concepts to those that ordinarily competent speakers reliably recognize to be so. The second problem is the argument’s need to aggregate descriptions extractable from individual judgments of what n, as used in w, refers to into a single reference-​fixing description that specifies what n designates at each world-​state (at which it is used). If the master description MD for n is to vindicate weak descriptivism, it must be non-​circular in two senses. First, satisfaction of MD, at any w must be the mechanism by which the referent of n at w is established. Second, MD can’t contain any other term for which descriptivism is supposed to hold, unless the referent of that term, at any world-​state, can be determined entirely independently of that of n. Somehow this must be guaranteed for all names, natural kind terms, and related expressions. To do this, Jackson must maintain either (a) that for each name, natural kind term, or related expression, we can correctly determine its referent in possible scenarios described without using any such terms, or (b) that there are some such expressions for which we can do this, which can then be used in describing world-​states in ways that allow us to extract non-​circular reference-​fixing descriptions for all remaining names, natural kind terms, and related expressions, or (c) that using all our pretheoretic intuitions about what refers to what in different scenarios, we can extract a single statement, expressible without any names, natural kind terms or related expressions, that fixes the referents of all such expressions in all worlds simultaneously. It has never been shown that this can be done. The third problem with Jackson’s argument is that he, like other weak descriptivists, mistakes the semantic question What does a term mean, and what do uses of it designate? with the presemantic question How did the term initially come to mean and refer to what it does, and how is that meaning and reference maintained? If n is a name, there will be an initial use, or series of uses, by one or more individuals to pick out a referent o, followed by a chain of reference transmission in which subsequent uses inherit the reference and semantic content of earlier uses. What does this have to do with semantics? That n designates o, which is also its semantic content, is a semantic fact—​a linguistic convention. For each different name in the language of a linguistic community, there is a different convention of this sort plus interpretive conventions governing other words, phrases, and sentences. The semantics of a language is a set of such conventions. By contrast, the fact that this, or that, individual originally introduced n to name o, after which n was passed down a historical chain to others without changing reference is a presemantic fact about how n came, and continues, to have the reference it does. It is neither a convention nor something present speakers need to know. Since they are familiar with how language is used to communicate, they will take it for granted that uses of names and many other words often inherit their reference from earlier uses. This is a general fact about linguistic communication, not a semantic fact about their language.10 The failure of causal descriptivists to vindicate weak descriptivism wasn’t their only problem. Taking themselves to have succeeded in vindicating descriptive reference-​fixing, Jackson and Lewis needed to rigidify their descriptions to match the modal profile of names, and to bring the epistemic profile of sentences containing their descriptions with those containing names. As suggested above, 88

Fruits of the Causal Theory of Reference

and argued in detail in ­chapter 2 of Soames (2002) and ­chapters 5–​10 of Soames (2005), this strategy fails at every step.

Natural Kind Terms Next a word extending the anti-​descriptivist account of names to natural kind terms like ‘water’, ‘tiger’, ‘green’, and ‘heat’, which apply to various objects, or quantities of stuff. Depending on their grammatical categories, these terms, or their close relatives, combine with the copula to form predicates—​‘is water’, ‘is a tiger’, ‘is green’, and ‘is hot’. In the case of the last of these, there is also the comparative form ‘is hotter than’ which is true of a pair the first of which contains more heat (molecular motion) than the second. Each of the four general terms designates a kind. What is the difference between natural kind terms and other terms? Though Kripke says little about this, an answer can be abstracted from his examples. Like names, natural kind terms are not synonymous with descriptions associated with them by speakers. Their introduction and transmission are also similar to that of names. Just as ordinary names are often introduced by stipulating that they are to refer to certain individuals with which one is acquainted, so natural kind terms can be introduced to designate kinds with which one is acquainted through their instances. For example, we may imagine ‘water’ and ‘green’ being introduced by the following stipulations: Let ‘water’ designate the property possessed by (nearly) all members of a certain class of samples that explains their most salient characteristics—​e.g., the fact that they boil and freeze at certain temperatures, that they are clear, potable, and necessary to life, etc. The predicate ‘is water’ applies (at a world-​state) to precisely those quantities that have the physical constitution which, at @, explains the salient features of (nearly) all our samples. Let ‘green’ designate the property possessed by (nearly) all members of a certain class of sample object surfaces that is causally responsible for the fact that they appear visually similar to us (and different from other surfaces). Thus, ‘is green’ applies (at a world-​state) to all and only those objects the surfaces of which have the characteristic that, at @, causally explains why our samples look similar to us (and different from other samples). These stipulations are, of course, idealized. ‘Water’ wasn’t explicitly introduced this way, but it could have been, and it behaves pretty much as if it had been. Presumably, speakers simply started calling certain quantities ‘water’, intending it to apply, not only to samples they had encountered, but to all other instances “of the same kind”. They noticed important properties P1…Pn of nearly all quantities they had been calling ‘water’, which they assumed to have a common unifying explanation, which, they thought, made them instances of the kind of stuff they are. Hoping to track this common element, they used ‘water’ to designate the property (which they then had no more informative way of designating) possession of which explained the fact that nearly all their samples had P1…Pn. Because their explanatory assumption turned out to be correct, the payoff was a term that can be used in counterfactual explanations and law-​like generalizations in ways that a mere conjunction of terms designating P1…Pn cannot. Other general terms—​e.g.‘tiger’,‘gold’,‘heat’, and (by extension) ‘hot’—​fit this picture. So does ‘green’, which designates a family of specific shades. The samples associated with the word are objects each of which is an instance of one of the shades. What the samples have in common is that they are judged by humans with normal color vision to look similar to one another (in various contexts and lights), and different from objects of other shades outside the family (in similar contexts and lights).‘Green’ designates the property of object surfaces that, at @, explains the perceived differences or similarities. So far, I have spoken only of how natural kind terms are introduced. Once introduced, they are typically passed from speaker to speaker in a chain of reference transmission, just as names are. Hence, 89

Scott Soames

it’s not surprising that arguments against descriptive analyses of the meaning or reference of natural kind terms largely parallel those against descriptive analyses of names. The most telling arguments against weak (and thereby against strong) descriptivism for natural kind terms are based on our ability to use them correctly despite the ignorance, error, and incompleteness of our descriptions of them. Most of us routinely use many natural kind terms that we can’t uniquely and non-​circularly describe. In my case these include ‘bauxite’, ‘tungsten’, ‘obsidian’, ‘wildebeest’, ‘sycamore’, ‘pulsar’, ‘insect’, and many more. Even when we can describe a natural kind, as (perhaps) in (2), our descriptions aren’t synonyms for the kind terms. (2) Light is the cause of visual experience. Heat is molecular motion. Humans are primates that are not apes, monkeys, or lemurs. Carbon dioxide is the chemical element extracted from the air and used by plants in photosynthesis. Moreover, although some of the propositions in (2) are necessary, none are a priori. In short, most of Kripke’s arguments against descriptive analyses of proper names carry over to similar analyses of natural kind terms.

Saving Important Insights of Causal Descriptivism Kripke was right that descriptions don’t give the meanings, or semantically fix the referents, of most ordinary names and natural kind terms. Post-​Kripkean nondescriptivists were right that their semantic contents are the objects or kinds they designate. Causal descriptivists were right that there is more to the meanings of these terms than those objects or kinds. If that sounds incoherent, it is probably because you are thinking, along with leading proponents on both sides of the debate, that meaning (for non-​indexical expressions)—​i.e. what competent speakers understand—​and semantic content, which either is or determines reference, are one and the same. They are not. What must be recognized, but in most quarters has not been, is that meaning, in the sense of what is understood and widely presupposed by competent language users, can affect the contents of assertions, and the beliefs expressed by them, without affecting semantic content.11 The semantic content of a term is what it contributes to the compositionally determined semantic contents of all sentential clauses in which it occurs, including those governed by modal operators (e.g. necessarily, possibly). This, in the case of names and natural kind terms, is the object or kind the term designates. But there is more to understanding certain special names and many ordinary natural kind terms than simply being able to use them to designate their semantic contents. For example, one who uses ‘Hesperus’ and ‘Phosphorus’ is expected know that users typically presuppose that ‘Hesperus’ stands for something visible in the evening and ‘Phosphorus’ stands for something visible in the morning. One who mixes this up misunderstands the names. Those who know enough to use the names are aware of this. With this in mind, suppose that A assertively utters (3) addressing B in a context in which they commonly presuppose that each understands the names. (3) Hesperus is Phosphorus. Although A thereby asserts the trivial proposition that Venus is Venus, which is the semantic content of (3), B extracts more information from A’s utterance. Presupposing that A understands the names, B reasons that A knows she will be taken to be committed to the claim that the unique object that is both Hesperus and visible in the evening is the unique object that is both Phosphorus and visible in the morning. Knowing that A expects B to see this, he correctly concludes that A asserted the descriptively enriched proposition. 90

Fruits of the Causal Theory of Reference

Although this proposition is contingent, A’s use of (4) asserts something true, and nothing false. (4) It is necessarily true that Hesperus is Phosphorus. The reason no falsehood is asserted is that what understanding the names requires is knowing that most agents who use them take, and expect others to take, ‘Hesperus’ to stand for something (actually) seen in the evening and ‘Phosphorus’ to stand for something (actually) seen in the morning. Since taking the names to refer to things actually seen at certain times tells one nothing about when they are seen at other possible world-​states, A and B don’t descriptively enrich the occurrences of the names under the modal operator in (4). A similar account applies to (5). (5) Water is H2O. We may imagine that one who understands (5) uses ‘is’ to stand for identity, ‘water’ to rigidly designate a natural kind k, and the name ‘H2O’ (which is related to, but semantically distinct from, the phrase ‘the substance molecules which consist of two hydrogen atoms and one oxygen atom’) to rigidly designate k. Since ‘water’ and ‘H2O’ have the same semantic content, the semantic content of (5) is the triviality that k = k, which isn’t what people intend to communicate. In most cases in which (5) is used, speaker-​hearers mutually presuppose that they understand the terms. In such cases, a speaker A  asserts the proposition that k  =  k, plus another proposition. Presupposing that A  understands the terms, B reasons that A  knows that she will be taken to be committed to the claim that k is both a chemical compound involving hydrogen and oxygen (which is what understanding ‘H2O’ requires) and one instances of which are clear and potable, necessary for life, and found in lakes and rivers (which understanding ‘water’ requires). Realizing that A expects him to so reason, B correctly concludes that A  asserted this informative, descriptively enriched proposition. The explanation of why, nevertheless, Necessarily water is H2O is true parallels the explanation of why Necessarily Hesperus is Phosphorus is true. The general lesson here is that our pretheoretic conception of meaning incorporates both elements of what is ordinarily called understanding and what theorists call semantic content. Understanding is a graded term. One can understand an expression or sentence more or less well. In addition to coming in degrees, understanding is also context sensitive; e.g., what counts as (adequately) understanding technical terms depends on the group or setting in which the language is used. All of this points to the need for richer and more integrated semantic-​pragmatic theories of the contents of illocutionary acts than were common in the heyday of disputes over causal descriptivism. If, as seems undeniable, asserted content arises from semantic contents plus contents of widespread presuppositions associated with understanding, then a more nuanced distinction between semantic content and illocutionary content is needed.12 This is only the beginning of a much longer story incorporating new analyses incorporating representationally identical but cognitively distinct propositions distinguished by the presence or absence of fine-​g rained cognitive, non-​Fregean, modes of presentation, new accounts of propositional attitudes sensitive to such differences, and new accounts of the relationship between sentences and the multiple propositions that single, unambiguous sentences may simultaneously express, and that single utterances of them may assert.13 This longer story is needed, in part, because the semantic-​ pragmatic frameworks in which the dispute between descriptivists and antidescriptivists took place were too impoverished to accommodate the respects in which each side was right, in which each went wrong.

91

Scott Soames

Further Reading Berger, Alan (2002) Terms and Truth. Cambridge, MA: MIT Press. Jackson, Frank (1998b) “Reference and Description Revisited”, Philosophical Perspectives 12: 201–​218. Kripke, Saul (1979) “Speaker’s Reference and Semantic Reference”, in Peter French, Theodore Uehling, and Howard Wettstein, eds. Contemporary Perspectives in the Philosophy of Language. Minneapolis, MN: University of Minnesota. Soames, Scott (2015) Rethinking Language, Mind, and Meaning. Princeton, NJ and Oxford:  Princeton University Press. Soames, Scott (forthcoming) “Philosophy of Language in the Twenty-​first Century”, in Piotr Stalmaszczyk, ed. The Cambridge Handbook of Philosophy of Language.

Notes 1 Frank Jackson (1998a, 82) puts things this way. 2 ‘D’ and ‘n’ are metalinguistic variables. The square quotes are called “corner” or “Quine” quotes. E.g., if ‘P’ and ‘Q’ are variables over sentences, the sentence For all sentences P and Q P & Q is a sentence that says For all sentences P and Q, the expression that consists of P, followed by ‘&’, followed by Q is a sentence. Similarly, For all sentences P, ‘P’ is true iff P says For all sentences P, the expression that consists of the left hand quote mark, followed by P, followed by the right hand quote mark, followed by ‘is true iff’, followed by P is true. 3 Kripke (1980, 57). 4 On this analysis, strong descriptivism maintains that both the meanings and semantic contents of names are the same as the meanings and semantic contents of dthat-​rigidified descriptions, though since both names and such descriptions are taken to be indexical, the meanings and semantic contents of each are distinct. 5 Considerations like these must be born in mind when people speak of different “intuitions” about what uses of names refer to when presented with Kripke-​style scenarios.The questions must be put carefully, indicating conversational dynamics. Even Kripke’s ‘Godel’-​‘Schmidt’ scenario (in which the one baptized ‘Godel’ steals the work of the one baptized ‘Schmidt’) can be contextualized to elicit the judgment that a given use of ‘Godel’ refers not to the one grounding the ‘Godel’ chain but to Schmidt. This can happen in a scenario in which what is being discussed is why at a certain stage of the work “Godel” made one argumentative move rather than another. Such cases don’t falsify Kripke’s platitudinous historical-​chain picture. 6 Lewis (1999, 332 n22). 7 For further discussion see Soames (2003, 366–​371) and Soames (2005, 299–​302). 8 Lewis missed this because he failed to see the circularity of his description ‘the place I have heard of under the name “Toromeo” ’. A similar example is given by Jackson (1998b, 210), which is discussed by Soames (2005, 186–​188). 9 Jackson (1998b, 212), my emphasis. 10 This point is discussed more fully in response to Jackson (1998a) by Soames (2005, 182–​184). 11 See ­chapter  4 of Soames (2015) and also my “Philosophy of Language in the Twenty-​first Century” (forthcoming). 12 See ­chapter 7 of Soames (2010) for some tentative first steps. 13 For an introduction, see Soames (forthcoming); for more details see ­chapters 2–​8 of Soames (2015).

References Berger, Alan (2002) Terms and Truth. Cambridge, MA: MIT Press. Evans, Gareth (1973) “The Causal Theory of Names”, Aristotelian Society Supplement 47: 187–​208. Jackson, Frank (1998a) From Metaphysics to Ethics. Oxford: Oxford University Press. Jackson, Frank (1998b) “Reference and Description Revisited”, Philosophical Perspectives 12: 201–​218. Kaplan, David (1989) “Demonstratives”, in Joseph Almog, John Perry, and Howard Wettstein, eds. Themes from Kaplan. New York: Oxford University Press. Kripke, Saul (1979) “Speaker’s Reference and Semantic Reference”, in Peter French, Theodore Uehling, and Howard Wettstein, eds. Contemporary Perspectives in the Philosophy of Language. Minneapolis, MN: University of Minnesota Press. Kripke, Saul (1980) Naming and Necessity. Cambridge, MA: Harvard University Press. Lewis, David (1999) “Naming the Colors”, Papers in Metaphysics and Epistemology. Cambridge:  Cambridge University Press. Soames, Scott (2002) Beyond Rigidity. New York: Oxford University Press.

92

Fruits of the Causal Theory of Reference Soames, Scott (2003) Philosophical Analysis in the Twentieth Century,Vol. 2. Princeton, NJ and Oxford: Princeton University Press. Soames, Scott (2005) Reference and Description. Princeton, NJ and Oxford: Princeton University Press. Soames, Scott (2010) Philosophy of Language. Princeton, NJ and Oxford: Princeton University Press. Soames, Scott (2015) Rethinking Language, Mind, and Meaning. Princeton, NJ and Oxford:  Princeton University Press. Soames, Scott (forthcoming) “Philosophy of Language in the Twenty-​first Century”, in Piotr Stalmaszczyk, ed. The Cambridge Handbook of Philosophy of Language.

93

7 THE PROBLEM OF REFERENCE CHANGE Harold Noonan

Reference Change and Reference Preservation Reference is the relation between a word—​proper name, common noun or description—​and the item in the world we use it to talk about. Both reference change—​the same word being used by later users to whom it has been passed on to refer to something different—​and reference preservation—​ the word retaining its reference when transmitted to other users—​are possibilities (rather, everyday actualities) which philosophers of language must accommodate. Both have figured prominently in the 20th-​century debate between descriptivists and their opponents. For descriptivists, inspired by and developing ideas from Frege and Russell, reference change has not appeared as a threat; rather the challenge has been to develop an account of reference which allows preservation of reference through changes in knowledge and belief which affect the descriptions speakers associate with names. A descriptivist account which requires every name to be associated with a reference-​determining description, or cluster of descriptions, is challenged by the evident fact that a name’s reference can remain the same despite the users no longer knowing things that were once well-​known and gaining many erroneous beliefs. In an obvious way legends and tall stories can accumulate about a famous figure, and the less interesting truths about him be forgotten, so it may be that present-​day users are still talking about the person who was the original reference of the name (King Arthur or Jonah or Goliath) despite complete ignorance of his activities and widespread erroneous belief. Ignorance and error are no bars to reference, it seems, and hence the onset of ignorance and the accumulation of error are likewise no bars to reference preservation. The opponent of descriptivism, following on from Kripke (1972), has no problem with explaining such preservation of reference. What has been perceived to be a problem for him is rather to explain how reference can change, without any intention that it does so, just because later users associate different information—​different descriptions—​with a name from those which earlier users associate with the name. The most famous example of such reference change in the philosophical literature is that of ‘Madagascar’, introduced by Evans (1973) as a challenge to a causal theory of names, and discussed inconclusively, by Kripke and Dummett (in Davidson et al. 1974; in Harman et al. 1974). ‘Madagascar’, or some earlier form of the name, was originally a name of part of the African mainland and became, without anyone’s intending a change of reference, via a process of transmission from African natives to Europeans, in which Marco Polo was involved (actual historical details are unclear, see Burgess 2014; Taylor 1898), a name of the great African island. The challenge to the opponent of 94

The Problem of Reference Change

descriptivism is to explain how the reference change could have happened, consistently with their anti-​descriptivist emphasis, which allows them so easily to explain how a name can preserve reference through radical change in the associated descriptions. The focus of this discussion will be on the apparent problem reference change presents to the Kripkean and the Madagascar example specifically. The conclusion will be that the problem the example highlights is a genuine one to which there is no obvious solution for the anti-​descriptivist.

Kripke and His Opponents Already in ‘Naming and Necessity’, Kripke acknowledged that two uses of the same word can, in some sense, be part of the same chain without preserving reference, if only because a user of the name might not intend to use it in the way, he believes, it was used by the person from whom he acquired it. He notes that if he decides to name his pet aardvark ‘Napoleon’ the reference of the name has changed: preservation of reference requires not only (if at all) a causal link between later users and earlier users but at least an absence of an intention to change reference. In the Addenda he reinforces the point by noting the possibility of a change of reference from something real to something fictional, as in the case of ‘St Nicholas’ (‘Santa Claus’). He then adds: ‘According to Evans, “Madagascar” was a native name for a part of Africa. Marco Polo, erroneously thinking that he was following native usage, applied the name to an island’ (Kripke 1972, 768). The key feature of the example is that the change of reference was inadvertent; no one ever intended it. Indeed, there was always an intention to preserve reference. In the Addenda Kripke briefly considers how to accommodate this and suggests that the social character of naming is crucial. But he does not go into further details. The possibility of reference change is also discussed in Dummett’s first reply to Kripke (Dummett 1973, 148–​151). He says that we can sometimes establish with reasonable probability that a name has been transferred from one bearer to another. But he says: ‘Kripke’s account leaves no room for the occurrence of a misunderstanding, since to speak of misunderstanding would presuppose that the name did in fact have a sense which could be understood’. Dummett illustrates the possibility of reference change by explaining that there is a German card game now called ‘Tarock’, which is different from the game called ‘Tarock’ played with the Tarot pack once played in Germany and still played in Austria. This change in reference in Germany may well have been inadvertent, Dummett suggests. At any rate, it is perfectly intelligible that it should have been, since we have a criterion for determining what game a name is being used as a name of independently of the historical origin of the name. Dummett’s complaint (1973, 150–​151) is that Kripke’s story cannot accommodate such reference change without becoming vacuous: Kripke expressly wishes to allow that the association with a name of a description which in fact does not apply to the person … for which the name was originally introduced does not deprive that name of reference to that person … it merely reveals a false belief about the referent.There is therefore no room in Kripke’s account for a shift of reference in the course of a chain of communication … Intuitively … there is no such guarantee [of preservation of reference] … Once this is conceded, the account crumbles away altogether. We are left with this: that a name refers to an object if there exists a chain of communication … at each stage of which there was a successful intention to preserve its reference. This proposition is indisputably true; but hardly illuminating. Dummett’s thought is that if no mention of ‘success’ is included we do not have a sufficient condition for preservation of reference. Whereas, if it is included, we have such a condition only trivially, because of the transitivity of identity. If X refers to A with N and Y’s intention is to refer to what X refers to then if Y’s intention is successful Y does refer to what X refers to, i.e., A, and so it goes on. 95

Harold Noonan

The mention of a causal historical chain falls away as irrelevant. Dummett elaborates this point in his contribution to a discussion with Kripke published in Synthese (Harman et al. 1974, 516–​517). Kripke (Harman et al. 1974, 518) replies by acknowledging that One might think, by some kind of transitivity … that if I refer to the same thing as he, and he refers to the same thing as me and so on, then it all goes back to the same things as the original man. But, he explains, this isn’t true because more than one intention of reference may be transmitted, even type of reference … utterance reference and so on. That’s where the picture needs to be qualified. It isn’t a matter of success or lack of success, but where he has a conflict of intentions … you may transmit only one point of conflict to the next man and that may be what happens. He illustrates what he has in mind with the famous case of Smith and Jones. I see someone raking leaves. In fact, he is Jones, but I take him to be Smith, so I say, ‘Smith is raking leaves’ (or ask ‘Why is Smith raking leaves?’ or whatever). I am using the name ‘Smith’, even here, as a name for Smith. If I am apprised of my mistake I will withdraw and admit that it isn’t Smith who is raking leaves. Smith is the semantic referent of the name. On the other hand, the primary interest of myself and my conversational partner may be the activities of the man raking leaves, i.e., Jones. So Jones is the utterance referent of the name. ‘The semantic referent of a designator’, Kripke explains, is given by a general intention to refer to a certain object … The speaker’s referent is given by a specific intention, on a given occasion, to refer to a certain object. If the speaker believes the object he wants to talk about … fulfils the condition for being the semantic referent, then he believes that there is no clash between his general intentions and his specific intentions. 1977, 264 Referential uses of singular terms, such as the use of ‘Smith’ in the Smith/​Jones scenario, are ones in which the speaker has a complex intention: ‘He has a specific intention, which is distinct from his general intention, but which he believes, as a matter of fact, to determine the same object as the one determined by his general intention’ (1977, 264). Kripke explains using this apparatus how he thinks reference shifts can inadvertently occur. A particular utterance intention (the utterance intention to refer to Jones, or the island, or the card game currently played in Germany) becomes a very general habitual intention. Our habitual primary intention is to refer to the utterance reference. If so, then the original semantic intention is over-​ridden by another semantic intention which has arisen out of the utterer’s reference becoming habitual and common among the speech community. And then tracing back the chain before this shift has occurred is illegitimate. Harman et al. 1974, 515 Kripke adds that one important thing that makes such shifts occur, is that this is what is most important in communication now. He emphasises that one reason why he thinks the ‘historical theory’ is true (he does say this, although he of course insists in ‘Naming and Necessity’ that he is offering at best a picture not a theory) and why the intention to use the word as you’ve heard it is primary is that ‘we 96

The Problem of Reference Change

use names to talk to each other, and if you’re using it with a different referent from the hearer just because of your own private beliefs, you’re not going to communicate with him’.Therefore, ‘we have a strong bias in favour of keeping the referent fixed even if our own beliefs about the object become erroneous’ (Harman et al. 1974, 515–​516). However, notwithstanding this motivation for allowing the intention to refer to what was previously referred to by the name to dominate, it can happen that this is overridden. It may be that if the speakers form another common intention in their speech community habitually to refer to that by the name which originally arose because of an error, then that is so primary in communication that it overrides what we would get to from a traceback. Harman et al. 1974, 516 Kripke acknowledges that this is vague and tentative but adds that it is what he has to say about this. However, it is not clear that Kripke need say more. After all, his primary concern is just to refute the description theory. So, as he has emphasised, notwithstanding his using the phrase ‘historical theory’ (in Harman et al. 1974, 513), he is not giving a theory of reference to replace the description theory, but merely a picture. So he is not intending to give a necessary and sufficient condition, or even a merely sufficient condition, for reference preservation. He is committed at most to the claim that a name typically refers to the thing its introducer baptised as long as there is not, as in the ‘Napoleon’ case, a conscious opting out of the initial referential practice and not, as in the Madagascar case, a widespread communal intention to which the users conform, which is in fact an intention to refer to something other than the thing initially baptised. But, of course, this claim, as Dummett insists, is consistent with the description theory, since it is consistent with the thesis that any user of a name must associate some identifying description with it that designates its bearer. So the question still remains, even if the description theory is refuted by Kripke, so long as it is allowed that some non-​trivial sufficient condition of reference preservation must exist, how any non-​descriptivist account can provide one. This is the question Gareth Evans addresses in ‘The Causal Theory of Names’ (1973).

Evans Evans’s paper has two parts.The first is an attack on an explicitly causal theory of reference and reference change, which he, of course, is cautious in ascribing to Kripke (though he does so). The second part is an attempt to develop an eclectic theory, bringing together the best element of the causal theory with the best element of the description theory (which he takes to be false—​ though not, in fact, refuted by Kripke’s arguments as given (Evans 1973, 196)), and to show how it can account for reference change. First, I shall look at Evans’s critique of the causal theory and then turn to his own eclectic theory. I think the key question addressed in his paper is whether any non-​descriptivist theory can account for reference change. If Evans’s eclectic theory can do so and can also be distinguished from a descriptivist account this is so. If not, Evans provides no reason to think that referential change can be accommodated by non-​descriptivist theorists. This is what I will conclude. The causal theory Evans targets is what one would extract from Kripke if one were to set aside his modestly cautious remark that he is only providing a picture. Evans’s statement of it is the following (1973, 191): A speaker, using a name ‘NN’ on a particular occasion, will denote some item x if there is a causal chain of reference-​preserving links leading back from his use on that occasion ultimately to the item x itself being involved in a name-​acquiring transaction such as explicit dubbing or the more general process whereby nicknames stick. I mention the notion of a reference-​preserving link to incorporate a condition that Kripke lays down; a speaker S’s 97

Harold Noonan

transmission of a name ‘NN’ to a speaker S′ constitutes a reference-​preserving link only if S [sic] intends to be using the name with the same denotation as he from whom he in his turn learned the name. With the typo (the second occurrence of ‘S’ should be ‘S′’) corrected, this states a sufficient condition for a name, on a particular occasion, denoting an item x. If any sufficient condition for reference preservation can be extracted from ‘Naming and Necessity’ it is this one. Evans (1973, 192) rejects it because it ignores the importance of surrounding context, and, he says, regards the capacity to denote something as a ‘magic trick’ which has somehow been passed on, and once passed on cannot be lost. He insists that what a name denotes depends upon what its user’s dispositions are bent towards. Hence what it names can change as the context changes, old dispositions are lost and new ones come into existence. His story (1973, 192)  about the pub conversation about Louis is intended to illustrate this. Conformably with Kripke’s ideas if I overhear a conversation in a pub about a certain ‘Louis’, get interested and even make contributions—​though I have no idea who Louis is apart from what I can glean from the conversation—​then, Evans agrees with Kripke, in that context I am using ‘Louis’ as the name of the person—​say Louis XIII of France—​the conversation is about. But that context may be lost. If long after, I use the name ‘Louis’ again when I have forgotten the conversation and by some confusion have associated it with a description which was not heard in the pub, say ‘a basketball player’, the mere fact that I am now using the same name and my use is causally connected to the long-​forgotten conversation does not, he insists, establish that I am in any interesting sense speaking of the French king. Evans goes on to illustrate, with various examples, the unsatisfactoriness of any theory which involves the general position that the denotation of a name in a community can be found by tracing a causal chain of reference-​preserving links back to some item so long as there is no intention to change reference. His key point is that reference preservation cannot involve wholly different mechanisms from reference acquisition. Generally, for meaning or reference: There aren’t two fundamentally different mechanisms involved in a word’s having a meaning: one bringing it about that a word acquires a meaning, and the other—​a causal mechanism—​which operates to ensure that its meaning is preserved. The former processes are operative all the time; whatever explains how a word gets its meaning also explains how it preserves it, if preserved it is. Indeed, such a theory [a causal theory of reference/​meaning preservation] could not account for a word’s changing its meaning. It is perfectly possible for this to happen without anyone’s intending to initiate a new practice with the word; the causal chain would then lead back too far. 1973, 195 The Madagascar case is introduced by Evans to illustrate the possibility of such unintended reference shift along a causal chain of ‘reference-​preserving’ links, i.e., ones satisfying the definition he gives (1973, 191), not ones that actually guarantee reference preservation. Evans also mentions a simple imaginary example. Two babies are mistakenly swapped soon after birth, but not before their mothers bestow names on them. So henceforth the man universally known as ‘Jack’ is so-​called because a woman dubbed some other baby with the name. Evans concludes that the causal theory unamended must be rejected. But he insists (1973, 196) that the description theory must be rejected also. For even if ‘Goliath was the Philistine David slew’ is the whole information connected with a name in a community, he says, this does not have the consequence that ‘Goliath’ is in that community a name of that man. (In fact, Biblical scholars now suggest 98

The Problem of Reference Change

that David did not kill Goliath but another Philistine, and the attribution of the slaying to Elhannan the Bethlehemite in 2 Samuel 21 xix is correct.) So Evans’s aim is to produce an eclectic theory which accommodates the possibility of reference change but cannot be brought under the descriptivist banner. The most important features of Evans’s eclectic theory are two: (i) the causal element is not, pace Kripke, a link between the initial baptism and subsequent uses of the name, but between the states and activities of the object which is the bearer of the name and the beliefs about that object possessed by the users of the name, (ii) not any such causal link is sufficient for an object to be the object named; rather the object named must be the dominant source of the name users’ information. What is really wrong with the description theory, Evans says, is not so much the idea that the intended referent of a name is determined by associated information, but the idea that the determination is a matter of fit. There is, he says, something absurd in the thought that the intended referent of a use of a name could be some item completely causally isolated from the user’s community just because it fits better than anything else the cluster of descriptions associated with the name. Comparably, I do not see something just because it matches perfectly my visual impressions. It must be the cause of them. Equally, Evans suggests, an item cannot be the intended referent of the name unless it is the causal source of (some of) the information associated with it. However, it is not sufficient for something to be the referent that it be a causal source of information associated with the name. For there may be more than one. Jones, the leaf raker, is the source of a body of information associated with the name ‘Smith’ by the observer, though the belief expressed is a belief about Smith. Again (Evans 1973, 200), ‘I may form the belief about the wife of some colleague that she has nice legs upon the basis of seeing someone else—​but the girl I saw is the source’. So Evans brings in the notion of dominance. A cluster or dossier of information can be dominantly of one item though it contains elements whose causal source is different. And persistent misidentification can bring it about that a cluster is dominantly of some item other than it is dominantly of originally. He illustrates this with a story about Napoleon. In scenario 1 an imposter takes over Napoleon’s role and name when he is an unknown army officer and goes on to rise to power and fame. In scenario 2 very late in Napoleon’s career an imposter takes over. In scenario 1 Evans says the information we have will be dominantly of the later man. So it will be correct for us to say that Napoleon impersonated an unknown army officer before rising to fame. In scenario 2, by contrast, the information we have will be dominantly of the man who has risen to fame and it will be correct to say that late in his career Napoleon was impersonated and, for example, did not fight at Waterloo. He says, ‘These differences seem to reside entirely in the differences in [our] reactions to the various discoveries, and dominance is meant to capture those differences’ (1973, 201). Evans brings the threads together with his story of Turnip. A  youth nicknamed ‘Turnip’ leaves his village to seek his fortune. Many years later a man comes to the village and lives as a hermit. Among the few surviving older villagers the belief springs up that this is Turnip returned. He isn’t. The name gets widely used as a name of the hermit. Among the older villagers the name remains a name of the young fortune-​seeker and they would agree that the hermit is not Turnip. But they may die off leaving a homogeneous community using the name to refer to the hermit. If he is the dominant source of the information associated with the name its reference will shift. On the other hand, it may be that the information passed on to the younger villagers is so important to them that the young fortune-​seeker remains the dominant source and the referent of the name for them, so that being informed of the facts they would acknowledge ‘the hermit isn’t Turnip after all’. Again, which person qualifies as the dominant source of the information they associate with the name is determined entirely by how they would react to the discovery of the true situation. The key question about this eclectic theory is whether it can account for reference change and also be distinguished from a sophisticated causal descriptivist theory (Kroon 1987; Lewis 1997, n22). Descriptivists can allow that some of the descriptions associated with names in their reference-​fixing 99

Harold Noonan

clusters may employ causal vocabulary, e.g., ‘the creator of such-​and-​such a work of art’, or ‘the causal origin of such-​and-​such uses of “N” ’. They still count as descriptivists because they insist that the reference of a name is determined by (maybe weighted, majoritarian (Kripke 1980, 64–​65, thesis (3))) fit (satisfaction), and take the association of a cluster with a name to be a matter of its being believed (Kripke 1980, 64–​65) by the users of the name to be satisfied by the referent (or, at least, the cluster’s satisfaction by its referent not being something of which the users of the name are totally ignorant). One way the eclectic theory might be distinct is by refusing a name a reference where a description theory (wrongly) assigns one. Evans envisages such a situation when he compares reference to perception and sketches the possibility of someone who is a perfect fit for all the descriptive information associated with a name in a community whilst being causally isolated from the community (1973, 197–​198). ‘The absurdity’, he says, that the denotation of our contemporary use of the name ‘Aristotle’ could be some unknown (n.b.) item whose doings are causally isolated from our body of information is strictly parallel to the absurdity in supposing that one might be seeing something one has no causal contact with solely upon the ground that there is a splendid match between object and visual impression. But the case is hard to understand, for isn’t the most important piece of information we have about Aristotle that he is the author of certain works, e.g., the Nicomachean Ethics, and the causal origin of our contemporary use of that name? The case thus does not provide any reason for distinguishing the eclectic theory from causal descriptivism. The other way in which the eclectic theory might be in conflict with descriptivism is if it gives a different (correct) verdict on the reference of a name. Suppose that X is the causal source of the information associated with a name ‘N’, and another object Y is the object best fitting the descriptions (and X is in fact the reference). Will this not show the distinctness of the eclectic theory? The difficulty, however, is that according to the eclectic theory its reference is not merely a source of the information associated with the name but the dominant source. Now the notion of dominance is explained entirely in terms of reactions to scenarios in which the facts are revealed. But then the notion can be applied independently of the eclectic theory.We can speak, in the same sense, of certain associated descriptions being dominant in determining the reference of a name (which is, I think, close to Kripke’s idea (1980, 64–​65) that a description theorist can allow that some descriptions can have more ‘weight’ than others, are more important in the ‘vote’ (see also Noonan 2013, 55)). But then the case we need to consider is one in which X is the dominant causal source of the information associated with ‘N’ and Y is the item which satisfies the dominant descriptions associated with the name. But if X is the dominant source, then Y is not the satisfier of the dominant descriptions associated with the name, since the reactions of users of the name would be to agree, given knowledge of the facts, that Y is not N. And if Y is the satisfier of the dominant descriptions, then X is not the dominant causal source of information associated with the name since the reaction of the users of the name would be to say that X is not N. ‘The dominant causal source’ and ‘the satisfier of the dominant descriptions’ cannot come apart since what makes for dominance is the user’s reactions and the user cannot react in incompatible ways at once. It appears, then, that Evans has not provided a theory of naming which is both clearly distinct from any form of descriptivism (in particular, a form of causal descriptivism) and can account for reference change. We are therefore still left with the question whether any non-​descriptivist theory of names can accommodate reference shift; in particular, inadvertent reference shift. What does seem clear is that any theory which can do so must incorporate the social dimension of name use.

100

The Problem of Reference Change

Back to Kripke But, as we said before, given the modesty of Kripke’s claims it is not clear how the possibility of inadvertent reference shifts can be thought of as a challenge to his own position. He does not claim to have a theory of reference or even a sufficient condition for reference preservation. So nothing he says is inconsistent with the proposition that in some cases, e.g., in the Madagascar case, such inadvertent reference shifts occur. It is straightforward to understand how writers after Kripke, such as Evans, who try to flesh out the anti-​descriptivist picture into a theory, are challenged by the Madagascar scenario. But it is not similarly straightforward to see how Kripke himself, who thinks that any theory of reference might well be subject to counter-​example, and so refrains from formulating one, is so challenged. However, I think that there is a challenge to be put to Kripke, namely, to explain why the Gödel scenario he envisages is a successful counter-​example to the description theory and reference has not changed so that Schmidt is not the reference of the name, while Madagascar (the island) is the reference of the name in the historical situation Evans describes so reference has changed. What is the difference? The Gödel scenario (Kripke 1980) is a situation in which Gödel, the man who in fact proved the incompleteness theorems, did not do so. He stole them from his friend Schmidt, and someone is told only ‘Gödel is the author of the incompleteness theorems’. The challenge to Kripke can be spelled out. The task is to explain why (1) ‘Madagascar’ is unambiguously used as the name of the island by later users who have received it from Marco Polo (directly or indirectly), despite the historical/​causal connection to (a baptism of) part of the mainland, whilst, on the other hand, (2) the fictional Gödel case as Kripke envisages it is a counter-​example to the description theory and Gödel is the reference of ‘Gödel’ as employed by the user of the name even though he has no identifying description at all of Gödel, and believes that ‘Gödel is the author of the theorems’ is true and so intends, when using the name, to be speaking of the author, who is, in fact Schmidt. To be clear about this difficulty it is important to be clear that Kripke’s position is that the fictional scenario he is considering (with details elaborated, of course), is a clear counter-​example to the description theory, a clear case in which the user of the name both is ignorant of any descriptions identifying Gödel and associates with the name a wholly erroneous description or set of descriptions. He does not, then, think of Gödel as the man to whom the proof of the theorems is commonly attributed, nor does he think of Gödel as the person referred to by the name by those from whom he has acquired it, nor as the person referred to by the name as used previously by himself, nor as what the users of the name in his community use the name to refer to. All of these are descriptions the description theorist can appeal to.1 But if we strip out of the case all these elements, as we must if the case is to be a candidate for a clear counter-​example to the description theory, why is it that he is referring to Gödel rather than Schmidt, whereas the later users of ‘Madagascar’ are not referring to the part of the mainland, but to the island? In his recent piece Burgess (2014) makes a proposal in response to this question. The difference, he says, is that the user of ‘Gödel’ in the fictional scenario is not acquainted with Schmidt (he has just heard of the incompleteness theorems, though no one has shown him any proofs or told him anything about their authorship). He now hears someone saying ‘Gödel is the author of the incompleteness theorems’ and so acquires the name ‘Gödel’ as a name for Gödel, though he has none of the Gödel-​directed intentions listed above which would make it unclear that the case is a counter-​ example to the description theory. On the other hand, the users of ‘Madagascar’ after Marco Polo are acquainted directly or indirectly with the island and so, despite their intention to be faithful in their use of the name to those from whom they got it, when it is passed on to them it is for them a name of the island. The problem with this proposal, I think, is that it neglects the insight in Evans’s remark that there are not two fundamentally different mechanisms involved in a word’s having a meaning, one bringing 101

Harold Noonan

it about that the word acquires the meaning and another which operates to ensure that its meaning is preserved. The former process is operative all the time. According to Kripke’s story (as noted above) the initial fixation of reference is in accord with the description theory. A name may be introduced for an object of acquaintance, but it need not be; its reference may be fixed by a description (a purely general description, a causal description employing causal vocabulary, a description containing an indexical, a metalinguistic description, or any combination thereof), or of course, a cluster of descriptions. So exactly the same may be true of the later uses of the name in which a new name-​using practice is inadvertently created. It is not essential to the story of Madagascar that later users, who use the name to refer to the island, are acquainted with it. What is essential is just that the descriptions they associate with the name that pick out the island are such that their intention to refer to the item satisfying them overrides, somehow, their intention to be faithful to the previous uses of the name they have encountered. And this can be so in the Gödel case too. What prevents it happening in the situation as we naturally imagine it, given Kripke’s brief description, is precisely that all who acquire the name are, and think of themselves as being, members of an ongoing community of users of the name and intend to conform to the usage of one another. But this social context must be eliminated by stipulation, as Kripke assumes it is, if the Gödel scenario Kripke envisages is not to be easily dismissed as a counter-​example to the description theory, for otherwise the user will have many alternative descriptions of Gödel available, e.g., ‘the man under whose name the theorems were published’ (Dummett 1973, 135–​136) (and then the difference between the Gödel case and the Madagascar case is easily explained by the descriptivist by appeal to the idea of the dominant descriptions or the notion of weighted majoritarian fit). However, it is only this social context that makes it intelligible that, despite his ignorance and mistaken belief, the user of ‘Gödel’ Kripke envisages does not refer to Schmidt, but yet successfully refers when he uses the name. So the Gödel example cannot be coherently understood as a counter-​example to the description theory. Hence Kripke cannot meet the challenge posed above. That is, he cannot explain both (1) why in fact ‘Madagascar’ undergoes a change of reference and later is unambiguously only a name of the island and (2) why the fictional Gödel case is a counter-​example to the description theory in which there is no change of reference.The significance of the Madagascar case for Kripke is thus that reflection on the difference between it and the putative Gödelian counter-​example to the description theory brings out that the latter cannot in fact be intelligibly understood as a counter-​example to the description theory at all.

Note 1 Note that given Kripke’s historical picture some uses of names are necessarily in accord with the description theory. Speaker intentions play two crucial roles in his story. There is an initial baptism in which the user intends the reference of the name to be fixed by a description. Then, when the name is passed on to a second user, he must intend to use it with the same reference as the first, and so on. Neither the baptiser’s first use, governed by his baptismal intention, nor any subsequent user’s first use governed by his deferential intention, can be a counter-​example to the description theory. Counter-​examples can only possibly emerge in subsequent uses of names by users.

References Burgess, J. 2014. Madagascar revisited. Analysis 2: 195–​225. Davidson, D., D. Kaplan, W.V. Quine, B. Partee, M. A. E. Dummett, G. Harman, H. Putnam, S. Kripke, W. Sellars, D. Lewis, and C. Parsons. 1974. First general discussion session (Proceedings of a conference on language, intentionality, and translation-​theory). Synthese 27: 471–​508. Dummett, M. A. E. 1973. Frege: Philosophy of Language. London: Duckworth. Evans, G. 1973. The causal theory of names. Proceedings of the Aristotelian Society, Supplementary Volumes 47: 187–​225.

102

The Problem of Reference Change Harman, G., W. V. Quine, S. Kripke, D. Lewis, M. A.  E. Dummett, and B. Partee. 1974. Second general discussion session (Proceedings of a conference on language, intentionality, and translation-​theory). Synthese 27: 509–​521. Kripke, S. 1972/​1980. Naming and necessity. In Semantics of Natural Language, eds. G. Harman and D. Davidson, 253–​355 and 763–​769. Dordrecht: Reidel. Reprinted in revised form as a monograph by Basil Blackwell, Oxford, 1980. Kripke, S. 1977. Speaker’s reference and semantic reference. Midwest Studies in Philosophy 2: 255–​276. Kroon, F. 1987. Causal descriptivism. Australasian Journal of Philosophy 65: 1–​17. Lewis, D. 1997. Naming the colours. Australasian Journal of Philosophy 75(3): 325–​342. Noonan, H. 2013. Routledge Philosophy Guidebook to Saul Kripke and Naming and Necessity. London: Routledge. Taylor, I. 1898. Names and Their History:  Handbook of Historical Geography and Topographical Nomenclature. London: Rivington’s.

103

PART III

Causal Theories and Cognitive Significance

8 COGNITIVE SIGNIFICANCE Aidan Gray

Frege’s Puzzle is a founding problem in analytic philosophy. It lies at the intersection of central topics in the philosophy of language and mind: the theory of reference, the nature of propositional attitudes, the nature of semantic theorizing, the relation between semantics and pragmatics, etc. The puzzle concerns the relation between the referential significance of a sentence (or utterance)—​ i.e., the way it portrays properties and relations as distributed over objects—​and its cognitive significance. ‘Cognitive significance’ is an umbrella term, used to pick out a range of cognitively relevant features of a sentence: the state of mind a speaker could express with it, the kinds of evidence the speaker takes to be relevant to it, the states of mind in others she can use it to report, etc. Frege’s Puzzle seems to show that reference and cognitive significance don’t align in the way that independent considerations suggest they should. This chapter is an overview of the puzzle and of the space of contemporary approaches to it.

1  Frege’s Challenge The background for discussions of cognitive significance is a puzzle introduced by (Frege 1892). Our interest is not historical, so I will present it in anachronistic form. It is generated by a tension between two plausible theses about meaning: one linking meaning with reference; another linking meaning with cognition. Millianism: The meaning of a proper name1 is its referent. Meaning and Cognitive Significance: The meaning of a sentence determines its cognitive significance. Millianism should be understood against the background of compositional theories of meaning. In a compositional theory, the meaning of a complex expression is a function of the meaning of its parts, and their mode of combination.The meanings of simple expressions are assigned directly. It is natural, in a theory of this kind, to hold that the meaning of a proper name is the object to which it refers; that is, the meaning “London” is London (the city), the meaning of “Bob Dylan” is Bob Dylan (the man), etc. This is the simplest hypothesis about the meaning of a name that explains the contribution a name makes to the conditions under which a sentence containing it is true. 107

Aidan Gray

Let’s call pairs of sentences that apparently differ only in the substitution of coreferential names ‘Frege-​pairs’. Given Millianism, compositionality, and the assumption that there are no ‘hidden’ semantic differences between Frege-​pairs, it follows that Frege-​pairs have the same meaning. So it follows, given that Robert Zimmerman is Bob Dylan, that (1a) and (1b) have the same meaning. (1a) Bob Dylan is Bob Dylan. (1b) Robert Zimmerman is Bob Dylan. Frege claimed that pairs like (1a)/​(1b) differ in cognitive significance. Identity statements of the form in (1a) are knowable a priori and analytic; while statements of the form in (1b) are not knowable a priori and contain “valuable extensions of our knowledge”. Frege took the difference in cognitive significance between (1a) and (1b) to entail that they differed in meaning. Here, he was relying on Meaning and Cognitive Significance. We can motivate Meaning and Cognitive Significance with simple reflection on the explanatory role of meaning. Meaning (or, perhaps, knowledge of meaning) bridges the gap between cognition and linguistic behavior. Given I believe that p why did I assert sentence S? Because S means that p. Given I had evidence e, why did I assent to S? Because S means that p and e is relevant to the truth of p. And so forth. We’ll see shortly that this story is too crude. But it illustrates why Meaning and Cognitive Significance is a natural place to start our semantic theorizing. Frege also introduced a more direct challenge for Millianism. In sentences that report propositional attitudes, or speech acts—​call these ‘ascriptions’ for short—​substitution of coreferential names can, apparently, alter truth-​values. Consider the following vignette. Smith knew Robert Zimmerman as a child. She is aware of the famous musician Bob Dylan, but is unaware that the boy she knew as a child is Dylan. As far as she knows, the boy she grew up with went on to lead an unmusical life. In relation to this situation, speakers judge (2a) to be true but (2b) to be false. (2a) Smith believes that Bob Dylan is a musician. (2b) Smith believes that Robert Zimmerman is a musician. If we accept these judgments, we have a more direct problem for Millianism. Relying only on the assumption that truth-​conditions are compositional and that (utterances of) (2a) and (2b) semantically differ only in the substitution of coreferential names, we predict that (2a) and (2b) must have the same truth-​value.

2  Sense and ‘The New Theory’ of Reference Frege responded to these considerations by rejecting Millianism. He held that singular terms had two kinds of linguistic significance: sense and reference.The sense of an expression, for Frege, was the way that the expression ‘presented’ its referent. The sense of an expression, for Frege, determines its reference. But two expressions with the same reference could ‘present’ it in different ways; so coreferential names could differ in sense. He proposed to explain differences in cognitive significance between Frege-​pairs with differences in sense, and so developed a theory of the role that sense played in linguistic understanding and semantics. Speakers understand an expression by being in cognitive contact with its sense (by ‘grasping’ it). Reference is the level at which compositional determination of truth-​conditions occurs. The sense of an expression determines its cognitive significance. The contrast between (1a) and (1b) shows that ‘Bob Dylan’ and ‘Robert Zimmerman’ have different senses. Frege held that ascriptions generate a ‘shift’ in meaning: when it is embedded under an attitude verb, an expression refers to its ‘customary’ sense (i.e. the sense it has when not so-​embedded). This 108

Cognitive Significance

allows Frege to capture the differences in truth-​conditions between (2a) and (2b). As they occur in those sentences, ‘Bob Dylan’ and ‘Robert Zimmerman’ have different referents (because they have different senses when they occur unembedded). So Frege can hold that (2a) and (2b) are not, contrary to appearances, related by substitution of coreferential expressions; so he can explain their difference in truth-​conditions without violating compositionality. This framework is elegant. But its plausibility depends, inter alia, on giving an account of sense. Frege didn’t offer an account. When he characterizes the sense of an expression, he tends to reach for descriptions. For example, the sense of ‘Aristotle’, for some speakers, might be the pupil of Plato and teacher of Alexander the Great. These remarks suggest a theory that formed the background to 20th-​ century work in the theory of reference: the sense of a name, n, for a speaker, S, is a description, d, that S associates with n; n refers to whatever satisfies d. Call this Descriptivism.2 Space does not allow us to recapitulate the arguments against Descriptivism.3 Suffice it to say that seminal work in the 1960s and 1970s—​by Donnellan, Kripke, Putnam, Kaplan, Burge, and others—​was taken by many to decisively refute it. In brief, the claim that linguistic reference is always mediated by descriptive identification of an object was taken to be implausible as an account of linguistic competence and to make false predictions about the epistemic and modal features of sentences involving names (Kripke 1980). In its place, the ‘new’ theory of reference held that reference is typically determined, in part, by factors that are ‘external’ to a speaker’s cognition—​for example, the chain of uses of a term that links a speaker’s use back to an original baptism.4 The work of the ‘new theorists’ is taken not just to refute Descriptivism, but to motivate Millianism. Though non-​Millians remain,5 Millianism forms the background for most contemporary discussions of cognitive significance. The ‘new theory’ is also taken to weaken, or complicate, our commitment to Meaning as Cognitive Significance. We motivated it above by reflecting on the explanatory role of meaning: grasp of meaning is the bridge between cognition and language-​use. But the externalism embodied in the new theory—​the idea that the meaning of a term for a speaker depends on facts that are external to the speaker’s psychology—​complicates this explanatory picture. Given externalism, it is possible for a speaker to be unaware that two expressions mean the same thing (because she is unaware, for example, that the chain of uses leading from her use of ‘Dylan’ and the one leading from her use of ‘Zimmerman’ converge on the same man). In the context of the new theory, difference in cognitive significance does not entail difference in meaning. A significant literature has developed around the issue of the extent to which externalism about meaning entails that meaning is not cognitively ‘transparent’ to speakers.6

3  Clarifying the Challenge(s) Frege’s Puzzle appears to offer two distinct, but related, challenges: to explain the difference in cognitive significance between Frege-​paired ‘simple sentences’; and to explain the apparent differences in truth-​conditions of Frege-​paired ascriptions. In this section, we examine the challenges in more detail.

3.1  Simple Sentences What, exactly, does Frege challenge us to explain? We introduced the challenge in relation to true identity statements, but it is broader. A competent, rational, speaker who doesn’t know that Dylan is Zimmerman, might accept (3a) while rejecting (3b). (3a) Dylan is a musician. (3b) Zimmerman is a musician. Identity statements dramatically illustrate a more general point: Frege-​paired declarative sentences can differ in the evidence that competent, rational, speakers would require to endorse them. And 109

Aidan Gray

given the kind of connection between meaning and cognition that Frege is assuming, this poses a perfectly general problem. Even if we reject Meaning and Cognitive Significance—​as new theorists might—​we are still left with a question: what is the difference between Frege-​pairs? How should we explain the different way they interact with linguistic cognition and communication? It is not uncommon to find discussions of Frege’s challenge framed in terms of the problem of explaining why a competent, rational, speaker might assent to, or accept, an utterance of a declarative sentence S while failing to assent to, or accept, an utterance of a Sʹ, when S and Sʹ are Frege-​pairs. But the challenge is broader than this in several ways. First, it shows up with non-​declarative sentences. Consider (4a)/​(4b). (4a) Is Dylan a musician? (4b) Is Zimmerman a musician? These pairs exhibit an analogous contrast to that exhibited by (3a)/​(3b), modulo the communicative role of interrogative sentences. A speaker who is unaware that Dylan is Zimmerman could respond differently to each pair in a way that is functionally analogous to the difference between acceptance/​ rejection in the case of declarative sentences. Such a speaker might treat (4b) but not (4a) as an attempt to change the subject (if the conversation had been about Dylan), or might treat (4a) but not (4b) as an attempt to reopen a question that had already been settled (if it was common ground in the conversation that Dylan was a musician). Analogous points could be made with respect to other non-​assertoric speech acts. There are also challenges associated with speakers who are aware of the relevant identities. Suppose Brown believes that Zimmerman is Dylan but is unaware he is a Minnesotan; contrast the effects on Brown of accepting the discourses in (5a)/​(5b). (5a) Dylan is a Minnesotan. But Dylan is not Zimmerman. (5b) Zimmerman is a Minnesotan. But Dylan is not Zimmerman. Notice two things. First, Brown would be in a different attitude state as a result of accepting (5a) as compared to accepting (5b): just consider how she would respond to the question, “Is Zimmerman a Minnesotan?”, after accepting each discourse. So the two discourses, as a whole, must differ in cognitive significance for her. But notice, also, that the second sentence of each discourse is the same. So if the two discourses differ in cognitive significance for Brown, it must be because the first uttered sentences differ in cognitive significance for her (or, equally puzzling, that the cognitive significance of a discourse is not determined by the cognitive significance of its component utterances). Recall, though, when the first sentence is uttered, Brown believes that Dylan is Zimmerman.This suggests that Frege-​ pairs can have different cognitive effects on speakers who believe that they are referentially equivalent. Theorists sometimes appeal directly to speakers’ judgments about sameness/​difference of meaning. We might simply ask speakers whether Frege-​pairs have the same meaning. Many theorists hold that untutored intuitions tell us that the pairs have different meanings. Perhaps this, itself, requires explanation. Similarly, Frege-​pairs appear to participate differently in intuitive evaluation of inferences. There are a few upshots to this. The first is that there appear to be a whole range of phenomena which are such that we are inclined, in explaining them, to appeal to the meaning of sentences or utterances, and such that Frege-​pairs participate differently in them (no doubt the list above could be extended). There is no simple, precise, statement of what it is for two sentences to differ in cognitive significance. So it’s not obvious how much uniformity there is in ‘the’ challenge posed by Frege’s puzzle. This has two important lessons for us. First, without a precise, pre-​theoretical, characterization of cognitive significance, we can’t put much weight on Meaning and Cognitive Significance. That the 110

Cognitive Significance

meaning of a sentence plays some, yet to be precisely characterized, role in each of the above phenomena is little reason to be troubled by the fact that sentences with the same meaning might play different roles. More interestingly, perhaps, the variety also suggests that we shouldn’t be too quick to assume that what is required is a unitary explanation. The literature is replete with apparently competing responses to Frege’s challenge. It would be salutary to keep in mind that apparently competing strategies might be complementary—​either by applying to different aspects of Frege’s challenge, or by working together in relation to explain some aspects.

3.2  Attitude Ascriptions Frege claimed that sentences like (2a) [“Smith believes that Bob Dylan is a musician”] and (2b) [“Smith believes that Robert Zimmerman is a musician”] can differ in truth-​value. The source of this claim is simply that there are circumstances in which competent speakers, even those who know that Dylan is Zimmerman, would accept (2a) while rejecting (2b) (and would even do so by calling one “true” and the other “false”). Contemporary theorists are cautious about moving directly from speakers’ judgments about the truth-​values of utterances to claims about the truth-​values of the uttered sentences. Put roughly, we now assume that speakers have generally reliable judgments, at most, about, what we might call, ‘the conveyed content’ of an utterance. Conveyed content is not always identical to the semantic content of the uttered sentence. It often includes substantial influence of various kinds of pragmatic processes. So we should be careful about how we describe the data. Instead of saying that the relevant pairs differ in truth-​value, we’ll say that they differ in ‘acceptability’.‘Acceptable’/​‘Unacceptable’ are technical terms, used to characterize speaker’s intuitive evaluations of uttered sentences in a way that doesn’t presuppose that speakers are making accurate judgments about the uttered sentence’s truth-​value.We’ll see below that some accounts of ascriptions insist that if we focus on the semantic properties of the uttered sentence, (2b) is true. When speakers reject it as unacceptable, the thought goes, they are responding to some feature of its conveyed content distinct from the semantic value of the uttered sentence. This redescription allows us to characterize a choice-​point in approaches to Frege’s challenge: should we give the same response to both aspects of the challenge? That is, do the same explanatory mechanisms explain the data about simple sentences and attitude ascriptions? Frege-​paired simple sentences cannot differ in truth-​value.7 This is why we had to gesture at the difficult-​to-​characterize idea of differences in cognitive significance. If we say the same thing about Frege-​paired ascriptions—​if we say that the relevant sentences cannot differ in truth-​value—​it is open to us to offer the same explanation of the perceived differences in meaning that we offer in the case of simple sentences. This is an important choice-​point. Some theorists approach the challenge in a hybrid way, positing a semantic difference between pairs of ascriptions while positing some other kind of difference between pairs of simple sentences. Others argue that a unified approach is possible, and preferable on the grounds of simplicity. It’s worth pausing to discuss features of the data that are relevant to the prospects of a unified approach. Let’s extend our vignette from above. Suppose Smith is friends with Jones. Jones and Smith are in the same situation with respect to Dylan/​Zimmerman. Jones would accept (1a) but not accept (1b); she would also accept (2a) but not accept (2b). So far, so good. But suppose that Jones, but not Smith, learns that Zimmerman is Dylan. Jones will now accept both (1a) and (1b). But it’s natural to think that so long as she thinks that Smith is still not in the know about Dylan’s identity, and that Smith would still avow the same beliefs, Jones would continue to accept (2a) and not accept (2b). It doesn’t seem that learning that Dylan is Zimmerman puts any pressure on Jones to revise her judgments about (2a) and (2b).This, presumably, is at the heart of Frege’s insistence that unlike simple sentences, the truth-​values of ascriptions can be altered by substitution of coreferential names. The differences in acceptability for ascriptions persist even when audiences (but not attributees) are ‘in the know’ about the relevant identities. 111

Aidan Gray

There are two related things to highlight. The first is just that speakers are especially resistant to treating reports like (2b) as true (in scenarios of the kind just imagined). In many, though perhaps not all, cases in which the conveyed content of an utterance exceeds the semantic content of an uttered sentence, ordinary speakers can be brought to distinguish between the two, and so can be brought to acknowledge, for example, that the semantic content is true when conveyed content is false. It isn’t clear that this is the case with (2b). Even philosophers—​well-​practiced in isolating semantic and pragmatic content—​often claim to be unable to detect the putative truth semantically expressed by it. On the other hand, some theorists claim that ordinary speakers, perhaps with a little philosophical tutelage, can be brought to see that (2a) and (2b) have the same semantic content.8 Second, whatever we say about that, there is still the initial contrast: once you’re in the know, you don’t treat (1a) and (1b) as differing in acceptability; but you do continue to treat (2a) and (2b) as differing in acceptability (at least until you receive philosophical therapy). And this, itself, is an aspect of the data. The uniformity we posit in our account of simple sentences and ascriptions cannot be so complete as to leave us unable to explain why differences in acceptability are substantially more robust with ascriptions. It’s important to note, though, that it isn’t clear that contrasts in acceptability for simple sentences always disappear for those in the know. Contrast (6a) and (6b) (imagine them in the context of a biography). (6a) Zimmerman left Minnesota but Dylan arrived in New York. (6b) Dylan left Minnesota but Zimmerman arrived in New York. Even if one is in the know, it’s quite natural to judge (6a) as acceptable and (6b) as unacceptable. Examples of this sort provide some support for a unified approach to simple sentences and ascriptions because they call into question the sharp cleavage in the data we claimed to find.9 How much support they provide will depend on the extent to which the explanation of the difference in acceptability between examples like this can be extended to ascriptions (consider: How much does the contrast between (6a) and (6b) depend on the narrative situation? How much does it depend on contrastive structure? How much does it depend on the fact that the speaker has chosen to use a second name rather than an anaphoric pronoun?). Finally, we should keep in mind that there is context-​sensitivity in the acceptability of ascriptions.10 Consider Kripke’s (1979) example: Pierre grew up in Paris, hearing about the beautiful English capital; so he would sincerely assert “Londres est jolie”. Having been transported to London at a later time, learning English there, failing to realize that the city he now lives in is the one he calls ‘Londres’, and finding it distasteful, he would sincerely assert “London is not pretty”. Consider (7). (7) Pierre thinks that London is pretty. If we imagine it uttered in the context of a discussion of Pierre’s beliefs about the relative attractiveness of capitals (“Pierre thinks that Paris is pretty, that Madrid is not pretty, that London is pretty …”), it appears acceptable. If we imagine it uttered in the context of a discussion of Pierre’s views about his current living situation (“Pierre loves living in London. He has a great job, he thinks that London is pretty”), it appears unacceptable. Part of answering Frege’s challenge is explaining this context-​sensitivity.

4  A Taxonomy of Approaches The literature on Frege’s Puzzle is voluminous and the variety of approaches is bewildering. In this section, I’ll provide an (admittedly incomplete) taxonomy. 112

Cognitive Significance

Consider the components of a linguistic exchange. On one side, we have a speaker’s cognitive situation:  their attitudes, representations of the speech situation, their communicative intentions, and their knowledge of their language. On the basis of that, the speaker utters a sentence with certain semantic features (relative to the context of utterance). Those features, in conjunction with facts about the communicative situation, determine the conveyed content of the utterance. On the basis of those things, the utterance has an effect on the cognitive state of the audience (principally on their attitudes). We can understand the space of responses to Frege’s challenge by starting from a set of ‘default’ assumptions about these components. We can think of the defaults as claims that theorists in this area ‘get for free’. This is not to say that they are uncontroversial or true. It’s only to say they are typically taken as an acceptable ground-​zero for theorizing about cognitive significance. Additions or modifications to them are justified by their role in responding to the challenge. The defaults are: (i)

(ii) (iii)

(iv)

(v)

Propositional content is Russellian. That is, propositions are complexes of objects, properties, and operators and they have a sentence-​like structure.11 So Russellian propositions are the contents of attitudes, uttered sentences, and communicative acts. The attitude state of a speaker can be modeled as a collection of relations to Russellian propositions (i.e. a set of propositions believed, a set of propositions desired, etc.). Meaning is Millian and compositional.The semantic contribution of an uttered name n in sentence S is the object n introduces into (the relevant part of) the proposition expressed by S. So Frege-​pairs semantically express the same proposition. In an attitude ascription ⌜X v’s that p⌝, v expresses a relation between individuals and propositions. The ascription is true iff the individual denoted by X stands in the relation denoted by v to the proposition denoted by p. No pragmatic effects—​assertoric enrichment, conversational implicature, pragmatic modulation, etc.—​are relevant to Frege’s challenge.

Jointly, these assumptions set the background for Frege’s challenge. They assume that sentences that appear only to differ in the substitution of coreferential names, really do only differ in that way. So Frege-​pairs have the same semantic value. And they assume that there are no post-​semantic mechanisms at play to generate different conveyed contents. And they offer no hint of how Frege-​ pairs might differ in relation to cognition. We can understand approaches to cognitive significance in terms of which of these assumptions they reject and how. Keep in mind that particular theories will often depart from more than one assumption.

4.1  Pragmatic Departures Perhaps the simplest, and probably the most popular, approach is to depart from (v). We can keep everything else in place and simply hold that Frege-​pairs, though they don’t differ in their semantic content, can nevertheless differ in conveyed content.12 The challenge for approaches of this kind is to specify the way in which the conveyed content diverges from the semantic content and to offer a non ad hoc account of the pragmatic mechanism that brings that divergence about.13 If this approach does not diverge from (i), it must locate the difference between semantic content and conveyed content in Russellian terms. This is done by positing descriptively enriched Russellian propositions as conveyed content. So, given the appropriate context, the conveyed content of an utterance of (3a) [“Dylan is a musician”] might include the proposition that Dylan, the famous sixties folk icon, is a musician, while the conveyed content of an utterance of (3b) [“Robert Zimmerman is a musician”] might include the proposition that Zimmerman, the son of Abram Zimmerman from Duluth, is a musician. Given that these are distinct Russellian propositions, the conveyed content 113

Aidan Gray

of the two utterances will be distinct. And the difference in cognitive significance between the two utterances will be explained in terms of differences in conveyed content. It isn’t clear that this approach can explain the cognitive difference in the sort of cases on which the debate usually focuses.14 Nor is it clear that accounts of this form will explain the full range of facts about cognitive significance canvassed in section 3.1. (Note: I don’t mean to suggest that it is clear that this approach won’t work either). There are different conceptions of the mechanisms that generate the conveyed content. We could hold that it is conversationally implicated15 or that it is asserted.16 In either case, the challenge is to show that the enrichments are independently motivated by our general theories of implicature and assertion, and not simply stipulated as convenient solutions to Frege’s challenge. A variant of this approach holds that utterances always express multiple layers of Russellian content, related according to general principles of information-​theory.17 And that differences in the way that semantic content is expressed—​for example, the difference between using “Zimmerman” or “Dylan”—​will generate different propositions at ‘token-​reflexive’ levels of conveyed content (roughly:  levels of conveyed content that are about how the semantic content of an utterance is determined). The most committed pragmatic theorist will offer the same explanation of simple sentences and ascriptions. As with simple sentences, we can hold that the extra content in ascriptions is asserted or that it is implicated. There are different approaches with respect to the nature of the extra content. Some theorists treat it as adding extra specification of the content of ascribee’s first-​order attitudes (Soames 2002); (for example, the conveyed content of an utterance of (2a) might contain the proposition that Smith believes that Dylan, the famous sixties folk icon, is a musician). Others treat the extra content as characterizing the ascribee’s cognitive state in terms of linguistic behavior or attitudes about language (for example, the conveyed content of (2a) might contain the proposition that Smith would assent to the sentence “Dylan is a musician”, or that Smith believes that this sentence is true).18 Still others characterize it in terms of psychological posits like ‘guises’ (Saul 1998). It isn’t clear that any non ad hoc pragmatic story will provide satisfying explanations here. It is worth noting, though, that if one can be developed in a non ad hoc way, it would likely explain the context-​sensitivity of ascriptions.

4.2  Propositional Departures Some theorists hold that responding to Frege’s challenge requires abandoning (i); that is, revising our account of propositions. Consider two subjects, each ignorant that Dylan is Zimmerman. Xavier believes (as we would naturally describe it) that Dylan is a musician and that Zimmerman is from Minnesota. Wyatt believes that Dylan is a musician and that Dylan is from Minnesota. Given the default assumptions, we have no way to distinguish the two attitude states: each believes the same Russellian propositions. This seems like a bad consequence; there are cognitive differences between the subjects that are relevant to their linguistic behavior (and their behavior more generally). We might, then, develop a more fine-​grained conception of propositions: one that distinguishes the propositional objects of the belief that Dylan is a musician and the belief that Zimmerman is a musician. This was an aspect of Frege’s response: he held that the objects of the attitudes are thoughts—​that is, abstract objects composed of senses (Frege 1918). The new theory of reference discouraged theorists from adopting a descriptivist version of Frege’s theory, but there are non-​descriptive approaches to sense. One tradition holds that given a proper understanding of semantic theorizing, differences in sense will be easy to capture. Suppose that a semantic theory takes the form of a collection of sentences assigning meanings to lexical items, and rules that generate the meanings of complex expressions compositionally. In that context, the clause that specifies the meanings of “Bob Dylan” might be (8a) while the clause that specifies the meaning of “Robert Zimmerman” might be (8b). 114

Cognitive Significance

(8a) “Bob Dylan” refers to Bob Dylan. (8b) “Robert Zimmerman” refers to Robert Zimmerman. These are distinct sentences in our theory of meaning, so it’s open to us to hold that they represent “Bob Dylan” and “Robert Zimmerman” as having different meanings (thus rejecting (iii)).19 This difference ramifies through the compositional rules and generates different meanings for sentences containing the names. So the theory assigns different meanings to Frege-​pairs. So we can locate the difference in cognitive significance between Frege-​pairs at the level of content. This kind of approach has some appeal; it inherits the elegance of Frege’s approach in that it offers a unified and simple response to the challenges. But it is often held to be unsatisfying. Note that the theory specifies the difference in meaning between “Bob Dylan” and “Robert Zimmerman” by using those expressions in the metalanguage. It thus provides no independent grip on that difference (something we might have hoped that our theory would provide). Proponents of the view insist that this demand is misplaced. They hold that (8a), for example, displays the sense of “Bob Dylan” even if it doesn’t describe it (because all it does is assign a referent to the name). Whether this sort of response is satisfactory will depend substantially on what one thinks we are entitled to expect from a theory of meaning. Suffice it to say for our purposes that many theorists want more.There are forms of non-​descriptive Fregeanism that offer substantive theories of sense; we will postpone discussion of those until the next section. There are other fine-​grained conceptions of propositions. Insofar as the problem non-​descriptive Fregeanism is the difficulty in giving a clear account of sense, we might be tempted to develop a conception of propositions that invokes entities we have a clearer antecedent grip on. An approach of this kind holds that linguistic expressions are, themselves, components (or, at least, ground the individuation) of propositions.20 We can start from Russellian propositions—​which, recall, are complexes of worldly entities—​and construct linguistically enhanced propositions by replacing every occurrence of a worldly entity with a pair comprising that entity and a linguistic expression that denotes it. So, for example, whenever we find Dylan in a proposition, we will replace him with the pair or the pair (or some other pair involving a different expression that denotes Dylan). In this way we retain the basic structure of Russellian propositions, but individuate propositions more finely by including linguistic expressions as propositional constituents. (This is an oversimplification, but will do for our purposes.) This approach can distinguish the object of the belief that Dylan is a musician from the object of the belief that Zimmerman is a musician. One proposition contains the name “Dylan” while the other contains the name “Zimmerman”. It appears to have other virtues: it offers an account of propositions that is finer-​grained than Russellianism, but also provides clear individuation conditions. It is also possible to supply a compositional theory that associates linguistically enhanced propositions with sentences. And it fits with a plausible background ideology that holds propositional thought is linguistically mediated (whether by public language or the language of thought). A worry about the approach is that it individuates propositions too finely; it entails, for example, that no sentences in distinct languages can express the same proposition.21 The extent of the issue here depends on how this account of propositions interacts with a semantics for ascriptions, so we will postpone discussion of it. Finally, we can mention relationist approaches to propositional content. Recall our initial problem in this section: how to distinguish Xavier and Wyatt’s attitude state. The options we have explored so far have proposed adding additional objects into propositional content—​either senses or linguistic expressions. An alternative adds additional relational structure into propositions. The crucial difference, it seems, between Xavier and Wyatt is that Wyatt’s attitude state encodes that the person who is represented to be a musician and the person who is represented to be from Minnesota are the same person. Other fine-​grained accounts of propositions understand this encoding of sameness as the re-​occurrence of propositional constituents. We could, instead, posit relations between propositional constituents. This is Semantic Relationism.22 It identifies the content of an attitude state with a 115

Aidan Gray

collection of Russellian propositions along with a coordination relation on those propositions. The coordination relation captures which propositional constituents are encoded as identical. Semantic Relationism is a fine-​g raining of propositional content that avoids awkward questions about the individuation of propositional constituents. It achieves this by replacing propositional constituents with relations. This benefit has a potential down-​side, though. It involves a significant departure from traditional approaches to compositionality and the aggregation of attitude states; the semantic contribution of a representation cannot be ‘localized’ to its immediate syntactic environment, because it might introduce semantically relevant relations to syntactically distant expressions.23

4.3  Cognitive Departures We just considered attempts to distinguish Xavier’s and Wyatt’s attitude state in terms of content. Other approaches agree that we need to distinguish those states, but deny that they need to be distinguished in terms of content.These approaches respond to Frege’s challenge by complicating our theory of the nature of attitude states without complicating our theory of their content (so rejecting (ii) but not (i)). The most standard version of this approach is to hold that a two-​place attitude relation—​X v’s that p—​holding between a subject and a proposition is grounded in a three-​place relation.The three-​ place relation holds between a subject, a proposition, and some third term—​options here include an expression in the language of thought, a character, a guise, a notion, a way of believing, etc.24 These are different proposals about features of the psychological state of a subject that partially ground the intentional features of their attitudes. The idea, for example, is that although Xavier’s belief that Dylan is from Minnesota and her belief that Zimmerman is from Minnesota relate Xavier to the same proposition, they differ in their third term. She is belief-​related to the same proposition ‘twice-​over’ in virtue of tokening different internal mental ‘symbols’. It is important to recognize that this claim is not, in the first instance, a claim about the semantics of ascriptions; it is not, for example, the claim that ‘believe’ semantically expresses a three-​place relation. Rather, it is a claim about the metaphysics of the attitudes. So without other additional hypotheses, it makes no new predictions about the semantics of attitude ascriptions. We will see below that some theorists incorporate this extra cognitive structure into the semantics of ascriptions. But some, like Salmon (1986), do not. He opts, instead, to mix this picture of the attitudes with a pragmatic account of the difference in cognitive significance between Frege-​paired ascriptions. Because accounts like this do not locate these differences at the level of content, they owe us some explanation of why differences in the ‘third term’ generate differences in cognitive significance. In the case of simple sentences, this is typically done by gesturing at the way that the third terms affect a subject’s linguistic behavior: Salmon (1986) suggests that they have an effect on one’s ability to ‘recognize’ propositions when they are encountered; Braun (1998) talks about their effect on dispositions to accept sentences; Fodor (1975) characterizes names in the language of thought in terms of their causal/​functional roles. These are, presumably, promissory notes awaiting the redemption of some future psychological theory. The central point here is that these approaches typically reject Meaning and Cognitive Significance. They hold that the cognitive differences between Frege-​pairs, at least for simple sentences, has no reflection at the level of meaning. It is psychological but not semantic. The cognitive difference between Frege-​pairs boils down to the different way that they interface with a speaker’s idiosyncratic psychology. A distinct class of approaches offers a relational account of the nature of the attitudes. They see attitudes as bundled into dossiers, files, webs, etc.25 Attitudes that are bundled in this way—​that are part of the same file or web, etc.—​are claimed to stand in important functional relations. In particular, they are functionally related so that the subject can ‘trade on the identity’ of their referents (in the sense of Campbell 1987). This means that the subject is disposed to deploy those attitudes in reasoning and 116

Cognitive Significance

action in a way that presupposes that they are about the same thing and does so without the presence of a representation of their identity. With this sort of functional organization in our theoretical repertoire, we can distinguish Xavier’s and Wyatt’s attitude states. In Wyatt’s case, the two attitudes are bundled together. In Xavier’s case they are not. In effect, bundling takes the place of the Fregean notion of sameness-​of-​sense.These functional differences will have a downstream effect on how Xavier and Wyatt are disposed to deploy their attitudes. The hope is that this will provide an explanation of the range of phenomena we canvassed above. Consider, for example, the phenomenon exhibited by (5a)/​(5b). If we hold the Zimmerman-​ bundle and the Dylan-​bundle can be distinct even when a subject believes that Dylan is Zimmerman (in file-​speak, if the files are linked but not merged), we could explain the cognitive difference between the two utterances.The first sentence in (5a) adds a belief to the Dylan-​bundle; the first sentence in (5b) adds a belief to the Zimmerman-​bundle. The relation between the bundle approach and Fregeanism is somewhat delicate. Certain versions of it conceive of themselves as a kind of neo-​Fregeanism.26 They conceive of mental files as senses and assign to them many of the semantic roles that Frege assigned to sense. From one perspective, then, we could think of these theories as offering a substantive account of non-​descriptive senses (and so, attempting to respond to the worry that non-​descriptive Fregeanism is non-​explanatory). On the other hand, some bundle theorists take a leaf from the three-​place relation theorist’s book and claim that differences in cognitive significance are psychological but not semantic. Geirsson (this volume), for example, holds that distinct names are psychologically associated with different webs of information and that utterances involving different names ‘elicit’ information from different webs. Finally we can mention a somewhat different kind of cognitive departure.The two approaches we have canvassed so far are alike in focusing on the nature of the attitudes. A different approach focuses on, what we can call, ‘evaluative heuristics’. According to this view, Frege-​pairs differ in the way that they interact with the psychological mechanisms by which speakers evaluate utterances for truth/​falsity.27 Theorists who take this approach hold that Frege-​pairs exploit a kind of systematic weakness in speakers’ ability to track utterance-​truth and a systematic reluctance to use known identities to make relevant inferences. This approach fits naturally with, though perhaps does not require commitment to, the relational accounts of cognitive architecture just mentioned.

4.4  Semantic Departures It is often thought that Frege’s challenge requires us to depart from the default picture of the semantics of ascriptions. The most popular idea is to start from the ‘three-​place’ relation approach to the attitudes. We can hold that ascriptions are semantically sensitive not only to which Russellian propositions an ascribee has attitudes towards, but also to how those attitudes are cognitively realized. Approaches differ with respect to how this cognitive structure is semantically encoded. They agree, though, that what particular cognitive information is encoded by the utterance of an ascription is dependent on context. On ‘hidden indexical’ approaches—​(Schiffer 1977, 1987, 1992; Crimmins and Perry 1989)—​ utterances of ⌜X v’s that p⌝ can make reference either to a particular mental symbol that carries (a part of) the content of p in a subject’s attitude state, or to a property of such symbols (perhaps the property of having a certain psychological or functional role). So the truth of an utterance of (2b) might require more than that Smith has a belief with the Russellian content that Dylan/​Zimmerman is a musician. It might require also that this representation of Dylan/​Zimmerman is carried by some particular one of Smith’s mental names for Dylan/​Zimmerman, or by a symbol that is connected functionally to the name “Zimmerman”. Given that it isn’t plausible that reference to mental symbols, or their properties, is carried by any piece of the syntax of ascriptions, Crimmins and Perry hold that they are ‘unarticulated constituents’ of the content of uttered ascriptions. It is part of the meaning of ascriptions that utterances of them can 117

Aidan Gray

be semantically filled-​out in this way (as, perhaps, the meaning of “It’s raining” requires that the location of the putative rain be supplied by context). So although Frege-​paired ascriptions, qua sentences, have the same linguistic meaning, utterances of them typically have different truth-​conditions. For example, the use of “Dylan” rather than “Zimmerman” to report Smith’s belief will typically generate requirements on the properties of the mental symbol by which Smith grasps the relevant proposition. The story, though, about how context supplies information about mental symbols must be more complicated (it cannot simply appeal to the names used in the report). Our example of ‘Pierre’ above already shows that. Richard (1990) offers a different approach for semantically capturing the context-​sensitivity of ascriptions. He posits two ‘layers’ of meaning associated with p in ⌜X v’s that p⌝. This approach to ascriptions is structurally similar to the linguistically enhanced approach to propositions canvassed above. The semantic contribution of p in an ascription is a complex consisting of Russellian content and the linguistic expressions that carry that content in the sentence (Richard calls these ‘RAMs’). Given the three-​place relation approach to attitudes, we can think of a subject’s attitude state, itself, as a collection of relations to RAMs. The truth of an utterance of ⌜X v’s that p⌝ requires that the v-​part of X ’s attitude state contains a RAM that stands in a contextually appropriate relation to the RAM expressed by p. An overarching worry for contextualist approaches to ascriptions is that we might lose the ability to capture cognitive relations between speakers, and systematic semantic relations between ascriptions. One danger for Richard’s view, for example, is that it makes the semantics of attitude ascriptions too fine-​g rained. Any difference in the expressions contained in p and pʹ will entail that they express different RAMs when embedded. It entails, for example, that (9a) and (9b) have distinct context-​ invariant meanings, given that their complement clauses express distinct RAMs (this example is from Larson and Ludlow 1993). (9a) Galileo believed the Earth moved. (9b) Galileo believed the Earth is non-​stationary. This is troubling insofar as it is natural to think that (9a) and (9b) make the same claim: it is hard to see how they could come apart in truth-​value. Related worries arise for inferential relations between attitude ascription (Richard, this volume) and more generally for psychological generalizations framed in terms of ascriptions (Goodman and Lederman 2019). Contextualist theories must respond to these worries by characterizing restrictions, or systematic patterns, in the way that context determines the class of RAMs that are taken to be contextually similar to the RAM expressed by p in an utterance of ⌜X v’s that p⌝. Perhaps, for example, in any normal context, the RAMs expressed by the complements in (9a) and (9b) will count as similar enough that any subject whose attitude state contains a RAM that stands in the contextually appropriate relation to one will thereby contain a RAM that stands in the contextually appropriate relation to the other. This would explain why it is hard to imagine contexts in which the truth-​values of (9a) and (9b) might come apart. (To clean up a left-​over strand from above, note that proponents of linguistically enhanced propositions offer essentially the same response to corresponding worries about their approach.) But as Goodman and Lederman (2019) show, it remains to be seen whether it’s possible to characterize contextual restrictions of this kind that capture all of the systematic relations between ascriptions that we might like.28 One more potential worry. A common complaint (see, e.g., Schiffer 1992; Braun 1998; Soames 2002) has been that it is implausible that ordinary speakers, when they utter ascriptions, are referring to, or more generally communicating about, the cognitive structures posited by three-​place theories of attitudes. The worry is that this approach imputes more theoretical sophistication to ordinary speakers than is plausible. The question for contextualists is whether the folk-​theoretical way that 118

Cognitive Significance

subjects keep track of other people’s psychological perspectives has enough determinacy to do the context-​determining work that the theory requires of them.

5  Interactions Space doesn’t allow us to evaluate the various strategies. But I  will emphasize something that is, perhaps, already clear. Much of the work in unraveling the issues here comes in understanding the interaction between the different departures. We have seen this already in the previous section. But I’ll offer a related impression: it seems unlikely that the full range of phenomena will be explained by any one approach. Recall, the challenge is quite open-​ended: to explain how Frege-​paired sentences can differ in relation to cognition. There is little reason to think that investigation will reveal that all such differences have exactly the same character, and little reason to think that many differences will be explicable without appealing to the interaction between different departures from the default. To justify this impression, I’ll offer one more example. It is plausible that at least part of the cognitive difference between Frege-​pairs is pragmatic; we have general reasons to believe that, very often, the conveyed content of an utterance is distinct from the semantic content of the uttered sentence. And the determinants of conveyed content are many and varied, including subtle linguistic choices and rich psychological background information. So suppose we grant that in some context the conveyed content of an utterance of (3a) might include the Russellian proposition that Dylan/​Zimmerman, the famous sixties folk icon, is a musician; while the conveyed content of an utterance of (3b) might include the Russellian proposition that Dylan/​Zimmerman, the son of Abram Zimmerman from Duluth, is a musician. There is still the question: why do utterances of those different sentences have different conveyed contents? Abstracting away from the details of any particular explanation it is unavoidable that we would need to appeal to some cognitive difference in the relation that agents stand in to the two sentences. But without departing from the default conception of semantics, propositions, or of cognitive structure, how could we capture any such difference? To see this, contrast two speech situations. On the one hand, we have Smith and Jones from above. Let’s stipulate that utterances of (3a) and (3b) would have the conveyed content just described. On the other hand, consider Bizzaro-​Smith and Bizzaro-​Jones, whose cognitive relation to “Dylan” and “Zimmerman” means that the conveyed contents of (3a) and (3b) would be reversed compared to Smith and Jones’ context. The conveyed content of an utterance of (3a), for them, would include the Russellian proposition that Dylan/​Zimmerman, the son of Abram Zimmerman from Duluth, is a musician. The conveyed content of (3b), for them, would include the Russellian proposition that Dylan/​Zimmerman, the famous sixties folk icon, is a musician. How, without departing from default assumptions, can we understand the difference between Smith and Jones, on one hand, and Bizzaro-​ Smith and Bizzaro-​Jones on the other? The details here will depend on what kind of pragmatic story one is telling. But any story will have to appeal to some difference like this: Smith believes that the name “Dylan” refers to a folk icon; she doesn’t believe that the name “Zimmerman” refers to a folk icon, etc. Bizzaro-​Smith believes that “Zimmerman” refers to a folk icon; she doesn’t believe that “Dylan” refers to a folk icon, etc. Thus Smith and Bizzaro-​Smith believe different Russellian propositions. But this is presumably not a brute fact about the psychology of Smith and Bizzaro-​Smith. It ought to be explained by something else. Focus on the following question: why is it that Smith believes that “Dylan” refers to a folk icon but Bizzaro-​Smith doesn’t believe that “Dylan” refers to a folk icon. Presumably this ought to be explained by a conjunction of linguistic knowledge and non-​linguistic knowledge.We might point to the fact that Smith believes the Russellian propositions in (10a)–​(10b). (10a) (10b)

that “Dylan” refers to Dylan/​Zimmerman that Dylan/​Zimmerman is a folk icon. 119

Aidan Gray

This might seem promising. But note that Bizzaro-​Smith also believes those Russellian propositions. She would express the (10b) proposition differently than Smith would; she would express it by saying “Zimmerman is a folk icon”. But this is precisely what the default story cannot explain.29 Recall from above that if all we say about Smith’s attitude state is that it contains a belief with the Russellian contents in (10a)–​(10b), we have said nothing that entails that Dylan is represented ‘as the same’ in those two attitudes. So we’ve said nothing that entails that Smith is in a position, for example, to infer from them that “Dylan” refers to a folk icon. So we can’t capture the difference between Smith and Jones’s context and Bizzaro-​Smith and Bizzaro-​Jones’s context that explains the difference in the conveyed content for utterances of (3a) and (3b). It seems, then, that any pragmatic story presupposes either a propositional or cognitive departure from the default assumptions. And given that, we should examine the interaction between those two kinds of explanation, and the relative contribution of each to the various challenges posed by cognitive significance.

Acknowledgments This chapter benefited from helpful feedback from the editors of this volume and from David Braun.

Further Reading Fine, Kit. 2007. Semantic Relationism. Wiley-​Blackwell. Frege, Gottlob. 1892. On Sinn and Bedeutung. In: The Frege Reader. Blackwell, 1997 151–​171. Goodman, Jeremy, and Lederman, Harvey. 2019. Perspectivism. Noûs. https://​doi.org/​10.1111/​nous.12322. Heck, Richard Kimberly. 2012. Solving Frege’s Puzzle. The Journal of Philosophy, 109, 132–​174 (originally published under the name “Richard G. Heck, Jr”). Recanati, François. 2012. Mental Files. Oxford University Press Richard, Mark. 1990. Propositional Attitudes:  An Essay on Thoughts and How We Ascribe Them. Cambridge University Press. Salmon, Nathan. 1986. Frege’s Puzzle. Ridgeview Publishing Company. Saul, Jennifer M. 2010. Simple Sentences, Substitution, and Intuitions. Oxford University Press.

Notes 1 For the sake of brevity, I’ll focus on proper names here. But analogous issues arise for other singular terms, and for expression of other semantic types. 2 For discussion see McLeod and Dunbar (this volume). 3 See Nelson (this volume). 4 For an overview, see Soames (this volume). 5 See Chalmers (2002), Jackson (this volume), Schoubye (this volume), and Sawyer (this volume) for different departures from Millianism. 6 See Boghossian (1992) for a classic discussion. 7 I’m glossing over some subtleties here. Most importantly, there are constructions which don’t appear to take sentential complements, yet appear to display the same behavior as ascriptions, for example ‘intensional transitive verbs’. 8 This claim is commonly made by Millians. One influential version of it motivates acceptance of the initially doubtful ascription by showing how it follows from other, intuitively acceptable, principles. An influential version of this idea can be found in Richard (1983). 9 This kind of example is introduced in Saul (1997) and explored in Braun and Saul (2002) and Saul (2010). 10 See Richard (this volume) for an overview. 11 See, e.g., Salmon (1986), Soames (1987a), and King (2007). This is a controversial component of the default. Many philosophers of language and mind hold that propositional content does not have sentence-​like structure. 12 There is also a possibility of claiming that utterances of Frege-​pairs don’t convey different contents, but they differ pragmatically in some other way; perhaps there are different pragmatic conditions on their appropriate utterance. Soames sometimes talks this way (see Soames 1987b).

120

Cognitive Significance 13 For overviews see Saul (1998) and Nelson (2019, section 5). For criticisms of Berg’s implicature approach, see Davis (2017). 14 See Sider and Braun (2006), Caplan (2007), and Speaks (2011). 15 See Richard (1986), Soames (1987b, 1987a), Salmon (1986, 1989), Berg (1988), Saul (1997, 1998). Sometimes these theorists speak generally about ‘conveyed content’ without specifying whether this is implicature or some other (unspecified) pragmatic mechanism. 16 Soames (2002). 17 Perry (2001), Korta and Perry (2011). 18 See Salmon (1986, 117), Soames (1987a), and Berg (1988, this volume). 19 This idea is most closely associated with Davidsonian approaches to semantics (see especially McDowell 1977). But it finds its way into model theoretic semantics approaches too. See discussion in Yalcin (2018). 20 See Harman (1972), Higginbotham (1991), Segal (1989), and Larson and Ludlow (1993). 21 Worries of this kind go back to Church’s (1950) discussion of Carnap (1947). 22 See Taschek (1995, 1998), Fine (2007, 2010b), Pinillos (2011, 2015), and Gray (2017). For related views, see Heck (2012) and Pryor (2016). 23 See Fine (2007, 2010a). For further discussion see Pickel and Rabern (2016, 2017). 24 For ‘expression in the language of thought’ see Fodor (1975, 1990); for ‘character’ see Kaplan (1989); for ‘guise’ see Salmon (1986, 1989); for ‘notion’ see Crimmins and Perry (1989), Perry (2001), and Crimmins (1992); for ‘way of believing’ see Braun (1998). The list could be extended. 25 Evans (1982), Forbes (1990), Perry (1980), Millikan (1997), Recanati (2012), Pryor (2016), Geirsson (this volume). 26 Evans (1982), Forbes (1990), Recanati (2012). 27 Braun (1998), Braun and Saul (2002), Saul (2010). 28 For related technical worries about approaches of this kind, see Sider (1995) and Soames (1995). 29 For versions of essentially the same worry, see Braun (2003) and Heck (2012).

References Berg, Jonathan. 1988. The pragmatics of substitutivity. Linguistics and Philosophy, 11, 355–​370. Boghossian, Paul A. 1992. Externalism and inference. Philosophical Issues, 2, 11–​28. Braun, David. 1998. Understanding belief reports. The Philosophical Review, 107(4), 555–​595. Braun, David. 2003. Review of Scott Soames. 2002. Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. Linguistics and Philosophy, 26(3), 367–​379. Braun, David, and Saul, Jennifer. 2002. Simple sentences, substitutions, and mistaken evaluations. Philosophical Studies, 111(1), 1–​41. Campbell, John. 1987. Is sense transparent? Proceedings of the Aristotelian Society, 88, 273–​292. Caplan, Ben. 2007. Millian descriptivism. Philosophical Studies, 133(2), 181–​198. Carnap, Rudolf. 1947. Meaning and Necessity: A Study in Semantics and Modal Logic. University of Chicago Press. Chalmers, David J. 2002. On sense and intension. Noûs, 36(s16), 135–​182. Church, Alonzo. 1950. On Carnap’s analysis of statements of assertion and belief. Analysis, 10(5), 97–​99. Crimmins, Mark. 1992. Talk about Beliefs. MIT Press. Crimmins, Mark, and Perry, John. 1989.The prince and the phone booth: Reporting puzzling beliefs. The Journal of Philosophy, 86(12), 685–​711. Davis, Wayne A. 2017. Berg’s answer to Frege’s Puzzle. Philosophia, 45(1), 19–​34. Evans, Gareth. 1982. The Varieties of Reference. Clarendon Press. Fine, Kit. 2007. Semantic Relationism. Wiley-​Blackwell. Fine, Kit. 2010a. Comments on Scott Soames “Coordination problems”. Philosophy and Phenomenological Research, 81(2), 475–​484. Fine, Kit. 2010b. Semantic necessity. In: Hale, Bob, and Hoffmann, Aviv (eds), Modality: Metaphysics, Logic, and Epistemology. Oxford University Press, 65–​80. Fodor, Jerry A. 1975. The Language of Thought. Harvard University Press. Fodor, Jerry. 1990. Substitution arguments and the individuation of beliefs. In: Boolos, George (ed.), Meaning and Method: Essays in Honor of Hilary Putnam. Cambridge University Press, 63–​79. Forbes, Graeme. 1990. The indispensability of Sinn. The Philosophical Review, 99(4), 535–​563. Frege, Gottlob. 1892. On Sinn and Bedeutung. In: The Frege Reader. Blackwell, 1997, 151–​171. Frege, Gottlob. 1918. The thought: A logical inquiry. In: The Frege Reader. Blackwell, 1997, 325–​345. Goodman, Jeremy, and Lederman, Harvey. 2019. Perspectivism. Noûs. https://​doi.org/​10.1111/​nous.12322. Gray, Aidan. 2017. Relational approaches to Frege’s Puzzle. Philosophy Compass, 12(10), e12429.

121

Aidan Gray Harman, Gilbert. 1972. Logical form. Foundations of Language, 9(1), 38–​65. Heck, Richard Kimberly. 2012. Solving Frege’s Puzzle. The Journal of Philosophy, 109, 132–​174 (originally published under the name “Richard G. Heck, Jr”). Higginbotham, James. 1991. Belief and logical form. Mind & Language, 6(4), 344–​369. Kaplan, David. 1989. Demonstratives. In: Almog, Joseph, Perry, John, and Wettstein, Howard (eds), Themes from Kaplan. Oxford University Press, 481–​563. King, Jeffrey C. 2007. The Nature and Structure of Content. Oxford University Press. Korta, Kepa, and Perry, John. 2011. Critical Pragmatics: An Inquiry into Reference and Communication. Cambridge University Press. Kripke, Saul. 1979. A puzzle about belief. In: Margalit, A. (ed.), Meaning and Use. Reidel, 239–​283. Kripke, Saul. 1980. Naming and Necessity. Harvard University Press. Larson, Richard K., and Ludlow, Peter. 1993. Interpreted logical forms. Synthese, 95(3), 305–​355. McDowell, John. 1977. On the sense and reference of a proper name. Mind, 86(342), 159–​185. Millikan, Ruth Garrett. 1997. Images of identity: In search of modes of presentation. Mind, 106(423), 499–​519. Nelson, Michael. 2019. Propositional attitude reports. https://​plato.stanford.edu/​archives/​spr2019/​entries/​prop-​ attitude-​reports/​. Perry, John. 1980. A problem about continued belief. Pacific Philosophical Quarterly, 61(4), 317–​322. Perry, John. 2001. Reference and Reflexivity. CSLI Stanford. Pickel, Bryan, and Rabern, Brian. 2016. The antinomy of the variable:  A Tarskian resolution. The Journal of Philosophy, 113(3), 137–​170. Pickel, Bryan, and Rabern, Brian. 2017. Does semantic relationism solve Frege’s Puzzle? Journal of Philosophical Logic, 46(1), 97–​118. Pinillos, N. Ángel. 2011. Coreference and meaning. Philosophical Studies, 154(2), 301–​324. Pinillos, N. Ángel. 2015. Millianism, relationism, and attitude ascriptions. In: Bianchi, Andrea (ed.), On Reference. Oxford University Press, 322–​334. Pryor, James. 2016. Mental graphs. Review of Philosophy and Psychology, 7(2), 309–​341. Recanati, François. 2012. Mental Files. Oxford University Press. Richard, Mark. 1983. Direct reference and ascriptions of belief. Journal of Philosophical Logic, 12(4), 425–​452. Richard, Mark. 1986. Attitude ascriptions, semantic theory, and pragmatic evidence. Proceedings of the Aristotelian Society, 87, 243–​262. Richard, Mark. 1990. Propositional Attitudes:  An Essay on Thoughts and How We Ascribe Them. Cambridge University Press. Salmon, Nathan. 1986. Frege’s Puzzle. Ridgeview Publishing Company. Salmon, Nathan. 1989. Illogical belief. Philosophical Perspectives, 3, 243–​285. Saul, Jennifer M. 1997. Substitution and simple sentences. Analysis, 57(2), 102–​108. Saul, Jennifer M. 1998. The pragmatics of attitude ascription. Philosophical Studies, 92(3), 363–​389. Saul, Jennifer M. 2010. Simple Sentences, Substitution, and Intuitions. Oxford University Press. Schiffer, Stephen. 1977. Naming and knowing. Midwest Studies in Philosophy, 2, 28–​41. Schiffer, Stephen. 1987. The ‘Fido’-​Fido theory of belief. Philosophical Perspectives, 1, 455–​480. Schiffer, Stephen. 1992. Belief ascription. The Journal of Philosophy, 89(10), 499–​521. Segal, Gabriel. 1989. A preference for sense and reference. The Journal of Philosophy, 86(2), 73–​89. Sider, Theodore. 1995. Three problems for Richard’s theory of belief ascription. Canadian Journal of Philosophy, 25(4), 487–​513. Sider, Theodore, and Braun, David. 2006. Kripke’s revenge. Philosophical Studies, 128, 669–​682. Soames, Scott. 1987a. Direct reference, propositional attitudes, and semantic content. Philosophical Topics, 15(1),  47–​87. Soames, Scott. 1987b. Substitutivity. In:  Jarvis Thomson, Judith (ed.), On Being and Saying:  Essays for Richard Cartwright. MIT Press, 99–​132. Soames, Scott. 1995. Beyond singular propositions? Canadian Journal of Philosophy, 25(4), 515–​549. Soames, Scott. 2002. Beyond Rigidity. Oxford University Press. Speaks, Jeff. 2011. Frege’s Puzzle and descriptive enrichment. Philosophy and Phenomenological Research, 83(2), 267–​282. Taschek, William W. 1995. Belief, substitution, and logical structure. Noûs, 29(1), 71–​95. Taschek, William W. 1998. On ascribing beliefs: Content in context. The Journal of Philosophy, 95(7), 323–​353. Yalcin, Seth. 2018. Semantics as model-​based science. In: Ball, Derek, and Rabern, Brian (eds), The Science of Meaning. Oxford University Press, 334–​360.

122

9 CONVERSATIONAL IMPLICATURE IN BELIEF REPORTS Jonathan Berg

A Question about Two Interpretations of Belief Reports It is natural to suppose that one can believe A is F without believing B is F, even when A=B. For instance, we are inclined to accept (1) Lois believes Clark Kent is a reporter, while rejecting (2) Lois believes Superman is a reporter (assuming the truth of the standard story, where Lois believes that there are two distinct individuals, one of whom is called ‘Clark Kent’ and is a reporter, the other of whom is called ‘Superman’ and is not a reporter, when in fact the two names refer to the same person). On the other hand, it is not hard to think of a context, in the very same state of affairs, where we would accept (2). Imagine someone marveling as follows at Superman’s skill in disguising himself: Look at what a master of disguise that Superman is! Why, when he puts on that suit and strolls into his office at the Daily Planet, he fools everybody—​Mr. White, Jimmy Olson; even Lois Lane believes Superman is a reporter. Arguably, this shift in the acceptability of (2) is due to a shift in how an utterance of (2) is interpreted. On this account the initial rejection of (2) comes from interpreting it “notionally”, or “de dicto”, while the acceptance of (2) in the given context of a discussion of Superman’s skillful disguise comes from interpreting it “relationally” or “de re”. Leaving aside the question of how best exactly to draw this distinction, I will assume only that the relational interpretation, but not the notional one, allows for the interchange of coreferential names salva veritate.1,2 Thus arises a question about the semantic status of these two interpretations of (2): are they two senses—​two literal semantic contents—​of (2)? Is (2) semantically ambiguous between these two interpretations? Or, is only one of these two interpretations a sense or semantic content, the other interpretation being some sort of pragmatic content of certain utterances of (2)? I will argue for the latter. 123

Jonathan Berg

First I will set the stage by recalling a couple of examples of this kind of pragmatic strategy, then I will briefly review the evidence against taking belief reports as semantically ambiguous; and then I will sketch a view of conversational implicature in belief reports.

Two Examples As Grice observes: It has, for example, been suggested that because it would be incorrect or inappropriate to say “He got into bed and took off his trousers” of a man who first took off his trousers and then got into bed, it is part of the meaning, or part of one meaning, of “and” to convey temporal succession. 1989, 8 So it seems that at least under one interpretation, from (3) Smith got into bed and took off his trousers we would infer (4) Smith took off his trousers after he got into bed. But (3)  can be interpreted as lacking this consequence about temporal order, as shown by the plausibility of (5) Smith got into bed and took off his trousers, but not in that order. Thus we arrive once again at a question about the semantic status of two interpretations: are these two interpretations of (3) two senses—​semantic contents—​of (3)? Is (3) semantically ambiguous between these two interpretations? Or, is only one of these two interpretations a sense or semantic content, the other interpretation being some sort of pragmatic content of certain utterances of (3)? We could say that ‘and’ has two senses, a temporal and a nontemporal sense, or we could say that ‘and’ has only the nontemporal sense, and that the temporal interpretation that we sometimes give to conjunctions is derived somehow pragmatically.3 Another example—​Donnellan’s distinction between referential and attributive uses of definite descriptions: A speaker who uses a definite description attributively in an assertion states something about whoever or whatever is the so-​and-​so. A speaker who uses a definite description referentially in an assertion, on the other hand, uses the description to enable his audience to pick out whom or what he is talking about and states something about that person or thing. 1966, 285 Thus, (6) Smith’s murderer is insane could be interpreted as requiring that someone insane murdered Smith, or it could be interpreted as requiring merely that whoever the speaker meant to be talking about is insane.

124

Implicature in Belief Reports

Are these two interpretations of (6)  two senses—​semantic contents—​of (6)? Is (6)  semantically ambiguous between these two interpretations? Or, is only one of these two interpretations a sense or semantic content of (6), the other interpretation being some sort of pragmatic content of certain utterances of (6)? Assuming that ‘Smith’s murderer’ is equivalent to ‘the murderer of Smith’, we could say that the word ‘the’ has two senses, an attributive one and a referential one; or we could say that ‘the’ (or the phrase ‘the murderer of Smith’) has only the attributive sense, and that the referential interpretation that we sometimes give to definite descriptions is derived somehow pragmatically.4 Kripke famously argues for the latter. I do not intend to discuss these two cases; I mention them only to put the current question in perspective.5 From my preferred perspective, the question about the notional use of belief ascriptions is like these questions about the temporal use of ‘and’ and the referential use of definite descriptions, where multiple interpretations suggestive of a semantic ambiguity could plausibly be viewed as a univocal semantic content together with some nonliteral (nonsemantic) content that is derived pragmatically. After a cursory review of the evidence against taking belief ascriptions as semantically ambiguous, I will outline a pragmatic account of the notional interpretation.

Why Aren’t Belief Reports Ambiguous? If belief ascriptions such as (2) Lois believes Superman is a reporter are semantically ambiguous between relational and notional senses, then if we are to avoid what Kripke calls “the lazy man’s approach in philosophy” (namely, “to posit ambiguities when in trouble”)—​then such ascriptions should display the same signs of ambiguity that we typically expect to see in cases of semantic ambiguity. But they don’t. Consider first the existence of disambiguating expressions. The lexically ambiguous word ‘board’, for instance, is disambiguated by the words ‘plank’ and ‘committee’, each of which has as its meaning a different one of these two senses of ‘board’. (For ease of exposition we may assume for now that these are the only senses of ’board’.) Are there disambiguating expressions that pick out two senses of the word ‘believe’ corresponding to the notional and relational interpretations of belief ascriptions? Not in English, nor presumably any language spoken by members of the audiences before whom I have discussed this. What, then, about languages that we don’t know? If you started learning a new language, would you expect it to have disambiguating expressions corresponding to notional and relational senses of the word ‘believe’? If you discovered that this language does have such disambiguating expressions, would it not be surprising? I would not expect to find such disambiguating expressions in any natural language, and in fact would be surprised to find them. Another standard sign of ambiguity is an expression’s resistance to “crossed interpretations” for sentences reduced by conjunction or ellipsis. For instance, Harry and Sally are both on the board could be interpreted as meaning that they’re both on some committee or that they’re both on some plank—​but not as meaning that one is on some committee and the other is on some plank. This supports the claim that ‘board’ is indeed ambiguous. What about ‘believes’? Consider this case of VP (verb phrase)-​ellipsis reduction: I believe Superman is a reporter and so does Lois.

125

Jonathan Berg

This clearly allows crossed interpretations, where my believing that Superman is a reporter is construed notionally, and Lois’s believing that Superman is a reporter is construed relationally. The plausibility of this interpretation is especially clear in light of this variation: I believe Superman is a reporter and so does Lois (though she wouldn’t put it in those words). So ‘believe’ doesn’t pass the crossed interpretations test for ambiguity. Aside from the fact that belief ascriptions such as (2) do not seem to meet the standard criteria for lexical ambiguity, another reason for thinking such ascriptions are not ambiguous arises from the great multitude of expressions to which the relational/​notional distinction can be applied. In particular, if ‘believes’ is ambiguous in the way considered, then so too must be all expressions synonymous with it—​including complex expressions such as ‘holds the view that’, ‘is of the opinion that’, and so on. This leads to two problems. First of all, exactly where in these complex expressions does the alleged ambiguity originate? Is ‘view’ ambiguous? Is ‘holds’? Is ‘opinion’? Or, instead of trying to locate the ambiguity in the particular words comprising these complex expressions, are we to take them (the complex expressions) as merely idiomatic? Neither option looks promising. A further problem with the strategy of extending the appeal to ambiguity from ‘believes’ to all expressions synonymous with it is that we generally don’t ascribe lexical ambiguity en masse. If some duality of interpretations seems to occur across a wide range of expressions, then instead of taking all those expressions as ambiguous in the same way, we would generally look for some other explanation. For instance, although a toy gun is not a gun, a toy car is not a car, a toy boat is not a boat, and so on, we wouldn’t say that the words ‘gun’, ‘car’, ‘boat’, etc., therefore each have at least two senses, one applying to guns, cars, boats, etc., and another that would apply to toys of these kinds. An alternative to taking belief ascriptions such as (2) Lois believes Superman is a reporter as lexically ambiguous is to take them as syntactically ambiguous. The most popular way to do this has been Russell’s method, whereby proper names are analyzed as definite descriptions, which are then eliminated in accordance with Russell’s theory of descriptions. Thus, (2) presumably gives way to something like this: Lois believes that the superhero from Krypton is a reporter; which, when Russell’s theory of descriptions is applied relative to the embedded sentence yields the narrow scope interpretation, (2a) Lois believes that one and only one entity is a superhero from Krypton and that that entity is a reporter; and which, when Russell’s theory is applied relative to the whole sentence yields the wide scope interpretation, (2b) One and only one entity is a superhero from Krypton, and Lois believes that entity is a reporter.6 Viewed in this way (2) seems to be syntactically ambiguous between (2a) and (2b). The main problem with Russell’s approach is that, for reasons I won’t go into here, each of his two central assumptions is notoriously problematic; both his theory of definite descriptions, and especially his view of common proper names as concealed definite descriptions, are widely contested and best 126

Implicature in Belief Reports

not relied upon. And in any case, the crossed interpretations test—​which (2) was already shown to fail—​works as well for syntactic ambiguity as for lexical ambiguity. These then are my main reasons for not construing belief ascriptions as semantically ambiguous between relational and notional interpretations.7 Actually these are only half my reasons; the other half consists in a pragmatic account of the notional interpretation, to which I now turn.

The Principle of Implicated Normalcy The pragmatic account I favor of the notional interpretation of belief reports—​that is, of our inclination (when such an inclination exists) to reject (2) Lois believes Superman is a reporter as false—​is based on the Principle of Implicated Normalcy: Speakers generally conversationally implicate that the circumstances regarding whatever they are speaking of are not abnormal in any significant, unanticipated, unindicated way. Berg 2012, 58 This principle derives from the simple fact that things are normally normal, and so we normally presume that things are normal. Of course, things aren’t always normal; normally people don’t like pain, yet sometimes they do. But it’s trivial that normally things are normal, and so we normally presume this. In light of the presumption of normalcy, it is not hard to see how the Principle of Implicated Normalcy obtains, on the basis of Grice’s first maxim of Quantity:  “Make your contribution as informative as is required (for the current purposes of the exchange)”. By the presumption of normalcy, listeners will tend to assume that the circumstances with regard to what the speaker is talking about are normal.Therefore, to whatever extent those circumstances are not normal, the listener will be left with the mistaken impression that they are normal—​unless the speaker does something to prevent the listener from getting this mistaken impression.Thus, if a speaker is following this maxim—​making her conversational contribution as informative as needed for the purpose of the exchange—​the listener may infer that if something important was not normal about the circumstances regarding what the speaker is speaking about, the speaker would have said so, to prevent the listener from getting the wrong idea. So, by not saying that the circumstances are not normal in some significant way, the speaker conversationally implicates that the circumstances are normal. The sort of conversational implicature predicted by the Principle of Implicated Normalcy can be worked out by the listener as follows: 1 Speaker S has said that p. (Assumption) 2 It is common knowledge (at least to S and me) that normally, if it is true that p, it is true that q. (Assumption) 3 S knows that from his assertion that p, I would normally infer that S thinks that q. (From 2) 4 S is following the maxim of Quantity, “Make your contribution as informative as is required (for the current purposes of the exchange)”. (Assumption) 5 Making one’s contribution as informative as is required requires preventing any significant potential misimpressions. (Assumption) 6 If S did not think that q, S would have done something to stop me from inferring that q (if it mattered for the purposes of the exchange). (From 1, 3, 4, 5) 7 S has done nothing to stop me from inferring that q. (Assumption) 8 S thinks that q. (From 6, 7) 127

Jonathan Berg

The Principle of Implicated Normalcy should not be controversial. As Jason Stanley observes, “when someone tells us that John ate this morning, we assume he did so in the normal way” (2007, 204). We assume, for instance, that he put the food in his mouth with silverware or with his fingers, not mouthing it up out of a doggie bowl on his hands and knees; that he ate a lot more than 1 gram and a lot less than 10 kilograms; that he knew what he was eating; and so on. If any of these assumptions of normalcy were known by our informant to be false, we would have, ceteris paribus, expected our informant to disabuse us of them. More generally, we expect cooperative speakers to tell “the whole truth”—​in particular, we expect them to avoid giving us any wrong ideas, that is, to correct any false beliefs they might think we have inferred from what they said. Of course, speakers cannot be expected to rule out every abnormality in the circumstances regarding whatever they’re talking about; the Principle of Implicated Normalcy is explicitly restricted to circumstances that are not abnormal in any significant, unanticipated, unindicated way. A speaker is not expected to rule out abnormal circumstances that are insignificantly so, that don’t matter for the purposes of the exchange. Nor must a speaker address abnormalities that are already known or expected. Speakers also needn’t warn their listeners about abnormalities that are indicated in some other way. It’s only when the circumstances regarding whatever speakers are speaking of are abnormal in any significant, unanticipated, unindicated way, that the maxim of Quantity kicks in, requiring speakers to set the record straight and prevent their listeners from getting the wrong idea due to their presumption of normalcy. It should be clear that if the Principle of Implicated Normalcy and the thinking behind it are correct, then there is a lot more that gets conversationally implicated than what we are explicitly aware of. Suppose, for instance, that in recounting to a friend what I did over the weekend, I say, “I went for a walk on the beach and watched a beautiful sunset”. According to the Principle of Implicated Normalcy, I would thereby conversationally implicate that there had been no beached whales in my path. For if there had been beached whales in my path, I should have said so, in order to be adequately informative—​since I assume that my listener presumes that the circumstances of my walk were normal, and they would not have been normal if there had been beached whales in my path; but I didn’t say there had been any beached whales in my path, thereby leaving my listener to surmise that there hadn’t been. Likewise, I would at the same time be implicating that the beach where I went walking was not being bombed, that it was not overrun by herds of wild elephants, and that it was not strewn with $100 bills. Clearly, then, I am assuming that speakers need not explicitly intend to convey, nor even be aware of, all that they conversationally implicate. This seems to be consistent with at least some of what Grice himself says—​for instance: A is standing by an obviously immobilized car and is approached by B; the following exchange takes place: (1)

A: I am out of petrol. B: There is a garage around the corner.

(Gloss: B would be infringing the maxim “Be Relevant” unless he thinks, or thinks it is possible, that the garage is open and has petrol to sell; so he implicates that the garage is open, or at least may be open, etc.) 1989, 32 It is entirely plausible (and I think most likely) that B makes his contribution to this conversation without giving the slightest thought to the garage’s being open or its having petrol to sell. Moreover, Grice’s use of the word ‘etc.’ at the end of this passage seems to point to a potentially never-​ending

128

Implicature in Belief Reports

list of conversational implicata not explicitly entertained by B. This interpretation of Grice is further supported by his remarks on calculability: Since, to calculate a conversational implicature is to calculate what has to be supposed in order to preserve the supposition that the Cooperative Principle is being observed, and since there may be various possible specific explanations, a list of which may be open, the conversational implicatum in such cases will be [a]‌disjunction of such specific explanations; and if the list of these is open, the implicatum will have just the kind of indeterminacy that many actual implicata do in fact possess. 1989, 40 I need not, however, rely on any particular interpretation of Grice. For my purposes here it will suffice to stipulate that I shall use the word ‘implicate’ in such a way that speakers need not explicitly intend to convey, nor even be aware of, all they implicate.

Normalcy and Belief How does the Principle of Implicated Normalcy bear on belief ascriptions? What’s normal for belief ascriptions? For one thing, I think that normally belief ascriptions are acceptable verbatim, in the sense that the person to whom the belief is being ascribed would—​normally—​accept that very ascription word for word as being true. So, for instance, if the ascription Meryl Streep believes her best performance was in Sophie’s Choice is true, then if we were to show her this ascription and ask her if it’s true, she would normally say it is. Of course, there are many ways that a true belief ascription might not be acceptable verbatim—​the purported believer might not understand all the words in the ascription, she might not know the intended referents of indexical expressions, she might be unwilling or unable to cooperate, and so on. But normally belief ascriptions are acceptable verbatim. Since belief ascriptions are normally acceptable verbatim, it follows by the Principle of Implicated Normalcy that normally in uttering a belief ascription one conversationally implicates that the ascription is acceptable verbatim. Thus, by uttering (2) Lois believes Superman is a reporter one would normally implicate that (2) is acceptable verbatim—​that Lois would accept this ascription. On the other hand, in suitably abnormal cases there is no such conversational implicature. In particular, when discussing how effectively Superman fools Lois in his Clark Kent disguise, an utterance of (2) does not carry the conversational implicature that (2) is acceptable verbatim by Lois, since the presumption of normalcy with regard to the verbatim acceptability of belief reports is suspended. Given that Lois is fooled into thinking that the names ‘Superman’ and ‘Clark Kent’ refer to two different people, the circumstances are anticipated to be abnormal in a way that prevents the listener from inferring that (2) is acceptable to Lois verbatim. Thus we have the outline of a pragmatic account of the inclination, where it exists, to reject (2), thereby obviating the need for positing a second semantic content for (2) corresponding to the notional interpretation of utterances of (2). However, extending this account to other belief ascriptions sometimes requires adverting to different sorts of conversational implicata. For instance, where we are ascribing a belief to someone who does not speak the language of the ascription, we surely would

129

Jonathan Berg

not be implicating that the ascription is acceptable verbatim. The relevant implicatum in such a case might be that a proper translation of the ascription to the language of the purported believer would be acceptable verbatim to her. Thus, it is important to recognize that the particular implicatum that will be relevant to the pragmatic account of the notional interpretation will depend on the particular case being considered.Verbatim acceptability of a given ascription is not always relevant.

Recap I have tried to show how the puzzle of apparent substitution failure of coreferential proper names in belief reports might usefully be viewed as a question about which interpretations of certain utterances of belief reports should be taken as semantic contents of the sentences uttered. After setting the stage with a couple of examples of this kind of discussion, as it arises with regard to the temporal interpretation of conjunctions and the referential use of definite descriptions, I reviewed the main evidence against taking belief reports as semantically ambiguous between notional and relational senses. I then explained how the notional interpretation of belief reports can be seen as conversationally implicated rather than as part of the semantic content of belief report sentences. At the heart of my argument lies the Principle of Implicated Normalcy, which says that speakers generally conversationally implicate that the circumstances regarding whatever they are speaking of are not abnormal in any significant, unanticipated, unindicated way. In particular, the utterer of a belief report normally implicates that her report would be acceptable in just those words to the reputed believer. This allows us to account for our inclination to reject a belief report such as (2) Lois believes Superman is a reporter without positing a notional sense of the sentence in addition to the relational sense. An added benefit of framing the issue this way—​as a question of whether to posit for any belief report sentence such as (2)  a notional sense in addition to the relational sense—​is that it neutralizes the most common and influential objection to pragmatic accounts of substitution failure:  that taking ascriptions such as (2)  as literally true is wildly counterintuitive. Such counterintuitivity about the truth-​values of belief ascriptions dissolves when attention is focused on the de re interpretation.8

Acknowledgment This research was supported by the ISRAEL SCIENCE FOUNDATION (grant no. 1220/​17).

Further Reading Berg, Jonathan. (2012) Direct Belief: An Essay on the Semantics, Pragmatics, and Metaphysics of Belief. Mouton Series in Pragmatics, 13. Berlin and Boston, MA: De Gruyter Mouton. Donnellan, Keith. (1966) “Reference and Definite Descriptions”. Philosophical Review, 75, 281–​304. Grice, Paul. (1989) Studies in the Way of Words. Cambridge, MA: Harvard University Press. Kripke, Saul. (1977) “Speaker’s Reference and Semantic Reference”. Midwest Studies in Philosophy, 2, 255–​276. Rpt. in Contemporary Perspectives in the Philosophy of Language. Ed. French et al., pp. 6–​27. Recanati, François. (1993) Direct Reference. Oxford: Blackwell. Salmon, Nathan. (1986) Frege’s Puzzle. Cambridge, MA: MIT Press. Soames, Scott. (2002) Beyond Rigidity. Oxford: Oxford University Press. Soames, Scott. (2008) “Drawing the Line between Meaning and Implicature—​and Relating both to Assertion”. Noûs, 42, 440–​465. Stanley, Jason. (2007) Language in Context. Oxford: Oxford University Press.

130

Implicature in Belief Reports

Notes 1 This discussion will be limited to the interchange of coreferential proper names. 2 “Interpretations” of utterances are to be distinguished from “senses”, understood as semantic contents of expressions. 3 Logical space also allows for saying that ‘and’ has only the temporal sense, and that the nontemporal interpretation that we sometimes give to conjunctions is derived pragmatically; but such a line looks like a nonstarter. 4 Once again, this time perhaps less implausibly, logical space also allows for a third possibility, namely, that ‘the’ has only the referential sense, and that the attributive interpretation that we sometimes give to definite descriptions is derived pragmatically. 5 For a thorough discussion of the referential-​attributive distinction see Anne Bezuidenhout, this volume. 6 For ease of exposition we may skip the Russellian analysis of the names ‘Lois’ and ‘Krypton’. 7 For more on the claim that belief ascriptions are semantically ambiguous between notional and referential ‎readings, see my Direct Belief (Berg 2012, 12–​24). I also argue against taking belief ascriptions as either indexical or semantically indeterminate (Berg 2012, 24–​48). 8 For more on the role of intuitions in the semantics of belief ascriptions see Recanati (1993) on “availability” and my replies (Berg 2012, 75–​91).

References Berg, Jonathan. (2012) Direct Belief: An Essay on the Semantics, Pragmatics, and Metaphysics of Belief. Mouton Series in Pragmatics, 13. Berlin and Boston, MA: De Gruyter Mouton. Donnellan, Keith. (1966) “Reference and Definite Descriptions”. Philosophical Review, 75, 281–​304. Grice, Paul. (1989) Studies in the Way of Words. Cambridge, MA: Harvard University Press. Kripke, Saul. (1977) “Speaker’s Reference and Semantic Reference”. Midwest Studies in Philosophy, 2, 255–​276. Rpt. in Contemporary Perspectives in the Philosophy of Language. Ed. French et al., pp. 6–​27. Recanati, François. (1993) Direct Reference. Oxford: Blackwell. Stanley, Jason. (2007) Language in Context. Oxford: Oxford University Press.

131

10 CONTEXT SENSITIVITY AND ‘BELIEVES’ Mark Richard

1 What should we understand by ‘context sensitivity’? David Kaplan’s Demonstratives (1989) has led many to understand it so. Uses of declarative natural language sentences have interpretations—​semantic values in linguist-​speak—​which determine truth conditions; a sentence use’s semantic value is systematically determined by its syntax and the semantic values of its parts on that use. Contexts are situations in which a sentence could be used, and thus relative to which it could have a semantic value. Contexts vary in indefinitely many ways—​in terms of who is speaking, who the audience is, what is common ground between speaker and audience, what the local weather is, whether the participants like one another, etc. Some phrases have meanings whose semantic values vary systematically with some aspect of context; a phrase like this it is said to be context sensitive.1 The poster kids for Kaplan-​style context sensitivity are indexicals like ‘ich’ and demonstratives such as ‘este’. A use of the German ‘ich’, zum Beispiel, refers to its user; a use of ‘este’, por ejemplo, to what’s demonstrated when it’s used. In the wake of Kaplan’s work on demonstratives and indexicals, theorists have investigated whether expressions which initially do not look to be context sensitive in fact are. This article is concerned with one such locus of putatively context-​sensitive expressions, propositional attitude verbs like ‘believes’ and ‘wonders’, and the verb phrases which they head, such as ‘believes that Lux likes Parker’ or ‘wonders whether Jade likes Lux’. As is usual in such discussions, attention here will be confined to ‘believes’ and the verb phrases it heads.

2 Conventional philosophical wisdom has it that ‘believes’ is no more context sensitive than run-​of-​ the-​mill transitive verbs like ‘summon’. Frege, for example, took the words after ‘that’ in (1) Victoria believes (today, 31 October 2019) that Selina has left to refer to the sense (‘thought’, in Frege-​speak) expressed by ‘Selina has left’, so that (1) as a whole says that Victoria stands in a relation (belief) to the thought that Selina has left; Fregean thoughts, in turn, are constituted from ‘ways of thinking’ (‘modes of presentation’, in Frege-​speak) of objects and properties. (The early) Russell held that attitudes were relations to propositions cum constructions 132

Context Sensitivity and ‘Believes’

from objects, properties, and relations, so that in (1) ‘believes’ picks out a relation to something whose parts are Selina and the property of having left. On such views, the idea that (1) could vary in truth value across contexts makes no more sense than the idea that (2) Selina summoned (today, 31 October 2019) Victoria could. Either Selina summoned Vicky or she didn’t; either Vicky believes that Selina left or she does not. Neither view is all that attractive. Frege seemed to think that (ignoring the embarrassment of words like ‘now’ and ‘I’), the sense of words in a public language does not much vary from speaker to speaker. Sense, as a ‘way of thinking’ of what one uses words to talk about, is supposed to capture the cognitive significance of phrases. But it is wildly implausible that we all have the ‘same way of thinking’ of the objects we refer to, much less associate the same way of thinking of an object with its name.This makes it hard to understand what is going on with an utterance of (1). If Victoria expresses a belief on Halloween with (3) Selina has left, I can reliably tell you that using (1). But there is no guarantee that Vicky and I assign the same sense to (3), so no guarantee that (1)  is correct, if in it ‘Selina has left’ refers to my sense of (3) instead of Vicky’s. While Frege did not think that ‘believes’ was sensitive to non-​linguistic context, he did think that that verb created a linguistic context which affected the semantic values of expressions occurring in it. Normally, my use of ‘Selina’ refers to Selina, my use of ‘has left’ refers to the property of having left. But when I utter (1), I refer to the senses I customarily express with these words. One might think that an epicycle on Frege’s views might be of help: why not say that in a sentence of the form Victoria believes that S the words in S refer to whatever sense Victoria assigns to them (or to a translation of them into Victoria’s language)? If so, my use of (1) can be true even if Victoria and I think of Selina in very different ways. Unfortunately, this proposal invalidates the argument from (1) and (4) Bulle believes (today, 31 October 2019) that Selina has left to (5) There is something Bulle and Victoria believe, since on this proposal (1) and (4) may be true even if they share no way of thinking of Selina. The early Russell’s view does not suffer from these sorts of problem: even if Victoria and Bulle think of Selina in different ways, that does not prevent them from ascribing the same property to the same person with (3) and ‘Selina a quitté’. But as is well known it looks to be unable to deal with the fact that, because of the different ways of thinking of Bulle which Victoria associates with ‘Bulle’ and ‘Marie-​France’ (which are both names of Bulle Ogier), it may be the case that (6) Victoria thinks that Bulle is Bulle (7) Victoria thinks that Bulle is Marie-​France diverge in truth value, something that can’t be the case if thoughts expressed by sentences of the form a is b simply ascribe identity to the subject and object of the sentence. One can retreat, as Russell did, to the idea that uses of ordinary proper names are ‘disguised definite descriptions’, thus explaining 133

Mark Richard

why ‘Bulle is Bulle’ expresses a belief of hers when Victoria uses it, while ‘Bulle is Marie-​France’ does not. But this tack makes it mysterious as to why we should be confident that, if Victoria and Bulle both express beliefs with ‘Selina has left’, we say something true using (5).2

3 That the philosophical tradition has a hard time telling a coherent story about sentences like (6) and (7) does not imply that verbs like ‘believes’ are context sensitive. What does? Recall poor Pierre, who was raised as a monolingual in France where he called London ‘Londres’. Spirted away to London, he learned English on the streets and so learned to call the city he lived in ‘London’. He never did get it, never realized that London is Londres.3 One day, Pierre finds himself in the Dublin airport, waiting for the next flight to London which, he is pretty certain, leaves from Gate 43. Barry stops him and asks ‘What gate is the flight to London?’, to which he replies ‘I think it’s Gate 43’. Barry’s wife asks him ‘Where does he think we should go?’, to which he responds (8) He thinks the flight to London is at Gate 43. Barbet, at that moment, having discerned from Pierre’s accent that he is French, asks him ‘A quelle porte se trouve le vol pour Londres?’, to which Pierre, replies (in French, of course) ‘Je n’en aucune idée; pas ici’, gesturing towards Gate 43. Barbet’s wife, who is English, asks him ‘Does he think this [Gate 43] is the gate for our flight?’, to which he responds (9) No, he doesn’t think the flight to London is at Gate 43. (8) and (9) are perfectly natural, completely acceptable things for Barry and Barbet to say. They strike many of us not simply as felicitous but as correct. But if that is so, then sentences like (8) look to vary in truth value across (simultaneous) contexts. But this does not appear to be due to a shift in meaning or reference of the words in (8)’s complement, and it is certainly not due to a shift in meaning or reference of the sentence’s subject. So something must have shifted. But what? One answer begins with the idea that something like the early Russell’s semantic view is correct: the job of semantics is to identify the semantic values of phrases, both simple and complex; those semantic values are the objects and properties we use those phrases to talk about. So semantics’ job is to assign something—​call it a ‘singular proposition’—​to a sentence, something structured and with semantic values as parts; this not only determines the truth conditions of the sentence but in some sense is or represents what the sentence says. Examples like Pierre make it obvious that given a singular proposition, there are ‘different ways’ to believe it: Pierre, after all, can think that London is in England ‘either in French or in English’. The example of Barry and Barbet makes it plausible that speakers may refer to, bring up, allude to, or otherwise invoke such ways of believing when using a phrase like ‘thinks the flight to London is at Gate 43’, for it seems plausible that while Barry is reporting on Pierre’s ‘English way’ of thinking that the flight leaves from 43, Barbet is reporting on his ‘French way’ of believing this. It is this—​what ‘ways of believing’ a proposition or of ‘having something in mind’ a user of (8) invokes—​that varies across contexts. Here are some versions of this answer. According to Stephen Schiffer (1977, 1992), sentences like (8) and (9) contain ‘implicit indexical component[s]‌requiring reference to a mode of presentation or type of mode of presentation’. (8) thus contains a complex, implicit indexical component which in use refers to a mode of presentation of the singular proposition that the flight to London leaves from Gate 43.When Barry utters (8), he refers to Pierre’s “French mode of presentation” of this proposition, correctly conveying that Barry believes the proposition “under” this mode of presentation; when Barbet utters a denial of (8), he refers to Pierre’s “English way of thinking” of the proposition, conveying correctly that Pierre does not believe the proposition under that way of thinking.4 134

Context Sensitivity and ‘Believes’

Mark Crimmins and John Perry (Crimmins and Perry 1989; Crimmins 1992) endorse a variant of this view. Simplifying: uses of (8) are (typically) about Pierre’s mental representations.When someone uses the sentence they have particular representations of London, and Gate 43 (and the relation of being at, and so on) in mind, and they convey that Pierre believes the singular proposition about London and Gate 43 in virtue of putting these representations together in a certain way.5 Mark Richard (1990, 2012) also holds that mental representations or ways of thinking are invoked by speakers. According to Richard, attitude ascription involves using the complement of an ascription to represent or ‘translate’ some (interpreted) representation of the ascription’s subject. Consider the example of Barry and Barbet, who are each focused on representations of Pierre’s: Barry is reporting a belief of Pierre’s expressed using ‘London’, Barbet one expressed using ‘Londres’. When Barry and Barbet utter (8) and (9), each uses ‘London’ to ‘translate’ a representation of Pierre’s: Barry uses it to translate ‘London’, Barbet to translate ‘Londres’. With (8) Barry is telling his wife that, relative to a policy of translating Pierre’s ‘London’ with London, one of Pierre’s belief (-​making representation)s can be translated with ‘that the flight to London is at Gate 43’; with (9) Barbet is telling his wife that, relative to a policy of translating Pierre’s ‘Londres’ with ‘London’, one cannot translate one of Pierre’s belief (-​making representation)s with ‘that the flight to London is at Gate 43’. In a bit more detail, Richard’s view is as follows. Take a representation r which determines a singular proposition p. Construct something with the structure of p by pairing each constituent of p with the constituent of r that determines it. If, for example, you begin with Pierre’s ‘Londres est jolie’ and the singular proposition that London is pretty (which we can, for exposition’s sake, identify with (10) ), you end up with (11) . Call such fusions of propositional determining representations with the propositions they determine articulated propositions; call the set of articulated propositions determined by the representations which realize a person’s beliefs her representational system.6 Now, the complement of a belief ascription like (12) Pierre believes that London is pretty, determines an articulated proposition –​in the case of (11) what’s determined is (11′) . Richard proposes that a speaker may intend, and expect her audience to recognize the intention, that expressions which she uses in an attitude ascription’s complement are meant to represent (sorts of) representations which the subject employs. These intentions and expectations establish a ‘translation manual’ geared to representing the way a speaker holds (or does not hold) a belief. In the example involving sentences (8) and (9), Barry’s intentions establish the manual: (M1) When speaking of Pierre, use to represent , while Barbet’s establish the manual (M2) When speaking of Pierre, use to represent . 135

Mark Richard

When someone utters (12), they are offering the articulated proposition (11′) as something which adequately translates some articulated proposition which is one of Pierre’s beliefs, with the standards of adequacy varying across contexts as does the ambient translation manual. Since the standards of adequacy vary across contexts, sentences like (8) and (12) are contextually sensitive.7 These accounts hardly exhaust the accounts of belief ascriptions that hold that, while ascriptions of the form a believes that S convey that someone believes something which is singular, what is conveyed depends upon the context. For example, Scott Soames (2002) argues that when someone uses sentence (8) or (9), the (semantic value of) the complement may undergo ‘pragmatic enrichment’, so that belief in a proposition more descriptively specific than the singular proposition that the flight to London is at Gate 43 is ascribed to Pierre. Barry’s utterance of (8) may be enriched with the property being called ‘London’ by Pierre, so that Pierre is ascribed belief in the claim, singular with respect to London and Gate 43, that the flight to the first, which Pierre calls ‘London’, is at the second.

4 Is there reason to prefer one of these options to the others? Schiffer’s and Crimmins and Perry’s views seem to suppose that, even holding the context fixed, the verb phrases of sentences of the form X believes that S typically ascribe different properties. This, in turn, makes it difficult to see how arguments like (13) Whatever Eve believes La Villianelle does. Eve believes that Lux is funny. So, La Villianelle believes that Lux is a funny. could be valid. Take Crimmins and Perry as an example. They take a typical use of (13)’s second premise to say something along the lines of (14) Eve believes the proposition that Lux is funny under r1 for some specific mental representation r1; the conclusion in turn ascribes to La Villianelle belief in the proposition under (presumably a quite different) representation r2. Mental representations being peculiar to their possessor, the representation referred to will be different from that referred to in a use of (13)’s conclusion. Crimmins and Perry do not discuss quantified attitude ascriptions, so it is not altogether clear how they would parse the first premise of (13). But given their view, one thinks that it is to be understood so: (15) For every p: if there’s an r such that Eve believes p under r, then there’s an r′ such that La Villianelle believes p under r′. But then uses of the premises of (13) rarely if ever provide guarantee of the truth of a use of its conclusion. This seems like a steep price to pay. The account Richard provides does not have this upshot. On Richard’s account, translation manuals are geared to individuals. If we are discussing Eve and La Villianelle’s beliefs and are focused on what they think about Lux in virtue of interacting with her, our translation manual might (abbreviating in the obvious way) look something like this: When discussing Eve, use ‘Lux’ to represent only representations of Lux linked to Eve’s interactions with her When discussing La Villianelle, use ‘Lux’ to represent only representations of Lux linked to her interactions with her. 136

Context Sensitivity and ‘Believes’

The meaning of ‘believes’, according to Richard, is something which takes a context’s translation manual m as input and has as output a function which (relative to a possible world) pairs an individual x with an articulated proposition p provided there’s an articulated proposition q such that q is in x’s representational system and p represents q, relative to m.8 So if we fix a context, the semantic value of ‘believes’ does not shift across the premises and conclusion in (13); if m is the manual our context provides, (13)’s main premise comes to t. For any articulated proposition p, if p represents an articulated proposition among Eve’s beliefs relative to m, then it represents one of La Villianelle’s relative to m. And this means (13) is valid.9

5 There are, of course, objections to the sorts of accounts discussed in the last sections, more than can be taken up in a brief article.10 Here are two. Objection. It is implausible to think that when one says ‘Ajax thinks that you are going for a walk’ we are referring to, describing, or otherwise invoking Ajax’s mental representations.Typically, we have no idea what the mental representations of others are like; even if we do, our claims about belief are usually indifferent to their nature. For that matter, the notion of mental representation these theories wield is borrowed from (speculative) psychology and cognitive science. Why should we think that everyday talk about what people think is infected with such theoretical baggage? Response. The views discussed in the last section are not committed to the claim that we invariably invoke mental representations in ascribing belief. Crimmins allows for what he calls de re ascriptions in which only the Russellian content of a belief is invoked; Richard thinks that often enough we are unconcerned with what mediates belief so that no particular way of rendering another’s mental idiom is presupposed.The question to ask about these views is whether speakers often intend to communicate information about ways of representing objects and properties, and whether when someone speaks with such an intention, that intention is routinely identified in the course of interpretation. And surely the answer to both questions is ‘yes’. Examples of this are very easy to come by. The example of Pierre in section 2 is one example. Here is another. We see Eleanor enter the room; we see that Sam and Dave see her enter. Sam nods at her; Eleanor leaves. Sam turns to Dave, says something, and Dave looks surprised and then turns to see Eleanor exit. I say to you ‘Ha. Sam and Dave both saw Eleanor come in, but only Sam realized/​knew that it was Eleanor’. Clearly I am trying to convey—​and you will understand me as trying to convey—​that while Sam, using some way w he has of re-​identifying Eleanor, thought that’s w, Dave did not, for any way w* he has of re-​identifying Eleanor think that’s w*. As for the claim that notions from psychology or cognitive science are being imputed to ordinary speakers: hogwash. Young children recognize that there are different ways to get information, even when their language lacks evidential markers:  a child of four is perfectly capable of determining such things as that if Nancy knows that Zoe hid the Barbie, it was because Anneliese told Nancy, not because Nancy saw Zoe hide it. To know such things is to have a theory which presupposes that there are different ways of thinking of/​knowing things about an object. It may be that one or another notion of mental representation to be found in cognitive science is descended from the ‘theoretical apparatus’ of everyday thought about the mental; it may not. But everyday thought about thought surely does deploy the idea that people have multiple ways of thinking about objects and their properties.

137

Mark Richard

Objection. Steven Schiffer objects that the sorts of views discussed in the last sections absurdly imply that when someone ascribes a belief, there is nothing they can be said to mean. The crux of the objection is that Meaning entails audience-​directed intentions, and one cannot mean something without intending to be understood. Part of meaning that such and such is intending one’s audience to recognize that that is what one meant, and—​a corollary—​part of referring to a thing is intending one’s audience to recognize that reference. Consequently, [a proposal of the sort discussed above] requires there to be some type of mode of presentation ϕ such that [what someone ascribing belief to Harold] means is that Harold believes [the relevant proposition] … under a mode of presentation of type ϕ, and … intends her audience to recognize … that she means [this]. Schiffer 1992, 515–​516 But, says Schiffer, there will be many (‘infinitely many’, says Schiffer) types of modes of presentation under which one believes a proposition, many of which will be equally salient in a context. Pierre in the example of Gate 43 believes the proposition that the flight to London is at Gate 43 under a mode of presentation (that ‘directly associated’ with his use of ‘the flight to London is at Gate 43’) of each of the following types (L) being expressed by ‘the flight to London leaves from Gate 43’ being expressed by the sentence Pierre used in response to Barry’s question being expressed by a sentence in English being associated with Pierre’s visual impression of Gate 43 being one that Pierre had in mind just a minute ago etc. These are all, Schiffer complains, equally salient in the communicative context. This makes it extremely implausible that of all the equally salient types of ways that [Pierre has of believing that the flight is at Gate 43, Barry] should mean—​and intend to be taken to mean—​a proposition about one definite one of them. Schiffer 1992, 516 As Schiffer himself notes, this objection is reminiscent of objections to accounts of incomplete definite descriptions which take a use of the F to be in one way or another elliptical for a description of the form the F which is G, with G being supplied by the (presumably tacit) intentions of the user.11 Suppose we are looking at a dog and I say ‘the dog looks hungry’. I know that it is common ground between us that the dog is: something I am watching; something you are watching; something that we are watching; a spotted dog in front of us; a hungry looking dog in front of us; Dani’s most beloved pet; etc. I do not intend any particular completion of my description to be thought by you to be one I intend for you to grasp, and so, it appears, I do not mean any of them. And so, it appears, there is nothing I mean. One thinks that the problem here is not a problem with the proposal about incomplete descriptions, or the idea that attitude ascriptions invoke representations or ways of thinking; rather, the problem is with an overly simple picture of what happens when one speaks assertively.12 Take the example of my utterance about the dog. It is common ground, when I say ‘the dog is hungry’, that I mean to be talking about the dog in front of me (and the dog in front of you, and …). So it is common ground that when I utter the sentence I intend that the common ground of the conversation be extended to include the claim that the dog in front of me is hungry; likewise for the claim that the dog in front of you is hungry, etc. When it is common ground that in using ‘the dog’, 138

Context Sensitivity and ‘Believes’

the dog that’s G picks out exactly one dog, a dog of which I intend to speak with my use of ‘the dog’, then I will, if I utter ‘the dog is hungry’ intend to convey that the claim dog that’s G is hungry. What I mean when I utter ‘the dog is hungry’ is not a single claim but the collection of all such claims: I intend for the common ground between us to be increased by adding all of these to it; you understand me if you understand me to be suggesting such a change in common ground. My meaning is large; I assert multitudes. Something similar can be said about Schiffer’s objection to contextual accounts of attitude ascription. When Barry speaks, he will think that there is some common ground about the way of thinking of London that he invokes when he uses ‘London’—​for example, that Barry is reporting on a belief Pierre expressed by response to Barry’s query. It will be common ground between Barry and his wife that this way of thinking is of various types, some of which are listed in (L) above. For each such type W, Barry’s utterance has the effect of adding to the conversation’s common ground that Pierre believes, making use of a way of thinking of type W, that the flight to London is at Gate 43. This is manifest to Barry and his wife, and manifest to both that this is manifest, etc. Producing such an increase in common ground is part of what Barry intends to do in speaking, and Barry’s wife understands him to have this intention. So Barry means, for each such type W, that Pierre believes, using a way of thinking of London of type W, that the flight to London is at Gate 43.13

6 Russell and Frege thought that belief and desire are relations to things we refer to with complements clauses. On their views, the sentence (PR)  Paul Ryan believes that Social Security is a Ponzi scheme reports that Paul is related to something named by ‘that Social Security is a Ponzi scheme’.They took such things to be constituted by (things which are or could be) natural language word meanings. Both thought that such meanings, and thus the objects of belief, are mind independent. In addition, Frege thought, and Russell seemed at least sympathetic to the idea, that rationality and mental motivation can be theorized in terms of the objects of belief: it is irrational, for instance, to believe a thought and its denial; that one believes the thought that A only if B and wishes that B can explain why one is disposed to attempt to make it the case that A. The views discussed in the last sections agree that (the sorts of things that are or could be the) semantic properties of public language expressions at least help to individuate attitudes; they accept contemporary views about such properties. For the most part, they also agree that whatever story we tell about the attitudes should cohere with the ideas, that those attitudes are explanatory of behavior, and that there is an intimate relation between such things as consistency of belief objects and rationality. But they think that this requires that states of belief—​the states ascribed by uses of predicates of the form believes that S—​are by and large states individuated in terms of ways of representing things which are idiosyncratic to the person under discussion. When one says that Pierre believes that London is lousy, one (typically) refers to or otherwise invokes Pierre’s mental particulars; it is this invocation which makes the ascription something we can use in our estimations of Pierre’s rationality or explanations of his behavior. On such views, the fact—​belief ascriptions report facts that vary with the (perspective of the ascriber on the) mentality of the ascribee—​does not only explain why the predicate is contextually sensitive; it reflects the fact that belief states are in some sense partially constituted by the idiosyncratic way the believer pictures the world. One might worry that these views toss out the baby with the dirty diapers. Surely, it will be said, there is an important sense in which speakers of English all mean the same thing by their words, and so share the concepts those words express. There is a single concept cat, expressed by pretty much 139

Mark Richard

every English speaker’s use of ‘cat’, be that speaker from Trenton, Tasmania, or Tunbridge Wells; likewise for the word ‘furry’. Thoughts are constructions from concepts with which we think, and so there is a thought, the thought that cats are furry, composed of these concepts. Speakers of English who express a thought by tokening ‘cats are furry’ are expressing that thought. And since to have a thought is to have its constituent concepts “in mind”, thought thus understood measures rationality and is explanatory of belief. The idea that we share meanings and concepts in some sense of ‘meaning’ and ‘concept’ is indeed plausible. But can we understand belief, given its role in describing rationality and motivation, as a relation to something built from shared concepts? Mark Sainsbury and Michael Tye (MnM, for short) say so.They say thinking, asserting, and other propositional attitudes are relations to thoughts, thoughts now being taken to be structures whose parts are concepts. Concepts are abstract objects (that is, they are not spatially located and “have no direct causal powers”) that have (unique) historical origins and (typically) limited temporal extent. Some concepts come into existence via intentional introduction; others arise via non-​intentional cognitive activity (as when a child forms a prototype based on the similarities among dogs she sees on walks). Concepts are associated with mental structures of thinkers, and these structures contribute to something like reproduction. When C is a concept. There is a concept-​originating event, an act of [a]‌subject S1, in which concept C is originated. This generates a C-​reproducing mechanism in S1 which can create copies or tokens of C. Being produced by this mechanism is what makes S1’s later use of a concept a use of C. If S1 transmits this concept to other thinkers, say S2, something similar occurs: a C-​reproducing mechanism is formed in S2. The mechanism counts as a C-​reproducing one in virtue of having been formed under the impact of S1’s uses of C. If S2 subsequently uses a concept derived from her C-​mechanism, that use is a use of C. Sainsbury and Tye 2012, 60 MnM take belief and the like to be “psychologically real” relations to thoughts cum structures constructed from concepts.They take attitude ascription to involve a complicated, contextually shifty process of conveying something about the object of someone’s thought: on a first pass, MnM say that (PR) comes to something like Paul Ryan bears the (“psychologically real”) relation BEL to a conceptual structure that stands in R to the conceptual structure that Social Security is a Ponzi scheme, where the relation R varies from situation to situation, but is something like the relation that conceptual structure p bears to conceptual structure q when, for the purposes at hand, p has important features in common with q. Like the views discussed above, this view posits a contextual shiftiness in the meaning of ‘believe’. Like those views, it aspires to make sense of such facts as the fact that beliefs with the same referential content may be distinct (since, for example, the concept expressed by ‘Hesperus’ will be different from the concept expressed by ‘Phosphorus’). But it is, I will argue, a bad picture of thought and its ascription.14 MnM’s concepts and conceptual structures are individuated externally.Whether tokens are tokens of the same concept is determined by things typically opaque to the thinker, like deference to the usage of others and historical origin. A standard objection to externalist views of the objects of belief is that they rob beliefs of motivational and epistemic properties they obviously have. I think MnM are vulnerable to some such objections. Here is one case they discuss. Paul and the French Nanny. Paul is raised by a French nanny who speaks to him in English, save that she uses the word ‘chat’ for cats. It is plausible to think that the French ‘chat’ and the English ‘cat’ express the same concept, so it is plausible to think that Paul acquires the 140

Context Sensitivity and ‘Believes’

concept cat from his Nanny. His parents, on their occasional visits, speak to him of ‘cats’. Paul acquires ‘cat’; presumably he associates it with the concept his parents do, the concept cat. It never occurs to Paul to ask himself whether chats are cats. One day he suddenly thinks to himself ‘Merde! Chats are cats! Incroyable!’ He seems to have learned something that he did not know, namely (as he puts it to himself) that chats are cats. MnM are willing to grant that ‘cat’ and ‘chat’ express the same concept. They deny that if this is so, then at story’s end Paul learns that cats are chats: for them the conceptual structure that cats are chats is the conceptual structure that cats are cats; Paul knows this as soon as he acquires the concept cat.15 They suggest that Paul learns something “meta-​conceptual”, that ‘cat’ expresses the concept chat (Sainsbury and Tye 2012, 129). This won’t do if the concept cat is the concept chat, for in that case the conceptual structure, that ‘cat’ expresses the concept chat, is the conceptual structure, that ‘cat’ expresses the concept cat; Paul already knew that. I take it this was a slip of the pen, and what MnM meant to say was something along the lines of: what Paul learns is (what he expresses when he utters) (P1) ‘chat’ and ‘cat’ express the same concept. But this response does not work for cases that surely should receive the same diagnosis. Consider, for example, the case of Pauline and the Scottish Nanny. The details of this case are like Paul’s, save the nanny is from Glasgow, speaks only English and pronounces ‘cat’ caht; Pauline’s parents are from New York and pronounce the word keat. As in Paul’s case, it is some time before Pauline cottons onto the fact that (as she thinks to herself) ‘keats are cahts’. In this case, Pauline acquires the same concept twice over, once from her nanny, once from her parents. And she acquires the same word twice over. Here, a meta-​linguistic or ‘meta-​conceptual’ explanation will not turn the trick if we individuate concepts as do MnM. On their view, what is thought by Pauline when she orally tokens (P2) ‘keat’ and ‘caht’ express the same concept expresses something Pauline knows as soon as she acquires the word ‘caht’ from her nanny. Here’s why. When Pauline’s nanny uses the word she pronounces caht—​that is, the English word ‘cat’—​Pauline gains, in virtue of her deferential use of ‘caht’, two concepts: the concept expressed by ‘caht’ (the concept cat), and a concept of the word used, the English word ‘cat’. The word concept Pauline acquires is presumably the concept, common to the nanny and to us, which we all express with the phrase “the word ‘cat’ ”. So what Pauline thinks to herself when she mentally tokens (P3) ‘caht’ expresses the concept caht is the very same conceptual structure that we think when we token (P4) ‘cat’ expresses the concept cat. Analogously, when Pauline hears her parents first use ‘keat’, her deferential use of their word makes it the case that she “reacquires” the same two concepts.When she hears her parents speak, she (speaking metaphorically) opens up a new concept file, labeled keat, that is deferential to her parents use of 141

Mark Richard

‘keat’. She also opens a new entry in her linguistic lexicon for the word—​‘keat’—​she acquires from her parents. The upshot is that what she thinks when she mentally tokens (P5) ‘keat’ expresses the concept keat is the conceptual structure her parents think when they token this. But the conceptual structure her parents token is, of course, just the one we think when we token (P4). (Remember: MnM’s concepts and conceptual structures are individuated externally.) It should now be clear that when Pauline tokens (P2) she expresses the very conceptual structure she does when she tokens (P6) ‘caht’ and ‘caht’ express the same concept, since she has known from the get-​go that ‘cat’ expresses the same concept as ‘cat’. I think it is obvious that Pauline has made a “substantive discovery” at the end of the case. But it doesn’t seem MnM can agree. I  am willing to describe the discovery as a matter of Pauline’s coming to believe a “conceptual structure” she did not believe before. But the conceptual structure is idiosyncratic—​it involves Pauline’s distinct concepts, caht and keat; she came to know, as she would put it, that cahts are keats. Of course, it is only with a certain amount of stage setting that we can describe what Pauline came to learn, for we do not begin with words for her idiosyncratic concepts. Some will object that we all share a concept we all express with our common word ‘cat’, namely the concept cat. But if so, then since I have described the case of Pauline as one in which she acquires that concept (twice), I have ruled out the idea that Pauline failed to know all along what she says, when she thinks (P7) Keats are cahts. For what she thinks is simply that cats are cats. I respond as follows. Consider four utterances about Pauline’s cat, Cipollotto: (C)

1. Nanny: 2. Pauline: 3. Dad: 4. Pauline:

Cipollotto is a caht. Cipollotto is a caht. Cipollotto is a keat. Cipollotto is a keat.

I think we all have fairly strong pre-​theoretic inclinations to say each of the following: 1

In (C1) and (C2), the nanny and Pauline use the same words and express the same concepts /​say the same thing. 2 In (C3) and (C4), Dad and Pauline use the same words and express the same concepts /​say the same thing. 3 In (C1) and (C3), the nanny and Dad use the same words and express the same concepts /​say the same thing. 4 In (C2) and (C4), Pauline does not use the same words and does not express the same concepts /​say the same thing. Now it appears that these claims are jointly inconsistent, and so something’s gotta give. But what? I’m inclined to say that questions along the lines of (Q)

Are these tokens of the same word? Are those people expressing the same concept with their words? Did those people say the same thing with their sentences? 142

Context Sensitivity and ‘Believes’

don’t have fixed, context and interest invariant answers. We sometimes individuate words in terms of mutually reinforcing chains of deference and overlapping usage. We sometimes individuate them by internal functional roles, roles grounded in part in entries in a “mental dictionary” used in parsing, interpretation, and production. Something similar is true of concepts. A concept can be individuated in more or less external terms—​in terms of public, shared categories or what those represent—​or in terms of (these and) internal psychological roles. Ditto for thoughts. For everyday purposes, these two modes of individuation often overlap. It takes a case like Pauline’s to make them come apart. The inclination to endorse all of (C1) through (C4) results from our applying different schemes of concept and thought individuation to different cases. When comparing thoughts across persons, we often individuate them in terms of publicly shareable properties, and thus endorse (C1) through (C3). In the intra-​personal case, there is good reason to employ a different scheme of individuation, one that reflects, as we might put it, internal identity and difference. Thus (C4). Rightly understood, (C1) though (C4) are not only jointly consistent—​they are spot-​on correct. An account of thought’s objects and their ascription should be flexible enough to acknowledge the validity of both ways of individuating thoughts. Not only does it need to allow that the token vehicles of thought can be individuated more finely than the vehicles provided by public language—​it needs to allow that the objects of attitudes expressed by such vehicles may be individuated as finely as the vehicles themselves. Pauline thinks one thing when she thinks ‘keats meow’, another when she thinks ‘cahts meow’. An account of these matters also needs to make room for the fact that we often identify beliefs and the like when they are realized by shared public language words or concepts; for that matter, that we sometimes identify attitudes when they are realized by sentences that express the same Russellian or possible worlds proposition. MnM’s account of thought, I fear, does not have the necessary flexibility. I think there is a place in theory for something a bit like the things MnM call concepts. But that place is not in a general account of thought and its ascription; rather it is in an account of linguistic understanding. A concept qua something that comes into being at a particular time, which is transmitted in something that resembles in some ways biological reproduction, which speakers in some sense ‘grasp’ and express with their words—​these are (some of) the defining properties of MnM’s concepts; I think an adequate account of linguistic meaning sees meanings as things with such properties.16 But meanings in this sense are simply not finely enough grained to ground an account of belief and other propositional attitudes which does justice to the idea that such attitudes are of use in thinking about rationality and motivation.

Notes 1 If semantic value is compositional, then phrases containing context-​sensitive expressions are also context sensitive. When speaking of Kaplanesque context sensitivity, I generally restrict ‘context sensitive’ to phrases none of whose proper constituents are context sensitive.   Context sensitivity is distinct from ambiguity and polysemy, though it is not easy to give a theory-​neutral account of the differences. In the case of simple phrases the difference is presumably a matter of ambiguous and polysemous phrases being associated with multiple ‘lexical entries’ stored in the speaker’s mind, with the meaning of a use determined by speaker intention. Context sensitivity, on the other hand, involves something like there being a single rule or routine which determines a semantic value for the phrase’s use. 2 The classic statement of Frege’s view is Frege (1952). Russell’s early views about propositions are found in Russell (1903); his midcourse correction appears in Russell (1911). A more extensive discussion of their and related views can be found in Richard (1990, 2012). 3 The story of Pierre is first told in Kripke (1979). 4 This needs to be said more carefully, since the indexical components may be assigned either particular modes of presentation or types of such. To make reading easy, I suppress this complication in expositing a view. 5 Does the user of (8) refer to the representations? Perry and Crimmins say that the representations are “unarticulated constituents” of what (8) says. They are, that is, constituents of the claim the use of (8) makes, but there is no element of (8) which is responsible for /​has as its semantic value /​articulates these representations. The archetypes for unarticulated constituency are uses of sentences about the weather: a use of ‘it’s raining’ says of a particular location (often but not invariably the speaker’s) that it is raining there; it is, according to

143

Mark Richard Crimmins and Perry, implausible to think that some piece of the sentence’s syntax is responsible for this. As I understand Crimmins and Perry, they wish to remain neutral as to whether, when c is an unarticulated constituent of what is expressed by x’s use of sentence S, this is a result of x’s referring to c. 6 There is a discussion of how the picture being sketched here applies to cases in which a believer’s belief are not grounded in linguistic representations in the introduction to Richard (2012). 7 More details are provided in the next section. 8 For all the details, see Richard (1990) and (2012). 9 There are other ways in which the three views mentioned in section 3 differ which are relevant to choosing among them. For instance, they vary as to whether it is the rule that we refer to or otherwise invoke particular modes of presentation, representations, or ways of describing in ascribing belief. 10 The first objection below is a mash up of objections in Soames (2002); the second, obviously, is in Schiffer (1992). Objections not taken up here are made in Crimmins (1992), Saul (1999), Sider (1995), and Soames (2002). Responses to those objections can be found in Richard (2006) and Richard (2012). 11 Schiffer credits this observation to Jim Higginbotham. It is fairly obvious that similar objections can be raised to proposals that quantified utterances involve tacit domain restrictions. 12 A diagnosis like (but, perhaps, not quite identical to) the following is proposed by Bowker (2017). Bowker’s paper concentrates on incomplete definite descriptions, though it is clear that he thinks what he says can be used to respond to Schiffer’s discussion of views on which ‘believes’ is contextually sensitive. Bowker is not to be held accountable for my extension of his discussion to the pragmatics of belief ascription. 13 There is discussion of and responses to other objections to the sorts of accounts presented in sections 4 and 5 in Richard (2012, Chapters 6, 7 and 13). 14 The balance of this section is adapted from a contribution to an Author Meets Critics session on Sainsbury and Tye (2012). I’m grateful to David Braun and Mark Sainsbury for spirited discussion of those remarks. 15 The argument depends on (I hope uncontentious) claims that learning is a relation to the sort of thing invoked by a complement clause, and that, as he already knew that cats are cats, Paul could not have learned (the identical thought) that cats are chats. 16 I have spelled this out in Richard (2019). The picture of meaning in that book agrees with MnM that meanings/​concepts are historical entities. But it identifies the meaning of a (simple) expression in a group with (roughly put) the evolving collection of assumptions such that it is common ground in the group, that users of the expression generally make those assumptions when using the expression and expect their audience to recognize them as being made.

References Bowker, M. 2017. Saying a Bundle: Meaning, Intention, and Underdetermination. Synthese [online]. Crimmins, M. 1992. Talk About Belief. MIT Press. Crimmins, M. and Perry, J. 1989. The Prince and the Phone Booth. The Journal of Philosophy 86, 685–​711. Frege, G. 1952. On Sense and Reference. In Geach, P. and Black, M., eds., Translations from the Philosophical Writings of Gottlob Frege. Blackwell. Kaplan, D. 1989. Demonstratives. In Almog, J., Perry, J., and Wettstein, H., eds., Themes from Kaplan. Oxford University Press. Kripke, S. 1979. A Puzzle about Belief. In Margalit, A., ed., Meaning and Use. Reidel. Richard, M. 1990. Propositional Attitudes. Cambridge University Press. Richard, M. 2006. Propositional Attitude Ascription. In Devitt, M. and Hanley, R. The Blackwell Guide to the Philosophy of Language. Blackwell. Richard, M. 2012. Context and the Attitudes. Oxford University Press. Richard, M. 2019. Meanings as Species. Oxford University Press. Russell, B. 1903. Principles of Mathematics. Cambridge University Press. Russell, B. 1911. Knowledge by Acquaintance and Knowledge by Description. Proceedings of the Aristotelian Society 11, 108–​128. Sainsbury, M. and Tye, M. 2012. Seven Puzzles of Thought. Oxford University Press. Saul, J. 1999.The Road to Hell: Intentions and Propositional Attitude Ascriptions. Mind and Language 14, 356–​375. Schiffer, S. 1977. Naming and Knowing. Midwest Studies in Philosophy 2, 28–​41. Schiffer, S. 1992. Belief Ascription. The Journal of Philosophy 89, 499–​521. Sider, T. 1995. Three Problems for Richard’s Theory of Belief Ascription. Canadian Journal of Philosophy 25, 487–​514. Soames, S. 2002. Beyond Rigidity. Oxford University Press.

144

11 A RETURN TO SIMPLE SENTENCES David Pitt

In 1997, Jennifer Saul (1997) introduced a fascinating new puzzle into the philosophy of language literature by identifying cases of substitution failure in apparently extensional contexts. Her first example was the pair of sentences (1) and (2): (1) Clark Kent went into the phone booth, and Superman came out. (2) Clark Kent went into the phone booth, and Clark Kent came out. We would expect that substituting ‘Clark Kent’ for the co-​referential ‘Superman’ would always result in a sentence with the same truth value. Intuitively, however, it does not, since it is possible for (1)  to be true while (2)  is false. Yet these sentences do not contain any of the familiar opacity-​ inducing propositional-​attitude, quotational, or modal constructions that are traditionally appealed to to explain such substitution failures. Why, then, does substitution fail? That is the puzzle. A flurry of responses followed, and Saul’s examples became a lively topic of debate. Ten years later she published a book, Simple Sentences, Substitution, and Intuitions (Saul 2007) in which she laid out her objections to all of the proposed solutions to her puzzle, as well as her own take on how best to solve it. My own contribution to this discussion (Pitt 2001) was an attempt to solve the problem at the semantic level. I argued that failure of substitutivity in sentences like (1) and (2) is due to the fact that the relevant terms, ‘Superman’ and ‘Clark Kent’, are in fact not co-​referential. What these names refer to is what I called “alter egos” of the Kryptonian Kal El. ‘Kal El’ refers to an individual, and ‘Superman’ and ‘Clark Kent’ refer to distinct alter egos of that individual, where an alter ego is an individual inhabiting a persona.1 I characterized a persona as consisting of certain ways of behaving or dressing that are deliberately contrived or adopted and kept separate from an individual’s ordinary self-​presentation. Superman is Kal-​El-​inhabiting-​the-​Superman-​persona, and Clark Kent is Kal-​El-​ inhabiting-​the-​Clark-​Kent-​persona. Since Kal El does not always inhabit these personas, he is not identical to either alter ego, and since he does not inhabit them simultaneously, they are not identical to each other. I further suggested that one may take alter egos—​the referents of alter-​ego names like ‘Superman’ and ‘Clark Kent’—​to be sums of four-​dimensional time-​slices of the underlying individual. Superman is the sum of time-​slices of Kal El during which he is inhabiting the Superman persona, and Clark Kent is the sum of time-​slices of Kal El during which he is inhabiting the Clark

145

David Pitt

Kent persona. Since these sums are distinct, the names are not co-​referential, and so are not substitutable salva veritate in extensional contexts. In this paper I would like to respond to several objections that have been made to my account. Saul (2007) claims that my account cannot handle a kind of case introduced by Joseph Moore (1999). Imagine that Kal El is sitting at a desk at the Daily Planet dressed as Clark Kent but talking to Lois Lane on the phone as Superman. Imagine further that Lois is simultaneously looking through a window at him at his desk. Saul argues that (3), which is, intuitively, a correct way to describe the situation (3) While talking on the phone to Superman, Lois looked through the window at Clark Kent cannot be true on the temporal slice account, since distinct temporal slices cannot be co-​present: Lois “cannot be talking on the phone to one temporal part and at the same time looking at a different one” (Saul 2007, 31). “Pitt needs to be able to answer the question”, she continues (2007, 32), “of which persona Kal El is occupying when he is making eye contact with Lois as Clark while speaking to her on the phone as Superman” (I guess we should also suppose that she is a not a good lip-​reader). Here is how I answer the question. In this situation Kal El is partially inhabiting both the Superman and the Clark Kent personas. That is, he is speaking as Superman while being dressed as Clark Kent. And while he is doing this he is neither Clark Kent nor Superman, but, rather, Kal El, exhibiting aspects of both personas (just as while he is in the shower he is neither Clark Kent nor Superman, and is exhibiting aspects of neither persona).This is no more problematic than someone dressed as Hamlet on stage talking and moving around like Ophelia. In this moment, the actor would be playing neither Hamlet nor Ophelia, but partially affecting characteristics of both. The Superman and Clark Kent personas are Kal El’s creations, and he can do what he wants with them. If he is trying to maintain the deception that his two alter egos are in fact different individuals, he will take care not to be seen or recognized as simultaneously exhibiting aspects of both. But he can freely mix and match such aspects as he chooses (though, again, in doing so he is being neither alter ego). And he could make radical changes to them. For example, if he got tired of walking around as Clark Kent, he could arrange to have Superman confer the power of flight upon Clark Kent (telling some story about Kryptonian powers). And he could also, as Superman, “disguise” himself as Clark Kent, and be known to be doing as much by everyone, in order to serve some purpose (e.g., convincing some criminal that Clark Kent had been shot dead). There are many possibilities. In the story as it is, however, Kal El wants to keep the alter egos separate, at least so far as the citizens of Metropolis are concerned. Saul also objects that my account gives “the very counterintuitive result that [(4)] and [(5)] below are false”. (4) Superman is Clark Kent. (5) Superman is Kal El. She claims that I try to explain why people wrongly judge these sentences to be true as due to the fact that “they understand [them] as meaning the same as” (6) and (7): (6) The person whose alter ego is Superman is the person whose alter ego is Clark Kent. (7) The person whose alter ego is Superman is Kal El. and that this is not likely to be true, since people who think that (4) and (5) are true are not aware of the facts about Kal El, Superman and Clark Kent (Saul 2007, 32–​33.) On the contrary, however, anyone who is unaware of these facts would take (4) and (5) to be false, not true. The average citizen of Metropolis has no idea of the connections among Kal El, Superman, and Clark Kent. Anyone who 146

A Return to Simple Sentences

was inclined to accept that (4) and (5) are true has become aware of a hitherto unknown, because deliberately concealed, intimate relationship between individuals they formerly thought to be distinct. Saul notes that “most of us—​myself included, until I read Pitt’s paper—​have forgotten (if we ever knew) that there is a third individual [, (Kal El)] … who adopted two alter egos” (2007, 33). I think Kal El is a red herring here. Anyone who had forgotten or never knew about Kal El would likely be stumped by (7). What needs to be explained is the anti-​substitution intuition—​that is, why someone who is “enlightened” about the Superman/​Clark Kent situation would accept that (1) and (2) could have different truth values. The “unenlightened” do not accept the substitution of names in (1) and (2) because they do not believe that Superman is Clark Kent—​that is, they do not believe that ‘Superman’ and ‘Clark Kent’ are co-​referential—​that they are names of the same individual. The simplest explanation for why anyone who is enlightened would have the same intuition is that they too do not believe that the names are co-​referential—​in spite of their assent to ‘Superman is Clark Kent’. It seems to me that what Saul’s case shows is that the enlightened did not really believe what they thought they believed; for if they had, they would unproblematically accept the substitution—​as they would in a case like (8) and (9): (8) Bon Iver went into the phone booth and Justin Vernon came out. (9) Bon Iver went into the phone booth and Bon Iver came out. The way “enlightenment” is often characterized in the literature on the puzzle makes this explanation unavailable: to be enlightened is to know that Superman is Clark Kent (i.e., that ‘Superman’ and ‘Clark Kent’ refer to the same individual). But I think this is not the right way to characterize enlightenment—​i.e., what one knows when one is in on the Clark Kent/​Superman story. What one comes to know upon being enlightened is that the individual who flies around Metropolis in the cape is the same individual who sits behind a desk at the Daily Planet in black horn-​r immed glasses, and that part of the explanation for the different get-​ups is his intention that this fact should not be obvious. If, once enlightened, an individual still uses the names differentially—​as evidenced by, for example, the fact that certain substitutions are still rejected—​there is reason to think that they have not come to believe that ‘Superman’ and ‘Clark Kent’ are co-​referential, and hence intersubstitutable in all extensional contexts. What I suggested in Pitt (2001) is that (4) and (5) are short-​hand ways of saying what (6) and (7) say. Such short-​hand uses of literally false identity sentences are in fact quite common. As I noted in my original paper, advertisements for movies, plays, etc., for example, often use sentences like (10) and (11) (10) Taron Egerton is Elton John. (11) Ruth Negga is Hamlet. though everybody knows that they are, literally, false. What they mean is that Taron Egerton is playing Elton John (representing him in a movie about his life; the two men cannot be identical), and Ruth Negga is playing Hamlet (inhabiting the role on stage; no person could be identical to a fictional character). Someone who did not know who Taron Egerton, Elton John, Ruth Negga, or Hamlet is might take them to be literally true, just as someone who did not know the whole Kal El story might take (4) and (5) to be literally true. Someone who did know and who utters (10) or (11) might be challenged about whether what they say could be literally true. And it might take them some time to make explicit what they really meant. I see Saul’s cases as presenting a challenge to anyone who would utter (4) or (5) thinking it could be literally true, given their rejection of the inference from (1) to (2). Anyone who rejects the inference must, whether or not they can easily make it explicit, be thinking something else. Solutions appealing to guises, aspects or senses (e.g., Forbes 1997, 1999; Moore 1999; Predelli 1999) take the same approach. 147

David Pitt

A third objection is that my differential treatment of Superman/​Clark Kent/​Kal El cases and Batman/​Bruce Wayne cases leads to problems (Saul 2007, 33–​34). Saul argues that if it turns out that Kal El as Clark Kent got into fewer fights than he did as Superman, then on my view (12) comes out false, which she finds unproblematic: (12) Clark Kent has been in more fights than Superman. On the other hand, since inhabitants do whatever their alter egos do, if Bruce Wayne has only been in one fight as himself, but has been in ten fights as Batman, then on my view (13) comes out true: (13) Bruce Wayne has been in more fights than Batman. But Saul finds this counterintuitive. After all, Batman got into ten fights, while Bruce Wayne, out of persona, has only been in one. So, it would seem, Bruce Wayne has been in fewer fights than Batman. Saul argues, further, that it is counterintuitive to treat the cases differently, since they are, intuitively, alike. However, it seems to me that the cases are not at all alike, since the first involves a comparison of two distinct alter egos, while the second involves a comparison of an alter ego and its inhabitant. More to the point would be a comparison between (13) and (14): (14) Kal El has been in more fights than Superman. I suspect that Saul would maintain that (14) is as counterintuitive as (13); but it seems to me that it is not. If we know the story, then we know that everything Superman does Kal El does, though not vice versa. So, if Superman has been in ten fights, Kal El was also in those fights. If Kal El got into one on a visit home to Krypton, as himself, it seems to me unproblematic to say that Kal El has gotten into more fights than Superman. He got into one as himself, and ten as Superman. That is eleven fights. Superman only got into ten. Eleven is greater than ten. Kal El got into more fights than the ones he got into as Superman. Likewise in the Batman case. While Bruce Wayne only got into one fight as himself, he got into ten as Batman. So, again, that is eleven fights: Bruce Wayne got into more fights than the ones he got into as Batman. I suspect that Saul’s qualms arise from treating the two cases in the same way—​that is, from thinking that Bruce Wayne is something separate from Batman in the way that Clark Kent is something separate from Superman. But Bruce Wayne is not separate from Batman in this way. Batman is a proper part of him; Superman is not a proper part of Clark Kent. Suppose someone slaps my right cheek ten times and slaps me in the side of the head once. Since my right cheek is part of my head, my head has been slapped more times (eleven) than my right cheek (ten); even if it sounds counterintuitive to say so. Finally, Saul objects that my account is ill-​equipped to deal with substitution failures involving names that are clearly not names for agents or their alter egos. St Petersburg has not created an alter ego, Leningrad; nor has Leningrad created St Petersburg as [an] alter ego. For Pitt, then, [(15) and (16)] must take the same truth value. Saul 2007, 34 (15) I visited St Petersburg once, but I never made it to Leningrad. (16) I visited St Petersburg once, but I never made it to St Petersburg.

148

A Return to Simple Sentences

But the essential feature of my account was to explain failure of substitutivity in the standard way—​ as owing to non-​co-​referentiality of substituted terms. The fact that the names in Saul’s original examples refer to alter egos on my account is incidental. I introduced the notion of an alter ego in order to substantiate the claim of non-​co-​referentiality in the examples Saul originally provided. But this does not commit me to taking the referent of either ‘St Petersburg’ or ‘Leningrad’ to be a person, or an alter ego. Nor does it follow that (15) and (16) must have the same truth value. I took the referents of ‘Batman’ and ‘Bruce Wayne’ to be four-​dimensional objects, where the former is a proper part of the latter. Precisely the same thing can be said of ‘Leningrad’ and ‘St Petersburg’. If we are to make sense of the intuition that (15) can be truly uttered, then we must suppose that the speaker is enlightened about the fact that ‘St Petersburg’ and ‘Leningrad’ are names of the same city, and means to be saying something other than what would be said in an utterance of (16). The most obvious interpretation is that the speaker meant to say that she had never visited St Petersburg when it was called “Leningrad”, and did so in a short-​hand sort of way, instead of uttering (17): (17) I visited St Petersburg once, but I never visited it when it was called ‘Leningrad’. This can easily be accommodated by taking the referent of ‘St Petersburg’ to be a four-​dimensional object, and the referent of ‘St Petersburg when it was called “Leningrad” ’ to be a spatiotemporal proper part of it. And the same approach can supply an explanation of how an utterance of (18) (Saul 2007, 19) could be assessed as true while (19) is assessed as false: (18) Shostakovich always signalled his connection to the classical traditions of St Petersburg, even if he was forced to live in Leningrad. (19) Shostakovich always signalled his connection to the classical traditions of Leningrad, even if he was forced to live in St Petersburg. The point of (18) is, of course, that the city Shostakovich was forced to live in had very different characteristics when it was called ‘Leningrad’ than when it was called ‘St Petersburg’. Since the relevant differences are not implicated by (19), (19) is, if not outright false, at least puzzling. But, again, the relevant failure of substitutivity salva whatever can be explained by taking the referents of ‘St Petersburg’ and ‘Leningrad’ to be distinct four-​dimensional objects. So both of these examples can be accommodated by the approach I took in my 2001 paper. Saul’s own solution to her puzzle relies on an analysis of intuitions and the psychology behind our judgments in her cases. She suggests that information associated with the name ‘Superman’ is stored separately from information associated with the name ‘Clark Kent’. She says that “despite our knowledge that ‘Superman is Clark Kent’ is true, we have a well-​motivated and deeply ingrained habit of not always integrating Superman and Clark information” (Saul 2007, 138–​139). “Sometimes, … enlightened speakers well aware of particular identities will fail to make all the inferences that they could from the relevant identity claims” (Saul 2007, 145). And this can occur even if the enlightened are invited to reflect upon the identity sentence as (or just before?) they evaluate the inference. What needs to be explained is why anti-​substitution intuitions persist “even as we are led through an inference that should (on my view) demonstrate that they are mistaken” (Saul 2007, 139). Thus, when we judge that the inference from (1) to (2) is invalid, it is because we are (for the moment?) under the influence of the separation of information about Superman and Clark Kent, forgetting, or ignoring the identity. This strikes me as very implausible. In order for the enlightened to find the inference from (1) to (2) invalid, our belief that Superman is Clark Kent must somehow become unavailable. When we focus on (1) and (2), we are momentarily blinded to a belief we rehearsed moments before. But it is

149

David Pitt

hard to believe that we are so easily distractible. And can one not hold all three propositions before one’s mind at once? Can one not think “This is weird. Though it’s true that Superman is Clark Kent, it’s also true that you can’t infer from the fact that Superman leaps tall buildings that Clark Kent leaps tall buildings.What’s going on here?”? Appreciating the puzzle requires comparing the intuitions and seeing that they are in tension. If we forgot the identity intuition every time we had the failure of substitutivity intuition, why would we find Saul’s cases so fascinating to begin with? Another critic of my solution to Saul’s puzzle is Stefano Predelli. Predelli (2004, 109) accuses me of egregious selectivity with respect to which intuitions a solution to Saul’s puzzle should save. In particular, he finds it “remarkable” that I end up denying that ‘Superman is Clark Kent’ is true. But it is inevitable that one of the intuitions contrasted by Saul will be false, since they are inconsistent. That is why we have a paradox. It cannot be the case that Superman is identical to Clark Kent, yet Superman has properties (e.g., leaping tall buildings) that Clark Kent does not. So either Superman is not identical to Clark Kent, or they share all the same properties. One of our intuitions must be wrong. Predelli finds it “even more remarkable” that I offer a different explanation for the Batman case than the Superman/​Clark Kent case. He says that while I  provide a “semantic rendering” of the intuitions in the Superman/​Clark Kent cases, which (according to me) involve an individual and two distinct alter egos, I do not in the Batman cases, which, I claim, involve an individual and a single alter ego. So, whereas my account explains the failure of the inference from (1) to (2) as due to the non-​co-​reference of ‘Superman’ and ‘Clark Kent’, it does not explain the failure of inference from (18) to (19) in this way (20) Batman wears a mask. (21) Bruce Wayne wears a mask. since, on my account, this inference does not fail. Hence, “the apparent parallel between [the cases] must be relinquished” (Predelli 2004, 110). This objection shows a misunderstanding of my account, a feature of which is that relations between alter egos and relations between alter egos and their inhabitants are different. To compare Superman and Clark Kent is to compare alter egos.To compare Batman and Bruce Wayne is to compare an alter ego to the individual inhabiting it. The proper example to compare to (20) and (21) is (22) and (23): (22) Superman wears a cape. (23) Kal El wears a cape. (21) is true because Bruce Wayne wears a mask when he is being Batman. (23) is true because Kal El wears a cape when he is being Superman. My account treats them in exactly the same way. Predelli also claims that my account fails because there are conceivable situations in which one might judge (24) to be true (24) Clark Kent leapt over a tall building last night. Whereas on my view this is never the case. But, again, Predelli has not completely understood my account. I did not claim that it is a necessary truth that Clark Kent does not leap tall buildings. It is a contingent fact that Kal El keeps his alter egos separate. He could change his mind at some point. (Recall the example above of Superman publicly conferring the power of flight upon Clark Kent.) Moreover, Kal El is free to fly home while dressed as Clark Kent; but, at least as things stand with respect to his management of his alter egos, in such a case he is not being Clark Kent, since flying is not a component of the Clark Kent persona. 150

A Return to Simple Sentences

Finally, Predelli (2004, n4) argues that my claim (Pitt 2001, 547) that (B) what is true of an alter ego is true of its inhabitant leads to contradiction, as shown by the following: (25) Clark Kent is unlike Superman in that Clark Kent is shy. Therefore, (26) Kal El is unlike Superman in that Kal El is shy. But (26) is equivalent to (27) (27) Superman is unlike Kal El in that Kal El is shy. From which, given (B), it follows that (28) Kal El is unlike Kal El in that Kal El is shy. which is of course a contradiction: Kal El cannot be unlike himself. Predelli is clearly right about this. It cannot be the case that whatever is true of an alter ego is true of its inhabitant, as can be shown by even simpler examples: (29) (30) (31) (32)

Superman is a proper spatiotemporal part of Kal El. Kal El is a proper spatiotemporal part of Kal El. Batman always wears a mask Bruce Wayne always wears a mask.

I concede that it is false that whatever is true of an alter ego is true of its inhabitant. But I do not think this constitutes a fatal problem for my account, since (B) is not an essential component of it. (B) was simply an incautious over-​generalization of what I had in mind at that point in the paper, which was that Kal El does whatever Superman and Clark Kent do, as they are doing it, and Bruce Wayne does whatever Batman does, as he is doing it. And this is because an alter ego of an individual is that individual inhabiting a persona, and an individual does whatever that individual inhabiting a persona does while inhabiting it. That said, I do not think Predelli’s examples support his objection. It is, I maintain, not true that Clark Kent is shy and that Superman is unlike him in this respect. Clark Kent is an alter ego, a person inhabiting a persona; and what it is to inhabit a persona is, among other things, to behave in certain ways (which typically are not ways the inhabitant behaves when out of persona). A persona, as I use the term, is something like a role one plays.When Kal El is being Clark Kent, he is acting. Hence, what is true is that Clark Kent acts shy while Superman does not. And this is independent of whether or not Kal El is shy on his own time. Therefore, (25) is false. What is true is (33): (33) Clark Kent is unlike Superman in that Clark Kent acts shy. It is certainly true that Kal El acts differently from Superman when he is in the Clark Kent persona (when he is being Clark Kent). But it does not follow that Kal El himself, on his days off, is, or acts, the way Clark Kent does. If it is a general truth about Clark Kent that he acts shy, it is because acting shy is part of the Clark Kent persona. But it cannot be inferred from this that it is a general truth about Kal El that he is, or acts, shy. So (25) also does not entail (34), from which a different contradiction could be derived: 151

David Pitt

(34) Kal El is unlike Superman in that he acts shy. What is true is that Kal El acts shy when he is being Clark Kent, but not when he is being Superman: (35) Kal El is unlike Superman in that he acts shy when he is being Clark Kent. And this is true because Superman never is Clark Kent.

Note 1 In Pitt (2001) I used the term ‘primum ego’ to denote the individual whose inhabitation of a persona is an alter ego. I have never liked this term. Moreover, it is, I am told, bad Latin: since ‘ego’ is masculine, it should be ‘primus ego’. But I do not like that either. Here I will refer to the individual whose inhabitation of a persona is an alter ego of that individual as its “inhabitant”.

References Forbes, G. 1997. “How Much Substitutivity”, Analysis 57, 109–​113. Forbes, G. 1999. “Enlightened Semantics for Simple Sentences”, Analysis 59, 86–​91. Moore, J. 1999. “Saving Substitutivity in Simple Sentences”, Analysis 59, 91–​105. Pitt, D. 2001. “Alter Egos and Their Names”, Journal of Philosophy 98, 531–​552. Predelli, S. 1999. “Saul, Salmon, and Superman”, Analysis 59, 113–​116. Predelli, S. 2004. “Superheroes and Their Names”, American Philosophical Quarterly 41, 107–​123. Saul, J. 1997. “Substitution and Simple Sentences”, Analysis 57, 102–​108. Saul, J. 2007. Simple Sentences, Substitution, and Intuitions, Oxford: Oxford University Press.

152

12 ELICITING AND CONVEYING INFORMATION Heimir Geirsson

As a young boy I  often spent time in the shop of our local printer. The printer sometimes tried to supplement his income by printing comics. He would receive reams of unbound comics that contained the pictures without words together with an original master copy that contained a dialogue. He then translated the original into Icelandic, set the text in lead blocks using a spider-​like machine that always seemed to be falling apart, and, finally, ran the reams through the press, thus adding the newly translated text to the pictures. During one of my visits I  picked up on some pages of a new comic, pictures only. I  browsed through several pages, trying to make sense of the story. When the printer saw what I was up to he stopped his work, walked over to me, placed a couple of pictures from the comic on a table, looked at me knowingly, smiled and winked as only he could do, and then he moved the two pictures toward each other until they overlapped. Not a word was said. But I got it. These two were the same! Many days later, when the printer had finished his work and the comics were ready, I learned that the names of the characters involved were “Clark Kent” and “Superman”. Almost without exception philosophers think of Frege’s puzzle in linguistic terms. Solving the puzzle means finding a semantic solution. However, my childhood experience shows that the puzzle can be raised and that one can discover the relevant identity without relying on semantics, which indicates that perhaps the focus when dealing with the puzzle should be on, not semantics, but rather some underlying mental states. After browsing through the pictures at the print shop I had gathered various information about, on the one hand, the person wearing a cape, and on the other hand, the office worker.What the printer helped me realize was that, because the two characters were the same, the two sets of information were about the same individual.What is important here is that a semantic solution to the puzzle will not be able to explain my discovery.1 There were no terms involved in my case and so there is no meaning of the relevant terms involved that captures and explains my discovery. The lesson that I want to draw from the story is that since Frege puzzle cases can extend beyond semantics, the attempted solutions to the cases should not focus solely on semantics. Solutions that do so can at best provide a partial solution to the puzzle. They will not provide a solution that explains the broader phenomenon; the one that includes my childhood case. Below I will provide a solution that accounts for the typical Frege case as well as my childhood case. The solution will, accordingly, not be a semantic solution. Instead it will focus on information we have on objects and how we

153

Heimir Geirsson

organize and access the information. The solution I will provide is psychological and not semantic in nature. At the same time I will show that the solution I provide aligns itself well with Millianism.

The Importance of Simple Sentences For well over two decades the discussion of cognitive significance and the substitutivity of coreferential names focused on names as they appear in embedded sentences, and then particularly in belief reports. Since the Superman stories provide a clear and vivid example that most are familiar with, assume with me that the story details real events, i.e., assume that Lois Lane and Clark Kent are as real as the rest of us and that Superman has, once again, saved Metropolis from a meteor impact. How, then, can we explain Lois assenting to (2) and not to (1), given that that “Clark Kent” and “Superman” refer to the same person? (1) Lois believes that Clark Kent saved the city. (2) Lois believes that Superman saved the city. There is no denying that we do have anti-​substitution intuitions when it comes to sentences like the ones above. How do we explain those intuitions? Those who accept a Frege-​like semantic view can explain the anti-​substitution intuition by arguing that the embedded sentences express different propositions. But those who accept direct reference generally cannot resort to that kind of a solution, since the direct reference theorists generally accept the view that the semantic meaning of a name is just the object named. It is particularly difficult to account for the anti-​ substitution intuition if one is a Millian, as I am, as the Millian further assumes that names only contribute their referents to propositions expressed by simple sentences in which they occur, that the resulting proposition is a singular proposition that can be represented as an ordered pair of an object and a property, and that embedded simple sentences express the same proposition as when not embedded.2 Given this, (3) Clark Kent saved the city and (4) Superman saved the city express the same proposition that can be represented as . Further, the embedded sentences in (1) and (2) express the same proposition and so (1) and (2) have the same truth value. There are good reasons to believe that the attention given to embedded sentences was to a large degree misplaced. In the introduction to Naming and Necessity Kripke pointed out that he never intended to argue for a doctrine of universal substitutivity of proper names. He pointed out that the sentence “Hesperus is Phosphorus” can sometimes be used to raise an empirical issue while “Hesperus is Hesperus” cannot be used in the same way. Sentences expressing identity that are not embedded, simple sentences, resist substitution as well as do sentences embedded in attitude contexts. That is, our intuitions about substitutions are not limited to embedded sentences. But our anti-​ substitution intuitions are not limited to simple sentences expressing identity. Chances are that people will resist substitution in (3) and (4), as well as in (5) and (6) 154

Eliciting and Conveying Information

(5) Clark Kent is Clark Kent. (6) Clark Kent is Superman. I believe that an approach that treats simple sentences and belief reports as raising fundamentally different problems is mistaken. Instead, I  believe that the preferred way to deal with substitution problems is to view substitution in simple sentences and embedded sentences as raising the same problem, in which case we should be able to find a unified solution. We have intuitions that tell us that we cannot substitute freely in simple sentences, and we have intuitions that tell us that we cannot substitute freely in belief contexts. I believe that the same intuitions are at work in both instances and the same kind of an explanation of these intuitions should apply to both cases. The solution that I will present focuses on singular propositions and ways of believing them and it embraces Millianism. I will try to honor our pre-​theoretic intuitions against free substitutions and at the same time agree with the Millian view that truth value is preserved when substituting coreferential names, both in simple sentences and in belief reports. The key move in accomplishing this is to distinguish what our intuitions are working with: semantics or information. Once we recognize that our intuitions are working on two levels, namely with information and with semantics, we can provide an account that recognizes both the Millian view that substitution preserves truth value and our uneasiness with substitutions. The idea that we can believe singular propositions in different ways is, I believe, fundamentally sound. But it is possible to advance this basic idea in various ways or tell different stories about how one can believe singular propositions in different ways. The story I will tell will not focus on a semantic solution, and it will not rely on one systematically mistaking one proposition for another, i.e., it will not be a story about pragmatic implicature. Instead, it will be a psychological story about how we represent objects and elicit information about objects when so prompted. But before I tell my story I will look at an account developed by David Braun and Jennifer Saul.

Braun and Saul’s Mistaken Evaluation David Braun and Jennifer M. Saul have, independently as well as together, presented an attractive account of beliefs of simple sentences.3 The crux of their view is that substitution failures are ultimately to be explained as mistaken evaluations.While I do like some aspects of their account, I will argue that the mistaken evaluation view does not work because it does not adequately account for our anti-​substitution intuitions. Braun and Saul develop a view that does not rely on pragmatic implicature.4 Consider the following sentences: (7) Superman leaps more tall buildings than Clark Kent. (8) Superman leaps more tall buildings than Superman. It appears that (7) is true while (8) is false. Braun and Saul suggest that one may maintain two cognitively separated sets of beliefs, or pools of information, about Clark Kent, one which is associated with the name “Superman” and another which is associated with the name “Clark Kent”, and that these sets attribute different properties to Superman/​Kent, and affect how one believes propositions containing Clark Kent. Braun and Saul write, These two pools of information attribute different properties to Superman/​Clark. For instance, the ‘Superman’ pool attributes to him the property of leaping tall buildings, while the ‘Clark’ pool does not. You also associate different images with the two names. When you quickly evaluated the proposition semantically expressed by [7]‌, your pools of information and images appeared to you to support that proposition’s truth. Therefore, you judged that [7] was true.You did not pause to consider the identity [Superman is Clark Kent] long 155

Heimir Geirsson

enough to notice its logical consequences. Or, if you did, you erred in not considering this good reason to alter your original judgment.5 Braun and Saul imply here that an enlightened speaker, one who knows that Superman is Clark Kent, and one who is attentive, would see, and accept, that both (7) and (8) express falsehood. By the same token, an enlightened and attentive speaker should accept that both (3) and (4) express truth, given that she knows that Superman has prevented the destruction of her city. Our standard intuitions about sentences like (3) and (4), and (7) and (8), are incorrect, they claim. Saul develops the Braun-​Saul view further in her Simple Sentences, Substitution, and Intuition.6 There she argues that we have a reason to believe that a subject often stores information about an individual in different nodes or folders when learning about an individual that is presented in different ways. So, why is it then that we tend to have the intuition that (7) is true? Saul’s answer is that we fail to reflect on the identity of Superman and Clark Kent and hence do not integrate the information about Superman and Clark Kent. Since we have good reasons for not reflecting on the identity and not making the relevant inferences, we don’t do so. Hence, Saul concludes, we take (7) to be true.7 But what if we are aware of the relevant identity and do reflect upon it and still don’t realize that (7) and (8) must have the same truth value? Saul’s answer is that even if we reflect upon the double life of Clark Kent that does not mean that we make all the inferences that one can make about him. Saul provides several possible explanations for this failure to make the relevant inferences. We might simply fail to make some of the inferences, just as students sometimes fail to make simple inferences. Or, it might be the case that we are more convinced of the truth and falsity of the relevant statements than the inference that shows that both have the same truth value. Or, it might be the case that one has greater confidence in one’s truth-​conditional judgments than one’s inference making.8 So, the fundamental answer that Saul provides for our reluctance to freely substitute coreferential names in simple sentences remains the same as before; we fail to make the relevant inferences. Saul further emphasizes her view that our intuitions against substitution depend on us failing to make the relevant inferences when she suggests an explanation as to why we speak the way we do. We do utter sentences such as (9) Shostakovich always signaled his connections to the classical traditions of St Petersburg, even if he was forced to live in Leningrad. Someone who utters (9)  may be aware of the identity in question, namely that St Petersburg is Leningrad, and so it is not lack of awareness of the relevant identity that explains this utterance. In fact, the speaker might take herself to be communicating something true and relevant. The speaker seems to be communicating a contrast between living in St Petersburg and Leningrad. However, if St Petersburg is Leningrad there cannot be such contrast. Presumably, the speaker would never utter (10). (10) Shostakovich always signaled his connections to the classical traditions of Leningrad, even if he was forced to live in Leningrad. So how can we explain that a speaker who is aware of the identity of St Petersburg and Leningrad utters (9) while she would never utter (10)? Here is Saul’s answer. The answer is simple. Sometimes … enlightened speakers well aware of particular identities will fail to make all the inferences that they could from the relevant identity claims … The utterer of [9]‌knows that St Petersburg is Leningrad, and indeed reflects on this at the time that he utters [9]. Nonetheless, he may think that [9] is true, and not odd. This is because he 156

Eliciting and Conveying Information

simply doesn’t use his knowledge of the identity in question to infer from [9] to [10], which might give him pause.9 Surely an unenlightened speaker could utter something like (9). However, Saul retains the primary explanation of the Braun-​Saul view when she argues that an enlightened speaker would only utter something like (9) if he fails to make the relevant inferences. Contrary to Braun and Saul, I will argue that even the enlightened speaker who makes all the relevant inferences could still have a good reason to utter something like (9), and still have a good reason not to freely substitute in simple sentences.That is, I will maintain that truth value is preserved with the substitution but that nevertheless our anti-​substitution intuitions should be respected. I will therefore present a view that is strongly opposed to Saul when she writes “If … I decide that Naïve Millianism is right, I will have good reason to believe that substitution of co-​referential names always succeeds—​in any context”.10 The Millian needs to tell a story that accounts for how (5) and (6), and (7) and (8), respectively, express the same propositions that, consequently, have the same truth value, while explaining how we rightly have strong anti-​substitution intuitions, even when enlightened about the relevant identity. It is such a story that I will tell. The story will rely on a distinction between (a) the proposition expressed by a sentence, and (b) the information conveyed and/​or elicited by a sentence expressing the proposition. The proposition expressed by a sentence is a function of the semantic value of its components and so a study of (a) is a study in semantics. But often two sentences that express the same proposition nevertheless convey and/​or elicit very different information. Sometimes the difference in information between two sentences is due to pragmatic implicature,11 but my concern is not with those, as we can provide an account of different information conveyed or elicited without resorting to pragmatic implicature. Instead we can rely on how different sentences convey different information to or elicit different information from the audience depending on how she organizes information. For example, (3) and (4) express the same propositions and so I cannot claim that (3) expresses a false proposition. However, if I tell someone that Clark Kent saved the city, using the sentence “Clark Kent saved the city”, then that certainly conveys the wrong and misleading information that he did so in his Clark Kent outfit and not in his Superman outfit. I believe that the basic explanation that Braun and Saul provide of our semantic intuitions is a good one. However, I do not believe that their account adequately explains why we have anti-​substitution intuitions. That is, I do not believe that the source of the problem and what lies behind our anti-​ substitution intuitions is mistaken evaluation. Even when I am informed about the relevant identity, that is, even when I am in the know, and consider the relevant statements and the relevant identity there is still something that gnaws at me. Even when I know that Superman is Clark Kent there is still this lurking feeling when I consider the relevant statements, a feeling that I share with Fregeans and many neo-​Russellians, that we cannot freely substitute coreferential names. Even though I understand the semantic machinery that underlies names and that allows me to claim that truth value is preserved, my intuitions still have me resist substitution and so there is a reason to believe that there is something else at work, something beyond semantics, that makes me resist substitution.

Mental Files and Information Webs It is plausible to maintain that we file information away in our minds and that we normally organize information about individuals so that different pieces of information about what we take to be the same individual are kept together in a single file.12 Several philosophers have used the analogy of a file, including my former self, suggesting that it provides a plausible explanatory account of how 157

Heimir Geirsson

information is stored. However, there are reasons to believe that not everything in a file is equally salient. Some pieces of information are more salient than other pieces of information, and thus more readily recalled when one thinks about the relevant object. A web analogy, much like Quine’s web of belief, captures the idea that some pieces of information are more central than other pieces of information, and that some pieces of information are therefore more stable and more readily evoked than other. For example, it is likely that central to most people’s web for Superman is that he is very fast, wears a cape, can fly, and is very strong. It is less important to most that his given name is “Kal El”, and still less important that his father’s name was “Jor-​L”. Accordingly, that information is likely to be less central in the information web and hence less likely to be recalled when one thinks about Superman. It is plausible to assume that in typical cases I form a file about what I take to be a new and unique object and then place information that I take to be of the object in the file. However, unlike many, I view the file metaphor as being just that, a metaphor. I am not granting a special metaphysical status to files. Instead, using the file metaphor is a convenient way of talking about us clustering information together. Also, unlike, e.g., Recanati, I am not assigning semantic values to files. For Recanati files refer and names derive their reference from the files they are associated with. On the account that I provide names are the primary referring items. However, we do use names as one of several ways to access the files they are associated with. And finally, unlike Recanati, I do not require subjects to stand in a special epistemic relationship, an ER relation, to an object the file is of. I see no difficulties in one having mental files of Santa Claus and Minnie Mouse without being able to stand in an ER to them, as one can clearly have a wealth of information about these characters. When my daughter was born, I already had a file in place for her, namely what we might call my Dagny file.When I had my son, I formed my Atli file. In the years since then, I have gathered all kinds of information in the two files. Some information is represented linguistically, either as predicates or propositions. Other information is not couched in linguistic form but rather in imagistic forms or non-​linguistic form. I can recall vividly and see it in my mind’s eye, much like if I was watching a film or a snapshot, how the kids ran around when younger, how they looked on a soccer field, and how they looked when sleeping in their beds. I can of course describe my mental episodes when I recall these moments, but I recall them not as propositions but rather as images.13 In some cases, I can vividly recall scent and mood from certain events in their lives and so the files have a varied and sometimes a vast amount of information.While my children’s elementary school teachers are likely to have files with information about my kids, the information that they have will be very different from what I have. One reason for that is more limited access to my kids than I had. Another reason is that the teachers are likely to focus on different aspects of my kids looks and behavior than do I, and they are likely to have a different view of what is important, and so they will gather different information and organize it in a different way than do I. Generally, if we mistake a person’s identity in such a way that we take one person to be two, such as when I do not recognize a student in my class when I see him working the cash register at a local store, then I form two files for the person. In this case, one will contain information about the person as a student, and the other will contain information about him as a store clerk. When I later find out that they are the same person, I will combine the two files into one. But there are instances when we do not combine files when discovering the identity in question. Suppose that I have one file about Clark Kent as Clark Kent and a second file about Clark Kent as Superman. If I now discover the identity of Clark Kent/​Superman, then I will likely retain my two files. There are several reasons why I might do this. Suppose that I initially formed two files because I thought I was dealing with two individuals and then later find out that they are one. One reason I might keep the files separated after the discovery of the identity is that most people do not know about the identity of Superman and Clark Kent. If I combine the two files and all that is in them, then it is likely that my future communication with others about Superman/​Clark Kent would be confusing and fraught with misunderstandings. If I, because I have combined the information in the 158

Eliciting and Conveying Information

two files, indicate in conversation that Clark Kent, using the name “Clark Kent”, is strong and faster than a speeding train, then it is likely that my respondent will be lost, as she has no such information in her Clark Kent file and, in fact, what I am saying contradicts much of what is likely to be the central information in her Clark Kent file. Second, I might know about the relevant identity of Superman/​Clark Kent from the time I first encounter Clark Kent and nevertheless form two files about him and file information accordingly. The reason for me doing so might be the same as above: avoiding confusion in communication. Third, if someone is developing or has developed a persona that is significantly different from the one he/​she normally displays, then a new file might be useful and even necessary to avoid misunderstandings and miscommunications. For example, when attending a performance by David Bowie, I expect one type of show and music. If, on the other hand, I were to attend a performance by Ziggy Stardust, then I expect a different type of a show and for the most part a different style of music than would be performed at a Bowie concert. The personas are different enough so that one has very different expectations about a Ziggy concert than one has of a Bowie concert. Hence, it is often useful to keep files of different personas separated even though one knows that they are of the same object. Accordingly, if we have good reasons to keep files separate even if we know the relevant identity, then we are hesitant to make inferences that result from combining the two files, as the results of such inferences are likely to be misleading. For example, if I know that I should expect a glam show at a Ziggy concert, it would be misleading to conclude that I should expect a glam show at a Bowie concert, even though I know the relevant identity. One of the reasons for keeping the Bowie file and the Ziggy file separated is the fact that the two put on very different shows. Given that there are many reasons to keep more than one file about someone, even when the identity in question is known, we need to revisit why it is that we form a file about an object. Earlier I claimed that we file information of what we take to be the same individual or object in a single file. That now must be amended. We file information about what we take to be the same individual or object in a single file except if that individual develops personas or displays aspects so different from his, her, or its usual self that we expect significantly different looks and/​or behavior from that persona/​aspect than we do from his, hers, or its usual self. Similarly, when we discover the identity of a person or object that we had taken to be two (or more) persons or objects and thus had created two (or more) files, then we tend to combine the information in the files unless the personas or objects are so different that we expect significantly different looks and/​or behavior from them.

Accessing Files I suggest that the file I have about an individual can be accessed by a marker. A marker directs one’s thought to where information about the relevant individual or object is stored and so allows one to deposit or retrieve information about that individual or object. A marker can be a name, a description, a scent, a picture, or a representation that we associate with a file in a given context. For example, the name “Superman” is a marker for a Superman file as is “The Man of Steel”, and as is a picture or an image of the flying superhero. Similarly, the name “Clark Kent” is a marker for my Clark Kent file and so when I see or hear the name my thoughts are directed to the Clark Kent file. So, when I hear or see a sentence containing the name “Clark Kent”, then that directs me to the appropriate file and information is added to or elicited from the Clark Kent file, and similarly, when I see or hear a sentence containing the name “Superman”, then information is added to or elicited from the Superman file. Similarly, when I see a picture of Superman, then that directs me to my Superman file and not to my Clark Kent file. Typically, the information that is most pertinent in a file represents the most prominent features of the person or object, and it is the information that is most readily accessible and so is the information that is most easily elicited. Accordingly, when, for example, a conversation about 159

Heimir Geirsson

Clark Kent prompts me to recall something about him, then some of his most prominent features are most likely to be elicited. The drab bespectacled reporter comes to mind as the drab bespectacled reporter, while his tie color or shoe size, the more peripheral information that is less salient, is less likely to be brought up. The information elicited from a file determines how I represent the object in the proposition believed and thus determines to a large extent how I believe the proposition. It is therefore of great importance how a proposition is presented to me. Since the names “Clark Kent” and “Superman” are markers for different files, it makes a big difference whether a proposition containing Clark Kent as a constituent is presented to me with a sentence that contains “Clark Kent” or a sentence that contains “Superman”. If the sentence contains “Clark Kent”, then I am directed to my Clark Kent file, and if it contains “Superman”, then I am directed to my Superman file. Accordingly, if someone tells me that Superman saved the city, then “Superman” elicits information from my Superman file, and I believe the proposition expressed accordingly. However, if someone tells me that Clark Kent saved the city, then “Clark Kent” elicits information from my Clark Kent file. The result is likely to be confusion or disbelief on my behalf, since the information elicited from the Clark Kent file has me represent the hero as Clark Kent and not as Superman and nothing in my Clark Kent file indicates that he, in that guise, has superpowers. We can expect a similar result if different types of markers for files are used. Suppose that I am shown a picture of Superman charging his way to a meteor and then moving it to a safe distance from Metropolis. The image of Superman is a marker for my Superman file and so that is the file that is accessed for past and additional information about the superhero. Superman’s acts are perfectly consistent with the information that I already have about him and so this is just one more heroic act. Suppose on the other hand that I am shown a picture of Clark Kent in his drab suit charging towards the meteor and subsequently moving it. The picture of the journalist has me access my Clark Kent file. This time around the new information does not fit what I already know about Clark Kent, the journalist. The result is, again, likely to be confusion and disbelief.

Enlightened and Unenlightened Subjects The file that I have on Clark Kent figures heavily in how I believe a singular proposition that has him as a constituent when the proposition is presented to me with a sentence containing “Clark Kent”. The file I have about him as Clark Kent is what I draw from when I entertain the proposition so presented, and it thus largely determines the way in which I believe the proposition. Similarly, when I believe a singular proposition that has Clark Kent as a constituent and the proposition is presented to me with a sentence containing the name “Superman”, then the file that I have about Clark Kent as Superman is accessed and information is elicited from my Superman file. Accordingly, I believe the proposition in different ways when it is presented to me with the two different sentences.14 The picture presented provides a simple explanation of why I  might assent to (4)  and not to (3) while being fully rational. The names used to refer to the object in the proposition expressed by the sentences elicit information from different files and so I believe the proposition expressed in different ways. If I am not aware of the identity of Clark Kent and Superman, then I treat the names as referring to different individuals and my files are set up on the assumption that they contain information about different individuals.The two sentences express the same singular proposition and so, as we wanted, substitution of coreferential names preserves truth value. However, we preserve our anti-​ substitution intuitions as we can explain those with different names used in the sentences expressing the proposition eliciting information from different files. If we focus on truth values, substitution is fine. Once we turn our attention to information, we resist substitution. Suppose that I  am enlightened about the identity of Clark Kent and Superman. My anti-​ substitution intuitions are still there, and for a good reason. If someone who has witnessed Superman 160

Eliciting and Conveying Information

saving the city tells me that Clark Kent saved the city, using the name “Clark Kent”, then the use of the name “Clark Kent” prompts me to think of Clark Kent as Clark Kent saving the city and so indicates that he saved the city while in his Clark Kent persona; a highly unlikely scenario. The use of the name “Clark Kent” results in the Clark Kent file being accessed, and since the file does not contain information about Clark Kent as Clark Kent having super strength, the report of Clark Kent saving the city indicates that the superhero has revealed his identity to the public. Again, substitution of coreferential names, while preserving truth values, conveys and elicits very different information. What if both I and my respondent are fully aware of Clark Kent being Superman? Even then we do not freely substitute the names when talking with each other. Since we are fully aware of the dual lives that Clark Kent lives, we have two files for him; one for Clark Kent as Superman and one for Clark Kent as Clark Kent. If my respondent now reports that Clark Kent has saved the city, using the name “Clark Kent”, then that conveys that Clark Kent has done so in his Clark Kent guise and so has revealed his identity. The use of “Clark Kent” would direct me to my Clark Kent file, prompting me to think of the drab journalist saving the city. The result would at first be disbelief, and if I accepted the new information, then doing so would require a rather massive overhaul of the information that I have on Clark Kent and Superman. Similarly, if a fellow music aficionado had told me in 2010 that Ziggy Stardust was preparing a tour, then I would have made efforts to attend one of the shows, fully expecting Ziggy to appear as Ziggy and playing Ziggy’s music with the Spiders from Mars, his usual band. I would not have made the same effort had I been told that David Bowie was preparing a tour. Even when I  am in the know about the identity of Superman and Ziggy, I  resist substitution in most situations because the information conveyed and/​or elicited by a sentence containing the names “Clark Kent” or “Ziggy Stardust” is different than the information conveyed and/​or elicited by a sentence containing the name “Superman” or the name “David Bowie”. Contrary to Braun and Saul, there is no mistaken evaluation here. I am in the know about the identity, I might make all the relevant inferences, and I still resist substitution, the reason being that even though the two sentences express the same proposition the information conveyed and elicited is very different. The information does not affect the truth value of the proposition expressed, but it does affect the cogency of free substitution. Let us revisit Saul’s example of St Petersburg and Leningrad. The two problematic sentences were (9) Shostakovich always signaled his connections to the classical traditions of St Petersburg, even if he was forced to live in Leningrad. and (10) Shostakovich always signaled his connections to the classical traditions of Leningrad even if he was forced to live in Leningrad. Saul wrote that an enlightened speaker might utter (9)  because she failed to make the relevant inferences. If she used her knowledge of the relevant identity and inferred (10) from (9), then that might give her pause. There is, I maintain, nothing odd about uttering (9), even after one has made all relevant inferences, because there is a systematic difference in the two uses of the name of the relevant city. It was known as Leningrad during most of the Soviet era, and St Petersburg both before the rise of the Soviet Union and after its fall. Because of this, it is very likely that people will speak of the city using the name “Leningrad” when they are talking about the city during the Soviet reign, and that they will use the name “St Petersburg” otherwise. For example, people talk about the siege of Leningrad and not the siege of St Petersburg during the Second World War. So, while the enlightened speaker will know that St Petersburg and Leningrad are one city, the enlightened speaker will also know that the names 161

Heimir Geirsson

refer to the city during different time periods. It makes sense, then, that the enlightened speaker will store information about the city from one time period in the St Petersburg file and information about the city from a different time period in the Leningrad file, and that includes information about a very different political climate during the time periods. Sentence (9) will therefore convey information to and/​or elicit information from two different files. One file contains information about St Petersburg during the Soviet era, the other does not. Soviet officials appear to have been attentive to a similar distinction, as the following is true: Shostakovich’s signaling his connections to St Petersburg instead of to Leningrad resulted in him having to make a public recantation for some of his works! Even when the relevant information is in place, there is nothing odd about the previous sentence.

When Substitutions are Permitted It is evident that our choice of names is of great importance when making assertions as well as when reporting beliefs. If Lois sincerely assents to (4) and not to (3), then it certainly appears that (2) is true and (1) is not true. And if I utter (3) then I wrongly convey the information that Clark Kent as Clark Kent saved the city. The anti-​substitution intuitions are alive and well. In spite of that the Millian holds that (3) and (4) as well as the embedded sentences in (1) and (2) express the same proposition and that the two simple sentences and the two belief reports have the same truth values. If I see Superman deal with the meteor and report his feat to Lois by uttering (3), then that is likely to be met with disbelief and/​or confusion, Lois not knowing about the identity. But it is likely to be met with the same reaction even if she knows about the identity, because my use of “Clark Kent” directs her to her Clark Kent file, and that file does not have information about Clark Kent doing these kinds of things in his Clark Kent guise. Before finding out about Clark Kent’s true identity Lois would readily accept (5)  while she would not accept (6). Once she is in the know about the identity we still cannot freely substitute “Clark Kent” and “Superman” due to the names directing us to different files and thus conveying and eliciting different information. Similarly, before finding out about the identity, Lois would readily accept (7) and not (8), even though the sentences express the same proposition. Once she knows about the identity, she is likely to still accept (7) due to how the proposition is expressed. The information elicited will include that Clark Kent leaps more tall buildings in his Superman guise than in his Clark Kent guise. In the know or not, (8) will still be puzzling, as the information elicited is not coherent. The discussion of Clark Kent and Superman helps explain why we readily accept (11) and not (12): (11) David Bowie released his last album in 2016. (12) Ziggy Stardust released his last album in 2016. Here we have different information associated with different personas and so we need to clearly separate information files on the two personas. Note also that the example can be presented without using the names “David Bowie” and “Ziggy Stardust”. We could, for example, use pictures of Bowie and Ziggy instead and we would readily accept the information with the Bowie picture and not with the Ziggy picture. The sentences as they stand express the same proposition, according to the Millian, but nevertheless we can only accept one of them as conveying or eliciting accurate information. In this case there is nothing odd about saying that Bowie’s last album bears little resemblance to Ziggy’s work.

162

Eliciting and Conveying Information

The examples discussed so far are significantly different from the following. Suppose that I am talking with someone who knew Robert Zimmerman as a child and who never figured out that Zimmerman is Dylan. Suppose further that I know about my respondent’s lack of knowledge about the relevant identity. In that case I choose my words so that they match the information I know my respondent possesses. Accordingly, I utter (13) and not (14). (13) Zimmerman had a successful music career. (14) Dylan had a successful music career. Alternatively, I might quickly inform my respondent that Zimmerman has taken the name “Dylan” and then I can proceed and utter (14), thus passing on the relevant information. However, there are contexts in which we can freely substitute since neither the truth value nor the information conveyed and/​or elicited is affected by the substitution. Generally, I can freely substitute names if my audience is in the know about the relevant identity and it is not the case that the person named has developed a new identity or a persona. In that regard Zimmerman’s name change is much like that of someone who takes the name of a long-​time partner. His or her old friends can, in that case, talk about the person using either the old or the new name. Here I am, of course, assuming that there is not a significant character change that accompanies the name change. What applies to simple sentences also applies to attitude reports. If I am reporting Stephen’s belief to someone who is not in the know about the Clark Kent/​Superman identity, then it is at the very least misleading to say that if I told the person that Stephen believes that Clark Kent saved the city then I have informed her that Stephen believes that Superman saved the city. Sure, the embedded sentences in my belief reports express the same proposition and so the reports have the same truth values. But the two embedded sentences convey and elicit different information due to different names being employed in them. Assuming that most people have similar key information in their Clark Kent file, the former report conveys the information that Stephen believes that Clark Kent as Clark Kent saved the city. The information I thus convey about Stephen’s beliefs is not accurate. On the other hand, I can accurately convey his beliefs with the latter report. And it would be outrageous to claim that if I report that Stephen believes that Kal El (Kent’s given name on Krypton) saved the city, my respondent never having heard the name “Kal El”, I thereby inform her that Superman saved the city. Even though the sentences “Kal El saved the city” and “Superman saved the city” express the same proposition, they convey and elicit very different information.15 While truth value is preserved with the substitution, the information conveyed and elicited prevents free substitution. If I am reporting Stephen’s belief to someone who is in the know about both of Clark Kent’s names, then the same reasoning as above applies. My choice of names conveys information about how Stephen believes the proposition and so I need to choose my names accordingly. Finally, in some cases we are not concerned with how a proposition is believed, but instead we are primarily concerned with retaining truth value. For example, when it comes to Zimmerman’s childhood friend, then I can use either “Zimmerman” or “Dylan” when reporting her belief that the singer left his hometown at an early age. Assuming that I am reporting her belief to someone who only knows the singer by the name “Dylan”, then I can substitute the name, thus adjusting to my audience, on the assumption that we are concerned with the proposition believed and not how it is believed. As Kripke pointed out, simple sentences that contain different but codesignating names can raise different issues. He did not advocate or accept singular propositions as I do here, and he even speculated that we might have to give up the apparatus of propositions. My goal here has been twofold. First, to show that the Millian has resources that enable her to both accept singular propositions and explain our intuitions about the failure of substitutivity of coreferential names. Everyone, including the Millian, feels the intuitive pull against free substitutions when faced with sentences

163

Heimir Geirsson

such as (3) and (4), and (5) and (6). Appealing to a three-​place belief relation or to the hearer who is in the know about the relevant identity failing to draw all the inferences she should or could draw does not explain the extent of our intuitions. Simple sentences do not contain the belief relation that is present in attitude reports, and even when I am in the know about the relevant identity it seems evident that (5) and (6) can convey and elicit very different information. My second goal has been to provide an account that does not depend on a semantic solution to the problem of substitutivity. The solution I have provided does not depend on the meaning of names, as the same issues can be raised without the use of names, and the same explanations can be provided regardless of whether the issues are raised with names, pictures, or scent.

Further Reading Geirsson, Heimir. Philosophy of Language and Webs of Information, New York: Routledge, 2013. Goodman, R., Genone, J., and Kroll, N. (eds). Singular Thought and Mental Files, Oxford:  Oxford University Press, 2020. Recanati, François. Mental Files, Oxford: Oxford University Press, 2012. Salmon, Nathan. Frege’s Puzzle, Cambridge, MA: MIT Press, 1986. Saul, Jennifer. Simple Sentences, Substitution, and Intuitions, Oxford: Oxford University Press, 2007.

Notes 1 Granted, some argue that pictures can have semantic properties. However, the case for that is far from settled and, at this time, I think we should be open to the idea that pictures do not have semantic properties. Additionally, one can construct examples involving Frege puzzle cases that are based on smell and not pictures and, arguably, smell does not have semantic properties. 2 Note that I do not discuss knowledge context in this paper. I have argued elsewhere that while one can plausibly claim that if one believes that Clark Kent is Clark Kent, then one believes that Clark Kent is Superman, since the two sentences express the same proposition, it doesn’t follow that if one knows the trivial identity statement one also knows that Clark Kent is Superman. Knowledge brings with it justification, and one can be justified in believing a proposition when apprehended in one way and not so justified when it is apprehended in a different way. While the Babylonian astronomers knew that Hesperus is Hesperus, it took them centuries to acquire the justification for believing that Hesperus is Phosphorus. For more on this, see Heimir Geirsson, “Justification and Ways of Believing”, Disputatio 12 (2002); Philosophy of Language and Webs of Information (New York: Routledge, 2013). 3 Jennifer Saul, “The Pragmatics of Attitude Ascriptions”, Philosophical Studies 92 (1998); David Braun, “Understanding Belief Reports”, The Philosophical Review 107 (1998); David Braun and Jennifer Saul,“Simple Sentences, Substitutions and Mistaken Evaluations”, Philosophical Studies 111 (2002). 4 See their “Simple Sentences, Substitutions and Mistaken Evaluations”. Saul continues to develop this line in Jennifer Saul, Simple Sentences, Substitution, and Intuitions (Oxford: Oxford University Press, 2007). 5 Braun and Saul, “Simple Sentences, Substitutions and Mistaken Evaluations”, 17. 6 Saul, Simple Sentences, Substitution, and Intuitions, 128–​130. 7 Ibid., 136. 8 Ibid., 137. 9 Ibid., 145. 10 Ibid., 143. 11 The main proponent of the pragmatic implicature account is Nathan Salmon. See, e.g., his Nathan Salmon, Frege’s Puzzle (Cambridge, MA: MIT Press, 1986). The account is further developed by Scott Soames in Scott Soames, Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity (New York: Oxford University Press, 2002). 12 The idea that we organize information in files or pools is a familiar one. It has been used in various forms in, to name a few, e.g., John Perry, The Problem of the Essential Indexical (New York: Oxford University Press, 1993); Krista Lawlor, New Thoughts about Old Things: Cognitive Policies as the Ground of Singular Concepts, ed. Robert Nozick, Studies in Philosophy (New York: Garland Publishing, 2001); Heimir Geirsson,“Justification and Relative Apriority”, Ratio 12 (1999); “Justification and Ways of Believing”; Philosophy of Language and Webs of Information; Braun and Saul, “Simple Sentences, Substitutions and Mistaken Evaluations”; Saul, Simple Sentences, Substitution, and Intuitions; François Recanati, Mental Files (Oxford: Oxford University Press, 2012).

164

Eliciting and Conveying Information 13 Compare this with Recanati’s files. In his contribution to this volume, “Reference and Singular Thought”, he describes files as containing descriptions. In Mental Files he writes “The role of a mental file based on a certain acquaintance relation is to store information acquired in virtue of that relation. The information in question need not be veridical; we can think of it in terms, simply, of a list of predicates which the subject takes the referent to satisfy” (Recanati, Mental Files, 37–​38, italics are mine). 14 Compare this with what Kripke writes in the introduction to Naming and Necessity:  “My view that the English sentence ‘Hesperus is Phosphorus’ could sometimes be used to raise an empirical issue while ‘Hesperus is Hesperus’ could not shows that I do not treat the sentences as completely interchangeable” (Saul Kripke, Naming and Necessity (Cambridge, MA: Harvard University Press, 1980), 20). 15 Had I  informed her of Superman saving the city, then she would have been justified in believing that Superman saved the city. I have argued elsewhere that the justification for believing a singular proposition is crucially tied to how one believes it, and so given how the proposition is presented one cannot assume that when one is justified in believing a proposition is one way, one is thereby justified in believing the proposition simpliciter.

References Braun, David. “Understanding Belief Reports”. The Philosophical Review 107 (1998): 555–​595. Braun, David, and Jennifer Saul. “Simple Sentences, Substitutions and Mistaken Evaluations”. Philosophical Studies 111 (2002): 1–​41. Geirsson, Heimir. “Justification and Relative Apriority”. Ratio 12 (1999): 148–​161. Geirsson, Heimir. “Justification and Ways of Believing”. Disputatio 12 (2002): 43–​53. Geirsson, Heimir. Philosophy of Language and Webs of Information. New York: Routledge, 2013. Kripke, Saul. Naming and Necessity. Cambridge, MA: Harvard University Press, 1980. Lawlor, Krista. New Thoughts about Old Things:  Cognitive Policies as the Ground of Singular Concepts. Studies in Philosophy. Edited by Robert Nozick. New York: Garland Publishing, 2001. Perry, John. The Problem of the Essential Indexical. New York: Oxford University Press, 1993. Recanati, François. Mental Files. Oxford: Oxford University Press, 2012. Salmon, Nathan. Frege’s Puzzle. Cambridge, MA: MIT Press, 1986. Saul, Jennifer. “The Pragmatics of Attitude Ascriptions”. Philosophical Studies 92 (1998): 363–​389. Saul, Jennifer. Simple Sentences, Substitution, and Intuitions. Oxford: Oxford University Press, 2007. Soames, Scott. Beyond Rigidity:  The Unfinished Semantic Agenda of Naming and Necessity. New  York:  Oxford University Press, 2002.

165

PART IV

Alternate Theories

13 CAUSAL DESCRIPTIVISM Olga Poller

Descriptivism is a theory which attempts to explain why proper names (and names of kinds) designate objects (kinds) and why we can refer to objects (kinds), using names. Its roots lie in classical Fregean understanding of what constitutes the meaning of a proper name. According to Frege (1918 [in] 1984, 359), every proper name is associated with the way in which it presents the object it designates (Gegebenheitsweise) and this way (a manner) of presentation is a way how the object can be determined (Bestimmungsweise, 1914 [in] 1980, 80). These two metaphors encapsulate the classical notion of the meaning of a proper name: there is information connected with a proper name which is shareable (can be grasped by many), presents the object of reference in some aspect (this aspect constitutes the name’s cognitive significance), and which provides the ‘route’ to the name’s reference (determines the name’s reference). Thus, since names are conventionally associated with a property which an object of reference must satisfy to be the referent of a conventional use of a name (cf. Evans 1982, 311), the knowledge of a proper name’s meaning (its referential feature) guarantees that one is able to successfully refer to the name’s semantic referent. Schematically, this classical notion of a proper name’s meaning can be encapsulated in the following two theses, one of which concerns a criterion of reference identification (‘Ref ID’) and the other knowledge (‘Ep ID’): (Ref ID) There is a condition, conventionally associated with a proper name, fulfilling of which is necessary and sufficient for an object to be the name’s referent; (Ep ID) A speaker knows a proper name iff she knows the necessary and sufficient condition an object must fulfill to be the name’s referent. The version of descriptivism criticized by Kripke (1980), is grounded in this classical notion of meaning—​according to it, there exists a descriptive condition associated with a proper name which is necessary and sufficient for an object to be the name’s reference and in order to know the name, the speaker must know the condition. Kripke noticed that if we take possible worlds into account, descriptivism could be understood either as a theory of reference or as a theory of a proper name’s meaning. Consider a possible world w. We might ask: with respect to a possible world w what object is the referent of a proper name in this world? For those who are proponents of a descriptivism understood as a theory of meaning, it is

169

Olga Poller

that object which satisfies the descriptive condition of a (particular) definite description(-​s) in the world w, because the name means the same as the description. On the other hand, for those who are proponents of descriptivism as a theory of reference, it is that object which satisfies a descriptive condition in our world w * because the name doesn’t mean the same as a description—​the description was used only as a tool in our world w * to fix the name’s reference. That is why Kripke’s modal argument (that names have different modal profiles than descriptions) only threatens the version of descriptivism which is understood as a theory of meaning. The possibility of treating the descriptive condition in two different ways has a notable impact on understanding ( Ep ID ) the thesis about knowledge. We can ask: what condition should be known to a speaker in order to know a proper name? Once again, the answer is dependent upon how the descriptive theory is understood—​as a theory of meaning or as a theory of reference-​fixing. For those who postulate that a name means the same as a (particular) definite description(-​s), there is exactly the same descriptive condition that an object should satisfy in any possible world w and this condition should be known to a speaker in order to know a name. But if descriptivism is understood as a reference-​fixing theory, then it is only required that an object should satisfy a descriptive condition in our world w *. In that way, the condition may not be necessary, being understood as mandatory only in our world w *—​in order to know the name, a speaker needs to know a descriptive condition an object fulfills in our world w * to be the name’s referent. That is why only some of Kripke’s semantic arguments threaten descriptivism as a theory of reference-​fixing: for example, if Gödel hadn’t been the author of the incompleteness theorem in a possible world w, then it is possible for a descriptivist to accept the consequence that with respect to a possible world w the name ‘Gödel’ refers to Gödel, despite the descriptive condition being false about Gödel with respect to w—​to refer to Gödel, it is enough that the condition is true about him with respect to our world w *.1 Kripke’s epistemic argument is the only argument that applies to both versions of descriptivism—​because of the ‘iff ’ condition in the epistemic thesis, it is a consequence of a descriptivism that a speaker should know a priori ‘If N exists, then N is ϕ ’ (where ‘N’ is a proper name and ‘ϕ ’ is the descriptive condition fixing the name’s reference). For a descriptive theory of reference-​fixing this statement is an a priori truth which is contingent.2 Now we can formulate both versions of descriptivism (understood either as a theory of meaning or as a theory of reference-​fixing): Descriptive theory of meaning 1 There is a descriptive condition ϕ , conventionally associated with a proper name N, and with respect to a possible world w N refers to whoever fulfills ϕ  in w; 2 A speaker knows N iff she knows the condition ϕ an object must fulfill in w in order to be the referent of N. Descriptive theory of reference-​fixing 1 There is a descriptive condition ϕ , conventionally associated with a proper name N, and with respect to a possible world w the name N refers to whoever fulfills ϕ in (the actual world) w *; 2 A speaker knows N iff she knows the condition ϕ an object must fulfill in (the actual world) w * in order to be the referent of N. In both versions of descriptivism, a condition of reference determination is descriptive (thesis 1) and it is precisely this condition which should be known to a speaker in order to know a proper name (thesis 2). I will refer to descriptivism based on a classical notion of meaning (a condition which determines the reference of an expression is precisely the condition a speaker should know in order to know the expression) as a classical descriptivism.3 170

Causal Descriptivism

It is important to keep in mind that there are very fundamental consequences stemming from the decision as to which expressions the descriptive condition may contain. In particular:  does a descriptive theory allow the descriptive condition to contain indexical expressions?4 An affirmative answer to this question not only has an impact on the notions of the shareability and accessibility of the descriptive condition, but the notion of reference determination as well. After Kripke convincingly argued that proper names are rigid designators and definite descriptions are not, it was quite natural for descriptivists5 to assume that definite descriptions identifying a name’s reference contain rigidifying operators, such as actual or dthat. Assuming a classical interpretation of these operators provided by Kaplan (1989), a definite description containing any of them with respect to a possible world w designates the object that it does with respect to a distinguished (actual) world w *. In that way, if a definite description identifying a name’s reference is rigidified, then a descriptive theory of meaning allows names to be rigid designators (note that for a descriptive theory of reference-​fixing it is without importance if a descriptive condition associated with a proper name contains rigidifying operators or not—​the condition is only used to fix the name’s reference in an actual world w *, and this is why the name designates rigidly, regardless of how the definite description designates). Scott Soames (2005, 307; 2002, 49–​50) pointed out that the costs of such rigidifying are very high for both versions of descriptivism. If the descriptive condition of a definite description is rigidified with the dthat operator, then such a description by definition (Kaplan 1989, 521)  not only designates rigidly but directly. If names appear to be directly referential according to a descriptive theory of meaning, then such a descriptive theory simply collapses into Millianism.Things are also far from simple in cases where definite descriptions contain the actual operator. Consider the following example (similar to the one given by Soames 2005, 303–​304; 2002, 43). Christopher Columbus did not know that he had discovered America, but he could have known. Let c stand for ‘Columbus’, let ‘America’ be synonymous with the definite description ‘the only part of land which is actually situated in the Earth’s western hemisphere’ (ιx.Actually ϕ ) and let us skip for simplicity’s sake the past tense operator. We can render formally ‘It is possible that Columbus knew that he discovered America’ as follows: 1 ( ) ◊K c D (c, ιx.Actually ϕ ) We will skip the reading of (1) in which the definite description lies outside of the ‘know’ operator because this reading is trivial—​Columbus knew about the discovered land that he had discovered it (this is not new information that he could have been aware of). In that way, the only non-​trivial reading of (1) is that in which the definite description containing the actual operator lies within the scope of ‘know’. Actual refers back to our world and as a consequence we attribute to Columbus a knowledge about our actual world, which intuitively is not the case. With a classical definition of actual (Kaplan 1989, 545, definition 10(ii)), it is impossible to co-​index this operator with a possibility operator, and that is why it is impossible to emphasize that in a counterfactual situation Columbus would have had knowledge about his actual world. The importance of Soames’s argument is in its general version: the descriptive condition of a definite description securing a name’s reference and known to speakers should be sensitive to changes in evaluation parameters, and this is why it cannot contain operators which fix parameters of evaluation.6

Non-​classical Descriptivism One of the important conclusions that could be drawn from Kripke’s Naming and Necessity is the opportunity to change our understanding of the notion of meaning and separate a condition of a reference identification from a speaker’s knowledge. According to Kripke’s theory, there exists a linguistic practice (going back to a dubbing ceremony), conventionally associated with a proper name, of passing the name from person to person. This linguistic practice is what secures the name’s 171

Olga Poller

reference (the name’s referent is whoever or whatever was dubbed by N in the first instance of the practice). To know the name, it is necessary and sufficient to participate in the practice (without any knowledge requirements). After Naming and Necessity, it became possible to separate a condition for reference identification from a speaker’s knowledge and this possibility allowed descriptivism to be non-​classical, that is, to hold a descriptive condition either in thesis 1 (a name with respect to possible worlds refers to whatever satisfies a descriptive condition but this condition may not be known to a speaker) or in thesis 2 (it is a linguistic practice, not a descriptive condition, that secures the name’s reference, but in order to know the name, a speaker must know a descriptive condition). There are several branches of non-​classical descriptivism: a metalinguistic descriptivism (there is a linguistic practice going back to baptism which secures the reference of a name, but in order to know the name a speaker should know a (partial) description ‘the bearer of N’),7 a demonstrative descriptivism8 (the reference is secured by a causal relation between an object of reference and a mental file labeled ‘N’, but in order to know the name a speaker should know a description ‘the subject of this dossier (the one I am now accessing)’), causal descriptivism,9 and two-​dimensional causal descriptivism.10

Causal Descriptivism Before we proceed to an explanation of what constitutes causal descriptivism, let us return for a moment to Kripke’s proposal. It can be encapsulated in the following two theses, the first concerning reference determination and the second tackles the knowledge of a name: C1 world w a name N refers to whoever was dubbed by N during the ( ) With respect to a possible dubbing ceremony in w *; C2 ( ) There is a linguistic practice, initiated by the dubbing ceremony and conventionally associated with a proper name N, of passing name’s tokens from person to person with intention to refer to the dubee; a speaker knows N iff he is participating in the practice in w *. Evans (1973) provided counterexamples to both directions of the biconditional in (C2): in order to know a name, it is neither necessary nor sufficient to participate in the linguistic practice of passing the name’s tokens. It is possible to master a name (that is, to learn it and successfully refer with it to the name’s referent) without participation in any chain of passing the name’s tokens with preserved reference to the dubee. For example, it is sometimes possible to ‘figure out’ rules for the giving of names (for example, in the tribe of Wagera Indians, the first born child in a family received the name of a paternal grandfather or, in the American street naming system, a street will receive an ordinal number as a name depending on a numerical order (Evans 1973, 275)) and successfully be able to use it. Also—​because there is no requirement for the disambiguation of what (whose) name is used during the name-​passing practice—​it is possible to learn coincidentally another name and successfully use it, saying something true about its referent. For example, it is possible to learn the name ‘Kingston’ from somebody who was making a racist remark about Kingston upon Thames (with ‘Kingston is the capital of Jamaica’) and provide the correct answer in a quiz (‘What is the capital of Jamaica?—​Kingston’, Evans 1973, 275).11 These counterexamples, however, have remained in the shadow compared to other examples provided by Evans against sufficiency of the passing practice (‘Anir’ and ‘Madagascar’ examples, 1973, 276–​277). For example, despite Marco Polo having participated in the practice—​he gained the name ‘Madagascar’ from Madagascar inhabitants with the intention of using it with the same reference and he passed the name further with the intention preserved, namely to refer to a portion of mainland Africa—​the name does not refer now to a portion of mainland Africa.12 The counterexamples provided by Evans inspired Kroon, Fumerton, and Lewis to fix Kripke’s theory by adding a descriptive condition to it. Thus causal descriptivism is a version of non-​classical descriptive theory of reference-​fixing in which the descriptive condition only secures the name’s 172

Causal Descriptivism

referent—​there is no requirement that this condition should be known by speakers (it is enough that they behave as if it was known, Kroon 1987, 15; Fumerton 1989, 112).The view can be reconstructed as the following two thesis: CD1 ( ) There is a descriptive condition ‘the causal source of that use (that token) of the name N’13 associated with any use of a proper name N, and with respect to a possible world w the name N refers in this use to whoever fulfills the descriptive condition in w *; CD2 ( ) A speaker knows N iff the causal source of her uses of N in w * is N’s referent. The essence of causal descriptivism lies in the idea that a chain of uses of the name tokens should be ‘cut’ on particular ‘links’, each having its own causal connection with its source. If every use of a name has its own causal connection with its source (with the name’s referent) then there is no need to be a participant in a chain of other uses of the name (in that way the theory accommodates Evans’s claim that in order to master a name, it is not necessary to gain it through participation in a linguistic practice of transmission of the name’s tokens). If the practice itself is not necessary to gain the name, then errors during passing tokens of a name (examples such as ‘Madagascar’) pose no threat to the theory. Let us consider some examples (Kroon 1987, 3, 4, 13 n15) to see how the theory is immune to Evans’s critique. Imagine that A is talking about Pliny the Younger. Somebody, say B, interrupts A and asks: ‘Who was Pliny the Elder?’ Consider another story. Imagine that old lady A has a penchant of naming her poodles after famous writers. A tells her friend B (who is aware of A’s penchant): ‘Dumas was the best friend I’ve ever had’. B responds: ‘Who was Dumas?’ Assume that in both stories B hears a name for the first time and uses the names ‘Dumas’ and ‘Pliny the Elder’ in his remarks for the first time. Now, if you share an intuition that B in his remarks refers to Pliny the Elder and Dumas (a writer), you can ask the following question—​how is such a reference possible? In both cases, the intention to refer to the same dubee is not preserved as well as there are no preceding uses of ‘Dumas’ and ‘Pliny the Elder’ in the context with the intention of referring to the writer or a Roman philosopher. Therefore, according to Kripke’s theory, B fails to refer at all. How then can causal descriptivism explain such a reference? According to Kroon (1987, 10) B refers to the individuals (whoever they are) because it is true of them that they have been the causal origin of the particular tokens of ‘Dumas’ and ‘Pliny the Elder’. One can test this intuition by considering a counterfactual situation, for example what if Dumas had never assumed the name ‘Dumas’, A would not have given the poodle the name, and B would never have entertained the belief ‘Dumas is the famous writer whose name A recycled in calling her poodle Dumas’ (cf. Kroon 1987, 16–​17). The counterfactual situation demonstrates that it is events involving the writer that have caused B’s belief. There are several known objections raised against causal descriptivism. Devitt and Sterelny (1999, 61) claimed that causal descriptivism is methodologically redundant. The relation of being the causal source of a particular use of a name N is sufficient to secure N’s reference—​it is the relation itself, not the description built around it, which determines N’s reference. Raatikainen (2006, 82) pointed out that the notion of causal origin is vague and may vary from person to person. If so, then such a semantic theory provides at best a speaker’s reference but not the semantic reference for a proper name, and a theory as such is hardly desirable. Moreover, in case of empty names, such as ‘Pegasus’, the theory provides odd referents. For example, the cause of my uses of ‘Pegasus’ is a particular book about Greek mythology but the book could hardly be counted as the ‘Pegasus’ referent (Raatikainen 2006, 82).

Two-​dimensional Causal Descriptivism Despite the fact that many people have taken issue with causal descriptivism, including the two-​ dimensional version of it,14 we will see that this version (held by Jackson 2010) is based on a quite different idea.Two-​dimensional causal descriptivism has as its integral parts ideas from two-​dimensional 173

Olga Poller

semantics, the causal-​ historical theory, and classical descriptivism. Let us start from the two-​ dimensional part. Two-​dimensional semantics is based on the Carnapian (Carnap 1947) proposal to model the Fregean distinction between meaning (sense) and the denotation of expressions in terms of intension and extension. Kripke convinced us that proper names designate rigidly, and acceptance of this thesis has consequences for a Carnapian. In particular, if two names co-​refer and designate rigidly, then they are necessary coreferential. In turn, necessary coreferentiality between proper names means on the Carnapian basis that identity statements such as ‘Hesperus is Phosphorus’ are analytically true and should be known a priori. But this is not so, as Frege’s ‘Hesperus /​Phosphorus’ puzzle shows. To avoid unwanted analyticity, one of the possible moves open to a Carnapian was to hold two dimensions and assume that proper names behave as indexicals do in Kaplanian semantics—​the reference of proper names may change with respect to centered worlds (possible worlds which contain the name’s referent and token of the name, Jackson 2010, 140–​141), but once the centered world is fixed, the reference of proper names remains the same with respect to all worlds of evaluation.15 In that way, identity statements with coreferential proper names express necessary truths which are not analytical. It is worth noting that despite the fact that two-​dimensionalism treats the actual operator in a standard way (the operator fixes an actual world with respect to worlds of evaluation), it is possible that the operator will fix another world considered as actual because it is assumed in this semantics that an actual world is shiftable (we may consider another world as actual and in such case actual will fix that very world). The possibility to shift an actual world gives the two-​dimensionalist a strategy with which to respond to the Soamsean argument against rigidified descriptivism: saying that Columbus could know that he had discovered America, we are talking about a possible situation in which—​if it is considered actual—​Columbus knew that he had discovered the only part of land which is situated in that situation in the Earth’s western hemisphere. Now let us turn to explanation of the causal and descriptive parts of two-​dimensional causal descriptivism. This view is a version of non-​classical descriptivism which can be reconstructed as the following two theses:16 ( 2DCD1) With respect to a possible world w and a centered world w * a name N refers to whoever was dubbed by N during the dubbing ceremony in w *; ( 2DCD 2) There is a linguistic practice, initiated by the dubbing ceremony and conventionally associated with a proper name N, of passing a name’s tokens from person to person with intention to refer to the dubee; a speaker knows N iff he is participating in the practice in w *and knows a descriptive condition ‘the causal source of that use (that token) of the name N’ that is true of the dubee in w *.17 As we can see, two-​dimensional causal descriptivism is a very weak version of descriptivism—​to know a name, a speaker should participate in the linguistic practice of passing name’s tokens and know the definite description describing the linguistic practice. The descriptive condition does not secure the name’s reference and is quite trivial by itself (it is not the descriptive condition but the name’s intension (primary or A-​intension18) which represents the name’s cognitive significance). So why is knowledge of it demanded at all? The possession of such knowledge enables one to understand what name is used in a context of utterance, whether it is ‘London, Ontario’, or ‘London, England’ (Jackson 2010, 143), because only one city is such that it stands in the causal connection which allows a speaker to conventionally use ‘London’ in the context. Beside this purely practical reason, there is a deep thought going back to Evans (1982) and ultimately to Russell (1911) that ‘[i]‌n order to understand an utterance containing a referring expression, the hearer must link up the utterance with some information in his possession’ (Evans 1982, 305), otherwise he doesn’t know which object is the one he is thinking about. The lack of knowledge prevents a speaker from being an adequate link on any chain of knowledge transmission (cf. Evans 1982, 387). In that way—​knowing the descriptive condition—​a speaker possesses knowledge about the referent of a proper name and this knowledge 174

Causal Descriptivism

gives the speaker both a way of identifying the named object and an object-​dependent thought. As noted earlier, any version of descriptivism holding that in order to know a name it is required to know a descriptive condition faces Kripke’s epistemic and semantic arguments. In two-​dimensional causal descriptivism, the truth of ‘If N exists, then N is the causal source of that use of the name N’ is known a priori because it is established by reflecting on the linguistic practice and the practice guarantees its truth.19 Precisely because the practice itself guarantees its truth, the description cannot be false and thus this makes the theory immune to Kripke’s epistemic and semantic objections (from ignorance and error). The triviality and irrefutability of the description allows us to answer Soames’s (2002, 20–​21) explanatory challenge to descriptive theories (concerning knowledge but not determinacy of reference because the descriptivism is non-​classical): ordinary speakers possess the description by being aware of the linguistic practice, that is why the description is privileged in the sense that ordinary speakers possess it (but do not use it to establish the reference of names). Note that the descriptive condition of the description is reflexive in the sense that it mentions a token of the name. Because the particular use of a name (the token of the name) is not accessible to all speakers, the description is not shareable nor accessible to all speakers as well (probably being different for most speakers), and that is why it can hardly be considered a name’s meaning.

Conclusion What makes descriptivism an attractive theory is its explanatory power, a fact which has even been admitted by Kripke (1980, 27–​29). Classical descriptivism provides a simple explanation to the following questions: • why can speakers refer to objects with names? They can because they know the condition an object must satisfy to be a conventional referent of a name; • how can speakers acquire and transmit information about a particular object, the referent of a proper name (how do we know whom we have talked about)? We acquire and transmit information about a particular object, using a proper name, because we use the name conventionally, that is, to talk about the object which satisfies the referential feature connected (conventionally) with the proper name; • how can we explain the difference between a semantic reference and a speaker’s reference in the case of proper names (Kripke’s ‘Jones /​Smith raking the leaves’ example20)? It is possible to have a misbelief that the person raking the leaves satisfies the descriptive condition conventionally associated with a proper name (a name’s referential feature) and uses ‘Jones’ instead of ‘Smith’ with (an erroneous) belief that this use of the name is a conventional use; • why are identity statements with coreferential (but different) proper names cognitively significant? They are significant because we come to learn that one and the same object satisfies two different descriptive conditions which can be used to identify it (different referential properties), which can be a valuable extension of our knowledge (and useful in practice); • why do sentences with empty proper names, such as ‘Pegasus does not exist’, express true or false statements? Because we truly assert that there is no individual which satisfies the descriptive condition connected with ‘Pegasus’ (there are no winged horses); • why does the substitution of (different) coreferential proper names salva veritate in ascriptional contexts fail? It fails because one but not another descriptive condition is a part of the ascription (for example, it is a part of the ascription that the referent of a proper name is identified in a descriptive way connected with one but not another proper name).21 In non-​classical descriptivism, a descriptive condition is only exploited in order to answer some of these questions, all of which vary for the different branches. Because non-​classical descriptivism is 175

Olga Poller

mixed with the causal-​historical theory of reference, it is unclear as to what answer it provides to the question of the semantic content of a proper name—​does a descriptive condition constitute a part of a proposition and, in case it does, is this part truth-​conditionally relevant? The answer to this question remains especially unclear if descriptivism is understood not as a theory of meaning but as a theory of reference-​fixing, and in particular in its non-​classical version. It is notable that the mentioned questions concern mostly the cognitive significance of names and speakers’ beliefs, and that is why the most promising version of non-​classical descriptivism is the one which postulates that the descriptive condition is known to speakers and constitutes a part of a proposition expressed with a sentence containing a proper name. The two-​dimensional version of descriptivism seems more promising in this respect. First, it represents a proposition as a set of possible worlds (as a conglomerate of the primary and secondary intensions of a sentence) and such representation allows it to provide a clear answer as to what should be considered the semantic value of a proper name—​it is the name’s primary and secondary intension. The primary intension of a proper name could be the same as the primary intension of a definite description and could be different for coreferential proper names. The variability of primary intensions of proper names allows one to answer questions about substitutivity and the cognitive significance of names. Second, the descriptive condition that is required to be known (‘the causal source of that use (that token) of the name N’) is irrefutable and allows questions as to why speakers can transmit knowledge about a particular object to be answered, as well what is the difference between semantic reference and a speaker’s reference of a proper name. Third, names in two-​dimensionalism are rigid designators (they have the same secondary intensions across possible worlds) which allows this version of descriptivism to accommodate Kripke’s thesis about the rigidity of names. In this manner, a two-​dimensional version of descriptivism has all of the virtues of traditional descriptivism whilst remaining immune to the objections and charges of Kripke.

Acknowledgments I owe many thanks to Katarzyna Kijania-​Placek for her valuable comments and criticism.

Notes 1 As I have noted, taking into account possible worlds leads to the differentiation between descriptivism as a theory of meaning and as a theory of reference-​fixing. It is possible to take a further step and ask the following question: what will change if we additionally take into account the time of evaluation? To know a name at a time t , must one know the relevant descriptive condition at that time or is it enough to know the condition at some other (probably somehow distinguished) time? Will such descriptivism survive Kripke’s objections and remain a non-​trivial semantic theory? (cf. Poller 2016, 2017). 2 See more about the epistemological argument in Jeshion (2002). 3 Glüer and Pagin (2006, 2008) hold a version of a classical descriptivism. 4 For example Strawson, who is considered a classical descriptivist, explicitly stated that it is impossible to free descriptions identifying the name’s reference from indexicals (1959, 182 n1). 5 See Plantinga (1978), Ackerman (1979), Searle (1983), Stanley (1997), Jackson (1998), Pettit (2004), Kallestrup (2012). Joseph Almog (1986, 224 n12) called the idea to rigidify definite descriptions ‘a semantical theft which substitutes an honest metaphysical toil’. 6 A possible response to this objection is to move to a two-​dimensional semantics, see Kallestrup (2012) and (Nelson 2002), and for a response to such move see Soames (2005, 305 n17) and Everett (2005, 120–​121). 7 See Kneale (1962), Loar (1976, 1980), Bach (1987, 2002), Katz (1994, 2001), and Justice (2001, 2007). Barbara Abbott (2004, 6) called the proponents of metalinguistic descriptivism ‘partisans’. 8 See Forbes (1989, 1990, 1993). 9 See Kroon (1983, 1985, 1987), Lewis (1984, 1997), and Fumerton (1989). 10 See Jackson (1998, 2010, and this volume). 11 A similar example is given by McKinsey (2011: 329–​330) and Soames (2005, 301, 419). 12 For more on reference change see Harold Noonan, this volume.

176

Causal Descriptivism 13 Cf. Kroon (1987, 1 n1, 2, 10, 15), Lewis (1997, 339 n22), Fumerton (1989, 114). 14 Raatikainen (2006, 79), Soames (2005, 287), Cumming (2016). 15 Note that the extension of indexicals, such as actual, may change with respect to centered worlds. This gives two-​dimensional semantics the potential to explain how it is possible to express a contingent a priori truth with sentences such as ‘Julius is the actual inventor of the zip’. 16 For more on this, see Jackson, this volume. 17 Thesis 1 cf. Jackson (2010, 15, 138, 143; 2007, 20). Thesis 2 cf. Jackson (1998, 203, 212; 2010, 143–​144). 18 See Jackson, this volume, for the notion of A-​intension explained. As I  have noted, to avoid unwanted analyticity of statements with coreferential names, one may hold two dimensions and let proper names behave as indexicals do in Kaplanian semantics—​their reference may change with respect to epistemic scenarios (i.e., epistemically possible situations), but once a scenario is fixed, the reference of proper names remains the same with respect to all worlds of evaluation. In that way, proper names are seen as having variable epistemic intensions (primary intension, a function from scenarios to objects) but stable metaphysical intensions (secondary intension, a function from possible worlds to objects with respect to a fixed scenario, Chalmers and Rabern 2014, 211–​212). Take for example coreferential names ‘Hesperus’ and ‘Phosphorus’. Despite their stable secondary intensions (which guarantee that names are rigid designators), the primary intensions of the names vary independently across epistemically possible scenarios, so there is a scenario in which the primary intensions differ. The primary intensions are supposed to correspond to a Fregean sense and could be reconstructed as the diagonals of two-​dimensional intensions of the names (a two-​dimensional intension is a mapping from scenario-​world pairs to extensions, Chalmers 2006, 102) (cf. Poller 2019). 19 A similar argument concerning ‘blue’ can be found in (Pettit 2004, 335). 20 Talking about the difference between the two, Kripke provided the following example (1977, 263): A and B see Smith and mistake him for Jones. A asks: ‘What is Jones doing?’ and B answers ‘Raking the leaves’. The name ‘Jones’ refers to Jones in the common language of A and B (Jones is the name’s semantic referent); nevertheless, in this small dialogue A and B both referred to Smith with ‘Jones’ (Smith is the speakers’ referent). 21 See Michael Nelson’s contribution to this volume for elaboration of these points.

References Abbott, B. 2004. Proper Names and Language. [In] G. N. Carlson and F. J. Pelletier (Eds.), Reference and Quantification: The Partee Effect (pp. 1–​19). Chicago, IL: CSLI Publications. Ackerman, D. 1979. Proper Names, Propositional Attitudes and Non-​Descriptive Connotations. Philosophical Studies, 35(1), 55–​69. Almog, J. 1986. Naming without Necessity. The Journal of Philosophy, 83(4), 210–​242. Bach, K. 1987. Thought and Reference. Oxford: Clarendon Press. Bach, K. 2002. Giorgione Was So-​called Because of His Name. Philosophical Perspectives, 16, 73–​102. Carnap, R. 1947. Meaning and Necessity. Chicago, IL: Chicago University Press. Chalmers, D. 2006. The Foundations of Two-​Dimensional Semantics. [In] M. García-​Carpintero and J. Maciá (Eds.), Two-​Dimensional Semantics (pp. 55–​140). Oxford: Oxford University Press. Chalmers D. and Rabern, B. 2014. Two-​Dimensional Semantics and the Nesting Problem. Analysis, 74(2), 210–​224. Cumming, S. 2016. Names. [In] E. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Fall 2016 Edition): https://​ plato.stanford.edu/​archives/​fall2016/​entries/​names/​. Devitt, M. and Sterelny, K. 1999. Language and Reality: An Introduction to the Philosophy of Language. Oxford: Wiley Blackwell. Evans, G. 1973.The Causal Theory of Names. [In] A. P. Martinich (Ed.), The Philosophy of Language (pp. 270–​283). New York: Oxford University Press, 1996. Evans, G. 1982. The Varieties of Reference (Ed. J. McDowell). New York: Oxford University Press. Everett, A. 2005. Recent Defenses of Descriptivism. Mind & Language, 20(1), 103–​139. Forbes, G. 1989. Languages of Possibility: An Essay in Philosophical Logic. Oxford: Basil Blackwell. Forbes, G. 1990. The Indispensability of Sinn. The Philosophical Review, 99(4), 535–​563. Forbes, G. 1993. Reply to Marks. Philosophical Studies, 69, 281–​295. Frege, G. 1914. VIII/​12 Frege to Jourdain Undated. [In] B. McGuinness (Ed.), Philosophical and Mathematical Correspondence, Oxford: Basil Blackwell, 1980. Frege, G. 1918. Thoughts. [In] B. McGuinness (Ed.), Collected Papers on Mathematics, Logic, and Philosophy, Oxford: Basil Blackwell, 1984. Fumerton, R. 1989. Russelling Causal Theories of Reference. [In] C. W. Savage and C. A. Anderson (Eds.), Rereading Russell (pp. 108–​118). Minneapolis, MN: University of Minnesota Press.

177

Olga Poller Glüer, K. and Pagin, P. 2006. Proper Names and Relational Modality. Linguistics and Philosophy, 29(5), 507–​535. Glüer, K. and Pagin, P. 2008. Relational Modality. Journal of Logic, Language and Information, 17, 307–​322. Jackson, F. 1998. Reference and Description Revisited. Noûs, 32 (Supplement: Philosophical Perspectives, 12, Language, Mind, and Ontology), 201–​218. Jackson, F. 2007. Reference and Description from the Descriptivists’ Corner. Philosophical Books, 48(1), 17–​26. Jackson, F. 2010. Language, Names and Information. Malden, MA and Oxford: Wiley Blackwell. Jeshion, R. 2002. The Epistemological Argument against Descriptivism. Philosophy and Phenomenological Research, 64(2), 325–​345. Justice, J. 2001. On Sense and Reflexivity. The Journal of Philosophy, 98(7), 351–​364. Justice, J. 2007. Unified Semantics of Singular Terms. The Philosophical Quarterly, 57(228), 363–​373. Kallestrup, K. 2012. Actually-​Rigidified Descriptivism Revisited. Dialectica, 66(1), 5–​21. Kaplan, D. 1989. Demonstratives. [In] J. Almog, J. Perry and H.Wettstein (Eds.), Themes from Kaplan (pp. 481–​563). New York: Oxford University Press. Katz, G. 1994. Names without Bearers. The Philosophical Review, 103(1), 1–​39. Katz, G. 2001. The End of Millianism: Multiple Bearers, Improper Names, and Compositional Meaning. The Journal of Philosophy, 98(3), 137–​166. Kneale, W. 1962. Modality De Dicto and De Re. [In] E. Nagel, P. Suppes and A. Tarski (Eds.), Proceedings of 1960 International Congress (ss. 622–​633). Stanford, CA: Stanford University Press. Kripke, S. 1977. Speaker’s Reference and Semantic Reference. Midwest Studies of Philosophy, 2(1), 255–​276. Kripke, S. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press. Kroon, F. 1983. The Problem of ‘Jonah’: How Not to Argue for the Causal Theory of Reference. Philosophical Studies, 43(2), 281–​299. Kroon, F. 1985. Theoretical Terms and the Causal View of Reference. Australasian Journal of Philosophy, 63(2), 143–​166. Kroon, F. 1987. Causal Descriptivism. Australasian Journal of Philosophy, 65(1), 1–​17. Lewis, D. 1984. Putnam’s Paradox. Australasian Journal of Philosophy, 62(3), 221–​236. Lewis, D. 1997. Naming the Colors. Australasian Journal of Philosophy, 75(3), 325–​342. Loar, B. 1976. The Semantics of Singular Terms. Philosophical Studies, 30(6), 353–​377. Loar, B. 1980. Names and Descriptions: A Reply to Michael Devitt. Philosophical Studies, 38(1), 85–​89. McKinsey, M. 2011. Understanding Proper Names. Linguistics and Philosophy, 33, 325–​354. Nelson, M. 2002. Descriptivism Defended. Noûs, 36(3), 408–​436. Pettit, P. 2004. Descriptivism, Rigidified and Anchored. Philosophical Studies, 118, 323–​338. Plantinga, A. 1978. The Boethian Compromise. American Philosophical Quarterly, 15(2), 129–​138. Poller, O. 2016. Obrona deskrypcyjnej teorii odniesienia nazw. Część I.  Odpowiedź na argumenty Kripkego: modalny i epistemiczny. [In defense of a descriptive theory of reference for proper names. Part I. A response to Kripke’s modal and epistemic arguments]. Principia, 63, 85–​108. Poller, O. 2017. Obrona deskrypcyjnej teorii odniesienia nazw. Część II. Odpowiedź na semantyczne argumenty Kripkego. [In defense of a descriptive theory of reference for proper names. Part II. A response to Kripke’s semantic arguments]. Principia, 64, 85–​112. Poller, O. 2019.Variability, Rigidity and the Nesting Problem. Theoria, doi:10.1111/​theo.12206. Raatikainen, P. 2006. Against Causal Descriptivism. Mind & Society, 5(78), 78–​84. Russell, B. 1911. Knowledge by Acquaintance and Knowledge by Description. Proceedings of the Aristotelian Society, 11, 108–​128. Searle, J. 1983. Intentionality. New York: Cambridge University Press. Soames, S. 2002. Beyond Rigidity:  The Unified Semantic Agenda of Naming and Necessity. Oxford:  Oxford University Press. Soames, S. 2005. Reference and Description: The Case againstTwo-​Dimensionalism. Princeton, NJ and Oxford: Princeton University Press. Stanley, J. 1997. Names and Rigid Designation. [In] B. Hale and C. Wright (Eds.), A Companion to the Philosophy of Language (pp. 555–​585). Malden, MA: Blackwell Publishers. Strawson, P. 1959. Individuals: An Essay in Descriptive Metaphysics. New York: Routledge, 1996.

178

14 REFERENCE-​FIXING AND PRESUPPOSITIONS Manuel García-​Carpintero

1  Introduction: Reference-​fixing, Semantics, and Metasemantics Kripke (1980, 55) distinguishes between using a description to fix the reference of an expression and using it to give its meaning. Kaplan and Stalnaker then articulated a distinction between semantics and metasemantics or foundational semantics, ascribing complementary roles to each. To the former category belong theories that assign meanings to their bearers, prominent among them linguistic expressions. To the latter belong theories that provide “the basis” for ascribing such meanings (Kaplan 1989b, 573–​574) or state “what the facts are” that give these meanings to their bearers (Stalnaker 1997, 535). This is a metaphysical undertaking on the grounding of meaning-​facts, on what determines, fixes, or constitutes them.1 This distinction is sometimes used to consign to “mere” metasemantics descriptive material that, on both intuitive and theoretical grounds, plays a linguistically significant role in the determination of the referents of names and other expressions.2 In previous work (García-​Carpintero 2000, 2006a, 2018a), I  appealed to reference-​fixing presuppositions with the aim of undermining this application of the distinction. Following Heim, I  assumed that linguistic presuppositions are features of linguistic meaning (García-​Carpintero 2018b). Indeed, in the ensuing years the view that descriptive reference-​fixing presuppositions are part of linguistic meaning has become mainstream in semantics (cf., e.g., Heim 2008; Hunter 2013; Maier 2010, 2016). For such descriptive material—​I have argued—​Kripke’s distinction only tracks a contrast between “planes” or “levels” of content—​“at issue” vs. “backgrounded”. However, it is not straightforward to understand how the relevant presuppositions that semanticists posit play the reference-​fixing role that Kripke was envisaging. In this contribution I want to confront this issue. I will approach it by assuming an account of the semantics vs. metasemantics divide that I have provided elsewhere (García-​Carpintero 2012, forthcoming-​a). I advance there a version of the Austinian normative approach originally promoted by Alston (1964), Austin (1962) and Searle (1969).3 Although displaced for a while by the influence of Gricean and Chomskian views on the issue, it is becoming popular again among philosophers and linguists. On this view, meaning-​facts about natural languages are determined by social norms and social conventions. In what follows, I will articulate in that framework the distinction between semantic and non-​ semantic (“pre”-​or “post”-​ semantic) facts in reference-​ fixing, presuppositionally understood.

179

Manuel García-Carpintero

A full argument for the view that I’ll present should be abductive. In addition to articulating it, I’ll offer a small portion of such an inference to the best explanation. A main consideration is that it assumes and contributes to vindicating an overarching, general account of semantic reference. It is an Austinian analog to the Gricean picture developed by Schiffer (1981) and Neale (2004, 2016), elaborating on Strawson’s (1950) ideas; I’ll highlight the parallel. Another consideration is how the proposal contributes to taking a principled stance on recent debates confronting “intentionalists” and “conventionalists” on the (derivative) metasemantics of tokens of indexicals and other referential expressions. Metasemantic debates presuppose views on the underlying semantic facts. They thereby also presuppose views about what languages are. In keeping with the Austinian stance, here I’ll assume that natural languages are essentially social tools. More specifically, I’ll assume that, in an explanatorily prior sense, languages are conventional devices put in place and backed by social rules. I will assume a minimal characterization of conventions—​a common core to the accounts by Lewis (1975), Bach and Harnish (1979, 120–​134), Davis (2003, 204–​219), Marmor (2009), and Geurts (2019). A convention is a social regularity in the behavior of a group, which serves a common interest (it solves a “coordination problem”, as Lewis puts it), and is arbitrary in that there is an alternative which would have solved it just as well.

2  Semantic Content: Character and Locution On the Austinian metasemantic picture for linguistic types that I  assume, languages are social tools devised for communicative purposes. Conventions concerning the specific lexical items and constructions of a particular language, grounded on a subpersonal combinatorial linguistic competence driven by our biology, assign to sentence-​types semantic contents that it is apt to characterize as “speech-​act potentials” (Bach and Harnish 1979; Alston 2000), in ways I’ll explain below.4 The underlying social and psychological facts determine which assignments are correct. The view of semantic content—​“what is said”—​on offer is thus “social” (Camp 2006, §6) or “forensic” (Perry 2009, 191). In what follows I’ll develop this rough picture. I’ll start with a distinction of Lewis’s (1980) in what I undiscriminatingly described in the previous paragraph as “semantic content”. Yalcin (2014) disambiguates the two notions as “semantic value” and “content”; others refer to them as, respectively, “compositional” vs. “assertoric” content.5 I will here use the (from my perspective) descriptively more accurate “sentence character” (or just “character” for brevity),6 and “locutionary meaning” (“locution”). Character is ascribed to sentences, given the widespread context-​dependence present in natural language, in order to fulfill central explanatory tasks for theories of natural languages, among them:  accounting for facts about systematicity and productivity in understanding, communication, and acquisition, and explaining judgments about entailments, truth-​value, or correctness relative to particular situations.7 Locutionary meaning is the linguistically determined speech-​act potential assigned to sentences in context. Ultimately, the data in need of explanation in those cases (systematicity, etc.) concern locutions, and hence characters should properly relate to them; but, as Lewis (1980) points out, it doesn’t follow that they should be identified, and there are reasons against this that we will now examine. On the Austinian picture of the basic metasemantic facts that I want to elaborate here, utterances play a central role.8 Utterances are intentional actions, speech acts. Ball (forthcoming) provides a compelling account of Austin’s (1962) notorious view of utterances as separable in different, but somehow embedded acts: a “phonetic”, a “rhetic”, a “locutionary”, an “illocutionary”, and a “perlocutionary” act.9 Rehearsing arguments from Moltmann (2018), Ball provides forceful replies to Searle’s (1968) criticism, and an account of the locutionary/​illocutionary distinction that I  find adequate.10 The rhetic act is the act of using lexical items and their grammatical modes of combination “with a certain more or less definite ‘sense’ and a more or less definite ‘reference’ (which together are equivalent to 180

Reference-Fixing and Presuppositions

‘meaning’)” (Austin 1962, 93). To individuate the locutionary act, as Ball construes it, we generalize away the specific lexical items.11 In contrast with Searle’s (1968) alternative view, locutions don’t merely consist on this picture of a propositional content, but they include specifications on illocutionary force too: they can be the presentation of a content with erotetic, directive, or assertoric force, and further forces to the extent that they are locuted, on which more below. Locutions don’t differ from standard propositional contents only in this; crucially for our topic, they also typically include separate “planes” of propositional content:  “at issue” vs. backgrounded—​presupposed or conventionally implicated (García-​Carpintero 2006b, 43–​47; 2008, 68–​76). Nonetheless, the locutionary act performed in an utterance should be distinguished from the illocutionary act.12 If someone utters ‘The vote was anonymous’, meaning that the vote was unanimous (Bach and Harnish 1979, 33), the locution is (let’s assume) the assertoric presentation of the at-​issue propositional content that a vote was anonymous, on the background presupposition that there was one and only one such vote in the context; but it is at the very least unclear whether the speaker becomes thereby assertorically committed to that content.13 But this is what is required for the illocutionary act to occur. In the proper context (for instance, if the audience is familiar with the fact that the speaker systematically incurs in this malapropism), the speaker might well become assertorically committed instead to the claim that the vote was unanimous;14 but this is not his locutionary act. Everybody in these debates wants to preserve the intuitive distinction between literal, or direct meanings, and nonliteral, implied, or indirect meanings.15 Locuted meanings are what is literally, directly conveyed: the presentation of a set of more or less well specified at issue and backgrounded contents with more or less specific forces. The previous case shows that the speaker need not be committed to the relevant content in the way constitutive of that force. Substitution implicatures and indirect speech acts in general (in contrast with additive implicatures, like Grice’s (1975) recommendation letter and gas petrol examples,16 cf. Meibauer (2009),Vandeveken (1991, 375–​376)) also support the point, for in such cases the speaker is not illocutionarily committed to the locuted content. Contemporary semanticists assign some semantic significance to at least the three moods apparently present in all languages: declarative, interrogative, and imperative (Charlow 2014; Starr 2014; Roberts 2018). Character, not just locution, thus includes information about potential illocutionary force. Hanks (2015, 9)  rejects what he characterizes as the taxonomic version of Frege’s traditional force-​content distinction, which I have assumed in the discussion so far. This is the idea influentially articulated by Stenius (1967) that there is a meaning-​component (a truth-​or fulfillment-​conditional component) common to speech acts of different illocutionary types, including questions, directives, and assertions. As Collins (2018, 3538–​3539) shows, Hanks’s reasons are not cogent. For current semantics also distinguishes meanings for noun phrases, meanings for verb phrases, and forceless meanings for phrasal combinations thereof, which are common constituents of the distinct semantic objects assigned to imperatives, interrogatives, and indicatives.The appeal to current semantics thus in fact legitimizes force-​endowed sentential meanings as much as their forceless “parts”. My (at-​issue or backgrounded) locuted contents are forceless traditional propositions. Now, in contrast to Hanks and others who have written recently on the traditional topic of the “unity of propositions”, I think we should adopt an attitude towards them that is as minimalist as possible. Propositions are force-​neutral, but they also lack any “structure”—​whatever that might mean for abstract entities (Keller 2013). I will assume the Stalnakerian view that they are just properties of verifying circumstances of evaluation.17 What are such circumstances? For Stalnaker they are complete and consistent possible worlds, for Lewisians centered possible worlds. I will think of them as “smaller” than full possible worlds, as in Situation Semantics or in “truthmaker semantics” (Fine 2017).18 Let’s rehearse now the Lewisian distinction between sentential character and locutionary meaning. Consider indexicals. They typically have three types of uses. A  sentence like ‘He is happy’ might be uttered on its own, the pronoun ‘he’ perhaps accompanied by a pointing gesture. In that case, philosophers after Kaplan (1989a) would think of its at-​issue locutionary content as a singular 181

Manuel García-Carpintero

proposition, ascribing the property of being happy to the male pointed at.19 The pronoun might be anaphoric on another expression, getting its referent from it: ‘Peter thinks that he is happy’. And it can be bound by an expression of generality: ‘Everybody/​somebody thinks that he is happy’. These are not isolated facts; other expressions exhibit the same variations, and they systematically re-​occur across languages. Considerations of explanatory power promote a unifying explanation. Semanticists have advanced different sophisticated frameworks to account for the data in such an explanatorily powerful way.20 For our purposes, however, we can make the point we need by using the simple and hopefully familiar technical apparatus of First-​Order Logic used in Tarskian-​Davidsonian truth-​conditional approaches.21 We can think that explicit indexicals like ‘he’, or the hidden counterparts linguists have reasons to posit in the logical forms corresponding to the sentences—​ the syntactically articulated representations that act as inputs to the semantic machinery—​behave semantically like variables in first-​order logic (FOL) languages. They are interpreted relative to assignments, which can be varied relative to (contextually specified) domains of quantification, suitable to capture the semantic behavior of expressions of generality. For ‘He is happy’, we get something like this representation: (1) On the presumption that x is the “demonstrated” male, x is happy. (1) is not right, in many ways. For one thing, the presupposed condition that the referent is male really belongs in a different content-​plane.22 Also, ‘demonstrated’ is placed inside scare quotes to point in the direction of the complications to be discussed below.23 Details aside, what we thus obtain by way of the semantic, compositionally determined content for (1) is a property like the one denoted by (2):24 (2) λx. On the presumption that x is the “demonstrated” male, x is happy. Lewis (1980) insists that there is no difficulty in deriving from semantic, compositionally given sentential meanings like (2) locutionary propositional contents that correspond better to what is intuitively said when our sentence is literally uttered. Writers in the semantic tradition I am rehearsing invoke at this point the following recourse: the “demonstrated” male in the context of a particular utterance is identified with the value of the relevant variable given by a particular assignment, which is said to have been selected by the utterance context C (Heim and Kratzer 1998, 243). We thus obtain representations like this for our locuted content: (3) [λx.On the presumption that x is the “demonstrated” male, x is happy.] a, where a = the C-​ “demonstrated” male, if any. This illustrates the distinction between sentential and locutionary meanings. I am calling the former ‘characters’, generalizing Kaplan’s (1989a) character/​content distinction to all context-​dependent expressions, along the lines illustrated here for ‘he’. Context-​dependence is pervasive (Bach 2012). Quantifiers exhibit it (Stanley 2000), also tense and modals (Kratzer 2012), gradable adjectives (Kennedy 2007), and many other expressions. The potentially universal phenomenon of polysemy (Ludlow 2014) makes it even more pervasive. Characters (expression-​ type meanings) are clearly semantic contents, and hence there is a metasemantics for them. As said, Lewis (1980) thought that locutions like (3)  (expression-​ use meanings) are equally semantic, by assuming a principle that I will call (L)inking:25 (L) The content communicated by a literal utterance of S in context C is caeteris paribus the semantic content of S in C. 182

Reference-Fixing and Presuppositions

In spite of assuming (L), Lewis (1980) distinguishes character and locution. Doubts about (L) suffice to show that the distinction is well-​taken, even if—​as I’ll suggest—​they should be defused. Contents like (2) are not those we intuitively put literally forward when we assert ‘He is happy’. Contents like (3) would uphold (L) for this case, as Lewis was assuming. But there are reasons for doubting that they are properly qualified as semantic. Nowak (forthcoming) offers some of them.26 On the one hand, even if (2) is all that counts as the semantic content of our sentence, there are conceptually and empirically solid ways to explain how speakers might nonetheless converge on something like (3); “postsemantic”, pragmatic resources—​those involved in the derivation of implicatures—​account for it. On the other hand, proposals to uphold (L) in these cases, outlined below, are highly controversial. Harris (2017, forthcoming) nicely articulates the more general considerations behind Nowak’s argument—​García-​Carpintero (2006b) develops a similar view. Harris argues that the semantics/​ pragmatics divide traces a natural, real divide in cognition:  one separating a sufficiently isolated Fodorian module (a Chomskian I-​language), from central inferential capacities. On this view, semantics only provides constraints on the intuitive literally conveyed meanings—​like (2) in our example—​and (L) fails. Harris also provides a reply to compositionality-​based objections to this view of semantics grounded on the fact that what it ultimately composes is the values of characters in context (cf. García-​Carpintero 2006b, 51–​52 and Michaelson and Woods (n.d.) for similar rejoinders).To account for systematicity and productivity, constraint semantics needs a characterization of compositionality that allows for “pragmatic intrusion” throughout the composition process; Pagin and Pelletier (2007) provide one such characterization.27 This is the view that so-​called “contextualists” in debates about the semantics/​pragmatics divide—​ like Bach (1994), Carston (2002), Neale (2005, 2016), and Schiffer (2003)—​have advanced, for reasons that Nowak (forthcoming) and Harris (forthcoming) bring to the fore.28 So-​called “minimalists” like Borg (2012) and Cappelen and Lepore (2005) are sympathetic to such motivations, but they nonetheless want semantics to deliver truth-​evaluable contents. It is unclear how stable the minimalist motivations and their proposals are: the motivations commit them to (L), but it is very unclear that their proposals in fact vindicate it.29 What separates minimalists from contextualists is their take on (L). Minimalists—​as much so as “indexicalists” like Stanley (2000) or King (2014)—​aim to uphold (L); contextualists just dismiss it. I agree that there are very strong reasons to support (L). As I mentioned earlier, the main initial data that semantic theories aim to explain concern locutions. If (L) obtains, there is also a metasemantics for utterances (uses of sentences in context), in addition to that for sentence-​types.30 Constraint semantics dispenses with it: no semantics for utterances, no need for a metasemantics. There might be a metasemantics for utterance meanings to the extent that there is one for pragmatically conveyed meanings, but there is none when it comes to the semantics of natural languages:  utterances (or sentences-​in-​context) don’t have a distinctive one, i.e., any one beyond that for the sentence-​types they instantiate. This, however, is too cavalier an attitude to take.31 Even if semantic content reduces to character, we do need to define characters, and this requires a well-​supported view on what the contents they determine relative to contexts are; (L) offers the most straightforward one. To illustrate this with the sort of case we have been discussing: Nunberg (1993) offers compelling examples of indexicals—​‘I am traditionally allowed to order whatever I like for my last meal’ uttered by a condemned prisoner—​ that do not seem to contribute objects to what is communicated, but rather properties.Why are these cases irrelevant to the semantic content of the relevant sentences, if indeed they are? Why do writers like Nowak who reject (L) nonetheless assume that characters for demonstratives assign them objects in context?32 What do they mean when they give an important theoretical role in their accounts to a notion of literal content (Harris 2017, 338; forthcoming)? It seems that to answer such questions we need locuted contents along the lines of (3), not just (2).33 There are two proposals to keep to (L) for indexicals and demonstratives, i.e., for their (utterance-​) metasemantics: an intentionalist view, and a conventionalist account; Kaplan (1989a) first endorsed the 183

Manuel García-Carpintero

latter, switching later to the former (Kaplan 1989b). Conventionalists hold that the semantic value of a demonstrative in context is given by cues in the conversational context, paradigmatically pointing gestures, but more in general salience or attentional prominence.34 Wettstein (1984), Reimer (1991), Gauker (2008, 2018), and Stojnić, Stone, and Lepore (2013, 2017) defend such views. Intentionalists appeal instead to what speakers have in mind, along lines suggested by Donnellan (1970) for proper names. Intentionalists base their view on cases in which the demonstrative succeeds in referring without any apparent cues (cf. Speaks 2016). We will now examine how these issues replicate when it comes to the semantic significance of force-​indicators, which will help presenting in section 4 my Austinian take on these issues. As it will transpire, the view has both intentionalist and conventionalist features, although it strongly verges towards the latter.

3  Force-​Indicators: Semantics and Metasemantics The declarative, interrogative, and imperative moods are identified by morphosyntactic paradigms and functional roles (König and Siemund, 2007, 282–​284).35 They are universal in human languages (Roberts 2018, 319). There are distinctive marks of the three moods under embeddings (‘John said that he was happy’, ‘John wondered whether he was happy’, ‘John told him to be happy’, Roberts 2018, 321). Conjunctions, disjunctions, or conditionals mix them (Murray and Starr forthcoming, §2.3).These are initial reasons to think that the distinction has compositional significance, and hence that each of the three moods make a distinctive contribution to locutions (Pendlebury 1986; Chierchia and McConnell-​Ginet 1990, 4.3). On a currently popular view, sentences of each of the three types are assigned three allegedly different types of semantic object:  a proposition to declaratives, a set of sets of propositions to interrogatives (encoding the class of possible answers), and a “to-​do list” for conversational participants to imperatives (Roberts 2018, 320; Portner 2018, 180–​181). The determination of the actual illocutionary force of an utterance of a sentence of any of these types is then left to the “postsemantics”, given a force-​linking principle that connects their distinctive semantic values with “default”, “typical” speech acts made with them (Roberts 2018, 320–​321, 327, 349). The principle is a particular application of (L) to the case at hand. The three default speech acts are characterized by means of a generalization of Stalnaker’s (1978) influential account of assertion and presupposition. On Stalnaker’s view, contexts are understood as a common ground of accepted propositions. The proposal generalizes this, including in addition questions under discussion (the questions mutually adopted for the conversation), and “to-​do lists” for the participants in the conversation—​for each agent, the properties it is mutually assumed that the agent is to make true. The three default speech acts are then understood as proposals to update the relevant “parts” of such contexts. I have argued (García-​Carpintero forthcoming-​a) that this “force-​ linking” principle is not adequate, for reasons like those mentioned in the above discussion of the Bach-​Harris-​Nowak view of demonstratives. The declarative can be used to make many different specific speech acts, including guesses and suppositions, for instance.The imperative can be used for commands, requests, invitations, to give permission, and so on. As Chierchia and McConnell-​Ginet (1990, 173)  contend, there is nothing nonliteral or indirect about these uses. Moreover, the compositional grammar appears to serve to indicate them, in English, Spanish, and other languages by means of appositive hedges like ‘I guess’, ‘I suppose’, or ‘I beg you’ (Benton and van Elswyk 2019); and these hedges appear to be semantic counterparts of evidentials in languages that have them, in which they clearly are part of the compositional machinery (Murray and Starr forthcoming, §2.3). The Portner-​Roberts view of moods thus fails to meet (L), and hence to explain how the semantics fixes the literal, conventional meaning of utterances of sentences in context. The semantics should provide more specific constraints on the objects that are the values of characters in context. The semantic objects ascribed to force-​indicators should identify specific forces conveyed by literal, 184

Reference-Fixing and Presuppositions

conventional utterances. This is what writers in the speech-​act tradition have been offering, either in a Gricean, descriptive-​psychological setting (Bach and Harnish 1979) or in an Austinian, normative one (Alston 2000; Austin 1962; Searle 1969). In a popular recent development, Williamson (1996)—​like Roberts—​also suggests that the universal moods have “default” meanings: “In natural language, the default use of declarative sentences is to make assertions” (Williamson 1996, 258). However, Williamson argues that there is a very specific speech act that is default-​conveyed; he refers to it as flat-​out assertion (1996, 246), and he offers an account that distinguishes it from other specific speech acts that we also make in conventional, literal uses of declarative sentences, like guesses, swears, or suppositions. The account is provided in the Rawlsian tradition of constitutive rules (García-​Carpintero 2004).Williamson claims that the following norm or rule (the knowledge rule) is constitutive of assertion, and individuates it: (KR)

One must ((assert p) only if one knows p).36

In the course of the debate that this proposal has generated, other writers have accepted the view that (flat-​out) assertion is defined by constitutive rules, but have proposed alternative norms; thus, Weiner (2005) proposes a truth rule (TR), Lackey (2007) a reasonableness rule (RBR), and I myself (García-​ Carpintero 2004, 2018b) a knowledge provision rule (KPR): (TR) One must ((assert p) only if p). (RBR) One must ((assert p) only if it is reasonable for one to believe p). (KPR) One must ((assert p) only if one’s audience is put thereby in a position to know p). The obligations these rules impose are sui generis, like those constitutive of games, the model on which Williamson bases his account: they do not have their source in norms of morality, rationality, prudence, or etiquette.They are not all things considered, but pro tanto; in any particular case, they can be overruled by stronger obligations imposed by other norms. They are intended to characterize what is essential or constitutive of assertion (and not, as it may seem at first glance, of correct assertion). The view is that assertion is an act essentially constituted by its being subject to the relevant norm. On Williamson’s view, assertion is the unique representational act such that, in performing it, one is committed to knowing the represented proposition; i.e., the propositional act such that, if one performs it without knowing the intended proposition, one is thereby contravening an obligation. There are additional features or rules contributing to a full characterization of assertion, as in Searle’s (1969) well-​known account or in Alston’s (2000) elaboration, i.e., the “sincerity” or “preparatory” conditions.The rules are intended to characterize what an act must “count as” for it to be an assertion, i.e., what Searle describes as its “essential rule”. As said, it is common ground among participants in these debates that assertion is what is done by default (i.e., unless conditions in an open-​ended list apply, such as those creating irony, fiction, or the presence of canceling parenthetical remarks such as ‘I conjecture’, etc.) by uttering declarative sentences. This gives us a pre-​theoretical handle on the phenomenon that we aim to characterize, and hence an independent standpoint from which to appraise them: it is the act, whatever its proper definition is, that is in fact associated with the indicative mood in natural languages as used on some occasions (the default, “flat-​out” cases), and which speakers intentionally purport to make by such means on such occasions. Intuitively, these are cases in which we mean to be taken “at our word”; i.e., cases in which we aim to impart information merely on the basis of our saying-​so. We thus present ourselves as believing what is said, and as assuming that our audience might be interested in the information. Answering a request for information, reporting on our day to our family, or reporting current events in a newspaper would thus be such default uses. We can similarly think of corresponding default uses for the imperative (say, cases in which someone with authority tells somebody under her authority to do something) and the interrogative (requests for flat-​out assertions, cases in which somebody asks for information). These are precisely 185

Manuel García-Carpintero

the sort of circumstances proffered to give the functional roles of the three moods in standard typological research (König and Siemund 2007, 282–​284). Gutzmann (2015) offers a revision of Potts’s (2005) multidimensional semantics, with the goal of providing a compositional formulation of Kaplan’s (1999) ideas on expletives and pejoratives such as ‘damn’.The idea is that a literal utterance of a sentence like ‘that damn Kaplan was promoted’ contributes to “at issue” locuted content the non-​evaluative proposition that Kaplan was promoted, but it also conventionally contributes a condition on felicitous use: that the utterance context is such that the speaker has a derogatory attitude regarding Kaplan. In addition to preserving compositionality,37 Gutzmann’s system has the great virtue of allowing for the semantic interaction between the two planes.38 In my view, Gutzmann’s proposal is not, however, fully adequate. In the first place, Gutzmann offers Gricean psychological conditions to characterize forces, inadequate to capture the proper normative conditions on the felicitous use of the relevant sentences. For a declarative like ‘Homer is bald’, for instance, Gutzmann (2015, 203)  has the use condition that the speaker wants the hearer to know that Homer is bald. However, a literal utterance of that sentence can be perfectly felicitous even if the speaker doesn’t care in the least whether the hearer comes to know that Homer is bald, and it can be infelicitous even if the speaker does want that (say, if he lacks the knowledge).39 Use-​conditions should rather concern a normative conversational scoreboard of the kind envisaged by Lewis (1979) and Thomason (1990). The default use condition for declaratives is that in felicitous uses the speaker meets the specific norm that defines the illocutionary force of flat-​out assertion—​in my view (KPR). Hedges like ‘I guess’ or ‘I think’ alter this, weakening the commitment. Second, Gutzmann’s framework doesn’t handle presuppositions, but it is essential for my proposals here that we preserve the benefits on this score of dynamic approaches. A  distinctive part of the scoreboard should gather the propositions to whose truth participants are committed as a result of previous utterances accepted as felicitous (same for shared questions under discussion or shared plans, García-​Carpintero 2015). Accepted declaratives (‘Homer is bald’) update this scoreboard constituent; presuppositions down the conversational line may then felicitously target it (‘Homer regrets that he is bald’). I will have to just assume that a fully adequate formal system can be developed, along the lines just outlined. Let’s now see how to account on that basis with semantic reference and reference-​fixing.

4  Semantic Reference in an Austinian Framework The universality of the three sentential moods suggests that the kinds they conventionally indicate are “natural”. Here ‘natural’ is not used in opposition to ‘social’, but rather to refer to properties and kinds in Lewis’s (1983) “sparse” (as opposed to “abundant”) sense. Illocutionary types such as assertion (like games) are “social constructs”, definable by social rules (García-​Carpintero forthcoming-​b). They have “Platonic” essences, in the terminology of Newman and Knobe’s (2019) generalized essentialism, as opposed to “causal” essences like that of water. Natural properties and kinds are those that play substantive explanatory roles, and hence have a “hidden nature” which only reveals itself through inquiry. The commonly known stereotypical features that allow us to pick up the kind are (in the case of flat-​out assertions) those mentioned above: they are utterances whose speakers present themselves as meaning to be taken at their word.The explanatory proposal I favor is that they are “tellings” (Fricker 2017): occasions in which speakers come to be beholden to the (KPR) norm. Tellings—​ flat-​out assertions—​are conventionally default-​indicated by the declarative mood, as explained, but they could be made by other means; for instance, indirectly—​by gestures, rhetorical questions, irony, fiction-​making, or implicatures (García-​Carpintero 2019a). Griceans need not disagree with any of this: they accept that moods conventionally convey illocutionary types, which are social natural kinds that can equally be conveyed in other ways. The disagreement concerns only the nature of the kind in question—​a psychological one for Griceans, normative for Austinians. All these points carry over to reference, which, like Searle (1969, ch. 4) and 186

Reference-Fixing and Presuppositions

Strawson (1950), I’ll treat as an ancillary speech act. Schiffer (1981) makes a proposal—​endorsed and elaborated by Neale (2004, 2016)—​of which my own is a natural Austinian counterpart, as I’ll now explain. In previous work (García-​Carpintero, 2019b) I have argued that to understand how kinds whose essence is given by constitutive rules impose their defining obligations, we need to assume a further, as it were “external” normativity. I have used this point to explain why Williamson’s (1996, 239) argument that such kinds cannot be conventional is not compelling. Williamson overlooks the fact that only normative kinds that are in force are really normative; and it might be by convention that they come to be so. Many kinds defined by constitutive norms are not in force, and hence are not really normative: they don’t give anybody a reason to act.40 Think of variations we can concoct on the rules that define actual games, and the “possible” games they define.They define putative kinds; but, if they are not in force, nobody is really obligated by them, and hence they are not truly normative. What would make such merely putative rule-​constituted kinds “actual” in a given community, practices whose constitutive norms players are beholden to? What this requires, I suggested, might just be a convention or agreement; more generally, this is determined by social norms (Bicchieri 2006), whether or not they are conventions. Applying this to our case, it is in virtue of such social norms that mood-​indicating devices like the three clause types—​declarative, interrogative, and imperative—​ get associated with kinds defined by constitutive rules, respectively (flat-​out) assertion, question, and command; and this is a central procedure through which the relevant normative kinds come to be in force.41 On these grounds, the Austinian view helps itself for metasemantic purposes to a version of the knowledge-​based Principle of Charity that Williamson (2007, 264)  promotes, avoiding its pitfalls (McGlynn 2012, 398–​400; García-​Carpintero 2018a, 1125). The social norms that explain the conventional association of the declarative clause type and the constitutive norms defining flat-​out assertion in fact afford a teleological explanation of why the norm is in force. Such a teleological explanation comes with a specification of “normal functioning”: the way in which past tokens of the type functioned when the type acquired its function. Of course, it is not just the illocutionary force that was conveyed in such cases: it came together with particular contents. Cases of normal functioning are those in which the norm was observed, and hence the content was known and therefore true. These contents were compositionally determined. Our present concern has to do with the systematic truth-​conditional contribution of the expressive devices that linguists mark with the +ref feature (Glanzberg 2018b) like indexicals and names, whose job is to conventionally indicate what a given discourse is about. Schiffer (1981) argues that what is distinctive of referential expressions is that they conventionally contribute to locutions whose contents are singular propositions—​singular in particular with respect to their compositional truth-​conditional contribution. It may be thought that the minimalist view of propositional contents endorsed before prevents us from capturing the relevant singularity, but this is not the case. As Russell (1903, 316) pointed out, even attributively used descriptions make truth-​conditional contributions that intuitively are “about” the things that satisfy them. However, unlike them, conventional referential expressions convey this in a “direct” way (García-​Carpintero and Palmira n.d.). This intuition of “directness” can be spelled out firstly in terms of the medieval intuitions that Quine (1956) invokes to distinguish attitudes directed at particular sloops from attitudes merely directed, say, towards relief from slooplessness, and secondly by Kripkean intuitions of rigidity (Quine 1956). Neale (2016) and Schiffer (2016) set into relief the well-​known difficulties that Gricean intentionalist views have to confront to properly develop Schiffer’s just outlined account. Neale focuses on an especially problematic issue, that of “aphonics”—​the unpronounced referential “expressions” that current semantics pose:  PRO, movement traces, indicators for domain of individuals or worlds, scales for quantities and degrees in them, and so on. I’ll say below how the Austinian account handles the issue, but I’ll present it first. 187

Manuel García-Carpintero

As indicated above, I take it as well established that some expressions in natural language semantically trigger presuppositions, and hence that the locutions they help to convey specify them. For instance, descriptions like ‘the King of France’ conventionally trigger the presupposition that there is a unique King of France. In previous work (García-​Carpintero 2018b), I have argued that these linguistic presuppositions should be viewed as ancillary speech acts of their own, defined in “flat-​out” uses by a common knowledge rule (PR): (PR)

One must ((presuppose p in a context C) only if p is common knowledge in C).

In the Gutzmann-​inspired framework suggested above, such norms would be additional conditions on the felicitous use of sentences including their triggers. Note that we should sharply distinguish descriptions of how conversations involving these expressions proceed from the normative account—​even though, of course, if correct the account should contribute to explaining it. Even in the case of communication involving only exchanges of flat-​out assertions (the central case for the empirical validation of our proposals, on the teleological account just suggested), conversations can proceed smoothly when presuppositions are infelicitous, and when they are known to be so, as Donnellan’s (1966, 14) famous ‘King in his counting house’ example witnesses:42 the presuppositions in those cases are said to be commonly “accepted”, even if they are not believed and fail to be knowledge. There is also the phenomenon of accommodation (I can impart information with ‘the king is in his counting house’ to someone who doesn’t know that there is a king) and there are exchanges of other kinds (say, irony, fiction), not bound by the norms for flat-​out uses of the moods. García-​Carpintero (2018b) discusses these issues. Referential expressions are a particular kind of presupposition trigger on the account I have been defending.They carry presuppositions of acquaintance or familiarity (García-​Carpintero 2000; cp. Roberts 2002; Cumming 2014; Maier 2010, 2016).What this comes to is, I suggest, that they trigger the application of norm (RR) below—​which I take to be a specification of (PR) for referential expressions—​ as a felicity condition on their uses: (RR) One must ((refer to o by means of e in making an act thereby about o) only if it is commonly known who or what o is, for such representational purposes). (RR) acknowledges the need to relativize knowing-​which/​who ascriptions to certain interests, which Boër and Lycan (1986) pointed out.43 The interest at stake is that of conveying a specific singular content. Now, descriptive accounts in the semantic tradition started by Karttunen (1976) add to contexts’ abstract discourse referents, which are then invoked to characterize presuppositions of familiarity in semantic accounts closely related to mine (cf., e.g., Roberts 2002; Cumming 2014; Maier 2010, 2016). Such discourse referents correspond to pieces of information that would prima facie allow for the norm (RR) to be met. They are conditions presumed to pick out individuals such as: playing the relevant utterance role (being its speaker, hearer, time, day, location) in the case of indexicals (‘I’, ‘you’, ‘here’, ‘now’, ‘today’); being explicitly demonstrated or otherwise perceptually salient, plus fulfilling the relevant φ-​features (being proximal or distal, male or female, plural or singular, and so on), in the case of deictic uses of demonstratives and pronouns (‘this’, ‘that’, ‘he’, ‘she’); meeting the relevant specifications laid down in previous discourse, plus perhaps being had in mind by the speaker, in the case of anaphoric uses of the same expressions; being picked out by the relevant naming-​practice, in the case of names.44 Discourse referents—​conditions or pieces of information—​are usually understood to be parts of attitudinal Stalnakerian contexts. In sync with the Austinian view, I instead think of them as further presuppositional constituents of the Lewisian scoreboard, beholden to the condition that they do pick out unique referents. The reader should keep in mind the remark above on the difference between 188

Reference-Fixing and Presuppositions

descriptions of conversations, and the normative account. In the relevant cases (discourses involving flat-​out assertions), if the conditions in question fail to pick out a unique object, the locuted content is “gappy” and the assertion cannot be evaluated as true or false. Nonetheless, there are cases in which it will be “accepted” that the discourse referent picks out something, and the conversation will proceed smoothly. And there are other cases, like those mentioned above: irony, fiction, and so on. Finally, the previous remarks about assertion apply also to reference. What I have been characterizing is something like Kripke’s (1977) semantic reference: reference contributing to locuted contents, made by means of expressions conventionally designed for that purpose. But the account leaves open that there are further cases, including Kripke’s speaker’s reference. In his much-​discussed account of de jure coreference, Fine (2007, 50)  speaks of “semantic requirements”. If we think of such requirements along the lines of Broome’s (2014) for related talk of rationality requirements, we conceive their sources—​semantics and rationality, respectively—​as particular codes, like legal regulations. This is what the Austinian account I have outlined suggests. The locutionary meaning of ‘The King of France is bald’, say, includes as normative use-​conditions the presence in the scoreboard of the discourse referent that a referent for ‘France’ is picked out by a particular naming-​practice; the satisfaction of the presuppositional requirement for it to be common knowledge that there is a unique king of that place; and the illocutionary requirement that the speaker knows that he is bald. In a literal use, these norms will be actually in force, participants being committed to them. It is in this clear-​cut sense that meanings (locutions) are speech-​act potentials. For norms to be in force in a community, there must be a “collective acceptance” of them in it (Kutz 2000). Norms should somehow “guide” the behavior of those bound by them, and hence speakers should agree to be committed to the ones advocated here.45 Fine (2007) does assume that semantic requirements are known by speakers with full understanding of the language. At least fully competent speakers should thus intend, say, the indexicals they use to refer to the object picked out by the relevant discourse referent. Intentionalism and conventionalism are hence not really incompatible; conventionalists should at most deny that intentions have a distinctive role to play in metasemantics. Lepore and Stone (2015, 200–​220) indeed argue that—​as Keiser (2018, 148)  and Simons (2018, 285) aptly put it—​the only intention that plays a role in meaning-​determination is the generic one to contribute the grammatically specified meaning of the utterance. However, this goes beyond what the outlined Austinian proposal requires. In their rejection of the Gricean intentionalist picture, Lepore and Stone go as far as to reject that there are conversational implicatures, speaker’s reference, and indirect meanings, whose metasemantics distinctively depends on speakers’ intentions. This is unwarranted, as I  have argued (García-​Carpintero 2016). There is no reason why the conventions determining locutions cannot rely on speakers’ intentions, at least to the extent that they are properly made manifest to conversational participants—​as Griceans have traditionally insisted, by relying on some form of the well-​established belief constraints on rational intentions (cf. Neale 2016, 275–​278). As Heck (2014, 336–​343) and Mount (2015, 11–​12) point out, if they do not ensue from referential communicative intentions, contextual cues such as pointing gestures, etc., do not determine referents; and the enormous variety and open-​endedness of “cues”, or forms of “salience” or “attentional prominence” by means of which speakers manage to commit themselves to singular claims by using demonstratives doesn’t fit well with the generality of norms, conventions, or rules. Furthermore, by allowing for semantic contributions whose metasemantics depends on specific speakers’ intentions, we make room for adequate explanations of lexical innovations compatible with sensible forms of conventionalism (Armstrong 2016), and more in general for explanations of how semantic-​determining conventions come to be in place (Carston 2016; Harris 2016, §6). Nonetheless, I  think that Lepore and Stone (2015) are right to point out—​and have made a very good case—​that the conventional resources constituting natural languages go well beyond what is sometimes presumed, including prosody, intonation, discourse coherence, narrative structure, 189

Manuel García-Carpintero

systematic polysemy, and so on.46 The Austinian picture I have been outlining here is entirely compatible with all these being resources that allow for the expression of semantic meanings. While Harris (forthcoming) relies on a psychological natural kind (the I-​language module) to trace the semantics/​ pragmatics divide, the Austinian view instead trusts social kinds—​on the proposal I just made, flat-​out assertions (tellings) and the form of the Principle of Charity that depends on their being in force. If consistently taken, Harris’s proposal leads to the meager views of semantics of the sort advocated by true Chomskians like Pietroski (2018)—​a direction that, sensibly, Harris is not prepared to take. I don’t need to deny that there are interesting explanatory distinctions among meanings that are just determined by the “language faculty”, and those that require more “pragmatic-​looking” resources like coherence relations (Pagin 2014); but on the Austinian view I have been promoting, this is compatible with their being distinctions in the class of semantic content. Although intentions thus may also play a distinctive role in metasemantics on the form of conventionalism I am advocating, their job is still subsidiary in contrast with the one they play in Gricean proposals. I lack the space here to go into any detail, but this alleviates well-​known objections to bare intentionalist accounts. In general, facts about what it takes to be intentionally subject to a code may help with the case of non-​fully competent speakers, including children. More specifically, consider the locution “refer to o by means of e in making an act thereby about o” in (RR) above. Neale (2016) provides a good explanation why we need it to characterize the semantic role of expressions, and the exchange between Neale (2016) and Schiffer (2016) manifest the difficulties that “aphonics” (unpronounced referential “expressions” contributing to semantic, truth-​conditional content) create. Bach (2017) emphasizes the problem such “encoding intentions” pose for giving intentions any role in the determination of semantic content.Viebahn (2018), however, offers good replies, clearly available I think given the role of intentions in the present account.

Conclusion In this paper I have articulated a version of the Austinian conception of metasemantics on which the semantic meanings of sentences are speech-​act potentials. On the suggested view they are complex, including different layers of propositional contents presented with different quite specific illocutionary forces. They are fit to be the meanings (including forces) to which speakers intuitively commit in literal uses. The proposal supports a reply to the invocation of the semantics/​ metasemantics divide by anti-​descriptivist philosophers to relegate to the later role all descriptive reference-​fixing material: presuppositions of acquaintance, those related to φ-​features and “being-​ named” metalinguistic descriptions in the case of names remain semantic, even though they don’t belong in “at-​issue” semantic content. Such semantic descriptive reference-​fixing material may be judged “weak and trivial” (Gómez-​Torrente 2019, 17), but it nonetheless shows that fully-​ fledged anti-​descriptivism, Millianism in particular, is wrong, which is philosophically significant. I have referenced a considerable amount of recent work on the semantics of names and indexicals, including their semantic contributions to attitude-​ascriptions that supports the proposal. Other than that, it amounts to an overarching account of linguistic reference in the tradition of Strawson (1950; cf. also Justice 2007). That, I submit, should be a crucial piece in a full abductive argument for it.

Acknowledgments Financial support was provided by the DGI, Spanish Government, research project FFI2016-​80588-​ R and FFI2016-​81858-​REDC, the award ICREA Academia for excellence in research, 2018, funded by the Generalitat de Catalunya. Thanks to Brian Ball, Eliot Michaelson, Craige Roberts and the editors for their comments, and to Michael Maudsley for the grammatical revision. 190

Reference-Fixing and Presuppositions

Notes 1 Burgess and Sherman also point this out in their introduction to a recent compilation (Burgess and Sherman 2014, 3). 2 Cf. e.g. Soames (2005, 183), Braun (2018, 216–​218), Gómez-​Torrente (2019, 10–​11). 3 This tradition owns more to the later Wittgenstein’s views than is sometimes acknowledged; cf. Harris and Unnsteinsson (2017). 4 The proposal to be developed cashes out in a particular way the slogan “meaning is normative”, which has of course been the topic of massive controversy. Glüer and Wikforss (2018) and Liebesman (2018) offer a good up-​to-​date picture. García-​Carpintero (2012, 411–​418) has my own take on it. 5 Dummett (1973), García-​Carpintero (2006b), Ninan (2012), Rabern (2013), and Stanley (1997), among others, make related distinctions. 6 Borrowing and generalizing from Kaplan (1989a), although I am not committed thereby to Kaplan’s entire panoply of views on context-​dependence. I mean the term as synonymous with sentence standing meaning (Maitra 2007). 7 Cf. Yalcin (2014, 18–​23) for more details on those explanatory goals. Systematicity concerns the fact that speakers who, say, competently understand ‘John loves Mary’ can equally understand ‘Mary loves John’; productivity, the fact that competent understanding is in principle unbounded: ‘the son of Mary swims’, ‘the son of the son of Mary swims’ … . 8 The utterances taken here into consideration involve the deployment of expressions, linguistic tokens—​ which should deflate concerns about “utterance semantics” (Neale 2016, 235) and still preserve an important theoretical role for them. My previous work on these matters (strongly influenced by Perry (cp. e.g. Perry 2012) emphasizes token-​reflexivity, but I was assuming tokens to be uttering events. 9 Korta and Perry (2011) and Recanati (2013) offer related accounts. 10 See also, for related views, García-​Carpintero (2006b, 43–​47; 2019a), Camp (2007, 208–​212), Recanati (2013, 624–​625). 11 To the extent that the expressions themselves are not part of what is conventionally meant; I am here putting aside issues raised here by token-​reflexivity, considered below; cf. Braun (2018) and Radulescu (2018a, 2018b) for good discussions. 12 My distinction here corresponds to the one between sentence force and utterance force by Chierchia and McConnell-​Ginet (1990, 171) and Murray and Starr (forthcoming, §1). 13 Cf. Davis (1999, 35 n), Unnsteinsson (2017). Unnsteinsson makes the very good point that, due to structural or lexical ambiguity, the locuted content will be in such cases typically too underspecified for the speaker to assertorically commit to it. In our example, nothing in the context rules out the interpretation that the vote lacked “individuality, distinction, or recognizability”, which is one of the meanings of ‘anonymous’. 14 Unnsteinsson (2017) claims that the speaker said (in a Gricean sense) that the vote was unanimous. This is consistent with the view advocated here (cf. García-​Carpintero 2019a). 15 Perhaps Cappelen and Lepore (2005, 57) are an exception; Travis is also a doubtful case (Fisher 2019, n 24). Unlike Bach and Harnish (1979, 70), I don’t distinguish between ‘literal’ and ‘direct’—​I consider that distinction otiose. 16 The first features a person who writes just “his command of English is excellent, and his attendance at tutorials has been regular” as a letter of reference for a pupil’s application for a philosophy job; the second, a man who, when speaking to someone who has run out of petrol, says “There is a garage around the corner”, thereby implicating that the garage is open and selling petrol (Grice 1975, 32). 17 Cf. Richard (2013). 18 “Smaller” not just spatiotemporally (say, events in our light cone), but also at the level of detail (say, domain of individuals, relevant features, etc.) at which events are specified. 19 Crucially for the present proposal, in a descriptively adequate framework the locution should also include another, not-​at-​issue propositional content, specifying a presupposition triggered by the pronoun including the φ-​feature that the individual in question is male (Heim 2008; García-​Carpintero 2000) ( see section 4). Cf. Del Prete and Zucchi (2017) for a sophisticated formal proposal compatible with the view developed here; cf. also Maier (2010, 2016). 20 See the works referenced in the previous note, and others referenced in them. 21 Larson and Segal (1996) is an excellent text-​book introduction to Davidsonian truth-​conditional semantics; Heim and Kratzer (1998) provides the standard text-​book introduction to the ideas I am about to summarize. 22 Cf. Heim and Kratzer (1998), Del Prete and Zucchi (2017), Maier (2010, 2016), and Hunter (2013) for more accurate accounts. 23 Cp. Kaplan’s (1989a, 525 n) nice metaphor that the semantic referent of a demonstrative is whoever appears in the ‘demonstration platform’.

191

Manuel García-Carpintero 24 The lambda operator is a formal procedure to construct complex predicates; for (2), being such that, on the presumption that it is the “demonstrated” male, it is happy. 25 Cf. Armstrong (2016, 91). Recanati’s (2018, 118)  Determination Thesis and Kennedy’s (2007, 36)  principle Interpretive Economy are in the same spirit. 26 Cf. also Smit (2012, 63–​64) and Speaks (2017).The early pages of Heck’s (2014) go in the same direction, but he seems to me overall closer to endorse some version of (L). 27 Cf. also Pagin and Westerståhl (2010). Philosophers who assume (L) also have reasons to invoke this notion of compositionality, cf. Glanzberg and King (forthcoming, n 40). 28 Yalcin (2014), Glanzberg (2018a), and Pietroski (2018) adopt a more radical view, close to Chomsky’s take on these issues, by making semantic contents even thinner than (2). 29 Cf. Camp’s (2007) and Maitra’s (2007) discussion of Cappelen and Lepore (2005), and García-​Carpintero’s (2013) of Borg (2012). 30 Cf. Stanley and Szabó (2000), Glanzberg (2007). 31 Cf. Stojnić (2017, 169), García-​Carpintero (2006b, 59, 62). Rabern (2017) appears to reduce semantic content to character content. He argues for the minimalist suggestion that contents like (2)—​or rather corresponding “diagonal” contents—​are what we literally convey, so that (L) is after all satisfied (2017, 196 n, 202–​205; cf. also García-​Carpintero 2006b, 56, 64). The determination of contents like (3) intuitively closer to what is literally communicated is left to “postsemantics”, i.e., pragmatics. I find this problematic, as suggested in the main text. 32 Note that King (2001) argues for a quantificational account of complex demonstratives; cf. Hawthorne and Manley (2012) for a related, more general view. I am just assuming the standard Kaplanian view for my purposes. 33 Heck (2014, 355) puts it nicely: “the most natural explication of the notion of what a sentence means is in terms of how it constrains the linguistic acts that can be performed by uttering it, and … these acts must be characterized not in terms of what is meant but in terms of what is said”. (It is passages like this that suggest to me the interpretation in note 26.) 34 Conventionalists hold something like Neale’s (2016, 229) thesis (TCS) in support of (L). Intentionalists like King also endorse a version of it in which speaker’s intentions play a decisive role; as said, contextualists like Bach and Neale just reject (L). 35 Speech acts like assertions can be made with subsentential utterances (Stainton 1995). They can either be regarded as elliptical for a full declarative (Merchant 2004), or pragmatically enriched into the content of one along the lines that Stainton suggests. 36 The internal parentheses mark the scope of the asserted content. The external ones properly delimit the scope of the obligation. If one asserts p when one doesn’t know p, one violates the constitutive rule of what one does. If the scope of the obligation were limited to the antecedent, it would be permissible to assert p when one doesn’t know it. 37 Understood in the sense that I think is needed for the explanatory purposes for which the notion is deployed, cf. the references considered in note 27 and related main text above. 38 I find also congenial the sort of architecture outlined by Geurts (2019), although, for reasons indicated above, Geurts’s blanket appeal to commitments should be made more specific, distinguishing the specific commitments resulting from different speech acts. 39 Cf.Vlach (1981), Alston (2000, ch. 2), Green (2007, ch. 3). 40 Cf. Broome (2014, ch. 2) for a Scanlon-​inspired sharp discussion of the “core” normative notions that I am assuming here. 41 This doesn’t mean that the assertoric practice itself (as opposed to its expression by the mood) is conventional; in fact I have suggested it isn’t (García-​Carpintero 2019b). 42 I think that a usurper occupies the throne, but his minions take him to be the legitimate king; wanting to see him, I say to one of them, ‘Is the king in his counting house’? (PR) is violated, because I fail to even believe that the guy is the one and only king. 43 We don’t need to discuss here whether this is just a pragmatic effect (cf. Braun 2006). 44 Cf. García-​Carpintero (2018a) for elaboration, and for a critical comparison of the view assumed here and currently popular “predicativist” views. The previous issues about prima facie violations of the rule in cases of accommodation, acceptance, and so on arise also for (RR). Consider thus the example by Michaelson that Speaks (2016, 328) discusses, “we are sitting on a couch in my house, and I think I see something quite surprising—​like a bird quickly flying past the doorway. You do not flinch, so I am almost sure that you did not see what I think I saw. But, to be sure, I might ask you: ‘Did you see that?’ ” I trust the general line I mentioned before works also here. Thanks to Eliot Michaelson and the editors. 45 Gómez-​ Torrente (2019) provides compelling considerations against proposals for reference-​ fixing (intentionalists and conventionalists) on the grounds that they don’t introduce “suitable indeterminacies”

192

Reference-Fixing and Presuppositions (2019, 45); the suitability at stake is imposed by a condition that reference-​fixing rules should be known by competent speakers (2019, 46). The Austinian account can, I think, accommodate the requirement and resulting indeterminacy in its proper light. 46 Del Pinal (2018) provides a multiple-​level semantics focusing on polysemy that, like Gutzmann’s, I take to be highly congenial to my proposals here. Del Pinal doesn’t mention the case of moods, but in fact I take the difference between using declaratives to convey flat-​out assertions and using them to convey, say, guesses as a case of polysemy.

References Alston, William (1964): “Linguistic Acts”, American Philosophical Quarterly 1(2), 138–​146. Alston, William (2000): Illocutionary Acts and Sentence Meaning, Ithaca, NY: Cornell University Press. Armstrong, Josh (2016): “The Problem of Lexical Innovation”, Linguistics and Philosophy 39, 87–​118. Austin, John (1962): How to Do Things with Words, Oxford: Clarendon Press. Bach, Kent (1994): “Conversational Implicitures”, Mind and Language 9, 124–​162. Bach, Kent (2012):  “Context Dependence”, in M. García-​Carpintero and M. Kölbel (eds.), The Continuum Companion to the Philosophy of Language, London: Continuum, 153–​184. Bach, Kent (2017): “Reference, Intention, and Context: Do Demonstratives Really Refer?”, in M. de Ponte and K. Korta (eds.), Reference and Representation in Thought and Language, Oxford:  Oxford University Press,  57–​72. Bach, K. and Harnish, R. M. (1979): Linguistic Communication and Speech Acts, Cambridge, MA: MIT Press. Ball, Brian (forthcoming): “Speech Acts, Actions, and Events”, in P. Stalmaszczyk (ed.), The Cambridge Handbook of the Philosophy of Language, Cambridge: Cambridge University Press. Benton, M. and van Elswyk, P. (2019): “Hedged Assertion”, in S. Goldberg (ed.), Oxford Handbook of Assertion, Oxford: Oxford University Press, DOI: 10.1093/​oxfordhb/​9780190675233.013.11. Bicchieri, Cristina (2006): The Grammar of Society: The Nature and Dynamics of Social Norms, Cambridge: Cambridge University Press. Boër, S. E. and Lycan, W. G. (1986): Knowing Who, Cambridge, MA: MIT Press. Borg, Emma (2012): Pursuing Meaning, Oxford: Oxford University Press. Braun, David (2006): “Now You Know Who Hong Oak Yun Is”, Philosophical Issues 16, 24–​42. Braun, David (2018): “Utterances and Expressions in Semantics and Logic”, in Ken Turner and Laurence Horn (eds.), Pragmatics,Truth and Underspecification: Towards an Atlas of Meaning, Leiden: Brill, 182–​258. Broome, John (2014): Rationality through Reasoning, Oxford: Wiley Blackwell. Burgess,Alexis and Sherman, Brett (2014): “Introduction”, in A. Burgess and B. Sherman (eds.), Metasemantics: New Essays on the Foundations of Meaning, Oxford: Oxford University Press, 1–​16. Camp, Elizabeth (2006): “Contextualism, Metaphor, and What is Said”, Mind and Language 129, 280–​309. Camp, Elizabeth (2007): “Prudent Semantics Meets Wanton Speech Act Pluralism”, in G. Preyer and G. Peter (eds.), Context-​Sensitivity and Semantic Minimalism, Oxford: Oxford University Press, 194–​213. Cappelen, H. and Lepore, E. (2005): Insensitive Semantics, Oxford: Blackwell. Carston, Robyn (2002): Thoughts and Utterances, Oxford: Blackwell. Carston, Robyn (2016): “Linguistic Conventions and the Role of Pragmatics”, Mind and Language 31(5), 612–​624. Charlow, Nate (2014): “The Meaning of Imperatives”, Philosophy Compass 9(8), 540–​555. Chierchia, G. and McConnell-​Ginet, S. (1990): Meaning and Grammar, Cambridge, MA: MIT Press. Collins, John (2018): “The Redundancy of the Act”, Synthese 195, 3519–​3545. Cumming, Sam (2014): “Discourse Content”, in A. Burgess and B. Sherman (eds.), Metasemantics: New Essays on the Foundations of Meaning, Oxford: Oxford University Press, 214–​230. Davis, Wayne A. (1999): “Communicating, Telling, and Informing”, Philosophical Inquiry 21,  21–​43. Davis, Wayne A. (2003): Meaning, Expression, and Thought, Cambridge: Cambridge University Press. Del Pinal, Guillermo (2018): “Meaning, Modulation, and Context: A Multidimensional Semantics for Truth-​ Conditional Pragmatics”, Linguistics and Philosophy 41, 165–​207. Del Prete, F. and Zucchi, S. (2017): “A Unified Non-​monstrous Semantics for Third-​person Pronouns”, Semantics and Pragmatics 10(10), DOI: 10.3765/​sp.10.10. Donnellan, Keith (1966): “Reference and Definite Descriptions”, Philosophical Review 75, 281–​304. Included in J. Almog and P. Leonardi (eds.), Essays on Reference, Language, and Mind, Oxford: Oxford University Press, 2012, 3–​30, to which I refer. Donnellan, Keith (1970): “Proper Names and Identifying Descriptions”, Synthese 21, 335–​358. Dummett, Michael (1973): Frege: Philosophy of Language; 2nd edition, 1981, London: Duckworth. Fine, Kit (2007): Semantic Relationism, Oxford: Blackwell.

193

Manuel García-Carpintero Fine, Kit (2017):  “Truthmaker Semantics”, in B. Hale, C. Wright and A. Miller (eds.), A Companion to the Philosophy of Language, 2nd edition, Oxford: Blackwell, 556–​577. Fisher, Sarah (2019): “Reassessing Truth‑evaluability in the Minimalism‑Contextualism Debate”, Synthese online first, DOI: 10.1007/​s11229-​019-​02245-​2. Fricker, Elizabeth (2017): “Norms, Constitutive and Social, and Assertion”, American Philosophical Quarterly 54(4), 397–​418. García-​Carpintero, Manuel (2000):  “A Presuppositional Account of Reference-​Fixing”, Journal of Philosophy 97(3), 109–​147. García-​Carpintero, Manuel (2004):  “Assertion and the Semantics of Force-​Markers”, in C. Bianchi (ed.), The Semantics/​Pragmatics Distinction, CSLI Lecture Notes, Chicago, IL: The University of Chicago Press, 133–​166. García-​Carpintero, Manuel (2006a):  “Two-​dimensionalism:  A Neo-​Fregean Interpretation”, in M. García-​ Carpintero and J. Macià (eds.), Two-​Dimensional Semantics, Oxford: Oxford University Press, 181–​204. García-​Carpintero, Manuel (2006b): “Recanati on the Semantics/​Pragmatics Distinction”, Crítica 38, 35–​68. García-​Carpintero, Manuel (2008): “Homophonic Prejudices”, Crítica (Mexico), 40, 67–​84. García-​Carpintero, Manuel (2012): “Foundational Semantics, I & II”, Philosophy Compass 7(6), 397–​421. García-​Carpintero, Manuel (2013): “Emma Borg’s Pursuing Meaning”, Mind 122, 515–​521. García-​ Carpintero, Manuel (2015):  “Contexts as Shared Commitments”, Frontiers in Psychology 6, 1932, DOI: 10.3389/​fpsyg.2015.01932. García-​Carpintero, Manuel (2016): “Indirect Assertions”, Polish Journal of Philosophy 10(1), 13–​49. García-​Carpintero, Manuel (2018a): “The Mill-​Frege Theory of Proper Names”, Mind 127(508), 1107–​1168, DOI: 10.1093/​mind/​fzx010. García-​Carpintero, Manuel (2018b): “On the Nature of Presupposition: A Normative Speech Act Account”, Erkenntnis, online first, DOI: 10.1007/​s10670-​018-​0027-​3. García-​Carpintero, Manuel (2019a): “Sneaky Assertions”, Philosophical Perspectives, DOI: 10.1111/​phpe.12116. García-​ Carpintero, Manuel (2019b):  “Conventions and Constitutive Norms”, Journal of Social Ontology, DOI: 10.1515/​jso-​2019-​0013. García-​Carpintero, Manuel (forthcoming-​a):“Metasemantics: A Normative Perspective (and the Case of Mood)”, in P. Stalmaszczyk (ed.), The Cambridge Handbook of the Philosophy of Language, Cambridge:  Cambridge University Press. García-​ Carpintero, Manuel (forthcoming-​ b): “How to Understand Rule-​ Constituted Kinds”, Review of Philosophy and Psychology. García-​ Carpintero, M. and Palmira, M. (n.d.):  “Singular Propositions and Propositional Vehicles”, unpublished ms. Gauker, Christopher (2008): “Zero Tolerance for Pragmatics”, Synthese 165, 359–​371. Gauker, Christopher (2018):  “Against the Speaker-​ Intention Theory of Demonstratives”, Linguistics and Philosophy, DOI: 10.1007/​s10988-​018-​9239-​x. Geurts, Bart (2019): “Communication as Commitment Sharing: Speech Acts, Implicatures, Common Ground”, Theoretical Linguistics 45(1–​2): 1–​30, DOI: 10.1515/​tl-​2019-​0001. Glanzberg, Michael (2007): “Context, Content, and Relativism”, Philosophical Studies 136(1): 1–​29. Glanzberg, Michael (2018a):  “Lexical Meaning, Concepts, and the Metasemantics of Predicates”, in Brian Rabern and Derek Ball (eds.), The Science of Meaning, Oxford: Oxford University Press, 197–​225. Glanzberg, Michael (2018b):  “About Convention and Grammar”, in G. Preyer (ed.), Beyond Semantics and Pragmatics, Oxford: Oxford University Press, 230–​260. Glanzberg, M. and King, J. (forthcoming): “Binding, Compositionality, and Semantic Values”, Philosopher’s Imprint. Glüer, Kathrin and Wikforss, Åsa (2018): “The Normativity of Meaning and Content”, The Stanford Encyclopedia of Philosophy (Spring 2018 edition), Edward N. Zalta (ed.), https://​plato.stanford.edu/​archives/​spr2018/​ entries/​meaning-​normativity/​. Gómez-​Torrente, Mario (2019): Roads to Reference: An Essay on Reference-​fixing in Natural Language, Oxford: Oxford University Press. Green, Mitchell (2007): Self-​Expression, Oxford: Oxford University Press. Grice, H. P. (1975):  “Logic and Conversation”, in P. Cole and J. Morgan (eds.), Syntax and Semantics, vol. 3, New  York:  Academic Press, 41–​58. Included in H. P. Grice, Studies in the Ways of Words, Cambridge, MA: Harvard University Press, 1989, 22–​40, from which I quote. Gutzmann, Daniel (2015): Use-​Conditional Meaning, Oxford: Oxford University Press. Hanks, Peter (2015): Propositional Content, Oxford: Oxford University Press. Harris, Daniel W. (2016):  “Intentionalism vs. the New Conventionalism”, Canadian Journal of Philosophy 47, 173–​201. Harris, Daniel W. (2017):  “Speaker Reference and Cognitive Architecture”, Croatian Journal of Philosophy 17, 319–​349.

194

Reference-Fixing and Presuppositions Harris, Daniel W. (forthcoming): “Semantics without Semantic Content”, Mind and Language. Harris, D.W. and Unnsteinsson, E. (2017): “Wittgenstein’s Influence on Austin’s Philosophy of Language”, British Journal for the History of Philosophy 26(2), 371–​395. Hawthorne, John and Manley, David (2012): The Reference Book, Oxford: Oxford University Press. Heck, Richard G. Jnr. (2014): “Semantics and Context-​Dependence: Towards a Strawsonian Account”, in A. Burgess and B. Sherman (eds.), Metasemantics:  New Essays on the Foundations of Meaning, Oxford:  Oxford University Press, 327–​364. Heim, Irene (2008): “Features on Bound Pronouns”, in D. Harbour, D. Adger and S. Béjar (eds.), Phi Theory, Oxford: Oxford University Press, 35–​56. Heim, I. and Kratzer, A. (1998): Semantics in Generative Grammar, Oxford: Blackwell. Hunter, Julie (2013): “Presuppositional Indexicals”, Journal of Semantics 30(3), 381–​421, DOI: 10.1093/​jos/​ffs013. Justice, John (2007): “Unified Semantics of Singular Terms”, Philosophical Quarterly 57, 363–​373. Kaplan, David (1989a):  “Demonstratives”, in J. Almog, J. Perry and H. Wettstein (eds.), Themes from Kaplan, Oxford: Oxford University Press, 481–​563. Kaplan, David (1989b):  “Afterthoughts”, in J. Almog, J. Perry and H. Wettstein (eds.), Themes from Kaplan, Oxford: Oxford University Press, 565–​614. Kaplan, David (1999): “The Meaning of ‘Ouch’ and ‘Oops’ ”, UCLA, ms. Karttunen, Lauri (1976): “Discourse Referents”, in J. McCawkey (ed.), Syntax and Semantics 7: Notes from the Linguistic Undergrounds, New York: Academic Press, 363–​385. Keiser, Jessica (2018):  “Varieties of Intentionalism”, in G. Preyer (ed.), Beyond Semantics and Pragmatics, Oxford: Oxford University Press, 147–​156. Keller, Lorraine (2013): “The Metaphysics of Propositional Constituency”, Canadian Journal of Philosophy 43(5–​ 6), 655–​678. Kennedy, Christopher (2007):  “Vagueness and Grammar:  The Semantics of Relative and Absolute Gradable Predicates”, Linguistics and Philosophy 30(1), 1–​45. King, Jeffrey (2001): Complex Demonstratives: A Quantificational Account, Cambridge, MA: MIT Press. King, Jeffrey (2014):  “The Metasemantics of Contextual Sensitivity”, in A. Burgess and B. Sherman (eds.), Metasemantics: New Essays on the Foundations of Meaning, Oxford: Oxford University Press, 97–​118. König, E. and Siemund, P. (2007): “Speech Act Distinctions in Grammar”, in T. Shopen (ed.), Language Typology and Semantic Description, Cambridge: Cambridge University Press, 276–​324. Korta, K. and Perry, J. (2011): Critical Pragmatics, Cambridge: Cambridge University Press. Kratzer, Angelika (2012): Modals and Conditionals, Oxford: Oxford University Press. Kripke, Saul (1977): “Speaker’s Reference and Semantic Reference”, in P. French, T. Uehling and H. Wettstein (eds.), Contemporary Perspectives in the Philosophy of Language, Minneapolis, MN: University of Minnesota Press, 255–​276. Kripke, Saul (1980): Naming and Necessity, Cambridge, MA: Harvard University Press. Kutz, Christopher (2000): “Acting Together”, Philosophy and Phenomenological Research 61,  1–​31. Lackey, Jennifer (2007): “Norms of Assertion”, Noûs 41(4), 594–​626. Larson, R. and Segal, G. (1996): Knowledge of Meaning: Semantic Value and Logical Form, Cambridge, MA: MIT Press. Lepore, E. and Stone, M. (2015): Imagination and Convention, Oxford: Oxford University Press. Lewis, David (1975):  “Languages and Language”, in K. Gunderson (ed.), Language, Mind and Knowledge, Minneapolis, MN:  University of Minnesota Press, 3–​35. Also in D. Lewis, Philosophical Papers, vol. 1, Oxford: Oxford University Press, 1983, 163–​188. Lewis, David (1979): “Scorekeeping in a Language Game”, Journal of Philosophical Logic 8, 339–​359. Lewis, David (1980): “Index, Context, and Content”, in S. Kanger and S. Öhman (eds.), Philosophy and Grammar, Dordrecht: Reidel. Also in David Lewis, Papers in Philosophical Logic, Cambridge: Cambridge University Press,  21–​44. Lewis, David (1983): “New Work for a Theory of Universals”, Australasian Journal of Philosophy 61, 343–​377. Liebesman, David (2018):  “The Normativity of Meaning”, in D. Star (ed.), Oxford Handbook of Reasons and Normativity, Oxford: Oxford University Press, 1015–​1039. Ludlow, Peter (2014): Living Words, Oxford: Oxford University Press. Maier, Emar (2010):  “Presupposing Acquaintance:  A Unified Semantics for de dicto, de re and de se Belief Reports”, Linguistics and Philosophy 32(5), 429–​474. Maier, Emar (2016): “Attitudes and Mental Files in Discourse Representation Theory”, Review of Philosophy and Psychology 7, 473–​490. Maitra, Ishani (2007): “How and Why to be a Moderate Contextualist”, in G. Preyer and G. Peter (eds.), Context-​ Sensitivity and Semantic Minimalism, Oxford: Oxford University Press, 112–​32. Marmor, Andrei (2009): Social Conventions, Princeton, NJ: Princeton University Press. McGlynn, Aidan (2012): “Interpretation and Knowledge Maximization”, Philosophical Studies 160, 391–​405.

195

Manuel García-Carpintero Meibauer, Jörg (2009):  “Implicature”, in J. Mey (ed.), Concise Encyclopedia of Pragmatics, Amsterdam:  Elsevier, 365–​378. Merchant, Jason (2004): “Fragments and Ellipsis”, Linguistics and Philosophy 27(6), 661–​738. Michaelson, E. and Woods, J. (n.d.): “Building Character”, unpublished ms. Moltmann, Friederike (2018):  “Levels of Linguistic Acts and the Semantics of Saying and Quoting”, in S. Tsohatzidis (ed.), Interpreting J. L. Austin, Cambridge: Cambridge University Press, 34–​59. Mount, Allyson (2015):  “Character, Impropriety, and Success:  A Unified Account of Indexicals”, Mind and Language 30, 1–​21. Murray, S. and Starr, W. (forthcoming): “The Structure of Communicative Acts”, Linguistics and Philosophy. Neale, Stephen (2004): “This, That, and The Other”, in M. Reimer and A. Bezuidenhout (eds.), Descriptions and Beyond, Oxford: Oxford University Press, 68–​182. Neale, Stephen (2005): “Pragmatism and Binding”, in Z. Szabó (ed.), Semantics vs. Pragmatics, Oxford: Oxford University Press, 165–​285. Neale, Stephen (2016):  “Silent Reference”, in G. Ostertag (ed.), Meaning and Other Things, Oxford:  Oxford University Press, 229–​342. Newman, G. and Knobe, J. (2019): “The Essence of Essentialism”, Mind and Language, DOI: 10.1111/​mila.12226. Ninan, Dilip (2012): “Propositions, Semantic Values, and Rigidity”, Philosophical Studies 158(6), 401–​413. Nowak, Ethan (forthcoming): “No Context, no Content, no Problem”, Mind and Language. Nunberg, Geoffrey (1993): “Indexicals and Deixis”, Linguistics and Philosophy 16, 1–​43. Pagin, Peter (2014): “Pragmatic Enrichment as Coherence Raising”, Philosophical Studies 168, 59–​100. Pagin, P. and Pelletier, J. (2007): “Content, Context and Composition”, in G. Preyer and G. Peter (eds.), Context-​ Sensitivity and Semantic Minimalism, Oxford: Oxford University Press, 25–​62. Pagin, P. and Westerståhl, D. (2010): “Pure Quotation and General Compositionality”, Linguistics and Philosophy 33, 381–​415. Pendlebury, Michael (1986): “Against the Power of Force: Reflections on the Meaning of Mood”, Mind 95, 361–​372. Perry, John (2009): “Directing Intentions”, in J. Almog and P. Leonardi (eds.), The Philosophy of David Kaplan, Oxford: Oxford University Press, 187–​201. Perry, John (2012): Reference and Reflexivity, 2nd edition, Stanford, CA: CSLI Publications. Pietroski, Paul (2018): Conjoining Meanings, Oxford: Oxford University Press. Portner, Paul (2018): Mood, Oxford: Oxford University Press. Potts, Christopher (2005): The Logic of Conventional Implicatures, Oxford: Oxford University Press. Quine, W. V. O. (1956): “Quantifiers and Propositional Attitudes”, Journal of Philosophy 53. Also in W. V. O. Quine, The Ways of Paradox, New York: Random House, 1966, 185–​196. Rabern, Brian (2013): “Monsters in Kaplan’s Logic of Demonstratives”, Philosophical Studies 164(2), 393–​404. Rabern, Brian (2017): “A Bridge from Semantic Value to Content”, Philosophical Topics 45(2), 181–​207. Radulescu, Alexander (2018a): “Token-​Reflexivity and Repetition”, Ergo 5(28), 745–​763. Radulescu, Alexander (2018b):  “Synonymy between Token-​ Reflexive Expressions”, Mind, DOI:  10.1093/​ mind/​fzy040. Recanati, François (2013): “Content, Mood, and Force”, Philosophy Compass 8(7), 622–​632. Recanati, François (2018): “From Meaning to Content: Issues in Meta-​Semantics”, in Brian Rabern and Derek Ball (eds.), The Science of Meaning, Oxford: Oxford University Press, 113–​137. Reimer, Marga (1991): “Demonstratives, Demonstrations and Demonstrata”, Philosophical Studies 63, 187–​202. Richard, Mark (2013): “What are Propositions?”, Canadian Journal of Philosophy, 43(5–​6), 702–​719. Roberts, Craige (2002):  “Demonstratives as Definites”, in K. van Deemter and R. Kibble (eds.), Information Sharing: Reference and Presupposition in Language Generation and Interpretation, Stanford, CA: CSLI Press, 89–​196. Roberts, Craige (2018): “Speech Acts in Discourse Contexts”, in D. Fogal, D. Harris and M. Moss (eds.), New Work in Speech Acts, Oxford: Oxford University Press, 317–​359. Russell, Bertrand (1903):  “On Meaning and Denotation”, in Alasdair Urquhart (ed.), The Collected Papers of Bertrand Russell,Volume 4: Foundations of Logic, 1903–​05, 324–​358, Routledge: London, 1994. Schiffer, Stephen (1981): “Indexicals and the Theory of Reference”, Synthese 49, 43–​100. Schiffer, Stephen (2003): The Things We Mean, Oxford: Clarendon Press. Schiffer, Stephen (2016):  “Gricean Semantics and Reference:  Responses to Anita Avramides, Stephen Neale, and Kent Bach”, in G. Ostertag (ed.), Meaning and Other Things, Oxford: Oxford University Press, 493–​527. Searle, John R. (1968): “Austin on Locutionary and Illocutionary Acts”, The Philosophical Review 77(4), 405–​424. Searle, John R. (1969): Speech Acts, Cambridge: Cambridge University Press. Simons, Mandy (2018): “Convention, Intention, and the Conversational Record”, in G. Preyer (ed.), Beyond Semantics and Pragmatics, Oxford: Oxford University Press, 284–​302. Smit, J. P. (2012):  “Why Bare Demonstratives Need Not Semantically Refer”, Canadian Journal of Philosophy 42(1),  43–​66.

196

Reference-Fixing and Presuppositions Soames, Scott (2005): Reference and Descriptions: The Case against Two-​Dimensionalism, Princeton, NJ: Princeton University Press. Speaks, Jeff (2016): “The Role of Speaker and Hearer in the Character of Demonstratives”, Mind 125, 301–​339, DOI: 10.1093/​mind/​fzv195. Speaks, Jeff (2017): “A Puzzle about Demonstratives and Semantic Competence”, Philosophical Studies 174, 709–​ 734, DOI: 10.1007/​s11098-​016-​0704-​5. Stainton, Robert (1995):  “Non-​sentential Assertions and Semantic Ellipsis”, Linguistics and Philosophy 18(3), 281–​296. Stalnaker, Robert (1978): “Assertion”, in P. Cole (ed.) Syntax and Semantics 9, New York: Academic Press, 315–​ 332; in R. Stalnaker, Context and Content, Oxford: Oxford University Press, 1999, 78–​95. Stalnaker, Robert (1997):  “Reference and Necessity”, in C. Wright and B. Hale (eds.), A Companion to the Philosophy of Language, Oxford: Blackwell, 534–​554. Stanley, Jason (1997): “Rigidity and Content”, in R. G. Heck (ed.), Language,Thought, and Logic, Oxford: Oxford University Press, 131–​156. Stanley, Jason (2000): “Context and Logical Form”, Linguistics and Philosophy 23, 391–​434. Stanley, J. and Szabó, Z. G. (2000): “On Quantifier Domain Restriction”, Mind and Language 15, 219–​261. Starr, William (2014): “Mood, Force and Truth”, Protosociology 31, 160–​181. Stenius, Erick (1967): “Mood and Language-​Game”, Synthese 17, 254–​274. Stojnić, Una (2017): “On the Connection between Semantic Content and the Objects of Assertion”, Philosophical Topics 45(2), 163–​179. Stojnić, U., Stone, M., and Lepore, E. (2013):  “Deixis (Even without Pointing)”, Philosophical Perspectives 27, 502–​525. Stojnić, U., Stone, M., and Lepore, E. (2017):“Discourse and Logical Form: Pronouns, Attention and Coherence”, Linguistics and Philosophy 40, 519–​547. Strawson, Peter (1950): “On Referring”, Mind 59, 320–​344. Thomason, Richmond H. (1990): “Accommodation, Meaning, and Implicature: Interdisciplinary Foundations for Pragmatics”, in P. R. Cohen, J. Morgan, and M. E. Pollack (eds.), Intentions in Communication, Cambridge, MA: MIT Press, 325–​363. Unnsteinsson, Elmar (2017): “A Gricean Theory of Malaprops”, Mind and Language 32(4), 446–​462. Vanderveken, Daniel (1991): “Non-​literal Speech Acts and Conversational Maxims”, in E. LePore and R. van Gulick (eds.), John Searle and His Critics, Oxford: Blackwell, 371–​384. Viebahn, Emanuel (2018):  “Ways of Using Words:  On Semantic Intentions”, Philosophy and Phenomenological Research DOI: 10.1111/​phpr.12526. Vlach, Frank (1981): “Speaker’s Meaning”, Linguistics & Philosophy 4, 359–​391. Weiner, Matthew (2005): “Must We Know What We Say?”, Philosophical Review 114, 227–​251. Wettstein, Howard (1984): “How to Bridge the Gap between Meaning and Reference”, Synthese 58, 63–​84. Williamson, Timothy (1996/​2000): “Knowing and Asserting”, Philosophical Review 105, 489–​523; included with some revisions as ­chapter 11 of his Knowledge and Its Limits, New York: Oxford University Press, 2000, from which I quote. Williamson, Timothy (2007): The Philosophy of Philosophy, Oxford: Blackwell. Yalcin, Seth (2014):  “Semantics and Metasemantics in the Context of Generative Grammar”, in A. Burgess and B. Sherman (eds.), Metasemantics: New Essays on the Foundations of Meaning, Oxford: Oxford University Press,  17–​54.

197

15 NAMES AS PREDICATES Sarah Sawyer

1  Introduction Names typically occur in bare, unmodified, singular form, as in (1) and (2): (1) Sophie is hosting the book club tonight. (2) I have known Sophie since she was a child. Because names typically occur in bare, unmodified, singular form, philosophical debates about the semantics of names have traditionally focused exclusively on examples such as these, and have traditionally seen their goal as that of providing an adequate account of names understood as singular terms; and given Kripke’s (1972) arguments against the descriptive theories of names offered initially by Russell (1905, 1911) and elaborated by Quine (1948, 1950), Referentialism, the view that a name is a directly referential expression, has come to dominate the philosophical literature.1 But names can also occur in modified form, taking, for example, the definite and indefinite articles, and occurring in plural and quantified forms, as in, respectively, (3)–​(6): (3) (4) (5) (6)

The Sophie that runs the café is extremely funny. A Sophie joined my dance class this week. There are three Sophies in my son’s year at school. Several Sophies work at the local cinema.

The fact that names can take the definite and indefinite articles, and occur in plural and quantified forms shows that names at least sometimes function like common count nouns, akin to terms such as ‘cat’ and ‘book’, and suggests that names are, at least sometimes, not singular, referential terms, but general, predicative terms. This means that the apparent generality of names needs to be accounted for just as much as their apparent singularity. Given the linguistic data, which shows that in surface grammar names function sometimes as singular terms and sometimes as general terms, an adequate theory of the semantics of names must do at least three things. It must provide an account of singular, referential uses of names; it must provide an account of general, predicative uses of names; and it must provide an account of the relation between the two. There are three broad approaches one might take. The first treats names as predicative terms 198

Names as Predicates

in all their occurrences, and faces the task of explaining singular uses of names. The second treats names as singular terms in all their occurrences, and faces the task of explaining predicative uses of names. The third treats names as polysemous (systematically ambiguous), and faces the task of explaining the connection between singular uses and predicative uses given that they are assumed to be semantically distinct. But one thing is clear: it is no longer adequate to maintain merely that names are singular, directly referential expressions. The generality of names evident in their predicative use must be accommodated in one way or another, and recognition of this fact is reflected in the recent philosophical literature on names. Given the topic of this chapter, I will focus on the first approach, according to which names are predicates in all their occurrences, but will mention alternative approaches when it would be helpful to draw comparisons.

2  The Predicate View (Predicativism) The Predicate View, or simply Predicativism, maintains that names play the semantic and syntactic role of predicates in all their occurrences.2 The view was initially defended by Sloat (1969) and, independently, by Burge (1973), who states the central thesis like this: ‘A proper name is a predicate true of an object if and only if the object is given that name in an appropriate way’ (Burge 1973, 428). A more recent defence of the view is given by Graff Fara, who says ‘in every case where a word occurs as a name, it is a predicate with (potentially) multiple application. The condition of its application is given by the being-​called condition, where the being-​called condition is stated in terms of the following schema’ (Graff Fara 2011, 496; see also 2015a, 64): (BCC) A name ‘N’ (when a predicate) is true of a thing just in case it is called N. Thus, just as ‘cat’ is a predicate true of an object if and only if it is a cat, ‘Sophie’ is a predicate true of an object if and only if it is a Sophie, where what counts as being a Sophie is understood variously as having been given the name ‘Sophie’ in the appropriate way (Burge 1973), being called Sophie (Graff Fara 2011, 2015a), or bearing the name ‘Sophie’ (cf. Bach 1997, 2002; Guerts 1997). The conditions are, as stated, broadly equivalent.3 There is an obvious metalinguistic element to names which makes them different from other predicates. Cats would have been cats even if the word ‘cat’ had never been used; but Sophies would not have been Sophies unless the word ‘Sophie’ had been used as a name. Nonetheless, a name need not be taken to be semantically equivalent to a metalinguistic predicate since the appropriateness condition, the being-​called condition, and the name-​bearing condition can be understood as conditions under which a name is true of an entity rather than as providing the semantic content of the relevant name. As Burge says, ‘[t]‌here is and need be no claim that a proper name abbreviates another predicate, even a roughly co-​extensive predicate such as ‘is an entity called “PN” ’. A proper name is a predicate in its own right’ (Burge 1973, 428–​429, original emphasis).4 One of the obvious benefits of Predicativism is its ability to accommodate general, predicative uses of names in a straightforward way. This is because, in maintaining that names are predicates in all their occurrences, the Predicate View allows for the fact that different individuals can share a name. A quick Google search reveals that there are 150 Sarah Sawyers in the US alone, and I have personally encountered three (other) Sarah Sawyers in my life. Referentialism, in contrast, treats names as context-​insensitive, individual-​denoting expressions. It is thus committed to the claim that people cannot strictly speaking share a name, and has to deal with apparent cases of name-​sharing as cases of ambiguity, with a name being multiply ambiguous in as many ways as it has bearers. Since ambiguous terms must be disambiguated in a semantic theory, the view requires, as Kripke says, that ‘uses of phonetically the same sounds to name distinct objects count as distinct names … for theoretical purposes’ (Kripke 1972, 260). The obvious way to represent this in the semantic theory is by means

199

Sarah Sawyer

of indexing:  ‘Sarah Sawyer1’; Sarah Sawyer2’; Sarah Sawyer3’, and so on. The same issue arises for the claim that names are abbreviated definite descriptions (cf. Russell 1905, 1911), which also treats same-​sounding names as multiply ambiguous, this time disambiguated by their descriptive semantic content. The multiple ambiguity thesis is counterintuitive and the Predicate View avoids the need to subscribe to it.5 But the Predicate View does need to explain how a name, understood as a predicate, can be used as a singular term to refer to a particular individual. That is, the Predicate View needs to explain bare singular names as exemplified in (1)  and (2). To this end, Predicativists are in broad agreement:  a bare singular name constitutes the predicative component of a denuded determiner phrase, a determiner phrase with an unpronounced determiner.6 But there is disagreement about the nature of the unpronounced determiner. According to Burge, bare singular names involve a demonstrative element, so that ‘[r]‌oughly, singular unmodified proper names, functioning as singular terms, have the same semantical structure as the phrase “that book” ’ (Burge 1973, 432). According to Graff Fara, in contrast, ‘the unpronounced determiner is the definite article, so that bare singular names actually constitute the predicative component of a denuded definite description’ (Graff Fara 2015a, 60); they thus have the same semantical structure as the phrase ‘the book’. This marks a clear division in the field. Predicativists who favour the former, demonstrative approach include Burge (1973), Elugardo (2002), and Sawyer (2010). Predicativists who favour the latter, definite article approach include Sloat (1969), Elbourne (2005), Matushansky (2005, 2006, 2008, 2015), and Graff Fara (2011, 2015a, 2015b).7 Thus the sentence ‘Sophie is tall’ presents in surface grammar as (7): (7)

Sophie

is

tall

but is understood by ‘that’-​Predicativists as (8), where Øthat is the unpronounced demonstrative: (8)

Øthat

Sophie

is

tall

and is understood by ‘the’-​Predicativists as (9), where Øthe is the unpronounced definite article: (9)

Øthe

Sophie

is

tall

I look at prima facie considerations in favour of ‘that’-​Predicativism in section 3 and prima facie considerations in favour of ‘the’-​Predicativism in section 4, but for now, suffice it to note that on suitable understandings, both versions have the result that a bare singular name refers to a contextually salient individual of whom the relevant name-​predicate is true. The Predicativist’s claim that there is a semantic connection between bare singular names on the one hand and general, predicative uses of names on the other provides a unified semantic account of

200

Names as Predicates

names in all their occurrences. As a result, the Predicate View provides a straightforward explanation of the connection between different uses of a given name, as exemplified by the apparent validity of inferences such as the following, which move from singular, referential uses to general, predicative uses and vice versa (cf. Hornsby 1976): (10) Sophie Laplane is a dancer, so there is at least one Sophie who is a dancer. (11) Sophies tend to be French, so Sophie Laplane is probably French. These ‘Hornsby entailments’ (as Graff Fara (2015a, 114)  calls them) are intuitively valid, and Predicativism explains their intuitive validity as a case of semantic validity.8 This may not be the only available explanation of the apparent validity of such inferences (cf. Leckie 2013; Rami 2014) but it has the prima facie virtue of simplicity over the alternative, pragmatic explanations that need to be offered if one adopts a Polysemy view, where such views need in addition to be supplemented by an adequate account of the relation between the two semantically distinct uses of names (cf. Leckie 2013; Rami 2014; Jeshion 2015c; Gray 2018; Delgado 2019; Schoubye 2020).

3  Singular Reference and Rigid Designation Two facts about bare singular names seem to favour ‘that’-​Predicativism over ‘the’-​Predicativism. The first concerns singular reference and the second concerns rigidity. According to the Predicate View, bare singular names contain an unpronounced determiner. If the unpronounced determiner is a demonstrative, we have a straightforward and immediate explanation of how a bare singular name, understood as a demonstrative-​predicate construction, can refer to a single individual on an occasion of use. After all, demonstrative-​predicate constructions (in their referential use9) generally enable singular reference to individuals that fall into the extension of the relevant predicate. This point is related to an evident similarity between sentences containing explicit demonstrative-​predicate constructions, such as (12), and sentences containing bare singular names, such as (13): (12) That cat is smart. (13) Sophie is smart. Such sentences are, as Burge notes, incompletely interpreted in the sense that they lack truth-​value independently of a context of application (Burge 1973, 432). That is, just as a contextual application of the phrase ‘That cat’ is required in order to determine which cat is being said to be smart, a contextual application of the name ‘Sophie’ is required in order to determine which Sophie is being said to be smart. The incomplete interpretation evident in the former is explained by the fact that at the level of logical form, a demonstrative construction contains a free variable designed to have different objects assigned to it in different contexts;10 and the incomplete interpretation evident in the latter is explained by the fact that at the level of logical form it is structurally analogous to the former and hence also contains a free variable designed to have different objects assigned to it in different contexts. If the unpronounced determiner is the definite article, in contrast, the Predicate View is faced with a problem: since a definite description is typically understood as involving a uniqueness operator, an account is needed of how a predicate that is not uniquely satisfied (there are many entities called Sophie, for example) can nonetheless be used to denote a unique individual in virtue of being coupled with a uniqueness operator. Burge explicitly rejects ‘the’-​Predicativism on these grounds. He says, ‘[m]‌ost of the proper names that a person is capable of using at a given time will be true of more than one object.We should therefore reject the claim that proper names in singular unmodified

201

Sarah Sawyer

form abbreviate the roles of the uniqueness operator and a predicate’ (Burge 1973, 431). He goes on to conclude his defence of the Predicate View with the following remarks: Insofar as proper names exemplify a fundamental way in which language relates to the world, they provide reason to focus not on individual constants, but on variables—​and not the variables of quantification, but free variables which represent demonstratives and which receive their interpretation extralinguistically, through the referential actions of language users. Burge 1973, 43911 The second fact about bare singular names that seems to favour ‘that’-​Predicativism over ‘the’-​ Predicativism concerns rigidity. A rigid designator is an expression whose referent in the actual world is identical with its referent relative to every possible circumstance of evaluation; it designates the same individual in every world in which it designates. Bare singular names are widely recognised as rigid designators. Since demonstratives used as referential expressions are also widely recognised as rigid designators, the claim that bare singular names are denuded complex demonstratives provides a straightforward explanation of the rigidity of bare singular names. In contrast, definite descriptions are widely thought not to be rigid designators.To illustrate the difference, note that the name ‘Jacinda Ardern’ refers to Jacinda Ardern both actually and counterfactually, whereas the definite description ‘the current Prime Minister of New Zealand’ denotes Jacinda Ardern actually, but a range of different people counterfactually. This difference in modal profile between names and definite descriptions is illustrated by the intuitive difference in truth-​value between (14), which is intuitively false, and (15), which is intuitively true: (14) Jacinda Ardern might not have been Jacinda Ardern. (15) Jacinda Ardern might not have been the current Prime Minister of New Zealand. And it is this difference in modal profile between names and definite descriptions that Kripke (1972) exploits in his arguments against Descriptivism: names, he concludes, are not definite descriptions. Advocates of ‘the’-​Predicativism have offered responses to both concerns. The first concern, as they point out, is simply a specific version of the more general concern about how an incomplete definite description (a definite description the predicative element of which is not uniquely satisfied) can be used to refer to a unique individual. Consider, for example, the definite description ‘the table’ in the following sentence: (16) Your dinner is on the table. Despite the fact that the definite description is incomplete (there are many tables), it can nonetheless be used to refer to a unique table when uttered in a given context. How is this possible? There are two broad approaches available. The first is partly semantic, since it incorporates additional semantic structure into incomplete definite descriptions in the form of an index or free variable, which is then taken to be assigned an individual relative to a context (cf. Larson and Segal 1995, 353–​355; Elbourne 2005; Hawthorne and Manley 2012, 219–​242). The second approach is purely pragmatic, since although it also maintains that an incomplete definite description will be assigned an individual relative to a context, it does not postulate additional semantic structure in the form of an index or variable in order to secure the assignment (cf. Russell 1919; Donnellan 1966; Neale 1990). But, as Graff Fara (2015a) points out, ‘the’-​Predicativists do not need to settle this local dispute. Rather, whatever the best account of incomplete definite descriptions turns out to be, ‘the’-​Predicativists can simply co-​opt it to account for singular, referential uses of names, with a bare singular name then understood as referring to a contextually salient individual of whom the name-​predicate is true. On any such account, ‘the’-​Predicativism, just like ‘that’-​Predicativism, will treat the referential variability 202

Names as Predicates

of bare names as a kind of context-​sensitivity rather than as the kind of lexical ambiguity to which the Referentialist is committed. The second concern comes from the claim that bare names are rigid designators while definite descriptions are not. This claim, however, turns out to be an oversimplification. First, incomplete definite descriptions can function as rigid designators. Consider the following sentences containing incomplete definite descriptions (taken from Graff Fara 2015c, 364): (17) Three guests outstayed their welcome at the party. (18) People might have left the party at midnight. (19) I’m guessing that everyone will leave the party early. The definite description ‘the party’ as it occurs in (17) is incomplete and yet designates a specific party rigidly. Moreover, the expression displays modal rigidity in (18) and temporal rigidity in (19). Graff Fara advocates the same ‘piggy-​backing’ strategy here as above, maintaining that whatever the best account of the rigidity of incomplete definite descriptions turns out to be,‘the’-​Predicativists can simply co-​opt it to account for the rigidity of bare singular names. She says: The upshot is that there is no onus on me to provide an explanation of how or why incomplete definite descriptions are rigid designators. That is the right of the piggybacker … I may content myself with the claim that whatever the semantics for incomplete definite descriptions is, and whatever the mechanism is that renders them rigid, denuded incomplete definite descriptions with names will follow suit. Graff Fara 2015a, 102–​10312 Further, not only can incomplete definite descriptions function as rigid designators, but bare singular names do not always display rigidity. In at least two notable cases, the non-​rigidity of incomplete definite descriptions is mirrored by the non-​rigidity of bare names (cf. Elbourne 2005; Graff Fara 2015a). The first example is that of role-​type incomplete definite descriptions, which are mirrored by role-​type bare names: (20) The car we rent should be an automatic. (21) Quintus should be your fifth son. The second example is that of co-​varying incomplete definite descriptions, which are mirrored by co-​varying bare names: (22) Every philosophy department that has a female logician wants the female logician to teach logic. (23) Every woman who has a husband called John and a lover called Gerontius takes only Gerontius to the Rare Names Convention (Elbourne 2005, 181). Thus ‘the’-​Predicativism may, after all, be able to account for both the singularity and the rigidity of bare names. The primary argument in favour of ‘the’-​Predicativism, however, comes from the ‘Sloat Chart’.

4  The Sloat Chart Sloat (1969) laid out a chart that compared the distribution of determiners with common count nouns (displayed on the left) and the distribution of determiners with names in the place of those 203

Sarah Sawyer

common count nouns (displayed on the right). He considers (i) the indefinite article, (ii) ‘sm’ which is the unstressed ‘some’ that occurs with plurals and mass nouns, (iii) ‘sóme’, which is the ‘some’ that is typically represented by the existential quantifier, (iv) bare plurals, and (v) the definite article. Asterisks indicate linguistically unacceptable sentences.

The Sloat Chart (Sloat 1969, 27)

A man stopped by. * Sm man stopped by. Sóme man stopped by. Sm men stopped by. Sóme men stopped by. Men must breathe. The clever man stopped by. The man who is clever stopped by. A clever man stopped by. The men stopped by. The man stopped by. * Man stopped by.

A Smith stopped by. * Sm Smith stopped by. Sóme Smith stopped by. Sm Smiths stopped by. Sóme Smiths stopped by. Smiths must breathe. The clever Smith stopped by. The Smith who is clever stopped by. A clever Smith stopped by. The Smiths stopped by. * The Smith stopped by. Smith stopped by.

Graff Fara (2015a, 82)  extends the Sloat Chart by adding the following examples involving demonstratives. Question marks indicate marginally acceptable sentences: That clever man stopped by. ?? That man who is clever stopped by. * That men stopped by. That man stopped by.

That clever Smith stopped by. ?? That Smith who is clever stopped by. * That Smiths stopped by. That Smith stopped by.

The (extended) Sloat Chart provides evidence in favour of the claim that names function like common count nouns in a wide range of cases. But it also highlights a syntactic difference between names and common count nouns, illustrated in Sloat’s final two examples: common count nouns, such as ‘man’ in ‘Man stopped by’ cannot occur as bare singulars in argument position; whereas names, such as ‘Smith’ in ‘Smith stopped by’, can. The first thing to note is that the syntactic difference between names and common count nouns does not itself constitute a problem for Predicativism. After all, there are syntactic differences between mass nouns and count nouns despite the fact that they are both forms of noun: mass nouns, such as ‘bread’ and ‘cement’, cannot combine with the determiners ‘a’ or ‘every’; they cannot occur in the plural; they cannot complement ‘how many’; but they can complement ‘how much’.There are well-​ motivated explanations for these differences. The syntactic difference between names and common count nouns, then, does not tell against ‘the’-​Predicativism, but it does incur an explanatory debt. Specifically, if ‘the’-​Predicativism is to be plausible, it needs to provide a well-​motivated account of the rules governing when the definite article can appear on the surface and when it cannot.13 Graff Fara, drawing on Sloat’s own response to this question, offers an account in the form of the following general principle (Graff Fara 2015c, 95; see also Graff Fara 2015a, 93 for a different but equivalent formulation):14 Where ‘Øthe’: When a name occurs in a definite description, the definite article must be unpronounced when the name is its structural sister, unless the name is stressed.

204

Names as Predicates

She says: One could think of this as a kind of obligatory contraction: when nothing occurs “between” the definite article and a name, they get smushed together, resulting in an unpronounced definite article (discounting those cases in which the article is stressed for emphasis). One way for a word or phrase to get between the definite article and a name is to be linearly between them, as in … ‘The ever-​popular Michael is teaching metaphysics this year’. But linear adjacency of the definite article and the singular name is not a sufficient condition for contraction. A branching node that dominates the name but not the definite article also gets “between” them in the sense that’s relevant for blocking contraction, as in … ‘[The [Ivan [on the roof]]] is howling. Graff Fara 2015a, 9315 The following provide graphic illustrations of the syntactic structure of pairs of sentences that differ with respect to whether the definite article is null or voiced: (24) Michael is teaching Metaphysics this year. (with ‘The’ null, since contraction of ‘The’ and ‘Michael’ is possible): S

DP

Øthe

VP

Michael is teaching Metaphysics this year

(25) The ever-​popular Michael is teaching Metaphysics this year. (with ‘The’ voiced, since contraction of ‘The’ and ‘Michael’ is not possible): S

DP

The

VP

NP

ever-popular

is teaching Metaphysics this year Michael

(26) Ivan, who is on the roof, is howling. (with ‘The’ null, since contraction of ‘The’ and ‘Ivan’ is possible):

205

Sarah Sawyer S

VP

DP

Øthe

RC

Ivan

is howling

who is on the roof

(27) The Ivan on the roof is howling. (with ‘The’ voiced, since contraction of ‘The’ and ‘Ivan’ is not possible): S

DP

The

VP

NP

is howling

Ivan

on the roof

The exception to the contraction rule concerns bare names that are stressed for emphasis: (28) A: I met an Emma Thompson today. B: That’s nothing! I met THE Emma Thompson today! The explanation for the definite article being voiced in such cases is simply that the emphasis is most naturally achieved by stressing the definite article (rather than the name), and it cannot be stressed without being voiced. The Where ‘Øthe’ principle accounts for a surprisingly wide range of cases, and provides prima facie syntactic support for ‘the’-​Predicativism, since it explains not only the syntactic similarity between bare names and common count nouns, but also the syntactic differences.16 Whether it ultimately provides support for ‘the’-​Predicativism over ‘that’-​Predicativism, as Graff Fara claims (cf. also Matushansky 2005, 2006, 2015), is less clear.17

5  Extended Uses of Names The examples of names considered so far have all been literal uses of names. But names are not always used literally. Consider the following examples involving extended uses of names: (29) Two Draculas came to the Halloween party. (30) I wish I had a class full of Sophies. 206

Names as Predicates

(31) The Picassos are in the east wing of the gallery. (32) Walter Cox is a Romanov. Such examples do not obviously fit the Predicativist claim that a name is a predicate true of an object if and only if it bears that name. In (29) ‘Dracula’ should be interpreted as true of an object if and only if it represents Dracula; in (30) ‘Sophie’ should be interpreted as true of an object if and only if it resembles Sophie in some salient way, such as being keen to learn; in (31) ‘Picasso’ should be interpreted as true of an object if and only if it was produced by Picasso; and in (32) ‘Romanov’ should be interpreted as true of an object if and only if it is descended from the dynasty of Romanovs. As such, extended uses of names have been cited as evidence against the Predicate View and in favour of a Polysemy view that treats singular, referential uses of names as primary (cf. Boer 1975; Jeshion 2015a, 2015b, 2015c; Delgado 2019).18 The Predicativist is faced with the task of explaining extended uses of names as non-​literal without thereby undermining the literal treatment of the original allegedly predicative examples given in (3)–​(6). To this end, there are two points to note about the extended cases. First, common count nouns, just like names, have non-​literal uses. For example, as noted by Graff Fara (2015a, 2015b), the (representational) function of the name ‘Dracula’ in (29) is analogous to the (representational) function of the common count noun ‘witch’ in (33): (33) Two witches came to the Halloween party. Similarly, the (resemblance) function of the name ‘Sophie’ in (30) is analogous to the (resemblance) function of the common count nouns ‘kitten’ and ‘tiger’ as they occur in (34): (34) I wish I had a class full of kittens rather than tigers. It may be harder to come up with common count nouns that function in the way that ‘Picasso’ does in (31) and ‘Romanov’ does in (32), but this may simply reflect the fact that production and dynasty are specifically traits valued in humans, rather than reflecting anything about the semantic or syntactic features of names. The mere fact that both names and common count nouns have non-​literal uses provides some prima facie reason to think that the extended cases could be accommodated by the Predicate View rather than telling against it, but the matter is far from settled (cf. Graff Fara 2015a, 2015b; Jeshion 2015a, 2015b; Matushansky 2015; Delgado 2019). The second point to note is that there is a significant difference between the original cases given in (3)–​(6) and the extended cases such as those given above. What is significant about the original cases is that understanding them requires no more than the basic recognition that they involve names. In contrast, the extended cases presented in favour of the Referentialist Polysemy view are cases the understanding of which requires substantive knowledge of a specific bearer of the relevant name. Such knowledge is also required to understand examples of names used as adjectives, as in (35) and (36) (taken from Delgado 2019, 386), and to understand examples of names used as verbs, as in (37) (taken from Clark and Clark 1979, 783) and (38) (taken from Delgado 2019, 386): (35) (36) (37) (38)

That dress is so Chanel. I enjoy watching Almodovar movies. My sister Houdini’d her way out of the locked closet. I Googled the title but didn’t find anything.

Daniel Dennett’s ‘The Philosophical Lexicon’19 provides a superb example of the way in which names can be adapted to play the role of syntactic types other than bare singular names.20 And here too we find that substantive knowledge of the philosophical views and styles of the philosophers 207

Sarah Sawyer

from whose names the entries are derived is required to understand the point of those entries. The Philosophical Lexicon is, essentially, an extended insider joke.The fact that such examples depend on derivations from bare singular names is taken by some to favour Referentialist Polysemy views; but it is worth noting that the difference between the original cases and the extended cases means that the Predicate View can, after all, distinguish between them without thereby undermining the literal treatment of the original predicative examples.Whether a move along these lines will be successful in defence of Predicativism is again a matter of debate. However, as Matushansky argues, extended uses of names ‘do not form a homogenous group yet can be clearly distinguished on both syntactic and semantic grounds from names involving a detectable naming component’ (2015, 335). If she is right, extended uses of names, far from telling against the Predicate View, might just support it.

Further Reading Burge, T. (1973) ‘Reference and Proper Names’, The Journal of Philosophy 70: 425–​439. Delgado, L. (2019) ‘Between Singularity and Generality: The Semantic Life of Proper Names’, Linguistics and Philosophy 42: 381–​417. Graff Fara, D. (2015) ‘Names are Predicates’, Philosophical Review 124: 59–​117. Jeshion, R. (2015) ‘Referentialism and Predicativism about Proper Names’, Erkenntnis 80: 363–​404. Matushansky, O. (2015) ‘The Other Francis Bacon: On Non-​Bare Proper Names’, Erkenntnis 80: 335–​362. Sawyer, S. (2010) ‘A Modified Predicate Theory of Proper Names’, in S. Sawyer, ed., New Waves in Philosophy of Language, Hampshire: Palgrave Macmillan: 206–​225.

Notes 1 To say that a name is a directly referential expression is to say that the semantic function of a name is to refer to its bearer directly rather than via descriptions. Its most austere form, Millianism, maintains that the semantic content of a name is exhausted by its referent (cf. Salmon 1986; Soames 2002). A  less austere, Fregean version allows co-​referential names to have distinct but non-​descriptive semantic content (cf. McDowell 1977; Evans 1982). 2 As noted by Graff Fara (2015a, 60), the view is consistent with different accounts of the semantical value of a predicate: it might be extensional, such as a function from entities to truth values; or it might be intensional, such as a property, or a function from possible worlds to extensions. 3 Graf Fara (2011) distinguishes the being-​called condition (an instance of which is: being called Sophie) from a bastardised being-​called condition (an instance of which is: being called ‘Sophie’). The two are not equivalent, since the mere fact that I call you ‘Sophie’ does not make you a Sophie; in order to be a Sophie, you need to be called Sophie. For a recent discussion of name-​bearing conditions see Gray (2014). 4 Burge’s target here is primarily traditional predicate accounts according to which a name abbreviates either a string of general, predicative terms (cf. Russell 1905, 1911), an artificially manufactured predicate (cf. Quine 1948, 1950), or a metalinguistic predicate (cf. Russell 1911; Kneale 1962). He notes that ‘[a]‌ppeals to abbreviation or manufacture are transparently ad hoc’ (Burge 1973, 427) and goes on to say that ‘[t]heoretically, it is undesirable to postulate abbreviation rules if they can be avoided’ (Burge 1973, 428). The complaint applies equally to later Descriptivist accounts such as Bach’s (1981, 1987, 2002) Nominal Description Theory, Guerts’s (1997) Quotation Theory and Katz’s (2001) Metalinguistic Descriptivism. 5 Burge (1973) argues in addition that such views face a dilemma: either they fail to provide a unified account of singular and general uses of names, or they cannot allow for unrestricted quantification over individuals who share a name since, for example, if a general use of a name were treated as equivalent to a disjunction of singular uses of same-​sounding names in a speaker’s language, a general use could only range over individuals for whom the speaker had a name. The force of the dilemma is elaborated and defended in Hornsby (1976). See also Sawyer (2010). 6 Gray (2018) argues that the best version of Predicativism should reject the unpronounced determiner view, but the view that results is better classified as an extended predicate version of the Polysemy approach. See note 18. 7 See also Segal (2001), who proposes that bare names be treated as referential definite descriptions, and Gray (2017), who argues that the most plausible version of Predicativism would treat bare names as non-​anaphoric definite descriptions. Neither author explicitly endorses the view they articulate. 8 Graff Fara (2011), drawing on work by Matushansky (2005, 2008), also argues that if names were not predicates in appellative position, ‘being called’ would itself be an ambiguous expression, where the name

208

Names as Predicates ‘Sophie’ occurs in appellative position in sentences such as the following: ‘My next door neighbour is called Sophie’. Predicativism avoids this unwelcome result. 9 We can distinguish three different uses of demonstrative constructions such as ‘that Sophie’. The first is as a referential expression, as in ‘That Sophie is tall’; the second is as a pronoun of laziness, as in ‘Sophie Laplane is a dancer. That Sophie works for the Scottish Ballet’; the third is as a kind of anaphoric expression (so-​called ‘donkey-​anaphora’), as in ‘If a friend brings a Sophie to a party, they probably know that Sophie’. It is clearly the first of these uses that is relevant here and we can leave the others to one side. 10 See Burge (1977) for an account of the logical form of demonstrative constructions along these lines. 11 In this respect, the claim that bare singular names contain an unpronounced demonstrative bears a certain similarity to Indexicalism, according to which the character of a name N is a function from a context to the contextually salient individual who bears the name N (cf. Recanati 1997; Pelczar and Rainsbury 1998; Pelczar 2001), and to Variabilism, according to which names function like pronouns (cf. Cumming 2008; Schoubye 2020). Indexicalism and Variabilism still need to explain predicative uses of names, however. For concerns about their ability to do so, see for example Delgado (2019). 12 See Schoubye (2018) for critical discussion, although it’s unclear the arguments work specifically against Graff Fara’s piggybacking strategy. 13 A range of objections to Predicativism on the grounds that names do not function syntactically like common count nouns can be found in Higginbotham (1988), Larson and Segal (1995), Segal (2001), and King (2006). For responses see Sawyer (2010) and Graff Fara (2015a, 2015c). For a set of recent objections to ‘that’-​ Predicativism, see Rami (2014). 14 The principle is simpler and accounts for more of the linguistic data than that offered by Sloat (1969). It also differs, as Graff Fara notes (2015a, 95), from the principle called m-​merger offered by Matushansky (2006). 15 According to Graff Fara, further evidence for the existence of a null definite article is given by the fact that a number of languages, including French, German, Italian, Portuguese, and Spanish, permit pronunciation of the definite article with singular uses of names (cf. Matushansky 2006, 2008). However, Delgado (2019) takes broader cross-​linguistic data to tell against ‘the’-​Predicativism and in favour of what she calls Polyreferentialism. 16 Hawthorne and Manley (2012) raise an objection to ‘the’-​Predicativism based on the syntactic difference between the following: (i) ‘In every race, the colt won’; (ii) ‘In every race, John won’. While (i) permits a covarying reading that allows for different races to have been won by different colts, (ii) can only be true if a single John won every race. Graff Fara (2015c) argues that Where Øthe, when combined with the account of nominal restriction given in Stanley (2005), not only explains the difference between the sentences, but predicts it. For critical discussion see Schoubye (2016). 17 Jeshion (2015c, 2017) argues that there are contexts in which sentences of the form ‘The Smith stopped by’ are in fact grammatical. This, she says, undermines the syntactic rationale for ‘the’-​Predicativism. See also Gray (2017). The grammaticality of the examples she provides is clear, but I am inclined to think, contra Jeshion, that the data further supports ‘that’-​Predicativism rather than Referentialism. 18 Although Polysemy accounts typically treat singular uses of names as primary, and hence are, as Rami puts it, extended individual constant views (Rami 2014, 849), they need not do so. Gray offers (although does not strictly endorse) an extended predicate view, according to which singular uses are ‘the articulation of individual-​denoting expressions … generated via lexical rules’ (Gray 2018, 5550) analogous to terms such as ‘Mum’ and ‘Grandma’; and Rami also gestures towards a ‘systematic non-​uniform equivalence account’ (Rami 2014, 855) according to which neither use is primary, but concedes that developing it would be ‘a rather difficult task’ (Rami 2014, 860). 19 The Philosophical Lexicon is available online: www.philosophicallexicon.com/​. 20 Delgado (2019) takes the fact that names can function as adjectives and verbs to undermine the initial rationale for Predicativism, since the mere fact that names function as predicates can no longer be seen to support the claim that names are fundamentally predicates.

References Bach, K. (1981) ‘What’s in a Name’, Australasian Journal of Philosophy 59: 371–​386. Bach, K. (1987) Thought and Reference, Oxford: Oxford University Press. Bach, K. (1997) ‘Do Belief Reports Report Beliefs?’, Pacific Philosophical Quarterly 78: 215–​241. Bach, K. (2002) ‘Giorgione Was So-​Called Because of His Name’, Philosophical Perspectives 16: 73–​103. Boer, S. (1975) ‘Proper Names as Predicates’, Philosophical Studies 27: 389–​400. Burge, T. (1973) ‘Reference and Proper Names’, The Journal of Philosophy 70: 425–​439. Burge, T. (1977) ‘Belief De Re’, Journal of Philosophy 74: 338–​362. Clark, E.V. and Clark, H. H. (1979). ‘When Nouns Surface as Verbs’, Language 55: 767–​811.

209

Sarah Sawyer Cumming, S. (2008) ‘Variabilism’, The Philosophical Review 117: 525–​554. Delgado, L. (2019) ‘Between Singularity and Generality: The Semantic Life of Proper Names’, Linguistics and Philosophy 42: 381–​417. Donnellan, K. (1966) ‘Reference and Definite Descriptions’, Philosophical Review 75: 281–​304. Elbourne, P. (2005) Situations and Individuals, Cambridge, MA: MIT Press. Elugardo, R. (2002) ‘The Predicate View of Proper Names’, in G. Preyer and G. Peter, eds, Logical Form and Language, Oxford: Oxford University Press: 467–​503. Evans, G. (1982) The Varieties of Reference, J. McDowell, ed., Oxford: Clarendon. Graff Fara, D. (2011) ‘You Can Call Me “Stupid”, … Just Don’t Call Me Stupid’, Analysis 71: 492–​501. Graff Fara, D. (2015a) ‘Names are Predicates’, Philosophical Review 124: 59–​117. Graff Fara, D. (2015b) ‘“Literal” Uses of Proper Names’, in A. Bianchi, ed., On Reference, Oxford:  Oxford University Press: 251–​279. Graff Fara, D. (2015c) ‘A Problem for Predicativism solved by Predicativism’, Analysis 75: 362–​371. Gray, A. (2014) ‘Name-​Bearing, Reference, and Circularity’, Philosophical Studies 171: 207–​231. Gray, A. (2017) ‘Names in Strange Places’, Linguistics and Philosophy 40: 429–​472. Gray, A. (2018) ‘Lexical-​Rule Predicativism about Names’, Synthese 195: 5549–​5569. Guerts, B. (1997) ‘Good News about the Description Theory of Names’, Journal of Semantics 14: 319–​348. Hawthorne, J. and Manley, D. (2012) The Reference Book, Oxford: Oxford University Press. Higginbotham, J. (1988) ‘Contexts, Models and Meanings: A Note on the Data of Semantics’, in R. Kempson, ed., Mental Representations:  The Interface Between Language and Reality, Cambridge:  Cambridge University Press: 29–​48. Hornsby, J. (1976) ‘Proper Names: A Defence of Burge’, Philosophical Studies 30: 227–​234. Jeshion, R. (2015a) ‘Names Not Predicates’, in A. Bianchi, ed., On Reference, Oxford:  Oxford University Press: 225–​250. Jeshion, R. (2015b) ‘A Rejoinder to Fara’s “Literal” Uses of Proper Names’, in A. Bianchi, ed., On Reference, Oxford: Oxford University Press: 280–​294. Jeshion, R. (2015c) ‘Referentialism and Predicativism about Proper Names’, Erkenntnis 80: 363–​404. Jeshion, R. (2017) ‘“The” Problem for the-​Predicativism’, Philosophical Review 126: 219–​240. Katz, J. (2001) ‘The End of Millianism: Multiple Bearers, Improper Names, and Compositional Meaning’, Journal of Philosophy 98: 137–​166. King, J. (2006) ‘Singular Terms, Reference, and Methodology in Semantics’, Philosophical Issues 16: 141–​161. Kneale, W. (1962) ‘Modality De Dicto and De Re’, in E. Nagel, P. Suppes, and A. Tarski, eds, Logic, Methodology and Philosophy of Science:  Proceedings of the 1960 International Congress, Stanford, CA:  Stanford University Press: 622–​633. Kripke, S. (1972) ‘Naming and Necessity’, in D. Davidson and G. Harman, eds, Semantics of Natural Languages, Dordrecht:  Reidel:  253–​ 355; reprinted, with an added preface, as Naming and Necessity, Cambridge, MA: Harvard University Press, 1980. Larson, R. and Segal, G. (1995) Knowledge of Meaning, Cambridge, MA: MIT Press. Leckie, G. (2013) ‘The Double Life of Names’, Philosophical Studies 165: 1139–​1160. Matushansky, O. (2005) ‘Call Me Ishmael’, in E. Maier, C. Bary and J. Huitnik, eds, Proceedings of SuB 9, Nijmegen: NCS: 226–​240. Matushansky, O. (2006) ‘Why Rose is The Rose:  On the Use of Definite Articles in Proper Names’, in O. Bonami and P. C. Hofherr, eds, Empirical Issues in Formal Syntax and Semantics, Paris: CSSP: 285–​307. Matushansky, O. (2008) ‘On the Linguistic Complexity of Proper Names’, Linguistics and Philosophy 31: 573–​627. Matushansky, O. (2015) ‘The Other Francis Bacon: On Non-​Bare Proper Names’, Erkenntnis 80: 335–​362. McDowell, J. (1977) ‘On the Sense and Reference of a Proper Name’, Mind 86: 159–​185. Neale, S. (1990) Descriptions, Cambridge, MA: MIT Press. Pelczar, M. W. (2001) ‘Names as Tokens and Names as Tools’, Synthese 128: 133–​155. Pelczar, M. and Rainsbury, J. (1998) ‘The Indexical Character of Names’, Synthese 114: 293–​317. Quine, W.V. O. (1948) ‘On What There Is’, The Review of Metaphysics 2: 21–​38. Quine, W. V. O. (1950) Methods of Logic, New York: Holt. Rami, D. (2014) ‘On the Unification Argument for the Predicate View on Proper Names’, Synthese 191: 841–​862. Recanati, F. (1997) Direct Reference: From Language to Thought, Oxford: Blackwell. Russell, B. (1905) ‘On Denoting’, Mind 14: 479–​493. Russell, B. (1911) ‘Knowledge by Acquaintance and Knowledge by Description’, Proceedings of the Aristotelian Society 11: 108–​128. Russell, B. (1919) Introduction to Mathematical Philosophy, London: George Allen and Unwin. Salmon, N. (1986) Frege’s Puzzle, Cambridge, MA: MIT Press. Sawyer, S. (2010) ‘A Modified Predicate Theory of Proper Names’, in S. Sawyer, ed., New Waves in Philosophy of Language, Hampshire: Palgrave Macmillan: 206–​225.

210

Names as Predicates Schoubye,A. J. (2016) ‘A Problem for Predicativism Not Solved by Predicativism’, Semantics and Pragmatics 9: 1–​11. Schoubye, A. J. (2018) ‘The Predicative Predicament’, Philosophy and Phenomenological Research 96: 571–​595. Schoubye, A. (2020) ‘Names are Variables’, The Philosophical Review 129: 53–​94. Segal, G. (2001) ‘Two Theories of Names’, Mind and Language 16: 547–​563. Sloat, C. (1969) ‘Proper Nouns in English’, Language 45: 26–​30. Soames, S. (2002) Beyond Rigidity, Oxford: Oxford University Press. Stanley, J. (2005) ‘Semantics in Context’, in Contextualism in Philosophy: Knowledge, Meaning and Truth, Oxford: Oxford University Press: 221–​253.

211

16 VARIABILISM Anders J. Schoubye

1  Introduction In philosophy of language, the view that proper names should be analyzed as variables is known as variabilism.While similar in certain respects, this view stands in opposition to millianism, which for long has been the predominant view of proper names. Proponents of variabilism maintain that an analysis of names as variables is empirically superior to a millian analysis since it not only explains the same data, but in addition explains a variety of data points that the millian analysis cannot. In this article, I introduce variabilism by way of comparison with millianism. I focus specifically on empirical data that the millian analysis struggles to account for and I explain why variabilism is better positioned to account for this data.

2  Millianism According to millianism, the meaning of a proper name is exhausted by its reference. That is, the meaning of a name is the individual to which it refers.This thesis was first put forward by J. S. Mill—​ hence the name millianism—​who famously distinguished between two kinds of meaning, namely what he referred to as denotation and connotation.1 The denotation of an expression corresponds to reference or extension, but what Mill had in mind with regard to the notion of connotation is more difficult to precisely pin down. Loosely, it might be described as a kind of implied descriptive content. However, with respect to proper names, Mill maintained that these expressions have no connotative meaning. Mill writes, A connotative term is one which denotes a subject, and implies an attribute. By a subject is here meant anything which possesses attributes. Thus, John, or London, or England, are names which signify an object only … None of these names, therefore, are connotative. Mill 1843, 31 As an argument in favor of this claim, Mill continues, A man may have been named John, because that was the name of his father; a town may have been named Dartmouth, because it is situated at the mouth of the Dart. But it is no 212

Variabilism

part of the signification of the word John, that the father of the person socalled bore the same name; nor even of the word Dartmouth, to be situated at the mouth of Dart. If sand should choke up the mouth of the river, or an earthquake changes its course, and remove it to be a distance from the town, the name of the town would not necessarily be changed. Mill 1843, 33 In this paragraph, Mill is identifying an important characteristic of proper names, namely that once an individual is named, the reference of the name intuitively becomes independent of the facts that may have grounded its being so named, e.g. relational or geographical facts. The river Dart could dry up, but the name ‘Dartmouth’ would still refer to Dartmouth. By contrast, a definite description denoting Dartmouth, for example ‘The town that sits at the mouth of the river Dart’, would shift its denotation if certain relevant geographical facts were changed, i.e. if the river changed its course. If the river simply ceased to exist, the description would, it seems, fail to refer even if the relevant town remained otherwise unchanged. Using contemporary tools of formal semantics where ‘⟦⋅⟧’ denotes a function from expressions to semantic values and a (for now) extensional semantic framework, the millian semantics for proper names can be stated as follows: (1) ⟦‘Alfred’⟧ = Alfred In non-​technical terms, the reference (and, hence, meaning) of the name ‘Alfred’, according to millianism, is just Alfred. Nothing more, nothing less. Millianism became orthodoxy in philosophy of language largely as a result of the seminal work of Kripke (1980).2 Millianism had been the subject of severe criticisms by chiefly Frege (1892) and Russell (1905) who both favored a descriptivist analysis, i.e. an analysis where names are treated essentially as definite descriptions. However, Kripke argued convincingly that names and definite descriptions behave markedly differently in various kinds of linguistic environments. For example, as alluded to above, names are not susceptible to the same kind of variability in metaphysical modal and counterfactual environments as definite descriptions.The latter are clearly capable of picking out distinct individuals across different metaphysical possibilities, which explains why a sentence such as (2) is generally judged true. (2) The president of France could have been someone other than Emmanuel Macron. By contrast, names seem to consistently pick out their reference at the world of the context of utterance, i.e. the actual world, even when occurring in the scope of a metaphysical modal. This explains why a sentence such as (3) is generally judged to be false. (3) Emmanuel Macron could have been someone other than Emmanuel Macron. This difference in truth conditions is, of course, difficult to explain if one assumes that a name is simply a covert description, for example that the meaning of ‘Emmanuel Macron’ is just ‘The president of France’. Upgrading to an intensional semantic framework, we can represent this difference between descriptions and names as follows. (4) ⟦‘The president of France’⟧w = [ιx.president-​of-​France(x) at w] (5) ⟦‘Alfred’⟧w = Alfred The pivotal difference between (4) and (5) is that the semantic value of the description is sensitive to a world of evaluation, whereas the semantic value of the name is not. So, if one of these 213

Anders J. Schoubye

expressions occurs in the scope of an operator whose function is to force an evaluation of the subsequent expressions at possible worlds other than the world of the context of utterance (e.g. a metaphysical modal), the description can pick out different individuals at these worlds whereas the name cannot. Kripke referred to this property of being immune to reference shifting by quantifiers over possible worlds as rigid designation.

Rigid Designation An expression τ is rigid iff for all possible worlds w and w′, ⟦τ⟧w = ⟦τ⟧w′ In addition to the observation that names are rigid, Millians also maintain that names are not context-​ sensitive, such as pronouns or indexicals.3 It is widely assumed that expressions such as ‘I’ or ‘she’ function essentially as variables that may have different values on different occasions of use and formally this is captured by relativizing the interpretation function ⟦⋅⟧w to two additional parameters, namely a context c and a variable assignment g where g is a function from variables to individuals. The variable assignment is determined by the context (typically as a function of the intentions of the speaker) and the assignment then determines directly the semantic values of any pronouns or indexicals that the speaker used.4 However, even though multiple individuals may be referred to using what is intuitively just one expression, for example ‘Alfred’, Millians generally assume that these individuals have distinct names that are simply homonyms, i.e. expressions that are phonologically or orthographically identical, yet have different meanings. So, even though there are multiple individuals who are referred to using the expression ‘Alfred’, Millians assume that each of these individuals have different names, viz. ‘Alfred1’, ‘Alfred2’, ‘Alfred3’ … ‘Alfredn’ (cf. Kaplan 1990). In short, on the millian view, names are constants rather than variables. One argument in favor of this analysis is that names do not appear to be sensitive to operators whose function is to shift variable assignments, e.g. nominal quantifiers such as ‘every’, ‘most’, and ‘many’. Pronouns, for example, have so-​called bound interpretations where they do not refer to specific individuals and their meanings instead appear to depend on values introduced by a higher quantifier. For example, the sentence (6) has one interpretation where every boy loves their own mother in addition to the interpretation that every boy loves the mother of whoever is demonstrated.The bound reading is indicated by co-​indexing between the quantificational antecedent and the pronoun. (6) [Every boy]1 loves his1/​2 mother. By contrast, it is not obvious that names can be used in this way. For example, there does not seem to be a corresponding bound interpretation of ‘Alfred’ in (7). (7) #[Every person named ‘Alfred’]1 loves Alfred’s1 mother. Relativizing the interpretation function to a context and a variable assignment, the millian analysis of proper names looks as follows: (8) ⟦‘Alfred’⟧c,g,w = Alfred In conclusion, although the millian analysis is remarkably simple, its explanatory potential is impressive. Millianism yields accurate predictions about the meaning of names in a very wide range of cases including complex modal and quantificational environments. Where millianism faces problems is with respect to comparatively less standard uses of names. 214

Variabilism

3  Empirical Challenges As mentioned above, I  will focus on three putative problems for millianism. Due to limitations of space, the presentation of these problems will be short (but hopefully sweet). These problems (and their variabilist solutions) are discussed in more detail in Schoubye (2017, 2020). The relevant problems are the following. • Bound uses of names • Shifted uses of names • Predicative uses of names Above I stated that names are not susceptible to bound interpretations along the lines of pronouns. However, this is not entirely accurate. It is true that names appear to resist binding in some of the linguistic environments where pronouns can be bound, but this does not mean that names have no bound interpretations at all. Consider the following examples. (9) If a child is christened ‘Bambi’, Disney will sue Bambi’s parents. (Geurts 1997) (10) Every woman who has a husband called John and a lover called Gerontius takes only Gerontius to the Rare Names convention. (Elbourne 2005) The names in these sentences do not function as referential terms, but rather as bound variables. For example, the sentence in (10) does not intuitively express a proposition about one specific individual, but rather a general proposition about individuals called Gerontius.The same applies mutatis mutandis to (9). These uses of names present an immediate challenge for millianism, because according to this view, names are logical constants, i.e. expressions whose semantic value is only determined by the model and therefore not shiftable by assignment shifters such as nominal quantifiers. Consequently, if millianism is correct, it should not be possible to use names in this way. Moreover, names also have cross-​ sententially bound interpretations. Consider the following example from Cumming (2007). (11) There is a gentleman in Hertfordshire by the name ‘Ernest’. Ernest is engaged to two women. As in the previous examples, there is a natural interpretation of (11) where the unquoted name is functioning as a bound variable rather than as a referential term. To see this, simply imagine a situation where there are two individuals called Ernest in Hertfordshire and both are engaged to two women. Furthermore, suppose that the speaker is not acquainted with either Ernest1 or Ernest2, but has deduced, say, on the basis of various kinds of statistical data concerning marriages in Hertfordshire, that at least one Ernest is engaged to two women. In this case, the speaker’s assertion of (11) is intuitively true, but in order to make sense of this on the millian analysis, one has to maintain that the speaker’s use of ‘Ernest’ in (11) refers to, and therefore expresses a proposition about, either Ernest1 or Ernest2. But this seems highly implausible. After all, which of them would it refer to and why? It seems more natural to think that the right analysis here is one where ‘Ernest’ is a variable bound by a quantifier in the previous sentence. Next, let’s consider shifted names. According to millianism, names are rigid. Consequently, operators whose function is to shift the world of evaluation (i.e. modal operators) should not be able to shift the reference of any name. As argued above, this prediction is clearly desirable for a variety of flavors of modal vocabulary, but it is not clear that the prediction is correct with respect to epistemic modal vocabulary. Consider the following case: it is believed by some that Robert Del Naja, a member of the Bristol-​based music trio Massive Attack, is in fact identical to the infamous street artist Banksy (whose identity is currently unknown).5 Given the current state of ignorance about Banksy’s

215

Anders J. Schoubye

identity and assuming that there is some plausibility to this theory about Del Naja, it seems that the sentence in (12) is intuitively true. (12) Del Naja might be Banksy. Moreover, the sentence in (12) is intuitively true even if in actual fact, but unbeknownst to us, Del Naja is not Banksy. That is, (12) is intuitively true now even if we later discover that its prejacent, viz. (13), is false. (13) Del Naja is Banksy. There is a natural explanation for the source of this judgment. We do not know the identity of Banksy and, moreover, it is compatible with what we know—​perhaps even probable given what we know—​that Del Naja is Banksy. Consequently, we judge it to be true that it is possible that Del Naja is Banksy. However, the intuitive truth of (12) is a problem for the standard millian view. If names are rigid, and hence immune to world-​shifting operators, it follows that identity sentences are either necessarily true or necessarily false. Consequently, a sentence such as (12) can only be true if its prejacent is true, i.e. if (13) is true. The same problem arises when sentences such as (13) are embedded under a non-​factive epistemic attitude verb such as ‘believe’. For example, the sentence in (14) could be true even though the sentence in (15) is false.6 (14) Goldie believes that Del Naja is Del Naja. (15) Goldie believes that Del Naja is Banksy. Finally, let’s consider so-​called predicative uses of names. It has been observed at least since Sloat (1969) and Burge (1973) that names can be used as predicates. Consider the following examples from Burge (1973). (16) (17) (18) (19)

There are relatively few Alfreds in Princeton. An Alfred joined the club today. The Alfred who joined the club was a baboon. Some Alfreds are crazy.

This observation about predicative uses of names is in fact completely general. That is, setting aside bare singular uses of names in argument position of a predicate, for example (12) and (13), names appear to have the same syntactic distribution as standard count nouns (see Graff Fara (2015) for an in-​depth discussion). In particular, names combine happily with indefinite determiners (16)–​(17), definite determiners (18), quantifiers (19)–​(20), and numerical determiners (21).They can even occur as bare plurals as in (22). (20) All/​most/​many Alfreds joined the club today. (21) Three/​four/​seventeen Alfreds joined the club today. (22) Alfreds are clever. Predicative uses of names present an immediate problem for millianism. After all, since these names denote properties, this directly contradicts the claim that the meaning of a name is its reference.These uses of names are simply not referential uses.

216

Variabilism

For reasons that are not entirely clear to me, predicative uses of names have been somewhat neglected or dismissed by proponents of millianism.7 It is sometimes argued that such uses of names are coerced, i.e. arbitrary instances of temporarily forcing an expression outside its normal grammatical category. Since such coerced uses are both rare and unsystematic, the argument is that they therefore do not pose a threat to the millian thesis. However, the coercion explanation does not seem to be supported by any empirical evidence. Predicative uses are considered completely standard to normal speakers of English and are readily available (without any extensive contextual setups) for more or less every name in the language. For example, I  am guessing that you, the reader, did not have a sense of immediate squeamishness when the name ‘Alfred’ was used predicatively earlier in this paper. Most likely, you did not even notice, because such uses are completely unremarkable. Another fairly typical response is to deny that these are actually names or, relatedly, to argue that names are ambiguous (between predicative and referential types). Since the millian analysis only intends to provide an analysis of the referential type, it thereby avoids having to also account for predicative uses. But, as has been pointed out by Hornsby (1976), Leckie (2013), and Schoubye (2017), there are many reasons to be skeptical of this response. I mention only two here. First, the meanings of predicative uses of names and referential uses of names are clearly closely related. This is evidenced, for example, by certain types of inferences that seem intuitively acceptable. For example, (23) and (24) below. (23) (24)

Alfred joined the club today. ∴ An Alfred joined the club today. No Alfred joined the club today. ∴ Alfred did not join the club today.

These inferences are, using Kaplan’s (1989) terminology, indexically valid. If the premise is true in a context c, then the conclusion must also be true in c. Now, if one subscribes to the millian view, this is difficult to explain.The reason is that referential names contain and express no semantic information about the name of their referents. Nothing in the millian analysis rules out the existence of a context where Alfred joined the club today but where no one is called Alfred. You might think that such a context should be ruled out, but all it takes for the sentence ‘Alfred joined the club today’ to be true (given a millian analysis) is that the referent of ‘Alfred’ joined the club—​not that the referent is called Alfred.8 Second, if the existence of predicative and referential uses of names is simply a manifestation of a brute ambiguity, then it becomes difficult to explain certain facts about competence. Mainly that competence with referential uses appears to suffice for competence with predicative uses. For example, suppose that some speaker s is competent with referential uses of ‘Alfred’. She has encountered such uses and has been capable of determining its meaning (i.e. its reference) on one or more occasions. Now, even if s has never encountered a predicative use of ‘Alfred’, it is highly implausible that s will be incapable of understanding or inferring the predicative meaning simply on the basis of her competence with the referential counterpart. Moreover, no elaborate contextual setups are generally needed in order for s to do this successfully. This observation about competence is difficult to reconcile with a millian analysis, because given that a name conveys no semantic information about the properties of its referent, it is mysterious why competence with a name would suffice for competence with a descriptive predicate. According to the millian, the meanings of referential and predicative names are just fundamentally unrelated. These are not knock down arguments against millianism and there are various avenues that proponents of millianism might pursue in response. However, for the remainder of this paper, I will focus on variabilism and explain why this view not only accounts for the standard data (that millianism also accounts for), but also avoids the three potential problems outlined above. 217

Anders J. Schoubye

4  Variabilism Variabilism has been defended in some form or other by several researchers including Yagisawa (1984), Recanati (1993), Haas-​Spohn (1995), Dever (1998), Heim (1998), Pelczar and Rainsbury (1998), Pelczar (2001), Cumming (2008), Santorio (2012), and Rami (2014). Most of these views are quite similar. Names are assumed to behave essentially like free variables.9 In this paper, I will focus on the version of variabilism that I have defended more recently, namely in Schoubye (2017, 2020). According to this view, names are free variables whose meaning depends on a variable assignment determined by the context of utterance. However, assignments of semantic values to names come with a precondition (a semantic presupposition), namely that the individual assigned as the semantic value of the variable bears the relevant name. So, the analysis ends up looking as follows:

This says that the semantic value of ‘Alfredi’ relative to a context c and an assignment g is whatever g assigns to i on the condition that g(i) is called Alfred in the world of c. If g(i) is not called Alfred in wc, the expression has no semantic value relative to that context and assignment, i.e. it fails to refer. This analysis treats names essentially along the lines of pronouns. According to the standard analysis of pronouns, pronouns are also free variables whose semantic value depend on a variable assignment, but again this assignment is constrained by certain grammatical features of pronouns, namely person, number and gender features (also called ϕ-​features).The standard analysis of pronouns looks as follows:

The background assumption here is that the speaker’s intention in context is what determines the operative variable assignment in that context. If a speaker intends to refer to a using ‘Alfred1’, the context will determine an assignment g such that g: 1 ↦ a. Whether the speaker then succeeds in referring will depend on whether a ∈ {x ∣ x is called Alfred in wc}. Although the variabilist analysis of names is obviously different from a millian analysis, it shares many essential features. First, names are singular terms and the meaning of a name is simply the individual to whom the name refers (in context). So, in standard cases, a name’s truth conditional contribution is simply an individual—​precisely as on the millian analysis. Names are also rigid designators in a variabilist analysis. Shifting the world parameter has no effect on the semantic value of a name, so the meaning of a name (relative to an assignment g) is constant across all possible worlds (given g). This means that a variabilist analysis is going to make the same predictions as the millian analysis for the vast majority of cases, for example in simple sentences involving names, but also cases such as (3). The question, then, is how and why a variabilist analysis is superior to a millian analysis when it comes to the three problems outlined above. Below, I provide a short overview of the answer to this question. The observations and arguments in this section are essentially abbreviated versions of observations and arguments detailed in Schoubye (2017, 2020).

5  Addressing the Empirical Challenges As regards bound names, it is fairly obvious why a variabilist analysis would have an advantage over millianism. If names are variables that range over individuals, there is no prima facie reason to assume that such variables cannot be bound. And, as demonstrated above, they can—​at least in certain 218

Variabilism

linguistic environments. So, since variabilism straightforwardly predicts the possibility of bound interpretations of names, the main question for proponents of variabilism is rather why bound interpretations of names are so much more limited than bound interpretations of e.g. pronouns. For example, names appear to not permit bound interpretations in many cases where one might think they should, for example in (10). However, building on work by Gundel et al. (1993), one can give a compelling psycholinguistic explanation of the relative infrequency of bound interpretations of names. Gundel et al. (1993) show that among referential/​anaphoric expressions such as demonstratives, pronouns, proper names, and descriptions, speakers have a preference for pronouns whenever the intended referent/​anaphoric anchor is cognitively activated, for example, by being retrievable from the immediate linguistic environment. They refer to this as the givenness hierarchy. Whenever the intended referent is given, using a pronoun is preferred because it is less cognitively demanding to process. For illustration, consider the three sentences below. These sentences could all be used to express the same proposition, but in normal contexts, most speakers have a strong preference for (27). (27) Alfred1 thinks he1 is a genius. (28) #Alfred1 thinks Alfred1 is a genius. (29) #Alfred1 thinks [the person called Alfred]1 is a genius. Now, notice that when a name occurs in a linguistic environment where it ought in principle to have a bound interpretation, i.e. where it has a suitable antecedent c-​commanding it, the name is in direct competition with a corresponding pronoun. Furthermore, the “intended referent”, viz. the anaphoric anchor, is in any such case cognitively activated. Consequently, the speaker ought to use a pronoun (since that promotes easy processing) if that interpretation is intended. If the speaker uses a name, then it would be perfectly rational to infer that the speaker is not intending a bound interpretation, because if she were, she should have used a pronoun. So, if a name is used rather than a pronoun, one naturally gravitates towards the interpretation where the name is not bound and this explains why names appear to resist bound interpretations in the vast majority of cases. Assuming that this explanation of the infrequency of bound interpretations of names is correct, it raises another question, namely in what types of cases is a bound interpretation of a name licensed? One candidate case is a situation where using a (bound) pronoun over a (bound) name would impede rather than promote processing of the intended interpretation. If the processing explanation is correct, we should expect a preference for a name over a pronoun in such cases. As it happens, we have already encountered such a case, namely Elbourne’s example in (10)—​repeated below. (10) Every woman who has a husband called John and a lover called Gerontius takes only Gerontius to the Rare Names convention. In this case, if a pronoun is substituted for the name ‘Gerontius’, there are now two potential anaphoric anchors both of which are activated, cf. below. (30) Every woman who has [a husband called John]1 and [a lover called Gerontius]2 takes only him1/​2 to the Rare Names convention. Determining that the speaker’s intended interpretation of the pronoun is one where the anaphoric anchor is ‘a lover called Gerontius’ rather than ‘a husband called John’ requires fairly laborious pragmatic reasoning. Consequently, the processing costs of using the pronoun are much higher than the costs of using a name since the latter provides immediate disambiguation. For this reason, a bound interpretation of the name is licensed. This looks like strong evidence in favor of the processing explanation outlined above.10 219

Anders J. Schoubye

Moving on to shifted uses of names, one might think that variabilism has no tangible advantage over millianism. After all, according to both views, names are rigid, and so their predictions with respect to names embedded under modals should be identical—​and they are. However, the solution to this problem is not to be found in the semantics of names (alone), but rather in the semantics of epistemic modals. In recent years, several researchers have argued that epistemic modals should not be analyzed simply as quantifiers over possible worlds. One of the main motivations for this analysis is precisely to account for cases of shifted uses, namely shifted uses of pronouns (including even so-​called “pure” indexicals) under epistemic modals. Here is an example of such a case: suppose Del Naja has been involved in an accident resulting in severe amnesia leaving his memory of the past twenty years extremely hazy. Suppose further that someone hands Del Naja the aforementioned article detailing all the evidence that he is Banksy. In this case, it seems clear that Del Naja could truly assert (31) even if it later turns out that Del Naja is not Banksy. (31) Ii might be Banksy. In order to capture this, it seems that the semantic value of one of the singular terms flanking the identity sign must somehow be assigned a different semantic value from the semantic values assigned at the world of the context. Shifted uses of pronouns were first discussed in detail by Santorio (2012). He argues that the best way to account for these cases is by assuming that the function of an epistemic modal is not merely to change the possible world that the prejacent is evaluated at, but moreover to change the variable assignment in accordance with the speaker’s information state. So, simplifying somewhat, rather than assuming (as is standard) that sentences of the form ‘might ϕ’ are true at w iff there is an accessible world w′ (where w′ is compatible with the speaker’s information state) such that ⟦ϕ⟧w′  =  1, one instead assumes that ‘might ϕ’ is true at ⟨w,g⟩ iff there is an accessible ⟨w′,g′⟩ (where ⟨w′,g′⟩ is compatible with the speaker’s information state) such that ⟦ϕ⟧⟨w′,g′⟩ = 1. In Santorio’s version, this is combined with a Lewisian style counterpart semantics where every individual in the domain has one (or more) epistemic counterpart(s) at each possible world. An individual a counts as an epistemic counterpart of b iff it is consistent with b’s information state at the world of the context that b is identical to a. Going back to the example above, given that Del Naja has no recollection of his whereabouts for the entire period where Banksy has been active as an artist, it is therefore consistent with Del Naja’s information state that he is Banksy. As a result, there is an accessible ⟨w′, g′⟩ such that the epistemic counterpart of Del Naja that is assigned to ‘I’ by g′ is identical to Banksy. So, (31) is true.11 The idea that epistemic modals have this kind of a dual function can also be motivated by more general considerations. Remember, epistemic modals are information modals, i.e. they are sensitive to a body of information available to a relevant party, typically the speaker. And since variable assignments are essentially representations of referential relations that obtain given certain features of the context, it makes sense to think that once an expression (that depends for its meaning on this assignment) is embedded under an epistemic modal, the assignment is shifted to represent the referential relations that are assumed to obtain given the relevant information state. In other words, when a speaker makes a statement whose meaning depends on her information state, it makes sense to think that what various referential expressions denote also depends on this information state. Santorio’s analysis of epistemic modals is, of course, perfectly suited to account for shifted uses of names given that these are also analyzed as variables (a point that Santorio also makes himself). However, if names are analyzed as constants, as the millians maintain, then the problem with shifted interpretations remains. Since names are not the type of expression that can be shifted by operators that shift variable assignments, Santorio’s modified analysis of epistemic modals is of no use. A further advantage of a variabilist analysis of names is that it makes it possible to provide a uniform explanation of all shifted uses, both pronouns and names. 220

Variabilism

Finally, let’s turn to predicative uses of names. Again, since variabilism assumes that names are singular terms, predicative uses of names appear to pose as much of a problem for variabilism as they do for millianism. However, while assuming that names are type-​ambiguous between referential and predicative uses is implausible given a millian analysis of referential names (as argued above), this is not the case for a variabilist analysis. Recall, two of the main problems with a type-​ambiguity view for millianism are (a) indexically valid inferences, cf. (23)–​(24) above, and (b) issues concerning competence. But these are not problems for a variabilist analysis. Let’s start with indexically valid inferences. Consider again (23), repeated below. (23)

Alfred joined the club today. ∴ An Alfred joined the club today.

The problem for millianism was explaining why whenever the premise (‘Alfred joined the club today’) is true in a context c, it follows that the conclusion (‘An Alfred joined the club today’) is also true in c. On the variabilist analysis outlined in section 4, this follows straightforwardly from the semantics (i.e. the analysis of bare singular names in argument position of a predicate, viz. referential uses of names). On the analysis of names in (25), the meaning of a name is simply the individual to whom the name refers (given the speaker’s intentions), but in order for a bare singular occurrence of ‘Alfred’, to refer, it is required that the intended referent is called Alfred. If the individual determined by the variable assignment does not satisfy this condition, the name fails to refer. Hence, in any context in which the premise of (23) is true, it follows straightforwardly that the referent of ‘Alfred’ is called Alfred in that context. Consequently, it also follows that there is an x such that x is called Alfred and x joined the club today.The same type of reasoning can be used, mutatis mutandis, to account for the indexical validity of (24). So accounting for these inferences is remarkably straightforward given the variabilist analysis in (25). What about competence? Again, variabilism offers a simple and elegant explanation. On the variabilist analysis, referential names encode a semantic presupposition that the intended referents must satisfy.This precondition is simply that the intended referents have a certain appellative property, namely that they are called that name. It follows that competence with any referential name requires understanding that the intended referents must be called that name. When a name is used as a predicate, syntax alone is going to alert the speaker to the fact that the expression in question denotes a property of some kind rather than an individual. And if the speaker is competent with referential uses of the name, the speaker will know what property serves as a precondition on the use of that name. Hence, the speaker can infer, on any given occasion, that if the name is being used to express a property, the property in question is most likely the one that serves as a precondition on the use of its referential counterpart.12 In conclusion, it is perfectly viable for a proponent of variabilism to assume that names are type-​ ambiguous between referential and predicative uses, because variabilism provides a straightforward explanation of why this ambiguity is systematic. Finally, it is worth pointing out that variabilism avoids the conclusion that referential names are also multiply lexically ambiguous. As mentioned above, according to millianism, each individual called ‘Alfred’ has a different name. Since there are potentially hundreds of thousands of people called Alfred, this leads to a quite significant inflation of the lexicon. It also assumes that whenever one encounters a new individual called Alfred, one learns a new word. This strikes me as rather implausible. As Pelczar (2001) writes: Why, for example, doesn’t a good dictionary list up all of the (supposedly) many meanings of a name, given that it is so unstinting in the case of a standardly recognized ambiguous term? Perhaps considerations of material economy and other practical limitations make it unfeasible to include all of the (again, supposedly) many meanings of a name like ‘David’; 221

Anders J. Schoubye

but surely the task is not so daunting when it comes to a name like ‘Cleopatra’ or ‘Willard Van Orman Quine’. Are we to account for the absence of these terms from the O.E.D. as a consequence of their obsolescence or marginality? But this is a book that contains an entry for ‘mammothrept’. 2001, 134 Of course, a variabilist view completely avoids this problem, because according to variabilism, there is just one name ‘Alfred’. It just so happens that this name can be used to refer to many different (indeed, hundreds of thousands of) individuals as long as they are called Alfred.

6  Concluding Remarks Above, I  introduced millianism about names and outlined three problems for this view. I  then introduced variabilism and showed that this analysis will make the same predictions as millianism in more or less all standard cases. In addition, I  argued that a variabilist analysis provides simple solutions to the three aforementioned problems without compromising or giving up any of the core virtues of millianism (e.g. direct reference and rigidity). But the explanatory potential of variabilism quite likely extends beyond the issues discussed in this short paper.Variabilism also provides a potential explanation of certain metasemantic problems for millianism, for example the so-​called problem of reference shift discussed by Evans (1985).13 And the analysis also seems well suited to explain what happens in cases where there is a mismatch between speaker’s reference and semantic reference. These are issues to be further explored and developed in future work. In conclusion, it seems clear that variabilism is a viable, and very attractive, alternative to millianism (and various forms of descriptivism).

Notes 1 See Mill (1843, 19–​46). 2 Ideas very similar to Kripke’s had, however, been published several years earlier, namely by Barcan Marcus (1961). 3 See e.g. Perry (1997). For a response to Perry, see Pelczar (2001). 4 I am simplifying somewhat as some researchers maintain that the semantic value of certain pronouns, e.g. ‘I’ and ‘you’ are determined directly by the context c and not the variable assignment. These are classified as indexicals rather than pronouns. I myself do not subscribe to this view, but this issue is orthogonal to the problems discussed in this paper. 5 Some of the reasons behind this theory are covered in the following story from The Independent:  www. independent.co.uk/​arts-​entertainment/​music/​news/​banksy-​robert-​del-​naja-​massive-​attack-​art-​who-​is-​he-​ identity-​real-​name-​graffiti-​music-​similarities-​a7805741.html. 6 This is, of course, just a variant of Frege’s (1892) famous argument against millianism, viz. Frege’s puzzle. 7 Exceptions include Leckie (2013) and Jeshion (2015) both of whom consider these uses very carefully and provide spirited defenses of millianism. I do not have space here to engage with these papers, but I refer the interested reader to Schoubye (2017, 2020) for further discussion. 8 A response to this point that I often encounter is that this is not a problem because the millian can just provide a pragmatic explanation for the seeming acceptability of the inference in (3). But that seems to me to miss the point, because a pragmatic explanation (whatever that may be) cannot explain why the inference is indexically valid (as indexical validity is a purely semantic notion). Of course, proponents of millianism could also deny that these inferences are indexically valid, but I am not sure on what grounds. 9 I am grouping together here so-​called “indexical” views of names with views where names are explicitly treated as variables. There will be important differences between indexical and variabilist views depending on what assumptions are made about the nature of these expressions. These views are nevertheless similar in spirit, and so I  group them together here under the banner of variabilism. Yagisawa is perhaps the only slight outlier since he assumes that names are variables that are bound by a covert existential quantifier. However since names strictly speaking are variables on this view, I think it deserves mentioning here.

222

Variabilism 10 To keep this fairly short, I am leaving out some important details here. For example, an attentive reader might have noticed that both Elbourne’s and Geurt’s cases of bound interpretations of names are donkey cases, i.e. cases where the standard structural constraints on binding are not satisfied (the binder does not c-​command the name). I discuss this complication (and others) in Schoubye (2020). 11 For a detailed outline of the semantics, see Santorio (2012). A quite similar strategy for dealing with shifted uses is also employed by Cumming (2007), Ninan (2012, 2018), and Rabern (2018). Cumming (2007) in particular employs the strategy to deal with shifted interpretations of names and argues that it also provides a way of solving Frege’s puzzle, i.e. cases such as (14) and (15). 12 Again, due to space limitations, I  am leaving out many important details here. The more detailed story involves the observation that predicative names can be understood as morphologically derived from referential names. There is evidence for this in not only English, but many other languages and this explains the systematic availability of predicative names in these languages (see Schoubye 2017 for discussion). Another issue that I do not have space to discuss here is predicative uses of names, but where the property expressed by the name is not a naming property. For example, uses of ‘Einstein’ to mean intellectual genius or ‘Picasso’ to mean ‘a replica of a painting by Picasso’. I discuss these uses and why they are not a problem for either variabilism or millianism in Schoubye (2020). 13 See Pelczar and Rainsbury (1998) for discussion.

References Barcan Marcus, Ruth 1961. Modalities and Intensional Languages. Synthese, 13, 4: 303–​322. Burge, Tyler 1973. Reference and Proper Names. The Journal of Philosophy, 70: 425–​439. Cumming, Samuel J. 2007. Proper Nouns. PhD thesis (Rutgers University). Cumming, Samuel 2008.Variabilism. The Philosophical Review, 117, 4: 525–​554. Dever, Joshua 1998.Variables. PhD thesis (University of California, Berkeley). Elbourne, Paul 2005. Situations and Individuals. Cambridge, MA: MIT Press. Evans, Gareth 1985. The Causal Theory of Names. In A. P. Martinich, ed. The Philosophy of Language. New York: Oxford University Press, pp. 292–​304. Frege, Gottlob 1892. On Sinn and Bedeutung. In Michael Beaney, ed., The Frege Reader. Blackwell Publishing, 1997, pp. 151–​172. Geurts, Bart 1997. Good News about the Description Theory of Names. Journal of Semantics, 14: 319–​348. Graff Fara, Delia 2015. Names are Predicates. Philosophical Review, 124, 1: 59–​117. Gundel, Jeanette K., Hedberg, Nancy, and Zacharski, Ron 1993. Cognitive Status and the Form of Referring Expressions in Discourse. Language, 96, 2: 274–​307. Haas-​Spohn, U. 1995. Versteckte Indexikalität und subjektive Bedeutung. Berlin: Akademie Verlag. Heim, Irene 1998. Anaphora and Semantic Interpretation: A Reinterpretation of Reinhart’s Approach. In Uli Sauerland and Orin Percus, eds., The Interpretive Tract, MIT Working Papers in Linguistics 25. Cambridge, MA: MIT Press, pp. 205–​246. Hornsby, Jennifer 1976. Proper Names: A Defence of Burge. Philosophical Studies, 30: 227–​234. Jeshion, Robin 2015. Referentialism and Predicativism about Proper Names. Erkenntniss, 80: 363–​404. Kaplan, David 1989. Demonstratives. In Joseph Almog, John Perry, and Howard Wettstein, eds., Themes from Kaplan. New York: Oxford University Press, pp. 481–​563. Kaplan, David 1990. Words. Proceedings of the Aristotelian Society, Supplementary Volumes, 64: 93–​119. Kripke, Saul 1980. Naming and Necessity. Cambridge, MA. Harvard University Press. Leckie, Gail 2013. The Double Life of Names. Philosophical Studies, 165: 1139–​1160. Mill, John Stuart 1843. A System of Logic, Ratiocinative and Inductive. In J. Robson, ed., The Collected Works of J. S. Mill, volume 7–​8. Toronto: University of Toronto Press, 1973. Ninan, Dilip 2012. Counterfactual Attitudes and Multi-​Centered Worlds. Semantics and Pragmatics, 5, 5: 1–​57. Ninan, Dilip 2018. Quantification and Epistemic Modality. Philosophical Review, 127, 2: 433–​485. Pelczar, Michael W. 2001. Names as Tokens and Names as Tools. Synthese, 128: 133–​155. Pelczar, Michael W. and Rainsbury, J. 1998. The Indexical Character of Proper Names. Synthese, 114: 293–​317. Perry, John 1997. Reflexivity, Indexicality and Names. In W. Kunne, ed., Direct Reference, Indexicality and Propositional Attitudes. Stanford, CA: CSLI. Rabern, Brian 2018. Binding Bound Variables in Epistemic Contexts. Inquiry. https://​ doi.org/​ 10.1080/​ 0020174X.2018.1470568. Rami, Dolf 2014. The Use-​ Conditional Indexical Conception of Proper Names. Philosophical Studies, 168: 119–​150. Recanati, François 1993. Direct Reference. Oxford: Blackwell. Russell, Bertrand 1905. On Denoting. Mind, 14: 479–​493.

223

Anders J. Schoubye Santorio, Paolo 2012. Reference and Monstrosity. The Philosophical Review, 121, 3: 359–​406. Schoubye, Anders J. 2017. Type-​Ambiguous Names. Mind, 126, 503: 715–​767. Schoubye, Anders J. 2020. Names are Variables. The Philosophical Review, vol. 129, 1:53–94. Sloat, Clarence 1969. Proper Nouns in English. Language, 45, 1: 26–​30. Yagisawa, Takashi 1984. Proper Names are Variables. Erkenntniss, 21, 2: 195–​208.

224

PART V

Two-​Dimensional Semantics

17 TWO-​DIMENSIONAL SEMANTICS Frank Jackson

1  The Informational Background The world (universe) is a huge object spread out in space and time.1 What is this object like? Words, especially those from the mouths, pens and keyboards of people we trust, help us answer this question. Newton talked of standing on the shoulders of giants.What he in fact stood on were, to continue the metaphor, the words of giants. How is it possible to get information about the world and its parts from words? To start with parts: by grasping how a part has to be for a word or string of words to apply to it.Take the objects in the extension of ‘chair’ (as used for an item of furniture). They have something in common over and above the fact that ‘chair’ applies to or is true of them. Each is, in addition, the right way for ‘chair’ to apply to it, and that right way is the information the word is able to give about the nature of the objects to which it applies. Or, to say it in terms of representation: when you describe something using ‘chair’, you represent it as being a certain way, and make the information that it is this way available to those who understand the language. There is, therefore, a sense in which a description theory of reference is true for some words: ‘chair’, for instance, applies to something just if it has a certain descriptive nature. Here ‘descriptive nature’ means what we mean when we talk of what things are like, their nature; we aren’t taking a stand on the ontology of properties.There is another question about reference—​the question of how it comes to be that, e.g., ‘chair’, applies to an object just if it has that certain descriptive nature.We won’t discuss that meta-​question, except to note that the answer to it will mention inter alia mental representation, the intentions of users of the word, and linguistic conventions. We can think of the information ‘chair’ is able to deliver in set-​theoretic terms.The word partitions all the objects there might be into those that are the right way for ‘chair’ to apply to them, versus the rest.We obtain the information the word is able to deliver by learning that some given object belongs to the sub-​set that is the right way. Here we are thinking of obtaining information about something as narrowing the open possibilities for how that something might be. Much the same goes for information about the world. In order for the sentences of those warning about global warming to be true, there is a way our world, and most especially the Earth part of it, has to be. Our grasp of that way is what worries us when we come across those sentences (from reliable sources), and is the information we extract from them to the extent we are confident of their truth. We can think of the information (in the inclusive sense that allows for false information) a sentence 227

Frank Jackson

is able to give us—​as a rough approximation to start with—​as the set of possible worlds at which the sentence is true: these are the worlds, in the sense of complete ways things might be, which are the right way to make the sentence true. The centrality of best explanation style reasoning in science (and everyday life) makes salient how important it is that we are able to grasp how a world has to be for a sentence to be true at it. Often we arrive at a view about how things are by (i) surveying a number of mutually inconsistent possibilities about how things might be, possibilities we describe in sentences, (ii) assessing the possibilities for initial credibility, something we could hardly do unless we knew, or had a reasonable grasp of, how things have to be in order for the sentences describing the possibilities to be true, (iii) deducing, for each possibility, what would happen in some specified situation were it how things are in fact; again, something we could hardly do unless we knew, or had a reasonable grasp of, how things have to be for the sentences describing the possibilities to be true, and (iv) checking, for each possibility, whether or not what would happen were it how things are does in fact happen in the specified situation. An example of interest to philosophers is the discovery that water is H2O, in the late 1700s. Chemists ended up agreeing that water’s being H2O was the best explanation of the experimental results, but making sense of the reasoning that led to this conclusion would not be possible without an understanding of which possibilities they were considering when they used ‘water’, ‘H2O’ and ‘—​’, where ‘—​’ was the word they used to refer to one or another alternative to H2O, in the sentences that formulated the competing hypotheses-​cum-​possibilities. Above, I link the worlds at which a sentence, S, is true with the information it delivers (to those who sufficiently understand the sentence and trust its origin).They are the worlds which are the right way for S to be true at them. What then is it for S to be true simpliciter? That will be the case if the worlds which are the right way include the way things actually are. In short, for S to be true is for S to be true at the actual world.This is a kind of correspondence theory, but shorn of the idea that truth is correspondence to the facts, and is a theory for sentences and not propositions. We can call the set of possible worlds at which some sentence S is true, its content or intension (I will use these terms interchangeably). It is not the same as S’s meaning. For we can give the same information using sentences that differ in meaning. For example, information about some given shape can be given in Cartesian or in polar co-​ordinates. The sentences couched in the two vocabularies differ in meaning but are alike in what they say the shape is. An obvious question is, why acknowledge two contents or intensions for some sentences? Why complicate matters? As we will see in the next two sections, for some sentences the account given above of the information they make available needs to be modified, in significant detail if not in spirit.

2  Rigidification As well as talking about how things are in fact—​about how the actual world is—​we talk about how things might or would have been. We talk, for example, about how things would be today had the corona virus pandemic been avoided or had the dinosaurs not gone extinct. When we do this, we often make comparisons between how things are in fact, that is, in the actual world, with how things are in various non-​actual possible worlds. When we compare how things are in the actual world with how things are in some non-​actual world, it can help to have words that make the extension of some bit of language at a non-​actual possible world depend in part on how things are at the actual world. Suppose I say that the winner of a lottery would not have been the winner if they had not bought a ticket. It might be asked, how can the winner fail to be the winner? All becomes clear if I explain that what I am saying is that the actual winner, the winner the way things actually are, would not have been the winner the way things would have been had they not bought a ticket. I use ‘the actual winner’ to talk about the winner at the actual world at worlds where they are not the winner. Or suppose I want to say that being poisonous for 228

Two-Dimensional Semantics

humans depends on our physiology. I might use ‘Had our physiology been different in various ways, the substances which are actually poisonous would not have been poisonous’. The ‘actually’ makes it clear that I am not making the absurd claim that a change in physiology could falsify the truth that what’s poisonous is poisonous. I am, rather, saying that the substances which poison us in the actual world do not poison us at certain counterfactual worlds where we have a different physiology. Our final example is from the philosophy of colour. Some hold, roughly, that blue is the property that causes things to look blue in circumstances, C (how to specify C does not matter for our purposes). Among these philosophers, there is debate over what to say about possible worlds where things look blue in C but the property that causes things to look blue differs from the property that does the job in our world (due, say, to differences in our physiology or the composition of light). One option is to hold that the blue objects in those worlds are those that possess the property that causes things to look blue in C in those worlds. The other option is to hold that the blue objects in those worlds are any objects which possess the property that causes things to look blue in C in the actual world. Philosophers who belong to the second camp can express their view by saying that blue is the property that actually causes things to look blue in C. We can think of the examples of the previous paragraph as examples of rigidification. Reference (extension) at non-​actual worlds is being restricted, in one way or another, by reference at the actual world. In our simplest case, ‘the actual winner’ refers at every world, W, to the winner at the actual world,WA, provided they exist at W. It is, therefore, a rigid designator. Its reference does not vary from world to world. Rigidification complicates truth at a world. Whether or not a sentence is true at W no longer depends solely on how things are at W. Typically, it will depend also on the relationship between how things are at W and how things are at the actual world. Take ‘The actual winner holds a ticket’, and assume that the definite description can be thought of in Russellian terms. The sentence will be true at W if and only if (i) there is one winner at WA, and (ii) the winner at WA exists in W (but may or may not be the winner in W) and holds a ticket in W. We can specify two contents or intensions for such sentences. One is the set of worlds, W, at which the sentence is true. The other is the set of worlds, W, at which the sentence is true if W is the actual world. In the first case, we consider the truth behaviour of the sentence as we go from world to world, leaving which world is the actual world undisturbed. In the second, we consider the truth behaviour of the sentence at each world under the supposition that it is the actual world. We can call the first the C-​intension (as every world but one is counterfactual), and the second the A-​intension (as each world is being treated as actual when evaluating the sentence). The A-​intension for ‘The actual winner holds a ticket’ is the set of worlds, W, at which there is one winner and they hold a ticket. This is one way to get two intensions for some sentences. When it happens, which intension is right for information? The A-​intension, for it charts the worlds whose actuality is consistent with the truth of the sentence. And this is the intuitively correct answer for ‘The actual winner holds a ticket’. It does not differ in what it says about how things are—​the information it purports to give about the world we occupy—​from ‘The winner holds a ticket’, and the A-​intension of the first is the same as the intension of the second (when there’s no need to chart two intensions, I will simply talk of the intension). Now we need to note a further reason to acknowledge two intensions or contents for some sentences.

3  De Se or Centred Information One way physical structures give information is by responding to their surroundings in distinctive ways. When this happens, often the information concerns how things are in relation to the structures themselves. It is de se information.The clicking of a Geiger counter tells us about radiation where the counter itself is. The same goes for human bodies. Our bodies respond to happenings around us in 229

Frank Jackson

ways that carry information. One kind of response is especially important, namely having a perceptual experience. Perceptual experiences carry information about how things are relative to ourselves. We hear cars as approaching us; we feel the ground as being beneath our feet; we see objects as being a certain distance and orientation from us; and so on. How should we conceive of this kind of information? Suppose a Geiger counter is clicking at a rate that indicates significant radiation. Does its clicking tell us something about the world in which it is located? Yes. It tells us that it is one of the worlds where there is significant radiation somewhere or other. However, the clicking tells us more than this. It tells us that the radiation is where the counter is located. But it does this without saying anything about that location over and above its being one where there is significant radiation, and without saying anything about the counter itself other than that it is located where there is significant radiation. It says, to put words in its mouth, ‘For each world that I might be in, I am one or other of the counters located where there is significant radiation in that world’. Here are two equivalent ways of spelling this out: (i) in terms of a function that goes from each world, W, and counter, C, in W, to truth just if C in W is located where there is significant radiation; and (ii) in terms of a set of centred worlds, each centre having significant radiation (this is perhaps the most common way of spelling it out). Here it is important to understand location as location in space-​time. The information is that the radiation is where the clicking is, in both the spatial and temporal sense. The same is true for perceptual experiences. They give information about how things are relative to the perceiver, but without saying anything about the perceiver over and above her or him being where things are that way relative to her or him.2 Suppose it feels hot to me where I am. My experience ‘says’ that I am somewhere hot, but does not say anything about me over and above my being in a hot location. Or, to say it in terms of representation: my experience represents its being hot where I am, without representing anything more about me.The upshot is that we should think of its content in terms of a function from subjects in worlds that goes to truth just when the subjects are located somewhere hot in their world, or, equally, in terms of a set of centred worlds with hot centres. We should not think in terms of a partition among worlds. What does this tell us about sentences? When it feels hot to me where I am, I am typically in a position to produce a sentence like ‘It is hot where I am’. Producing the sentence delivers information but does not create it. We should, therefore, construe its content in the same way as the content of the experience, at least if we are thinking of content in terms of information. Its content is not a set of worlds; it is a function from subjects in worlds to truth just when they are located somewhere hot in the worlds. Equally—​because putting a belief into words doesn’t create a new belief—​when I express what I believe using ‘It is hot where I am’, the content of my belief is not a set of worlds but is a function from subjects in worlds to truth just when the subjects are located somewhere hot in the worlds or, equivalently, what I believe is that I belong to the class of those who are somewhere hot in whatever world it is that I occupy. What has this got to do with two-​dimensionalism? Take some given token of ‘It is hot where I am’ produced by me, FJ, at some given time. It will be true at a world, W, just if it is hot where FJ is in W at that time. This means that there are two contents for a token of the sentence. One is the function we have lately been talking about from subjects in worlds to truth when those subjects are located somewhere hot in those worlds (in the space-​time sense, as we noted earlier—​a sense most easily understood in terms of the function going from temporal parts in worlds to truth, but opponents of temporal parts will want to do things differently); the other is a set of worlds, the worlds where FJ is hot at the time in question. The upshot is that we have two contents or intensions. If you think, as I and many do, that mental content is more fundamental than linguistic content—​the view that comes down to us from, e.g., Locke (1690, Book III)—​the more fundamental one is the function from subjects in worlds to truth, for that captures the content of the experience (and the belief based on it). This makes it reasonable to call it the primary content or intension, as is often done, and the other the secondary content or intension. 230

Two-Dimensional Semantics

I emphasise that the conclusion reached is that the sentence has two contents or intensions. The experience and belief have (in my view, but there is disagreement) just the one content or intension, the one given by the primary or A-​intension, the function from subjects in worlds to truth—​that indeed was part of the rationale for the term ‘primary’.

4  A Helpful Thought Experiment I say above that the content of the belief I might express using ‘It is hot where I am’ isn’t a set of worlds, because (i) the belief is justified by an experience whose content is not a set of worlds, and (ii) putting a belief into words doesn’t create a new belief. Let me reinforce the case for holding that there are beliefs whose content is not given by a set of worlds with a thought experiment of a piece with many in the literature (see, e.g., Kaplan 1989; Lewis 1979). I am looking at a mirror but think I am looking out a window. In the mirror I see someone being approached from behind by a trick-​or-​treater. I am seeing myself but believe I am seeing someone else. Do I have a belief about myself? In one sense, yes. I am seeing someone; the someone is me; and I am in an information-​preserving causal relation with myself. But there is also a sense in which the answer is no. For I do not believe that I belong to the class of people being approached from behind by a trick-​or-​treater. The thought experiment reinforces the point that we need to acknowledge that some beliefs are more than beliefs about the kind of world we occupy. The belief I do not have is the belief whose content is given by a function from bodies in worlds to truth just if the body in the worlds is being approached from behind by a trick-​or-​treater, for, as we noted, I do not believe that I belong to the class of those being approached from behind by a trick-​or-​treater. Here’s another way to say it: the subjects in worlds that might be me, consistently with how I take things to be, do not include subjects being approached from behind by trick-​or-​treaters.

5  Natural Kind Terms I have given examples of sentences with two contents. Some have two contents because of rigidification, and some because they provide de se or centred information, and sometimes both are in play. Are sentences containing natural kind terms also examples of sentences with two contents? I  (controversially) think the answer to this question is yes. I  will develop the argument using the familiar example of water and H2O. The chemists in the late 1700s, mentioned earlier, knew that there exists a liquid-​at-​room-​ temperature kind, which is potable, clear, falls from the sky, fills the lakes, etc.—​the watery stuff, as it is often tagged. They took it to be a kind in the sense of having a composition, wanted to know its composition, and proceeded to show that it is H2O. They reported this discovery using the word ‘water’ as in ‘Water is H2O’ (or the French or German etc. equivalents). Why was it justified, and how did they know it was justified, to report their discovery using ‘water’? When economists gathering information on individual house prices report what they have learnt using, as it might be, the word ‘median’, it is reasonable to ask what justifies their using this word, and we know how to go about answering this question. We should expect to be able to do much the same for the word ‘water’. The obvious answer is that ‘water’ is known to be a word for telling about the watery stuff—​that’s its informational role—​for what was discovered was that the watery stuff is H2O. This gives us two pressing questions. What about the Twin Earth thought experiment, which is designed to show that the watery stuff on Twin Earth isn’t water (Putnam 1973); and, secondly, how should we explain the widely recognised fact that, at every possible world where water exists, water is H2O, despite the fact that at some possible worlds the watery stuff isn’t H2O? I think it is important to consider these two questions in order, and will start with Twin Earth. Astronomers discover a distant planet. Amazingly, it is very like Earth; so alike that they call it ‘Twin Earth’. It contains watery stuff, which is called ‘water’ by its inhabitants. However, the watery 231

Frank Jackson

stuff on Twin Earth is XYZ, not H2O. Is XYZ water? Does ‘water’ in our mouths apply to XYZ as well as to H2O? It is a matter of record that philosophers who have thought carefully about these questions give three different answers. One party says no. If we Earthians want to talk about XYZ without using ‘XYZ’ itself, we need to coin a special word—​‘retaw’ as it might be. A second party says yes. They hold that water is multiply realisable in the way that, as many hold, mental states are, and the discovery of Twin Earth would be the discovery that water is in fact multiply realised. They may point to the example of Jade, which comes in two forms, and note that the discovery of Twin Earth would mean we should stop talking of the watery stuff. A third party says that our usage of ‘water’ allows us to go either way; the matter is, in this sense, indeterminate.3 What would explain going the first way, the way that insists it is a mistake to hold that XYZ is water? Imagine I am a team captain with a favourite coin for tosses at the beginning of matches. I name it ‘Lucky’ in honour of its track record, and keep it in my sports bag. Unknown to me, my rival for the captaincy replaces the coin with an identical looking one, hoping to break my lucky streak. At the beginning of the next match, I take the substituted coin from my bag and say ‘This is Lucky’. Have I spoken truly? No, obviously. Despite the fact that the substituted coin shares very many properties with the old coin, ‘Lucky’ does not refer to it. And the reason is clear. I introduced the name ‘Lucky’ as a name for a coin that shares a certain causal history with me. The substituted coin lacks this de se history. We can now understand why some go the first way. They discern a de se element in our usage of ‘water’. They think of ‘water’ in our mouths as a word for giving information about the watery stuff we Earthians have come across, and that stuff is H2O, not XYZ.Why do some go the second way, the way that thinks of water as multiply realisable? They do not discern a de se element in our usage of the word, or at least in their usage of the word (we are allowed to use words in different ways). Perhaps they are more interested in the potability etc. side of things. Why do some go the third way? They think that, whatever may be the case for one or another individual user of ‘water’, speakers of English as a whole have left it open whether or not ‘water’ has a de se element. The upshot is that there are three possible views about the informational content of ‘water’ sentences: (i) that we use them to give information about the watery stuff we ourselves have come across; (ii) that we use them to give information about the or any watery stuff that there is; and (iii) that it is indeterminate which of the two we do, but as the chance of there being watery stuff different in kind from the watery stuff we have come across is very low—​there is no chance that Twin Earth in fact exists—​the indeterminacy does not matter. We are now in a position to address the second question: how to explain the widely recognised fact that, at every possible world, water is H2O. It is the product of rigidification. The word ‘water’ is used for that which actually is watery, is the watery stuff in the actual world, where to be watery may have a de se element, may lack one, or the matter may be indeterminate. It does not matter; it is in fact H2O on any of the three possibilities.That’s why ‘water’ refers to H2O across logical space, and means of course that ‘water’ sentences have two contents. Some will feel that our discussion above misses the key message of Twin Earth, namely, the role of causation in securing the reference of natural kind terms: ‘water’ refers to the predominant cause of tokens of ‘water’ (roughly), and that’s why it refers to H2O across logical space. (And ‘water’ in the mouths of Twin Earthians refers to XYZ because XYZ is the predominant cause of ‘water’ tokens on Twin Earth.) There are, however, familiar problems with this reading of the message of Twin Earth. The reference of some natural kind terms cannot go by the predominant cause of tokens of them. Physicists and chemists are sometimes certain of the existence of a thing or things of such and such a kind, on the basis of one or another theoretical consideration rather than on the basis of interactions with them. They give things of this kind a name. Sometimes the name is temporary, an example is ‘eka-​silicon’ which was replaced by ‘germanium’; sometimes not, as in the names of some subatomic 232

Two-Dimensional Semantics

particles. Either way, they got a name before becoming the predominant cause of the name. Also, as we note above, philosophers who hold that ‘water’ in our mouths does not refer to XYZ, sometimes coin the word ‘retaw’ in order to refer to XYZ. But XYZ is not the predominant cause of their tokens of ‘retaw’. Finally, it isn’t in fact true that H2O is the predominant cause of tokens of ‘water’. Most tokens of ‘water’ are caused by what speakers believe. An example is someone asserting ‘This glass contains water’, when they desire to speak truly and believe the glass contains water. Moreover, typically what causes them to believe the glass contains water will be the fact that the stuff in the glass is a potable liquid, which is pretty much tasteless and odourless and is, they believe, the stuff in lakes and rivers. The stuff will, of course, very likely be H2O, but its being H2O is not what typically causes the beliefs that lead them to use ‘water’ in the sentences they utter (especially if they are speaking in, say, 1705). What does the causing is how the stuff looks and tastes etc.

6  Names Sentences containing names are also sources of examples of sentences with two contents. I start with the easy case. It is relatively non-​controversial that there are descriptive names, names which are rigidified definite descriptions (Evans 1979). We are not certain who first proved that there are infinitely many primes. Euclid gives an early proof but may not have been its author. (Thanks here to Benjamin Morison.) Despite our ignorance, we can coin a name for the person, say ‘Primer’, and use it to say, surely truly, ‘Primer was very smart’. It is plausible that this sentence is equivalent to ‘The person who actually was the first to prove there are infinitely many primes was very smart’—​the ‘actually’ being needed to ensure that ‘Primer’ is name-​like in the sense of never referring to different persons at different worlds. If this is right, then ‘Primer was very smart’ has two contents: one is the set of worlds where it is true; the other is the set of worlds, W, such that if W is the actual world, the sentence is true. There can also be descriptive names with de se content. I return to my car in the supermarket carpark. It has a new dent in it. I am certain that exactly one person was responsible but know nothing that picks them out apart from their relationship to myself, courtesy of my car. I name them ‘Denter’.When I say, as it might be,‘Had Denter not hit my car, I would be happier than I am’, what am I saying? I am saying that had the person who actually hit my car not done so, I would be happier than I am.‘Denter’ is working as a rigidified definite description with a de se element: I am using the relationship to myself to pin down the person I want to talk about, and the role of the (implicit) ‘actually’ is to ensure that I can sensibly talk about this person and how things would be in counterfactual situations where they do not hit my car. The hard cases are proper names like ‘London’ and ‘Roger Federer’, names that have an established use in our language. The contrast is with names introduced in special contexts where there are obvious candidates to be the definite descriptions that get rigidified. (When I told you about the names ‘Primer’ and ‘Denter’, the contexts delivered which descriptions should be rigidified.) All the same, if our focus is on information, there is an obvious way to find plausible candidates to be the rigidified de se definite descriptions for established names: ask what is common knowledge about such names, knowledge that bears on their role in delivering information about how things are. As a start on answering this question, I note that, included in the information we often seek, is information about individual objects:  the street where a party is to be held; which person is the greatest tennis player; a racehorse we want to back; and so on. And it helps if these objects have marks that help distinguish them in known ways from other similar objects. That’s why we give names to streets (and complain when the name of a street we are looking for has been vandalised), people, cities and racehorses, etc. Surely, this is common knowledge. Do we have to tell the folk that names get assigned, and that the assigned names can be helpful via the way they assist in discriminating among objects? It is also common knowledge that names are elements in causal chains designed to deliver information about the objects to which they have been assigned. Parents know that a daughter of theirs 233

Frank Jackson

got her name through a process—​obviously, for they were involved in the process—​and that when they use the name in notes to, for example, her school, they are using the name to deliver information about their daughter, and of course the school also knows this. Or take Roger Federer. Don’t we all know that he got the name ‘Roger Federer’ through a process that linked that name with him (without, unless we are Swiss, knowing the fine detail), which was then fit for the purpose of passing on information about him? I labour what seems to me obvious only because it is often suggested that the causal-​historical theory of reference for proper names—​which is, of course, what is sketched above—​isn’t common knowledge. As against this, I urge that Kripke’s (1973) signal achievement was to tell us which bit of common knowledge—​the bit about names getting assigned and the way they are used to transmit information—​constitutes the bit that delivers the correct theory of reference for proper names. I am not though suggesting that Kripke would agree with this way of putting matters, which is a way that fits best with causal descriptivism about names (Kroon 1987), not the causal theory as typically presented. (See chapters by Nelson and Soames (this volume) for further discussion of causal descriptivism.) This means we can think of established names as rigidified descriptions. The informational job of a sentence of the form ‘N is F’ is to give information about the object assigned ‘N’ or some predecessor name (names change over time and in translation) through the way ‘N’ and any predecessors are used to pass on information about the object so named to the effect that it is F. How can I talk of the object? Very many names are names of more than one object. What we need to do is think of the tokens as picking out a single object if all goes well. When we hear ‘I am happy’, we get useful information about a single person (if all goes well) because we use the token to identify the person; in particular, we look for the causal source of the token of ‘I’. Mutatis mutandis for names. Although many objects will likely have the name ‘N’, for a given token of ‘N is F’ there will be (if all goes well) a single object lying at the far end of the information-​preserving causal chain that was set in motion by an assignment of ‘N’ or its predecessor(s). Earlier I mentioned Euclid. Not much is known about him. How, in that case, can we be confident that the sentence ‘Euclid gave us an early proof that there are infinitely many primes’ is true? Because we are confident there are causal paths that start with someone assigned a name that morphed into ‘Euclid’ over time, paths made with sentences containing that name in one or another form, and created by speakers and writers in the business of passing on information, which end up in token sentences containing ‘Euclid’, and that these sentences tell us that the someone in question wrote down a proof that there are infinitely many primes. Tokens of ‘Euclid’ in the sentences we come across today are a bit like the fossil record—​something we use to trace back to how things were long ago. The account just given of what to say about the fact that objects do not, as a rule, have unique names treats the information coming from sentences containing names as de se information: we use our relationship to certain tokens to single out the objects we seek to give information about. This means that the A-​intensions of these sentences will be a set of centred worlds, or a function from subjects and worlds to truth just if the subjects are the right way in their worlds; whereas their C-​ intentions will be the set of worlds W where the actual thing assigned the token name in question is the right way in W (assuming it exists in W).

7  Learning from the De Se How do I learn about how things are by coming across a sentence token I understand and whose source I trust? In the simplest case, I know the set of worlds at which the sentence is true. Because I trust the source, I take it that the sentence token belongs to that set. I know that I am in the same world as the sentence token. I accordingly locate myself in that set of worlds (or maybe reinforce my belief that I am so located).That’s how I learn. But what happens when I come across a sentence with 234

Two-Dimensional Semantics

de se content—​a sentence whose A-​intension, whose informational content, is a set of centred worlds or a function from subjects in worlds to truth, and not a set of worlds? Take our example from before, ‘It is hot where I am’. It determines a function from subjects in worlds to truth just when the subject is located somewhere hot in the world. This means that when I come across a token of this sentence and trust its producer, I learn that it is hot somewhere or other in the world I share with the token. But where is the hot location in question? I need to draw on my understanding of how sentences containing ‘I’ work. The subject is the producer of the token sentence.This means I learn that it is hot where the producer of the sentence is located.This will, in turn, typically be information that is de se. The location is the location that I see or hear the producer of the token as occupying relative to where I myself am located. The information I acquire will be captured by centred worlds or functions from subjects in worlds to truth. Some find this puzzling but here is an example to reassure. Pilots coming in to land at regional airports learn from windsocks. They know a windsock gives information about how the wind is in relation to it itself, gives, that is, de se information—​information captured by a function from windsock–​world pairs to truth just if the direction and strength of the wind at the windsock in the world is the right way. But that’s not a problem for the pilots. They know how they themselves stand vis-​à-​vis some token windsock, and use that information to get the needed information about how the wind is in relation to where they themselves are.They end up with information about how things are vis-​à-​vis themselves.Very often we do what the pilots are doing (especially when we learn from sentences containing proper names, pronouns, phrases like ‘over here’, etc.).

Acknowledgements My debts are too many to enumerate but I must mention David Braddon-​Mitchell, David Chalmers, Lloyd Humberstone, David Lewis, Robert Stalnaker and Daniel Stoljar, and very helpful comments from the editors.

Further Reading Chalmers, David J. 2004. ‘The foundations of two-​dimensional semantics’, in Manuel Garcia-​Carpintero and Josep Macia (eds.), Two-​Dimensional Semantics: Foundations and Applications, Oxford: Oxford University Press, pp. 55–​140. Big paper on the many different ways of understanding two-​dimensionalism from someone who is a supporter. Davies, Martin and Lloyd Humberstone 1980. ‘Two notions of necessity’, Philosophical Studies, 38 (1):  1–​30. Detailed account of how actuality operators work, connecting their operation with explaining the contingent a priori, a topic not covered in this entry. Jackson, Frank 1998. From Metaphysics to Ethics:  A Defence of Conceptual Analysis, Oxford:  Clarendon Press. Chapters 2 and 3 concern the connections between conceptual analysis and the distinction between A-​and C-​ intensions. Jackson, Frank 2010. Language, Names, and Information, Oxford: Wiley-​Blackwell. Fuller coverage of some of the issues in this entry, along with a detailed argument in Lecture Four to the conclusion that what’s metaphysically possible isn’t a sub-​set of what’s conceptually possible, a topic not covered in this entry. Kripke, Saul 1973. Naming and Necessity, Oxford: Blackwell. Together with Putnam (1973), this is a principal impetus for two-​dimensionalism. The precise relationship between what Kripke and Putnam say and two-​ dimensionalism is controversial. Lewis, David 1979. ‘Attitudes de dicto and de se’, Philosophical Review, 88: 513–​543. Argues that all attitudes should be thought of as de se. Putnam, Hilary 1973. ‘Meaning and reference’, The Journal of Philosophy, 70 (19): 699–​711.Together with Kripke (1973), this is a principal impetus for two-​dimensionalism.The precise relationship between what Kripke and Putnam say and two-​dimensionalism is controversial. Stalnaker, Robert 1984. Inquiry, Cambridge, MA:  MIT Press. Expounds the possible worlds way of thinking about content and information. Tichý, Pavel 1983. ‘Kripke on necessity a posteriori’, Philosophical Studies, 43 (2): 225–​241. Presents the two-​ dimensionalist way of thinking about the necessary a posteriori, a topic not covered in this entry.

235

Frank Jackson Weber, Clas 2013. ‘Centered communication’, Philosophical Studies, 166: 205–​223. Detailed account of communication in terms of centred worlds.

Notes 1 In this entry I seek to explain, as simply as possible, why I am a believer in one version of two-​dimensionalism, the version expounded here. I do not offer reviews of alternative approaches. 2 Often relative to the body of the perceiver, as perceptual experiences in general represent orientation relative to a body: in front and to the left, for example. 3 The first answer used to be orthodoxy (with dissenters) but for evidence that things are changing see the remarks in Levine (2018, 16).

References Evans, Gareth 1979. ‘Reference and contingency’, The Monist, 62 (2): 161–​189. Kaplan, David 1989. ‘Demonstratives’, in Joseph Almog, John Perry and Howard Wettstein (eds.), Themes from Kaplan, New York: Oxford University Press, pp. 481–​564. Kripke, Saul 1973. Naming and Necessity, Oxford: Blackwell. Kroon, Fred 1987. ‘Causal descriptivism’, Australasian Journal of Philosophy, 65 (1): 1–​17. Levine, Joseph 2018. Quality and Content, Oxford: Oxford University Press. Lewis, David 1979. ‘Attitudes de dicto and de se’, Philosophical Review, 88: 513–​543. Locke, John 1690. An Essay Concerning Human Understanding, London: Thomas Bassett. Putnam, Hilary 1973. ‘Meaning and reference’, The Journal of Philosophy, 70 (19): 699–​711.

236

18 TWO-​DIMENSIONAL SEMANTICS AND IDENTITY STATEMENTS Kai-​Yee  Wong

1  Introduction In contrast to standard possible worlds semantics, possible worlds in a two-​dimensional semantic framework play two kinds of roles, rather than just one.This allows the framework to assign two kinds of intensions to expressions, rather than just one. Its fruitful use in explicating modal operators and the meanings of referential expressions like indexicals has led to two-​dimensional accounts that seek to revive the Fregean conception of meaning, or more specifically the descriptivist view of reference, which has fallen into disrepute due to intense criticisms, most famously, by Saul Kripke’s seminal work Naming and Necessity (1972).1 This entry provides a critical overview of the two most prominent of such accounts, proposed by Frank Jackson and David Chalmers. Unfortunately, there is not space to describe either account in detail, so I rely largely on brief descriptions and plenty of references to primary and secondary sources. More importantly, I rely on focusing on how the accounts explain the phenomenon of necessary a posteriori identity that Kripke’s well-​known examples (such as ‘Hesperus is Phosphorus’) have brought to light.2 Focusing on and structuring our discussions around this phenomenon is by no means just an expository convenience. These cases are of central importance in themselves. They epitomize a large part of the contribution of Naming and Necessity to the theory of reference.This is why section 2 outlines Kripke’s critique of the Fregean theory of referential expressions and section 3 explains rigid designation and identity statements. Section 4 discusses rigidification and the notion of a world considered as actual, leading to the introduction of two-​ dimensional functions in section 5.  I  discuss, finally, Frank Jackson’s and David Chalmers’s two-​ dimensionalism in sections 6 and 7. Section 8 makes a few critical remarks on Jackson’s handling of the semantic argument. I conclude (section 9) by discussing the important distinction between semantics and metasemantics.

2  Names, Millianism, and Descriptivism Traditionally and pre-​theoretically it has been supposed by many that there is a kind of singular term that names, or serves as a tag of something, rather than describes it.This simple and intuitive notion of a genuine naming device has found its way into various philosophical theories of singular reference and linguistic reference. In J. S. Mill’s view, ordinary proper names are genuine names: ‘the only names of

237

Kai-Yee Wong

objects which connote nothing are proper names; and they, strictly speaking, have no signification’ (Mill 1879, 67). By contrast, in Russell’s view (1905, 1920), only ‘logically proper names’ are genuine naming devices. Ordinary proper names are ‘disguised’ (or ‘truncated’, ‘abbreviated’) descriptions; and according to Russell, (definite) descriptions are not semantically self-​contained singular terms but incomplete symbols introduced by contextual definitions. For Frege, proper names (and in fact all singular terms) are also assimilated to descriptions. But Frege (1892) counted descriptions as genuine singular terms. Despite the significant dissimilarity between the views of Russell and Frege, there is nevertheless considerable agreement between them concerning singular terms, and, in particular, ordinary names. And this area of agreement constitutes the basis of what is commonly called descriptivism or the description theory. Proponents of this theory include Carnap (1947), Dummett (1973), Linsky (1977), and Searle (1958). Both names and descriptions can be used to construct simple identity sentences. Clearly some of these sentences, e.g., ‘Hesperus is Phosphorus’ and ‘The author of Pnin is Nabokov’, carry empirical information and are thus a posteriori (can only be known on the basis of empirical evidence). This spells trouble for the Millian view, for it seems to imply (1) Hesperus is Phosphorus, or any identity statement with names (only), can assert nothing but the self-​identity of an object, and should, therefore, be considered as semantically equivalent to ‘Hesperus is Hesperus’, which is hardly informative. ‘The author of Pnin is Nabokov’ faces no such puzzle (or so-​called Frege’s Puzzle) because the ‘the author of Pnin’ is clearly not just a tag of the individual it describes. So, it has been proposed that we may solve the puzzle by rejecting the Millian view in favor of descriptivism. Standard descriptivism of proper names claims that they are descriptional, where a singular term a is descriptional if there is associated with it a definite phrase β such that a and β are synonymous. Similarly, a general term a is descriptional if there is associated with it a descriptive (and often complex) phrase β such that a and β are synonymous. Descriptivism of natural kind terms holds that natural kind terms are also descriptional. Accordingly, for any singular term a that is descriptional, there is a descriptive phrase β such that the referent of a with respect to a possible world w is whatever denoted by β at w. Similarly, for any descriptional general term a there is associated with it a descriptive (and often complex) phrase β such that the class of things to which a applies with respect to a possible world w is whatever class of things β applies to with respect to w. It is important to clarify that descriptivism need not hold that a descriptional term in a language must abbreviate or be synonymous to a certain description in the language. The core idea of descriptivism is that a word refers to that which has the property or properties the speaker associates with it. The fundamental mechanism for words to refer to things, according to descriptivism, is via possession of associated properties or (in Jackson’s words) representational properties. A descriptivist theory of proper names holds that descriptions and proper names are the same in the sense that they both share this fundamental way of securing reference through the mediation of a set of representational properties, or Sinn (sense) in the Fregean terminology.3 It is ‘only in this sense that names (in a rich enough language) can be thought of as abbreviated definite descriptions’ (Jackson 1998a, 207). It is also worth noting that according to the now-​standard characterization of descriptivism, there are three roles that representational properties are supposed to fill: the representational role, the informative role, and the reference-​fixing  role.4

238

Semantics and Identity Statements

The Millian view of proper names argues that names are not descriptional, i.e., that they do not refer via a descriptive sense that serves all the three roles. The Millian view is also called ‘the theory of direct reference’.5 The Millian theory is also a kind of semantic externalism.This contradicts internalism, according to which what fixes the reference of a name supervenes on what is internal to the user, e.g., the properties that users associate with the name, such that anyone who understands it adequately will be in a position to grasp a priori that something falls under it if and only if it satisfies the relevant properties. As a positive theory about semantic values, the Millian theory holds that the semantic values (or propositional contents) of Millian terms are the objects or individuals they refer to. ‘Millianism’ is sometimes used to refer to this positive theory, most explicitly and rigorously defended by Salmon (1986, 1991) and, particularly for indexicals and demonstratives, by Kaplan’s seminal work (1989a, 1989b). The positive theory needs to be supplemented by an external account of referent fixing, i.e., of the mechanism by means of which a Millian term comes to refer to a particular object. In this regard, the picture Kripke offers is a causal one: a proper name typically has its reference fixed by ostension in a baptism, and the reference is preserved in the community to which the name belongs by a certain causal-​communicative connection between a use of the name and an object in the baptism. Putnam (1975) famously held a similar account for natural kind terms. Using his famous ‘Twin Earth’ thought experiment, he argued that ‘water’ is not descriptional in terms of general properties that may be included in the internal psychological states of a competent speaker. He presented an alternative picture, which he shared with Kripke, according to which natural kind terms such as ‘water’ or ‘tiger’ are typically introduced or taught by ostension to paradigmatic samples.6 In Naming and Necessity, Kripke rejects the Fregean picture in favor of the Millian view. His famous modal argument contends that the following holds true for any typical ordinary proper name ‘N’: for any description,‘D’, allegedly appropriate for association with ‘N’ in a descriptivist way, it may come to pass that N turns out not to be D. This contradicts the description theory, which implies that ‘N is D’ is necessarily true. The intuition here is that proper names, unlike descriptions, are rigid designators: they refer to the same object in all possible worlds. Another of Kripke’s arguments is of particular interest to our discussion below. It is the semantic argument or the arguments from error and ignorance. The argument is widely regarded as the most persuasive of Kripke’s arguments against descriptivism.7 In the simplest terms, it is argued that there are intuitively successful examples of reference where: [Error cases] The descriptivist analysis implies wrong conclusions about what names refer to. [Ignorance cases] The speakers lack the conceptual resources that the descriptivist analysis requires them to have.8 Consider a property, say being the discoverer of the incompleteness theorem, which might be associated with the name ‘Gödel’ as giving its sense according to the descriptivist analysis. Kripke asks us to imagine, as a hypothesis of what the world actually is, that Gödel never had that property: he only stole the proof from a dying mathematician—​say, Schmidt—​and published it as his own. It is not essential to the argument that the mistake be community-​wide. It can be a mistaken belief shared only by a minority of the community. The important point is that it is perfectly coherent to imagine that we are in such an ‘error situation’ and yet we are not always referring to Schmidt when we use ‘Gödel’. If we come to realize the error, we will say we were mistaken about Gödel’s contribution to logic, not, as the description theory implies, that we used ‘Gödel’ to refer to Schmidt. For the cases of ignorance, we are asked to imagine a speaker who uses, e.g., the word ‘Feynman’ to talk about Feynman despite knowing only that Feynman is, say, a famous physicist. Again, that such a story is coherent seems to fly in the face of the description theory, according to which the speaker’s word cannot refer to Feynman unless he can associate it with some description that picks out the physicist.

239

Kai-Yee Wong

Similar cases, not restricted only to proper names, have also been offered in plentitude by Donnellan, Kaplan, and Putnam. For example, Putnam claims that he does not know what separates elms from beeches but insists that he succeeds in referring to elms when he says that he does not know how beeches differ from elms (Putnam 1975, 226).

3  Identity Statements In Naming and Necessity, Kripke questions the traditional assimilation of metaphysical modalities and epistemic modalities.9 He argues that ‘necessary’ and ‘a priori’, as applied to statements,‘are not obvious synonyms’ (Kripke 1972, 38), but also that they are not even coextensive. Among the examples he gave are true identity statements such as (1). Since the statement can only be known on the basis of empirical evidence, the traditional view can be refuted if it is shown that (1) is necessarily true. In standard modal semantics an identity statement ‘a = b’ is true with respect to a possible world w if and only if the denotations of ‘a’ and ‘b’ with respect to w is identical. Now if ‘a = b’ is true in the actual world, then, given that both ‘a’ and ‘b’ are rigid, their denotations have to be the same in all possible worlds. So, assuming that names are directly referential (and so, rigid), it can be proved that (1) is necessarily true. The intuition about identity behind this proof is an integral part of the orthodox semantics for modal logic (SML). To see this, we may consider the seemingly paradoxical character of (i) (x) (y) [(x=y)→(x =y)]. According to SML, (i) is valid. On the face of it, it seems to have the paradoxical consequence that no identity statements can be contingent. In Quine’s (1961) well-​known critique of the intelligibility of quantified modal logic, he argues that the paradox is a symptom of modal logic’s commitment to what he thought is a metaphysical jungle, i.e., Aristotelian essentialism—​the thesis that an object has some of its traits necessarily and others contingently. It is true that (i) is essentialist in the sense that it asserts that any object identical with x is necessarily identical with x. But only x itself can be identical with x, so (i) asserts no more than that everything is necessarily itself, or (ii) (x) □(x =x). In fact, from (ii) and (iii) (x) (y) ((x = y)→(□(x = x)→□(x =y)), which is a substitution instance of the law of substitutivity of identity (iv) (x) (y) ((x = y)→ (Fx→ Fy)), we can derive (i). Since (ii) is by no means paradoxical, so neither is (i). How then would one come to think that (i) entails that all identities are necessary, including for instance (v) The evening star is Venus, where ‘the evening star’ abbreviates ‘the first star to appear in the evening’? The answer is scope ambiguity. If descriptions are treated in accordance with the conventional Russellian account, 240

Semantics and Identity Statements

(vi) □ (The evening star = Venus), is ambiguous between a wide (W) and a narrow (N) reading: (W) Just one thing x has the property of being the evening star and it is necessary that x = Venus. (N) It is necessary that just one thing x has the property of being the evening star and x = Venus. In order to correctly derive (vii) (The evening star = Venus) → □ (the evening star = Venus) from (i), ‘the evening star’ in the consequent of (vii), i.e. (vi), must be interpreted as having a wide scope.10 That is to say, one must read (vi) as (W). But (vi) is incompatible with the claim that (v) is contingent only when it is read as (N). So, what (i) allows us to infer is that, given that something is in fact the evening star and is identical with Venus, it is necessary that that thing is Venus. This is hardly incompatible with the claim that it is only contingent that the evening star is identical to Venus. What then does (i) really establish? In SML, (i) is true (in w) under an assignment V, iff for every alternative assignment V*, the formula (viii) (x = y) → □(x = y) is true. Free individual variables under an assignment are paradigms of directly referential terms, and so their denotations do not vary across possible worlds. In other words, they are rigid. Thus what (i) establishes is that any true identity statement constructed from two rigid designators is necessarily true. To deny that ‘Hesperus is Phosphorus’ is necessary, given that it is true, while maintaining that names are rigid, therefore, amounts to rejecting the way identity is handled in the SML. There are two different types of examples of necessary a posteriori identity in Kripke’s work: (a) Singular identities: identity claims involving singular terms that are rigid, such as ‘Tully is Cicero’ and ‘Hesperus is Phosphorus’. (b) Identities involving natural kind terms: we learned from Kripke and Putnam that ‘water’ can be seen as a rigid designator with its reference fixed as whatever substance that possess such superficial properties as being colorless, tasteless, the liquid that fills our lakes, and so forth—​in short, the property of being the watery stuff. Since it is an empirical, scientific discovery that the watery stuff in the actual world is H2O, ‘Water is H2O’ is necessary a posteriori, given that ‘H2O’ is taken as a rigid designator for a chemical structure. Similar examples include ‘Gold is the metal of atomic number 79’, ‘Heat is molecular motion’, and so on.11 In what follows, ‘the Kripkean examples’ will be use as a shorthand for these two types of examples when there is no need to distinguish them.

4  Rigidification and a World Considered as Actual Descriptivists need a strategy to respond to the Kripkean examples. The strategy must be able to:12 (S1) find associated descriptions (for names and natural kind terms) capable of withstanding the modal argument and (S2) the semantic argument, and

241

Kai-Yee Wong

(S3) explain the Kripkean examples. Two-​dimensionalists like Jackson and Chalmers think that two-​dimensional semantics provides a way to execute (S3) that will preserve Kripke’s insights.13 Let’s get there in steps and start with (S1). In addition to thinking about how things actually are, we are often disposed to evaluate the truth of sentences at various counterfactual states of the world or possible worlds. In such situations, we employ sentences with modal constructions such as ‘possibly’, ‘necessarily’, and ‘at such and such world w’ so as to talk about how things are in various counterfactual circumstances. But in some contexts things are more complicated. Consider how the following two sentences differ in meaning: Necessarily, if the sky is green then the sky is green. Necessarily, if the sky is green then the sky is actually green. As it is true in every possible world that if the sky is green then it is green, the first sentence makes the trivially true claim that the sky is green in any possible world in which the sky is green. The second, however, is false. The clause ‘the sky is actually green’ talks about what is going on in the actual world even though it is inside the scope of a modal operator, where normally we would be talking about other possible worlds. Using ‘actual’ allows us to refer back to the actual world when evaluating the truth of a sentence even though it is inside the scope of a modal operator. When it comes to the extension of a term, ‘actual’ allows us to refer back to the actual extension and thus achieve rigidification. Treating a description ‘the F’ as ‘the actual F’, therefore, will actualize it, turning it into a rigid designator.This, as Jackson proposed, provides a way to satisfy (S1). He thinks that the description theory should be refined using rigidification: names and natural kind terms are to be considered equivalent to actualized descriptions. With respect to a language that is rich enough, his proposal is in effect that any name or natural kind term is synonymous to ‘the actual F’, for some appropriate predicate F.14 Rigidification not only helps satisfy (S1) but also serves to connect (S1) to (S3) because it provides one more way to get an intension for the description that is rigidified. In possible worlds semantics, to get the (standard) intension for ‘the actual watery stuff is sweet’, we ask what the truth-​value of the sentence at each possible world w is, holding the actual world fixed. But what if we do the same thing but with each w considered as actual, or under the supposition that it is the actual world? We get a different kind of intension. With two kinds of intensions, we get the two-​dimensional semantic framework, from which comes a two-​dimensional way to satisfy (S3).15

Considering a Word as Actual Here we need to turn to two contrasting ‘attitudes’ towards possible worlds—​considered-​as-​actual and considered-​as-​counterfactual—​because of their connections to the two ways of getting intensions. Suppose I draw a card with its face down from a deck. I then turn it over and see a ♥2. That allows me to say that, of the fifty-​two possible outcomes, fifty-​one of them are merely possible or counterfactual. Let’s travel back to the moment right before the card drawn is turned over. For each of the fifty-​two alternatives, I can now consider each of them as actual, despite the fact that all except one of them (♥2) are merely possible. I can do that because each of them can be thought of as a supposition of the way the actual outcome turns out. On the other hand, for each of the fifty-​two possible outcomes I can consider it as merely possible or counterfactual, provided that some other outcome is being considered as actual. This includes even ♥2, which, given the way things actually are, is not a merely possible outcome.16 These contrasting attitudes are part of the intuitive basis of two-​dimensional semantics.To connect them to the notions of extension and truth and eventually to two-​dimensional intensions, we will 242

Semantics and Identity Statements

do well to consider Stalnaker’s important discussion on two different ways facts enter into the determination of truth and content. Let’s start with some familiar concepts in possible worlds semantics. A proposition is a representation of the world as being a certain way. When I  say ‘the sky is blue’, whether the proposition expressed is true is determined by what the world is like. Thus a proposition may be understood as the way truth-​values depend on facts, or, a function from possible worlds to truth-​values. Every proposition is then a way of picking out a set of possible worlds—​all those for which the proposition takes the value true. In this picture, each possible world w plays the role of circumstance of evaluation, and, as far as that role is concerned, it doesn’t seem to matter whether we consider w as actual or as counterfactual. Given that facts in w are such that snow is white in w, ‘Snow is white’ is true with respect to w. Considering w as actual or as counterfactual will not change the result of the evaluation, although knowing whether w is actual may sometimes matter. This said, how a world is considered as is important to the determination of the truth-​values because of a second way possible worlds, or facts, enter into such determination. Stalnaker writes, It is a matter of fact that an utterance has the content which it has. What one says—​the proposition he expresses—​is itself something that might have been different if the facts had been different. 1978, 70 This can best be illustrated by utterances containing indexicals, such as ‘you’, ‘I’, and ‘now’. Changing the context for ‘You are a fool’ will determine different contents for the sentence, which may be given different truth-​values with respect to the same evaluative circumstance. Here is a quick example. I said ‘You are a fool’ to Ted. Ted is a fool, so what I said was true (as evaluated at the actual world). Ted said the same to Carol, who is no fool, so, in that context, the sentence has a different content that is false with respect to the same evaluative circumstance, i.e. that actual world. Constraints on semantic content posed by facts about context of use gave rise to what has been called ‘two-​dimensional modal logic’ in the 1970s.17 It did not take long for it to be noticed that the formal two-​dimensional apparatus might shed light on Kripke’s problematic examples of identity statements. It is the basics of such an apparatus we now turn to. We learned from Kripke and Putnam that ‘water’ is a rigid designator for the substance that is the watery stuff. Suppose ‘water’ was introduced by way of a reference-​fixing description in terms of watery properties. Because the watery stuff in the actual world, W1, is the chemical compound H2O, ‘water’, being rigid, refers to H2O with respect to all counterfactual worlds. Using just a limited number of possible worlds, we may represent the extension of ‘water’ in each possible world by a one-​ dimensional matrix specifying what the word applies to in each world: W1

W2

W3

H2O

H2O

H2O

This function is the intension of ‘water’. In contrast, the intension of the description ‘the watery stuff ’ is: W1

W2

W3

H2O

XYZ

H2O

(assuming that W2 is like Putnam’s Twin Earth, in which the watery stuff is composed of a different chemical structure XYZ and that the watery stuff in W3 is H2O.)18 Given that the watery stuff is XYZ in W2, ‘water’ refers (rigidly) to XYZ under the supposition that W2 is the actual world: 243

Kai-Yee Wong

W1

W2

W3

XYZ

XYZ

XYZ

The intension of ‘water’ under the supposition that W3 is the actual world is the same as the first matrix. To represent the dependence of the intension of ‘water’ upon different suppositions of which world is actual, we need to use a two-​dimensional function: Ma W1

W2

W3

W1

H2O

H2O

H2O

W2

XYZ

XYZ

XYZ

W3

H2O

H2O

H2O

Matrices like Ma give a clear representation of the double role of a possible world. The vertical dimension represents possible worlds in their role as the actual world. They are A-​worlds, worlds considered as actual.The horizontal dimension represents possible worlds in their role as evaluation circumstance. They are C-​world (because only one of them is not counterfactual).19 The cells on the diagonal (the shaded ones) have a unique and crucial theoretical role to play: they represent a world playing the double role of an A-​world and a C-​world.20

5  Two Kinds of Intensions We can now define two kinds of extensions for any term T. The extension of T at a world w is what T applies to at w given whatever world is the actual world.21 Call this the C-​extension (simpliciter) of T. This is ‘extension’ in its unadorned sense. More generally, The C-​extension-​v (C-​extension relative to v) of T at w is what T applies to in w under the supposition that v is the actual world.22 The A-​extension of T at w is what the term applies to at w under the supposition that w is the actual world. In terms of C-​and A-​extensions, we define the C-​intension and A-​intension  of T: The C-​intension (simpliciter) of T is the function that, for each world w returns the C-​extension (simpliciter) at w. Generally, The C-​intension-​v of T is the function that, for each world w, returns the C-​extension-​v at w. The A-​intension of T is the function that, for each world w, returns the A-​extension of T at w, or equivalently, the function that, for each world w, returns what T applies to at w under the supposition that w is actual. For all terms, the A-​and C-​extensions at the actual world must be the same. For some terms, such as ‘circle’ and ‘the color of the sky’, the A and C-​extension in a world are always the same. To represent 244

Semantics and Identity Statements

the A-​or C-​intension of such a term, a one-​dimensional matrix will suffice. But this is generally not true for rigidified descriptions like ‘the actual winner of the race’, natural kind terms like ‘water’, and proper names like ‘Hesperus’. Let’s turn to sentences. Consider again ‘Water is H2O’. Given Ma and how things are in the actual world, the sentence (2) Water is H2O expresses a necessarily true proposition, represented by the top (or equivalently the bottom) row of the matrix below. Yet under the supposition that W2 is the actual world, (2) expresses a necessarily false proposition because what satisfies the relevant reference-​fixing description in W2 is not H2O but XYZ. So, we have the following matrix: Mb W1

W2

W3

W1

T

T

T

W2

F

F

F

W3

T

T

T

That is, apart from the ordinary, C-​proposition (i.e., the top row), we get another intension. It is the A-​intension or diagonal proposition, the function which gives the truth-​value of (2) at each world w under the supposition that w is the actual world. Mb is what Stalnaker (1978) calls a propositional concept. It is a two-​dimensional function from possible worlds to propositions (where propositions are themselves C-​intensions) or a function from pairs of worlds to truth-​values. We are now in a position to say what a general two-​dimensional modal analysis of necessary a posteriori identity is like. Using (2) as an example again, this can be done succinctly as follows because the key elements of the analysis have been encapsulated in our explanation of Mb: Given how things are actually, the statement that water is H2O states a necessary truth, as captured by the top row of T’s in Mb, i.e., the necessary C-​intension (simpliciter) of (2). Since it is a contingent fact that what fixes the reference of ‘water’, the watery properties, picks out the particular stuff that it does, one cannot rule out a priori that water is not H2O. In other words, (2) is true in some but not all worlds considered as actual. This feature is captured by the fact that the diagonal of Mb is contingent, that is, by the fact that (2) has a contingent A-​intension. In general, necessary a posteriori identity is a phenomenon to be explained by the different modal statuses of the C-​and A-​intensions involved.

6  Jackson’s Two-​Dimensionalism Jackson distinguishes two senses in which we understand some sentences without knowing the conditions under which they are true.The propositions expressed by our ‘water’ sentences depend on how things are in our world. Anyone who does not know that the watery stuff of our acquaintance is H2O does not know the truth conditions under which ‘Water covers most of the earth’ is true, in the sense that they ‘could know all there is to know about some counterfactual world without knowing whether the sentence is true in that world … through their ignorance about the actual world’. Nevertheless, they understand ‘Water covers most of the earth’ in the sense that they know ‘how the proposition expressed depends on context of utterance—​in this case, how it depends on which stuff in the world of utterance is the watery stuff of our acquaintance in it’ (1998b, 73–​74, emphasis added). It is not difficult to see that the proposition involved in the first sense is the C-​intension and in the second the A-​intension. Given this distinction, Jackson claims, the explanation of the Kripkean examples of identity is now straightforward. Indeed, it follows the general analysis presented above. ‘Water is 245

Kai-Yee Wong

H2O’ is necessary because its C-​proposition is necessary, and it is so because natural kind terms are rigidified descriptions. But understanding requires only knowing the A-​proposition. Therefore, one can understand the sentence without knowing enough to see that the sentence is necessary or even that it is true (see Jackson 1998b, 77). The truth here can be factored into a part that is necessary, the C-​proposition, and a part that can be known only a posteriori, the A-​proposition, which, however, is contingent. (S3) is thus satisfied. Jackson’s approach to the abstract two-​dimensional framework is based on a kind of contextual interpretation of A-​worlds. On this understanding, to consider a world as actual is to imagine a possible world in which one is located. So, an A-​world can be modeled on a world centered on a subject uttering a token of utterance, or a so-​called ‘centered world’.23 Indeed Jackson has used a paradigmatic case of context dependence to explain how one may understand ‘water’ sentences without knowing the [C-​] proposition expressed. Suppose I hear someone say ‘He has a beard’ without knowing who is being spoken of. In such a case, ‘I will understand what is being said without necessarily knowing the conditions under which what I said is true … [that is] I may not know which proposition is being expressed’ because ‘I know how to move from the appropriate contextual information … to the proposition expressed’ (Jackson 1998b, 73). He made ‘a similar point’ about ‘water’ sentences. Just as I can understand ‘He has a beard’ without knowing what ‘he’ refers to, those who do not know that ‘water’ refers to H2O can understand ‘Water covers most of the earth’ in the sense that they know how the proposition expressed depends on context of utterance—​in this case, how it depends on which stuff in the world of utterance is the watery stuff of our acquaintance in it. Generally, knowing the linguistic meaning of a sentence is a matter of ‘knowing how the proposition expressed depends on context’ (Jackson 1998b, 75).

7  Chalmers’s Two-​Dimensionalism The broadly rationalist tradition, as Chalmers noted, posited a ‘golden triangle’ of connections between three central notions of philosophy—​meaning (sense), reason (apriority), and modality (necessity).24 The Kripkean examples have threatened to break the triangle by severing the link between reason (a priority) and modality (necessity). Chalmers’s two-​dimensional semantics seeks to restore the triangle by re-​establishing a rationalist account of linguistic meaning that satisfies a thesis that very much sums up the triangle. He calls it the ‘Core Thesis’: Core Thesis: For any sentence S, S is a priori if and only if S has a necessary primary [A-​] intension. Chalmers 2006, 64 The Core Thesis in effect constrains how two-​dimensional semantics should be interpreted if it is to vindicate the rationalist conception of meaning. Chalmers has argued, however, that the contextual interpretation of A-​worlds is incompatible with the Core Thesis. Sentences such as ‘Utterances exist’ and ‘Language exists’ pose a pressing problem for the contextual interpretation. Such sentences, Chalmers argues, are obviously true in all contexts in which they are uttered and thus would have a necessary A-​intension (primary intension, in Chalmers’s words) on the contextual account. But such sentences are not plausibly a priori.25 So, the contextual approach cannot vindicate the Core Thesis. To satisfy the Core Thesis, Chalmers has proposed to understand A-​worlds as what he calls ‘scenarios’. On this understanding, scenarios are themselves characterized in terms of a priori coherence, that is, claims that cannot be ruled out for all one can tell on the basis of idealized a priori reasoning. Since the primary intension is defined in terms of scenarios, it is thus tied constitutively to the epistemic domain and satisfies the Core Thesis. 246

Semantics and Identity Statements

Scenarios afford a distinctively epistemic kind of possibilities. A scenario is an exhaustive description of the world that is a priori coherent. It represents a maximal hypothesis about what the world might be like for all one can know on the basis of idealized a priori reasoning.The fundamental intuition backing the use of scenarios is that there is a strong dependence of an expression’s extensions on scenarios: [If] we come to know that the world has a certain character, we are in a position to conclude that the expression has a certain extension. And if we were to learn that the world has a different character, we would be in a position to conclude that the expression has a different extension. That is: we are in a position to come to know the extension of an expression, depending on which epistemic possibility turns out to be actual. Chalmers 2004, 177–​178 Accordingly, grasping the meaning of an expression or a sentence is a matter of possessing everything required to know, in an idealized condition of a priori reflection, what the expression applies to or whether the sentence is true relative to any scenario sufficiently described. Let us spell out what all this means for the Core Thesis. If a sentence S has a necessary primary intension, then it is true in all a priori coherent hypotheses. That means any hypothesis that entails ~S must be a priori incoherent and so ~S can be ruled out a priori. Hence, S is true a priori. On the other hand, if S has a contingent primary intension, then S is false in some a priori coherent hypothesis.That means it cannot be ruled out a priori that ~S. Hence S is not true a priori.The Core Thesis is satisfied. In effect, on the epistemic approach, we take an epistemic notion (apriority) as basic and use it to define a modal space (the space of epistemic possibility), which is then used to define the epistemic (primary) intension. On this interpretation, the diagonal intension isolates an aspect of meaning that is constitutively connected with reason and modality. If we put aside the subtleties to do with Chalmers’s claim that ordinary terms like ‘water’ and ‘Hesperus’ must not be used in the construction of scenarios, the explanation of the Kripkean examples, on the epistemic approach, follows that of the general two-​dimensional modal analysis outlined above. Take the example of (2) again. The linguistic meaning of ‘water’ and (2) implies that Ma and Mb correctly model the two kinds of intensions associated with ‘Water’ and ‘Water is H2O’, given that A-​worlds are understood as epistemic possibilities. The sentence (or an utterance of it) is necessary as it has a necessary secondary (C-​) proposition. But given the Core Thesis, it is a posteriori on account of having a contingent primary (A-​) proposition. (S3) is thus satisfied.26

8  The Semantic Argument and Recipe Having explained how (S3) may be satisfied in a two-​dimensional way, I shall comment briefly on Jackson’s causal-​descriptivist answer to (S2). The purpose is twofold: not to leave (S2) unaddressed and to help connect the two-​dimensional answer to (S3) with the important interpretative problems to be discussed in the next section. In the Fregean tradition, the content of a descriptional term is not only the mechanism by which the reference of the term (with respect to a possible world) is semantically determined, but also a conceptual representation of the referent which a competent speaker associates with his or her use of the term. In other words, descriptivists are committed to the notion—​let’s call it Recipe—​that a fully competent speaker of a descriptional term “grasps” or possesses a kind of conceptual or internal “recipe” for securing an object as the referent of the term in every situation where his or her use of term successfully refers. More generally, Recipe says that semantic competence is sufficient for reference determination: to know the meaning of a referential expression consists in grasping a veridical criterion for determining to which thing it applies in every possible situation.

247

Kai-Yee Wong

The semantic argument threatens to invalidate Recipe. Recall that the argument rests on intuitively coherent scenarios (in the ordinary sense of the word, not as Chalmers uses it) in which a fully competent speaker S’s use of a name N refers successfully to an object O in spite of error or ignorance. Let’s call such scenarios ‘EI-​scenarios’. If Kripke’s causal account is correct, S’s use of N in an EI-​scenario refers to O because of some underlying causal link between his use of the word and O. To handle the EI-​scenarios, Jackson begins by remarking that in such debates as engendered by Twin Earth scenarios, there is ‘considerable agreement about what to call “water” and what not to call “water” in these various scenarios’ (Jackson 2004, 273). The impact and importance of the writing of critics of descriptivism, Jackson points out further, derive from this fact. That is, they derive ‘precisely from the intuitive plausibility of many of their claims about what refers, or fails to refer, to what in various possible worlds’ (Jackson 1998a, 212). But then, if speakers can say what to call ‘water’ when various possibilities are described to them, we can identify the representational property for the word ‘water’: it is the property that, often implicitly, guides them when they say which stuff, if any, in each possibility to call ‘water’ when presented with the various scenarios. When the guidance is implicit, the pattern that underlies the various verdicts will be one they cannot state in words. Jackson 2004, 272–​273 And ‘the same goes for the considerable agreement about what to call “gold”—​and what to call “Gödel”, “Aristotle”, or “life”, if it comes to that’ (Jackson 2004, 273). In other words, Jackson seems to think that the semantic argument raises no special problems for Recipe. In all EI-​scenarios, he would say, there is some guidance, albeit implicit, that can be recovered to serve as the representation property for the word. While it may be obvious what constitutes the guidance in a case like that of ‘water’—​the watery properties—​it seems doubtful that in the case of ‘Gödel’ or ‘Feynman’, or in any typical Error Case or Ignorance Case, there must be something that can be said to be strictly analogous to the watery properties. More importantly, the impact of the semantic argument derives precisely from the observation that in an EI-​scenario the speaker needs no guidance, implicit or otherwise, that he or she can be said to grasp by virtue of understanding the meaning of the word. The word manages to refer, if Kripke is correct, simply because it is causally linked in certain suitable ways to the referent. From the fact that speakers, thinking about the scenarios, or we philosophers, can say what to call ‘Gödel’ or ‘Feynman’ when certain EI-​scenarios are described to them, or us, it does not follow, as Jackson says, ‘we can identify the representation property of the word’.There may be no associated descriptions or conceptual recipe to be identified on which speakers must rely to secure reference. We can say what to call ‘Gödel’ or ‘Feynman’ in these scenarios because enough detail of a certain sort is described to us about each scenario. Such detail concerns causal facts that make it the case that the word refers to what it refers to. (And as such they are, arguably, facts pertaining to metasemantics rather than semantics; more on this in the next section.) It is one thing to say that there must be such facts or else the word would not have referred to what it refers, but it is another to say that such facts give us the needed representational properties as required by the description theory, that is, as a kind of meaning or sense in the broadly Fregean sense. But given that such causal facts are available, is there any obstacle preventing the descriptivist from exploiting them in a way that suits their need? A description that mimics such causal facts would seem available in each case. This is the causal-​descriptivist solution to the semantic argument, which Jackson endorses:

248

Semantics and Identity Statements

the description theorist can explain how I managed to refer in this kind of case by appeal to an association with the property of being a certain kind of causal origin (and maybe the property of being apt to be recognized as such by me). Jackson 1998a, 209 Causal-​descriptivism faces objections at two levels. Firstly, it can be objected that the properties that mimic or are ‘recoverable’ from the underlying causal facts would simply be so fancy as to make Recipe too demanding of the speakers’ conceptual resources. And the problem is compounded by reference borrowing, the well-​noted phenomenon that speakers often effectively ‘borrow’ their reference from speakers earlier in the causal chain, though borrowers need not be able to say who the lenders are.27 Secondly, it may be argued that appealing to the idea of an associated causal-​description that mimics the underlying causal chain is ill-​advised at a fundamental level. That is, it can be objected that the causal theory of reference is not a descriptive account of what the semantic values of names and natural kinds are but rather a metasemantic account. This brings us to the metasemantic/​semantics distinction, which provides a larger perspective for viewing the two-​dimensional approach to explicating the Kripkean examples.

9  Semantics and Metasemantics ‘Semantic value’, as we are using it, is a general and neutral term for the things that a semantic theory assigns to the expressions it interprets. We may distinguish two kinds of questions regarding semantic values: ‘descriptive questions’ and ‘foundational questions’, as Stalnaker labeled them. The distinction is introduced in Kaplan (1989a) to address a ‘crucial question’ about the status of the mechanisms of reference hypothesized by the causal theory of reference for names. Kaplan asks whether those mechanisms are part of the semantics of a language that states a semantic value for proper names, or whether they belong to a ‘metasemantic’ account that ‘tells us the basis for determining a semantic value for proper names’ (1989a, 574). In several of his influential works, Stalnaker applies the distinction to clarify the accomplishments of Kripke’s works and to cast light on interpretive questions regarding the two-​dimensional framework.28 According to Stalnaker (1997), a descriptive semantics or semantics is ‘a theory that says what the semantics for the language is without saying what it is about the practice of using that language that explains why that semantics is the right semantics’ (1997, 172). By contrast, the task of foundational semantics or metasemantics is to give an account of ‘what the facts are that give expressions their semantics values, or more generally, what makes it the case that the language spoken by a particular individual or community is a language with a particular descriptive semantics’ (1997, 166–​167). Clarifying the distinction allows one to see that the two sorts of enquires need not receive a single kind of answer. As a case in point, Kripke defends a Millian answer to the descriptive question and a causal answer to the metasemantic question regarding the semantic values of proper names and natural kind terms. One may object to either or even both of Kripke’s answers. But one must not equivocate between the two kinds of questions. It is helpful here to consider how one may agree with Kripke on his Millian approach but contest his answer to the foundational question, as in such special cases as that of Gareth Evans’s famous example, ‘Julius’, a (rigid) proper name stipulated to be the person (whoever he might be) who invented the zip (Evans 1979). In this example, what fixes the reference is not some causal mechanism but the satisfaction of an associated description, i.e., ‘the inventor of the zip’.29 Yet the name has the kind of semantic value that the Millian thinks it has. It is important, therefore, ‘not to bypass the metasemantic question by equivocating between the two kinds of questions’ (Stalnaker 2003a, 211). Such equivocation, Stalnaker argues (I think rightly), is part of what motivates descriptivism of names and natural kind terms.The objections above against causal-​descriptivism bore this out. Committed to a single answer for the two kinds of questions 249

Kai-Yee Wong

and finding the causal theory too convincing to reject, the causal-​descriptivist was committed to a semantic theory that mimics the causal theory of reference and so was driven to countenance an implausibly fancy kind of causal-​representational property, as already noted. Applying this distinction to the abstract two-​dimensional apparatus, we may have a semantic interpretation or a metasemantic one. On the former interpretation, the two-​dimensional intensions are what a correct descriptive semantic theory for a language should assign to expressions, and so it is part of the descriptive semantics of English that, say, ‘water’ and ‘Water is H2O’ should be associated, respectively, with the two-​dimensional functions Ma and Mb. These intensions are within the epistemic reach of anyone who knows the semantics of the language: they should have some kind of a priori access to the intensions by virtue of their linguistic competence. A speaker of English can understand ‘water’ without knowing water is H2O because linguistic understanding is a matter of knowing the diagonal proposition of Ma, which says how the reference of the term depends on what stuff might turn out to be watery. If this seems to echo what was said above about Jackson’s or Chalmers’s accounts, this is because their approaches to two-​dimensional semantics are semantic. For them, the A-​intension and C-​intension are two kinds of linguistic meanings. In other words, they have proposed their two-​dimensional accounts as a conception of linguistic meaning and, in doing so, subscribed to the semantic interpretation. The contrasting, metasemantic, interpretation assumes less about what a proper account of the semantics of a language associates with expressions of the language. The metasemantic understanding begins with a certain observation and then seeks to give a descriptively adequate representation of it metasemantically. The observation is the hardly controversial one that there is a second way in which facts entered into the determination of the truth-​value of sentences, as noted in section 4: it is a matter of empirical fact that the expressions have the semantic values that they have. The semantic values of an expression may be different in two possible worlds for a variety of reasons: meaning change, context dependence, different objects being the origin of a certain causal chain of reference, or different individuals satisfying a stipulation in a ‘Julius’-​sort of way. Whatever these facts are and whatever the semantic value a semantic theory associates an expression with, a two-​dimensional intension can be used to represent the way the semantic value of the expression depends on such facts (assuming that the semantic value is a one-​dimensional function). On the metasemantic interpretation, two-​dimensional intensions as a whole are not regarded as belonging to the semantics itself, but rather derivative from the semantic values that an expression has in different possible worlds. Regarding necessary identity statements such as ‘Hesperus is Phosphorus’ and ‘Water is H2O’ that seem to convey contingent information, the general metasemantic strategy for clarifying them is to begin by asking not how it is that the sentence manages to convey that information but rather ‘what kind of possible world would be excluded by one who told someone that the statement was false’ (Stalnaker 2007, 260).30 Given the assumption that ‘Hesperus is Phosphorus’ is interpreted in the standard way,31 it is a world in which ‘Hesperus’ denotes a different object from what ‘Phosphorus’ does. What an utterance of ‘Hesperus is Phosphorus’ expresses in our world, according to the Millian story, will still be true in this kind of possible world but the diagonal proposition defined on a metasemantic two-​dimensional function ‘for the utterance in question will be false in such a world, and so will exclude it’ (Stalnaker 2007, 261). The exclusion of this kind of possible world gives an adequate descriptive representation of the information conveyed by the statement. Following the proposal of Stalnaker (1978, 2004), there is another way to tell very much the same the story, but in terms of the Gricean maxims of conversation. In a rational discourse, it is presumed that certain maxims of conversation are being followed. This presumption constrains the interpretation of what is said. Speakers may exploit this presumption by saying things that would manifestly violate certain maxims if interpreted in the standard way so that a reinterpretation will be required. Since necessary truths by themselves are hardly informative, an assertion of ‘Hesperus is Phosphorus’, if it is to

250

Semantics and Identity Statements

be considered informative, will be a prima facie violation of the maxim that speakers presume that their addressees understand what they say. So, we are driven by the Gricean mechanism to reinterpret the sentence by diagonalization so as to bring the statement into conformity with conversation rules. On this account, diagonalization is reinterpretation, not the standard, semantic interpretation. Hence, ‘Hesperus is Phosphorus’ is a posteriori not because the correct semantics says that it has a contingent A-​intension but because an utterance of it can be reinterpreted in the way just described. One cannot stress enough the importance of the metasemantics/​semantics distinction for the very idea of a two-​dimensional explanation of the necessary a posteriori. The semantic interpretation is a kind of semantic internalism that builds the means by which a term acquired its reference into the semantic properties of the term and thus assumes that two-​dimensional meanings are cognitively accessible to a person who uses the terms with those semantic properties. A correct semantics of a language will then provide two-​dimensional meanings that are determined by the internal states of the speakers of the language, or put in another way, that are a matter of the decisions of the speakers to speak a language with that semantics. For those who favor the kind of externalist picture of reference and intentionality in Kripke’s work, it seems more plausible to think that semantics provide no such things.32 Indeed we have seen how Jackson’s causal-​descriptivist response to the semantic argument has rendered Recipe implausibly demanding with regard to speakers’ conceptual resources. And it is at least arguable that Chalmers’s appeal to idealization only fudges a similar and pressing problem faced by his account.33 Semantically interpreted two-​ dimensional semantics provides cognitively accessible two-​ dimensional meanings that are determined by the internal states of speakers. This seems to promise an explanation of the Kripkean examples and, at the same time, a reconciliation of his insights (and similar ones by Putnam and others) with the Fregean-​descriptivist picture of reference and meaning. From the externalist perspective, however, such attraction has come largely from a kind of equivocation of the semantic and the metasemantic.

Further Reading Chalmers, David J. (2006). ‘The Foundations of Two-​dimensional Semantics’, in M. Garcia-​Carpintero and J. Macia (eds.), Two-​Dimensional Semantics, Oxford, Oxford University Press, pp. 55–​140. The core of this paper develops the epistemic understanding of two-​dimensional framework. It also provides a helpful taxonomy of different approaches to the two-​dimensional ideas. Jackson, Frank (1998). From Metaphysics to Ethics:  A Defence of Conceptual Analysis, Oxford, Clarendon Press. Chapters 2 and 3 elucidate the connections between the two kinds of intensions and conceptual analysis, an important topic not covered in this entry. LaPorte, Joseph (2013). Rigid Designation and Theoretical Identities, Oxford, Oxford University Press. The book offers an elaborate account of the connections between the reference of words for properties and kinds, and theoretical identity statements. Essays in Philosophical Studies Volume 118, Issue 1–​2, March 2004. Special issue on the two-​dimensional framework, entitled ‘The Two-​Dimensional Framework and Its Applications: Metaphysics, Language, and Mind’. It includes a detailed account of formal features of the framework and its application, as well as the contingent a priori, a topic not covered in this entry. Salmon, Nathan (1982/​2005). Reference and Essence, 2nd ed., New York, Prometheus Books. Almost four decades on, Salmon’s definitive characterization of the descriptional theory and clarification of direct referentiality remains unsurpassed. Stalnaker, Robert (1997). ‘Reference and Necessity’, in Bob Hale and Crispin Wright (eds.), A Companion to the Philosophy of Language, Oxford, Basil Blackwell. Stalnaker argues that the most important accomplishment of Naming and Necessity was to separate semantic questions about reference from metaphysical questions about objects of reference. Stalnaker, Robert (2004). ‘Assertion Revisited: On the Interpretation of Two-​Dimensional Modal Semantics’, Philosophical Studies 118: 299–​322. This paper concerns the applications of two-​dimensional modal semantics to the explanation of the contents of speech and thought and different interpretations and applications of the apparatus are contrasted.

251

Kai-Yee Wong

Notes 1 For a historical account of the development of the two-​dimensional ideas and the two-​dimensional framework, see Chalmers (2006) and the editors’ introduction to Garcia-​Carpintero and Macia (2006). 2 This means I have to pass over two related topics Kripke has discussed. (a) Kripke’s work also shed light on another phenomenon that manifests mainly in identity statements, that is, the contingent a priori. See Davies (2006), Evans (1979), Geirsson (1991), Stalnaker (2001), and Williamson (1986). (b) Apart from identities sentences, Kripke has discussed another type of alleged necessary a posteriori truths, i.e., attributions of essential properties to things (e.g., properties to do with origins or constitution). Discussing them would mean considering some complicated issues to do with compossibility and the principle of cross-​world identification. See, e.g., Forbes (1980, 1985) and Salmon (2005, ch. 7). 3 According to certain refinements of Frege’s account, the sense of an expression is identified with a ‘cluster’ of properties, rather than a conjunction. There have been other refinements. See Searle (1958, 1967), Linsky (1977), and Salmon (2005). See also Chalmers (2002). In this entry I  will leave undiscussed debates over Kripke’s interpretation of Sinn and how close Frege’s own view really is to the kind of descriptivist picture depicted in Naming and Necessity. See Dummett (1981, appendix to ­chapter 5), Kripke (2008), Linsky (1977), Salmon (2005), and Soames (2002). 4 This is based on Burge’s trifurcation in Burge (1977). As Salmon (2005, ch. 1) points out, the three-​way identification of the three different senses (corresponding to the three different roles) of a descriptional term is a very strong theoretical claim on the part of the Fregean theorist. 5 The expression ‘directly referential’ was introduced in Kaplan (1977). I  will not offer reviews of the contributions of other champions of the theory of direct reference. For such a review, see Salmon (2005). See also Soames (2002). 6 It should be noted that the causal account does not deny that one may also fix the reference of a name by means of a description without any causal contact with the referent so introduced. See the ‘Julius’ example below. 7 The semantic argument is not one but a set of similar arguments. Though the arguments have been put forward by Donnellan, Kaplan, Kripke, and Putnam, they are generally regarded as due mainly to Donnellan (1972) and Kripke (1972). It should also be noted that in addition to the modal and semantic arguments, Kripke has put forward an ‘epistemological argument’, which will not be discussed here. See Salmon (2005, section 2.2). for the essence of the argument. Regarding this argument, Speaks (2010) deserves mention. The paper argues that Chalmers’s epistemic two-​dimensionalism is open to a close relative of the epistemological argument. 8 Here I am rephrasing a succinct formulation by Stalnaker (2008, 12). 9 For the traditional view, see Ayer (1946) and Pap (1958). For some related discussion, see Horowitz (1985), Plantinga (1974), Quinton (1964), and Swinburne (1975). 10 Kripke (1971, 139ff) and Smullyan (1948) explain this clearly. 11 It is worth noting that identity statements with natural kind terms have given rise to some difficult interpretive issues, such as whether they really are identity statements and what it means for general terms or natural kind terms to be rigid designators. For some excellent discussion on these issues, see LaPorte (2013), Salmon (2005), and Soames (2002). 12 See Soames (2005, 38), for a slightly different version of such a strategy. 13 It has been suggested that two-​ dimensionalists can do this partly because there are sources of two-​ dimensionalism in Kripke’s works. See, for instance, Chapter  2 of Soames (2005) (‘Roots of Two-​ Dimensionalism in Kaplan and Kripke’) and Garcia-​Carpintero and Macia’s introduction to their 2006 book, which starts with ‘Kripke’s 2-​D Intimations’. 14 Jackson says that things would be easy if ‘everyone agrees that “water” is equivalent to “the actual watery stuffy” ’. For if so, this ‘would make “There is water” and “There is H2O” true at exactly the same worlds’, as Kripke would have us think. ‘However’, he adds, ‘it is far from true that everyone agrees that “water” is equivalent to “the actual watery stuff ” ’ (Jackson 2004, 263.) He then proceeds to discuss three levels of such disagreement and indicate how each of them can be settled in favor of the ‘actualized description’ view. Handling Kripke’s modal argument with actualized descriptions is not new. See Stanley (1997) for a detailed discussion of this approach. See also Pettit (2004). 15 This way to satisfy (S3) is the general two-​dimensional modal analysis of necessary a posteriori identity to be outlined at the end of section 5. The analysis derives from writings on Naming and Necessity by Davies and Humberstone (1980), Kaplan (1989a, 1989b), Lewis (1981), and Stalnaker (1978). 16 Theorists discussing two-​dimensionalism sometimes seem to suggest that these two contrasting attitudes can be ‘clarified’ in terms of two contrasting types of conditional judgments, that is, the counterfactual and the indicative. But it seems plausible to think that the former contrast is intuitively as fundamental as, if not more

252

Semantics and Identity Statements fundamental than, the latter contrast. Hence, I eschew this kind of rather common way of ‘clarification’ (see, e.g. Schroeter 2005). 17 From the early 1970s through mid-​1980s, constraints on semantic content posed by facts about context of use and the influence of context on the determination of truth-​values constituted the central topics in such areas as formal pragmatics, context-​sensitive semantics, logical pragmatics, double-​index semantics, and two-​ dimensional modal logic. The approach to the necessary a posteriori discussed in this entry is an application of an abstract two-​dimensional semantic framework that is a synthesis of results from investigations in the above areas. See e.g. Åqvist (1973), Bar-​Hillel (1954), Hansson (1974), Kaplan (1979, 1989b), Kamp (1971), Lewis (1972, 1981), Montague (1970), Segerberg (1973), and Stalnaker (1978, 1980, 1981). 18 In Putnam’s original thought experiment, Twin Earth is another planet that is exactly like Earth except in its watery aspect. The problems with thinking of Twin Earth as a possible world are discussed in Putnam (1990). 19 I am following Jackson’s terminology here, who calls the two kinds of intensions ‘A-​ intensions’ and ‘C-​intensions’. 20 Whether the A-​worlds are indeed worlds is a substantive question since the A-​worlds might outstrip the C-​worlds—​i.e., there might be scenarios that don’t correspond to any C-​world, as, e.g., some a posteriori physicalists think, and correspondingly, there might be what Chalmers sometimes calls strong necessities. (I owe this remark to the editors.) 21 In phrases like ‘the extension of a term in (or at) world w’ and ‘what a term applies to in (or at) w’, the two prepositions are used to mean (more accurately) ‘with respect to’. 22 That is, the extension (simpliciter) of T in w is the C-​extension-​v of T in w when v is replaced by an implicit index ‘our world’. I will simply use ‘C-​extension’ without a world-​index when the context is clear as to what the index is. 23 The term ‘centered worlds’ is due to Quine (1969). He means by ‘centered possible world’ an ordered pair consisting of a possible world and a center (consisting of a time and an individual in the world). The center is necessary for cases involving indexical terms such as ‘I’. See also Chalmers (1996, 60–​61). 24 The triangle resulted from the combined effort of Kant, Frege, and Carnap. Necessary truths are knowable a priori and vice versa (Kant), and, as Frege’s notion of sense suggests, cognitive significance (reason) is an aspect of meaning. Since Carnap’s notion of intension is more well-​defined than Frege’s sense, it promised to ground, as Carnap intended, a Fregean link between meaning and reason via modality. (Carnapian intensions are functions from state-​descriptions (possible worlds) to extensions.) See Chalmers 2006. 25 Schroeter (2005) powerfully argues that Chalmers’s account faces the very same type of objections he takes to challenge the contextualist interpretation. Schroeter argues that ‘in order to get an authoritative interpretation of a token representation like “water”, we must assume that that very token is causally hooked up in an appropriate way with the subject’s past representational practice with the term and ultimately with items in the subject’s environment’. Thus, if she is right, ‘no extension can be assigned for such words unless language, thought, and a thinking subject exist in the world considered as actual’ (2005, 346). 26 Here I gloss over some subtleties that for our purpose won’t matter. Suffice it to note the following: on the epistemic approach to scenarios, most ordinary proper names and familiar expressions must not be used in the construction of a scenario on pain of severing the close link between epistemic possibility and metaphysical possibility. The preservation of such a link is required if certain kinds of inferences from conceivability to metaphysical possibility are to be validated. This point bears on some highly elaborated arguments Chalmers has put forward for generalized scrutability, a bold thesis which says that no matter how the world had turned out, all truths would have been a priori scrutable (on idealized reflection) from a compact base. We may clarify this point briefly in a simple way relevant to our issues at hand. If epistemic intension holds the key to explaining the aposteriority of ‘Water is H2O’, then epistemic possibility and metaphysical possibility must be connected in such a way as to allow the construction of at least one scenario E at which ‘Water is not H2O’ is true. But given the necessity of the identity between water and H2O, ‘Water is not H2O’ cannot figure in the construction of such a scenario E without rendering E metaphysically impossible. The case can be generalized to identity statements using only proper names. So E must be specified using a restricted vocabulary in a canonical language that excludes ‘water’ without depriving it a reference with respect to the scenario. The same applies to a large class of ordinary expressions, including ordinary natural kind terms, proper names, and descriptions. Chalmers argued in painstaking detail for the scrutability thesis in Chalmers (2012). The thesis is the focal point of a Carnap-​inspired project of developing a picture of reality on which all truths can be derived from a limited class of basic truth. 27 To use an example given by Jackson himself: ‘It is the property of having the representational property, whatever it is, of the word “quark” for the user that they are borrowing from. And what is the representational property for this group of users for the term “quark”? Either they are the experts, in which case it is property Q (whatever it is), or they are not the experts, in which case it is the property of having the representational property, whatever

253

Kai-Yee Wong it is, of the word “quark” for the user that they in turn are borrowing from. And so on’ (Jackson 2004, 271). Compare a similar proposal made more recently by McGinn: ‘When I use the name “Plato”, the description that is implicit in my understanding is “the person whose body was at the origin of that causal chain”, where with the demonstrative I refer to the chain of speech acts that originates in the baptism of Plato (his little baby body squirming there) and continues on through the centuries to the uses of “Plato” today … In other words, we incorporate the existent causal chains involving uses of names into a demonstrative description of such chains and then identify that description with the sense of the name’ (McGinn 2012, 116).Versions of causal descriptivism have also been suggested by, e.g., Kroon (1987) and Lewis (1984) and have enjoyed some popularity. 28 Mainly Stalnaker (1997, 2001, 2003b, and 2004). 2001 and 2003b in particular are devoted to bringing out the significance of the contrast between the semantic and the metasemantic interpretation of the two-​ dimensional apparatus; they also serve as the basis of our discussion below on this contrast. 29 Kripke did recognize that the reference of a name could be fixed in the way ‘Julius’ was. But he argued that the references of names were typically not fixed that way. 30 My articulation of this strategy follows closely Stalnaker (2007, 260–​261). 31 The standard way is:  names and natural kinds are rigid designators and their references are fixed causal-​ historically in the kind of way Kripke has suggested. 32 Stalnaker tries to press this point home in (2001, 2003b, 2004, and 2007). 33 One may think that the epistemic account is immune to the problem as it appeals to canonical descriptions of scenarios and idealization. Idealization, colorfully characterized in terms of what Chalmers calls a ‘cosmoscope’ (Chalmers 2012, ch. 3), ‘provides a convenient bridge between “worlds” considered in their logical roles and the “conceptual grasp” that accompanies a speaker’s current cognitive capacities’ (Wilson 2014, 238). But idealization, like Jackson’s appeal to implicit knowledge, only fudges the issue. Since idealization is built into the notion of a scenario, the primary intension is also idealized. How cognitively demanding Chalmers’s epistemic intensions are is therefore difficult to pin down. Considering also that scenarios can only be described completely in a canonical language not including ordinary names, it makes it doubly difficult to pin down how demanding his account is of the conceptual or cognitive capacity of a competent but not idealized English speaker who grasps the meaning of, say, ‘Gödel’.

References Åqvist, L. (1973).‘Modal Logic with Subjunctive Conditionals and Dispositional Predicates’, Journal of Philosophical Logic 2: 1–​76. Ayer, A. J. (1946). Language,Truth and Logic, 2nd ed., London, Gollancz. (1st ed. 1936.) Bar-​Hillel,Yehoshua (1954).‘Indexical Expressions’, Aspects of Language, Jerusalem, Magnes Press (1970), pp. 69–​88. Burge, Tyler (1977). ‘Belief De Re’, Journal of Philosophy 74: 338–​362. Carnap, Rudolf (1947). Meaning and Necessity:  A Study in Semantics and Modal Logic, 2nd ed., Chicago, IL, University of Chicago Press. Chalmers, David (1996). The Conscious Mind, Oxford, Oxford University Press. Chalmers, David (2002). ‘On Sense and Intension’, Philosophical Perspectives, 16: 135–​182. Chalmers, David (2004). ‘Epistemic Two-​Dimensional Semantics’, Philosophical Studies 118: 153–​226. Chalmers, David (2006).‘The Foundations of Two-​dimensional Semantics’, in M. Garcia-​Carpintero and J. Macia (eds.), Two-​Dimensional Semantics, Oxford, Oxford University Press, pp. 55–​140. Chalmers, David (2012). Constructing the World, Oxford, Oxford University Press. Davies, Martin (2006).‘Reference, Contingency, and the Two-​Dimensional Framework’, in M. Garcia-​Carpintero and J. Macia (eds.), Two-​Dimensional Semantics, Oxford, Oxford University Press, pp. 141–​175. Davies, Martin and I. L. Humberstone (1980). ‘Two Notions of Necessity’, Philosophical Studies 38: 1–​30. Donnellan, Keith. (1972). ‘Proper Names and Identifying Descriptions’, in D. Davidson and G. Harman (eds.), Semantics of Natural Language, Dordrecht, Reidel, pp. 356–​379. Dummett, Michael ([1973]1981). Frege: Philosophy of Language, 2nd ed., London, Duckworth. Evans, Gareth (1979). ‘Reference and Contingency’, The Monist 62: 161–​189. Forbes, Graeme (1980). ‘Origin and Identity’, Philosophical Studies 37, 353–​362. Forbes, Graeme (1985). The Metaphysics of Modality, Oxford, Clarendon Press. Frege, G. (1892). ‘On Sinn and Bedeutung’, in M. Beany (ed.), The Frege Reader, Oxford, Blackwell (1997), pp. 151–​171. Garcia-​Carpintero, M. and J. Macia (2006). Two-​Dimensional Semantics, Oxford, Oxford University Press. Geirsson, Heimir (1991). ‘The Contingent A Priori:  Kripke’s Two Types of Examples’, Australasian Journal of Philosophy 69: 195–​205. Hansson, B. (1974). ‘A Program for Pragmatics’, in S. Stenlund (ed.), Logical Theory and Semantic Analysis, Dordrecht, D. Reidel, 163–​174.

254

Semantics and Identity Statements Horowitz, T. (1985). ‘A Priori Truth’, Journal of Philosophy 82: 225–​239. Jackson, Frank (1998a). ‘Reference and Description Revisited’, Philosophical Perspectives 12: 201–​218. Jackson, Frank (1998b). From Metaphysics to Ethics, Oxford, Oxford University Press. Jackson, Frank (2004). ‘Why We Need A-​Intensions’, Philosophical Studies, 118: 257–​277. Kamp, J. A. W. (1971). ‘Formal Properties of “Now”’, Theoria 37: 227–​273. Kaplan, David (1977). ‘Demonstratives’, unpublished manuscript, UCLA Department of Philosophy, later published as Kaplan (1989b). Kaplan, David (1979). ‘On the Logic of Demonstratives’, in P. A. French, T. E. Uehling, and H. K. Wettstein (eds.), Contemporary Perspectives in the Philosophy of Language, Minneapolis, MN, University of Minnesota Press, pp. 401–​414. Kaplan, David (1989a).‘Afterthoughts’, in J.Almog, J. Perry, and H.Wettstein (eds.), Themes from Kaplan, New York, Oxford University Press, pp. 565–​614. Kaplan, David (1989b). ‘Demonstratives:  An Essay on the Semantics, Logic, Metaphysics, and Epistemology of Demonstratives and Other Indexicals’, in J. Almog, J. Perry, and H. Wettstein (eds.), Themes from Kaplan, Oxford, Oxford University Press, pp. 481–​564. Kripke, Saul (1971). ‘Identity and Necessity’, in M. Munitz (ed.), Identity and Individuation, New York, New York University Press, pp. 135–​164. Kripke, Saul (1972/​1980). ‘Naming and Necessity’, in D. Davidson and G. Harman (eds.), Semantics of Natural Language, Dordrecht, Reidel. Book edition:  Naming and Necessity, Oxford, Basil Blackwell, 1980. (Pages numbers refer to book edition.) Kripke, Saul (2008). ‘Frege’s Theory of Sense and Reference: Some Exegetical Notes’, Theoria 74: 181–​218. Kroon, Frederick (1987). ‘Causal Descriptivism’, Australasian Journal of Philosophy 65: 1–​17. LaPorte, Joseph (2013). Rigid Designation and Theoretical Identities, Oxford, Oxford University Press. Lewis, David (1972). ‘General Semantics’, in D. Davidson and G. Harman (eds.), Semantics of Natural Language, Dordrecht, Reidel, pp. 169–​218. Lewis, David (1981). ‘Index, Context, and Content’, in S. Kanger and S. Ohmann (eds.), Philosophy and Grammar, Dordrecht, Reidel, pp. 79–​100. Lewis, David (1984). ‘Putnam’s Paradox’, Australasian Journal of Philosophy 62: 221–​236. Linsky, Leonard (1977). Names and Descriptions, Chicago, IL, University of Chicago Press. McGinn, Colin (2012). Truth by Analysis, Oxford, Oxford University Press. Mill, John Stuart (1879). ‘Of Names’, in E. Nagel and R. B. Brandt (eds.), Meaning and Knowledge: Systematic Readings in Epistemology, New York, Harcourt, Brace & World, 1965, pp. 60–​87. Montague, Richard (1970). ‘Pragmatics and Intensional Logic’, Synthese 22: 68–​94. Pap, Arthur (1958). Semantics and Necessary Truth: An Inquiry into the Foundations of Analytic Philosophy, New Haven, CT,Yale University Press. Pettit, Philip (2004). ‘Descriptivism, Rigidified and Anchored’, Philosophical Studies 118: 323–​338. Plantinga, Alvin (1974). The Nature of Necessity, Oxford, Clarendon Press. Putnam, Hilary (1975). ‘The Meaning of “Meaning”’, in Mind, Language and Reality:  Philosophical Papers 2, Cambridge, Cambridge University Press, pp. 215–​271. Putnam, Hilary (1990). ‘Is Water Necessarily H2O’, in James Conant (ed.), Realism with a Human Face, Cambridge, MA, Harvard University Press, pp. 54–​79. Quine, W. V. (1969). ‘Propositional Objects’, in Ontological Relativity and Other Essays, New  York, Columbia University Press, pp. 139–​160. Quinton, Anthony (1964). ‘The A Priori and the Analytic’, Proceedings of the Aristotelian Society 64: 31–​54. Russell, Bertrand (1905). ‘On Denoting’, Mind n.s. 14: 479–​493. Russell, Bertrand (1920). ‘Descriptions’, in Introduction to Mathematical Philosophy, 2nd ed., London, Allen and Unwin, Ch. 16. Salmon, Nathan (2005). Reference and Essence, 2nd ed., New  York, Prometheus Books. (First edition by Blackwell, 1982.) Salmon, Nathan (1986). Frege’s Puzzle, Cambridge, MA, MIT Press. Salmon, Nathan (1991). ‘How Not to Become a Millian Heir’, Philosophical Studies 62: 165–​177. Salmon, Nathan (2005). ‘Are General Terms Rigid?’, Linguistics and Philosophy 28: 117–​134. Schroeter, Laura (2005). ‘Considering Empty Worlds as Actual’, Australasian Journal of Philosophy 83: 331–​347. Searle, John (1958). ‘Proper Names’, Mind 67: 66–​173. Searle, John (1967). ‘Proper Names and Descriptions’, in P. Edwards (ed.), The Encyclopedia of Philosophy, Vol. 6, New York, Macmillan and The Free Press, pp. 487–​491. Segerberg, K. (1973). ‘Two-​Dimensional Modal Logics’, Journal of Philosophical Logic 2: 77–​96. Smullyan, A. F. (1948). ‘Modality and Description’, Journal of Symbolic Logic 13: 31–​37. Soames, Scott (2002). Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity, Oxford, Oxford University Press.

255

Kai-Yee Wong Soames, Scott (2005). Reference and Description:  The Case against Two-​Dimensionalism, Princeton, NJ, Princeton University Press. Speaks, Jeff (2010). ‘Epistemic Two-​ Dimensionalism and the Epistemic Argument’, Australasian Journal of Philosophy 88: 59–​78. Stalnaker, Robert (1978).‘Assertion’, in P. Cole (ed.), Syntax and Semantics,Vol. 9: Pragmatics, New York, Academic, pp. 315–​332. Stalnaker, Robert (1980). ‘Logical Semiotic’, in E. Agazzi (ed.), Modern Logic:  A Survey, Dordrecht, Reidel, pp. 439–​456. Stalnaker, Robert (1981). ‘Indexical Belief ’, Synthese 49: 129–​151. Stalnaker, Robert (1997). ‘Reference and Necessity’, in Bob Hale and Crispin Wright (eds.), A Companion to the Philosophy of Language, Oxford, Basil Blackwell. (Page numbers refer to reprint in Stalnaker 2003a.) Stalnaker, Robert (2001). ‘On Considering a World as Actual’, Proceedings of the Aristotelian Society. (Page numbers refer to reprint in Stalnaker 2003a.) Stalnaker, Robert (2003a). Ways a World Might Be:  Metaphysical and Anti-​metaphysical Essays, Oxford, Oxford University Press. Stalnaker, Robert (2003b). ‘Metaphysics and Conceptual Necessity’, in Ways a World Might Be: Metaphysical and Anti-​metaphysical Essays, Oxford, Oxford University Press, pp. 201–​216. Stalnaker, Robert (2004). ‘Assertion Revisited: On the Interpretation of Two-​Dimensional Modal Semantics’, Philosophical Studies 118: 299–​322. Stalnaker, Robert (2007). ‘Critical Notice of Scott’s Case against Two-​Dimensionalism’, The Philosophical Review 116: 251–​266. Stalnaker, Robert (2008). Our Knowledge of the Inner World, Oxford, Oxford University Press. Stanley, Jason (1997). ‘Names and Rigid Designation’, in Bob Hale and Crispin Wright (eds.), A Companion to the Philosophy of Language, Oxford, Basil Blackwell, pp. 555–​585. Swinburne, R. G. (1975). ‘Analyticity, Necessity and A Priority’, Mind 84: 225–​243. Williamson, Timothy (1986). ‘The Contingents A Priori:  Has It Anything to Do with Indexicals?’, Analysis 46: 113–​117. Wilson, Mark (2014). ‘David Chalmers versus the Boll Weevil’, Philosophy and Phenomenological Research, 89: 238–​248.

256

19 TWO-​DIMENSIONALISM AND THE FOUNDATION OF LINGUISTIC ANALYSIS Andrew Melnyk

Can two-​dimensional semantics provide a foundation for linguistic analysis, i.e., the conceptual analysis of words in a natural language? In what follows, I make a case for skepticism. I understand linguistic analysis to be the use of a certain method—​the method of hypothetical cases—​to gain a priori knowledge of certain necessary truths. The method consists of asking oneself, for a variety of hypothetical situations, whether one would apply a given word to something in that hypothetical situation. The necessary truths that this method (together with further reflection) supposedly enables one to know a priori are necessary truths that can be formulated by using the given word, e.g., necessary truths that express necessary and sufficient conditions for the correct application of the word. To illustrate, consider how (according to advocates of linguistic analysis) one might come to know that, necessarily, being an unmarried man is a necessary and sufficient condition of being a bachelor. Suppose that, for all of the (sufficiently varied) hypothetical situations one considers, one would say that ‘bachelor’ applies to all the unmarried men in that situation. That suggests that being an unmarried man is a sufficient condition for being a bachelor. And suppose that, for all of the (sufficiently varied) hypothetical situations one considers, one would say that ‘bachelor’ doesn’t apply to all people who fail to be unmarried men in that situation. That suggests that being an unmarried man is also a necessary condition of being a bachelor. Now if one comes to know that bachelors are unmarried men through linguistic analysis, one must have understood the term, ‘bachelor’. And whatever exactly is the process by which one comes to understand words, it surely requires perceptual experience of the world. So understanding ‘bachelor’ requires that one have had perceptual experience of the world. But it does not follow that the knowledge that, necessarily, being an unmarried man is a necessary and sufficient condition of being a bachelor would be a posteriori if acquired through linguistic analysis. Such knowledge would still count as a priori if, to gain it, one would require no further perceptual experience than is required for coming to understand ‘bachelor’ in the first place. And advocates of linguistic analysis claim, of course, that this requirement is met. The method of hypothetical cases can only be used by someone who understands ‘bachelor’; but if one does understand ‘bachelor’, then one needs no further perceptual experience either to use the method of hypothetical cases or to draw a conclusion from the results of doing so. And ‘further’ perceptual experience includes two things: not only perceptual experience had after one comes to understand ‘bachelor’, but also perceptual experience, had before one comes

257

Andrew Melnyk

to understand ‘bachelor’, that goes beyond what is required for coming to understand ‘bachelor’. Intuitively, the knowledge that, necessarily, being an unmarried man is a necessary and sufficient condition for being a bachelor would count as a priori if acquired through linguistic analysis because merely understanding ‘bachelor’ is sufficient (together with certain capacities for a priori reasoning) for one to acquire the knowledge.1 Linguistic analysis is the use of a certain method to achieve a certain goal. But it is appropriate to use a method to achieve a goal only if the method is capable of achieving the goal. So linguistic analysis is appropriate only if the method of hypothetical cases is capable of generating a priori knowledge of necessary truths. But how it does so cannot be magic. So linguistic analysis is appropriate only if there is some explanation of how the method of hypothetical cases is capable of yielding a priori knowledge of necessary truths, an explanation in terms of what goes on in our minds when we use the method. It seems inevitable that any such explanation will construe the necessary truths discovered by the method of hypothetical cases as arising somehow from the meanings of words.2 The explanation may then be able to explain how one acquires knowledge of necessary truths naturalistically, but still preserve the a priori character of this knowledge, by not requiring knowers to have had any more perceptual experience than they needed to come to understand certain words. The appropriateness of linguistic analysis therefore turns on whether such an explanation of how the method of hypothetical cases can yield a priori knowledge of necessary truths—​what I will call a ‘foundation’ for linguistic analysis—​exists. In section 1, I present what I call the ‘classical foundation’ for linguistic analysis. In section 2, for a reason that will emerge in section 4, I present a semantically externalist view of the meaning and reference of general terms, according to which no foundation for linguistic analysis is possible.3 In section 3, I present a two-​dimensional account of the same terrain, focusing exclusively on Frank Jackson’s version of two-​dimensionalism, and sketch how it can, it seems, provide a foundation for linguistic analysis.4 Finally, in section 4, I suggest a reason why the two-​dimensional account might not in fact be able to provide such a foundation.

1  The Classical Foundation for Linguistic Analysis What I am calling the ‘classical foundation’ for linguistic analysis is intended to be a reconstruction of what practitioners of linguistic analysis during its heyday would have offered as a rationale for their practice if pressed to give one.5 It presupposes a cluster of inter-​related theses about meaning, reference, and linguistic understanding that are nowadays very widely rejected, though initially attractive, and it will provide a useful foil for views discussed later. The central thesis that the classical foundation presupposes is that the meaning of a general term (e.g., ‘gold’, ‘bachelor’) is a particular set of properties that is conventionally associated with the term. In one sense of the word ‘concept’, such a set of properties can be said to be a concept, because the set of properties specifies what it is to be F (or maybe what it would be to be F), where ‘F’ is the general term conventionally associated with the set of properties; and what it is (or would be) to be F is the conceptually necessary and sufficient condition for being F.6 It is concepts in this sense that linguistic analysis analyzes; hence the traditional name for linguistic analysis, ‘conceptual analysis’.7 For example, ‘gold’ might be conventionally associated with the properties of being shiny, yellow, and malleable, so that ‘Gold is highly prized’ would mean that the shiny, yellow, and malleable stuff is highly prized; and being shiny, yellow, and malleable is what it is (or would be) to be gold, the conceptually necessary and sufficient condition for being gold. Likewise, the empty but still meaningful term, ‘unicorn’, might be conventionally associated with the properties of being a horse and having a single horn in the center of the forehead.8 A corollary of the central thesis that is crucial to grounding linguistic analysis is that understanding a general term, ‘F’, requires knowing, in some fashion, what it is to be F—​requires knowing, in some fashion, that ‘F’ applies to something iff it has the properties that are conventionally associated with 258

The Foundation of Linguistic Analysis

‘F’. Thus, understanding ‘gold’ requires knowing, in some fashion, that ‘gold’ applies to something iff it is shiny, yellow, and malleable. But understanding ‘F’ does not require the knower to verbalize the knowledge; a perfectly competent speaker-​hearer of ‘F’ could go a lifetime without doing so. Understanding ‘F’ is merely implicit knowledge of the meaning of ‘F’, and manifests itself chiefly in one’s competence as a speaker-​hearer. The central thesis about meaning and its corollary about understanding naturally suggests an account of how the method of hypothetical cases manages to yield a priori knowledge of necessary truths. The necessary truths that can be known a priori are, of course, the conceptually necessary truths that specify the necessary and sufficient condition for a general term to apply to something—​ for something to be F if the term is ‘F’. And, according to the corollary, understanding a general term gives one implicit knowledge of the conceptually necessary and sufficient condition for the term to apply to something. Let us now add the plausible claim that one is drawing upon this implicit knowledge not only when, as a competent speaker-​hearer, one considers actual circumstances and judges whether or not a general term applies to something in those circumstances but also when one considers hypothetical circumstances and judges whether or not a general term would apply to something in those circumstances. Because one is drawing upon knowledge, one’s judgments are correct. And on the basis of sufficiently many such judgments about a general term, one can formulate an explicit statement of the conceptually necessary and sufficient condition for the term to apply to something. Knowledge of this statement counts as a priori because gaining it requires no perceptual experience of the world beyond what is required for coming to understand the term.

2  The Externalist Challenge to Linguistic Analysis If the theses about meaning and linguistic understanding presupposed by the classical foundation for linguistic analysis are true, then linguistic analysis seems to have a solid foundation. On the other hand, linguistic analysis lacks a foundation if a semantically externalist view of the same terrain is true instead. Let me sketch a generic form of such an externalist view, without attributing it to any particular author(s), and then spell out its consequences for linguistic analysis. According to this form of externalism, a general term, ‘F’, does not have a meaning of the sort envisaged by the classical view: there is no set of properties conventionally associated with ‘F’ such that to be F is to have the properties in the set.9 Two important consequences follow. First, the reference of ‘F’ cannot be determined by its meaning; its reference cannot be the thing, if any, that has the properties in the set of properties conventionally associated with ‘F’. Second, understanding a general term cannot consist in implicitly knowing that ‘F’ applies to something iff it has the properties that are conventionally associated with ‘F’. Given these consequences, this form of externalism must provide alternative accounts of how the reference of a general term is determined, and of what understanding a general term consists in. For present purposes, however, detailed accounts are not needed, and so I will merely sketch. According to an externalist account of reference-​determination, a general term refers to the real thing to which it stands in a certain non-​semantic and natural relation, perhaps, but not necessarily, a causal-​historical relation;10 standing in this relation to a real thing is what referring is.11 The account doesn’t claim, however, that the term is conventionally associated with (the property of) standing in this relation to a certain real thing. It also doesn’t claim that understanding the term requires knowing that the term applies to whatever stands in this relation to a certain real thing. For example, on Saul Kripke’s externalist ‘picture’ of reference, the term ‘gold’ on my lips refers to gold because it stands at the end of a lengthy historical chain of reference-​borrowing, passing through countless users of the term, that got started when someone successfully introduced the term ‘gold’ as a term for a certain stuff which was in fact gold (Kripke 1980, 93–​97).12 But Kripke’s picture does not claim that there is a conventional association among English-​speakers between ‘gold’ and (the property of) standing at the end of this chain of reference-​borrowing; and it does not claim that speaker-​hearers who 259

Andrew Melnyk

understand ‘gold’ need to know, even implicitly, that ‘gold’ applies to whatever stands at the end of this chain.13 There is no standard externalist account of understanding a general term. But one externalist proposal is that understanding ‘F’ merely requires the ability to translate back and forth between occurrences of ‘F’ in English sentences and tokenings of a co-​referential concept in thoughts with the same truth-​conditional contents as the sentences (see Millikan 1984, 147–​148; Devitt and Sterelny 1999, 187–​190). Such an ability need not require propositional knowledge of meaning. Because inaccurate, or less accurate, habits of translation would tend to produce less success in interacting with other people and with the physical world, the ability might be acquired by exercising comparatively unsophisticated psychological capacities to abandon less successful behavior in favor of more successful behavior (e.g., the capacity to be conditioned). If the externalist view of meaning and understanding that I have sketched is true, then the method of hypothetical cases seems incapable of yielding a priori knowledge of necessary truths. For on this view there are no conceptually necessary truths, which would require meaning of the sort envisaged by the classical view; so obviously the method of hypothetical cases cannot yield a priori knowledge of them.14 Consistently with this view there may, of course, be metaphysically necessary truths to the effect that something is F iff it is the G (e.g., that something is gold iff it is the element with atomic number 79). But understanding ‘F’, on the externalist view, does not require knowledge, not even implicit knowledge, of metaphysically necessary truths about being F; such knowledge would have to be acquired a posteriori (as in the case of gold). And there appears to be no further kind of necessary truth that might lead to a vindication of the claim that the method of hypothetical cases yields a priori knowledge of necessary truths.

3  A Two-​Dimensional Foundation for Linguistic Analysis? Let me now explain how two-​dimensional semantics—​or at least my reconstruction of Frank Jackson’s well-​known version of it—​might provide a foundation for linguistic analysis, starting with what it has to say about meaning, reference, and linguistic understanding. I  limit myself to Jackson’s well-​known version of two-​dimensionalism, and only insofar as it applies to general terms, because its theoretical goals are reasonably clear and explicitly include providing a foundation for linguistic analysis (Jackson 1998, 2001, and in this volume; Chalmers and Jackson 2001). By ‘two-​dimensionalism’ I  will henceforth mean my reconstruction of Jackson’s two-​ dimensionalism as it applies to general terms. Two-​dimensionalism agrees with the classical account in identifying the meaning of a term, ‘F’, with a particular set of properties that is conventionally associated with ‘F’. But there is a major difference between how the two accounts view the set of properties. On the classical account (section 1), the set of properties conventionally associated with ‘F’ plays two roles: first, it specifies what it is to be F; second, it determines the reference of ‘F’—​the reference of ‘F’ is the thing, if any, that has the properties in the set. On two-​dimensionalism, the conventional association between ‘F’ and certain properties is along these lines: (i) ‘F’ is the term that people conventionally use when they want to talk about the thing that actually has the associated properties and (ii) when people hear other people utter ‘F’, they conventionally take them to be talking about the thing that actually has those properties.15 Because the unique bearer of the properties in the actual world may not bear those properties in other possible worlds, the set of properties conventionally associated with ‘F’ does not play the first role. But it still plays the second, reference-​determining role: the reference of ‘F’ is the actual bearer of the properties in the set.16 Thus, the conventionally associated properties for ‘water’ might include falling from the sky as rain and filling the lakes and rivers, in which case the reference of ‘water’ would be the actual stuff that falls from the sky as rain, fills the lakes and rivers, and so forth; in some possible worlds, however, that actual stuff—​H2O, as it turns out—​does not fall from the sky as rain, fill the lakes and rivers, and so forth. Even though, on two-​dimensionalism, the conventionally associated properties do not specify what it is to be F, so that there is no conceptually necessary and sufficient condition 260

The Foundation of Linguistic Analysis

of being F, there can still be a metaphysically necessary and sufficient condition of being F. However, this metaphysically necessary and sufficient condition is not entailed by the meaning of ‘F’, i.e., by the properties conventionally associated with ‘F’. Understanding ‘F’, on the two-​dimensionalist view, requires knowing the meaning of ‘F’. But because this meaning does not specify a conceptually (or metaphysically) necessary and sufficient condition of being F, understanding ‘F’ does not require knowledge of such a condition; here, of course, two-​dimensionalism breaks with the classical view and agrees with externalism. Understanding ‘F’ is also claimed not to require the ability to state which properties fix the reference of ‘F’. Understanding ‘F’ is claimed to be merely implicit knowledge of which properties fix the reference of ‘F’, taking the form of an ability correctly to judge as true or false (or perhaps as indeterminate) every ‘F’-​involving ‘application conditional’ (Chalmers and Jackson 2001), that is, every conditional of the form, ‘If the actual world turns out to be thus and so, then “F” applies to such-​and-​such’ (e.g., ‘If the actual world turns out to be one in which D2O [deuterium oxide] is the stuff that falls from the sky as rain, fills the lakes and rivers, and so forth, then “water” refers to D2O’). Because, on the two-​dimensionalist view, understanding ‘F’ does not require implicit knowledge of a conceptually (or metaphysically) necessary and sufficient condition of being F, two-​dimensionalism cannot explain how the method of hypothetical cases manages to yield a priori knowledge of conceptually (or metaphysically) necessary truths. But it need not try to do that, for it claims that there is another kind of necessary truth, and it seeks to explain how the method of hypothetical cases manages to yield a priori knowledge of necessary truths of that kind. Necessary truth is traditionally understood as truth in all possible worlds. But two-​dimensionalism draws an important distinction between possible worlds ‘considered as counterfactual’—​ways the world might have been—​and possible worlds ‘considered as actual’—​ways the actual world might turn out to be. Possible worlds ‘considered as counterfactual’ are possible worlds as traditionally construed. Intuitively, ‘What if your spouse had been a space alien?’ is a very different question from ‘What if your spouse turns out to be a space alien?’ Both questions concern the same possibility; but the first question considers this possibility as a way the world might have been while the second question considers it as a way the world might turn out to be. To envisage the possibility that one’s spouse should turn out to be a space alien is to envisage the possibility that one is dramatically mistaken as to how the world has actually been all along. The meaning of ‘F’, on two-​dimensionalism, can be represented formally as a function from each possible world considered as actual to the referent of ‘F’ (if any) in that world; this function Jackson calls the ‘A-​intension’ of ‘F’. The A-​intension therefore takes each possible world considered as actual onto the thing (if any) in that world that has the properties conventionally associated with ‘F’, and it will typically take different possible worlds considered as actual onto different referents.17 Perhaps the A-​intension of ‘water’ takes the possibility that D2O turns out to be the stuff that falls from the sky as rain and so forth onto D2O, but takes the possibility that alcohol turns out to be the stuff that falls from the sky as rain and so forth onto alcohol.The reference of ‘F’ can also be represented formally, as a constant function from each possible world considered as counterfactual to the thing that, in the actual world, has all the properties in the set conventionally associated with ‘F’; this function is what Jackson calls the ‘C-​intension’ of ‘F’.18 Thus, on the assumption that H2O is the actual stuff that falls from the sky as rain, fills the lakes and rivers, and so forth, the reference of ‘water’ can be represented formally as a constant function from each possible world considered as counterfactual to H2O. The distinction between possible worlds considered as counterfactual and possible worlds considered as actual enables a distinction to be drawn between two kinds of necessary truth. The first kind of necessary truth, often called ‘metaphysically’ or ‘broadly logically’ necessary, is truth in all possible worlds considered as counterfactual, and may be called ‘C-​necessary’. The second kind of necessary truth, which has no traditional name, is truth in all possible worlds considered as actual, and may be called ‘A-​necessary’. For example, the sentence, ‘Water falls from the sky as rain’, is not C-​necessary, because had the world been too cold for water ever to evaporate then water would not 261

Andrew Melnyk

have fallen from the sky as rain. But if the A-​intension of ‘water’ can be paraphrased as ‘the actual stuff that falls from the sky as rain and fills the lakes and rivers’, then the sentence is A-​necessary—​true in all worlds considered as actual—​because whatever turns out to fall from the sky as rain and fill the lakes and rivers, whether H2O, D2O, alcohol, or something even better, it falls from the sky as rain. It is now easy to see how two-​dimensionalism might explain how the method of hypothetical cases can yield a priori knowledge of necessary truths. The first move is to claim that the necessary truths in question are A-​necessary truths.19 Admittedly, A-​necessary truths are not conceptually necessary truths. So linguistic analysis of ‘F’ as vindicated by two-​dimensionalism would not produce a priori knowledge of what it is to be F; it would not achieve the longstanding philosophical goal of discovering the nature or essence of things. Jackson claims that it would nonetheless play an indispensable role in solving what he calls ‘location problems’ in philosophy (Jackson 1998). The second move is to appeal to the two-​dimensionalist view that understanding ‘F’ requires implicit knowledge of the A-​intension of ‘F’: the ability to evaluate (judge as true or false or indeterminate) every ‘F’-​ involving application conditional correctly, i.e., in conformity with the actual A-​intension of ‘F’. This view promises to secure the a priori character of our knowledge of A-​necessary truths. For suppose that one has the ability correctly to evaluate every ‘F’-​involving application conditional. Then whenever, using the method of hypothetical cases, one considers whether ‘F’ applies to something in a hypothetical situation, one can simply treat the hypothetical situation as a way the actual world might turn out to be, and then exercise the ability. And having answered the question about the applicability of ‘F’ with regard to sufficiently many hypothetical situations considered as actual, and having reflected on the situations’ commonalities, one can rightly conclude (without the help of further perceptual experience) that something is F iff it is the actual bearer of certain properties. If this conclusion is true, it seems to constitute knowledge, and indeed a priori knowledge, because reaching it requires no perceptual experience beyond what was required to acquire understanding of ‘F’ in the first place.

4  A Problem for the Two-​Dimensional Foundation The two-​dimensional account of how the method of hypothetical cases can yield a priori knowledge faces a problem. It takes no official stand on the nature of knowledge itself. But it would presumably be served best by some version of a reliabilist theory of knowledge, so that the products of linguistic analysis count as knowledge because they are produced by what is in fact a reliable belief-​forming process, even if the subject has no evidence that it is reliable. But a reliabilist theory of knowledge must, in my view, include a ‘no defeater’ condition in order to avoid certain counterexamples (see Lycan 1988, 109–​110). A true belief that arises from the operation of a reliable belief-​forming process is not knowledge if the subject has a defeater, either a rebutting defeater (evidence that the belief is false) or an undermining defeater (evidence that the belief-​forming process is not reliable); subjects know only if they have no defeater. I will argue that, even if the two-​dimensionalist account of linguistic analysis is true, practitioners of linguistic analysis who reflect on the account have an undermining defeater for the belief-​forming process that is claimed to operate in linguistic analysis. The defeater is the fact that, given the available evidence, the two-​dimensionalist account of linguistic understanding is not clearly more probable than an externalist account on which, as we saw in section 2, linguistic analysis does not constitute a reliable belief-​forming process. One can, of course, practice linguistic analysis without reflecting on the two-​dimensionalist (or any) account of it; but philosophers who practice linguistic analysis are likely to reflect on it. An undermining defeater for a belief-​forming process need not show that the belief-​forming process is unlikely to produce true beliefs. For example, suppose my teenage son tells his mother that he has done all his homework. Testimony, of course, is in general a reliable belief-​forming process. But my son has often told his mother that he has done all his homework when in fact he hasn’t, and 262

The Foundation of Linguistic Analysis

she knows this. Also, the present episode seems no different from earlier episodes. She therefore has an undermining defeater, so that, even if in an excess of charity she believes him and he has told the truth, she does not know that he has done all his homework. But the prior occasions on which his testimony on homework has been inaccurate need not have formed a majority of all prior occasions on which he provided testimony on homework; all his mother knows is that his testimony on homework has often been inaccurate. I will not attempt further exploration of what it takes to be an undermining defeater. I  will argue that philosophers who reflect on the two-​dimensionalist account of linguistic analysis have an undermining defeater at least as strong, intuitively, as the one my wife has in knowing that my son’s testimony has often been inaccurate. These philosophers should therefore judge that the products of their linguistic analysis do not constitute knowledge; at the very least they should withdraw any claim that they do constitute knowledge. In principle, the defeater that these philosophers have could itself be defeated, but only by the discovery of further empirical evidence. So, even if linguistic analysis can in the end produce knowledge, it would still not be a priori knowledge. Suppose, then, that one practices linguistic analysis, and that the two-​dimensionalist account of linguistic analysis is true. On reflection, one will notice that, according to the two-​dimensionalist account, engaging in linguistic analysis constitutes a reliable belief-​forming process only if understanding a general term, ‘F’, requires that one can evaluate all ‘F’-​relevant application conditionals correctly, i.e., in conformity with the actual A-​intension of ‘F’. Next, one will naturally wonder whether understanding ‘F’ really does require this particular ability. If it does, that must be because (i) understanding ‘F’ is implicitly knowing that ‘F’ means so-​and-​so, and (ii) what ‘F’ means is a certain particular A-​intension, so that (iii) understanding ‘F’ is implicitly knowing that ‘F’ means this particular A-​intension, i.e., is being able to evaluate all ‘F’-​relevant application conditionals in conformity with this particular A-​ intension.20 According to two-​dimensionalism, however, for the meaning of ‘F’ to be a certain particular A-​intension (and not another one) just is for ‘F’ to be conventionally associated, in the right reference-​fixing sort of way, with a certain set of properties (and not another set).21 But it is a posteriori whether ‘F’ is conventionally associated, in the right reference-​fixing sort of way, with one set of properties rather than another; and so it is also a posteriori whether understanding ‘F’ really does require that one can evaluate all ‘F’-​relevant application conditionals in conformity with the actual A-​intension of ‘F’ (and not with some other A-​intension). All that being so, what would actual empirical evidence for the view that understanding ‘F’ requires this particular ability look like? However the view is arrived at, it trivially implies the universal psychological generalization that everyone who understands ‘F’ has the ability to evaluate all ‘F’-​relevant application conditionals in conformity with the actual A-​intension of ‘F’.The view therefore predicts that numerous universal psychological generalizations hold among people who understand general terms in English. For instance, the view predicts that everyone who understands ‘water’ has the ability to evaluate all ‘water’-​relevant application conditionals in conformity with the actual A-​intension of ‘water’, and hence (since there is only one A-​intension of ‘water’) that everyone who understands ‘water’ would give the same answers as everyone else to every question of the form, ‘If the actual world turns out to be thus and so, does “water” apply to such-​and-​such?’ Does the available empirical evidence suggest that such universal psychological generalizations actually hold? Presumably no one has ever discovered evidence for them as a social scientist would, by identifying a representative sample of people who understand (say) ‘water’, quizzing them with a representative sample of the right sort of questions, and then observing whether their answers exhibit uniformity. A different approach would exploit the fact that two-​dimensionalists grant that the A-​intension of a general term can often be paraphrased by a rigidified definite description, at least roughly. Suppose ‘water’ is such a term. Then the hypothesis that everyone who understands ‘water’ can evaluate all ‘water’-​relevant application conditionals in conformity with the A-​intension of ‘water’ predicts that everyone who understands ‘water’ will think, more or less explicitly, that water is the actual G, where ‘G’ roughly paraphrases the A-​intension of ‘water’. And in principle there could be evidence that this prediction is correct. Thinking that water is the actual G does not, of course, 263

Andrew Melnyk

entail being disposed to use ‘the actual G’, rather than any other rigidified definite description true of water, to evaluate ‘water’-​relevant application conditionals;22 but evidence that any prediction of a hypothesis is correct is some evidence for the hypothesis. Presumably, however, no one has ever identified a representative sample of people who understand ‘water’, asked them whether they think water is the actual G, and found that they all do. We must therefore fall back on casual empiricism. But when we do, I will now argue, such evidence as we find does not clearly favor, and may not favor at all, the conclusion that those who understand a general term, ‘F’, all think that F is the actual G, for some predicate, ‘G’. First, one must have done more than hear someone use a term a few times to determine with reasonable confidence that the speaker thinks that a certain definite description is true of the term’s referent. How confident do you feel about which rigidified definite descriptions your friends and family think are true of the referents of ‘butter’ or ‘soil’ or ‘ingenuity’, terms that you must have heard them use scores of times? Either one must have examined a fair bit of speakers’ term-​using speech or one must actually question speakers; and one is not likely to have met either condition with regard to more than a few speakers. One’s casual observation, then, has probably only revealed small numbers of people who understand ‘F’ and all think that the term’s referent is the actual G, for some ‘G’. Such small samples constitute weak evidence. Second, however many confirming instances one’s casual observation reveals, they are unlikely to form a representative sample. One is most unlikely to have selected speakers so that they form a representative sample: they are just people one has run into through one’s job or reading habits or hobbies. And probably one has positive reason to fear that the sample is unrepresentative (e.g., if the speakers are highly educated, reflective, and articulate people from the distinctive sub-​culture of academic analytic philosophy). One should hesitate to extrapolate far beyond it. Third, positive instances of a generalization confirm the generalization only if there are no counterexamples. On the face of it, however, there are many counterexamples to the generalizations in question:  people who understand a general term but who appear not to think that any definite description is true of the term’s referent.23 The common philosophical example of ‘water’ may give a misleading impression, because all humans directly experience water every day in similar ways. But consider the better-​known terms for chemical elements. Many of us count ourselves as understanding, say, ‘caesium’, even though all we know about caesium is that it is a highly reactive metallic element, and we know that caesium is not the only highly reactive metallic element. Many of us count ourselves as understanding ‘argon’ and ‘krypton’, even though all we know about argon and krypton is that they are noble gases, though not the only ones.24 Such speakers, and others like them, are prima facie counterexamples to the hypothesis that, for every general term ‘F’, those who understand ‘F’ all think that F is the actual G, for some predicate, ‘G’. Such speakers, however, are no threat to an externalist account of understanding a general term, the chief rival to the two-​dimensionalist account. Suppose the externalist account mooted above is true: that understanding a general term merely requires the ability to translate back and forth between occurrences of the term in English sentences and tokenings of a co-​referential concept in thoughts with the same truth-​conditional contents as the sentences. Even if what explains how speakers can translate back and forth in this way is sometimes their thinking that a rigidified definite description is true of the term’s referent, it needn’t be the same rigidified definite description for everyone who understands the term; it needn’t even be the same rigidified definite description for a single speaker at different times in their life. Such an externalist account is therefore not committed to such psychological generalizations as that everyone who understands ‘water’ thinks, more or less explicitly, that water is the actual G, where ‘G’ roughly paraphrases the A-​intension of ‘water’. At the same time, an externalist account need not deny that sometimes there is a single definite description that all people who understand a term think is true of the term’s referent. The account is therefore consistent with the evidence for the two-​dimensionalist account that we have reviewed. 264

The Foundation of Linguistic Analysis

Let us now take stock. Those who reflect on the two-​dimensionalist account of linguistic analysis have no systematic evidence for the entailed account of understanding a general term. They may have some unsystematic but weak evidence.There are, however, prima facie counterexamples to the account, cases that a rival externalist account of understanding a general term naturally accommodates.What to conclude? One might reasonably conclude that the externalist account wins by a whisker. Perhaps one might reasonably conclude that the two-​dimensionalist account wins by a whisker.What one may not reasonably conclude is that on balance the evidence clearly favors the two-​dimensionalist account; there is a good chance that the account is false, and that its externalist rival is true.Therefore, even if the two-​ dimensionalist account of linguistic analysis is in fact true, practitioners of linguistic analysis who reflect on the account along the preceding lines have an undermining defeater for the belief-​forming process that, according to the account, operates in linguistic analysis. Further empirical evidence might defeat this defeater, and then the products of linguistic analysis would constitute knowledge (if the two-​ dimensionalist account of linguistic analysis is true); but they would not constitute a priori knowledge. Even if the two-​dimensionalist account of understanding a general term is not true, we still have dispositions to return confident answers to questions of the form, ‘If the actual world turns out to be thus and so, does “F” apply to such-​and-​such?’ The extent to which we have such dispositions—​for how many general terms and for how many possibilities considered as actual—​is not obvious from casual observation. But most of my students, for example, can be persuaded that, if what falls from the sky as rain and fills the lakes and rivers turns out to be D2O, rather than H2O, then ‘water’ refers to D2O. Something must explain their dispositions to judge in this way, but if it is not their ability to evaluate all ‘water’-​relevant application conditionals in conformity with the A-​intension of ‘water’, then what is it? The question would be a good one for psychological research. I speculate that their dispositions arise, not from their first-​order semantic competence, but from their tacit endorsement of a proto-​theory of reference-​determination—​which may be no more than the vague idea that words refer to those things in the world that we have dealings with when we use the words. Their (tacit) reasoning would then be: if what falls from the sky as rain and fills the lakes and rivers turns out to be D2O, then our dealings with what we have been calling ‘water’ have in fact all along been dealings with D2O, and so ‘water’ refers to D2O. But this vague idea cannot require two-​dimensionalism, because one way of developing it would eventuate in Millikan’s radically externalist account of reference-​determination according to which, very roughly, tokens of ‘water’ refer to the stuff that these tokens must correspond with for them to be able to perform their proper function of helping to produce true beliefs (Millikan 1984, ch. 6).

Further Reading Byrne, Alex, and James Pryor. ‘Bad Intensions’. In Two-​Dimensional Semantics, M. García-​Carpintero and J. Macià (eds.), New York: Oxford University Press, 2006. Chalmers, David, and Frank Jackson. ‘Conceptual Analysis and Reductive Explanation’, The Philosophical Review, 110, 2001: 315–​360. Jackson, Frank. From Metaphysics to Ethics:  A Defence of Conceptual Analysis, New  York:  Oxford University Press, 1998. Laurence, Stephen, and Eric Margolis. ‘Concepts and Conceptual Analysis’, Philosophy and Phenomenological Research, 67, 2003: 253–​282. Millikan, Ruth Garrett. Language,Thought, and Other Biological Categories: New Foundations for Realism, Cambridge, MA: MIT Press, 1984. Millikan, Ruth Garrett. Language: A Biological Model, New York: Oxford University Press, 2005.

Notes 1 Does acquisition of knowledge count as a priori if it relies on a belief that is innate? On the one hand, though acquiring the belief does not require the knower to have had perceptual experience, it may require that the knower’s ancestors had perceptual experience. On the other hand, a priori knowledge may not have a definition sharp enough to permit a definitive answer to the question.

265

Andrew Melnyk 2 The meanings of words, despite being partly constituted by mental phenomena, are mind-​independent, just as minds themselves are mind-​independent: their existence is at most causally dependent on thought about them. If the necessary truths allegedly discovered by the method of hypothetical cases arise from the meanings of words, then they are mind-​independent too. 3 I treat only general terms; no one thinks we should do linguistic analysis on names. Only a portion of Jackson’s exposition of two-​dimensionalism in the present volume concerns general terms. 4 I focus exclusively on Jacksonian two-​dimensionalism because the view is clearly intended as an account of public language—​which cannot be said of, for example, David Chalmers’ two-​dimensionalism (Chalmers 2006; Schroeter 2017). 5 A sophisticated version of it has been defended more recently, for example, by Christopher Peacocke (1993). 6 The necessity is just truth in all possible worlds; I call it ‘conceptual’ to suggest that it is owed (somehow) to the nature of concepts. 7 In what I take to be a very different sense of ‘concept’, a concept is a kind of mental representation—​perhaps a word in the language of thought, as in Jerry Fodor’s theory (Fodor 1998). 8 A corollary of this central thesis (though it plays no part in grounding linguistic analysis) is that the reference of a general term, ‘F’, is the thing, if any, that has all the properties in the set of properties conventionally associated with ‘F’—​the thing, if any, that meets the conceptually necessary and sufficient condition for being F. 9 Or such that to be F merely requires having the properties in the set. A weaker form of externalism could allow that there is a set of properties conventionally associated with ‘F’ such that, if something is F, then it has the properties in the set, so that, for example, ‘tiger’ applies to something only if it is an animal. But the considerations invoked to support externalism as against a classical view tend to militate also against the weaker externalist view. For example, I once heard Berent Enç say that, early in their acquaintance, Fred Dretske nearly succeeded in persuading him that a buckeye was an animal. Enç was semantically competent with ‘buckeye’, for he had heard of buckeyes, knew, for instance, that they are common in Ohio (which is The Buckeye State!), and was weakly inclined to think that they are plants. He would not have lost that competence had Dretske succeeded. So we do not take semantic competence with ‘buckeye’ to require believing that buckeyes are plants. (Buckeyes are a kind of tree, though the term ‘buckeye’ is also used to refer to the tree’s seeds, which resemble the eyes of deer.) 10 Ruth Millikan’s externalist view doesn’t take this relation to be causal-​historical (Millikan 1984, 2005, 2017). 11 So empty terms do not refer. They may, however, be viewed as the products of processes that are supposed to produce non-​empty terms, but that sometimes fail to do so; and they may play many of the same functional roles as non-​empty terms (Millikan 1984, 136). 12 The theory originated by Kripke has been developed, for example, by Philip Kitcher (1993) and by Michael Devitt and Kim Sterelny (1999). 13 Proponents of causal descriptivism, of course, make both these claims. 14 The scope of the generic form of externalism that I have been presenting includes all general terms, even ‘bachelor’. A version restricted to natural-​kind terms would entail the weaker claim that there are no conceptually necessary truths about natural kinds. Stephen Biggs and Ranpal Dosanjh argue that externalism is true of many general terms, not just natural-​kind terms (this volume). 15 Compare Jackson, in this volume, section 5: “the obvious [answer] is that ‘water’ is a word for telling about the watery stuff—​that’s its informational role”. 16 In Kripke’s terminology, the properties in the set are said merely to fix the reference (rather than, more ambitiously, to give the meaning) of ‘F’ (Kripke 1980). 17 The exceptions are when the properties are necessary properties; for example, 2=√4 in all possible worlds, so ‘the square root of 4’ refers to 2 in all worlds considered as actual. 18 The reference is a constant function because ‘F’ is a rigid designator, referring, in every possible world in which it refers to anything, to whatever it refers to in the actual world. ‘F’ is rigid, because the definite description that fixes its reference is a rigidified definite description, ‘the actual G’. 19 And that they are in principle expressible in a public language—​because, on Jackson’s two-​dimensionalism, the A-​intensions that ground A-​necessary truths are the A-​intensions of public terms. Versions of two-​ dimensionalism on which A-​intensions (or primary intensions) belong to, say, terms in individual idiolects rather than to public terms will not have this consequence. 20 The meaning in question is what I have been calling ‘meaning’, which is distinct from reference. 21 The fact that a term is conventionally associated, in the right reference-​fixing sort of way, with a certain set of properties makes it the case that the term has the A-​intension that it has; but it does so only because the term’s having the A-​intension that it has just is the term’s being conventionally associated, in the right reference-​ fixing sort of way, with a certain set of properties. One could conceivably claim instead that the semantic fact merely emerges from the social-​cum-​psychological fact, so that the two facts are distinct; but that claim is clearly not Jackson’s.

266

The Foundation of Linguistic Analysis 22 To believe that water is the actual G is not yet to treat ‘the actual G’ as fixing the reference of ‘water’. 23 Putnam famously made this point with his example of ‘elm’ and ‘beech’ (Putnam 1975). Kripke made the analogous point for proper names with his example of ‘Richard Feynman’ and ‘Murray Gell-​Mann’; both Feynman and Gell-​Mann are known to some people who understand these names only as ‘a famous American physicist’ (Kripke 1980). The point is elaborated by Ruth Millikan (1984, ch. 9). 24 I know that argon is something called ‘argon’, but not that it is the only thing called ‘argon’. (In fact, I know a person called ‘Argon’.)

References Chalmers, David. 2006. ‘Two-​Dimensional Semantics’. In The Oxford Handbook of Philosophy of Language, E. Lepore and B. Smith (eds.). Oxford: Oxford University Press, pp. 575–​606. Chalmers, David, and Frank Jackson. 2001. ‘Conceptual Analysis and Reductive Explanation’, The Philosophical Review, 110, 315–​360. Devitt, Michael, and Kim Sterelny. 1999. Language and Reality:  An Introduction to the Philosophy of Language. Cambridge, MA: MIT Press. Fodor, Jerry. 1998. Concepts: Where Cognitive Science Went Wrong. New York: Oxford University Press. Jackson, Frank. 1998. From Metaphysics to Ethics:  A Defence of Conceptual Analysis. New  York:  Oxford University Press. Jackson, Frank. 2001. ‘Précis of From Metaphysics to Ethics’, Philosophy and Phenomenological Research, 62, 617–​624. Kitcher, Philip. 1993. The Advancement of Science. New York: Oxford University Press. Kripke, Saul. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press. Lycan, William. 1988. Judgement and Justification. Cambridge: Cambridge University Press. Millikan, Ruth. 1984. Language, Thought, and Other Biological Categories: New Foundations for Realism. Cambridge, MA: MIT Press. Millikan, Ruth. 2005. Language: A Biological Model. New York: Oxford University Press. Millikan, Ruth. 2017. Beyond Concepts:  Unicepts, Language, and Natural Information. New  York:  Oxford University Press. Peacocke, Christopher. 1993 ‘How Are a Priori Truths Possible?’, European Journal of Philosophy, 1, 175–​199. Putnam, Hilary. 1975. ‘The Meaning of “Meaning”’. In Mind, Language, and Reality:  Philosophical Papers, Vol. 2. Cambridge: Cambridge University Press. Schroeter, Laura. 2017. ‘Two-​Dimensional Semantics’, The Stanford Encyclopedia of Philosophy (Summer 2017 Edition), Edward N. Zalta (ed.), https://​plato.stanford.edu/​archives/​sum2017/​entries/​two-​dimensional-​ semantics/​.

267

20 A PUZZLE ABOUT ASSERTION Gregory Bochner

This paper presents a puzzle case for Stalnaker’s two-​dimensionalist theory of assertion. Section I  introduces the two-​dimensionalist framework. In section II, internalist and externalist versions of the framework are contrasted. Section III outlines Stalnaker’s pragmatic-​externalist theory of assertion. In section IV, a problem is posed: the theory predicts the wrong truth value for some referential assertions. Section V offers a diagnosis:  the components of the pragmatic account which create the problem are ‘holism about context’ and the denial of a ‘reality principle’ about reference determination. The paper does not purport to offer a solution to the problem, but the final section VI argues that, in order to solve it, the pragmatic account should have to be made compatible with a more thorough form of externalism (‘actual externalism’).

I 

Introduction: Two-​Dimensionalism

The starting observation common to all versions of two-​dimensionalism is that there are two ways in which the truth value of a statement may depend on the facts. First, the facts play an interpretive role when they determine what is said. Second, they play an evaluative role when they determine whether what was said is true or false. Roughly, the facts playing the interpretive role are facts about possible situations in which the statement is made (‘contexts’), whereas the facts playing the evaluative role are facts in possible situations about which the statement is made (‘circumstances’). The two roles of facts are evident in the case of indexical statements. In Kaplan’s (1989a) semantics, the ‘character’ conventionally associated with an indexical sentence-​type determines, relative to a context, what is said (a content). When Vandana Shiva utters the sentence “I am an environmental activist”, what she says is that Vandana Shiva is an environmental activist. When Jair Bolsonaro utters it, what he says is that Jair Bolsonaro is an environmental activist. Same sentence, different contexts of use, different contents. The ‘content’ (or intension) contextually associated with a particular use of the sentence determines, relative to a circumstance, a truth value (or extension). What Bolsonaro said in using the sentence “I am an environmental activist” is false, given the facts in the actual world. But the facts could have been different. There are other possible worlds in which Bolsonaro is an environmental activist, and when what he said is evaluated relative to worlds like these, it is true. Same content, different circumstances of evaluation, different truth values. The Kaplanian form of two-​ dimensionalism, which was motivated by the needs to provide a compositional account of indexical

268

A Puzzle about Assertion

sentences and to capture an intuitive notion of what is said (‘content’) distinct from their conventional meanings (‘character’), is nowadays widely accepted. Other and more controversial forms of two-​dimensionalism go further. They generalize the double dependency: the truth value of any statement, whether overtly indexical or not, depends on a context and a circumstance. Moreover, any statement is associated with two notions of contents. The motivations for those additional moves are cognitive more than semantical. They include an ambition to reconcile referentialist and descriptivist insights about reference. Frege (1892) had highlighted that referentialism, the thesis that the contents of statements may directly involve concrete referents, conflicts with observations concerning the cognitive values of some statements. For instance, although the names ‘Hesperus’ and ‘Phosphorus’ have the same referent (Venus), the statements “Hesperus is Hesperus” and “Hesperus is Phosphorus” have different cognitive values: the former is trivial (its truth value is known a priori), while the latter can be informative (its truth value is known a posteriori). So these statements seem to convey different information. Referentialism, however, predicts that they express the same (trivial) content: that Venus is Venus. From such cognitive observations, Frege concluded that referentialism is false and embraced descriptivism, the view that the contents of statements involve only descriptive conditions, which determine a referent (given a circumstance). According to descriptivism, “Hesperus is Hesperus” says something trivial, viz. that the brightest celestial body in the evening sky is the brightest celestial body in the evening sky, but “Hesperus is Phosphorus” says something informative, viz. that the brightest celestial body in the evening sky is the brightest celestial body in the morning sky.Thus descriptivism provides an explanation for the difference in cognitive value. However, Kripke (1980) refuted descriptivism by showing that it yields the wrong predictions about the truth values of some statements. For instance, descriptivism predicts that the statement “Hesperus is the brightest celestial body in the evening sky” is true relative to all possible circumstances. Yet although the statement is true at the actual world (w0), it is not necessarily true, i.e. true at all possible worlds of evaluation. Thus there is a possible world w1 in which Venus is the brightest celestial body in the morning sky but where it is Mars which is the brightest celestial body in the evening sky. Relative to w1, the statement (as made in w0) is evaluated as false. From such semantical observations, Kripke concluded that descriptivism is incorrect:  the descriptions commonly associated with proper names do not determine a (potentially different) referent at each possible circumstance. Instead, proper names are ‘rigid designators’, i.e. terms that designate the same referent (viz. their actual referent) at all possible worlds of evaluation. Since the Kripkean revolution, one of the central challenges in the philosophy of language and mind has been to reconcile Fregean and Kripkean insights about reference within a unified account.1 Controversial two-​dimensionalism responds to that challenge by claiming that two distinct contents (intensions) are associated with statements involving referential terms: one referential, the other descriptive.2 The referential content is a function from possible circumstances into truth values, or, equivalently, the set of possible circumstances at which the statement is true. Provided that proper names are rigid designators, ‘Hesperus’ and ‘Phosphorus’ co-​refer (to Venus) at all possible worlds. It follows that the statements “Hesperus is Hesperus” and “Hesperus is Phosphorus” express the same referential content, true at all possible worlds of evaluation. Consider three possible worlds: the actual world w0 (where Venus is the brightest celestial body both in the evening sky and in the morning sky), the world w1 introduced above (where Venus is the brightest celestial body in the morning sky but Mars is the brightest celestial body in the evening sky), and w2 where Mars is the brightest celestial body both in the evening sky and in the morning sky. Confining our attention to those worlds for simplicity of exposition, we can represent the referential content of the statement “Hesperus is Phosphorus” (as made in the actual world w0) by means of the following matrix: w0 T

w1 T

w2 T

269

Gregory Bochner

In this one-​dimensional matrix, the worlds w0, w1, and w2 on the horizontal axis play the evaluative role. The referential content of the statement is the horizontal proposition obtained by enumerating the truth values relative to each world on the horizontal axis. It amounts to a necessary proposition, true at all the possible worlds of evaluation, hence to the following set: {w0, w1, w2}. Now, this referential content does not do justice to the fact that “Hesperus is Phosphorus” can be informative. In modal accounts of content, the informativeness of a statement is naturally explained in terms of the possibilities that its truth excludes, i.e. by the possibilities at which it is evaluated as false. When a content is necessary, it is true at all possibilities, so there is no possibility to be excluded. Thus this proposition does not capture the new information that the statement will convey to someone who did not know that Hesperus, the shiniest celestial body in the evening sky, is identical to Phosphorus, the shiniest celestial body in the morning sky. At that point, controversial two-​dimensionalism says that there is a second, descriptive content associated with the statement. That content is obtained by considering not just what the statement says in the actual world w0, but what it says when it is made in other possible worlds. Assuming that ‘Hesperus’ is associated with the descriptive condition the brightest celestial body in the evening sky and ‘Phosphorus’ with the descriptive condition the brightest celestial body in the morning sky, the statement “Hesperus is Phosphorus” says different things in w0, w1, and w2. In w0, it says that Venus is Venus. In w1, it says that Venus is Mars. In w2, it says (as in w1) that Mars is Mars. This further dependency on the facts playing the interpretive role is reflected in a two-​dimensional matrix, representing a function from possible contexts to contents:

w0 w1 w2

w0 T F T

w1 T F T

w2 T F T

As this matrix shows, the statement expresses different horizontal propositions in the different worlds of interpretation on the vertical axis. Each of these propositions is necessary and therefore uninformative. Now someone who is not aware that the brightest celestial body in the evening sky is the brightest celestial body in the morning sky will not know which world among w0, w1, and w2 is actual. Such a person, being ignorant of certain contextual facts relevant to the interpretation of the statement, will not know which horizontal proposition the statement expressed. Still, she will know that if the statement was made in w0 or in w2, then it is true, whereas if it was made in w1, then it is false. That is, she will know that if the statement is both interpreted and evaluated relative to w0 or w2, it is true, whereas if it is both interpreted and evaluated relative to w1, it is false. This knowledge is represented by the diagonal proposition in the matrix, obtained by enumerating the truth values from upper left to lower right. According to the controversial two-​dimensionalist, this diagonal proposition is what captures the cognitive value of the statement: it amounts to the descriptive and contingent proposition that the brightest celestial body in the evening sky is the brightest celestial body in the morning sky, which is true when evaluated at w0 and w2, but false when evaluated at w1. The informativeness of the statement is then explained by invoking the fact that the truth of the diagonal proposition rules out at least one world—​here, w1—​from the set of possible worlds compatible with what is known or believed.

II  Internalist and Externalist Versions of Two-​Dimensionalism According to Stalnaker (2001/​2003b, 2003a, 2006), there is an important difference between two interpretations of the two-​dimensionalist framework. The distinction Stalnaker has in mind here parallels the contrast he has long been emphasizing between the Kaplanian version of two-​dimensionalism (a semantic theory for indexical sentences) and his own version (a pragmatic theory of indexical and nonindexical utterances).3 According to the semantic interpretation of two-​dimensionalism (which 270

A Puzzle about Assertion

Stalnaker also calls “the generalized Kaplan interpretation”), the diagonal proposition associated with a statement is the actual meaning of the corresponding sentence:  it is what is primarily grasped by competent speakers who understand the statement and what semantically determines different horizontal propositions in different possible worlds of interpretation. The statement expresses the horizontal contents it expresses in the various possible worlds of interpretation because it has, in the actual world, the diagonal content it has.4 According to the metasemantic interpretation, the order of explanation is reversed: a statement has the diagonal content it has because it expresses the horizontal contents it expresses in the various possible worlds of interpretation.5 Crucially, the semantic facts in those possible worlds of interpretation can be different from the semantic facts in the actual world, and the horizontal content expressed by the statement in a possible world of interpretation depends not on its actual meaning, but on the meaning that the statement has there.6 The two interpretations may agree on the two-​dimensional matrices determined by problematic statements but, at a deeper level, they are underlain by different approaches to fundamental issues. First, while the semantic interpretation of two-​dimensionalism is associated with the revival of an internalist (and descriptivist) approach to Intentionality (meaning is in the head and it determines reference via the satisfaction of descriptive conditions), Stalnaker has insisted in many places that the metasemantic interpretation is of a piece with his radically externalist project (meaning is not in the head and reference is fundamentally determined by causal relations).7 The metasemantic tack presupposes the Millian and causal picture defended by Kripke, according to which the meaning of a proper name is exhausted by its (unique) bearer and reference is fixed at the presemantic or metasemantic level. On that picture, names are not indexical terms whose meaning would semantically fix a referent given a context of use. Then, if ‘Hesperus’ has a referent other than Venus in some possible world of interpretation w, this cannot be because its actual (indexical) meaning (its character) would be a descriptive condition fixing different referents (horizontal contents) in different contexts: it must be because ‘Hesperus’ has a different (Millian) meaning in w. Second, while the semantic interpretation of two-​dimensionalism is concerned to give an account of the meanings of particular sentences and statements, the metasemantic interpretation belongs to a pragmatic theory of the dynamics of conversations, which has further independent motivations. In my opinion, the metasemantic interpretation, with its externalist and pragmatic underpinnings, is more appealing than the semantic interpretation. But I will not argue for this here. What I want to do, rather, is to pose a simple but deep problem arising for Stalnaker’s metasemantic-​pragmatic theory: this predicts the wrong truth values for some referential statements.The problem shows, I will argue, that the pragmatic theory should have to be revised so as to become compatible with a more thorough form of externalism. In what follows, I will outline Stalnaker’s pragmatic model, pose the problem it faces, and propose a diagnosis. Before doing this, I wish to make two bald remarks which, for reasons of space, I will not seek to defend here. First, two distinct theses are intertwined in what Stalnaker calls his externalist approach.8 One is externalism about content: horizontal-​referential contents come first, and they are not determined by what is in the head. The other is anti-​realism about Intentionality: there is no fact of the matter as to what the content of a given linguistic or mental representation is independently of facts about contexts of attribution in which some attributor would describe it. The two theses are separable. I myself do not accept Intentional anti-​realism, which is an extreme and very controversial thesis. Second, even though my point in the present paper will be that the pragmatic-​metasemantic account should make room for an even more externalist notion of content, I also believe that, unless Intentional anti-​realism was assumed from the start, the motivations for controversial two-​dimensionalism, and the presuppositions in terms of which some of the problems it addresses are posed, involve a tacit commitment to some (other) internalist notion of content. 271

Gregory Bochner

III  Stalnaker’s Pragmatic Theory of Assertion Stalnaker’s pragmatic model of discourse, which is based on his theory of assertion, involves two key notions: context and content.9 His notion of content is a purely modal notion. The content of any assertive utterance is a proposition, where a proposition is identified with the set of possible worlds at which it is true. His notion of context is different from Kaplan’s. In Kaplan’s semantics, a ‘context’ is a centered world: it is a situation consisting of a world w, an agent a, a time t, and a location l. (A proper context identifies a metaphysically possible world w in which a exists and is located in l at t.) If Eva utters the sentence “I am happy now” in the world w0 at time t0 and in location l0, the Kaplanian context of her utterance will be the centered world (w0, t0, l0, Eva). By contrast, in Stalnaker’s pragmatics, a ‘context (set)’ is a set of possible worlds: it corresponds to the body of information, concerning the actual world as a whole, that is presumed to be shared by all the conversational participants at the time of utterance. This information represents what is common ground among them at that point of their exchange, where the notion of common ground is defined in terms of the pragmatic presuppositions that they share, and where in turn the notion of pragmatic presupposition is analyzed as an individual propositional attitude, such that an individual a (at some time t) presupposes a proposition p iff (at t) a accepts (i.e. takes for granted for the purposes of the exchange) that p and a accepts that other participants accept that p. So the Stalnakerian context (set) for Eva’s utterance is the intersection of all the propositions that are presupposed by Eva and her interlocutors: e.g. the propositions that she is speaking, that it is sunny, but also e.g. the propositions that the earth is round, that the Berlin Wall fell in 1989, etc. Both Kaplan and Stalnaker claim that context determines content, but they mean different things. Kaplan means that, when a sentence contains indexical expressions (such as personal pronouns or tenses), its content is semantically bound to depend on certain features of the objective situation of speech. Stalnaker means that, for any utterance, its content is pragmatically bound to depend on an intersubjective description of the objective situation of speech. Unlike the Kaplanian context, the Stalnakerian context is not a particular situation (the objective situation of speech) but a set of possible situations (the ones that are compatible with the presuppositions that the conversational participants share about the objective situation of speech). It is a proposition presupposed about the objective situation of speech (the actual world). That proposition, whether or not it is in fact true, is what delineates the space of possible worlds (the context set), figuring in the two-​dimensional matrix, where particular utterances are both interpreted and evaluated. Once contexts are thus identified with contents (propositions), it becomes possible to characterize the dynamics of conversations as a double interaction between contexts (contents presupposed) and contents asserted. In normal communication, the conversational participants agree on what was asserted and what was asserted is informative. When an assertion is understood in the same way by all conversational participants, it is interpreted as expressing one and the same horizontal proposition in every possible world of the context set. The proposition thus expressed in every possible world of interpretation is the content which is asserted by the speaker. When the assertion is informative, the proposition asserted by the speaker is contingent: it is evaluated as false at some but not all possible worlds of the context set. The effect of an informative assertion, if it is accepted by the other participants, is to exclude the possible worlds at which the proposition asserted is false from the context set. To put it another way, the informativeness of the assertion is captured by the possible worlds that are eliminated when the context set is intersected with the content asserted. Now, we often presuppose things that are false, whether by mistake or by pretense. When we do, the common ground is false: the objective situation of speech (the actual world) does not belong to the set of possible situations compatible with what we are presupposing (the context set) about it. In such cases, as in all cases, our utterances are interpreted in the possible worlds of the context set—​ even, that is, if the actual world is not one of its members. It is then the proposition which is expressed in all those (nonactual) worlds of interpretation which, if accepted by other participants, excludes 272

A Puzzle about Assertion

some of those worlds from the context set at the next stage of conversation. Importantly, this implies that the facts in the actual world play only an accidental role in the interpretation of speech. When they affect what is said, this is only because the presuppositions about the actual facts relevant to the interpretation happen to be true.This critical feature of the pragmatic version of two-​dimensionalism is not shared by Kaplan’s semantics and the generalized Kaplan paradigm: The semantics that is relevant to determining the proposition that is the value of the function for a given possible world is the semantics that applies to the sentence in that possible world. The actual character of the sentence is irrelevant, except relative to possible worlds (such as the actual world) in which the sentence has that character. Stalnaker 2007, 259–​260 When the semantic facts that obtain in the actual world play a role in the interpretation of what is said, this is only because the same semantic facts obtain in the possible worlds of the context set.

IV  The Problem Now consider the following case, adapted from Barwise and Etchemendy (1987, 29, 121). Commenting on a poker game, I point to one of the players and say: “She has a good hand”. Like everyone else at the table, I take this person to be Celia. It never occurred to any of us that she might be someone else. We know Celia well, as she often plays poker with us. Still, this person is not Celia but her identical twin, Alice. None of us have ever heard or suspected that Celia had a twin sister. Alice is the player at our table in the actual world, Celia is the player at our table in some counterfactual worlds. As a matter of fact, Alice has a good hand. At the same time, by coincidence, Celia happens to be playing poker in another part of town, and she does not have a good hand there.10 The first thing to note about this case is that, unlike what happens with similar cases discussed in the literature, we have a very clear intuition about the truth value of my assertion here: it is true. Of course, all of us believe that the person sitting there is Celia. And so all of us believe that what I said in pointing to that person was that Celia has a good hand.Yet if we were right about this, what I said should be false, since Celia does not have a good hand at that time. But what I said is true, so that is not what I said. Rather, as all the players would acknowledge once apprised of the facts, what I said was that Alice—​that person over there to whom I am in fact pointing—​has a good hand, and since Alice does in fact have a good hand, what I said is true. As Donnellan might have put it, the point of my utterance was clearly to say something about that person (Alice) whose game I am observing. My referential intention is unambiguous because it is manifest that my evidence for attributing the property of having a good hand derives from the person whose game I am observing and to whom I am pointing. The reason I am pointing to that person is not because I believe that she is Celia, and the reason I am attributing that property to that person is not because I would believe on independent grounds that Celia has that property. I just want to say something about that person whose game I am observing, whoever she is, and incidentally, I happen to also believe that that person is Celia, but evidently that further belief is beside the point. Now the problem is this: Stalnaker’s account incorrectly predicts that what I asserted in this case is that Celia has a good hand, hence that what I asserted is false. Let us see how it predicts this. On Stalnaker’s analysis, in this scenario, the conversational participants are presupposing that the person who is sitting there is Celia. Here, as in Fregean cases, what we are presupposing cannot be the necessarily false proposition that Alice is Celia, since this amounts to the empty set of possible worlds. Instead, what we are presupposing must be some contingent proposition. As by stipulation we all know Celia well—​we see her on a regular basis, we have lots of true beliefs about her, we do not confuse her with anyone else, etc.11—​it is reasonable to suppose that Celia is represented rigidly across our alternatives. So the proposition we are presupposing should rather be the proposition that the 273

Gregory Bochner

person sitting at that table, whoever she is, is Celia. That proposition is contingent: it is true at some possible worlds, where the person sitting at the table is Celia, and false at other possible worlds (including the actual world), where the person sitting at the table is someone else. Now, once that contingent proposition is being presupposed, it is true at all the possible worlds of the context set that the person sitting at that table is Celia. From there, when everyone sees that I am pointing to the person sitting at that table and saying, “She has a good hand”, it is true at all the possible worlds of the context set that I am pointing to Celia and saying that she has a good hand. One and the same horizontal proposition, the proposition that Celia has a good hand (at that time), is expressed in all possible worlds of the context set.12 Hence Stalnaker’s account predicts that that proposition is what I said. However, this result is incorrect, since that proposition is false, whereas what I said was true.13

V  A Diagnosis It seems to me clear that this example creates a problem for Stalnaker’s metasemantic-​pragmatic account of assertion. Where does the problem come from? I think that the problem arises from two interconnected outcomes of the account. The first is holism about context: because the common presuppositions of all the participants at any given time of a conversation are all represented together by means of one set of possibilities, forming the context, any local change in what was presupposed necessarily alters the whole context. The second is that reference must always be determined in the world as we take it to be. When the facts in the actual world do play a role in determining reference, this is only indirectly and accidentally, because our contextual presuppositions about those actual facts happen to be true. This accidental role of the actual world violates a reality principle. The second outcome results from the first. Let us start with holism about context. As noted above, in the problematic example, I incidentally believe that the person whose game I am observing is Celia. This belief is totally irrelevant to the point of my statement, which is just to say, about that person whose game I am observing, that she has a good hand. But then, could Stalnaker not say that the presupposition that the person whose game I am observing is Celia does not play any role in determining what I said? No, he cannot say this. In his account the context (or “context ‘set’ ”) is represented by a set of possible worlds: the set of possible worlds compatible with what is presupposed by all participants at a given time. Any set of possible worlds is a proposition, so the context is a proposition. Once we come to presuppose the proposition that the person whose game I am observing is Celia, that proposition is added to the context: the posterior context then corresponds to the intersection of that proposition (a set of possible worlds) with the prior context (another set of possible worlds). From there, it becomes true at all worlds of the posterior context that the person whose game I am observing is Celia. At that point, it is too late: it is no longer possible to isolate that proposition from the rest of what is presupposed, so as to prevent it from affecting what is asserted. Then, what my utterance says as made within each possible world of the context is that that person/​Celia has a good hand, and so, according to Stalnaker’s account, the proposition that Celia has a good hand is what I asserted. In short, as all the presuppositions are packed together into one set of possibilities, there is no way to make only some but not all of those presuppositions operative in the determination of what is said. Let us now turn to the second outcome. Given holism about context, as the presupposition that the person whose game I am observing is Celia is false, the whole context is false: the actual world is not an element of the context. In particular, the facts determining the reference of my use of ‘she’ are different, and lead to different referents, in the actual world and in the world as we take it to be. Stalnaker’s account here predicts that the reference of the assertion I made in the actual world must be determined in the world as we take it to be, not in the actual world. However, the intuitions regarding the truth value of my assertion at the actual world show that the reference is fixed in the actual world. This highlights that Stalnaker’s account violates a ‘reality principle’:  in at least some 274

A Puzzle about Assertion

fundamental cases involving perception, the referents of the statements we make in and about the actual world are fixed directly in the actual world. Even internalists will grant this. But as it stands Stalnaker’s account is incompatible with this reality principle, for it predicts instead that reference is always determined in the world as the conversational participants take it to be. Stalnaker would probably be willing to grant the reality principle. And he might want to claim that it is compatible with his account. After all, he might insist, the horizontal proposition that my statement expresses in the actual world of interpretation, even if the actual world is not a member of the context set, is the proposition that Alice has a good hand, and that proposition is true at the actual world of evaluation—​as desired. However, it is important to see that there is an ambiguity in the claim ‘that my statement actually expresses the proposition that Alice has a good hand’. Everyone agrees that I was talking in the actual world. And clearly, when the statement I made in the actual world is interpreted in the actual world, what it says is that Alice has a good hand. So in one sense, indeed, it is true to say ‘that my statement actually expresses the proposition that Alice has a good hand’.Yet in another sense, which is the one that matters here, Stalnaker’s account predicts that it is false. On that account, the statement I made in the actual world is not interpreted in the actual world, but in the possible worlds of the context set, of which (in this case, as in many others) the actual world is not a member. Across the possible worlds of the context set, considered in their interpretive role, one and the same contingent proposition is expressed by the statement I made there. That proposition is the proposition that Celia has a good hand. Hence Stalnaker’s account predicts that that proposition is what I asserted. So in that sense, which is the one relevant to our problem, it is false to say ‘that my statement actually expresses the proposition that Alice has a good hand’. As strange as it may sound, in that relevant sense it should be true to say that what my actual statement actually expresses given that it is interpreted in the (nonactual) worlds of the context set of our actual conversation (the proposition that Celia has a good hand) is not what it would have expressed if it had been interpreted in the actual world (the proposition that Alice has a good hand). Now to remove this ambiguity in the (only) way that would dissolve our problem, Stalnaker might opt for the claim that ‘what my statement actually expresses’ unambiguously stands for what it expresses when interpreted in the actual world (even if the actual world is not in the context set). Given this, a new distinction would emerge between two notions: what is said (what a statement actually expresses: here, the proposition that Alice has a good hand) and what the conversational participants take what is said to be, given their presuppositions (what the statement expresses in the possible worlds of the context set: here, the proposition that Celia has a good hand). This move would dissolve the problem. However, it is important to realize that it would amount to a dramatic revision of Stalnaker’s pragmatic framework. First, it squarely implies that context—​in the pragmatic sense of common ground—​does not, after all, determine content. The pragmatic context continues to determine ‘what the conversational participants take what is said to be’, but ‘what is said’ is now determined in the actual world, independently of the pragmatic context. So this move avoids the problematic result that the actual world plays no role in determining what is said, but at the cost of jettisoning the central thesis that the pragmatic context determines what is said. Second, unlike the current metasemantic-​pragmatic version, it implies that what the conversational participants take what is said to be is not always identical to what is said. But then, it will generally be ‘what the participants take what is said to be’, and no longer ‘what is said’, that gets added to the common ground and thereby accounts for the rational dynamics of conversations. The content asserted can be what modifies the pragmatic context only by accident, when what the participants take what is said to be happens to be what is said—​i.e. when their presuppositions regarding what was said are true. This deprives ‘what is said’ of its theoretical roles of accounting for the distinctions that conversational participants take their assertions to be drawing among their common alternatives, and of being that which, once accepted and added to the common ground, reshapes the space of those alternatives. So this possible move, which prima facie looks innocuous, actually threatens the whole idea of a double interaction between context and content asserted. 275

Gregory Bochner

VI  Concluding Remarks: Actual vs. Presupposed Externalism What exactly does the problem have to do with the distinction between internalist and externalist versions of two-​dimensionalism? As I said, Stalnaker holds that his metasemantic interpretation of two-​dimensionalism is consonant with externalism about content because on that interpretation the horizontal propositions, which come before diagonal propositions in the order of explanation, are themselves determined in an externalist fashion. Within each possible world of interpretation w for a given utterance u, the horizontal proposition expressed by u in w is fixed by externalistic facts in w. But now of course, the latter shows no more than that, according to the metasemantic interpretation, conversational participants usually presuppose externalism to be true—​where that presupposition may be true or false. In order to assess whether the metasemantic interpretation necessitates the truth of externalism, we need to focus on its predictions regarding what is said in the actual world. But at that point, as we saw, the metasemantic account makes the locution ‘what a statement actually expresses’ ambiguous. To be sure, the facts determining which horizontal proposition ‘my statement actually expresses’ when it is interpreted in the actual world are externalistic facts holding in the actual world. Yet, when the actual world is not a member of the context set, the facts determining which horizontal proposition ‘my statement actually expresses’ given that it must be interpreted in the nonactual worlds of that context set are externalistic facts holding in those worlds. And in some of those worlds the externalistic facts may be different from the externalistic facts in the actual world: thus in our puzzle case, the externalistic facts in all the worlds of the context set involve a causal relation to Celia, while the externalistic facts in the actual world involve a causal relation to Alice. At that juncture, it will be useful to draw an explicit contrast between two versions of externalism: presupposed externalism and actual externalism. According to presupposed externalism, what is said by a referential statement made in the actual world is determined by externalistic facts determining reference in the possible worlds compatible with what the conversational participants presuppose. According to actual externalism, what is said by a referential statement made in the actual world is determined by externalistic facts determining reference in the actual world.The main points of the present paper may be put by saying (i)  that the metasemantic-​pragmatic interpretation of two-​dimensionalism is consonant with presupposed externalism but not with actual externalism, (ii) that presupposed externalism as a general theory is false, and so (iii) that, in one way or another, the metasemantic-​pragmatic interpretation will have to be revised so as to necessitate the truth of actual externalism. To conclude, I want to highlight that presupposed externalism has an internalistic flavor. Actual externalism provides a thoroughly externalist answer to the foundational question, ‘What are the facts in virtue of which a statement expresses the proposition that it expresses?’ But, perhaps surprisingly, upon close inspection, the answer yielded by presupposed externalism is internalist in spirit. Again, according to presupposed externalism, the facts that determine what is said within each possible world of interpretation w of a context set C at a given time t are externalistic facts holding in w. But now we may ask: in virtue of what facts is the context set C at t made up of those and only those possible worlds w? And the answer to that (meta-​)metasemantical question should now be this: for any world w found in C at t, the facts in virtue of which w is found in C at t include the fact, holding in the actual world, that at t the conversational participants share a certain description of the actual world. So in our puzzle case, the (meta-​)metasemantic facts in virtue of which my statement “She has a good hand” actually expresses (in the problematic sense) the proposition that Celia has a good hand involve, not externalistic facts determining reference in the actual world, but instead the fact that in the actual world the participants all presuppose a certain description of such externalistic facts. In other words, according to presupposed externalism, it is not the causal facts holding in the actual world, but the descriptions we entertain in the actual world concerning such facts, that determine which proposition my statement actually expresses (in the problematic sense).This means that, at some fundamental level, presupposed externalism actually amounts to a form of causal descriptivism, a thesis that is usually 276

A Puzzle about Assertion

associated with internalist responses to the externalist revolution, according to which it is internalized descriptions of causal relations, and not causal relations themselves, that determine reference.14

Acknowledgments The project leading to this publication has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Sklodowska-​Curie grant agreement No 799162. I began to write this paper in 2019 while I was a EURIAS Junior Fellow at the Paris Institute for Advanced Study (IAS), and it also benefited from my involvement in an interuniversity research project in Belgium funded by the Fonds National de la Recherche Scientifique (F.R.S.-​FNRS, PDR T.0184.16, 2016–​2020).

Notes 1 Beyond the various versions of two-​dimensionalism, which in many respects may count as neo-​descriptivist theories, other influential approaches include Evans’s (1982) non-​descriptivist Fregeanism, Salmon’s (1986) account invoking belief guises, Fine’s (2007) relationism, Fodor’s (2008) appeal to vehicles in the language of thought, or Recanati’s (2012) development of the mental file framework. 2 Influential pleas for controversial two-​dimensionalism include Stalnaker (1978/​1999a), Chalmers (1996, 2006), and Jackson (1998a). 3 On this contrast, see Stalnaker (1981/​1999a; 1987/​1999a, 122–​123; 1999b, 10). 4 Stalnaker cites Chalmers (1996, 2006) and Jackson (1998a) as the leading advocates of the semantic interpretation. 5 Within the same pragmatic family to which the metasemantic interpretation belongs, there is a further interpretation of the framework, which in Bochner (2013) I have called metasyntactic. It seems to me that the metasyntactic version, as stated there, will not help with the problem which I pose in the present paper for the metasemantic version. 6 On the distinction between semantics and metasemantics, see Kaplan (1989b, 573–​576) and Stalnaker (1997/​ 2003b; 2001/​2003b, 192). In a nutshell, the semantic facts about an expression specify the semantic value that the expression has, whereas the metasemantic facts are the facts in virtue of which that expression has the semantic value that it has. 7 For his externalist approach, see Stalnaker (1984, 2008). For the claim that his metasemantic interpretation is of a piece with his externalist approach, see Stalnaker (1999b; 2001/​2003b, 193–​194; 2003a, 206–​209; 2006). 8 This is particularly clear in Stalnaker (2008). 9 See Stalnaker (1978/​1999a, 1999b, 2014). 10 The original example in Barwise and Etchemendy (1987) involved a proper name:  “Claire has a good hand”. In that example, whose point was different, there were conflicting intuitions about the referent of the utterance. In my adapted example with the demonstrative, there is no conflict: we have a steady intuition about the referent of the utterance. 11 Of course, we are confusing her twin with Celia. But that does not mean that we are confusing Celia with her twin. 12 Note two points. First, the context is nondefective: the presuppositions of all the participants are the same, none of them is mistaken about what is common ground. On ‘nondefective contexts’, see Stalnaker (1978/​ 1999a, 85; 2014, 4). Second, no one has any doubt about the identity of the person I am observing (everyone is presupposing that she is Celia) and about what is being said: that is how the same horizontal proposition is expressed in all possible worlds of the context set. 13 Should we not conclude instead that, because the presupposition that the person I am observing is Celia was false, Stalnaker’s account predicts that what I said was neither true nor false? No. As Stalnaker (1970/​1999a, 39) writes: “Presuppositions, of course, need not be true. Where they turn out false, sometimes the whole point of the inquiry, deliberation, lecture, debate, command or promise is destroyed, but at other times it does not matter much at all. Suppose, for example, we are discussing whether we ought to vote for Daniels or O’Leary for President, presupposing that they are the Democratic and Republican candidates, respectively. If our real interest is in coming to a decision about who to vote for in the Presidential election, then the debate will seem a waste of time when we discover that in reality, the candidates are Nixon and Muskie. However, if our real concern was with the merits of the character and executive ability of Daniels and O’Leary, then our false presupposition makes little difference. Minor revisions might bring our debate in line with new presuppositions”. Stalnaker would clearly regard the present case as falling into that second category, as the

277

Gregory Bochner false presupposition that the person I am observing is Celia makes no difference to my point. Again, once apprised of the facts, the participants would regard what I said as true, not as neither true nor false. 14 Causal descriptivist claims are found in Loar (1976, 1980), Searle (1983, ch. 9), Lewis (1984), Kroon (1987), Jackson (1998b), and Chalmers (2002). Stalnaker officially dismisses causal descriptivism, and criticizes its internalist underpinnings, in Stalnaker (2004, 2006).

References Barwise, J. and J. Etchemendy (1987). The Liar: An Essay onTruth and Circularity. New York: Oxford University  Press. Bochner, G. (2013). “The Metasyntactic Interpretation of Two-​ Dimensionalism”, Philosophical Studies 163(3): 611–​626. Chalmers, D. (1996). The Conscious Mind. Oxford: Oxford University Press. Chalmers, D. (2002). “On Sense and Intension”, Philosophical Perspectives 16(16): 135–​182. Chalmers, D. (2006). “The Foundations of Two-​Dimensional Semantics”. In M. García-​Carpintero and J. Macià (Eds.), Two-​Dimensional Semantics: Foundation and Applications. Oxford: Oxford University Press, 55–​140. Evans, G. (1982). The Varieties of Reference. Oxford: Clarendon Press. Fine, K. (2007). Semantic Relationism. Oxford: Blackwell. Fodor, J. (2008). LOT 2: The Language of Thought Revisited. Oxford: Oxford University Press. Frege, G. (1892). Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik 100: 25–​50. Translated as “On Sense and Reference” by P. Geach. In P. Geach and M. Black (Eds.), Translations from the Philosophical Writings of Gottlob Frege. Oxford: Blackwell, 1970, 56–​78. García-​Carpintero, M. and Macià, J. (2006). Two-​Dimensional Semantics: Foundation and Applications. Oxford: Oxford University Press. Jackson, F. (1998a). From Metaphysics to Ethics. Oxford: Oxford University Press. Jackson, F. (1998b). “Reference and Description Revisited”. Philosophical Perspectives 12(12): 201–​218. Kaplan, D. (1989a). “Demonstratives”. In J. Almog, J. Perry, and H. Wettstein (Eds.), Themes from Kaplan. Oxford: Oxford University Press, 481–​563. Kaplan, D. (1989b). “Afterthoughts”. In J. Almog, J. Perry, and H. Wettstein (Eds.), Themes from Kaplan. Oxford: Oxford University Press, 565–​614. Kripke, S. (1980). Naming and Necessity. Cambridge, MA.:  Harvard University Press. (From “Naming and Necessity”. In D. Davidson and G. Harman (Eds.), Semantics of Natural Language. Dordrecht: Reidel, 1972, 253–​355.) Kroon, F. (1987). “Causal Descriptivism”, Australasian Journal of Philosophy 65(1): 1–​17. Lewis, D. (1984). “Putnam’s Paradox”, Australasian Journal of Philosophy 62(3): 221–​236. Loar, B. (1976). “The Semantics of Singular Terms”, Philosophical Studies 30(6): 353–​377. Loar, B. (1980). “Names and Descriptions: A Reply to Michael Devitt”, Philosophical Studies 38(1): 85–​89. Recanati, F. (2012). Mental Files. Oxford: Oxford University Press. Salmon, N. (1986). Frege’s Puzzle. Cambridge, MA: MIT Press. Searle, J. (1983). Intentionality: An Essay in the Philosophy of Mind. Cambridge: Cambridge University Press. Stalnaker, R. (1970). “Pragmatics”, Synthese 22: 272–​289. (Reprinted in his Context and Content. 1999a. Page references to the latter.) Stalnaker, R. (1978). “Assertion”. In P. Cole (Ed.), Syntax and Semantics 9. New York: Academic Press, 315–​332. (Reprinted in his Context and Content. 1999a. Page references to the latter.) Stalnaker, R. (1981).“Indexical Belief ”, Synthese 49(1): 129–​151. (Reprinted in his Context and Content. 1999a.) Stalnaker, R. (1984). Inquiry. Cambridge MA: MIT Press. Stalnaker, R. (1987). “Semantics for Belief ”, Philosophical Topics 15(1): 177–​190. (Reprinted in his Context and Content. 1999a. Page references to the latter.) Stalnaker, R. (1997). “Reference and Necessity”. In C. Wright and B. Hale (Eds.), A Companion to the Philosophy of Language. Oxford: Blackwell, 534–​554. Stalnaker, R. (1999a). Context and Content. Oxford: Oxford University Press. Stalnaker, R. (1999b). “Introduction”. In Context and Content. Oxford: Oxford University Press, 1–​28. Stalnaker, R. (2001). “On Considering a Possible World as Actual”, Proceedings of the Aristotelian Society, (Supplementary Volume) 75: 141–​156. (Reprinted in his Ways a World Might Be: Metaphysical and Anti-​ Metaphysical Essays. 2003b, 188–​200. Page references to the latter.) Stalnaker, R. (2003a). “Conceptual Truth and Metaphysical Necessity”. In Ways a World Might Be: Metaphysical and Anti-​Metaphysical Essays. Oxford: Clarendon Press, 201–​215. Stalnaker, R. (2003b). Ways a World Might Be: Metaphysical and Anti-​Metaphysical Essays. Oxford: Clarendon Press. Stalnaker, R. (2004). “Lewis on Intentionality”, Australasian Journal of Philosophy 82(1): 199–​212.

278

A Puzzle about Assertion Stalnaker, R. (2006). “Assertion Revisited”. In M. García-​Carpintero and J. Macià (Eds.), Two-​Dimensional Semantics: Foundation and Applications. Oxford: Oxford University Press, 293–​309. Stalnaker, R. (2007). “Critical Notice of Scott Soames’s Case against Two-​Dimensionalism”, The Philosophical Review, 116(2): 251–​266. Stalnaker, R. (2008). Our Knowledge of the Internal World. Oxford: Oxford University Press. Stalnaker, R. (2014). Context. Oxford: Oxford University Press.

279

PART VI

Natural Kind Terms and Rigidity

21 RIGIDITY OF GENERAL TERMS Genoveva Martí and José Martínez-​Fernández

The Notion of Rigidity The notion of rigidity, introduced by Saul Kripke in his 1970 lectures on Naming and Necessity (Kripke 1980), was at the origin of an important revolution in semantics. Kripke provided powerful arguments against the well-​established descriptivist theory, a theory that postulated that names are in fact abbreviated definite descriptions. According to Kripke, proper names and definite descriptions have a very different semantic behavior. For instance, the truth conditions of a sentence such as ‘Aristotle was a philosopher’ depend on Aristotle. However, the truth conditions of ‘the tutor of Alexander the Great was a philosopher’ do not depend only on Aristotle, even though Aristotle was in fact the tutor of Alexander the Great, for in a counterfactual situation or a possible world in which Philippus decided to hire the palace cook to tutor his son, the sentence that contains the description would be false, even if in that same possible world Aristotle had decided not to switch careers, and so, ‘Aristotle was a philosopher’ would still be true. Differences in truth conditions reveal semantic differences and in the case we just considered the difference is due to the substitution of a definite description for a proper name. Those differences led Kripke to characterize names as rigid designators, terms that designate the same individual in every possible world. Not only names, according to Kripke, are rigid designators, some definite descriptions are rigid too. For instance, the description ‘the successor of 8’ designates or denotes the number 9 in every possible world and if being the offspring of gametes X and Y captures an essential property possessed uniquely by Aristotle, ‘the offspring of gametes X and Y’ would be rigid too. Kripke proposes also “an intuitive test” for rigidity: although the tutor of Alexander might have failed to be the tutor of Alexander, Aristotle could not have failed to be Aristotle and similarly the successor of 8 could not have failed to be the successor of 8 and the offspring of gametes X and Y could not have failed to be the offspring of gametes X and Y, which according to Kripke shows that ‘Aristotle’, ‘the successor of 8’, and ‘the offspring of gametes X and Y’ are rigid designators. Thus, Kripke proposes, in fact, two characterizations of what it is for a term to be rigid. The first one appeals to the semantic idea of sameness of designation and sameness of truth conditions in every possible world. The second one, associated with the intuitive test, on the other hand, is based on a metaphysical notion of necessity or essence.1 A consequence of both characterizations of rigidity that will be relevant for the discussion below is that true sentences of the form t is t′ are necessary, if t and t′ are rigid designators. So, on Kripke’s 283

Genoveva Martí and José Martínez-Fernández

view, the rigidity of proper names and of some definite descriptions accounts for the necessity of true sentences such as ‘Hesperus is Phosphorus’ or ‘Aristotle was the offspring of gametes X and Y’.

Extending Rigidity beyond Singular Terms: Three Desiderata The notion of rigidity is introduced by Kripke to apply to singular terms, but Kripke also claimed that the distinction between rigid and non-​r igid terms applies to at least some general terms: [M]‌y argument implicitly concludes that some general terms, those for natural kinds, have a greater kinship with proper names than is generally realized. This conclusion holds for certain for various species names, whether they are count nouns, such as ‘cat’, ‘tiger’, ‘chunk of gold’, or mass terms such as ‘gold’, ‘water’, ‘iron pyrites’. It also applies to certain terms for natural phenomena, such as ‘heat’, ‘light’, ‘sound’, ‘lightning’ and, presumably, suitably elaborated, to corresponding adjectives—​‘hot’, ‘loud’, ‘red’. Kripke 1980, 134 Hence, according to Kripke, proper names and natural kind terms are semantically very similar. But it is not obvious how the notion of rigidity applies to those terms: whereas a rigid singular term has the same denotation, or extension, in every possible world, the general terms that Kripke considers rigid could have different extensions: had the world been different, there would be more cats, less gold, or fewer red things. Since Kripke did not make explicit what the distinction between rigid and non-​rigid general terms consists in, it is an open question whether the notion can be coherently extended to those terms.The most prominent criticism of the very idea of attempting to extend the distinction between rigid and non-​r igid terms beyond the realm of singular terms comes from Scott Soames (2002). The application of rigidity to general terms, according to Soames (2002, 263), should satisfy three basic desiderata: (i) it should be a natural extension of the notion of rigidity that Kripke introduces for singular terms, (ii) it should not overgeneralize, that is, it should characterize natural kind terms as rigid and other terms as non-​r igid,2 (iii) it should contribute to explaining the necessity of theoretical identifications, of sentences such as ‘water is H2O’ or ‘cats are felines’, much in the same way that the rigidity of ‘Hesperus’ and ‘Phosphorus’ plays a role in accounting for the necessity of ‘Hesperus is Phosphorus’. Soames argues that no coherent extension of the notion of rigidity to general terms can satisfy the three desiderata. In the case of general terms (much as was the case for singular terms) there are also two ways of understanding the distinction between rigid and non-​r igid. One of them is based on the distinction between essential and non-​essential properties and the other one exploits the idea that a term is rigid if it designates the same entity in every possible world. We will examine them separately.

The Essentialist Characterization of Rigidity for Kind Terms Rigid Application On the essentialist characterization of the rigidity of kind terms, a general term is rigid just in case it expresses a property that is essential to its bearers, where an essential property is taken to be a property

284

Rigidity of General Terms

an object cannot fail to possess if it exists. Michael Devitt, for instance, characterizes the notion as follows: [A]‌general term ‘F’ is a rigid applier iff it is such that if it applies to an object in any possible world, then it applies to that object in every possible world in which the object exists. Devitt 2005, 1463 Hence a tiger cannot fail to be a tiger and a sample of water cannot fail to be a sample of water, in the same way that Aristotle cannot fail to be Aristotle. This way of capturing the notion of rigidity satisfies the first desideratum proposed by Soames, since it is an extension of the intuitive test for rigidity that Kripke proposes for singular terms. This characterization of rigidity satisfies also the second desideratum, since it does not overgeneralize: although it characterizes natural kind terms as rigid, artifactual terms (‘pencil’, ‘computer’ …) and non-​natural kind terms (‘bachelor’, ‘pediatrician’ …) turn out to be non-​r igid.4

Essentialist Rigidity and the Necessity of Theoretical Identifications The satisfaction of the third desideratum is, however, much more problematic. The essentialist characterization of rigidity treats natural kind terms as predicates. Hence a theoretical identification such as ‘water is H2O’ should be interpreted as a universally quantified biconditional: ∀x (Water x ↔ H2O x). Since this is a true sentence that contains two rigid predicates, this sentence should be necessary if the analogy with the case of singular terms held, or so it seems. But the necessity of this sentence does not follow from its truth plus the rigidity of the terms involved. In general, the satisfaction of the third desideratum, if rigidity is understood as essentiality, should make the following schema valid: ∀x (P x ↔ Q x), P and Q rigid ∴  ∀x (P x ↔ Q x).

But in general this schema is invalid. By the first premise, P and Q have the same extension at the actual world, and by the second premise, if P applies to an individual in any given world it applies to that individual in any world in which the individual exists, and the same goes for Q. All that does not preclude that there could be an individual a that exists at a world w and does not exist in the actual world such that a is P in w (and in all those worlds in which it exists) without being Q in w (nor in any world in which it exists). Under those conditions obviously the sentence  ∀x (P x ↔ Q x) is not true and, hence, the necessity of theoretical identifications is not accounted for by the rigidity of the predicates involved. Michael Devitt (2005) accepts that this is a serious problem for the characterization of rigidity that he endorses. But Mario Gómez-​Torrente (2004, 2006) has proposed a refinement of the approach that is meant to free the essentialist characterization of this problem. According to Gómez-​Torrente, the quantifiers in theoretical identification sentences should be interpreted as possibilist quantifiers, namely, quantifiers that range over actual and possible individuals. Moreover, rigid predicates should be understood as obstinately essential, meaning that if an essentialist predication Pa is true at a world, Pa is true in all possible worlds, including worlds in which a does not exist. The commitment to possibilist quantification, which may be seen as objectionable by some on the basis of metaphysical qualms, is, according to Gómez-​Torrente, entirely justified, for theoretical identifications are not supposed to be just statements about what there actually is and how it behaves. Rather, they are claims about the properties of any possible object or sample of the kind in question.

285

Genoveva Martí and José Martínez-Fernández

Obstinate essentiality, on the other hand, is the counterpart of the notion of obstinate rigidity for singular terms, proposed by Salmon (1981, 34). Gómez-​Torrente (2006, 252) contends that Kripke himself is inclined to treat names as obstinately rigid, hence, there should be no problem accepting the notion of obstinate essentiality.Therefore, in Gómez-​Torrente’s view, the essentialist characterization of the rigidity of general terms can overcome the most pressing objection posed by Soames and satisfy the third desideratum. If someone objects to obstinate rigidity and obstinate essentiality, Gómez-​Torrente (2004, 2006)  argues also that there are weaker notions of rigidity and essentiality that justify the argument from a theoretical identification sentence to the necessity of a qualified counterpart of the conclusion—​a counterpart restricted to those worlds in which the relevant objects exist. Ahmed (2009) has argued that a weak notion of rigidity that would justify only the qualified necessity of true theoretical identifications does not fulfill the role Kripke intends theoretical identifications to play: revealing the essential properties of natural kinds and contributing crucially in his criticism of psychophysical identifications. Gómez-​Torrente (2009) disagrees with Ahmed’s interpretation of Kripke’s objectives and defends the appeal to the necessitation of qualified theoretical identifications.5 Soames (2006) has also raised concerns.

Other Issues There are other issues that affect the overall assessment of the essentialist characterization. Note that whether a property is essential or accidental may well depend on the possessor of the property: one and the same property may be essential to something and accidental to something else. For instance, having the shape of a cube is not essential to an avant-​garde bench in a park, but it is essential to pyrite crystals. And, as Jessica Wilson has pointed out,6 being charged may be essential to an electron, but it is inessential to persons. As rigid essentiality is characterized, a property is essential or accidental tout court (given the presence of the universal quantifier in the definition). It is not clear that the relativity of essence can be acknowledged and how the definition of a rigid, essentialist, predicate should be refined in order to account for it. The essentialist characterization of rigidity privileges the predicative function of kind terms as the basic one. Uses of kind terms to talk about species or substances are considered derivative and based on the fundamental function. Soames (2002, 245–​248) argues that even though the singular names-​ like use of kind terms may seem plausible for mass nouns, in general it is the predicative form that has priority. Often, we learn what it means to say that a is an instance of the kind, or property, P* by being told, or figuring out, that necessarily a is an instance of P* iff a is P, where the latter, a is P, is something we already understand and take as basic. For example, Jim’s shirt is red is necessarily equivalent to Jim’s shirt is an instance of the color red, and that object is circular is necessarily equivalent to that object is an instance of circularity; but it is the predicative form, is red and is circular, that is basic, and the nominal form, the color red and circularity, that is derivative. Soames 2002, 246–​247 So, there are substantial reasons to focus primarily on the predicative use of kind terms. However, neglecting other uses has, as a result, the consequence that it is unclear that the essentialist characterization satisfies the first desideratum, namely, that it is a natural extension of the notion of rigidity for singular terms. Because of their focus on the predicative function of kind terms, Devitt and Gómez-​Torrente do not make distinctions between essences of kinds and essences of individuals, between what a kind may or may not fail to be and the predicates that can or cannot fail to apply to individuals. Consequently, it is questionable that Devitt and Gómez-​Torrente are applying the correct

286

Rigidity of General Terms

counterpart of the intuitive test for rigidity that Kripke uses for singular terms: … could not have failed to be _​_​_.​ As Ilhan Inan has pointed out, In ordinary discourse we could … [say] ‘the colour blue might not have been the colour blue’ … and ‘blue things might not have been blue’ … It seems to me clear that when Kripke introduced this test to decide whether a particular general term is rigid, he meant the former and not the latter … [It] would be incorrect to say ‘Truth might not have been truth’, but correct to say ‘true propositions might not have been true’ for at least some propositions. From the latter claim, nothing follows about the rigidity of the term; it is the former that is important. Inan 2008, 219 According to the essentialist characterization of rigidity, the sentence ‘water could not have failed to be H2O’ is true. Hence applying Kripke‘s intuitive test, we should conclude that ‘water’ and ‘H2O’ should be classified as rigid. But we think that it is clear that the relevant intuitions here have to do with what we take to be the essence of the kind, the substance water, not with the essence of the individual samples that instantiate being water.7 This issue does not only arise for mass terms. When we consider count nouns such as ‘cat’, our judgment of sentences such as ‘cats could not have failed to have the type cat* of DNA’ is based on considerations about what it takes to be a cat, not on intuitions about the essential properties of Nana and other individual cats. By focusing primarily on the predicative function of kind terms, the essentialists do not acknowledge the importance of the distinction between essences of kinds and essences of individuals, and they overlook the role that the essences of kinds play in the evaluation of sentences like the ones above. That casts doubt on the satisfaction of the first desideratum, because the type of essentiality relevant in the application of Kripke’s test has to do with the essential properties of kinds, not with rigid application. Moreover, Soames has also put into question that the essentialist characterization of rigidity satisfies the second desideratum: some of the natural kind predicates that Kripke considers rigid are not essentialist, for instance, an object that is actually hot may fail to be hot in counterfactual situations.8 The essentialist characterization certainly does not overgeneralize rigidity, but it appears rather that it undergeneralizes it.

The Denotationalist Characterization of Rigidity for General Terms Rigidity as Sameness of Designation in Every Possible World On the denotationalist characterization, a general term is rigid just in case it refers to the same kind of thing (substance, species …) in every possible world. The metaphysics that underlies this characterization takes those kinds to be universals, abstract objects that general terms refer to. Hence, from this perspective, the primary semantic role of kind terms is to denote kinds. The predicative function of those terms can then be understood by appeal to the relation of instantiation. Both the denotationalist and the essentialist characterization of general terms share the previously mentioned schema proposed by Soames: ‘a is an instance of P* iff a is P’ (Soames 2002, 246) but they disagree on which side of the biconditional is the primary one. Soames himself acknowledges that when we consider mass terms such as ‘water’, it is initially plausible to think that we use the word to talk about a substance. This characterization of rigidity extends that referential function to all general terms just as the essentialist characterization extends the more natural predicative reading of count nouns to all general terms. According to the denotationalist stance, terms such as ‘gold’, ‘the substance with atomic number 79’, ‘blue’, or ‘cat’ are rigid because they refer to the same substance, the same color or the same

287

Genoveva Martí and José Martínez-Fernández

species in every possible world. Terms such as ‘the precious metal wedding rings are traditionally made of ’, ‘Mary’s favorite color’, or ‘the furry pets that scratch and purr’ are not rigid since they can designate different substances, colors, or species in different possible worlds. This characterization obviously satisfies the first desideratum, as the denotationalist definition of rigidity for kind terms coincides with the definition for singular terms: a term, singular or general, is rigid if it designates the same entity in every possible world. It is unclear, however, whether this view can satisfy the second and the third desiderata. On the one hand, non-​natural or artifactual kind terms such as ‘philosopher’ or ‘pencil’ arguably designate the same kind in every possible world, so this characterization of rigidity overgeneralizes. On the other hand, it can be argued that this approach fails to account for the necessity of theoretical identifications. Sentences such as ‘cats are felines’ and ‘water is H2O’ are taken to have respectively the form ∀x (P x → Q x) and ∀x (P x ↔ Q x). Even if P and Q are rigid, these two sentences may fail to be necessary, since P and Q may designate different universals whose extensions happen to satisfy the conditional or biconditional in the actual world, and do not satisfy them in some other worlds. Actual examples are not easy to come by, since this is fundamentally a logical point, but perhaps one example would be the predicates ‘renate’ and ‘cordate’. Although in the actual world something is a renate iff it is a cordate, these two predicates, although rigid, may fail to have the same extension in a different possible world if it is possible to have a heart without having kidneys.9 Before discussing the impact of these worries, there is a more pressing issue that affects the denotationalist characterization of rigidity.10

The Problem of Trivialization Soames (2002) and Schwartz (2002) pointed out that this approach trivializes the notion of rigidity, for it makes it impossible to distinguish between rigid and non-​r igid kind terms. Any kind term P, simple or complex, can be understood as rigidly designating P-​ness. Any sentence of the form ‘a is P’ or ‘a is a P’ is true at a given world just in case a instantiates P-​ness at that world.Thus, for instance, a kind term such as ‘the color of the sky’ can designate non-​r igidly different colors at different possible worlds: in the actual world it designates the color blue but in a world where the chemical composition of the atmosphere is different it designates the color green. But there is another way of reading ‘the color of the sky’ according to which the term designates the same property in every possible world, namely, it designates rigidly the property of being the color of the sky.This property is not a color; it is the property that things have at a world w when they have the same color as the sky in w.11 Consider a sentence such as ‘my true love’s eyes are the color of the sky’, understood as a sentence in which ‘the color of the sky’ is used as a predicate.12 On the non-​r igid reading, no doubt, the most natural one, this sentence is true in the actual world just in case e, the referent of ‘my true love’s eyes’ instantiates blueness; and at a different possible world w in which ‘the color of the sky’ designates the color green the sentence is true if e instantiates greenness. In the case of the rigid reading the sentence is true at any given world just in case e instantiates the property of being the color of the sky at that world. Since the sky is blue in the actual world, the sentence is true if e instantiates blueness and in w, where the sky is green, the sentence is true if it instantiates greenness. Since this can be generalized to all worlds, no difference in truth conditions arises between the rigid and the non-​r igid reading of ‘the color of the sky’. Moreover, the argument applies to any general term whatsoever. In general, take any sentence of the form ‘a is P’ (or ‘a is a P’) in which P is a non-​r igid general term and consider any possible world w. By hypothesis, P designates in w a certain property P(w)* and ‘a is P’ is true at w iff a belongs to the extension of P(w)* at w. But now consider P as rigidly designating the property R* whose extension at any world w′ coincides with the extension of P(w′)* at w′. Then ‘a is P’ is true at w iff a belongs to the extension of R* at w iff (by the definition of R*) a belongs to the extension of P(w)* at w. The truth conditions of ‘a is P’ in the non-​r igid reading of P coincide with the truth conditions in the rigid reading of P. So we might as well consider all kind terms rigid, because we are not going to detect any difference in truth conditions between the intuitive non-​r igid readings and the rigid ones. 288

Rigidity of General Terms

Joseph LaPorte has argued that treating general terms as designators and defining rigidity as sameness of designation does not trivialize the notion of rigidity: My preferred account does not trivialize rigidity. It is simply not the case that every kind designator rigidly designates its kind … ‘soda pop’, ‘soda’, and ‘pop’ all rigidly designate the soda pop kind; but ‘the beverage my uncle requests at Super Bowl parties’ only accidentally designates the kind. LaPorte 2000, 296 and 299 Dan López de Sa (2006) argues that if a sentence such as ‘being blue is being the color of the sky’ is contingent, at least one of the general terms must be interpreted non-​r igidly. LaPorte’s and López de Sa’s remarks, however, do not show that there is any difference in truth conditions between rigid and non-​r igid interpretations of general terms when those have a clearly predicative role. Even Soames has agreed that there is a difference between rigid and non-​r igid readings of kind terms when they figure in sentences that say something about the kinds those terms designate: I now think that there is a natural way of extending Kripke’s distinction between rigid and non-​r igid designators from singular to general terms—​even though it does not extend to predicates constructed from those terms. Soames 2004, 95 The sentence ‘blue is the color of the sky’ on the non-​r igid reading of ‘the color of the sky’ is true and contingent, but on the rigid reading of ‘the color of the sky’ it is false, simply because the universal blue is different from the universal being the color of the sky. But the trivialization problem remains, when those complex general terms are used as predicates. It is in those cases that no difference in truth conditions surfaces. Another reply to the trivialization objection has been proposed by Nathan Salmon (2003). Salmon asks us to consider the following valid argument: My true love’s eyes are the color of the sky. The color of the sky is blue. Therefore, my true love’s eyes are blue. Salmon argues that the natural interpretation of ‘the color of the sky’ in the second premise, a premise in which ‘the color of the sky’ and ‘blue’ are not used predicatively, is a non-​rigid interpretation, since the sentence is a true contingent identity.13 In order to explain the argument’s intuitive validity, Salmon contends, we should interpret ‘the color of the sky’ also as non-​r igid in the first premise, in which ‘the color of the sky’ is used predicatively. The assumption is that in order for the argument to be valid, the terms involved must have the same semantic value, lest we commit a fallacy of equivocation.The alleged conclusion is that since there is a unique non-​r igid reading that makes the second premise true, the non-​r igid reading of the general term in the first premise is non-​optional either and hence the denotational notion of rigidity is not trivialized.14 However, it is not clear that ‘the color of the sky’ has to be interpreted consistently as non-​r igid to preserve the intuitive validity of the argument. For instance, the following argument, that contains an explicit rigid reading of ‘the color of the sky’ in the first premise and a non-​r igid reading in the second one (clearly a contingent sentence), is also valid: Loch Ness has the property of being the color of the sky. The color of the sky happens to be blue. So, Loch Ness is blue. 289

Genoveva Martí and José Martínez-Fernández

Moreover, the first premise of Salmon’s argument, where ‘the color of the sky’ has a predicative function, has the same truth conditions whether ‘the color of the sky’ is interpreted rigidly or non-​ rigidly, so the trivialization problem persists. One can try to dismiss entirely the whole problem of trivialization by arguing that the entities that are rigidly designated by the rigid readings of expressions such as ‘the color of the sky’ are gerrymandered or unnatural. For instance, LaPorte discusses the possibility that the general term ‘the beverage my uncle requests at Super Bowl parties’ designates the kind ‘beverage my uncle requests at Super Bowl parties (call it the “BMURASP-​kind”). The latter kind is instantiated, in any possible world, by the beverage my uncle requests at Super Bowl parties in that world’ (LaPorte 2000, 300). It could be argued that these entities are not really things that exist, things that we may have any interest in talking about.15 But, although there are cases in which this seems definitely the right diagnosis, for instance when we consider “the kind” BMURASP, there are other cases in which the property rigidly designated by a complex general term has real interest: it figures in relevant empirical regularities and it is arguably an object of scientific study. The property of being the color of the sky designated rigidly by ‘the color of the sky’ is one such property. Large quantities of water (such as lakes) do have this property because of their constitution and reflectance properties. The term ‘the highest altitude commercial planes can fly at’ has a non-​r igid reading according to which it designates different altitudes at different worlds. But it also has a rigid reading according to which it designates a property that depends, among other things, on the density of the air and which is of interest to engineers to talk about and to study. As long as we accept that there are some natural rigid readings of complex general terms, trivialization is a problem, and it requires a solution. The problem, as we have seen, arises when complex general terms are used as predicates. In Martí and Martínez-​Fernández (2010, 2011), we propose a solution to the problem. Relying on ideas originally suggested by Bernard Linsky (1984 and 2006), we argue that there are sentences in which rigid and non-​rigid readings of general terms make a difference to truth conditions. Those differences do arise when complex general terms are placed under the scope of different intensional operators. The presence of those operators makes the difference in truth conditions surface.16

Overgeneralization and the Necessity of Theoretical Identifications Let us consider now the two other desiderata that the notion of rigidity for general terms is supposed to satisfy. First, let us focus on the second desideratum, according to which the extension of rigidity to general terms should not overgeneralize, namely, nearly all natural kind terms should be rigid, whereas many other general terms should be non-​r igid.17 The examples of general terms that Kripke proposes as semantically similar to proper names are terms for natural kinds (species, substances, natural phenomena). The natural assumption has been that Kripke meant to exclude terms such as ‘bachelor’, ‘funny’, or ‘pencil’ from the category of rigid terms.18 Clearly the view that rigid general terms designate the same universal in every possible world fails to make the distinction between terms that designate natural kinds or phenomena and other general terms. So this view does not satisfy the second desideratum proposed by Soames. Other authors, however, reject that such a distinction should be made and hence deny that the second desideratum should be satisfied. On the denotationalist characterization of rigidity, ‘tiger’ designates rigidly the kind tiger or the property of being a tiger just as ‘bachelor’ designates rigidly the kind bachelor or the property of being a bachelor. Moreover, if we focus primarily on semantic behavior, simple general terms are like proper names: names name individuals, simple general terms name kinds or universals. Proper names and simple general terms are equally rigid, and the rigidity of the latter is, on this view, independent of whether they designate natural or non-​natural kinds.19 290

Rigidity of General Terms

Schwartz (2018) objects to the extension of rigid designation to non-​natural kinds. According to Schwartz, treating all general terms uniformly as designators obscures the fact that there is an important semantic difference between a sentence such as ‘water is H2O’ and a sentence such as ‘a pediatrician is a doctor’. The former is a posteriori, and the latter is a priori. The necessity of the latter derives directly from the definition of the words ‘pediatrician’ and ‘doctor’. It is an analytic sentence. ‘Water is H2O’, on the contrary, is not analytic. It is a theoretical identification that, according to Kripke, is necessary in virtue of the rigidity of the terms involved. This brings us to the discussion of the third desideratum. Kripke argues that sentences of the form ‘t1 is t2’ are necessary if true whenever t1 and t2 are rigid designators. The same should apply to general terms: if P and Q are rigid, sentences of the form ‘P is Q’ or ‘Ps are Qs’ should be necessary if true. On the assumption that only natural kind terms are rigid (as defenders of the second desideratum contend), these sentences have been taken to be scientific or theoretical identifications (as defenders of the third desideratum claim), but obviously the point is more general. Even if one rejects the second desideratum, as the denotationalist view does, the validity of the claim that sentences of the form ‘P is Q’ or ‘Ps are Qs’ should be necessary if true is not threatened; its validity just does not depend on whether P and Q designate natural or non-​natural kinds or properties.20 Natural language sentences of the form ‘P is Q’ or ‘Ps are Qs’ such as ‘water is H2O’ or ‘bees are Apis mellifera’ can be interpreted as sentences in which the general terms function as predicates. In that case, the form of these sentences can be captured in terms of a universally quantified statement such as ∀x (Px ↔ Qx) and, as we have seen, the rigidity of P and Q and the truth of the statement does not guarantee its necessity, for P and Q may well rigidly designate two different universals that happen to apply to the same things. But sentences such as ‘water is H2O’ or ‘bees are Apis mellifera’ can also be interpreted as sentences that identify two kinds, sentences that are true if the universal designated by P and Q is the same. In these cases the rigidity of P and Q does guarantee the necessity of the statement.21 The third desideratum, as it is stated by Soames, is satisfied by the denotationalist view but only if theoretical identifications of the form ‘P is Q’ or ‘Ps are Qs’ are understood as claims about kinds and not as universally quantified sentences about the members or samples of the kind.22

Flying Rigidity Both the essentialist and the denotationalist characterizations of rigidity constitute arguably a coherent extension of the notion of rigidity beyond singular terms. It is true though that in both cases such an extension requires adjustments and reinterpretations of what should be required of an acceptable theory of rigidity. This has led some authors to eschew the application of the notion of rigidity beyond the realm of singular terms. For instance, Soames thinks that the essentialist characterization of rigidity cannot satisfy the third desideratum and that the denotationalist characterization cannot satisfy the second desideratum and moreover that it faces the trivialization problem. This leads him to advise ‘to reserve the terminology of rigidity exclusively to singular terms’ and explore another ‘semantic property of natural kind predicates that is importantly similar to a corresponding property of proper names’ (Soames 2002, 263). In fact, Soames proposes nondescriptionality as the relevant semantic property that should be discussed in this debate. Soames focuses primarily on Kripke’s claim that ‘natural kind terms have a greater kinship with proper names than is generally realized’ (Kripke 1980, 134). Some of the most powerful arguments in Naming and Necessity are designed to prove the nondescriptionality of proper names, and simple natural kind terms are in this respect very similar to proper names, in that the correct application of a simple kind term is not determined by the descriptions speakers associate with those terms. One could question, however, whether nondescriptionality constitutes the dividing line between natural and non-​natural kind terms. Hilary Putnam (1975) himself casts doubt on the alleged descriptionality of artifactual terms: 291

Genoveva Martí and José Martínez-Fernández

Let us consider for a moment the names of artifacts—​words like ‘pencil’, ‘chair’, ‘bottle’, etc. The traditional view is that these words are certainly defined by conjunctions, or possible clusters, of properties … (But) ‘pencil’ is not synonymous with any description—​not even loosely synonymous with a loose description. When we use the word ‘pencil’, we intend to refer to whatever has the same nature as the normal examples of the local pencils in the actual world. Putnam 1975, 242–​243 And other authors, as we have seen, reject that there is any semantic difference between natural kind terms and other general terms.23 Nimtz (2019), however, claims a special status for proper names and simple natural kind terms, although he does not base that claim on the nondescriptionality of those expressions. Nimtz agrees with Soames that the application of the notion of rigidity beyond the realm of singular terms is ultimately fruitless, but he argues that natural kind terms and proper names are both types of paradigm terms in that their application is governed by a relation anchored in paradigm actual objects.24 In any case, the question as to whether rigidity can, or should be extended to general terms, is still a debatable issue that is connected with semantic, metaphysical and epistemic questions.

Acknowledgments We are grateful to Alexander Bird and Joseph LaPorte for helpful comments and discussions. We acknowledge the support of the Ministerio de Ciencia e Innovación and European Union research projects FFI2015-​70707-​P (Localism and globalism in logic and semantics) and PID2019-​107667GB-​ I00 (Worlds and truth values: challenges to formal semantics).

Notes 1 This mixture of semantic and metaphysical considerations has been criticized by several authors. See, for instance, Salmon (1981) and Almog (1986). 2 The term ‘overgeneralization’ is due to Dan López de Sa (2008). 3 The idea was also suggested by Cook (1980) and Devitt and Sterelny (1999). 4 It is unclear whether color predicates, which according to Kripke are also rigid, are rigid under the essential characterization. On the one hand, they seem not to be: my house is yellow, but it could be brown. On the other hand, some might argue that the paint that colors the walls of my house has chemical properties that make it be the color it is, and therefore color terms would turn out to be rigid, as Kripke intended. 5 See also Jandrić (in press) for discussion. 6 In the public lecture ‘Trope Determination and Contingent Characterization’, delivered at the Bellingham Summer Philosophy Conference (August 2008). 7 Essences of kinds can be conceived as properties that objects or samples have to have in order to be members of the kind. They are definitory of the kind and they are necessary conditions for belonging to it. Those essential properties should be distinguished from essential properties of individuals or of samples. It is usually taken for granted that the definitory essences of a kind are also essential properties of the individual members or samples of the kind. This may be a plausible assumption, but it is a substantial metaphysical assumption. The conclusion that a property P is essential to an individual member or sample of kind K does not follow from the premise that P defines, or is an essence of, kind K, for the same reason that  ∀x (Kx → Px) does not entail ∀x (Kx →  Px). Some authors (Schwartz 2009; Bird 2018) have even argued that there are cases in nature where an individual changes their kind. That is an interesting metaphysical discussion that we will not address here. Our point is that assuming essences of kinds does not justify automatically the attribution of essential properties to the kind members. And the point stands independently of whether there are such cases in nature. We thank S. Marc Cohen for discussion of this issue. 8 See Soames (2004) for a discussion of this objection and others. 9 For more discussion, and other examples, see Martí and Martínez-​Fernández (2010, section 4). 10 A pioneering work on the formal representation of the denotationalist conception of general terms can be found in Mondadori (1978) and Linsky (1984). It is discussed in Schwartz (1980).

292

Rigidity of General Terms 11 There is also a rigidified description ‘the actual color of the sky’ that designates rigidly the color blue. In the discussion of the problem of trivialization, the relevant rigid reading is not this one, but the one introduced in the text. 12 The example is used by Salmon (2003). 13 Robert May (unpublished comments) has argued that ‘the color of the sky’ is a singular term even in the first premise. Salmon responds to May in Salmon (2005). 14 See Zerbudis (2013) for discussion of the structure and interpretation of Salmon’s argument. 15 See Schwartz (2002, especially section 3), for a discussion of allegedly unnatural or abstruse kinds, such as BMURASP. 16 Martí and Martínez-​Fernández (2010) contains the informal presentation of the solution and Martí and Martínez-​Fernández (2011) develops the formal apparatus. Haraldsen (2018) has criticized our proposal, arguing that it really does not provide a solution to the trivialization problem. 17 Note that the trivialization problem is different from the overgeneralization problem, although some authors conflate the two. The solution to one of them does not constitute a solution to the other. 18 Kripke himself suggests that terms such as ‘foolish’ or ‘fat’ have complex meanings that express necessary and sufficient conditions and hence that they are not rigid designators, unlike ‘cow’ or ‘tiger’ (Kripke 1980, 127–​128). See Macbeth (1995, 264 and ff.) for a formulation and discussion of the assumption. 19 See Putnam (1975, 242–​245) and, more recently, LaPorte (2000, 298–​299), López de Sa (2008), Inan (2008, 219–​220), and Martí and Martínez-​Fernández (2010, 51–​56). 20 See LaPorte in this volume for a discussion of the status of theoretical identifications. See also LaPorte (2010, 2013). Bird (2010) criticizes some of LaPorte’s theses. 21 As regards sentences such as ‘cats are felines’ or ‘gold is a metal’, the denotationalist view should analogously interpret them in terms of the necessary subsumption of one kind under the other rather than as quantified conditionals. 22 For another discussion of the issues that the extension of rigidity to general terms raises, see Fernández Moreno (2016, section 4.2). 23 Wikforss (2010) and Wikforss and Häggqvist (this volume) also think that natural kind terms are not semantically different from other general terms. In their view, they are all descriptional. For a Fregean approach to natural kind terms, see Noonan (2010). See also Briggs and Dosanjh (this volume) for a contrasting view inspired in Kripke’s and Putnam’s externalism. 24 Interestingly, Nimtz finds inspiration in Putnam (1975). A similar view is also presented by Haukioja (2012).

References Ahmed, Arif (2009): ‘Rigidity and essentiality: reply to Gómez-​Torrente’. Mind, 118 (469): 121–​133. Almog, Joseph (1986): ‘Naming without necessity’. Journal of Philosophy, 83 (4): 210–​242. Bird, Alexander (2010): ‘Discovering the essences of natural kinds’. In Beebee, Helen and Sabbarton-​Leary, Nigel (eds.) The Semantics and Metaphysics of Natural Kinds. New York: Routledge: 125–​136. Bird, Alexander (2018): ‘The metaphysics of natural kinds’. Synthese, 195: 1397–​1426. Cook, Monte (1980): ‘If ‘cat’ is a rigid designator, what does it designate?’. Philosophical Studies, 37: 61–​64. Devitt, Michael (2005): ‘Rigid application’. Philosophical Studies, 125 (2): 139–​165. Devitt, Michael and Sterelny, Kim: (1999) Language and Reality. Cambridge: Cambridge University Press. Fernández Moreno, Luís (2016): The Reference of Natural Kind Terms. Frankfurt: Peter Lang. Gómez-​Torrente, Mario (2004):  ‘Beyond rigidity? Essentialist predication and the rigidity of general terms’. Crítica, 36 (108): 37–​54. Gómez-​Torrente, Mario (2006): ‘Rigidity and essentiality’. Mind, 115 (458): 227–​259. Gómez-​ Torrente, Mario (2009):  ‘Essentiality and theoretical identifications:  reply to Ahmed’. Mind, 118 (469): 135–​148. Haraldsen, Fredrik (2018): ‘Rigidity and triviality’. Synthese 195 (5): 1993–​1999. Haukioja, Jussi (2012): ‘Rigidity and actuality-​dependence’. Philosophical Studies 157: 399–​410. Inan, Ilhan (2008): ‘Rigid general terms and essential predicates’. Philosophical Studies, 140 (2), 213–​228. Jandrić, Andrej (in press): ‘Rigid application, theoretical identifications and metaphysical necessity’. In Borstner, Bojan and Todorovic, Tadej (eds.) The Many Faces of the Philosophy of Michael Devitt. Berlin: De Gruyter. Kripke, Saul (1980): Naming and Necessity. Cambridge, MA: Harvard University Press. LaPorte, Joseph (2000): ‘Rigidity and kind’. Philosophical Studies, 97: 293–​316. LaPorte, Joseph (2010): ‘Theoretical identity statements, their truth, and their discovery’. In Beebee, Helen and Sabbarton-​Leary, Nigel (eds.) The Semantics and Metaphysics of Natural Kinds. New York: Routledge: 104–​124. LaPorte, Joseph (2013): Rigid Designation and Theoretical Identities. Oxford: Oxford University Press. Linsky, Bernard (1984): ‘General terms as designators’. Pacific Philosophical Quarterly, 65: 259–​276.

293

Genoveva Martí and José Martínez-Fernández Linsky, Bernard (2006): ‘General terms as rigid designators’. Philosophical Studies, 128: 655–​667. López de Sa, Dan (2006): ‘Flexible property designators’. Grazer Philosophische Studien, 73 (1): 221–​230. López de Sa, Dan (2008):  ‘The over-​generalization problem:  predicates rigidly signifying the “unnatural”’. Synthese, 163 (2): 263–​272. Macbeth, Danielle (1995): ‘Names, natural kind terms, and rigid designation’. Philosophical Studies, 79: 259–​281. Martí, Genoveva and Martínez-​Fernández, José (2010):  ‘General terms as designators:  a defense of the view’. In Beebee, Helen and Sabbarton-​Leary, Nigel (eds.) The Semantics and Metaphysics of Natural Kinds. New York: Routledge: 46–​63. Martí, Genoveva and Martínez-Fernandez, José (2011): ‘General terms, rigidity and the trivialization problem’. Synthese, 181: 277–​293. Mondadori, Fabrizio (1978): ‘Interpreting modal semantics’. In F. Guenthner and C. Rohrer (eds.) Studies in Formal Semantics, Amsterdam: North Holland Publishing: 13–​40. Nimtz, Christian (2019):  ‘Kripkean meta-​semantics and generalized rigidity’. The Philosophical Quarterly, 69 (275): 332–​353. Noonan, Harold (2010):  ‘The commonalities between proper names and natural kind terms:  a Fregean perspective’. In Beebee, Helen and Sabbarton-​Leary, Nigel (eds.) The Semantics and Metaphysics of Natural Kinds. New York: Routledge: 84–​103. Putnam, Hilary (1975):  ‘The meaning of “meaning”’. In Mind, Language and Reality, volume 2. Cambridge: Cambridge University Press: 215–​271. Salmon, Nathan U. (1981): Reference and Essence. Amherst, NY: Prometheus Books. Salmon, Nathan (2003): ‘Naming, necessity, and beyond’. Mind, 112: 475–​492. Salmon, Nathan (2005): ‘Are general terms rigid?’ Linguistics and Philosophy, 28: 117–​134. Schwartz, Stephen (1980): ‘Formal semantics and natural kind terms’, Philosophical Studies, 38, 189–​198. Schwartz, Stephen (2002):  ‘Kinds, general terms, and rigidity:  a reply to LaPorte’. Philosophical Studies, 109: 265–​277. Schwartz, Stephen (2009): ‘The essence of essence’, Australasian Journal of Philosophy, 87 (4): 609–​623. Schwartz, Stephen (2018): ‘Against rigidity for natural kind terms’. Synthese. https://​doi.org/​10.1007/​s11229-​ 018-​1857-​x (last accessed August 30, 2020). Soames, Scott (2002): Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. Oxford: Oxford University Press. Soames, Scott (2004): ‘Reply to Ezcurdia and Gómez-​Torrente’. Crítica, 36 (108): 83–​114. Soames, Scott (2006): ‘Response to critics’. Philosophical Studies, 128: 711–​738. Wikforss, Åsa (2010): ‘Are natural kind terms special?’. In Beebee, Helen and Sabbarton-​Leary, Nigel (eds.) The Semantics and Metaphysics of Natural Kinds. New York: Routledge: 64–​83. Zerbudis, Ezequiel (2013): ‘Second order descriptions and general term rigidity’. Crítica, 45 (135): 3–​27.

294

22 THE PSYCHOLOGY OF NATURAL KIND TERMS Emily Foster-​Hanson and Marjorie Rhodes

When he was three years old, the child of one of the authors came running into the living room to proclaim to her, “I know the dolphins in daddy’s closet are really mammal-​dolphins, like you said, and they breathe outside of water. But they look like fish-​dolphins, so I’m going to pretend that’s what they are”. There were no dolphins in the closet—​not real ones, not toys, not pictures. They existed only in his imagination. But even in his imagination, this three-​year-​old child did not treat the categories of fish, mammal, or dolphin as arbitrary. His words signaled that he viewed these categories as real, objective facts about the world. He accepted that whether a dolphin was a fish or mammal was not a matter of his opinion (or something he could figure out by relying on his own perception), but a matter decided by experts. His beliefs were so firm about this that he couldn’t think otherwise—​even in his imagination—​without signaling that he knew he was doing so “only for pretend”. In this example, the child used the words,“dolphin”,“fish”, and “mammal” as natural kind terms—​ words that speakers and listeners understand as referring to categories that they view as natural kinds. Not all category labels are natural kind terms because not all categories are natural kinds. Some are groupings that people agree are a matter of choice. For example, people can disagree over whether a hotdog is a sandwich, a beanbag is a chair, or whether sugar-​free cookies belong in the category of diet foods (Gelman and Coley 1991; Malt 1990). Natural kind terms, in contrast, refer to those categories that we believe “carve nature at its joints” (Plato 380 B.C./​1974; Mill 1884; e.g., Kripke 1972; Putnam 1973, 1975; Quine 1977; Schwartz 1979; Wilson 1999). From this perspective, grouping things into these kinds of categories is a matter of discovering these joints, not creating them. These intuitions are often wrong (Gelman and Rhodes 2012; Mayr 1982; Leslie 2013; Shtulman 2017), but their implications for psychology, linguistics, and conceptual development are profound. In this chapter, we review what it means to think of a category as a natural kind, how natural kind terms get their status in psychology and language, and how the language of natural kinds shapes cognition, development, and behavior.

Categories as Natural Kinds The prototypical example of a category that people think of as a natural kind is a basic level species (e.g., tiger, rose, bee). In contrast to other types of groupings (e.g., ad-​hoc categories, goal-​directed groups, and even categories of human-​made artifacts), people’s conceptual representations of natural 295

Emily Foster-Hanson and Marjorie Rhodes

kinds are distinctive in several ways (Gelman and Coley 1990). As illustrated in the above example, people think that the boundaries of these categories are set by nature, and therefore are an objective matter of fact, not up to individual opinion. Following what Putnam described as the division of linguistic labor (1973), people think it is reasonable to say, “according to experts, that’s a planet”, but not “according to experts, that’s a toy”, because we rely on experts to figure out the status of celestial bodies (whose status we view as set by nature and discovered by humans) but not to figure out the status of a toy (which could be whatever a child chooses to play with; Malt 1990). Similarly, people think that there are many flexible and subjective ways to classify human-​made artifacts but only one correct, objective (natural) way to classify animals (Kalish 1998a; Rhodes and Gelman 2009a). People also think that natural kinds have absolute boundaries:  Although natural kind categories have a graded structure, and some category members are viewed as more representative and generalizable than others (Rosch and Mervis 1975), even an unusual bird (e.g., a penguin) is still fully a bird, while an unusual piece of furniture (e.g., a beanbag) can be “sort of ” a chair (Diesendruck and Gelman 1999; Estes 2003, 2004; Kalish 2002; Rhodes and Gelman 2009b). As another consequence of thinking of natural kinds as carving nature at its joints, people think of these categories as capturing fundamental patterns of similarities and differences—​that members of the same kind are deeply similar to each other and different from others in many observable and unobservable ways. Thus, people readily generalize the features of individual members of natural kinds to other members of their categories, more than they do for other types of groupings (Gelman 1988; Gelman and O’Reilly 1988; Rips 1975). Finally, people expect membership in natural kind categories (e.g., whether an individual animal is a tiger) to be stable, even across superficial transformations (e.g., if someone paints over its stripes; Gelman and Wellman 1991; Keil 1989) or changes in environment (e.g., if it grows up in a community of peaceful sheep; Gelman and Wellman 1991; Waxman, Medin, and Ross 2007). It feels like we discover natural kinds in the world, and that the reason we think of some categories this way (e.g., animal species) more than others (e.g., human-​made artifacts) is because of the objective structure of nature—​that the world of animals simply is composed of discrete and homogeneous natural kinds while the world of artifacts is not. But this is not the case.The distinction between natural kinds and other types of categories is psychological. This is illustrated by two key considerations. First, whether we construe a particular category as a natural kind is not entirely dependent on its ontological status. It might feel like the objective structure of the world dictates that there is an obvious answer to whether Pluto is a planet but reasonable disagreement over whether a hotdog is a sandwich, but people’s intuitions about which kind of category is which originate in human culture. People in most (but not all; Astuti 1995) cultures studied to date think of basic level species categories as natural kinds—​as having objective, discrete boundaries and picking out coherent, homogeneous categories (Atran 1990; Taverna, Medin, and Waxman 2016). But even in the biological domain, beliefs about the status of particular categories varies by experience (Medin, Ross, Atran, Cox, Coley, Proffitt, and Blok 2006; Medin, Waxman, Woodring, and Washinawatok 2010; Ross, Medin, Coley, and Atran 2003) and expertise (Bailenson, Shum, Atran, Medin, and Coley 2002; Burnett, Medin, Ross, and Block 2005; Lynch, Coley, and Medin 2000; Medin, Ross, Atran, Burnett, and Blok 2002). For example, most diners don’t realize that when they order “grouper” from a restaurant menu, they could be served any of what the US Food and Drug Administration considers to be sixty-​four different species of fish (Lowell, Mustain, Ortenzi, and Warner 2015). To be clear, this view of natural kind concepts is separable from arguments about whether there really are any objective categories. Our point is more limited—​that natural kind concepts do not necessarily reflect objective truth, despite people’s strong intuitions that they do (Boyd, this volume; Haslanger 1995). While natural kinds may exist in the world, natural kind concepts exist only in people’s minds. Second, our intuitions about natural kinds are incompatible with modern views of the structure of the world. Even basic level animal species—​the prototypical example of natural kinds—​do not have the type of absolute, objective, and stable boundaries that people usually think they do, nor do 296

The Psychology of Natural Kind Terms

they reliably pick-​out categories that are highly homogeneous and distinct from other kinds of things (Gelman 2003; Leslie 2013; see also Kalish 1998a). For example, despite people’s intuitions that animal species are stable across time and have immutable boundaries, different species like polar bears and grizzly bears sometimes breed in nature, resulting in hybrids that are “sort of ” members of both kinds (Mallet 2008).While sometimes our intuitions about animal species will fit with reality as we experience it (e.g., an animal will still be a ferocious tiger if we paint over its stripes), these intuitions are cognitive shortcuts and are misleading in a broader sense (e.g., species boundaries change over evolutionary time: Gelman and Rhodes 2012). Even if people revise their intuitions in light of empirical evidence, natural kind concepts reflect a psychological, not a metaphysical, reality. In this way, they are a very important type of psychological concept, but can be a confusing and misleading construct in any theory of biology or metaphysics. Variability in natural kind beliefs is particularly striking in how people represent different social divisions. For instance, some people represent race as a natural kind—​and think that race is a fundamental and objective way of categorizing people—​whereas others (accurately) understand it as a social construction (Rhodes and Gelman 2009a). How people view race and ethnicity varies by their own political views (Mandalaywala, Amodio, and Rhodes 2018; and for children, those of their parents; Rhodes and Gelman 2009a), the diversity of their local environment (Deeb, Segall, Birnbaum, Ben-​Eliyahu, and Diesendruk 2011; Mandalaywayla, Ranger-​Murdock, Amodio, and Rhodes 2019), and their own membership in a majority or minority group (Kinzler and Dautel 2012; Mandalaywala, Amodio, and Rhodes 2018; Roberts and Gelman 2016). Similarly, whether people think of religion as marking natural kinds depends on their own religious upbringing (Chalik, Leslie, and Rhodes 2017; Diesendruck and Haber 2009) and whether religion is associated with social and political conflict in their environment (Diesendruck, Goldfein-​Elbaz, Rhodes, Gelman, and Neumark 2013; Smyth, Feeney, Eidson, and Coley 2017). Even gender, which people often think of as the prototypical example of a natural kind division in the social world (Rothbart and Taylor 1992; Prentice and Miller 2007; Taylor 1996; Taylor, Rhodes, and Gelman 2009), is understood by some to be socially constructed and arbitrary (Fast and Olson 2018; Rhodes and Gelman 2009a). Representations of social categories illustrate several key things of natural kind beliefs:  they are quite often inaccurate (in many—​if not all—​of these cases, a natural kind representation does not accurately reflect the objective structure of the social world), but they are also pervasive, variable, and potentially highly problematic (as will be discussed in more detail below).

How Natural Kinds Get Their Status in Psychology and Language Given that natural kind concepts are psychologically created, there are two related questions about how they get their status in psychology and language. First: where does the tendency to think of some categories as marking natural kinds come from? That is, if natural kind representations are inaccurate, why do people have them? Second: why do people come to represent some particular categories (and not others) in this way?

Where Do Natural Kind Representations Come From? One might think that natural kind beliefs arise as a product of formal science instruction or other aspects of Western education. It seems natural to assume that learning taxonomic systems of categorization, about DNA, and so on, could lead to the development of these beliefs. For example, learning about how genes are passed from parents to offspring provides an intuitive framework through which people might think of a wide range of properties—​from skin color to musical talent—​as heritable and fixed. Indeed, many biology curricula reinforce thinking of the biological world as composed of natural kinds, often to the detriment of students’ conceptual understanding of the processes that underlie evolution, the consequences of climate change, and other important scientific concepts (Bean, Sinatra, 297

Emily Foster-Hanson and Marjorie Rhodes

and Schrader 2010; Donovan 2014; Short and Hawley 2015; Shtulman and Calabi 2013; Shtulman 2017; Swiney, Bates, and Coley 2018). But, although some curricula do (perhaps inadvertently) reinforce natural kind representations, formal education does not create them. Natural kind representations of the biological world develop even in cultures with diverse educational practices and even with little exposure to formal science instruction at all (Atran 1990; Taverna, Medin, and Waxman 2016). Further, these beliefs develop in children prior to the onset of formal schooling (Gelman and Kremer 1991; Gelman and Wellman 1991;Waxman, Medin, and Ross 2007). For example, before the age of six, children think that people might disagree over how to categorize human-​made artifacts but that there is only one objectively accurate way to classify animals (Diesendruck, Goldfein-​Elbaz, Rhodes, Gelman, and Neumark 2013; Rhodes and Gelman 2009a; Rhodes, Gelman, and Karuza 2014). Young children are also certain than a penguin is either 100 percent a bird or not a bird at all (even if they aren’t sure which) but agree that sunglasses are “sort of ” clothing, for example. Finally, young children expect babies born to tiger parents to grow up to be ferocious tigers no matter what they look like or how they are raised (Gelman and Wellman 1991; Keil 1989), and they expect members of basic level species categories to share fundamental properties, even if they look different from one another (Gelman and Markman 1986; Gelman 1988). Young children across diverse cultural contexts hold these natural kind beliefs not only about animal species, but also about human gender categories (Rhodes and Gelman 2009a) and other social divisions that are salient in their environment (Diesendruck, Goldfein-​Elbaz, Rhodes, Gelman, and Neumark 2013; Diesendruck and Haber 2009; Mandalaywala et  al. 2019; Smyth, Feeney, Eidson, and Coley 2017). For example, children think it is objectively accurate to categorize people as boys or girls (Rhodes and Gelman 2009; Diesendruck, Goldfein-​Elbaz, Rhodes, Gelman, and Neumark 2013), and they think that girls have many features in common with one another despite differences in appearance, personality, or preferences (Berndt and Heller 1986; Biernat 1991; Gelman, Collman, and Maccoby 1986; Taylor, Rhodes, and Gelman 2009).Young children sometimes hold these beliefs even in communities where adults have more flexible beliefs (Astuti, Gregg, Solomon, Carey, Ingold, and Miller 2004; Rhodes and Gelman 2009a). That natural kind representations emerge so early in children’s development suggests that they are the product of fundamental and basic conceptual biases, rather than the result of direct instruction (Gelman 2003; Rhodes and Mandalaywala 2017). Children rely on basic conceptual and explanatory biases to actively construct their understanding of the world (Wellman and Gelman 1992; Gopnik and Wellman 2012). In the course of this process, they use basic perceptual and conceptual processes to categorize entities into distinct groups (e.g., to distinguish tigers from wolves) and basic explanatory biases to make sense of the patterns they observe (e.g., after observing that tigers have stripes and wolves do not, they could assume that there is something intrinsic to tigers and wolves that explains these differences; Cimpian and Steinberg 2014; Cimpian and Salomon 2014).The process of observing and explaining clusters of properties (e.g., having stripes, being orange, etc.) can also lead children to develop higher-​order beliefs about the structure of the category—​to assume that because tigers and wolves differ in several obvious ways, they probably differ in other ways not yet discovered as well, and that these differences are generally caused by intrinsic mechanisms (even if it isn’t yet clear what those are). Finally, children can then develop beliefs at still higher levels of abstraction—​wherein after learning about tigers, wolves, birds, monkeys, and so on, they then assume that the biological domain—​in general—​is composed of natural kinds. This abstract framework theory about the structure of the domain can then guide how they represent and think about new species categories as they encounter them. Indeed, by age four, children hold domain-​specific hypotheses about the structure of categories: They expect animal categories, in general, to be natural kinds (Gelman and Markman 1986; Keil 1989), but they recognize that social categories have more variable structure, and not all groupings of people reflect deep homogeneity across members (Diesendruck, Goldfein-​Elbaz, Rhodes, Gelman, 298

The Psychology of Natural Kind Terms

and Neumark 2013; Rhodes and Gelman 2009; Kalish and Lawson 2008). Importantly, natural kind beliefs describe a set of inter-​related beliefs about the structure of categories (e.g., homogeneity, stability, fixed boundaries, and so on) which can operate in different ways across categories and domains (Noyes and Dunham 2019; Rhodes and Mandalaywala 2017).

Why Do People Come to Represent Any Particular Category as a Natural Kind? Although children do not need to be explicitly taught that a particularly category is a natural kind, the processes by which they identify particular categories (or domains of categories) as natural kinds are highly sensitive to testimony from others and embedded in culture (Butler and Markman 2012; Butler and Tomasello 2016; Callanan 1990; Gelman 2009; Gelman,Ware, Manczak, and Graham 2013; Csibra and Gergely 2009; Harris and Koenig 2006). Just as adults defer to experts to know the “true” status of natural kind categories (Malt 1990), children look to the experts in their communities (i.e., adults) to learn how to divide up the world into meaningful units of experience (Danovitch and Keil 2004; Gelman and Markman 1986; Jaswal 2010; Jaswal, Lima, and Small 2009; Noyes and Keil 2017; VanderBorght and Jaswal 2009; Wilson and Keil 1998). Language plays a powerful role in shaping these processes (Vygotsky 1962; Quine 1977). For example, labels mark categories as distinct kinds of things (Kripke 1972; Putnam 1973, 1975). Hearing noun labels helps even three-​month-​old infants group objects into categories (Fulkerson and Waxman 2007; Waxman and Hall 1993; Xu 2002). For example, young babies who see perceptually dissimilar members of a category (e.g., fish), categorize them into the same group if they are presented with the same label (e.g.,“Look at this toma!”), but not otherwise (Ferry, Hespos, and Waxman 2010). Labels facilitate categorization because they are reliable cues of the speaker’s intent to communicate that individuals are similar in some meaningful way (Ferguson and Lew-​Williams 2016).While labels facilitate categorization, many labeled categories (e.g., toys, hammers, crackers) are not represented as natural kinds. Thus, simply being used as a category label does not turn a noun into a natural kind term. Using labels to provide generic descriptions of categories, however, may begin to do so (Carlson 1977; Gelman 2003; Leslie 2008). Generic statements (e.g.,“birds lay eggs”,“girls wear pink”) describe the properties of abstract categories, in contrast with quantified statements about specific category members (e.g.,“these birds lay eggs”,“Anna wears pink”; Carlson and Pelletier 1995). By at least thirty months, children understand generics to describe abstract categories, instead of particular examples. For example, upon looking at a picture of two penguins, two-​year-​old children respond “no” when they are asked, “Do these birds fly?” but “yes” (when looking at the same picture), when they are asked “Do birds fly?” (Gelman and Raman 2003; Graham, Nayer, and Gelman 2011; Hollander, Gelman, and Star 2002; Rhodes, Leslie, Bianchi, and Chalik 2017). Thus, by age two, children can track the difference between generic, abstract references to categories and references to particular members. By age three (and perhaps earlier), children also recognize generics as describing generalizable information about what category members are usually like, and they expect that the properties described by generics are common across category members (Cimpian, Brandone, and Gelman 2010; Leslie, Khemlani, and Glucksberg 2011), even when this is not always the case (for instance, the statement, “mosquitos carry the west Nile virus” is true of less than 1 percent of mosquitos, yet it can still be expressed generically). Thus, within the first few years of life, children understand that generic statements communicate generalizable information about abstract categories. Generics communicate information about shared properties, which can trigger the processes of developing higher-​order beliefs (e.g., that tigers have a lot in common in general), as well as explanatory assumptions that these patterns probably reflect intrinsic causes, as described above. Critically, children also defer to the expertise of adult speakers to use generics to signal that the category of interest is the type of category that carves nature at its joints. When an adult says,“Tigers have stripes”,“Tigers are ferocious”,“Tigers run very fast”, children learn not only about three features of tigers, they also learn that a knowledgeable adult in their community 299

Emily Foster-Hanson and Marjorie Rhodes

thinks that tigers is a meaningful way of dividing up the natural world.Thus, via multiple mechanisms, generic language contributes to the development of natural kind beliefs. Once a particular category label is described with generics, it can begin to take on the status of a natural kind term. For generics to play this role in the creation of natural kind beliefs and natural kind terms, and to account for the culturally variable spread of natural kind beliefs about particular groupings, then two things must be true. First, hearing categories described with generics must trigger the formation of natural kind beliefs in people (including in young children), and second, knowledgeable speakers (including teachers and parents) must produce such generics for categories for which they themselves hold these representations. If so, generic language would serve as a vehicle by which natural kind beliefs about particular categories spread across communities. There is indeed empirical support for both of these processes. First, generic language leads labels to take on the status of natural kind terms (in other words, hearing them elicits natural kind beliefs about referenced categories). Even hearing just a small number of generic statements about new categories of animals or people leads children and adults to view them as more homogeneous, stable, and causally powerful than they would otherwise (Gelman,Ware, and Kleinberg 2010; Rhodes, Leslie, and Tworek 2012). For example, hearing a series of generic statements about a new category (e.g., “Zarpies love to eat flowers”, “Zarpies flap their arms when they’re happy”, and so on) leads people to think that Zarpies represents a meaningful kind and that individual Zarpies will share many other features that were not mentioned. Indeed, even two-​year-​old children form new social categories (i.e., they can identify members of the category Zarpies) when they hear both noun labels and generics (e.g., “Zarpies eat flowers”), but not when they hear noun labels alone (e.g., “This Zarpie eats flowers”; Rhodes, Leslie, Bianchi, and Chalik 2017; Rhodes, Leslie, and Tworek 2012). Second, the production of generic language reflects speakers’ beliefs. Speakers use more generics when talking about categories they view as natural kinds (e.g., animal kinds, gender categories) than about categories they do not view in this way (e.g., artifacts; Brandone and Gelman 2009; Gelman, Goetz, Sarnecka, and Flukes 2008; Gelman, Taylor, and Nguyen 2004; Goldin-​Meadow, Gelman, and Mylander 2005; Pappas and Gelman 1998; Segall, Birnbaum, Deeb, and Diesendruck 2015). Inducing beliefs that a category is a natural kind even leads people to describe it using more generics (Rhodes, Leslie, and Tworek 2012). And this type of input is ubiquitous:  generics are frequent in parent-​child conversations (Gelman, Goetz, Sarnecka, and Flukes 2008; Graham, Nayer, and Gelman 2011) and—​like natural kind concepts—​across languages (Pirahã:  Everett 2009; Spanish:  Gelman, Sánchez Tapia, and Leslie 2016; Quechua:  Mannheim, Gelman, Escalante, Huayhua, and Puma 2010; Mandarin: Tardif, Gelman, Fu, and Zhu 2012). Thus, children hear knowledgeable speakers use generics to communicate their natural kind beliefs in their daily lives, and hearing categories described with generics triggers the formation of natural kind beliefs. In this way, generic language serves as a powerful tool for communicating natural kind beliefs about particular categories across communities. As described above, one way that generic language shapes beliefs that a particular category is meaningful and distinct is by communicating that a knowledgeable speaker thinks it is. In order to understand what people mean when they talk, listeners engage in complex calculations about a speaker’s knowledge state and the set of relevant alternatives—​the things the speaker could have said but chose not to (Clark 1996; Grice 1975; Horn 1984). To the extent that a set of other possibilities is salient in the communicative context (e.g., a knowledgeable speaker clearly could have generalized about children, but instead chose to generalize only about girls), the choice to use a generic to talk about a particular grouping is in itself informative. In support of this account, children infer that categories that a knowledgeable speaker describes with generics denote natural kinds, even if the particular properties the generics assert about the category later turn out to be wrong (Foster-​Hanson, Leslie, and Rhodes 2019; see also Moty and Rhodes 2019).

300

The Psychology of Natural Kind Terms

How the Language of Natural Kinds Shapes Cognition, Development, and Behavior Once people come to think of a particular category as a natural kind, these beliefs have a range of consequences for how they use that category in their daily lives. As described above, people use natural kinds to generalize information to new kind members (Gelman 1988; Gelman and O’Reilly 1988; Rips 1975), to infer fixed category boundaries (Diesendruck and Gelman 1999; Estes 2003, 2004; Kalish 2002; Rhodes and Gelman 2009b), and to reason about heritability (Gelman and Wellman 1991;Waxman, Medin, and Ross 2007).While these inferences provide useful cognitive shortcuts and can simplify learning, people’s inferences about natural kinds can also be problematic. For example, beliefs that animal kinds are homogenous and discrete are at odds with modern views of evolutionary change (Gelman and Rhodes 2012; Mayr 1982) and can lead people to ignore category variability (Emmons and Kelemen 2015; Shtulman and Shulz 2008). Thinking of animal categories as stable can even impede people’s ability to understand processes like metamorphosis (French, Menendez, Herrmann, Evans, and Rosengren 2018). Thinking of human social categories (e.g., gender, race, or religion) as objectively true can be particularly troublesome, as such groupings are often determined by history and cultural convention, rather than the objective structure of the world (Hirschfeld 1996). For instance, thinking of race as a natural kind is associated with more prejudice towards Black people and more endorsement of existing social hierarchies (Mandalaywala, Amodio, and Rhodes 2018; Williams and Eberhardt 2008). Adults’ beliefs that gender categories reflect the objective structure of the world are also correlated with more endorsement of gender stereotypes (Brescoll and LaFrance 2004; Brescoll, Uhlmann, and Newman 2013) and more support for boundary-​enhancing initiatives like laws mandating that transgender people use restrooms corresponding to their biological sex (Roberts, Ho, Rhodes, and Gelman 2017). Inducing essentialist beliefs about novel social categories even leads four-​to six-​year-​old children to share fewer resources with them (Rhodes, Leslie, Saunders, Dunham, and Cimpian 2017). Describing behaviors that children are familiar with using category labels (e.g., “helpers”, “drawers”) can even shape their attitudes and behavior (Cimpian, Arce, Markman, and Dweck 2007; Foster-​Hanson, Cimpian, Leshin, and Rhodes 2018; see also Gelman and Heyman 1999). Finally, hearing categories described using generics leads young children to infer not only what category members usually are like, but also what they should be like, and to say it’s wrong to do things differently from the group (Roberts, Ho, and Gelman 2017).

Other Interpretations of Generics While generics can communicate natural kind beliefs, they communicate other information and are open to alternate interpretations as well. For instance, generics also communicate information about the prevalence of particular properties (Cimpian, Brandone, and Gelman 2010), as people assume that the properties described by generics are more prevalent for the referenced category than for other similar categories (Tessler and Goodman 2019). More broadly, people’s concepts are situated within causal frameworks about the structure of the world (Murphy and Medin 1985).Thus, people interpret generic claims about categories in the context of their previous expectations about the structure of the category and the domain. In some cases, for example, people might have expectations that category regularities are not caused by intrinsic mechanisms, but instead result from extrinsic or structural forces (Vasilyeva, Gopnik, and Lombrozo 2018). For example, some people might assume that “girls wear pink” because girls naturally love pink (an intrinsic cause) whereas others might assume that this is because stores mostly sell pink clothes for girls (a structural cause). When people have these structure-​based beliefs, generic descriptions are unlikely to elicit natural kind representations.

301

Emily Foster-Hanson and Marjorie Rhodes

Thus, whether generics lead people to think of category labels as natural kind terms does not depend on the language alone, but on how it interacts with their developing conceptual frameworks. As described above, a critical component of these frameworks is whether people think that category regularities are explainable by intrinsic or structural causes. While children can understand structural explanations when they are given (Vasilyeva, Gopnik, and Lombrozo 2018), they are harder for children to generate on their own. For example, structural causes are often not readily observable, and understanding them relies on relational reasoning (Gentner 1983, 2005; Richland, Morrison, and Holyoak 2006) and thinking counterfactually about how things could have been (Beck, Robinson, Carroll, and Apperly 2006; Rafetseder, Cristi-​Vargas, and Perner 2010), both capacities that emerge relatively late in development. In contrast, internal causes for category regularities are often easier for people to learn and bring to mind (Cimpian and Salomon 2014; Hussak and Cimpian 2018). A preference for intrinsic causes may even be a basic cognitive bias: Young infants often infer internal causes when no external causes are apparent (Gelman 1990; Premack 1990; Spelke, Phillips, and Woodward 1995; Stewart 1984). Thus, understanding generics as describing the underlying nature of categories appears to be a cognitive bias (Cimpian and Cadena 2010; Cimpian and Markman 2011; Gelman 2004; Leslie 2014; Lyons 1977; Prasada and Dillingham 2009), although possibly one that varies by culture (Carstensen, Zhang, Heyman, Fu, Lee, and Walker 2019; see also Choi, Nisbett, and Norenzayan 1999).

Concluding Thoughts In summary, natural kind terms refer to natural kinds—​categories that people think are objectively true, but that in fact reflect a psychological—​rather than a metaphysical—​reality. In this way, natural kind terms do not directly refer to the true state of the world (whatever it may be), but rather to people’s beliefs about the true state of the world. People communicate their beliefs about which categories are natural kinds through the way they talk, especially by describing categories using generic noun phrases like, “Tigers have stripes”, Gold is a precious metal”, and “Boys just like to get messy”. Once people come to see a category as a natural kind, these beliefs shape their cognition and behavior in a variety of ways, some of which (e.g., stereotyping and prejudice) can be harmful. Given the many ways that natural kind terms shape how people think and act, understanding how they are transmitted across cultures and generations is therefore crucial to the study of human thought.

Acknowledgments Preparation of this chapter was supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under Award Number R01HD087672 (to Rhodes) and F31HD093431 (to Foster-​Hanson). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

References Astuti, R. (1995). People of the Sea:  Identity and Descent among the Vezo of Madagascar. Cambridge:  Cambridge University Press. Astuti, R., Gregg, E., Solomon, A., Carey, S., Ingold, T., and Miller, P. (2004). Constraints on conceptual development:  A case study of the acquisition of folkbiological and folksociological knowledge in Madagascar. Monographs of the Society for Research in Child Development, 69(3), 1–​161. Atran, S. (1990). Cognitive Foundations of Natural History: Towards an Anthropology of Science. New York: Cambridge University Press. Bailenson, J. N., Shum, M. S., Atran, S., Medin, D. L., and Coley, J. D. (2002). A Bird’s eye view:  Biological categorization and reasoning within and across cultures. Cognition, 84(1), 1–​53. https://​doi.org/​10.1016/​ S0010-​0277(02)00011-​2.

302

The Psychology of Natural Kind Terms Bean, T. E., Sinatra, G. M., and Schrader, P. G. (2010). Spore: Spawning evolutionary misconceptions? Journal of Science Education and Technology, 19(5), 409–​414. https://​doi.org/​10.1007/​s10956-​010-​9211-​1. Beck, S. R., Robinson, E. J., Carroll, D. J., and Apperly, I. A. (2006). Children’s thinking about counterfactuals and future hypotheticals as possibilities. Child Development, 77(2), 413–​426. https://​doi.org/​10.1111/​ j.1467-​8624.2006.00879.x. Berndt, T. J., and Heller, K. A. (1986). Gender stereotypes and social inferences: A developmental study. Journal of Personality and Social Psychology, 50(5), 889–​898. https://​doi.org/​10.1037/​0022-​3514.50.5.889. Biernat, M. (1991). Gender stereotypes and the relationship between masculinity and femininity: A developmental analysis. Journal of Personality and Social Psychology, 61(3), 351. https://​ doi.org/​ 10.1037/​ 0022-​3514.61.3.351. Brandone, A. C., and Gelman, S. A. (2009). Differences in preschoolers’ and adults’ use of generics about novel animals and artifacts: A window onto a conceptual divide. Cognition, 110(1), 1–​22. https://​doi.org/​10.1016/​ j.cognition.2008.08.005. Brescoll, V., and LaFrance, M. (2004). The correlates and consequences of newspaper reports of research on sex differences. Psychological Science, 15(8), 515–​520. https://​doi.org/​10.1111/​j.0956-​7976.2004.00712.x. Brescoll,V. L., Uhlmann, E. L., and Newman, G. E. (2013). The effects of system-​justifying motives on endorsement of essentialist explanations for gender differences. Journal of Personality and Social Psychology, 105(6), 891–​908. https://​doi.org/​10.1037/​a0034701. Burnett, R. C., Medin, D. L., Ross, N. O., and Blok, S.V. (2005). Ideal is typical. Canadian Journal of Experimental Psychology/​Revue Canadienne de Psychologie Expérimentale, 59(1), 3–​10. https://​doi.org/​10.1037/​h0087453. Butler, L. P., and Markman, E. M. (2012). Preschoolers use intentional and pedagogical cues to guide inductive inferences and exploration. Child Development, 83, 1416–​1428. https://​doi.org/​10.1111/​ j.1467-​8624.2012.01775.x. Butler, L. P., and Tomasello, M. (2016). Two-​and 3-​year-​old children integrate linguistic and pedagogical cues in guiding inductive generalizations and exploration. Journal of Experimental Child Psychology, 145, 64–​78. https://​doi.org/​10.1016/​j.jecp.2015.12.001. Callanan, M. A. (1990). Parents’ descriptions of objects: Potential data for children’s inferences about category principles. Cognitive Development, 5(1), 101–​122. https://​doi.org/​10.1016/​0885-​2014(90)90015-​L. Carlson, G.  N. (1977). A unified analysis of the English bare plural. Linguistics and Philosophy, 1(3), 413–​457. https://​doi.org/​10.1007/​BF00353456. Carlson, G. N., and Pelletier, F. J. (1995). The Generic Book. Chicago, IL: University of Chicago Press. Carstensen, A., Zhang, J., Heyman, G. D., Fu, G., Lee, K., and Walker, C. M. (2019). Context shapes early diversity in abstract thought. Proceedings of the National Academy of Sciences, 201818365. https://​doi.org/​10.1073/​ pnas.1818365116. Chalik, L., Leslie, S. J., and Rhodes, M. (2017). Cultural context shapes essentialist beliefs about religion. Developmental Psychology, 53(6), 1178–​1187. https://​doi.org/​10.1037/​dev0000301. Choi, I., Nisbett, R. E., and Norenzayan, A. (1999). Causal attribution across cultures: Variation and universality. Psychological Bulletin, 125(1), 47–​63. https://​doi.org/​10.1037/​0033-​2909.125.1.47. Cimpian, A., Arce, H. M. C., Markman, E. M., and Dweck, C. S. (2007). Subtle linguistic cues affect children’s motivation. Psychological Science, 18(4), 314–​316. https://​doi.org/​10.1111/​j.1467-​9280.2007.01896.x. Cimpian, A., Brandone, A. C., and Gelman, S. A. (2010). Generic statements require little evidence for acceptance but have powerful implications. Cognitive Science, 34(8), 1452–​1482. https://​doi.org/​10.1111/​ j.1551-​6709.2010.01126.x. Cimpian, A., and Cadena, C. (2010). Why are dunkels sticky? Preschoolers infer functionality and intentional creation for artifact properties learned from generic language. Cognition, 117(1), 62–​68. https://​doi.org/​ 10.1016/​j.cognition.2010.06.011. Cimpian, A., and Markman, E. M. (2011). The generic/​nongeneric distinction influences how children interpret new information about social others. Child Development, 82(2), 471–​492. https://​doi.org/​10.1111/​ j.1467-​8624.2010.01525.x. Cimpian, A., and Salomon, E. (2014). The inherence heuristic: An intuitive means of making sense of the world, and a potential precursor to psychological essentialism. Behavioral and Brain Sciences, 37(5), 461–​480. https://​ doi.org/​10.1017/​S0140525X13002197. Cimpian, A., and Steinberg, O. D. (2014). The inherence heuristic across development: Systematic differences between children’s and adults’ explanations for everyday facts. Cognitive Psychology, 75, 130–​154. https://​doi. org/​10.1016/​j.cogpsych.2014.09.001. Clark, H. H. (1996). Using Language. New York: Cambridge University Press. Csibra, G., and Gergely, G. (2009). Natural pedagogy. Trends in Cognitive Sciences, 13(4), 148–​153. https://​doi.org/​ 10.1016/​j.tics.2009.01.005. Danovitch, J., and Keil, F., (2004). Should you ask a fisherman or a biologist?: Developmental shifts in ways of clustering knowledge. Child Development, 75, 918–​931. https://​doi.org/​10.1111/​j.1467-​8624.2004.00714.x.

303

Emily Foster-Hanson and Marjorie Rhodes Deeb, I., Segall, G., Birnbaum, D., Ben-​Eliyahu, A., and Diesendruck, G. (2011). Seeing isn’t believing: The effect of intergroup exposure on children’s essentialist beliefs about ethnic categories. Journal of Personality and Social Psychology, 101(6), 1139–​1156. https://​doi.org/​10.1037/​a0026107. Diesendruck, G., Goldfein-​Elbaz, R., Rhodes, M., Gelman, S., and Neumark, N. (2013). Cross-​cultural differences in children’s beliefs about the objectivity of social categories. Child Development, 84(6), 1906–​1917. https://​ doi.org/​10.1111/​cdev.12108. Diesendruck, G., and Haber, L. (2009). God’s categories: The effect of religiosity on children’s teleological and essentialist beliefs about categories. Cognition, 110(1), 100–​114. https://​doi.org/​10.1016/​j.cognition.2008.11.001. Diesendruck, G., and Gelman, S. A. (1999). Domain differences in absolute judgments of category membership: Evidence for an essentialist account of categorization. Psychonomic Bulletin and Review, 6(2), 338–​346. http://​dx.doi.org/​10.3758/​BF03212339. Donovan, B. M. (2014). Playing with fire? The impact of the hidden curriculum in school genetics on essentialist conceptions of race. Journal of Research in Science Teaching, 51(4), 462–​496. https://​doi.org/​10.1002/​tea.21138. Emmons, N. A., and Kelemen, D. A. (2015).Young children’s acceptance of within-​species variation: Implications for essentialism and teaching evolution. Journal of Experimental Child Psychology, 139, 148–​160. http://​dx.doi. org/​10.1016/​j.jecp.2015.05.011. Estes, Z. (2003). Domain differences in the structure of artifactual and natural categories. Memory and Cognition, 31, 199–​214. https://​doi.org/​10.3758/​BF03194379. Estes, Z. (2004). Confidence and gradedness in semantic categorization: Definitely somewhat artifactual, maybe absolutely natural. Psychonomic Bulletin and Review, 11, 1041–​1047. https://​doi.org/​10.3758/​BF03196734. Everett, D.  L. (2009). Pirahã culture and grammar:  A response to some criticisms. Language, 85(2), 405–​442. https://​doi.org/​10.1353/​lan.0.0104. Fast, A. A., and Olson, K. R. (2018). Gender development in transgender preschool children. Child Development, 89(2), 620–​637. https://​doi.org/​10.1111/​cdev.12758. Ferguson, B., and Lew-​Williams, C. (2016). Communicative signals support abstract rule learning by 7-​month-​ old infants. Scientific Reports, 6, 25434. https://​doi.org/​10.1038/​srep25434. Ferry, A. L., Hespos, S. J., and Waxman, S. R. (2010). Categorization in 3-​and 4-​month-​old infants: An advantage of words over tones. Child Development, 81(2), 472–​479. https://​doi.org/​10.1111/​j.1467-​8624.2009.01408.x. Foster-​Hanson, E., Cimpian, A., Leshin, R. A., and Rhodes, M. (2018). Asking children to “be helpers” can backfire after setbacks. Child Development. https://​doi.org/​10.1111/​cdev.13147. Foster-​Hanson, E., Leslie, S., and Rhodes, M. (2019, January 20). Speaking of kinds:  How generic language shapes the development of category representations. https://​doi.org/​10.31234/​osf.io/​28qf7. French, J. A., Menendez, D., Herrmann, P. A., Evans, E. M., and Rosengren, K. S. (2018). Cognitive constraints influence an understanding of life-​cycle change. Journal of Experimental Child Psychology, 173, 205–​221. https://​doi.org/​10.1016/​j.jecp.2018.03.018. Fulkerson, A. L., and Waxman, S. R. (2007).Words (but not tones) facilitate object categorization: Evidence from 6-​ and 12-​month-​olds. Cognition, 105(1), 218–​228. https://​doi.org/​10.1016/​j.cognition.2006.09.005. Gelman, R. (1990). First principles organize attention to and learning about relevant data:  Number and the animate-​ inanimate distinction as examples. Cognitive Science, 14(1), 79–​106. https://​doi.org/​10.1207/​ s15516709cog1401_​5. Gelman, S.  A. (1988). The development of induction within natural kind and artifact categories. Cognitive Psychology, 20(1), 65–​95. https://​doi.org/​10.1016/​0010-​0285(88)90025-​4. Gelman, S. A. (2003). The Essential Child: Origins of Essentialism in Everyday Thought. Oxford Scholarship Online. https://​doi.org/​10.1093/​acprof:oso/​9780195154061.001.0001. Gelman, S. A. (2004). Psychological essentialism in children. Trends in Cognitive Sciences, 8(9), 404–​409. https://​ doi.org/​10.1016/​j.tics.2004.07.001. Gelman, S. A. (2009). Learning from others: Children’s construction of concepts. Annual Review of Psychology, 60, 115–​140. https://​doi.org/​10.1146/​annurev.psych.59.103006.093659. Gelman, S. A., and Coley, J. D. (1990). The importance of knowing a dodo is a bird: Categories and inferences in 2-​year-​old children. Developmental Psychology, 26(5), 796–​804. https://​doi.org/​10.1037/​0012-​1649.26.5.796. Gelman, S. A., Collman, P., and Maccoby, E. E. (1986). Inferring properties from categories versus inferring categories from properties: The case of gender. Child Development, 57(2), 396–​404. www.jstor.org/​stable/​1130595. Gelman, S. A., Goetz, P. J., Sarnecka, B. S., and Flukes, J. (2008). Generic language in parent-​child conversations. Language Learning and Development, 4(1), 1–​31. https://​doi.org/​10.1080/​15475440701542625. Gelman, S. A., and Heyman, G. D. (1999). Carrot-​eaters and creature-​believers: The effects of lexicalization on children’s inferences about social categories. Psychological Science, 10(6), 489–​493. https://​doi.org/​10.1111/​ 1467-​9280.00194. Gelman, S. A., and Kremer, K. E. (1991). Understanding natural cause: Children’s explanations of how objects and their properties originate. Child Development, 62(2), 396–​414. https://​doi.org/​10.1111/​j.1467-​8624.1991. tb01540.x.

304

The Psychology of Natural Kind Terms Gelman, S. A., and Markman, E. M. (1986). Categories and induction in young children. Cognition, 23(3), 183–​ 209. https://​doi.org/​10.1016/​0010-​0277(86)90034-​X. Gelman, S. A., and O’Reilly, A. W. (1988). Children’s inductive inferences within superordinate categories: The role of language and category structure. Child Development, 59(4), 876–​887. https://​doi.org/​10.2307/​ 1130255. Gelman, S. A., and Raman, L. (2003). Preschool children use linguistic form class and pragmatic cues to interpret generics. Child Development, 24, 308–​325. https://​doi.org/​10.1111/​1467-​8624.00537. Gelman, S. A., and Rhodes, M. (2012). Two-​thousand years of stasis. In Rosengren, K. S., Brem, S. K., Evans, E. M., and Sinatra, G. M. (Eds.), Evolution Challenges (pp. 3–​21). Oxford Scholarship Online. https://​doi.org/​ 10.1093/​acprof:oso/​9780199730421.003.0001. Gelman, S. A., Sánchez Tapia, I., and Leslie, S.-​J. (2016). Memory for generic and quantified sentences in Spanish-​speaking children and adults. Journal of Child Language, 43(6), 1231–​1244. https://​doi.org/​10.1017/​ s0305000915000483. Gelman, S. A., Taylor, M. G., and Nguyen, S. P. (2004). Mother-​child conversations about gender: Understanding the acquisition of essentialist beliefs. Monographs of the Society for Research in Child Development, 69(1), 93–​118. https://​doi.org/​10.1111/​j.0037-​976X.2004.00279.x. Gelman, S. A.,Ware, E. A., and Kleinberg, F. (2010). Effects of generic language on category content and structure. Cognitive Psychology, 61(3), 273–​301. https://​doi.org/​10.1016/​j.cogpsych.2010.06.001. Gelman, S. A., Ware, E. A., Manczak, E. M., and Graham, S. A. (2013). Children’s sensitivity to the knowledge expressed in pedagogical and nonpedagogical contexts. Developmental Psychology, 49(3), 491–​504. https://​doi. org/​10.1037/​a0027901. Gelman, S. A., and Wellman, H. M. (1991). Insides and essences:  Early understandings of the non-​obvious. Cognition, 38(3), 213–​244. https://​doi.org/​10.1016/​0010-​0277(91)90007-​Q. Gentner, D. (1983). Structure-​mapping:  A theoretical framework for analogy. Cognitive Science, 7(2), 155–​170. https://​doi.org/​10.1016/​S0364-​0213(83)80009-​3. Gentner, D. (2005). The development of relational category knowledge. In Gershkoff-​Stowe, L. and Rakison, D. H. (Eds.), Building Object Categories in Developmental Time (pp. 263–​294). New York: Psychology Press. Goldin-​Meadow, S., Gelman, S. A., and Mylander, C. (2005). Expressing generic concepts with and without a language model. Cognition, 96(2), 109–​126. https://​doi.org/​10.1016/​j.cognition.2004.07.003. Gopnik, A., and Wellman, H. M. (2012). Reconstructing constructivism:  Causal models, Bayesian learning mechanisms, and the theory theory. Psychological Bulletin, 138(6), 1085–​1108. https://​doi.org/​10.1037/​ a0028044. Graham, S. A., Nayer, S. L., and Gelman, S. A. (2011). Two-​year-​olds use the generic/​nongeneric distinction to guide their inferences about novel kinds. Child Development, 82(2), 493–​507. https://​doi.org/​10.1111/​ j.1467-​8624.2010.01572.x. Grice, H. (1975). Logic and conversation. In Cole, P. and Morgan, J. L. (Eds.), Syntax and Semantics (Vol. 3, pp. 41–​58). New York: Academic Press. Harris, P. L., and Koenig, M. A. (2006). Trust in testimony: How children learn about science and religion. Child Development, 77(3), 505–​524. https://​doi.org/​10.1111/​j.1467-​8624.2006.00886.x. Haslanger, S. (1995). Ontology and social construction. Philosophical Topics, 23(2), 95–​125. https://​doi.org/​ 10.5840/​philtopics19952324. Hirschfeld, L.  A. (1996). Race in the Making:  Cognition, Culture, and the Child’s Construction of Human Kinds. Cambridge, MA: MIT Press. Hollander, M.A., Gelman, S.A., and Star, J. (2002). Children’s interpretation of generic noun phrases. Developmental Psychology, 38(6), 883–​894. https://​doi.org/​10.1037/​0012-​1649.38.6.883. Horn, L. (1984).Toward a new taxonomy for pragmatic inference: Q-​based and r-​based implicature. In Schiffrin, D. (Ed.), Meaning, Form, and Use in Context: Linguistic Applications (pp. 11–​42). Washington, DC: Georgetown University Press. Hussak, L. J., and Cimpian, A. (2018). Memory accessibility shapes explanation: Testing key claims of the inherence heuristic account. Memory and Cognition, 46(1), 68–​88. https://​doi.org/​10.3758/​s13421-​017-​0746-​8. Jaswal,V. K. (2010). Believing what you’re told: Young children’s trust in unexpected testimony about the physical world. Cognitive Psychology, 61(3), 248–​272. https://​doi.org/​10.1016/​j.cogpsych.2010.06.002. Jaswal, V. K., Lima, O. K., and Small, J. E. (2009). Compliance, conversion, and category induction. Journal of Experimental Child Psychology, 102(2), 182–​195. https://​doi.org/​10.1016/​j.jecp.2008.04.006. Kalish, C. (1998a). Natural and artifactual kinds: Are children realists or relativists about categories? Developmental Psychology, 34(2), 376–​391. Kalish, C. (1998b). Reasons and causes: Children’s understanding of conformity to social rules and physical laws. Child Development, 69(3), 706–​720. https://​doi.org/​10.1111/​j.1467-​8624.1998.tb06238.x. Kalish, C. W. (2002). Children’s predictions of consistency in people’s actions. Cognition, 84(3), 237–​265. https://​ doi.org/​10.1016/​S0010-​0277(02)00052-​5.

305

Emily Foster-Hanson and Marjorie Rhodes Kalish, C. W., and Lawson, C. A. (2008). Development of social category representations: Early appreciation of roles and deontic relations. Child Development, 79(3), 577–​593. https://​doi.org/​10.1111/​j.1467-​8624.2008. 01144.x. Keil, F. C. (1989). Concepts, Kinds, and Conceptual Development. Cambridge, MA: MIT Press. Kinzler, K. D., and Dautel, J. B. (2012). Children’s essentialist reasoning about language and race. Developmental Science, 15(1), 131–​138. https://​doi.org/​10.1111/​j.1467-​7687.2011.01101.x. Kripke, S. A. (1972). Naming and necessity. In Davidson, D. and Harman, G. (Eds.), Semantics of Natural Language (pp. 253–​355). Dordrecht: Springer. Leslie, S.  J. (2008). Generics:  Cognition and acquisition. Philosophical Review, 117(1), 1–​47. https://​doi.org/​ 10.1215/​00318108-​2007-​023. Leslie, S.  J. (2013). Essence and natural kinds:  When science meets preschooler intuition. Oxford Studies in Epistemology, 4, 108–​166. https://​doi.org/​10.1093/​acprof:oso/​9780199672707.003.0005. Leslie, S. J. (2014). Carving up the social world with generics. Oxford Studies in Experimental Philosophy, 1, 208–​232. Leslie, S. J., Khemlani, S., and Glucksberg, S. (2011). Do all ducks lay eggs? The generic overgeneralization effect. Journal of Memory and Language, 65(1), 15–​31. https://​doi.org/​10.1016/​j.jml.2010.12.005. Lowell, B., Mustain, P., Ortenzi, K., and Warner, K. (2015). One name, one fish:  why seafood names matter. Oceana, 1–​12. http://​usa.oceana.org/​sites/​default/​files/​onenameonefishreport.pdf. Lynch, E. B., Coley, J. D., and Medin, D. L. (2000). Tall is typical: Central tendency, ideal dimensions, and graded category structure among tree experts and novices. Memory and Cognition, 28(1), 41–​50. https://​doi.org/​ 10.3758/​BF03211575. Lyons, J. (1977). Semantics (Vol. 2). Cambridge: Cambridge University Press. Mallet, J. (2008). Hybridization, ecological races and the nature of species: Empirical evidence for the ease of speciation. Philosophical Transactions of the Royal Society B: Biological Sciences, 363(1506), 2971–​2986. https://​ doi.org/​10.1098/​rstb.2008.0081. Malt, B. C. (1990). Features and beliefs in the mental representation of categories. Journal of Memory and Language, 29(3), 289–​315. https://​doi.org/​10.1016/​0749-​596X(90)90002-​H. Mandalaywala,T. M., Amodio, D. M., and Rhodes, M. (2018). Essentialism promotes racial prejudice by increasing endorsement of social hierarchies. Social Psychological and Personality Science, 9(4), 461–​469. https://​doi.org/​ 10.1177/​1948550617707020. Mandalaywala,T. M., Ranger-​Murdock, G., Amodio, D. M., and Rhodes, M. (2019).The nature and consequences of essentialist beliefs about race in early childhood. Child Development, 90(4), e437–​e453. https://​doi.org/​ 10.1111/​cdev.13008. Mannheim, B., Gelman, S. A., Escalante, C., Huayhua, M., and Puma, R. (2010). A developmental analysis of generic nouns in Southern Peruvian Quechua. Language Learning and Development, 7(1), 1–​23. https://​doi.org/​ 10.1080/​15475441003635620. Mayr, E. (1982). The Growth of Biological Thought: Diversity, Evolution, and Inheritance. Cambridge, MA: Harvard University Press. Medin, D. L., Ross, N., Atran, S., Burnett, R. C., and Blok, S.V. (2002). Categorization and reasoning in relation to culture and expertise. In Psychology of Learning and Motivation (Vol. 41, pp. 1–​41). Academic Press. https://​ doi.org/​10.1016/​S0079-​7421(02)80003-​2. Medin, D. L., Ross, N. O., Atran, S., Cox, D., Coley, J., Proffitt, J. B., and Blok, S. (2006). Folkbiology of freshwater fish. Cognition, 99(3), 237–​273. https://​doi.org/​10.1016/​j.cognition.2003.12.005. Medin, D., Waxman, S., Woodring, J., and Washinawatok, K. (2010). Human-​centeredness is not a universal feature of young children’s reasoning: Culture and experience matter when reasoning about biological entities. Cognitive Development, 25(3), 197–​207. https://​doi.org/​10.1016/​J.COGDEV.2010.02.001. Mill, J. S. (1884). A System of Logic, Ratiocinative and Inductive: Being a Connected View of the Principles of Evidence and the Methods of Scientific Investigation (Vol. 1). London: Longmans, Green, and Company. Moty, K., and Rhodes, M. (2019, August 29). The unintended consequences of the things we say: What generics communicate to children about unmentioned categories. https://​doi.org/​10.31234/​osf.io/​zkjyr. Murphy, G. L., and Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92(3), 289–​316. https://​doi.org/​10.1037/​0033-​295X.92.3.289. Noyes, A., and Dunham, Y. (2019, April 16). Separating kindhood from naturalness: When two dimensions are better than one. https://​doi.org/​10.31234/​osf.io/​q3zg5. Noyes, A., and Keil, F. C. (2017). Revising deference: Intuitive beliefs about category structure constrain expert deference. Journal of Memory and Language, 95, 68–​77. https://​doi.org/​10.1016/​j.jml.2017.02.003. Pappas, A., and Gelman, S. A. (1998). Generic noun phrases in mother-​child conversations. Journal of Child Language, 25, 19–​33. https://​doi.org/​10.1017/​S0305000997003292. Plato, and Grube, G. M. A. (1974). The Republic. Indianapolis, IN: Hackett Publishing.

306

The Psychology of Natural Kind Terms Prasada, S., and Dillingham, E. M. (2009). Representation of principled connections:  A window onto the formal aspect of common sense conception. Cognitive Science, 33(3), 401–​448. https://​doi.org/​10.1111/​ j.1551-​6709.2009.01018.x. Premack, D. (1990).The infant’s theory of self-​propelled objects. Cognition, 36(1), 1–​16. https://​doi.org/​10.1016/​ 0010-​0277(90)90051-​K. Prentice, D. A., and Miller, D. T. (2007). Psychological essentialism of human categories. Current Directions in Psychological Science, 16(4), 202–​206. https://​doi.org/​10.1111/​j.1467-​8721.2007.00504.x. Putnam, H. (1973). Explanation and reference. In Pearce, G. and Maynard, P. (Eds.), Conceptual Change (pp. 199–​221). Dordrecht: Springer. Putnam, H. (1975). The meaning of “meaning”. In Mind, Language and Reality: Philosophical Papers (Vol. 2, pp. 215–​271). Cambridge: Cambridge University Press. Quine, W. V. (1977). Intensions revisited. Midwest Studies in Philosophy, 2, 5–​11. Rafetseder, E., Cristi-​Vargas, R., and Perner, J. (2010). Counterfactual reasoning: Developing a sense of “nearest possible world”. Child Development, 81(1), 376–​389. https://​doi.org/​10.1111/​j.1467-​8624.2009.01401.x. Rhodes, M., and Gelman, S. A. (2009a). A developmental examination of the conceptual structure of animal, artifact, and human social categories across two cultural contexts. Cognitive Psychology, 59(3), 244–​274. https://​ doi.org/​10.1016/​j.cogpsych.2009.05.001. Rhodes, M., and Gelman, S. A. (2009b). Five-​year-​olds’ beliefs about the discreteness of category boundaries for animals and artifacts. Psychonomic Bulletin & Review, 16(5), 920–​924. https://​doi.org/​10.3758/​PBR.16.5.920. Rhodes, M., Gelman, S. A., and Karuza, J. C. (2014). Preschool ontology: The role of beliefs about category boundaries in early categorization. Journal of Cognition and Development, 15(1), 78–​ 93. https://​ doi.org/​ 10.1080/​15248372.2012.713875. Rhodes, M., Leslie, S. J., Bianchi, L., and Chalik, L. (2017).The role of generic language in the early development of social categorization. Child Development. https://​doi.org/​10.1111/​cdev.12714. Rhodes, M., Leslie, S. J., Saunders, K., Dunham,Y., and Cimpian, A. (2017). How does social essentialism affect the development of inter-​group relations? Developmental Science. https://​doi.org/​10.1111/​desc.12509. Rhodes, M., Leslie, S., and Tworek, C. M. (2012). Cultural transmission of social essentialism. PNAS Proceedings of the National Academy of Sciences of the United States of America, 109(34), 13526–​13531. https://​doi.org/​ 10.1073/​pnas.1208951109. Rhodes, M., and Mandalaywala,T. M. (2017).The development and developmental consequences of social essentialism. Wiley Interdisciplinary Reviews: Cognitive Science, 8(4), e1437. https://​doi.org/​10.1002/​wcs.1437. Richland, L. E., Morrison, R. G., and Holyoak, K. J. (2006). Children’s development of analogical reasoning: Insights from scene analogy problems. Journal of Experimental Child Psychology, 94(3), 249–​273. https://​doi.org/​ 10.1016/​j.jecp.2006.02.002. Rips, L. J. (1975). Inductive judgments about natural categories. Journal of Verbal Learning and Verbal Behavior, 14(6), 665–​681. https://​doi.org/​10.1016/​S0022-​5371(75)80055-​7. Roberts, S. O., and Gelman, S. A. (2016). Can White children grow up to be Black? Children’s reasoning about the stability of emotion and race. Developmental Psychology, 52(6), 887–​893. https://​doi.org/​10.1037/​dev0000132. Roberts, S. O., Ho, A. K., and Gelman, S. A. (2017). Group presence, category labels, and generic statements influence children to treat descriptive group regularities as prescriptive. Journal of Experimental Child Psychology, 158, 19–​31. https://​doi.org/​10.1016/​J.JECP.2016.11.013. Roberts, S. O., Ho, A. K., Rhodes, M., and Gelman, S. A. (2017). Making boundaries great again: Essentialism and support for boundary-​enhancing initiatives. Personality and Social Psychology Bulletin, 43(12), 1643–​1658. https://​doi.org/​10.1177/​0146167217724801. Rosch, E., and Mervis, C. B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7(4), 573–​605. https://​doi.org/​10.1016/​0010-​0285(75)90024-​9 Ross, N., Medin, D., Coley, J. D., and Atran, S. (2003). Cultural and experiential differences in the development of folkbiological induction. Cognitive Development, 18(1), 25–​47. https://​doi.org/​10.1016/​ S0885-​2014(02)00142-​9. Rothbart, M., and Taylor, M. (1992). Category labels and social reality: Do we view social categories as natural kinds? In Semin, G. R. and Fiedler, K. (Eds.), Language, Interaction and Social Cognition (pp. 11–​36). Thousand Oaks, CA: Sage. Schwartz, S.  P. (1979). Natural kind terms. Cognition, 7(3), 301–​315. https://​doi.org/​10.1016/​0010-​0277(79) 90003-​9. Segall, G., Birnbaum, D., Deeb, I., and Diesendruck, G. (2015). The intergenerational transmission of ethnic essentialism: How parents talk counts the most. Developmental Science, 18, 543–​555. https://​doi.org/​10.1111/​ desc.12235. Short, S. D., and Hawley, P. H. (2015).The effects of evolution education: Examining attitudes toward and knowledge of evolution in college courses. Evolutionary Psychology, 13(1), 147470491501300105. https://​doi.org/​ 10.1177/​147470491501300105.

307

Emily Foster-Hanson and Marjorie Rhodes Shtulman, A. (2017). Scienceblind:  Why Our Intuitive Theories about the World are So Often Wrong. New  York:  Basic Books. Shtulman, A., and Calabi, P. (2013). Tuition vs. intuition: Effects of instruction on naive theories of evolution. Merrill-​Palmer Quarterly, 59(2), 141–​167. https://​doi.org/​10.13110/​merrpalmquar1982.59.2.0141. Shtulman, A., and Schulz, L. (2008).The relation between essentialist beliefs and evolutionary reasoning. Cognitive Science, 32(6), 1049–​1062. https://​doi.org/​10.1080/​03640210801897864. Smyth, K., Feeney, A., Eidson, R. C., and Coley, J. D. (2017). Development of essentialist thinking about religion categories in Northern Ireland (and the United States). Developmental Psychology, 53(3), 475–​496. https://​doi. org/​10.1037/​dev0000253. Spelke, E. S., Phillips, A., and Woodward, A. L. (1995). Infants’ knowledge of object motion and human action. In Sperber, D., Premack, D. and Premack, A. J. (Eds.), Symposia of the Fyssen Foundation. Causal Cognition: A Multidisciplinary Debate (pp. 44–​78). Oxford: Clarendon Press/​Oxford University Press. Stewart, J. (1984). Object motion and the perception of animacy. Presented at the meeting of the Psychonomic Society, San Antonio, TX, November 1984. Swiney, L., Bates, D. G., and Coley, J. D. (2018). Cognitive constraints shape public debate on the risks of synthetic biology. Trends in Biotechnology, 36(12), 1199–​1201. https://​doi.org/​10.1016/​j.tibtech.2018.09.002. Tardif,T., Gelman, S.A., Fu, X., and Zhu, L. (2012).Acquisition of generic noun phrases in Chinese: Learning about lions without an ‘-​s’. Journal of Child Language, 39(1), 130–​161. https://​doi.org/​10.1017/​S0305000910000735. Taverna, A. S., Medin, D. L., and Waxman, S. R. (2016).“Inhabitants of the earth”: Reasoning about folkbiological concepts in Wichi children and adults. Early Education and Development, 27(8), 1109–​1129. https://​doi.org/​ 10.1080/​10409289.2016.1168228. Taylor, M.  G. (1996). The development of children’s beliefs about social and biological aspects of gender differences. Child Development, 67, 1555–​1571. DOI:10.1111/​j.1467-​8624.1996.tb01814.x. Taylor, M. G., Rhodes, M., and Gelman, S. A. (2009). Boys will be boys; cows will be cows: Children’s essentialist reasoning about gender categories and animal species. Child Development, 80(2), 461–​481. https://​doi.org/​ 10.1111/​j.1467-​8624.2009.01272.x. Tessler, M. H., and Goodman, N. D. (2019).The language of generalization. Psychological Review, 126(3), 395–​436. https://​doi.org/​10.1037/​rev0000142. VanderBorght, M., and Jaswal, V. K. (2009). Who knows best? Preschoolers sometimes prefer child informants over adult informants. Infant and Child Development: An International Journal of Research and Practice, 18(1), 61–​ 71. https://​doi.org/​10.1002/​icd.591. Vasilyeva, N., Gopnik, A., and Lombrozo, T. (2018). The development of structural thinking about social categories. Developmental Psychology. https://​doi.org/​10.1037/​dev0000555. Vygotsky, L. S. (1962). Thought and Language. Cambridge, MA: MIT Press. Waxman, S. R., and Hall, D. G. (1993). The development of a linkage between count nouns and object categories:  Evidence from fifteen-​to twenty-​one-​month-​old infants. Child Development, 64(4), 1224–​1241. https://​doi.org/​10.1111/​j.1467-​8624.1993.tb04197.x. Waxman, S., Medin, D., and Ross, N. (2007). Folkbiological reasoning from a cross-​cultural developmental perspective:  Early essentialist notions are shaped by cultural beliefs. Developmental Psychology, 43(2), 294–​308. https://​doi.org/​10.1037/​0012-​1649.43.2.294. Wellman, H. M., and Gelman, S. A. (1992). Cognitive development:  Foundational theories of core domains. Annual Review of Psychology, 43, 337–​375. Williams, M. J., and Eberhardt, J. L. (2008) Biological conceptions of race and the motivation to cross racial boundaries. Journal of Personality and Social Psychology, 94, 1033–​1047. https://​doi.org/​10.1037/​0022-​3514.94.6.1033. Wilson, R.  A. (1999). Realism, essence, and kind:  Resuscitating species essentialism. In Species:  New Interdisciplinary Essays (pp. 187–​207). Cambridge, MA: MIT Press. Wilson, R. A., and Keil, F. (1998). The shadows and shallows of explanation. Minds and Machines, 8(1), 137–​159. Xu, F. (2002). The role of language in acquiring object kind concepts in infancy. Cognition, 85(3), 223–​250. https://​doi.org/​10.1016/​S0010-​0277(02)00109-​9.

308

23 PERVASIVE EXTERNALISM Stephen Biggs and Ranpal Dosanjh1

1  Introduction Putnam (1975) argued that underlying features of referents, not their superficial features, determine reference for natural kind terms. Many philosophers quickly agreed. Nearly fifty years later, this externalism about natural kind terms is the dominant view. Philosophers, nevertheless, almost never consider why natural kind terms have this sort of semantics. Here, we provide an account, suggesting that natural kind terms have the semantics Putnam says they do because (1) our concepts develop in a way that conforms to our demand for simple, powerful explanations and (2) natural kind terms label kinds that are important even if their referents have nothing deep in common. This account—​and we think any plausible account is relevantly similar to ours—​has surprising consequences for the semantics of various kind terms (e.g., ‘bully’, ‘voluntary’), as well as for statements and theorizing, including philosophical theorizing, involving those terms.2

2  The Semantics of Natural Kind Terms Putnam (1975) holds that ‘water’ refers to whatever and only whatever has the same so-​called hidden structure as various samples that we have called ‘water’.3 A hidden structure for water is whatever enough samples that have been called ‘water’ share such that it produces at least many of the properties by which we ordinarily recognize samples of water as water (p. 157)—​the structure is hidden in that discovering that water has it requires having experience beyond what is required to be competent with ‘water’. Let wateriness be the collection of superficial properties by which we ordinarily recognize samples of water as water, including water’s characteristic appearance and behavior, and call whatever has these properties watery. According to Putnam, then, if enough samples that we have called ‘water’ share a hidden structure, whatever it is, then ‘water’ refers to whatever and only whatever has that structure, even in worlds where the most common watery stuff lacks that structure, and even in worlds where entities having that structure aren’t watery. But suppose that half of the samples that we have called ‘water’ are watery because they are H2O and half are watery because they are some other compound, XYZ.4 Would ‘water’ be an empty term? Not according to Putnam (1975, 160). Instead, ‘water’ would refer to whatever and only whatever has one of those two hidden structures; it would be like ‘jade’, which invariably refers to whatever and only whatever is jadeite or nephrite. More generally, if enough samples that we have called ‘water’ 309

Stephen Biggs and Ranpal Dosanjh

have any of some not-​too-​unruly number of hidden structures, and each of those is had by enough samples, then ‘water’ refers to whatever and only whatever has one of those structures, even in worlds where the most common watery stuff lacks any of those structures, and even in worlds where entities having those structures aren’t watery. But suppose that no two samples that we have called ‘water’ have anything in common beyond each being watery. Would ‘water’ be an empty term then? Not according to Putnam. In that case, ‘water’ would refer to whatever and only whatever is watery. That is, if samples that we have called ‘water’ have so many different hidden structures that ‘ “hidden structure” becomes irrelevant’, then ‘superficial characteristics become the decisive ones’ (1975, 160). A series of conditionals, the W-​conditionals, capture these commitments (see Bealer (2002, 109) and Korman (2006, 507) for similar conditionals): (W1) If enough samples that we have called ‘water’ have one hidden structure, and few enough samples that we have refused to call ‘water’ have that structure, then (in all worlds) ‘water’ refers to whatever and only whatever has that structure. (W2) Otherwise, if enough samples that we have called ‘water’ have one of two hidden structures, and enough samples have each structure, and few enough samples that we have refused to call ‘water’ have those structures, then (in all worlds) ‘water’ refers to whatever and only whatever has one of those two structures. ⋮ (Wn) Otherwise, if enough samples that we have called ‘water’ have one of n hidden structures, and enough samples have each structure, and few enough samples that we have refused to call ‘water’ have those structures, then (in all worlds) ‘water’ refers to whatever and only whatever has one of those n structures. (W*) Otherwise, (in all worlds) ‘water’ refers to whatever and only whatever is watery. Here, (W1)–​(Wn) capture the natural kind sense of ‘water’, while (W*) captures its superficial kind sense; a number of hidden structures greater than n is too unruly for water to be a natural kind.5 Generalizing gives the N-​conditionals for any (or at least many) natural kind term N : (N1) If enough samples that we have called N have one hidden structure, and few enough samples that we have refused to call N have that structure, then (in all worlds) N refers to whatever and only whatever has that structure. (N2) Otherwise, if enough samples that we have called N have one of two hidden structures, and enough samples have each structure, and few enough samples that we have refused to call N have those structures, then (in all worlds) N refers to whatever and only whatever has one of those two structures. ⋮ (Nn) Otherwise, if enough samples that we have called N have one of n hidden structures, and enough samples have each structure, and few enough samples that we have refused to call N have those structures, then (in all worlds) N refers to whatever and only whatever has one of those n structures. (N*) Otherwise, (in all worlds) N refers to whatever and only whatever has the superficial properties by which we ordinarily recognize the samples that we have called N as members of a kind designated by N. (Here, n will depend on N.) Any account of natural kind terms that implies instances of these N-​ conditionals for at least many natural kind terms is Putnamian.6 310

Pervasive Externalism

Putnamian accounts admit of two competing specifications. Perhaps ‘water’ has a complex intension that the W-​conditionals capture—​where an intension is a function that takes a context of utterance and context of evaluation as input and delivers an extension as output. Perhaps, instead, each of the W-​conditionals corresponds to a discrete, simple intension, only one of which ‘water’ has—​where which it has depends on which of the W-​conditionals has a true antecedent.We presume the former specification until the Appendix, where we show that our central points hold (mutatis mutandis) given the latter.

3  Why There are Natural Kind Terms Why do (at least many) natural kind terms have the semantics that a Putnamian account says they do? Our answer has two related parts.The first part explains why each natural kind term has a natural kind sense. The first and second parts together explain why each has a superficial kind sense. Here is the first part of our answer: as a matter of (possibly contingent) fact, our concepts develop in a way that is connected to our explanatory practices, specifically to conform to our demand for simple, powerful explanations. While this demand can sometimes be satisfied by causal-​historical explanations, it is typically satisfied by structural explanations when we deploy natural kind terms. Our argument below is independent of the origin of this demand, but two possible origins suggest themselves. One is that we have a significant psychological disposition to implicitly, compulsively posit such explanations.There is extensive evidence that good explanation plays a central role in cognition, and that natural kind terms play a central role in reasoning, all from a very early age.7 Another possible origin is in the pragmatics of natural kind terms. It may be that in order to function as part of an informative utterance, a natural kind term is expected to do explanatory work. Either way, natural kind terms have a natural kind sense because our concepts conform to our demand for simple, powerful explanations. The explanatory necessity of this proposal (or a relevantly similar one) could be supported with a dialectical challenge to those who reject it. The challenge: explain why natural kind terms have the semantics that they do without making a similar presupposition. Meeting this challenge is difficult partly because we conceive of natural kinds as being defined by hidden structures even when the only reason to believe that the hidden structures exist is an appeal to explanatory considerations—​as the atomists of ancient Greece conceived of various kinds as defined by a hidden structure long before modern atomic theory. Here is the second part of our answer: when we demand explanations for the unity of the samples that comprise kinds, it is because those kinds are antecedently important to us, that is, important to us apart from the explanation itself.The general idea is this: for natural kinds, there would be no impetus to explain their kindhood by positing a hidden structure if they wouldn’t comprise a kind even if the samples had nothing in common beyond the superficial, from which it follows that terms that refer to them have a superficial kind sense. To support the plausibility of this proposal, consider that our language includes superficial kind terms (like ‘bachelor’, which refers to all and only unmarried men) exactly because the kinds to which they refer are antecedently important (few if any of their members share anything relevant to their kind-​membership beyond the superficial). Correspondingly, we expect natural kinds for which we demand explanations to be antecedently important kinds. This means that the concepts that conform to our demand for explanations will be those expressed by kind terms for such antecedently important kinds. In sum, for any (or at least many) natural kind term N, N has a natural kind sense because our concepts develop in a way that conforms to our demand for simple, powerful explanations, N has a superficial kind sense because it identifies a kind that is important even if relevant samples have nothing deep in common, and so, N has the semantics that a Putnamian account says natural kind terms do. Below, we elaborate this version of what we call the Explanatory Account.8 We show how our proposal explains the semantics of natural kind terms, using ‘water’ as a paradigm case. We do this by 311

Stephen Biggs and Ranpal Dosanjh

first explaining why the instension of ‘water’ implies the existence of hidden structures in general, and then showing why it implies each of the W-​conditionals. To begin, consider a standard account of how we come to think of water as a kind. We learn that the properties comprising wateriness cluster together in a given sample drawn from a lake. We learn that these properties cluster in each sample drawn from the lake. We learn, thereby, that the samples appear and behave similarly across a wide range of circumstances. We learn the same for samples drawn from streams, oceans, rain, and so on. Consequently, we conceive of these samples as unified, as comprising a (non-​arbitrary) kind. Because concepts develop to conform to our demand for explanations, we don’t conceive of the samples foremost as a mere superficial unity, as members of a kind that’s defined by the motley wateriness. Rather, we conceive of them foremost as members of a kind that’s defined by that which explains the facts by which we come to think of it as a kind. We are concerned with the following explananda: the properties that comprise wateriness cluster in any given sample (clustering); the samples are similar to one another in appearance and behavior (similarity); and the samples behave similarly across a wide range of circumstances (stability). The theory that the samples have hidden structures, if true, would explain clustering. This is because the hidden structures of water, if there are any, necessitate wateriness in those samples in a way that is explanatory. Suppose, for example, that the samples’ hidden structure necessitates their wateriness by causing them to be watery. Then, given familiar connections between explanation and causation (see, for example, Salmon 1984), the theory that samples of water share a common hidden structure explains why they are watery. Suppose, instead, that the samples’ hidden structure necessitates their wateriness by grounding it. Then, given familiar connections between explanation and grounding (see, for example, Fine 2001; de Rosset 2013), the theory that samples share a common hidden structure explains why they are watery. Although the nature of explanation is controversial, we take for granted that the necessitation at issue is relevantly explanatory, or would be if it were realized. Likewise, the theory that there is a hidden structure would explain similarity and stability. Having a common hidden structure, whatever that structure is, explains, for any two samples with that structure, why they are similar in appearance and behavior, and also why that appearance and behavior remains similar in a variety of contexts. If water has more than one hidden structure, then the similarity and stability of two samples with distinct structures can be explained (less simply and less powerfully) by the necessitation relation described above. For similarity:  each has a structure that necessitates wateriness. For stability: the behaviors necessitated by the structures are the same across a range of circumstances. (These latter explanations are less powerful in that they depend on the ‘cosmic accident’ that these particular distinct structures happen to necessitate the same superficial properties and behaviors.) We are now in a position to explain why the intension of water implies (W1). Because concepts develop to conform to our demand for explanations that are simple and powerful (roughly, as simple as they can be without losing power), we don’t conceive of that which explains these explananda foremost as multiple, distinct entities. Rather, we conceive of the explainer foremost as a single entity.The theory that the samples have a single common hidden structure is exceptionally simple and powerful. It appeals to just one entity to explain clustering. It appeals to just that same entity to explain similarity and then again to explain stability. We can also explain why the intension of ‘water’ implies (W2)–​(Wn). The theory that some of the relevant samples have one hidden structure and the rest have another, if true, also explains targeted explananda, but less well. The two-​hidden-​structures theory can explain clustering. Each of the two hidden structures separately necessitate wateriness, so any sample with either of those hidden structures is watery. The two-​hidden-​structures theory also explains similarity. Any two samples will either have the same hidden structure, or will have distinct hidden structures that separately necessitate those appearances and behavior. The two-​hidden-​structures theory also explains stability. Any 312

Pervasive Externalism

two samples will either have the same hidden structure, or have distinct hidden structures that separately necessitate common behavior across a variety of circumstances. Nevertheless, the explanations provided by the one-​hidden-​structure theory are simpler than those provided by the two-​hidden-​structures theory. The latter implies two kinds where the former implies one. Also, the latter offers multiple explanations of clustering (one for each structure), and of similarity and stability (one for pairs of samples with a common hidden structure, and one for pairs of samples with distinct hidden structures), where the former offers one. The explanations of the one-​hidden-​structure theory are also more powerful than those of two-​hidden-​structures theory. For the latter, there is no explanation for the cosmic accident by which distinct hidden structures happen both to necessitate wateriness, and by which the behaviors happen to be stable across a range of circumstances. But for these same reasons, the two-​hidden-​structures theory is simpler and more powerful than any three-​hidden-​structures theory, which is simpler and more powerful than any four-​hidden-​structures theory, and so on. Since the two-​hidden-​structures theory explains much of what we want to explain and is quite simple and powerful, and since concept development conforms to our demand for such explanations, we conceive of the samples as being defined by two hidden structures if there isn’t just one, three if there aren’t just two, and onward until the number of hidden structures becomes too unruly. The number becomes too unruly where explanations that appeal to hidden structures lose too much simplicity and power (though how much is too much remains an open question). Consequently, since ‘water’ expresses our concept of the samples, its intension implies (W2)–​(Wn). Finally, we can explain why the intension of ‘water’ implies (W*). Once the number of hidden structures becomes too unruly—​at the limit, when each sample that we have called ‘water’ has a distinct hidden structure—​explanations that appeal to hidden structures lose their simplicity and power. At that point, the demand for simple and powerful explanations is not satisfied by conceiving of the samples as their hidden structure. That ‘water’ refers to watery things instead of being an empty term is a result of the role that water plays in our lives. Water is antecedently important to us. The relevant samples comprise an important kind even if they are unified only by their superficial properties; after all, watery things nourish life, erode soil, freeze at zero degrees Celsius, and so on. Because of the importance of wateriness, we conceive of the samples as comprising a kind, however superficial, even if there are no relevant hidden structures. Consequently, since ‘water’ expresses our concept of those samples, the intension of ‘water’ implies (W*). We maintain that this is the most plausible explanation for why natural kind terms have the semantics that they do. But again, even if the details are wrong, something in the neighborhood of this account is right, since any plausible explanation for why each natural kind term refers to whatever and only whatever shares a certain hidden structure (if there is one), rather than whatever and only whatever shares relevant superficial properties, appeals to the (potential) explanatory value of positing common hidden structures, and any plausible explanation for why, with few exceptions, each natural kind term refers to a superficial kind if there is no common hidden structure, rather than not referring at all, appeals to the importance of the superficial kind. This is crucial because the points that we advance in the next section could be advanced given any explanation that appeals to the (potential) explanatory value of positing common hidden structures and the importance of the kind. Any such explanation for why natural kind terms have the semantics that they do is a version of the Explanatory Account.9

4  Pervasive Externalism According to a Putnamian account, at least many natural kind terms have a superficial kind sense; their intensions include conditions for reference in the absence of hidden structures, as the intension of ‘water’ implies (W*). This wouldn’t be so if such conditions were invariably superfluous. Given 313

Stephen Biggs and Ranpal Dosanjh

Putnamian accounts, then, we should expect that many superficial kind terms have a natural kind sense, and would refer to whatever and only whatever has a certain hidden structure if there were a relevant one. It follows that we should expect the intensions of superficial kind terms to imply instances of the N-​conditionals. The arguments in the previous section apply not just to natural kind terms, but equally well to many superficial kind terms. Which terms? Putnam says ‘there is a strong tendency for words that are introduced as “one criterion” words to develop a “natural kind” sense’ (1975, 163). He offers ‘pediatrician’ as an example, holding that it has a natural kind sense (and thus, would refer to a hidden structure if there were one) even though it was introduced as shorthand for ‘child doctor’. But, beyond this, he provides little guidance.10 The Explanatory Account provides further guidance. Given the account, for any typical kind term K, the intension of K implies instances of the N-​conditionals if: (1) the theory that samples of that kind share some not-​too-​unruly number of hidden structures would, if true, have sufficient explanatory value, and (2) the relevant samples would comprise a sufficiently important kind even if there weren’t relevant hidden structures. Consider an example of a superficial kind term that meets these conditions. Suppose that all and only bullies share biological nature B, which makes them habitually cruel to the weak.11 In that case: one who wonders why she is habitually cruel to the weak (despite her best efforts not to be) can answer that she has B; one who wonders why she and a friend are both habitually cruel to the weak can answer that both have a common biological nature (and that it is B); and one who wonders why her otherwise aggressive friend ignores smaller and weaker potential targets can answer that the friend doesn’t have B. This suggests that the theory that bullies share a hidden structure would, if true, have significant explanatory value, as would the theory that they share any not-​too-​unruly number of hidden structures. And, even if bullies don’t share any not-​too-​unruly number of hidden structures, they still comprise a kind that’s important—​after all, we use ‘bully’ regularly even though (presumably) bullies don’t in fact share a hidden structure. Given the Explanatory Account, then, the intension of ‘bully’ is likely to imply the following instances of the N-​conditionals, which we can call the B-​conditionals: (B1) If enough people who we have called ‘bully’ have one hidden structure, and few enough people who we have refused to call ‘bully’ have that structure, then (in all worlds) ‘bully’ refers to whomever and only whomever has that structure. (B2) Otherwise, if enough people who we have called ‘bully’ have one of two hidden structures, and enough of those people have each structure, and few enough people who we have refused to call ‘bully’ have those structures, then (in all worlds) ‘bully’ refers to whomever and only whomever has one of those two structures. ⋮ (Bn) Otherwise, if enough people who we have called ‘bully’ have one of n hidden structures, and enough of those people have each structure, and few enough people who we have refused to call ‘bully’ have those structures, then (in all worlds) ‘bully’ refers to whomever and only whomever has one of those n structures. (B*) Otherwise, (in all worlds) ‘bully’ refers to whomever and only whomever is habitually cruel to the weak. Any Putnamian account requires some version of the Explanatory Account to explain why natural kind terms have the semantics that they do. So, advocates of a Putnamian account should accept that ‘bully’ has a natural kind sense. (Again, they must accept that some superficial kind terms have a natural kind sense; at issue here is only which ones.) These considerations generalize, suggesting that advocates of a Putnamian account should accept that many superficial kind terms have a natural kind sense; that is, they should accept Pervasive Externalism. 314

Pervasive Externalism

To illustrate this further, it is worth considering some examples where erstwhile superficial kind terms (introduced by some description) are now in fact deployed as natural kind terms. One such example is the term ‘solid’.The classical sense of the term contrasts being hollow or being gappy, both for material objects (‘solid gold’) and for geometric figures (‘solid sphere’). It also has the sense of rigid or impervious, which, by the seventeenth century, allowed it to be applied to a state of matter contrasting fluids. In this classical sense, a solid is a superficial kind. But by the early twentieth century, ‘solid’ developed a natural kind sense that contradicted the classical one.12 Most of the mass of an atom was discovered to be concentrated in the nucleus, with most of the remaining volume empty space. So ordinary, paradigmatically solid matter is not solid in the original sense: all such matter is hollow or gappy. Nevertheless, the term ‘solid’ continues to be applied to ordinary matter: we don’t say that matter is not solid because it is gappy. The new discoveries are used to explain how matter is rigid and impervious, rather than to show that it is not solid. This demonstrates that, despite being introduced as a superficial kind term, ‘solid’ developed a natural kind sense that stripped away some of these superficial features (not being hollow or gappy) but that explained some of the others (being rigid and impervious). Here is another example. The classical sense of ‘vacuum’ is nothingness or emptiness. While Democritus and other atomists held that atoms moved in a vacuum, Aristotle held that a vacuum could not be (there was no vacuum in nature). It wasn’t until Newtonian conceptions of absolute space were adopted that the idea of empty regions began to be widely accepted. A vacuum can thus be treated as a limiting case of a superficial kind: any region of space with nothing is a vacuum, and any region of space with something is not. Candidate regions of nothingness include deep space (in cosmology) and the space between a nucleus and electrons (in atomic physics); each of these has been called a vacuum. However, the current (quantum field theoretic) understanding of such regions gives them a complex structure with non-​zero energy, filled with virtual particles and spontaneous particle-​antiparticle creation and annihilation. If the term ‘vacuum’ had only a superficial kind sense, then we would have expected to return to the idea that a vacuum doesn’t exist in nature. But instead, the vacuum is said to have a structure (incompatible with being nothing or empty), and anything with that structure is a vacuum. (In practice, the vacuum is treated as the state that obtains when everything that can be removed from a region of space has been removed.) This once more illustrates a superficial kind term developing a natural kind sense accompanied by an increase in explanatory power. In both these cases, superficial kind terms developed their natural kind senses owing to metaphysically contingent facts about the world. Had the world been such that matter was a continuum, then ‘solid’ would have been a superficial kind term. And had the world been such that the space between bits of matter was empty, then ‘vacuum’ would have been a superficial kind term. We have every reason to believe that this generalizes to those superficial kind terms, like ‘bully’, that happen, again as a matter of contingent fact, not to have a hidden structure, or to have an unruly number of hidden structures. Pervasive externalism has at least two important consequences. First, consider that, on a standard understanding of natural kind terms, some traditionally synthetic statements (like ‘water is H2O’) express a posteriori necessary truths. Pervasive externalism goes further. It implies that, contrary to standard thinking, some traditionally analytic statements (like ‘all bullies are habitually cruel to the weak’) express a posteriori necessary truths.This follows from Pervasive Externalism’s analysis of superficial kind terms. Second, and correspondingly, Pervasive Externalism suggests that empirical investigation plays a greater role than has been acknowledged in some philosophical disputes. For example, although the dispute between compatibilists and incompatibilists about free will is often taken to be foremost semantic, if ‘voluntary’ has a natural kind sense (as the Explanatory Account suggests), then which of compatibilism or incompatibilism is true depends on whether or not enough actions that we have called ‘voluntary’ share a hidden structure, and if they do, how that hidden structure fits into the ordinary (deterministic or spontaneous) unfolding of events, an issue that is broadly empirical. There 315

Stephen Biggs and Ranpal Dosanjh

is much to say about these and other consequences, but we leave that for another time, though we revisit it briefly in the conclusion.

5  The Limits of Pervasive Externalism Based on the Explanatory Account, we should expect most superficial kind terms to have natural kind senses and most natural kind terms to have superficial kind senses. Nevertheless, there may be some superficial kind terms that don’t have natural kind senses, and (perhaps less likely) some natural kind terms that don’t have superficial kind senses. We see three exceptions under which, for any given kind term, the Explanatory Account does not imply that it has both a natural kind sense and a superficial kind sense. These are the limits of Pervasive Externalism. For the first exception, recall the first part of our proposal: concepts conform to our demand for explanations. These demands are plausibly diachronic, so we should expect that kind terms tend to develop a natural kind sense over time, not that they are invariably introduced with such a sense.This likely happened in the example cases of ‘solid’ and ‘vacuum’, described above. Other examples illustrate the same pattern. A word introduced to refer to whatever and only whatever is metaphysically simple (‘atom’) might have a descriptive sense initially, but then develop a natural kind sense as it is used repeatedly to refer to actual atoms. If it already has a natural kind sense when we discover that atoms are compound, it would explain why we say that atoms are divisible rather than saying that hydrogen is not an atom after all.Therefore, recently introduced or rarely used superficial kind terms might not have a natural kind sense.13 For the second exception, recall the second part of our proposal: we demand explanations for those kinds that are antecedently important to us. This claim was used in the argument for why natural kind terms have superficial kind senses. According to the Explanatory Account, the intension of a kind term implies an instance of (N*) if the relevant samples would comprise a sufficiently important kind even if there weren’t relevant hidden structures. Water and bullies are important to us even if neither has a hidden structure, so they imply instances of (N*). But some kinds may be important only insofar as its members share a hidden structure—​perhaps ‘phlogisticated substance’ is an example. The intension of any term for such a kind would imply instances of (N1)–​(Nn) but not an instance of (N*). Thus, kinds that are important only insofar as their members share a hidden structure might not have a superficial kind sense. For the third exception, recall the argument for why kind terms have natural kind senses.According to the Explanatory Account, a kind term will develop a natural kind sense if the theory that relevant samples share some not-​too-​unruly number of hidden structures would, if true, have sufficient explanatory value. A kind term where such a theory does not have sufficient explanatory value might not develop a natural kind sense. Characterizing sufficient explanatory value is difficult. The explanatory value of a theory is a function of both the facts that the theory promises to explain and our independent interest in explaining those facts. Any theory that explains multiple facts that we have significant independent interest in explaining has sufficient explanatory value. Any theory that explains no fact that we have any independent interest in explaining lacks sufficient explanatory value. Between these is a spectrum of explanatory values. This characterization is vague since it doesn’t specify who ‘we’ are or what constitutes ‘significant’ interest, and it’s incomplete since it doesn’t address the explanatory value of theories that either explain only one fact or are of only intermediate independent interest. Nevertheless, this characterization can help one think about particular cases. Consider again the term ‘bully’. Recall that the theory that bullies share a hidden structure, regardless of whether or not we ever consider it, promises to explain many facts. It would explain: why a given bully is habitually cruel at all; why they direct their cruelty toward smaller, weaker targets (habitually cruel people aren’t always so discriminating); why any two bullies behave so similarly. Moreover, many of us (parents, teachers, psychologists, victims, bullies) have an independent interest 316

Pervasive Externalism

in these facts; we wonder these things of many bullies that we encounter, and we would value relevant explanations. By the characterization above, the theory that bullies share a hidden structure would, if true, have sufficient explanatory value. So ‘bully’ has a natural kind sense. Now, consider ‘bachelor’. The theory that bachelors share a hidden structure (a single structure that produces in any bachelor both being unmarried and being male) would, if true, explain multiple facts—​for example, it would explain of each bachelor why he’s unmarried, and of any two why neither is married. Do we have significant independent interest in explaining these facts? We suspect that few of us do, which suggests that the theory lacks sufficient explanatory value. But someone who’s obsessed with marriage and maleness, as some are, might have significant independent interest in explaining these explananda, which suggests that the theory in question would, if true, have sufficient explanatory value for him. Perhaps, then, ‘bachelor’ is polysemous, with most uses lacking a natural kind sense but some uses having one.14

6  Putnam and Pervasive Externalism Although we aim to advance Pervasive Externalism given the truth of Putnamian accounts of kind terms, not to argue that Putnam would accept it, we pause to note that Putnam should be sympathetic. Suppose we introduce ‘bully’ by stipulation to be shorthand for some description (‘those who are habitually cruel to the weak’), and ‘bully’ hasn’t changed its meaning since it was introduced. In that case, on our view, giving up ‘all bullies are habitually cruel to the weak’ requires changing the meaning of ‘bully’. Nevertheless, because concepts conform to our demand for simple, powerful explanations, together with our interest in explaining relevant facts, ‘bully’ has developed a natural kind sense. Thus, giving up ‘all bullies are habitually cruel to the weak’ doesn’t require changing the intension of ‘bully’, even though it requires affirming the antecedent of a different part of that complex intension. Putnam agrees with this general picture. He agrees that ‘if we give up a principle that is analytic in this trivial sense’ of having been introduced by stipulation to have a certain meaning, ‘then we have clearly changed the meaning of a word’ (1962, 40). He also agrees that ‘[c]‌onventionality isn’t “a lingering trait” of statements introduced as truths by stipulation’ (1962, 41) since words introduced by stipulation to have a certain meaning don’t invariably retain their ‘conventional character’ (1962, 38). But he thinks that words only lose their conventional character when their referents figure into scientific explanations and laws. In contrast, we contend that it depends on our independent, often only implicit interest in explaining what such laws would explain, if they obtained. Consider an example. Putnam treats ‘all bachelors are unmarried’ as a ‘trivial … analyticity’ (1962, 40). This is because words only lose their conventional character when their referents figure into scientific explanations and laws, and ‘one is simply not going to find any laws … about [bachelors]’ (1962, 57). On our picture, whether or not ‘bachelor’ loses its conventional character doesn’t depend on whether or not one explicitly expects to find laws about bachelors, but rather, on our independent interest in explaining what such laws would explain, if they obtained. Moreover, despite this claim about ‘bachelor’, Putnam accepts the following. First, since ordinary uses of ‘bachelor’ would refer to anything with a certain hidden structure if there were such laws, ‘all bachelors are unmarried’ is analytic only given that there are no such laws. Second, we can’t know a priori that there aren’t such laws. Third, one can be competent with ‘bachelor’ without knowing whether there are such laws. Together, these three claims imply that we can’t know a priori that ‘all bachelors are unmarried’ since ‘bachelor’ has a natural kind sense. Putnam’s externalism, then, may be no less pervasive than ours.15 Whatever Putnam thought about ‘bachelor’ in 1962, his 1975 discussion of ‘pediatrician’ suggests that he would endorse Pervasive Externalism. He says that ‘pediatrician’ has a natural kind sense even though it was introduced as shorthand for ‘child doctor’; it’s unlikely that he thinks that we’re 317

Stephen Biggs and Ranpal Dosanjh

going to find relevant laws for pediatricians. He also extends externalism to artifact terms and even some verbs.16 Putnam’s discussion of ‘hunter’, however, suggests that he would reject our assessment of ‘bully’. He says that ‘terms whose meanings derive from a [syntactic] transformation’ don’t develop a natural kind sense (1975, 164). The meaning of ‘hunter’, he thinks, derives from a syntactic transformation of ‘one who hunts’, and thus, ‘a hunter is one who hunts’ expresses a truth in all worlds regardless of whether or not hunters share a hidden structure. Presumably, he’d at least consider saying the same for ‘a bully is one who bullies’. But Putnam never says why ‘hunter’ doesn’t develop a natural kind sense. Perhaps he would point to the fact that it was introduced by a syntactic transformation. But he can’t treat a term’s origin as decisive because he holds that ‘pediatrician’ developed a natural kind sense even though it was introduced as shorthand for ‘child doctor’. Perhaps he thinks that the psychological tie created by the syntactic transformation overrides the psychological push to develop a natural kind sense. We find this implausible—​not to mention that ‘pediatrician’ is a syntactic transformation of ‘pediatric’. At any rate, Putnam offers no compelling rationale for treating ‘pediatrician’ and ‘hunter’ differently. The possibility that some verbs have a natural kind sense deserves more exploration than it has received. We suspect that nouns and corresponding verbs might change their senses more or less together, where a change in one precipitates a change in the other. On this view, ‘a hunter is one who hunts’ might start out a priori (with both ‘hunter’ and ‘hunts’ having only superficial kind senses), become a posteriori as ‘hunter’ takes on a natural kind sense, and then return to being a priori as ‘hunts’ also takes on a closely related natural kind sense. Similarly, ‘a bully is one who bullies’ might initially be a priori (with both ‘bully’ and ‘bullies’ having only superficial kind senses), become a posteriori when the ‘bully’ takes on a natural kind sense, and then become a priori again when ‘bullies’ takes on a closely related natural kind sense, too.

7  Conclusion We conclude by noting two issues that merit further exploration. First, we expect that Pervasive Externalism applies to many important terms in philosophy. We noted above that it applies to terms, such as ‘voluntary’, that are central to theorizing about free will. We expect it also applies to terms that are central to theorizing in, among other domains, epistemology (e.g., ‘knowledge’), ethics (e.g., ‘justice’), and philosophy of mind (e.g., ‘consciousness’). Future work will explore these applications. Relatedly, we noted above that the fact that ‘voluntary’ has a natural kind sense has important consequences for discussions of free will, such as the disagreement between compatibilists and incompatibilists, rendering their dispute empirical rather than semantic. We expect similar results in other domains. Second, despite our focus on hidden structures, the Explanatory Account can be extended to other sorts of explanations. In particular, there is no reason that the demand for simple, powerful explanations to which our concepts conform can’t be satisfied by causal-​historical explanations, or with a combination of multiple kinds of explanation. For example, jumping desert rodents (composed of individual kangaroo rats, jerboas, and hopping mice) have the following superficial properties: long hind legs, small rounded bodies, and desert-​adapted behavior. The theory that they have a common set of hidden structures would, if true, explain the unity (clustering, similarity, and stability) of these individuals with respect to these properties. But, as a matter of fact, each individual has one of several hidden structures (one genomic structure for each species), which provides only a weak explanation of similarity and stability among individuals with different structures.17 The theory that they have a common type of causal history would, if true, also explain the unity of the individuals with respect to those properties. And, as a matter of fact, the common type of environment and ecological niche of their ancestors (leading to convergent evolution) provides a better explanation of similarity (and possibly stability) among individuals with different structures. 318

Pervasive Externalism

Consider a final example. Diamonds are natural kinds: they are carbon crystals with a particular geometric structure. The hidden structure provides a structural explanation for the unity of samples of diamonds. However, for most of human history, the only samples of diamonds to which people had access had formed naturally through geological processes. This causal history is also part of an (efficient causal) explanation of the unity of the samples of diamonds. According to the Explanatory Account, how these explanations drive the development of the concept expressed by ‘diamond’ depends on which facts we have an independent interest in explaining, that is, in the explanatory value of the relevant theories. This expectation is met when we look at the following case study.18 Those interested in regulating trade in diamonds traditionally defined ‘diamond’ as something having both a particular type of hidden structure and a particular type of causal history. For example, since 1956, the US Federal Trade Commission (FTC) defined ‘diamond’ as ‘a natural mineral consisting essentially of pure carbon crystallized in the isometric system’ (emphasis ours) before listing some of its physical and chemical properties. With the advent of laboratory-​grown stones with the same hidden structure, a (commercially) important decision had to be made: are the stones with the type-​identical hidden structure but type-​distinct causal history really diamonds? The question was addressed in 2018. Those who manufactured the stones argued that being natural is not a necessary attribute of diamonds. The FTC agreed, and struck the word ‘natural’ from the definition. So, according to the FTC, diamonds are indeed natural kinds, irrespective of their causal history. Those who mine diamonds argued that consumers have an expectation that diamonds are produced in the earth over the course of billions of years, and that their value arises in part owing to their causal history.The FTC agreed, and required that laboratory-​grown diamonds be clearly labeled as such in marketing materials, while mined diamonds require no modifiers. This case study shows that the factors considered by the FTC to modify the trade definition of ‘diamond’ track those described by the Explanatory Account. Once there was a tension between the structural explanation and the causal-​historical one, our independent interest in explaining the various facts about diamonds became a factor in the development of the definition. The parties to the debate have significant independent interest in explaining the (non-​ structural) physical and chemical properties of diamonds as well as significant independent interest in explaining their value. These interests pulled in different directions, but ultimately the former (and stronger) interest prevailed, and the trade definition was brought in line with the common definition, with some constraints on the use of the term to accommodate the latter (and weaker) interest.

Appendix We noted earlier that Putnamian accounts admit of two specifications. On one, ‘water’ has a complex intension that the W-​conditionals capture. On the other, ‘water’ has a simple intension, and which it has depends on which of W-​conditionals has a true antecedent. Call the former specification Complex Intension and the latter specification Simple Intension. On Complex Intension, ordinary uses of ‘water’ correspond to the same (complex) intension, regardless of which W-​conditional has a true antecedent. To see this, compare a token ordinary use in a scenario in which the antecedent of (W1) is true with a token ordinary use in a scenario in which the antecedent of (W*) is true. On the former, ‘water’ would be a natural kind term. On the latter, ‘water’ would be a superficial kind term. But the two uses correspond to the same intension (specifically, the intension that implies the W-​conditionals). The uses, then, are synonymous. On Simple Intension, which (simple) intension an ordinary use of ‘water’ corresponds to depends on which of the W-​conditionals has a true antecedent. To see this, compare a token ordinary use in a scenario in which the antecedent of (W1) is true with a token ordinary use in a scenario in which the antecedent of (W*) is true. On the former, ‘water’ would be a natural kind term. On the latter, 319

Stephen Biggs and Ranpal Dosanjh

‘water’ would be a superficial kind term. And no intension corresponds to both utterances. The uses, then, are homonymous. The Explanatory Account can be adjusted to fit Simple Intension. The adjusted account would hold that (W1) captures the simple intension of ordinary uses of ‘water’ if its antecedent is true, and does so because of our demand for simple, powerful explanations. And so on, mutatis mutandis, for each of (W2)–​(Wn). The adjusted account would hold that (W*) captures the simple intension of ordinary uses of ‘water’ if its antecedent is true, and does so because water is an important kind even if relevant samples aren’t unified in any deep way. Given Complex Intension, the Explanatory Account explains for each natural kind term a semantic fact: that it has the complex intension that it does. Given Simple Intension, the adjusted account explains for each natural kind term a semantic fact and n metasemantic facts: the semantic fact that it has the simple intension that it does; and the metasemantic facts that it would have the simple intension captured by an instance of (Ni) if there were i hidden structures (for each i ≤ n such that it doesn’t have i hidden structures), and would have the simple intension of an instance of (N*) if there were no relevant hidden structures. So, on Simple Intension, the adjusted account becomes a story about how candidate simple intensions are determined and ranked, rather than a story about how a complex intension develops. But the spirit of the story is the same. The case for Pervasive Externalism is preserved under the Simple Intension specification. On Complex Intension, ‘bully’ has a natural kind sense in a (fairly) literal sense of ‘sense’:  something semantic, the intension of ‘bully’, foremost directs us to any relevant hidden structures. On Simple Intension,‘bully’ has a natural kind sense in an attenuated sense of ‘sense’: something metasemantic (our psychology, the pragmatics of language, or whatever is responsible for our demand for explanations) privileges an intension that directs us to any relevant hidden structures over other possible intensions. So, on Simple Intension, the semantic similarity between ‘water’ and ‘bully’ described in section 5 becomes a description of their metasemantic similarity. Simple Intension likewise doesn’t imply that we can know a priori that ‘all bullies are habitually cruel to the weak’ expresses a necessary truth. On Complex Intension, we can’t know a priori whether or not this statement expresses a necessary truth because we can’t know a priori which of the B-​conditionals has a true antecedent—​even if any competent user of ‘bully’ can identify its complex intension. On Simple Intension, we can’t know a priori whether this statement expresses a necessary truth because we can’t know a priori which consequent captures the intension of ‘bully’, and thus, we can’t even identify the intension of ‘bully’ without experience beyond that which is required to be competent with ‘bully’.19 (Notice that, on Simple Intension, while identifying the intension of ‘bully’ requires determining how many hidden structures people who we have called ‘bully’ share, being competent with ‘bully’ doesn’t require determining the same, but being competent with ‘bully’ does require being able to identify the candidate simple intensions that the B-​conditionals capture.20)

Notes 1 This chapter is thoroughly co-​authored. The order of the authors is merely alphabetical. 2 Kripke (1980) also argued influentially for semantic externalism about natural kind terms. Putnam and Kripke (seem to) disagree about various details. Although we focus on Putnam’s way of talking and thinking, most of what we say could be reframed to apply to Kripke’s semantic externalism. See note 20 for a brief discussion of how. 3 Symbols inside single quotes refer to terms or statements unless otherwise noted, where terms and statements are respectively literally interpreted words and sentences, which themselves are un-​interpreted strings of symbols. The discussion could be reframed to avoid this commitment to literal interpretations (for example, by focusing on particular utterances, as we occasionally do). 4 Although we often presume that H2O is the hidden structure of water, what matters here is the intension of ‘water’, not the nature of water. Our argument, then, sidesteps objections to the claim that H2O is the hidden structure of water (from, e.g.,Weisberg 2005 and Needham 2011)—​as well as more general concerns about the claim that many natural kind terms correspond to things with certain hidden structures (from, e.g., Leslie 2013).

320

Pervasive Externalism 5 We use ‘natural kind’ to refer not only to kinds that have a single, unified hidden structure, but also to some kinds (like jade) with disjunctive hidden structures. We follow Putnam when we say that terms that refer to such kinds have a natural kind sense (1975, 163): they pick out individuals with certain hidden structures in all possible worlds, regardless of the superficial properties of those individuals in those worlds. 6 McLaughlin and Tye (1998), Bealer (2002), and Korman (2006), among others, adopt Putnamian accounts. Opponents include both those who reject semantic externalism outright (for example, Wikforss 2005), and externalists who hold that any natural kind term is empty unless there is a relevant hidden structure (for example, Nuccetelli 2003). 7 On the significant role of explanation in cognition and its early appearance, see, e.g., Gelman (2003), Sobel, Yoachim, Gopnik, Meltzoff, and Blumenthal (2007), Gopnick (2012), and Walker, Lombrozo, Legare, and Gopnik (2014). On the significant role that natural kind terms play in reasoning and its early appearance, see, e.g., Gelman and Markman (1986), Nazzi and Gopnick (2000), Walker et al. (2007), and Gelman (2003). Nazzi and Gopnick find, for example, that by around age four and a half most children presume that a novel term that is introduced to name an object (in a context where both hidden structure and superficial properties are salient) names whatever and only whatever shares the same hidden structure as that object, rather than whatever and only whatever shares its superficial properties. Walker et al. find that (in a similar context) children as young as three who are prompted to think about explanation use novel labels to track hidden structure, rather than superficial properties. For a useful discussion of findings that support the more general claim that we are strongly disposed to think of members of many kinds as sharing hidden structures, and that this disposition is ‘traceable to a deep-​seated cognitive outlook’ (Leslie 2013, 109), see Leslie (2013, section 1). 8 While something along the lines of the Explanatory Account may be implicit in many people’s thinking about natural kind terms, the most similar discussion comes in Leslie (2013). Leslie agrees with a version of the Explanatory Account that our dispositions to apply kind terms in ways that are consistent with semantic externalism stems from a psychological compulsion. But she takes that compulsion to be a brute, compulsive positing of essences, rather than, as per the Explanatory Account, a compulsive pursuit of simple, powerful explanations that manifests, among many other ways, as a compulsive positing of hidden structures; and she doesn’t hold that the compulsion at issue impacts semantics, merely that it impacts how we apply terms. 9 Two notes. First, as an aside, although the Explanatory Account appeals to how we conceive of various entities, eliminativists about concepts can accept an appropriately reframed version of the account. They can hold, for example, that the intension of ‘water’ implies (W1) because our demand for simple, powerful explanations directly influences behavior, disposing us to use ‘water’ to refer to whatever and only whatever has a certain hidden structure, provided that enough samples share a common hidden structure. Correspondingly, our presupposition that our demand for simple, powerful explanations influences concept formation could be replaced by the presupposition that the demand directly influences linguistic behavior. More broadly, the points we make throughout this essay can be reframed to suit a wide range of theories of mental states (for example, eliminativists theories) and general metasemantical theories (for example, use theories). Second, although we can’t explore this issue here, conjoining a Putnamian account with the Explanatory Account generates replies to many objections to a Putnamian account (e.g., objections from Häggqvist and Wikforss 2007). 10 Although Putnam’s (1975) suggestion that semantic externalism extends to some superficial kind terms (e.g., ‘pediatrician’) has points of contact with ours, subsequent commentary on his discussion of this extension (e.g., Schwartz 1978; Kornblith 1980) has little bearing on what we say. 11 One who doubts that biological kinds have hidden structures (see, e.g., Dupré 1981; Okasha 2002) might be wary of this appeal to B. But this wariness would be misplaced:  the supposition that bullies share a hidden structure is used to test how ‘bully’ would apply if they did, but it doesn’t imply that the supposition corresponds to a metaphysical possibility, let alone that actual biological kinds have hidden structures; and arguments that inspire this wariness challenge the claim that biological kinds in fact have hidden structures, not the coherence of the idea that biological kinds have hidden structures. 12 Rutherford’s (1911) description of atomic nuclei likely initiated this change. 13 To be clear, as per as Evans (1973), two uses of a word can express distinct concepts even if the uses are, in some sense, part of the same ‘chain’; correspondingly, uses of a word that have a natural kind sense and uses that don’t may be part of the same chain. 14 Putnam (1975, 1990)  acknowledges a loosely related role for interests. To be clear, our claim is that our interests impact our terms, not hidden structures themselves. 15 See Putnam’s example of all bachelors sharing a particular neurosis (1962, 58), although the extent of the resemblance between that case and a case of bachelors sharing a hidden structure is unclear since he never says whether or not the neurosis in question is a hidden structure for bachelors; specifically, he never says whether or not the neurosis makes one unmarried and male, being an unmarried male makes one have the neurosis, a single hidden structure makes one unmarried, male, and appropriately neurotic, or whatever.

321

Stephen Biggs and Ranpal Dosanjh 16 We find the extension to artifacts (at least those whose origins are known) less plausible because we doubt that we have significant independent interest in explaining what relevant theories, if true, would explain. 17 Likewise for the parallel explanation that appeals to a relevant common ancestor. 18 See Federal Trade Commission (2018), especially section R. 19 This point resembles McLaughlin and Tye’s (1998) claim that competent users of any given natural kind concept C can’t know a priori whether or not C is a natural kind concept or a superficial kind concept because (i) they can’t know a priori whether or not there is a relevant hidden structure, (ii) C picks out things with a relevant hidden structure since there is one, and (iii) C would pick out a relevant superficial kind if there weren’t a relevant hidden structure. McLaughlin and Tye invoke this aposteriority to respond to McKinsey’s (e.g., 1991) challenge to externalism about thought content. Others appeal to the same kind of aposteriority to address McKinsey’s challenge while either holding that any given natural kind term would have no extension if there were no relevant hidden structure (e.g., Nuccetelli 2003) or remaining ambivalent between that view and a broadly Putnamian account (e.g., Brown 2004). 20 Plausibly, Kripke (1980) presumes Simple Intension. Accordingly, our central claims could be applied given Kripke’s way of talking and thinking with appropriate changes in form.

References Bealer, George. 2002. ‘Modal Epistemology and the Rationalist Renaissance’, in Tamar Gendler and John Hawthorne (eds.), Conceivability and Possibility, New York: Oxford University Press. Brown, Jessica. 2004. Anti-​Individualism and Knowledge, Boston, MA: MIT Press. de Rosset, Louis. 2013. ‘Grounding Explanations’, Philosopher’s Imprint 13: 1–​26. Dupré, John. 1981. ‘Natural Kinds and Biological Taxa’, Philosophical Review 90: 66–​90. Evans, Gareth. 1973. ‘The Causal Theory of Names’, Proceedings of the Aristotelian Society, Supplementary Volume 47: 187–​208. Federal Trade Commission. 2018. Statement of Basis and Purpose: Final Revisions to the Jewelry Guides. www.ftc.gov/​ public-​statements/​2018/​07/​statement-​basis-​purpose-​final-​revisions-​jewelry-​guides. Fine, Kit. 2001. ‘The Question of Realism’, Philosopher’s Imprint 1: 1–​30. Gelman, Susan. 2003. The Essential Child: Origins of Essentialism in Everyday Thought, Oxford: Oxford University Press. Gelman, Susan and Markman, Ellen. 1986. ‘Categories and Induction in Young Children’, Cognition 23: 183–​209. Gopnik, Alison. 2012. ‘Scientific Thinking in Young Children: Theoretical Advances, Empirical Research and Policy Implications’, Science 337: 1623–​1627. Häggqvist, Sören and Wikforss, Åsa. 2007. ‘Externalism and A Posteriori Semantics’, Erkenntnis 67: 373–​386. Korman, Daniel. 2006. ‘What Externalists Should Say about Dry Earth’, The Journal of Philosophy 103: 503–​520. Kornblith, Hilary. 1980. ‘Referring to Artifacts’, The Philosophical Review 89: 109–​114. Kripke, Saul. 1980. Naming and Necessity, Princeton, NJ: Princeton University Press. Leslie, Sarah-​Jane. 2013. ‘Essence and Natural Kinds:  When Science Meets Preschooler Intuition’, in Tamar Gendler and John Hawthorne (eds.), Oxford Studies in Epistemology, Oxford: Oxford University Press. McKinsey, Michael. 1991. ‘Anti-​Individualism and Privileged Access’, Analysis 54: 9–​16. McLaughlin, Brian and Tye, Michael. 1998. ‘Is Content-​Externalism Compatible with Privileged Access?’, The Philosophical Review 107: 349–​380. Nazzi, Thierry, and Gopnik, Alison. 2000. ‘A Shift in Children’s Use of Perceptual and Causal Cues to Categorization’, Developmental Science 3: 389–​396. Needham, Paul. 2011. ‘Microessentialism: What Is the Argument?’, Noûs 45: 1–​21. Nuccetelli, S. 2003. ‘Knowing that One Knows What One Is Talking About’, in S. Nuccetelli (ed.), New Essays on Semantic Externalism and Self-​Knowledge, Cambridge, MA: MIT Press. Okasha, Samir. 2002. ‘Darwinian Metaphysics: Species and the Question of Essentialism’, Synthese 131: 191–​213. Putnam, Hilary. 1962. ‘The Analytic and the Synthetic’, in Mind, Language, and Reality: Philosophical Papers, Volume 2, 1975, Cambridge: Cambridge University Press. Putnam, Hilary. 1975. ‘The Meaning of “Meaning”’, Minnesota Studies in the Philosophy of Science 7: 131–​193. Putnam, Hilary. 1990. ‘Is Water Necessarily H2O?’ in J. Conant (ed.), Realism with a Human Face, Cambridge, MA: Harvard University Press. Rutherford, Ernest. 1911. ‘The Scattering of α and β Particles by Matter and the Structure of the Atom’, Philosophical Magazine 21: 669–​688. Salmon, Wesley. 1984. Scientific Explanation and the Causal Structure of the World, Princeton, NJ:  Princeton University Press. Schwartz, Stephen. 1978. ‘Putnam on Artifacts’, The Philosophical Review 87: 566–​574.

322

Pervasive Externalism Sobel, David, Yoachim, Caroline, Gopnik, Alison, Meltzoff, Andrew, and Blumenthal, Emily. 2007. ‘The Blicket Within: Preschoolers’ Inferences about Insides and Causes’, Journal of Cognitive Development 8: 159–​182. Walker, Caren, Lombrozo, Tania, Legare, Cristine, and Gopnik, Alison. 2014. ‘Explaining Prompts Children to Privilege Inductively Rich Properties’, Cognition 133: 343–​357. Weisberg, Michael. 2005. ‘Water is Not H2O’, in D. Baird et al. (eds.), Philosophy of Chemistry: Synthesis of a New Discipline, New York: Springer. Wikforss, Asa. 2005. ‘Naming Natural Kinds’, Synthese 145:  65–​87.

323

24 THEORETICAL IDENTITIES AS NECESSARY AND A PRIORI Joseph LaPorte

As I  understand ‘theoretical identity statement’, such a statement identifies an object with an essence. Here ‘object’ is broadly understood to include kinds or properties. Theoretical identity statements presented in the literature often concern chemical and biological natural kinds:  e.g., ‘Water  =  H2O’ and ‘Whale-​kind  =  the clade including P and all P’s descendants’, or alternatively, ‘Whale  =  P and all P’s descendants’, where P is a population.1 Some theoretical identity statements address concrete objects but I  will address just those about kinds in this entry. An example of a theoretical identity statement about a concrete object suggested in the literature would be ‘Cicero = the organism descended from sperm s and egg e’, where ‘s’ and ‘e’ are names for the right sperm and egg. Not all identity statements widely accepted as necessary specify a theoretically interesting essence in the relevant respect. ‘Water = aqua’, ‘Whale = Cetacea’, and ‘Cicero = Tully’ do not qualify. Each sentence contains two names for an object but no essence-​clarifying designator. After Kripke, all of the foregoing statements, both theoretical identity statements and identity statements containing just names, have come to be accepted as necessarily true if true at all. Correspondingly, designators on both sides of the identity sign are supposed to be rigid: roughly to designate the same object with respect to all possible worlds. Given that, say ‘whale’ and ‘Cetacea’ both designate the same kind, namely the kind whale, with respect to any given possible world w, the relevant identity statement will come out true with respect to w.

Discovering Essence A Posteriori According to a tradition made familiar by Kripke (1980) and Putnam (1975), scientists inquire into the nature of kinds that were named before much science was known. Scientists discover the essences of the kinds in question.When scientists have discovered enough about the essence of a kind, they can correct past speakers who supposed that certain objects did or did not have what it takes to belong to the kind. For example, past speakers called whales “fish”. Scientists have allegedly corrected this statement. Empirical research has resulted in scientists’ “discovering that ‘whales are mammals, not fish’ is a necessary truth” (Kripke 1980, 138). Kripke writes, “In general, science attempts, by investigating basic structural traits, to find the nature, and thus the essence (in the philosophical sense) of the kind” (1980, 138).

324

Theoretical Identities

The claim that there are a posteriori statements about essence captured the imagination of the philosophical community. It is “a conclusion that went against the assumption of centuries of philosophers”, as Linda Zagzebski relates (2017, 213). Putnam explains the “startling consequences”, for epistemology thus (1975, 232): Since Kant there has been a big split between philosophers who thought that all necessary truths were analytic and philosophers who thought that some necessary truths were synthetic a priori. But none of these philosophers thought that a (metaphysically) necessary truth could fail to be a priori. Putnam 1975, 233; see also Schwartz 2002, 270ff The necessary aposteriority in question generated much excitement. However, in the decades following Kripke’s exposition, important considerations have come to light to moderate enthusiasm for the necessary a posteriori, without impugning the importance of theoretical identities.

Direct Reference The theory of “direct reference” traces back most directly to Kaplan (1989a: see 1989b, 571). We could think of definite descriptions as contributing to a statement some sort of structured content specifying a single object, the designatum, by way of some property or set of properties belonging to the object: “the constituent of the proposition will be some sort of complex, constructed from various attributes by logical composition”. Now contrast that with a name. For Kaplan, “in the case of a singular term which is directly referential, the constituent of the proposition is just the object itself ” (Kaplan 1989a, 494). So according to the theory of direct reference, ‘Hesperus = Phosphorus’ expresses the same thing as ‘Hesperus = Hesperus’. What it expresses is therefore a priori and not after all a posteriori. Kripke initially proposed that the ancient Babylonians discovered a posteriori the truth of the proposition that Hesperus  =  Phosphorus but he later retreated to an agnostic position about whether that proposition is a posteriori or whether, as direct-​reference theorists maintain, it is after all a priori (1980, 21). Kripke concedes, even so, that the “spirit” of his views “suggests that a Millian line should be maintained as far as is feasible”, the Millian antidescriptive line being in harmony with direct reference (Kripke 1979, 248). Direct-​reference theorists face hard questions like, “If ‘Hesperus = Phosphorus’ expresses the same as ‘Hesperus = Hesperus’, how can the one be informative but not the other?” There’s no need to arbitrate between direct-​reference theorists and their opponents here. Rather, the point in raising the theory of direct reference is to point out that if the theory of direct reference is correct, then theoretical identities might be a priori after all, just as Hesperus = Phosphorus is. Thus Dowell and Sobel (2017) recognize pressure, from broadly Kripkean theories of reference, to say that “the proposition expressed by ‘Hesperus is Phosphorus’ is the very same as that expressed by ‘Hesperus is Hesperus’. By similar reasoning, the sentences ‘water is H2O’ and ‘water is water’ both express the same (necessary) proposition” (Dowell and Sobel 2017, 162). Dowell and Sobel’s suggestion is more plausible than might first appear to be the case. One might try to dismiss the view out of hand, on grounds that ‘H2O’ is allegedly descriptive, including as it does information about hydrogen and oxygen. If ‘H2O’ contains descriptive information, you might suppose, then it could not be a candidate for direct reference. But this seems too fast. Consider that ‘open-​hearted friendliness’ informs us about the referent by virtue of its compositionality, but it isn’t obviously functioning as a definite description because it does not obviously describe the designatum. Rather than serving as a definite description for the designatum, the expression seems to be something like a compositional name for the property open-​hearted friendliness, whose content is just that

325

Joseph LaPorte

property. Although ‘open-​hearted friendliness’ expresses what it does compositionally, nevertheless compositionality is not the same as having the content of a definite description, which is the hallmark of those expressions, so-​called “descriptive expressions”, typically contrasted with directly referential expressions (on the contrast see, e.g., Teresa Robertson 2009, 132). Perhaps we could say that ‘open-​ hearted friendliness’ provides us with something “like a picture of what’s named”, without denoting the property designated by describing it (to adapt an observation from Mark Richard 1993, 207). Something similar could be taken to account for the name-​like status of ‘H2O’ (for further discussion, see LaPorte 2013, 53–​55, 138–​139).

Apriority and Aposteriority after Direct Reference Part of Kripke’s work was to see distinctions where concepts had been run together before. The received tradition ran together the notions necessary, a priori, and some other notions like analytic:  “usually they’re not distinguished”, as Kripke reported (1971, 149). Kripke’s arguments indicated, at least initially, that necessity and apriority have to be distinguished. Now the theory of direct reference calls into question the necessary a posteriori. Are we back to where we started from before Kripke’s distinctions? No, we are not. Again, it would go beyond the limits of this entry to try to settle whether the theory of direct reference is true. Let us suppose it is. Even so, rather than return us to where we were before Kripke’s distinctions, the theory of direct reference invites us to refine further. Further distinctions will indicate that something consequential remains of aposteriority. The sentence or statement (interpreted sentence) ‘Hesperus  =  Phosphorus’ is necessarily true even though we can find this out only a posteriori (Fitch 2004, 110–​113; Hughes 2004, 106–​107; cf. Kripke 1980, 20–​21). The proposition that Hesperus  =  Phosphorus is a different matter. Direct-​reference theorists can grant that the statement ‘Hesperus = Phosphorus’ is a posteriori. Direct-​reference theorists might say that the relevant speaker is not in a position to know a priori whether ‘Hesperus = Phosphorus’, on its present interpretation, is true because, for direct-​reference theorists, this speaker would need astronomical investigation to determine that the sentence ‘Hesperus  =  Phosphorus’ expresses the proposition that Hesperus = Phosphorus, which proposition is known a priori to be true. Something similar would hold for theoretical identity statements like ‘water = H2O’ if they express a priori propositions by virtue of direct reference.2 The relevant aposteriority for statements, as opposed to propositions, is still interesting. Although perfectly competent speakers were unable to see, without astronomical observation, that the statement ‘Hesperus  =  Phosphorus’ is true, that statement is necessary, contrary to what seemed impossible to earlier generations of philosophers, who failed to distinguish between apriority, necessity, and analyticity. To them, it seemed apparent that the statement ‘Hesperus = Phosphorus’ is contingent; otherwise, like ‘Hesperus = Hesperus’, it would be a priori. As Kripke reported, “very few people, if any, distinguish between the concepts of statements being a priori and their being necessary” (1980, 34). So the necessary aposteriority of various statements expressing propositions knowable a priori, including statements about natural kinds, is appropriately surprising for someone with once-​ ubiquitous presuppositions.

Descriptivism After Putnam (1975) it is common to say that meanings are not “in the head”. Natural-​kind terms get their meaning and reference by way of causal baptisms: “Raccoon is that kind of animal” (the speaker points to a raccoon). Then it is up to empirical enquiry to determine what delimits raccoons as a group. So it would be an a posteriori matter (in the respect adumbrated in the foregoing section). Descriptivists maintain that a word like ‘raccoon’ gets hooked to a referent by the old-​fashioned account Kripke and Putnam made unpopular. Speakers associate a priori certain descriptive phrases 326

Theoretical Identities

with the word: has a bear-​like body, a long striped tail, and a mask over the eyes, etc. Whatever satisfies these descriptions is designated. Descriptivism, if successful, would vindicate the apriority of theoretical identity statements about kinds. But descriptivism faces well-​known criticisms from ignorance and error (see Häggqvist and Wikforss (2017) for a recent revival of descriptivism; Hoefer and Martí (2019) re-​issue concerns of ignorance and error). For one thing, not all raccoons have the foregoing traits. For another, speakers often don’t have enough information about raccoons to distinguish them from other animals. There are speakers who could recognize a raccoon face but who would be hard pressed to offer a distinguishing characteristic necessary and sufficient to be a raccoon. Furthermore, it does not seem to be a matter of definition that raccoons have the traits that would commonly be associated with them; on the contrary, it seems we could discover that these traits have been misattributed, in the same way that blueness is supposed to have been misattributed to blue jays. Descriptivism seems strongest for some terms that have been coined relatively recently, after the rise of modern science. Beebee and Sabbarton-​Leary (2010, 165), for example, argue that chemical names “are descriptors”. They give the example of ‘ununbium’ whose meaning is compositionally formed to convey that ununbium is the element with atomic number 112: un = 1; un = 1, bi = 2. Hence, “a chemist can come to know a priori” the truth of the theoretical identity statement ‘Ununbium = the element with atomic number 112’ (Beebee and Sabbarton-​Leary 2010, 165). There is no need to wait for any ununbium to be discovered or created in the lab, for there to be causal grounding and then a posteriori discovery of what the essence is; scientists know the essence before encountering the substance, according to Beebee and Sabbarton-​Leary. ‘Ununbium  =  the element with atomic number 112’ would seem to be analytic, at least by a reasonable understanding of ‘analytic’. Santana (2019) presents a similar example from mineralogy. Certain series of gradations can allow scientists to define according to a composition that, so far as we know, is uninstantiated: “merely theoretical”. So again, certain natural-​kind terms can be defined by their essence even without baptizing samples in the physical world. Even the most compelling descriptivism like that of Beebee and Sabbarton-​Leary has to face certain worries about the alleged a priori status of theoretical identity statements. It is not obvious that speakers would be incorrect to count a substance as ununbium were a proton to be replaced by an exotic subatomic particle, perhaps of antimatter. Thus, for example, scientists have managed to insert a little antimatter into certain elements and compounds, which do not thereby become some other chemical. But were ununbium to receive a surrogate proton, it need not, were the story to be developed right, qualify straightforwardly as having atomic number 112 because it would not contain 112 protons. In the same way, scientists could conceivably learn that elements in the periodic table can after all be converted to other elements by way of yet-​to-​be discovered chemical means instead of merely by way of physical means, so that these items are not after all chemical elements, at least in the intended sense. These discoveries would seem to belie the theoretical identity statement ‘Ununbium  =  the element with atomic number 112’, at least as real speakers actually use that sentence. One could stipulate hard and fast, come what may, that a term is to name the item with such and such essence, which would yield a priori knowledge of the nondefeasible truth of the relevant theoretical identity statement; but real speakers of natural and scientific languages do not seem ordinarily to stipulate so strictly. Perhaps these situations belying ‘Ununbium = the element with atomic number 112’ go contrary to physical laws and perhaps (as Bird would say (2007, 213)) lawful necessity is necessity in the strongest respect. If so, the theoretical identity statement in question is metaphysically necessary; but we would not know a priori, on these grounds, that it is necessary. Since it is not obvious that we could never learn that ‘Ununbium = the element with atomic number 112’ is false by learning more about the physical world, it is not obvious that this theoretical identity statement is true a priori, or in what qualified ways it might be. Even so, something of a descriptivist a priori case can be restored, despite worries that elements like ununbium might not 327

Joseph LaPorte

after all be elemental or might not have the expected atomic number. Perhaps a limited descriptivism could claim the following: ‘Ununbium’ is to designate the element with atomic number 112 provided that current physics and chemistry describe the physical world to a sufficient approximation (cf. Kripke 1980, 22 and Putnam 1975, 225 on similar “defeasible definitions”, as Putnam would call them). So if chemistry turns out to be on a wrong track about the elemental status of substances in the periodic table, say, then the “definition” of ununbium is to be canceled or defeated. But both the definition and the defeasibility are, arguably, knowable a priori. The details of what it takes to defeat the intended definition Ununbium = the element with atomic number 112 will be hard to specify beyond a sketchy approximation; perhaps we could recognize defeat when it happens but not articulate explicitly the conditions for defeat. This does not necessarily undermine apriority for a range of safe cases that could be recognized, a priori, as such.There is some reason to take recognitional ability to be a priori. If I can recognize how I would or would not apply a term under various relevant circumstances presented to me as epistemic possibilities, then it would seem I should be able to recognize the extension of the term, as I use it, a priori (there are affinities here with Chalmers’s thesis of the scrutability of reference: Chalmers 2012). At least we could say that it is a priori in this, perhaps limited, respect: that I would not have to consult a social scientist a posteriori, say, to find out my semantic commitments. Instead, I could rely on my own semantic intuitions as I judge by recognition how the term should be used in different epistemic circumstances that might be presented to me. Of course, I might recognize that I defer to experts; but that just pushes the obligation to experts, within the division of linguistic labor, to secure reference, by being able to recognize the relevant physical and socially coordinated circumstances for reference to take place.3

Paring Open Texture by Stipulation There is another account, besides descriptivism, by which theoretical identity statements come out defeasible and according to which their defeasible truth is knowable a priori in principle. According to that account, speakers adjust their use of terms over time, in order to reduce unclear application of the terms exposed by scientific investigation. On this view that open texture is honed away, scientific discovery teaches speakers a posteriori that their terms are too imprecise to secure any interesting theoretical identity statement. Stipulation then gives a more precise use of the relevant natural-​kind term, so that a theoretical identity statement indicating the relevant kind’s essence is known to be true (again, presumably defeasibly) a priori. Stipulation would result in a change of meaning. This goes contrary to the tradition from Kripke and Putnam: “scientific discoveries of species essence do not constitute a ‘change of meaning’; the possibility of such discoveries was part of the original enterprise” (Kripke 1980, 138; see also Putnam 1975, 224–​225). So past speakers did not speak a language in which ‘fish’ had a meaning, since-​ discarded, that would allow whales to be in the extension. Whales were never in the extension of ‘fish’: earlier speakers were just wrong about whether whales are fish, the story goes. Granted, as we now use ‘fish’, it seems incorrect to call whales, including porpoises, “fish”. But did former speakers use the word ‘fish’ with the same meaning we do, so that they were just mistaken to say that whales are “fish”?4 Boyd (1979, 395–​396) is an early voice suggesting, on the contrary, that refinement in the meaning of ‘fish’ has taken place, instead. For Boyd, it was “undetermined” (p. 396) whether, according to earlier speakers’ use of the sentence, ‘Whales are fish’ is true. We could have continued to apply the name ‘fish’ to whales “with equal justification” (p. 395). There is reason to suppose that Boyd is right that we could have continued to apply the name ‘fish’ to whales “with equal justification” or at least with sufficient justification. Earlier speakers called whales “fish” in full knowledge of substantial differences between whales and other “fish”. John 328

Theoretical Identities

Locke, for example, knew that “Porpoises have the warm Blood and Entrails of a Hog” (Essay (1975), III.vi.12, 447). Granted, such a realization eventually led speakers to change their use of ‘fish’ in order to keep the extension natural: they came to count salmon and eels together as “fish” to the exclusion of hogs and whales. But there were other alternatives than to alter their application of ‘fish’ in this way: speakers might instead have decided to count the “fish” an artificial assemblage. That would have allowed for the conservative conclusion that the extension of ‘fish’ refers, in accordance with precedent set by former application, to salmon, eels, and whales, but not swine. In that way, whales might have continued to be counted “fish”. Had that happened, experts would have said things like this by way of educating the laity: “The kind fish is not natural, since both whales and salmon are fish. From a biological perspective it is not helpful to recognize any group including just whales and gilled fish”. That the use of ‘fish’ was not settled in favor of either including or excluding whales, as science developed, so that in order to go one way or the other, a decision to refine had to be made stipulatively, is suggested by pressure speakers felt as science developed, to leave ‘fish’ to refer to whales along with salmon and eels. Tensions about how to use ‘fish’, after the affinities between whales and other mammals became clearer, suggest that the proper use of the term was unclear. In the eighteenth century Oliver Goldsmith testifies to pressures to leave ‘fish’ to apply to what speakers had been calling “fish”. Goldsmith wonders whether the whale is a “non-​fish” that had been misclassified by the general public and by earlier naturalists, or whether the whale is a bona fide member of the heterogeneous extension of ‘fish’. His own judgment was to conclude that the whale is a lofty type of “fish”: “it is best to let them rest in the station where the generality of mankind have assigned them; and as they have been willing to give them all from their abode the name of fishes, it is wisest in us to conform” (1791, vol. 7, 1). Other learned authors had taken the same tack, and would continue to do so. For example, the great naturalist John Ray had counted whales as “fish” about a century before Goldsmith, in order to be “in accordance with common usage” (Romero 2012, 23). In 1812, an American court case over the sale of “fish oil” honored ‘fish’’s “common and popular usage”, by which “the words ‘fish oils’ must be considered as embracing the oil of the whale, as well as that of any other fish” (DeCou 2018). As it happens, common use would afterward narrow to exclude whales. But this seems to have resulted from a choice to use ‘fish’ in one way rather than another, when neither alternative had been clearly right before. Interestingly, the question of whether whales are “fish” arose a second time in the twentieth century as classification became historical, reflecting the order in which groups have evolved. Again, something like stipulative meaning refinement has arguably occurred. The position that our terms are stipulatively adjusted in their use, after scientists disclose more about the unclarity of how those terms are to be used, could be raised more simply without appeal to correction. The first European speakers to baptize ‘raccoon’ might have pointed to a raccoon in North America and said, “raccoons are that kind of thing”. The question then arises whether ‘raccoon’ refers just to that species or whether, say, kinkajoos would count. Kinkajoos are members of the same family, although not the common raccoon species. Kinkajoos look like a cross between a cat and a weasel. They have a long prehensile tail by which they grab branches as they climb. They lack the distinguishing features of common raccoons, like a mask and striped tail. It seems hard to believe that when ‘raccoon’ was baptized, kinkajoos straightforwardly belonged to the extension of that term on any correct use of ‘raccoon’. Even so, they are now considered raccoons, broadly speaking. As the common raccoon is the best-​known species in the family, we recognize a “raccoon family”. Kinkajoos are members then of the “raccoon” family, just as lions and tigers are members of the cat family: apparently then the baptism of ‘raccoon’ managed to secure a name that would come to apply to kinkajoos, at least by one use. Kinkajoos are raccoons just as lions are cats though whether kinkajoos are “raccoons” might have been yet undetermined at the baptism. Suppose speakers had baptized raccoons as “that kind of species” (pointing) as opposed to “that kind of thing”. Such a baptismal pronouncement could have reduced the resulting unclarity about the proper extent of the named kind and its essence.Yet unclarity about the breadth of the essence of the 329

Joseph LaPorte

named kind would evidently have remained, because systematists are not agreed about the essence of species. Some authorities recognize up to about ten species, while other authorities recognize one species with different races or subspecies. Differences depend on criteria to which systematists appeal, in order to break organisms into broader groups or narrower ones. There are many similar examples for animals and plants (see Ludwig 2016, 1255–​1256). There are other reasons for the unclarity about species, too (Clarke and Okasha (2013) offer one helpful discussion). So the question as to what the essence is, of the baptized sample, depends on how broad our baptized species is. There is still unclarity about that matter, for raccoons; and it would seem to be overly optimistic to expect a discovery to resolve the unclarity in order to vindicate a theoretical identity statement of the form ‘Raccoon = the species with … essential characteristics’ that will have been discovered a posteriori to be true. Such a conclusion would probably be a choice with other options. Return to chemical kinds. Putnam says that were we to find a water-​like substance with a different chemical composition, we would recognize that it is not “water”. We could call it “twater”. We might dispute this with a similar example: jade. When earlier speakers found a new substance like the “jade” they had been familiar with, they still counted it “jade”, because of its similar properties (Hacking 2007; LaPorte 2004). As Hacking says, much of how speakers respond to scientific inquiry seems to be a matter of accident. Hacking suggests that “we have no idea what we would say, let alone should say, if a Twin-​Earth were ever to be discovered. If philosophers were to pay a little more attention to the real-​life historical contingencies of language use, they would be less inclined to construct zealously over-​confident arguments about what we would or should say, if ” (Hacking 2007, 275). Kripke (1980, 128–​129) takes chemical composition to be sufficient for the application of ‘water’. He suggests that if a new substance were found that had the chemical composition H2O but that had observable properties that were very different from those that we expect water to have, speakers would still call it “water”. Again, history suggests otherwise. After the discovery of the chemical composition of charcoal, scientists hardly expected the composition of this humble substance to be shared by any impressive material. But chemists were surprised by what they found after investigating diamond: the chemical composition was found to be exactly the same as that of charcoal. Still, speakers have not called diamond “charcoal”. Yet speakers might, and arguably with sufficiently similar fidelity to earlier use, have affirmed the theoretical identity statement ‘Diamond = C’, by which charcoal would count as “diamond”. There was some pressure to go in this direction. When the scientist Smithson Tennant first learned of the chemical affinity between diamond and charcoal, it seemed unclear whether he should say that diamonds are “charcoal in an unusual state”. Tennant himself concluded that diamond “consists entirely of charcoal, differing from the usual state of that substance only by its crystallized form” (Tennant 1797, 124). Language as a whole did not move to honor this use of ‘charcoal’, but it might easily have done so. The case of “water” seems similar. Speakers might not call a substance “water” even if it has the very same chemical composition, H2O. H2O that is unlike our familiar water might be said to be “non-​water”, with a similar degree of fidelity to our use of ‘water’ up to that point. That is not to say that it would clearly be non-​water. Again, it would be a vague case, and speakers might go either way. As it is, then, we have not confirmed the theoretical identity statement ‘Water = H2O’. Were we to affirm that sentence in order to clarify after discovering twater or H2O with the wrong properties, we would be refining use stipulatively, according to the thesis of open-​texture refinement.5

Other Kinds of Kinds and Other Versions of Apriority The literature on theoretical identity statements includes a discussion of many kinds of kinds. There has been increasing discussion over the years of human kinds, like woman, as well as institutional kinds like president, and the like. Theoretical identity statements about the essence of such kinds might be a priori for the same reason as theoretical identity statements about biological or chemical kinds. Thus, 330

Theoretical Identities

Engelhardt (2019, 1870) suggests a stipulative honing away of indeterminacy for various kinds from culture including “natural, legal, and social terms”. He elaborates, for example, on the case of law, for which a judge’s ruling sets a precedent that, in effect, helps to stipulate the proper use of kind terms at issue. The judge’s role in the economy of “linguistic labor on this view is to introduce a norm that constrains how a term can be appropriately applied” (Engelhardt 2019, 1866; for a similar view about “decisions as regards application or non-​application of terms”, see Martí and Ramírez-​Ludeña 2018). Theoretical identity statements about human kinds and institutional kinds and the like might be thought a priori for reasons unlike those already discussed in the foregoing, too. These kinds are often supposed to be in some way “socially constructed”. As Ron Mallon (2016, 192–​193) says, elaborating on the literature from Amie Thomasson and others, if “institutions emerge from collective acceptance”, then there’s a case that those engaging in the institutional practice have an a priori understanding of what it is to be the relevant institution: “members of that population will have a kind of a priori knowledge regarding the imposed properties” that bring about the existence of the institution (Mallon 2016, 193). A priori knowledge of what it is to be the relevant institution would generate a priori knowledge of the truth of the relevant theoretical identity statement indicating the institution’s essence.

Case by Case and Qualified Apriority that is Still Substantive The foregoing examples do not rule out the metaphysical possibility of discovering the truth of theoretical identity statements a posteriori, and such a possibility seems very hard to rule out. Indeed, the foregoing case for the apriority of some theoretical identity statements suggests only that in many actual cases, not necessarily all, what has been thought a matter of a posteriori discovery is not. On the other hand, it is interesting if many theoretical identity statements that are thought to be discovered to be true a posteriori are in fact known to be true a priori. That they are so has been suggested or conceded by many now, including some of the best-​known defenders of a Kripke-​Putnam line, broadly understood, about the a posteriori necessity of theoretical identity statements. So Robin Hendry acknowledges that claims like that of stipulative refinement “ought to be made out on a case-​by-​case basis” (Hendry 2015, 260; see also Hoefer and Martí 2019). So the foregoing case for apriority is case by case. Furthermore, the apriority at issue demands qualification, as already suggested in the discussion of descriptivism. Suppose we stipulate that whales are not to be counted as what we will henceforth call “fish”, rather than discover a posteriori that whales are not what we have all along called “fish”. Stipulation, or what roughly comes to stipulation in the refinement of natural language, would then seem to provide us a priori access to information that, as we use the term ‘fish’, whales do not have the right essence to belong in the extension of ‘fish’ with its newly refined meaning. But we would probably stipulate defeasibly, so that the stipulation would be issued with the condition that it be canceled if the scientific information we presuppose as we stipulate a refined use turns out to be wildly mistaken: e.g., if whales turn out to be cold blooded and gilled relatives of salmon and their supposed lungs turn out to be a different organ used for something other than breathing. Dramatic misunderstanding of the world like this would defeat our motion to refine the use of a term, at least barring artificially strictly enacted stipulation that would be unusual for real speakers of natural and scientific languages.This calls for qualification of the apriority of a theoretical identity statement. It can be known a priori to be true defeasibly: on the defeasible assumption that dubbers are roughly right about scientific information presupposed as a condition for their stipulating successfully from the start. The apriority at issue, though case by case and qualified, seems interesting. Most obviously, perhaps, the stipulation raises Kuhnian worries about whether past and present speakers talk past each other. The Kripke-​Putnam thesis that meanings for theoretical expressions remain constant seems to put to rest Kuhnian worries that speakers before current science spoke incommensurably of different things than later speakers, using the same words. Take, for example, ‘planet’. For Kuhn, the revision 331

Joseph LaPorte

in our use of ‘planet’ began with Copernicus, after whom speakers “were changing the meaning of ‘planet’ so that it could continue to make useful distinctions” (Kuhn 1970, 128). Accordingly, the usual worries arise that past and present speakers talk past each other so that scientific change does not mark genuine advancement. Whether or not Kuhn is right about a change in ‘planet’’s meaning centuries ago or the reasons for it, a plausible case can be marshaled that ‘planet’ has been stipulatively redefined in recent times, in light of unclear cases, especially Pluto. As Bokulich (2014) explains, scientists who vote on a definition of ‘planet’, including those who voted at a convention determining the fate of Pluto, have different options available to them, in selecting a definition: which of the “options scientists choose to adopt is largely a matter of convention”, Bokulich says; but because Pluto exposed unwelcome unclarity in the extension of ‘planet’, “what was not an option for the scientists was to leave the traditional definition and extension of the term planet intact, while having it remain a scientifically useful concept” (Bokulich 2014, 488). The usual Kuhnian worries suggest themselves. If we stipulatively revise our use of ‘planet’, “how can we make sense of genuine learning, rather than changing the subject?”, as Leech worries (2013, 256). There are also broadly Quinean issues that arise from the stipulative refinement, the defeasibility, and the theoretical disruption at issue in the foregoing sections. Can a distinction be made between theory change and meaning change, given that there seems to have been both theory change and meaning change in salient cases of conceptual change issuing in stipulation? Quine would recognize no distinction between our learning that whales are not “fish”, on the one hand, and our changing meanings by stipulating “whales are not fish”, on the other hand. This is just the sort of example Quineans tend to raise, to dispute any distinction. These broad Kuhnian and Quinean worries cannot be addressed here but they mark some of the interest in much of the apriority articulated in this paper.6

Acknowledgment I thank the editors for helpful criticism that resulted in improvements to this paper.

Further Reading Beebee, Helen and Sabbarton-​Leary, Nigel. 2010. The Semantics and Metaphysics of Natural Kinds. London: Routledge. Boyd, Richard. 1988. “How to Be a Moral Realist”. In Essays on Moral Realism, edited by G. Sayre-​McCord, pp. 181–​228. Ithaca, NY: Cornell University Press. Donnellan, Keith. 1983. “Kripke and Putnam on Natural Kind Terms”. In Knowledge and Mind, ed. C. Ginet and S. Shoemaker, pp. 84–​104. New York: Oxford University Press. Dupré, John. 1993. The Disorder of Things:  Metaphysical Foundations of the Disunity of Science. Cambridge, MA: Harvard University Press. Putnam, Hilary. 1992. “Is it Necessary that Water is H2O?” In The Philosophy of A.  J. Ayer, ed. L. E. Hahn, pp. 429–​454. La Salle, IL: Open Court. Stanford, P. Kyle and Kitcher, Philip. 2000. “Refining the Causal Theory of Reference for Natural Kind Terms”. Philosophical Studies 97: 97–​127.

Notes 1 There are venerable traditions according to which there can be no identity sign flanked by terms for properties. I discuss accommodations for purely predicative characterizations of “theoretical identities” that eschew the identity sign elsewhere (LaPorte 2013, ch. 5). 2 There are other strategies besides distinguishing between propositions and statements: for one general strategy, see Geirsson (2002). 3 See Miller (1992), on the one hand, and Devitt and Sterelny (1999) on the other hand for a comparison between a recognition theory and descriptivism and their connection to causal grounding. 4 I use double quotation marks here so as not to commit to using ‘fish’ as we ordinarily use it: instead, I use the term as earlier, less scientifically sophisticated speakers used it.

332

Theoretical Identities 5 Doubt is sometimes expressed over whether stipulation issues an a priori understanding (e.g., Horowitz 1983). There are other controversies.The account offered here is broadly Carnapian in recognizing meaning change. For Carnap, ‘fish’’s narrowing of use would represent a change of meaning, although a slight one and a useful one. However, Carnap presupposes a sort of descriptivism abandoned after Kripke, and not assumed here. Biggs and Wilson, who also take inspiration from Carnap (Biggs and Wilson 2016, Appendix A; 2019) take a position according to which there is no meaning change. Rather than to refine by replacing an earlier use of a term by a similar one with a slightly altered meaning to broaden or narrow a formerly unclear extension, we instead choose between candidate intensions to narrow our application of a term without altering its meaning, by accepting the intension with the best explanatory credentials. It may be more elegant, simpler, and more compatible with maintaining the truth of speakers’ beliefs over time, to have ‘fish’ as a term for today’s “fish” rather than a term for today’s “fish” plus whales. Like stipulation, the position embraced in the text here, the abductive process could be defended as broadly a priori (Biggs and Wilson 2017; see also 2020). The process articulated by Biggs and Wilson suggests more optimism than the one articulated in the text here: optimism that speakers would typically refine how to use a term without happenstance, by rational criteria selecting a candidate as uniquely correct. Biggs and Wilson grant that if the proper use of a term is to be determined by intuitions, or what we conceive about a kind whose open texture has been exposed, then the different possible resolutions of a term’s use “will render merely verbal what are intuitively substantive disagreements about the extension of the expression” (2019).That indicates that resolution on one or another precisification, after open texture is exposed, might change a term’s meaning. Whether or not abductive criteria solve this problem seems to depend on whether or not criteria like fruitfulness and consistency with speakers’ beliefs over time determine a unique resolution for a term’s sharpened use after open texture is exposed, and whether or not speakers are generally successful at pinpointing that unique resolution. The position in the text would indicate that happenstance is more common than pinpointing a unique proper use, in cases where speakers resolve indeterminacy. However, the text concedes that there is more or less happenstance, and so more or less strength for the position that there has been meaning change, depending on the case at hand: see below in the text. 6 I address the suggested worries from Kuhnian and Quinean traditions in LaPorte (2004, chs. 5 and 6 respectively).

References Beebee, Helen and Sabbarton-​Leary, Nigel. 2010.“On the Abuse of the Necessary A Posteriori”. In The Semantics and Metaphysics of Natural Kinds, ed. H. Beebee and N. Sabbarton-​Leary, pp. 159–​178. London: Routledge. Biggs, Stephen and Wilson, Jessica. 2016. “Carnap, the Necessary A Posteriori, and Metaphysical Anti-​realism”. In Ontology after Carnap, ed. Stephan Blatti and Sandra Lapointe, pp. 81–​104. Oxford: Oxford University Press. Biggs, Stephen and Wilson, Jessica. 2017. “The a Priority of Abduction”. Philosophical Studies 174:  735–​758. DOI: 10.1007/​s11098-​016-​0705-​4. Biggs, Stephen and Wilson, Jessica. 2019. “Abduction versus Conceiving in Modal Epistemology”. Synthese. https://​doi.org/​10.1007/​s11229-​019-​02117-​9. Biggs, Stephen and Wilson, Jessica. 2020. “Abductive Two-​Dimensionalism:  A New Route to the A  Priori Identification of Necessary Truths”. Synthese 197: 59–​93. Bird, Alexander. 2007. Nature’s Metaphysics. Oxford: Oxford University Press. Bokulich, Alisa. 2014. “Pluto and the ‘Planet Problem’:  Folk Concepts and Natural Kinds in Astronomy”. Perspectives on Science 22: 464–​490. DOI: 10.1162/​POSC_​a_​00146. Boyd, Richard. 1979. “Metaphor and Theory Change: What is ‘Metaphor’ a Metaphor For?” In Metaphor and Thought, ed. A. Ortony, pp. 356–​408. Cambridge: Cambridge University Press. Chalmers, David. 2012. Constructing the World. Oxford: Oxford University Press. Clarke, Ellen and Okasha, Samir. 2013. “Species and Organisms”. In From Groups to Individuals: Evolution and Emerging Individuality, ed. F. Bouchard and Philippe Huneman, pp. 55–​76. Cambridge MA: MIT Press. DeCou, Christopher. 2018. “When Whales Were Fish”. Lateral. www.lateralmag.com/​articles/​issue-​29/​ when-​whales-​were-​fish. Devitt, Michael and Sterelny, Kim. 1999. Languages and Reality: An Introduction to the Philosophy of Language. 2nd edn. Cambridge, MA: MIT Press. Dowell, J. L. and Sobel, David. 2017. “Advice for Non-​Analytical Naturalists”. In Reading Parfit, ed. Simon Kirchen, pp. 153–​171. New York: Routledge. Engelhardt, Jeff. 2019. “Linguistic Labor and its Division”. Philosophical Studies 176: 1855–​1871. https://​doi.org/​ 10.1007/​s11098-​018-​1099-​2. Fitch, G. W. 2004. Saul Kripke. Chesham: Acumen. Geirsson, Heimir. 2002. “Justification and Ways of Believing”. Disputatio 1: 43–​53.

333

Joseph LaPorte Goldsmith, Oliver. 1791. An History of the Earth and Animated Nature, 8 vols. London: F. Wingrave. Hacking, Ian. 2007. “The Contingencies of Ambiguity”. Analysis 67: 269–​277. Häggqvist, Sören and Wikforss, Åsa. 2017. “Natural Kinds and Natural Kind Terms: Myth and Reality”. The British Journal for the Philosophy of Science 69: 911–​933. http://​dx.doi.org/​10.1093/​bjps/​axw041. Hendry, Robin Findlay. 2015. “Are Chemical Kinds Natural Kinds?” In Recent Developments in the Philosophy of Science: EPSA13 Helsinki, ed. U. Maki et al., pp. 251–​261. London: Springer. Hoefer, Carl and Martí, Genoveva. 2019. “Water has a Microstructural Essence after All”. European Journal for Philosophy of Science 9: 1–​23. https://​doi.org/​10.1007/​s13194-​018-​0236-​2. Horowitz, Tamara. 1983. “Stipulation and Epistemological Privilege”. Philosophical Studies 44: 305–​318. Hughes, Christopher. 2004. Kripke: Names, Necessity, and Identity. Oxford: Clarendon. Kaplan, David. 1989a. “Demonstratives”. In Themes from Kaplan, ed. J. Almog, H. Wettstein, and J. Perry, pp. 481–​563. Oxford: Oxford University Press. Kaplan, David. 1989b. “Afterthoughts”. In Themes from Kaplan, ed. J. Almog, H. Wettstein, and J. Perry, pp. 565–​ 614. Oxford: Oxford University Press. Kripke, Saul. 1971. “Identity and Necessity”. In Identity and Individuation, ed. M. K. Munitz, 135–​ 164. New York: New York University Press. Kripke, Saul. 1979.“A Puzzle about Belief ”. In Meaning and Use, ed.A. Margalit, pp. 239–​283. Dordrecht: Reidel. Kripke, Saul. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press. Kuhn, Thomas S. 1970. The Structure of Scientific Revolutions. 2nd edn. Chicago, IL: University of Chicago Press. LaPorte, Joseph. 2004. Natural Kinds and Conceptual Change. Cambridge: Cambridge University Press. LaPorte, Joseph. 2013. Rigid Designation and Theoretical Identities. Oxford: Oxford University Press. Leech, Jessica. 2013. Review of The Semantics and Metaphysics of Natural Kinds, ed. H. Beebee and N. Sabbarton-​ Leary. Mind 122: 253–​257. Locke, John. 1975. An Essay Concerning Human Understanding, ed. with a Foreword by Peter Nidditch. Oxford: Oxford University Press. Ludwig, David. 2016.“Ontological Choices and the Value-​Free Ideal”. Erkenntnis 81: 1253–​1272. DOI: 10.1007/​ s10670-​015-​9793-​3. Mallon, Ron. 2016. The Construction of Human Kinds. Oxford: Oxford University Press. Martí, Genoveva and Ramírez-​Ludeña, Lorena. 2018. “Tolerance, flexibility and the application of kind terms”. Synthese. https://​doi.org/​10.1007/​s11229-​018-​01898-​9. Miller, Richard B. 1992.“A Purely Causal Solution to One of the Qua Problems”. Australasian Journal of Philosophy 70: 425–​434. Putnam, Hilary. 1975. “The Meaning of ‘Meaning’”. In Mind, Language and Reality: Philosophical Papers, vol. 2, pp. 215–​271. Cambridge: Cambridge University Press. Richard, Mark. 1993. “Articulated Terms”. Philosophical Perspectives 7: 207–​230. Robertson, Teresa. 2009. “Essentialism and Reference to Kinds: Three Issues in Penelope Mackie’s How Things Might Have Been: Individuals, Kinds, and Essential Properties”. Philosophical Books 50: 125–​141. Romero, Aldemaro Jr. 2012.“When Whales Became Mammals: The Scientific Journey of Cetaceans from Fish to Mammals in the History of Science”. CUNY Academic Works. https://​academicworks.cuny.edu/​bb_​pubs/​371. Santana, Carlos. 2019. “Mineral Misbehavior: Why Mineralogists Don’t Deal in Natural Kinds”. Foundations of Chemistry. https://​doi.org/​10.1007/​s10698-​019-​09338-​3. Schwartz, Steven P. 2002. “Kinds, General Terms, and Rigidity:  A Reply to LaPorte”. Philosophical Studies 109: 265–​277. Tennant, Smithson. 1797. “On the Nature of the Diamond”. Philosophical Transactions of the Royal Society of London 87(97): 123–​127. Zagzebski, Linda. 2017. Exemplarist Moral Theory. Oxford: Oxford University Press.

334

25 THE NEED FOR DESCRIPTIVISM Sören Häggqvist and Åsa Wikforss

1  Introduction Ever since Naming and Necessity (Kripke 1980), it has been widely held that natural kind terms (hereafter “NKTs”) constitute a distinct semantic category of terms. Kripke’s basic idea is that NKTs, unlike other kind terms, function just like proper names and should be given a Millian semantics. They are rigid designators and they lack all descriptive content: their semantic content is exhausted by the term’s extension (across possible worlds), just like reference is held to exhaust the semantic content of proper names. This means that some “underlying property” said to be picked out by an NKT provides both necessary and sufficient conditions for the term’s application. Kripke combined the Millian semantics with an externalist metasemantics. On this view, NKTs latch on to the underlying, essential property through an initial baptism. A scientifically ignorant speaker can do this through ostension, pointing to a set of items while intending the term to pick out the underlying, essential kind “instantiated by most or all of those items” (Kripke 1980, 135–​136). When people argue for the need for descriptivism, they typically appeal to the need to provide a semantics that accounts for the cognitive role of these terms: the role they play in reasoning and action. There are well known difficulties here, having to do with cases where a term is empty or where there appear to be meaning differences not accounted for by the Millian. As a result of these and related concerns, a number of proposals have been put forth that try to combine the basic Millian idea with some descriptivist component. There are now several versions of neo-​descriptivist theories which seek to accommodate Kripke’s claim either by adding such components or by invoking rigidified descriptions. While considerations to do with cognitive role certainly pose difficulties for a purely Millian account, we shall set them aside here and focus on what we take to be prior, foundational challenges to the Millian theory—​challenges deriving from the philosophy of science and from recent experimental data concerning speaker usage. Semantic theories are empirical theories: they have to mesh with evidence about how terms are (and have been) used. Moreover, a semantic theory has to cohere with what is true, as far as the relevant sciences inform us, about the worldly relata or referents: in the case of NKTs, the categories identified by various scientific disciplines. We shall argue that these considerations, taken together, suggest the need for a more thoroughgoing descriptivism than what has been proposed by the neo-​descriptivist theories—​a version of the traditional cluster theory.

335

Sören Häggqvist and Åsa Wikforss

The plan is as follows. First, we briefly address the question of which expressions are NKTs. This question is essential if one accepts semantic exceptionalism, the idea that the NKTs have a special semantics, as both Kripke and the neo-​descriptivists do. Next, we note that what a Millian semantics requires, namely something that unifies a kind across possible worlds, is not plausibly attributed to many kinds figuring centrally in the sciences. (Here we draw on Häggqvist and Wikforss (2018).) We then review recent empirical research on lay speakers’ usage, focusing on a number of studies investigating case judgements involving NKTs. Along the way we briefly indicate how a cluster descriptivist theory is compatible with extant data.

2  Which are the Natural Kind Terms? Philosophers of language have tended to answer this question by listing instances. Most such lists include the names of biological taxa at the level of species (or taxa taken to be species), chemical substances and elements, and—​echoing Kripke 1980—​terms for natural phenomena such as heat or light. If semantic exceptionalism is true, however, it should surely be possible to characterize the terms in question in some other way than by merely listing examples. Soames (2002, 246) speculates that many readers of Kripke (1980) simply, and despite Kripke’s explicit reservations, have taken NKTs to be names of kinds, thus assimilating them to singular terms. This is probably right. But although they often serve grammatically as names, their primary function, as Soames argues, is as predicates or general terms.The exceptionalist thesis thus concerns a putatively separable class of predicates. But just which?1 The driving idea is easily stated: the NKTs are just those that pick out, or purport to pick out, categories united independently of our parochial and often ignorant ways of sorting things together on the basis of perceived, often superficial, similarities; in short, independently of us. If NKTs are simply those thought or intended to designate natural kinds, this leaves some elasticity between the metaphysics of kinds and the putative semantic category, and might allow identification of something as a NKT without much knowledge about the category (if any) it purports to pick out. But on an externalist metasemantics, what is held to determine the extension of a term is an intention to pick out a unique kind in conjunction with a typically unknown specific sameness maker shared by an ostended (or otherwise selected) sample and the rest of the kind. This line threatens to render many NKTs empty, since it imposes a uniqueness condition which is neither guaranteed to be met by the world, nor always actually met. It is therefore problematic from the start. A further empirical question is whether the requisite speaker intentions in fact exist. We will consider this below (section 4). The narrower exceptionalist option of taking NKTs to be just those which pick out natural kinds threatens the problem (discussed in Häggqvist and Wikforss 2007) of making the purportedly special semantic category of NKTs inscrutable without substantive scientific knowledge. Moreover, scientific findings can only help in establishing whether members of a term’s extension meet antecedently imposed criteria for natural kind-​hood (such as having a unifying, “superexplanatory” essence; cf. Godman et al. (2020)). These criteria are not deliverances of science. And in view of the ever-​ fragmenting philosophical discussion about the metaphysics of kinds, prospects for learning this look bleak. Is light a natural kind (despite scientists counting as light only electromagnetic radiation in the visible spectrum, hence contrary to Kripke such radiation as causes in us the sensation of light (Gray 2006))? Is diamond?2 Ian Hacking surveyed the range of philosophical accounts of natural kinds and found “a slew of distinct analyses directed at unrelated projects” (Hacking 2007, 203). Ludwig (2018) notes that Hacking’s admonitions didn’t stop philosophers of science from producing new accounts; far from converging, however, these—​e.g. Khalidi (2018), Slater (2015), Franklin-​Hall (2015), Magnus (2011), Martínez (2015)—​add further fragmentation:  as Ludwig notes, “this diversity actually reinforces Hacking’s worry that general notions of natural kind have become an obstacle for understanding classificatory practices” (Ludwig 2018, 32). 336

The Need for Descriptivism

These proposals are not futile: they valuably pinpoint and draw on interesting ways categorizations actually go in science. But they do not point towards any catholic, general account of natural kind-​ hood. As many, including Hacking and Ludwig, have pointed out, one plausible reason why such an account isn’t forthcoming is that natural kinds have been supposed to accomplish many different philosophical tasks, and that only recently are we becoming aware of the infeasibility of multi-​tasking. Some such tasks are epistemic, some semantic, others metaphysical. Consider projectibility. It is hardly sufficient for natural kind-​hood.3 Is it necessary? Well, it comes in degrees. And indeed, most current general accounts of natural kinds like those mentioned above take naturalness to be a matter of degree, as suggested by Boyd’s notion of homeostatic property clusters.4 But a gradable notion of natural kinds does not sit at all well with the idea that the NKT form a special class of terms with distinctive semantic or metasemantic properties. For the thoroughgoing descriptivist, however, these problems simply do not arise. She may grant that some terms play a (or several) special roles in science, without being semantically special (see Wikforss 2010). On a cluster theory, for instance, all predicates have the same semantics (with definable terms as a limiting case where necessary and sufficient conditions can be stated).

3  The Untenability of Essentialism Ever since Kripke (1980) and Putnam (1975) popularized the idea that science has shown how water is essentially H2O, philosophers have treated the example as a paradigm of empirically identified essences. The claim itself has survived debates over the provenance of its presumed modal status (Salmon 1981; Haukioja 2015; Nimtz 2018)  and the purported rigidity of NKTs (upshot:  such terms are neither singular (Needham 2012) nor rigid (Soames 2002)).The notion of H2O as a model essence has itself survived widespread acknowledgement that biological species do not have intrinsic essences (pace Devitt 2008). If we take “H2O” to predicate an essence of water, however, problems abound. As standardly used in chemistry, this indicates what water is composed of, hydrogen and oxygen, and their molar proportions, 2:1. But composition is not a tenable candidate for essences, even if we restrict ourselves to chemical substances. Some substances are distinct but share composition (structural isomers). These may be distinguished by structural formulas: ethyl alcohol and dimethyl ether are composed of carbon, hydrogen, and oxygen in the same proportions (C2H6O) but the former’s structural formula is C2H5OH while the latter’s is (CH3)2O. The idea that structural formulas like these, or H-​O-​H, represent essences of chemical kinds is, however, implausible. The most reasonable interpretation of the proposal that structural, but not compositional, formulas express essential properties seems to be that water necessarily is a variably vast collection of H-​O-​H molecules, and that being such a collection is also sufficient for being water. But unlike some substances, for instance methane (see Needham 2011, 10), water is not actually (let alone necessarily) such a collection of individual particles. If it were, its freezing and boiling points would be much higher, like methane’s. Obviously, there is structure beneath composition in substances, which partly explains many of their characteristic features. The conductivity of water, its relatively high freezing and boiling points (compared with compounds of similar molecular weight such as methane) and the relative density of liquid water and ice are all explained by various aspects of what might reasonably be called microstructure: thus, e.g., conductivity in liquid water is due to the proton transfer afforded by H-​bonded networks of various ionic species, although the finer details remain partly unknown, as does the precise nature of the (various species of) H-​bond itself (Needham 2013). Pace Hoefer and Martí (2019), however, these microstructural details don’t follow from water’s being “composed predominantly of H2O molecules” (as it in fact isn’t in the liquid and solid phases); nor is such detail given by saying that “water is the substance formed by bringing together H2O molecules and allowing them to interact spontaneously” (Hendry 2006, 872)—​as Needham points out, “this is just to say that the molecular 337

Sören Häggqvist and Åsa Wikforss

character of the gas phase at low pressure disappears on condensation; it doesn’t actually give the microstructure of liquid water” (2011, 16). Water is thus a kind with very many, extremely complex, and partly unknown microstructures. Identifying the substance doesn’t require prior knowledge of such structures; investigation of the detailed structures does, however, require prior identification of the substance (Needham 2011). This is afforded by its characteristic macroscopic (not to be confused with “observational”) properties such as boiling, triple and inversion of density points.5 In addition, substances are individuated by the chemical reactions they enter into. The fact that we appeal to microstructural properties like molecular geometry to explain the differences in reactivity between, say, various allotropes of phosphorus or the two stereoisomers of thalidomide (Bursten 2014) doesn’t support microessentialism: these substances were distinguished prior to such explanation, and the fact that microstructural properties may be incorporated into the descriptions characterizing them doesn’t suggest that such properties dictate kind individuation independently of human intervention—​not even by dint of somehow being “aimed at” by scientifically ignorant speakers. The negotiability of natural kind individuation with respect to various candidate microstructural properties was emphasized at length in LaPorte (2004). One of LaPorte’s cases concerns the contrast between ‘topaz’ and ‘ruby’ (2004, 100–​ 102). ‘Topaz’ was originally used for a yellow gemstone, but came to include rose and blue samples with the same basic composition (Al2SiO4(F,OH)2); ‘ruby’, by contrast, kept being reserved for red samples of (impure) Al2O3 even after the discovery of blue, yellow, and pink instances with the same basic composition. Bursten (2014) notes that the colour variations are due to microstructural differences (minute presences of chromium, small aberrations in internuclear distances), and that the two terms may be seen as located at different taxonomic levels, with ‘ruby’ occurring at a finer-​g rained chemical level than ‘topaz’. However, although the distinction between ruby and non-​ruby Al2O3 can be drawn in microstructural terms, the localization of ‘ruby’ at a lower taxonomic shelf than ‘topaz’ was effected by humans, not nature.This confirms LaPorte’s main point. Although he may have used the term ‘chemical structure’ somewhat carelessly (following usage in the literature), his moral stands:  the world doesn’t supply a unique referent/​extension, and speaker intentions at early or introductory uses do not dictate further refinement of usage in the face of unforeseen empirical discoveries. The point isn’t that, as Putnam put it, it “may take an indeterminate amount of scientific investigation to determine” what the extension-​determining sameness relations are (Putnam 1975, 142). It’s that there just is no select such relation onto which introducers’ intentions can fasten.6 This undermines the view that even very flexible speaker intentions determine what is an essence of a kind dubbed before much knowledge about it is acquired (Haukioja 2015; Law 2015; cf. Häggqvist and Wikforss 2018). And at any rate, what Millianism requires are de re essences possession of which determines membership in a unique objective category.7 Apart from the problem we have discussed so far of finding candidates for such essences, a further problem for natural kind essentialism is that natural kinds do not in general (unlike biological taxa, for which intrinsic essences are hopeless anyway) form a hierarchy, but sometimes crosscut each other (Khalidi 1998). A neat illustration is afforded by the periodic table (Hacking 2007). Any sample of a given element will belong not just to that element, but to its group of elements (a column in the table) and to its block of elements (a row). But other elements in the same group need not belong to the same block, or vice versa. So the sample itself will not determine a unique kind to which it belongs. Nor will it determine a series of hierarchically nested kinds. It is therefore insufficient to try to counter the qua problem (Devitt and Sterelny 1987) by invoking an intended “level of taxonomy” at which the sample’s properties are to determine a kind. Blocks and groups are not nested as species and genera. But even a mild scientific realism surely recognizes these kinds as real. One could opt for a more austere notion of natural kinds in order to salvage the hierarchy thesis. But to do so only 338

The Need for Descriptivism

in order to avoid the qua problem would seem more like revisionary metaphysics than empirical semantic theorizing (cf. Havstad 2018, 2020). It is true that some scientifically important kinds afford characterization in terms of relatively brief statements of necessary and sufficient conditions. For chemical substances like water, such conditions may, as we saw, be stated in terms of specific macroscopic properties for purposes of chemistry, although even there, different contexts will select different criteria (determining whether D2O counts as water, whether different spin isomers count as the same or different substances, etc.). Such cases obviously do not impugn descriptivism. Of course, succinct “definitions” of terms are desirable, if available. Lay talk of water does not afford such logical precision—​whether speakers accept a substance as water is a more complex matter (cf. Malt 1994; Strevens 2019; and section 4 below). But a cluster theory may allow that which properties weigh most heavily varies with context—​even for short-​ lived contexts. Such a theory seems capable of handling fickle lay usage. It also accommodates the fact that important types of kinds have typically been construed in rival ways both simultaneously and diachronically—​compare ‘gene’ and ‘species’. The existence of incompatible criteria or definitions, sometimes based on mistaken empirical assumptions, does not prevent communication or understanding. This is particularly important in connection with the accusation that descriptivism imperils stable reference across theory change and with it scientific realism. Unlike Aristotle, we don’t now take liquidity to be definitive of water.Yet we understand what Aristotle referred to. We don’t take phlogiston, as defined (by Stahl) to be that which is emitted in combustion, to exist. Yet we can understand that Priestley succeeded in referring to oxygen with some tokens of ‘dephlogisticated air’, while others failed to refer at all (Kitcher 1978). A cluster theory should construe this difference as different contexts imposing different weights on different properties—​the property of being the result of removing Stahl’s suggested definiens for ‘phlogiston’ from air, versus the property of being a gas emitted when the results of burning mercury are heated.8 Hence, a cluster theory should acknowledge that both the weights of various descriptions/​properties and what belongs to the cluster at all may vary across different occasions—​and certainly over extended time.This will obviously not render reference completely stable. But that is as it should be; we credit Kitcher with the crucial insight that referential variation doesn’t entail incommensurability or incomprehensibility.

4  Evidence from Ordinary Usage Whatever precisely a natural kind is in general, however, and whether such kinds have essences or not, Millianism requires speakers to comport with its prediction that they believe such essences to characterize natural kinds, and to determine the extension of the corresponding terms. In this section we survey the empirical findings regarding this question. Psychologists were the first to examine the Millian theory experimentally. Braisby, Franks, and Hampton (1996) tested the claim that essential properties determine reference independently of people’s beliefs.The experiments test the extent to which people allow underlying features to govern their use of natural kind terms such as ‘cat’, ‘water’, ‘tiger’, and ‘gold’, and the conclusion is that speaker usage does not support the Kripke-​Putnam view. Although in some scenarios the majority (73 per cent) took underlying features to play an important role, a substantial minority did not, and the experiments do not support the conclusion that only underlying features play a decisive role (as required by the Millian position). Inspired by this experiment, Jylkkä, Railo, and Haukioja (2009; henceforth “JRH”) designed their own experiments, testing what they call “externalistic essentialism”, i.e. the hypothesis that “speakers take possessing E [the essence] to be a necessary and sufficient condition for belonging in the extension of [kind term] C” (2009, 41).This suggests they intend to test Millianism. Using a set of scenarios involving six fictive natural kinds the experiments test whether subjects take underlying properties to 339

Sören Häggqvist and Åsa Wikforss

be decisive. The authors conclude that externalism was largely confirmed: 69 per cent of the answers were externalistic (out of which 33 per cent gave purely externalistic answers), 28 per cent were internalistic, and 3 per cent were compromises (Jylkkä, Railo, and Haukioja 2009, 52). As we have argued elsewhere (Häggqvist and Wikforss 2015) it is far from clear that JRH’s data support externalistic essentialism. First, there is a need to explain the substantial minority who gave internalist answers (as recognized by JRH). Second, the cluster theory would seem to provide a better explanation of the mixed results of the experiments since it implies that whether a sample falls in the extension of the term depends on whether it fits the weighted majority of the associated descriptions. The cluster theory is not committed to “superficialism”, to the idea that only observable properties belong to the cluster of descriptions. This means that both macro-​and micro-​level properties may matter, and that the intuitions about twin cases depends on how individual speakers weigh these properties. Third, the experiments do not in fact test the sufficiency claim. Twin cases concern whether underlying properties matter, which speaks to the necessity claim, but not whether they are the only thing that matters, whether they provide sufficient conditions. To test the sufficiency claim one would need a scenario, such as Kripke’s tiger-​reptile case, in which the appearance properties are completely different but the underlying properties the same (Häggqvist and Wikforss 2015, 125). Haukioja, Nyquist, and Jylkkä (2020; henceforth “HNJ”) provide the first attempt within experimental philosophy to test the sufficiency claim. They present results from three experiments on ordinary speakers’ usage of natural kind terms, using both Twin Earth scenarios (experiment 1) and what they call reverse Twin Earth scenarios (experiment 2), i.e. scenarios where deep structure but not appearance was shared with the standard samples. Subjects were given a description of a case, either a Twin scenario (such as Putnam’s Twin Earth) or a reverse Twin scenario, such as a planet where a substance is found that is H2O but has none of the properties associated with water on Earth (it is solid, not liquid, greenish, and does not support life on the planet). HNJ conclude that in Twin Earth scenarios externalist judgements were the norm, indicating that shared underlying structure is necessary, but when it comes to the reverse Twin Earth scenarios the opposite was the norm indicating that the subjects did not think sharing underlying structure was sufficient. Experiments 1 and 2, they summarize, “indicate that neither a watery substance with a structure completely different from H2O, nor a completely non-​watery substance with structure H2O, belong in the extension of ‘water’, and similarly for other natural kind terms” (Haukioja, Nyquist, and Jylkkä 2020, 30). These speakers thus did not share Kripke’s intuition that underlying properties are sufficient for the application of NKTs. Other studies similarly suggest that no theory that only takes one set of properties to matter, macro-​or micro-​level, will fit the data. Genone and Lombrozo (2012) set out to address the debate between causal and descriptive theories of reference in the case of concepts. Using four types of vignettes crossing the predictions of a descriptive theory with those of a causal theory, Genone and Lombrozo examined whether participants tend to respond on the basis of a single factor (descriptive or causal) or on the basis of a mix of factors. Genone and Lombrozo conclude that most participants use both descriptive and causal information in making reference judgements, and that “the findings support the hypothesis that reference judgments are a function of both descriptive and causal factors” (2012, 728). They also conclude that when descriptive and causal factors were in conflict, responses were mixed, suggesting that the participants adopted distinct strategies for reconciliation. Genone and Lombrozo propose some kind of ‘hybrid’ theory to account for this, incorporating both descriptive and causal factors as relevant to determining reference. While they do not spell out what such a theory might look like, they suggest that existing attempts to combine descriptive and causal elements fail to accommodate the fact that speakers genuinely combine these two elements. In particular, they stress, there is no theory that can account for the lack of consistency within the individual. Such intra-​individual inconsistencies “suggest that they are using both causal and descriptive information in making judgments about reference rather than switching from one resolution of 340

The Need for Descriptivism

the ambiguities to the other between the two parts” (Genone and Lombrozo 2012, 735). This should worry those who suggest that natural kind terms are systematically used in such a way as to express two distinct intensions, one in accordance with descriptivism and one in accordance with Millianism (as in Chalmers 1996, 52–​71; also see Jackson, this volume).9 By contrast, the cluster theory predicts inconsistency of the sort observed by Genone and Lombrozo, since it does not hold that there are certain contexts in which one or the other component systematically holds sway. Rather, on the cluster theory, all the features in the cluster always have to be weighed against one another. Tobia, Newman, and Knobe (2020) also present experimental data suggesting that both superficial, observable properties and deeper causal properties serve as criteria for membership in natural kind categories. Unlike Genone and Lombrozo, they suggest that the proper conclusion is what they label the “Each Separately” view, according to which natural kind terms exhibit a type of ambiguity. When the two criteria conflict, as in a Twin Earth case, people assent to the claims that there is a sense in which the Twin Earth liquid is water and a sense in which it is not. They distinguish the “Each Separately” view from what they call the “Both Together” view which, roughly, they describe as the view that “water” refers to anything that has most of the relevant properties, or the weighted most (presumably they have the cluster theory in mind). While the “Each Separately” view involves two distinct categorization processes, implying ambiguity, the “Both Together” view involves a single process which is influenced by both superficial and causal properties. Tobia et al. suggest that the distinctive predictions of the “Each Separately” view only comes out in Twin Earth scenarios, where a sample has all the superficial properties associated with a natural kind but not the causal ones. And their conclusion is that the data support some version of this view since participants, in Twin Earth cases, were largely inclined to say that the sample was not a member of the kind in one sense, but was a member in another sense.10 More experimental work is no doubt required to evaluate the ambiguity proposal, but there are some reasons to question it. First, to properly test the theory one would have to use not only Twin Earth scenarios but also reverse Twin Earth scenarios, along the lines of JNH. If there is ambiguity and there is one sense in which the underlying property is decisive, then the prediction would be that in such a scenario the participants would say that there is one sense in which the kind is water and another sense in which it is not. This prediction is not borne out by the data provided by HNJ (2020). Second, as noted by Tobia et al. (forthcoming), on their proposal it is hard to explain the fact that many people did not choose the ambiguity option, but judged instead that the liquid is simply water or simply not water (Haukioja, Nyquist, and Jylkkä 2020, 16).11 They hypothesize that this is because context plays a decisive role, and that in some contexts (scientific) underlying properties matter more, and in others (legal contexts) they matter less. They provide an additional experiment testing this and conclude that there was an effect of context on participants’ judgements of category membership (21). The question is whether these data deliver the kind of systematic contextual differences required to support the ambiguity view. Tobia et al. (2020) compare with the case of ‘book’, referring to a physical object (in one context) and to an abstract object (the associated work) in another, and note that unless a general rule of this sort could be discerned, the ambiguity proposal is not supported. However, that some speakers (but not all) tend to pay more attention to underlying properties in scientific contexts need not suggest such a rule, since it may be explained as a salience effect within the confines of a “Both Together” view (such as the cluster theory): When the scientific context is stressed speakers pay more attention to underlying properties since they assume that is what matters to scientific classifications.12 Nichols, Pinillos, and Mallon (2016) also defend a version of the ambiguity theory.They set out to test descriptivist and causal theories and they conclude that in some contexts the reference relation for a kind term is descriptivist, in others it is causal. Their proposal, however, is not that the relevant factors are whether the context is scientific or not, along the lines of Tobia et al. Instead, they appeal to philosophical concerns that have been prominent in the debate. For instance, descriptivist theories 341

Sören Häggqvist and Åsa Wikforss

better accommodate failed reference cases whereas causal theories do better accommodating continuity of reference in the history of science. Nichols, Pinillos, and Mallon predict that they could switch people’s intuitions depending on what type of case they were presented with, concerning the continuity of reference or existence. The predictions draw some support. For instance, responding on a six-​point scale people tended to give significantly less descriptivist answers in cases triggering continuity of reference concerns. They also noted inconsistencies that need explaining if one does not adopt an ambiguity theory. The experiments are interesting and it would be valuable to try to replicate the data on a larger sample size (the sample size is merely thirty-​two subjects). In particular, it would be of interest to see whether it would be possible to get a shift from a clear agreement with the causal-​historical response to a clear agreement with the descriptivist response.This, after all, is what an ambiguity theory would predict, and in the first three studies this is not a pattern that can be discerned. The authors note this, and in an additional study they make the causal-​historical elements more salient, and as a result the answers fell more clearly on the causal-​historical side (while remaining descriptivist in existence cases). However, to settle the matter more data are needed. Even in the additional study responses do not fall neatly into the two categories that an ambiguity theory would predict.13 Also, the type of descriptivism tested is not the cluster theory, even though this theory is mentioned (Nichols, Pinillos, and Mallon 2016, 146), but a more definitional version where the associated descriptions are limited to observational properties. And, again, data from reverse Twin Earth cases would be essential to properly test the ambiguity theory.14

5  Conclusion We have argued that there are reasons to question the widely shared assumption that there is a distinct semantic category of kind terms, ‘the natural kind terms’, which require a wholly non-​descriptivist account. We have appealed to three types of considerations: first, demarcation challenges, concerning how to distinguish the alleged category of terms from other kind terms. Second, evidence from the philosophy of science suggesting that the essentialist assumptions undergirding Millianism fail. Third, recent experimental data concerning speaker usage where several studies give clear evidence that speakers do not privilege underlying properties, but consistently appeal to appearance properties as well in the application of natural kind terms. This suggests that descriptions do not merely serve a reference-​fixing function (as allowed by Kripke) but go into the semantic content of these terms. While it is too early to tell what the semantic upshot is, whether the ambiguity theory or the cluster theory, one conclusion can be drawn with some confidence: it is time to break with the Kripkean orthodoxy. There is a need for descriptivism.

Notes 1 For explicit discussion of this, see Biggs and Dosanjh (this volume). They interestingly argue that most kind terms have a natural kind sense due to the way explanatory interests inform our concepts. Also see Boyd (this volume). He argues that the definitions of particular natural kinds are relative to human projects and disciplines and that in that sense are social constructions, which would seem to suggest that there is no sharp distinction between the natural kind terms and other kind terms. 2 While diamond is a kind, it is presumably not a substance in the chemical sense since it doesn’t persist through phase change. For an account on which it is a separate chemical kind, see Bursten (2014). Haukioja et al. (2020) investigate lay subjects’ usage of inter alia ‘diamond’, and explicitly take it to be a natural kind term. For an account agreeing that it is one, also see Biggs and Dosanjh (this volume). 3 Natural kinds have, for instance, been supposed not just to permit, but also to explain induction, by offering a metaphysical basis for projection and discovery. But as Hacking (2007, 236) points out, artefacts ground stable induction well; and as Godfrey-​Smith (2003) notes, unnatural categories (teenagers) permit statistical induction from fair samples. 4 A notable exception is Ellis (2001).

342

The Need for Descriptivism 5 In water, inversion of density occurs at around 4°C: at higher temperatures, warmer water rises; at lower, it sinks. The triple point is the temperature and pressure at which three phases are at equilibrium. As Needham (2011) stresses, the latter property is quite theoretical (rather than observational), but still macroscopic. 6 Thanks to Stephen Biggs for prompting this clarification. 7 If our dispositions to apply kind terms are agile enough, perhaps they can be determinate even in the absence of essences, or more generally, cooperation on the world’s part. For a proposal that they are, drawing on the idea that kind concept formation is driven by explanatory concerns, see Biggs and Dosanjh (this volume). For similar appeals to how explanatory needs shape natural kind concepts, also see Strevens (2019). Biggs and Wilson argue in a series of papers (2018, 2019, 2020) that kinds concepts are shaped by abductive thinking in a way that imposes determinate rankings between competing intensions, hence curbing extensional indeterminacy. For an opposed view, closer to ours, see LaPorte (this volume). See also Boyd (this volume) who argues that there is a great deal of indeterminacy, especially within biology. 8 We thus propose a different explanation than Kitcher for reference shifts between tokens of the same word type: he traces them causally to different “initiating events” (1978, 537–​539). 9 See also Glüer and Pagin (2012) for an alternative proposal involving different intensions. 10 They also suggest (Tobia et al. forthcoming, 25) that the data might be taken to support the “Just Causal Properties” view, according to which the sense in which twin-​water can be called ‘water’ is a non-​literal sense. However, appealing to the literal/​non-​literal distinction seems like a rather question-​begging way of dismissing all the data that goes against non-​descriptivist theories. 11 Jylkkä, Railo, and Haukioja (2009) also tested the hypothesis that natural kind terms are ambiguous, having two senses (one internalistic, the other externalistic). They conclude that the hypothesis is not supported. Given the choice of providing an ambiguous answer (“yes on the one hand, but …”), only 17 per cent chose this option and 22 per cent still preferred the internalist reply (2009, 58). 12 This, in turn, might be explained by the observation that people tend to assume that science is in the business of finding underlying essences, as suggested by studies of psychological essentialism. 13 One might also worry that the probe in fact begs the question, since it contains the statement that “many scientists, in the middle ages, when faced with wildebeest would often call them catoblepas” (Nichols, Pinillos, and Mallon 2016, 159). This primes the participants into thinking that there must be continuity of reference. 14 See also Foster-​Hanson and Rhodes (this volume), for a discussion of experimental data on psychological essentialism.They suggest that the essentialist intuitions people (sometimes) have are more a reflection of our psychology than of the metaphysical reality of natural kinds.

References Biggs, S. and J. Wilson (2018) Carnap, the necessary a posteriori, and metaphysical anti-​realism, in S. Blatti and S. Lapointe (eds.), Ontology after Carnap. Oxford: Oxford University Press, pp. 81–​104. Biggs, S. and J. Wilson (2019) Abduction versus conceiving in the epistemology of modality, Synthese, doi.org/​ 10.1007/​s11229-​019-​02117-​9. Biggs, S. and J.Wilson (2020) Abductive two-​dimensionalism: a new route to the a priori identification of necessary truths, Synthese, doi.org/​10.1007/​s11229-​017-​1444-​6. Braisby, N., B. Franks and J. Hampton (1996) Essentialism, word use, and concepts, Cognition, 59, 247–​274. Bursten, J. R. (2014) Microstructure without essentialism, Philosophy of Science 81, 633–​653. Chalmers, D. (1996) The Conscious Mind. Oxford: Oxford University Press. Devitt, M. (2008) Resurrecting biological essentialism, Philosophy of Science 75, 344–​382. Devitt, M. and K. Sterelny (1987) Language and Reality, Cambridge, MA: MIT Press. Ellis, B. (2001) Scientific Essentialism, Cambridge: Cambridge University Press. Franklin-​Hall, L. (2015) Natural kinds as categorical bottlenecks, Philosophical Studies 172, 925–​948. Genone, J. and T. Lombrozo (2012) Concept possession, experimental semantics, and hybrid theories of reference, Philosophical Psychology, 25, 717–​742. Glüer, K. and P. Pagin (2012) General terms and relational modality, Noûs 46, 159–​199. Godfrey-​Smith, P. (2003) Goodman’s problem and scientific methodology, Journal of Philosophy 100, 573–​590. Godman, M.,A. Mallozzi and D. Papineau (2020) Essential Properties are Super-Explanatory:Taming Metaphysical Modality, Journal of the American Philosophical Association 6, 316–334. Gray, R. (2006) Natural phenomenon terms, Analysis 66, 141–​148. Hacking, I. (2007) Natural kinds:  rosy dawn, scholastic twilight, Royal Institute of Philosophy Supplement 61, 203–​239. Häggqvist, S. and Å., Wikforss (2007) Externalism and a posteriori semantics, Erkenntnis 67, 373–​386. Häggqvist, S and Å. Wikforss (2015) Experimental semantics: the case of natural kind terms, in J. Haukioja (ed.), Advances in Experimental Philosophy of Language, London: Bloomsbury, pp. 109–​138.

343

Sören Häggqvist and Åsa Wikforss Häggqvist, S and Å. Wikforss (2018) Natural kinds and natural kind terms: myth and reality, British Journal for the Philosophy of Science 69, 911–​933. Haukioja, J. (2015) On deriving essentialism from the theory of reference. Philosophical Studies 172, 2141–​2151. Haukioja, J., M. Nyquist and J. Jylkkä (2020) Reports from Twin Earth: both deep structure and appearance determine the reference of natural kind terms, Mind & Language,  1–​27. Havstad, J. (2018) Messy chemical kinds, British Journal for the Philosophy of Science 69, 719–​743. Havstad, J. (2020) Complexity begets crosscutting, dooms hierarchy (another paper on natural kinds), Synthese, doi.org/​10.1007/​s11229-​020-​02539-​w. Hendry, R. (2006) Elements, compounds, and other chemical kinds, Philosophy of Science 73, 864–​875. Hoefer, C. and G. Martí (2019) Water has a microstructural essence after all, European Journal for Philosophy of Science 9, doi.org/​10.1007/​s13194-​018-​0236-​2. Jylkkä, J., H., Railo, and J. Haukioja (2009). Psychological essentialism and semantic externalism: evidence for externalism in lay speakers’ language use, Philosophical Psychology, 22, 37–​60. Khalidi, M. A. (1998) Natural kinds and crosscutting categories, Journal of Philosophy 95, 33–​50. Khalidi, M. A. (2018) Natural kinds as nodes in causal networks, Synthese 195, 1379–​1396. Kitcher, P. (1978) Theories, theorists and theoretical change, Philosophical Review 87, 519–​547. Kripke, S. (1980) Naming and Necessity. Cambridge, MA: Harvard University Press. LaPorte, J. (2004) Natural Kinds and Conceptual Change, Cambridge: Cambridge University Press. Law, S . (2015) Natural kinds of substance, Australasian Journal of Philosophy, doi: 10.1080/​00048402.2015.1046384. Ludwig, D. (2018) Letting go of “natural kind”: toward a multidimensional framework of non-​arbitrary classification, Philosophy of Science 85, 31–​52. Magnus, P. D. (2011) Drakes, sea-​devils, and similarity fetishism, Biology and Philosophy 26, 857–​870. Malt, B. (1994) Water is not H2O, Cognitive Psychology 27, 41–​70. Martínez, M. (2015) Informationally-​connected property clusters, and polymorphism, Biology and Philosophy 30, 99–​117. Needham, P. (2011) Microessentialism: what is the argument?, Noûs 45:1, 1–​21. Needham, P. (2012) Natural kind thingamajigs, International Studies in the Philosophy of Science 26, 97–​101. Needham, P. (2013) Hydrogen bonding: homing in on a tricky chemical concept, Studies in History and Philosophy of Science 44, 51–​65. Nichols, S., N. A. Pinillos and R. Mallon (2016) Ambiguous reference, Mind 125, 145–​175. Nimtz, C. (2018) How science and semantics settle the issue of natural kind essentialism, Erkenntnis, doi.org/​ 10.1007/​s10670-​018-​0098-​1. Putnam, H. (1975) The meaning of ‘meaning’, in Philosophical Papers, vol. 2: Mind, Language, and Reality, Cambridge: Cambridge University Press, pp. 215–​271. Salmon, W. (1981) Reference and Essence, Oxford: Blackwell. Slater, M. (2015) Natural kindness, British Journal for the Philosophy of Science 66, 375–​411. Soames, S. (2002) Beyond Rigidity:  The Unfinished Semantic Agenda of Naming and Necessity, Oxford:  Oxford University Press. Strevens, M. (2019) Thinking off Your Feet, Cambridge, MA: Harvard University Press. Tobia, K., G. E. Newman and J. Knobe (2020), Water is and is not H2O, Mind & Language 35(2), 183–​208. Wikforss, Å. (2010) Are natural kind terms special?, in H. Beebee and N. Sabbarton-​Leary (eds.), The Semantics and Metaphysics of Natural Kinds, New York: Routledge, pp. 64–​83.

344

26 THE ACCOMMODATION THEORY OF REFERENCE Richard Boyd

Realism and Theories of Reference Theories of reference are important in lots of disciplines. In philosophy they’re important in discussions of, e.g., mental representation, modal logic, analytic metaphysics, philosophical theology, and metaethics. Although it’s been applied in other areas, the accommodationist theory (henceforth: accommodationism), like many other causal theories of reference, was (and is) especially intended to be a contribution to the articulation of a realist conception of scientific knowledge. Realists maintain, against logical empiricists, that (approximate) knowledge of unobservable ‘theoretical’ entities is possible. They reject the position of neo-​Kantian philosophers like Kuhn (1970) that fundamental laws in a discipline or ‘paradigm’ are conceptual truths rather than descriptions that correspond (or fail to correspond) to extra-​paradigm realities. The realist conception instead requires a correspondence conception of accurate representations. The truth, or approximate truth, or falsity of a sentence about electrons, for example, is supposed to be a matter of the real properties of real electrons, not, e.g., of the correct application of a conventional operational definition of ‘electron’, nor of conformity with conventional truths involving the term ‘electron’. Central to the realist project has been the idea that, when things go well epistemically, scientific understanding often progresses by a dialectic of successive approximation: approximately correct theories may give rise to more reliable methods and vice versa, even when successive theories embody fundamentally different laws and concepts. A key point of all realist causal theories of reference has been to underwrite that idea by explaining how scientific terms can sometimes continue to refer univocally even in the face of changes in concepts, ‘operational definitions’, framework principles, etc. Accommodationism is designed to fulfill those requirements.

Reference and Natural Kinds are Two Aspects of a Single Phenomenon Accommodationism differs from other realist naturalist theories in two respects. First, accommodationism treats reference and natural kinds (and natural relations, properties, and magnitudes; henceforth ‘natural kinds’, for simplicity) as two aspects of a single phenomenon of accommodation between linguistic, conceptual, inferential, and observational/​experimental practices, on the one hand, and epistemically relevant causal structures on the other. Second, it treats truth,

345

Richard Boyd

approximate truth, and other forms of representation as special cases of accommodation. Let’s begin with some straightforward examples.

Ground Squirrels Consider cases of non-​human animal signaling. Paul Sherman (1977, 1985) identified two sorts of alarm calls in Belding’s ground squirrels. One sort (a) warns of aerial predators, the other (t) warns of terrestrial ones. The squirrels respond differently to the two sorts of call, engaging in different evasive behaviors appropriate to the different sorts of predators. Sherman thus confirmed a semantic hypothesis: a refers to the natural kind aerial predators; t refers to the natural kind terrestrial predators. The a signals are not invariably associated with aerial predators nor t signals with terrestrial ones; there are false positives and false negatives. Instead, what makes a and t refer to the kinds aerial and terrestrial predators respectively is that (1)  there is a tendency for a signals to be produced in response to aerial predators rather than to other survival-​relevant features of the squirrels’ environment, (2)  there is a similar tendency for t signals to be produced in response to terrestrial predators, and (3)  these facts figure centrally in explaining how Belding’s ground squirrels avoid predation. Sherman’s referential hypotheses are components in the explanation of an achievement. They explain how the linguistic, perceptual, cognitive, and social structures in Belding’s ground squirrels are accommodated to relevant causal features of their environment so as to facilitate predator avoidance. The same accommodationist semantics is appropriate for other cases of non-​human signaling. Three features of accommodationism are important. (1) The semantic hypotheses, and the natural kinds assigned as referents, are species-​and adaptive-​ achievement specific. Belding’s ground squirrel alarm calls refer to kinds of predators on ground squirrels, not, for example, to aerial predators on sunfish or terrestrial predators on zebras. Alarm calls and mating calls are associated with different adaptive achievements and the different sorts of natural kinds to which they refer reflect that fact. (2) The natural kinds referred to by alarm calls are socially constructed by the relevant organisms. The referential relations obtain because of the roles of the alarm calls in socially coordinated behaviors and the definitions of the predator kinds reflect the adaptive value of those behaviors. In some cases—​including that of vervet monkeys (Seyfarth and Cheney 1986)—​the skills required to make correct alarm calls are to some extent socially learned, and the semantics of particular alarm calls is a matter of local culture:  monkey populations in different environments with different predator communities use the same calls to refer to different kinds of predators (Fichte and Kappeler 2011). (3) There’s no philosophical or scientific reason to think that ‘social construction’ in the sense of (1) –​(2) casts doubt on the reality of the natural kinds or of the semantic relations in question. The category aerial predator on Belding’s ground squirrels is not somehow irreducibly rodential in an ontologically compromising way because it’s defined in terms of the behavioral ecology of a particular rodent species. The fact that alarm calls and mating calls have different achievement-​ related semantic properties doesn’t make them ‘interest dependent’ in a way that diminished their reality or the reality of the natural kinds to which they refer. Nor are the categories of predators referred to by alarm calls in different vervet monkey communities, or the referential relations to them, ontologically diminished by being socially learned and constructed.

The Human Case: First Approximation For humans, too, the kind natural kind is itself a natural kind in the study of how classificatory practices, and their linguistic and conceptual, inferential, and observational manifestations, help to 346

The Accommodation Theory of Reference

underwrite the accommodation of human inferential architectures to causal factors in the world, thus underwriting the epistemic reliability of those practices. Natural kind definitions reflect the particular achievements being explained; natural kinds in chemistry need not be natural kinds in geography. The definition of any given K depends on the inferential architecture of the relevant discipline: the characteristic inferential connections between the term referring to K and all of the other natural kind terms within the discipline and the observational and experimental practices associated with them. So the referents of natural kind terms in a discipline, and the definitions of their referents are determined, not one term at a time, but instead, by the overall practices in the discipline. Here’s a simplified, initial approximation. [NB: I’ll continue to write about natural kinds, but accommodationism offers exactly analogous accounts of natural properties, relations, and magnitudes.] Let M be a disciplinary matrix and let t1, … ,tn be the natural kind terms deployed within the discourse central to the inductive/​explanatory successes of M. Then the families F1, … ,Fn of causal properties provide definitions of the kinds referred to by t1, … ,tn, and determine their extensions, just in case: 1. There is a systematic, causally sustained, tendency—​established by the causal relations between practices in M and causal structures in the world—​for what is predicated of ti within the practice of M to be approximately true of things which satisfy Fi, i=1, … , n. In particular, there is a systematic tendency for things of which ti is predicated to have (some or most of) the properties in Fi. 2. This fact, together with the causal profiles of things satisfying these definitions, helps to causally explain how the use of t1, … ,tn in M contributes to accommodation of the inferential architecture and practices of M to relevant causal structures. It helps to explain whatever tendency there is for participants in M to identify causally sustained generalizations, to obtain correct explanations, or to obtain successful solutions to practical problems. It’s a matter of controversy among philosophers sympathetic to the HPC [homeostatic property cluster] conception to what extent that conception applies in cases of ordinary everyday kinds and categories (see Boyd 1999a and Wilson 1999 for contrasting approaches). As this initial approximation indicates, reference and natural kind definitions are aspects of the very same phenomenon—​accommodation.

Partial Denotation, and Other Complications The initial approximation can be improved significantly by recognizing two complications. First, some terms in a discipline may not refer at all (more on this possibility later). Second, some terms partially denote and may undergo denotational refinement sensu Field (1973) (see also Boyd 2010).

One More Factor: Property Clusters The accommodationist literature has been closely associated with discussions about the homeostatic property cluster (HPC) conception of some natural kinds and about related cluster conceptions (Boyd 1999a, 2010, 2018; Griffiths 1999;Wilson, Barker and Brigandt 2007). According to any cluster definition-​friendly version of accommodationism, the accommodation of some natural kind concepts and terms to epistemically relevant causal structures requires (!) that those terms and concepts have indeterminate extensions. ‘Precisified’ versions of such kinds would be less well accommodated to relevant causal structures—​would correspond less well to reality. For example, biological species are HPC natural kinds. Suppose that the kinds represented by particular species names were precisified by a refinement of usage that established strict species boundaries. These boundaries would not 347

Richard Boyd

correspond to explanatorily or inductively relevant biological features. So any such refinement would mislead biologists about important biological factors. It would either mislead researchers towards treating as important distinctions that are irrelevant to causal explanation or to induction, or towards ignoring biologically important similarities. So, correspondence representation in the accommodationist realist’s sense implies that some terms correspond to partly indeterminate extensions and some sentences involving perfectly OK kind terms lack determinate truth values. Realism does not imply bivalence.

Social Construction, Realism, and ‘Mind-​Independence’ The accommodationist conception for human languages treats natural kinds and their naturalness as features of humans’ communicative and inferential practices.The definitions of the kind ‘natural kind’ and of the relation ‘reference’, and the definitions of particular natural kinds are relative to human projects and disciplines, and reference to those kinds is a distinctly social discipline-​specific process. So kinds and reference in the human case are thus social constructions. As in non-​human cases, this sort of social construction or ‘mind dependence’ doesn’t compromise the reality either of reference, or of the kinds in question. Human minds, social organizations, disciplinary aims and practices, etc. are just as real causal phenomena in nature as are similar phenomena in other species, and their inductively and explanatory relevant causal powers are equally as real. The mind dependence in question is a matter of causal relations between perfectly real features of the natural world (Boyd 1999b, 2010, 2019).

Some Consequences1 Reference is Neither ‘Pure Causal’ Nor ‘Causal-​Descriptive’ Many philosophers distinguish between ‘pure causal’ theories of reference—​according to which the reference of term is fixed by something like a baptismal or dubbing event followed by a chain of intentions to co-​refer—​and hybrid ‘causal-​descriptive’ conception—​according to which reference is determined by causal connections of the sort posited by ‘pure causal’ conceptions, together with conceptually central descriptions or causal/​inferential roles associated with a term by fully competent users. According to accommodationism neither is true. Reference is an ongoing process and the initial introduction of a term need not play any special role in establishing its referent. Moreover, the most fundamental concepts or inferential practices that scientists associate with natural kind terms as a matter of linguistic or conceptual competence need not play any special role in establishing reference. One of the (perhaps the) key point of causal or naturalistic theories of reference is that the most conceptually fundamental beliefs or inferences associated with a scientific term need not be approximately true of, or reliable with respect to, its referent. Of course, in the human case the satisfaction of the epistemic access and accommodation conditions will ordinarily depend on term-​users having lots of approximately true beliefs and approximately reliable inferential practices associated with the term, but none of these need be fundamental, and which ones they are need not be discernable by ‘conceptual analysis’ or by consulting ‘linguistic intuitions’. Cognitive, inferential or social factors are not treated as distinct from other causal factors:  they’re simply among the relevant causal factors in reference. Accommodationism is not a hyphenated ‘causal-​descriptive’ theory. Instead, accommodationism is a purely causal theory of reference: it treats the conceptual, inferential, and social factors involved in reference as ordinary causal features of the natural world. But it’s not anything like a ‘pure causal’ theory of reference as that description is ordinarily understood. A referential hypothesis about the terms in a discipline (or, more generally, an accommodationist semantic hypothesis in the sense discussed below) will be an hypothesis about what causally explains the inferential and explanatory successes in the discipline. 348

The Accommodation Theory of Reference

Definitions, in the Accommodationist Sense, are not Representational Entities The definitions anticipated by accommodationism and other naturalistic approaches are supposed to correspond approximately to ‘real essences’ in Locke’s (1689/​1975) sense—​to the real metaphysics underwriting the uses of natural kind terms. Examples like ‘Water  =  H20’ may suggest that such definitions are representational entities: a sentence, like ‘Water = H20’, or the proposition expressed by such a sentence, or a formula like ‘H20’, or the concept associated with one of these items, or something like that. Of course, that’s one of the scientifically and philosophically important notions of ‘definition’. On the accommodationist conception, however, the fundamental notion is that of a family of properties that figure in the right way in explaining the epistemic contribution of a term’s usages. The real metaphysical definition of natural kind is a non-​representational entity that’s itself represented by terms or concepts that refer to the kind. That conception is required because accommodationist natural kinds and their definitions figure in causal explanations of the epistemic successes underwritten in practice by the inferential architecture of scientific disciplines. A definition understood as a set or family or cluster of inductively or explanatorily important properties whose co-​occurrence tends to regulate the use of a term can figure in such an explanation whether or not anyone has the conceptual resources to formulate a definition of the kind it defines. A definition understood as a representational entity could play such an explanatory role only when researchers have something like the conceptual resources to formulate it. A central point of all naturalistic theories of natural kinds and their definitions is that uses of natural kind terms can contribute to explaining such successes even when no one has such resources. It’s an interesting and difficult question whether or not the accommodationist conception of reference can accommodate (pun) a notion of rigid designation and thus underwrite a conception of a posteriori necessary truths about natural kinds, magnitudes, properties, etc. (henceforth: natural kinds). On the one hand, an affirmative answer might be defended along the lines indicated by this outline. Suppose that in the actual world, A, a natural kind term, t, refers to kind K. Suppose that t, used in the actual world, refers to K’ in some possible world, W. Here’s (perhaps) why K=K’ according to accommodationism. Accommodationism implies that the definition in A of any natural kind depends on the inferential architecture of the relevant discipline, D. So, accommodationism implies that transworld identity for kinds tracks transworld identity for disciplinary architectures. So, if t used in A refers in W to K′, K′ plays in W the causal role associated with the term, t, in the W-​world inferential architecture of D. But, according to accommodationism the inferential architecture of a discipline includes, in additional to inferential, experimental, and instrumental practices, the reference-​ constituting causal relations between the terms in the discipline and their actual referents. So, t, used in A, refers in W to the kind associated, in the disciplinary architecture of D in W, with t. But that’s K. So K=K′. On the other hand, the whole thrust of the accommodationist conception of reference is that determinate reference is a perhaps rare phenomenon in the history, and is a species of a much looser relation—​accommodation—​between concepts and terms and disciplinarily relevant causal structures. It’s not ordinarily to be expected that a natural kind terms refers determinately in the actual world. Perhaps then it’s even less to be expected that a conception according to which natural kind terms rigidly designate phenomena across possible worlds would figure in a successful treatment of the myriad ways in which counterfactuals or claims about definitions figure in scientists’ (and others’) inferential architectures. I’m inclined to favor the second approach. In any event it’s useful to remember that the literature on causal or naturalistic theories of kinds and reference approximately bifurcates into a literature concerned mainly with issues in modal logic and analytic metaphysics and a literature concerned 349

Richard Boyd

mainly with issues about referential continuity and kind definitions in the actual (and very nearby possible) worlds (Beebee and Sabbarton-​Leary 2010b). The literature influenced by accommodationism is very definitely in the latter camp.

Published Definitions are Usually Hypotheses about Accommodationist Definitions Of course, philosophers and scientists do publish definitions of natural kinds. Almost always these are best understood as hypotheses about what the real definitions (in the accommodationist sense) of those kinds are: think biologists’ many definitions of species. Sometimes they reflect hypotheses about partial denotation and/​or proposed new definitions to resolve partial denotation and achieve denotational refinement:  think Kitcher (1984) on species. Even when denotational refinement is thereby achieved, the proposed published definitions need not correctly reflect the real definitions of the disambiguated natural kinds. Finally, sometimes scientists offer ‘operational definitions’ of scientific terms: specifications of the measurement, detection, or assessment procedures associated with those terms as they are used in particular empirical studies. An operational definition of a term is best understood as working hypotheses about how to assess, in particular practices, the extent to which something exhibits the properties in the term’s accommodationist definition. Despite the empiricist philosophical history of the term ‘operational definition’, operational definitions are not matters of linguistic stipulation but scientific proposals subject to theoretical and empirical challenges.

Definitions Don’t ‘Fix Reference’ The definitions posited by accommodationism and by other naturalistic theories of natural kinds and reference don’t fix reference, nor do the other sorts of definitions. Instead reference (and kind definitions) are fixed, to the extent that they are fixed, by the interactions between the aspects of the inferential architecture of a discipline and relevant causal structures. Almost the whole point of naturalistic conceptions of reference and natural kinds is that scientists (and other people) routinely refer successfully to kinds about whose definitions they either haven’t a clue or have an utterly mistaken conception. Accommodationism is not a theory of how definitions (of any sort) fix reference (they don’t). It’s a theory of how complex interactions between the uses of terms in a discipline and methodologically relevant families of properties in the world determine reference and kind definitions simultaneously. It’s not either a theory focused on scientists’ (and others’) definitional practices. These practices it treats as ordinary components of theoretical and instrumental work. They’re not, in general, assigned a special significance in the establishment of reference relations. Definitions don’t fix reference.

Reference as a Dialectical Process: Reference is Often not ‘Fixed’ The term ‘reference fixing’ is misleading: reference is really a dialectical process. Partial denotation and related phenomena are commonplace in the sciences and other intellectual disciplines. It is pretty common for terms to be, for a long time, associated with a variety of different partial denotata while nevertheless contributing significantly to accommodation. Often it takes a long time for reference to be ‘fixed’. Similarly, the dialectical process may involve the introduction of new concepts and terms for newly recognized phenomena. We should think of reference, not as a relation between terms or signals and determinate kinds but instead as the dialectical process of accommodation between the uses of terms or signals and methodologically relevant families of properties that explains how classificatory practices contribute to inductive explanatory and practical successes (Boyd 1993, 2001, 2013). If things go well, reference—​understood as an ongoing process—​may result in the establishment of 350

The Accommodation Theory of Reference

the determinate reference suggested by the idealizations reflected in formal logical systems, but that sort of situation may be somewhat rare.

Rethinking Truth: For More Correspondence not Less Accommodation without Approximate Theoretical Truth Realist conceptions maintain that, in many important cases, scientific research proceeds by successive approximations, both theoretical and methodological:  approximate truth and approximately reliable methods go hand in hand, with reliable methods depending on approximately true theoretical commitment and contributing to more accurate theories and even better methods and so on. In some important cases the situation is messier. In many historical cases, however, empirical successes—​ successful empirical approximations—​were achieved without scientists’ methods being undergirded by anything like approximately true basic theories. Many ancient and medieval astronomers achieved remarkably empirically adequate conceptions of astronomical phenomena without having approximately true theories about the nature of those phenomena. Darwin’s achievements rested on an enormous body of scientific lore about species, animal behavior, biogeography, inheritance, etc. but not on deep theoretical knowledge in these domains. Phlogiston theorists made important empirical contributions to chemistry, as did chemists whose theory of acidity posited that all acids contain oxygen. Much empirically successful work in optics proceeded without an approximately true theory of the nature of light, etc. The accommodationist conception permits the articulation of a fully general realist conception of the epistemology of science that accommodates (pun intended) both sorts of cases. According to the general conception, empirically successful science depends causally on there being a suitable accommodation between inferential architecture and relevant causal structures.That accommodation may be (and is in the case of many recent sciences is) significantly underwritten by approximately true theoretical commitments about the causal factors that underwrite the empirically predicted phenomena, but it can be underwritten to a significant extent by the uses in the relevant discipline of observational terms whose uses in relevant practices are in fact aligned with the operation of those factors even in the presence of a fundamentally false theory about them—​or no theory at all (Boyd 2019, esp. section 3.4).

Theoretical Approximation without Determinate Reference A related phenomenon sometimes occurs in successful contemporary sciences where background theories that underwrite reliable methods involve terms that lack determinate reference. Consider, for example, substantively and methodologically insightful theories that deploy the term ‘species’ and related taxonomic terms. It’s a serious position in biology and the philosophy of biology that those terms, and many particular species names, fail to have determinate referents. That could happen in a way basically favorable to the picture of progress by successive approximation to the truth. For example, suppose that Kitcher (1984) is right that ‘species’ partially denotes slightly different loci of local evolutionary stability in different sub-​disciplines of biology and is in need of denotational refinement. Absent such refinement, some theoretically important claims about species must, strictly speaking, lack a determinate subject matter and so cannot be true, or approximately true (or false, for that matter). Nevertheless, in that case and in similar cases one might be able to argue along something like the following lines. The different partial denotata of the term in question are similar enough that many of the statements that scientists accept are usefully approximately true about each of the denotata. When the term later undergoes denotational refinement its precisified versions are even better accommodated to relevant causal structures and make possible even more fruitful approximations to truths about each of them. Accommodationism treats reference 351

Richard Boyd

as a dialectical process like that, so it can preserve the realist idea that epistemic successes are often achieved by successive approximations to the truth. So could any other realist theory of reference that acknowledges partial denotation and denotational refinement.

Scientific Progress without Approximate Truth The conception that sciences progress through successive approximations to the truth is more seriously challenged when, in highly successful contemporary scientific inquiry, findings and theories are sometimes formulated in sentences that lack truth values because some key terms lack even partial denotata. Sometimes, theories and findings framed using such terms make positive but messy contributions to scientific knowledge and progress, even though the ways in which the uses of the terms in question contribute to knowledge is too messy to be treated as matters of partial denotation or other referential ambiguities. In many such cases scientists are eventually able to resolve the messiness only by introducing a new suite of terms and concepts to replace the non-​referring and non-​partially denoting ones. Nevertheless, scientists’ ability to clarify issues in that way will depend on their being able initially to muddle through, guided by approximate knowledge that is reflected, in part, in their earlier very messy theories and findings that weren’t (even modulo partial denotation) approximately true (Boyd 2013, 2019). Here’s a possible example. Many philosophers and biologists (certainly a minority) have expressed the view that there’s no determinate species category (see, e.g., Mishler 1999; Ereshefsky 1998, 1999, 2010; Hendry et al. 2000; Hey et al. 2003). Suppose for the sake of argument that they’re right. It might then be necessary to replace reference to ‘species’, ‘sub-​species’, ‘species families’, etc. with altogether different terminology. Something like that is suggested by these excerpts from Willis’s (2017) analysis of the species complex of Amazonian peacock “bass” cichlids (Cichla pinima sensu lato). Results suggest that the historical narrative for these populations is more complex than can be portrayed by recognizing them as one, two, or four species: their history and contemporary dynamics cannot be unambiguously rendered as discrete units (taxa) at any level without both choosing the supremacy of one delimitation criterion and obscuring the very information that provides insight into the diversification process. This calls into question the utility of species as a rank, term, or concept, and suggests that while biologists may have a reasonable grasp of the structure of evolution, our methods of conveying these insights need updating (italics mine). … In this way, “species” is a red herring: something that serves to confuse and distract from the goal, which is to document and understand evolutionary processes. Instead, researchers should focus on quantifying the diversity in nature, at whatever hierarchical level it occurs, rather than applying labels that are counterproductive to our goals, and work to interpret what that variation means about contemporaneous and historical processes that contributed to that diversity. Although proposing a system to refer to evolutionary phenomena as an alternative to “species” is beyond the scope of this paper, it would seem less ambiguous, and thus ultimately more efficient, for studies using lineages as the basis of their work to include informal verbal descriptions of these meta-​population lineages accompanied by visual aids (photos, maps, phylogenies, frequency distribution plots, etc.) to describe the units under study. Suppose for the sake of argument that something like Willis’s position is correct. We’ve got a ‘reasonable grasp’ of the structures of the sorts of evolutionary phenomena we now describe using 352

The Accommodation Theory of Reference

‘species’-​ish terms and concepts but they’re the wrong terms and concepts. We may suppose that terms like ‘genus’, ‘species’, ‘sub-​species’, ‘species family’, etc. often serve, in particular applications, to mark out some biologically important distinctions but that, even within investigations by different researchers of a given set of populations, they could be employed differently by different researchers to mark out the different basic distinctions. There is, let’s assume, insufficient uniformity in how they can be fruitfully used for them to be understood as having anything like determinate partial denotata (that seems to be Willis’s claim). On these assumptions we’ve been muddling through using non-​ referring (and non-​partially denoting) terms and concepts, but we’ve still got a reasonable enough grasp of the relevant issue (that is: reasonable enough approximate knowledge) to begin to move towards replacing the defective terms and concepts with very different ones. The implications are extraordinary. Millions of published findings about ‘genera’, ‘species’, ‘sub-​ species’, ‘species families’, etc. (1)  have had no truth values or determinate propositional content, even allowing for ambiguity and partial denotation and (yet), (2) they contribute to our having a reasonable grasp of the structure of evolution, and embody sufficient approximate knowledge to help biologists underwrite the articulation of an entirely new set of terms and concepts. They represent approximate knowledge about evolution without having truth values or determinate propositional contents. Let’s call terms that behave as ‘genera’, ‘species’, ‘sub-​species’, ‘species families’, etc. are said to do on Willis’s conception insightful terms. Given the actual inferential architectures of the relevant disciplines, they contribute to the accommodation of methodological practices to epistemically relevant causal structures. They make possible insights—​ instances of a sort of approximate knowledge—​that help scientists muddle through even though scientific findings reported with them altogether lack truth values because they don’t come close to having determinate subject matters. Let’s call the relation insightful terms bear, in actual practices, to the subject matter(s) of their sciences insightful accommodation. Like determinate and partial denotation, insightful accommodation will involve a term’s uses being, to some extent, causally regulated by the properties of methodologically relevant causal structures, but the specificity of that regulation will not be sufficient to underwrite even partial denotation. On Willis’s account of species-​ish terms, for example, the uses of those terms will, in almost any particular study, be reliably regulated by the properties of evolutionarily important families of populations but—​because species-​ish terms don’t align systematically with important properties of such families across different studies—​partial denotation is not achieved. In fact, cases of insightful accommodation can be somewhat messier than this (see Boyd 2013, 2018, where cases of insightful accommodation are described but the term ‘insightful accommodation’ is not used). The posited future history of species-​ish terms and concepts is, of course, conjectural. Similar conjectures are plausible about terms like ‘gene’, ‘population’, ‘monophyly’, ‘organism’, ‘homology’, and so on. Other candidates for a similar status are of some importance to philosophers: ‘proposition’, ‘instinct’, ‘reduction’, ‘law’, ‘meaning’, and ‘race’ (see Biggs and Dosanjh (this volume) for relevant discussion). Almost certainly there are important scientific (and philosophical terms) that are insightful in this way. They—​and some of the methods and findings they figure in—​are onto something(s) but in a way that doesn’t fit neatly (even modulo partial denotation) into a logician’s conception of reference and approximate truth. As in cases of uniquely referring and partially denoting terms, their uses are causally regulated by features of some inductively or explanatorily important causal structures but the regulation is pretty messy. Still, some of the sentences involving insightful terms are such that, given the prevailing inferential architecture, they approximately represent important features of the world—​and contribute to the formulation and discovery of more ordinary approximate representations—​they’re insightful approximate representations without themselves being approximately true in any ordinary sense. So, instead of always being epistemic distractions, sometimes the uses of non-​referring (and non-​partially denoting) terms and concepts contributes positively to scientific knowledge. 353

Richard Boyd

Referential Relations and Truth as Special Cases of Accommodation Pretty clearly there’s something like a continuum between unique reference and partial denotation and between partial denotation and insightful accommodation of terms to relevant causal structures. That’s what you should expect given that reference and reference-​like relations are processes of which exactly determinate reference is a special case. For that reason, accommodationism treats insightful accommodation—​epistemically relevant regulation of term uses by relevant causal structures—​as the fundamental referential relation. Unique reference and partial denotation are special cases. In fact, the fundamental semantic relation between signal systems and the world is accommodation itself: the alignment between inferential practices and causal structures that underwrites epistemic successes achieved using those systems. So, reference, partial denotation, and insightful accommodation are special cases of the ways in which inferential architectures involving human-​language signaling systems are aligned with causal structures in an epistemically favorable way. Other features of inferential architectures are also special cases of accommodation: extent of the reliability of perception and of instrumentation in a discipline; extent of the reliability of information transmissions through journals, graduate training, conferences, etc. Importantly, all these special cases are holistically related. Partial denotation, for example, is a special case of accommodation only because of (among other things) the extent of the epistemic reliability of related scientific instrumentation and modes of information transmission. Conversely, the extent of the reliability of instrumentation depends—​g iven the theory-​dependence of instrumentation—​on the approximate accuracy of theories framed using partially denoting terms, as does the extent of the epistemic reliability of information transmission involving those terms. In the same sense truth, approximate truth, and insightful approximate representation are also special cases of accommodation (Boyd 2013, 2018). For inferential architectures that make use of the discursive resources of recursive languages, like human inferential architectures, these factors play the same sort of role in establishing and maintaining accommodation as do the other causal contributors to accommodation.

Accommodation and Our Animal Friends The proposal to see reference, truth, and approximate truth as special cases of accommodation, when coupled with the idea that reference and accommodation are ongoing processes, perfectly covers the sort of refinement of usages proposed by Willis for ‘species’, where the insights framed in terms of earlier concepts and terms are to be reframed with concepts and terms even better accommodated to relevant causal structures. It does something more. It shows that truth in human languages is a special case of a sort of representation where the representing signal need not have anything like determinate truth conditions or determinate propositional content. Approximate representation by signals is a matter of epistemically important alignment between the structures of signals and relevant causal structures underwriting epistemic achievements, even when that alignment looks very much unlike truth à la Tarski. That in turn allows us to see approximate representations in human languages and in animal signaling systems, like the alarm calls discussed earlier, as instances of the same natural phenomenon of alignment with causal structures underwriting epistemic achievements. We can understand epistemically successful representations in signaling systems of our animal—​and plant, and fungal, and prokaryotic—​friends as being representations in the same sense that approximate representations in human languages are representations. This is, of course, as it almost certainly should be. It’s extremely plausible that human language capacities had ancestral roots in earlier primate signaling systems that lacked the recursive structures necessary to underwrite exactly determinate truth conditions or propositional contents as we ordinarily understand them.We should expect that human languages emerged 354

The Accommodation Theory of Reference

from earlier signaling systems and that there should be some continuity in how they represented relevant phenomena. The accommodationist conception provides the conception of representation appropriate to that expectation.

Approximation and the Growth of Knowledge The continuity between approximate representations in non-​human signaling systems and truth, approximate truth, and merely insightful representations in human language suggests the possibility of understanding the category of knowledge to extend to some approximate representation in non-​ human signaling systems. Consider the epistemic successes of those of our animal friends who lack the resources to have representations with determinate (or even approximately determinate) truth conditions or to engage in the sorts of rationalizing deliberation and justification that our recursive languages sometimes makes possible. Some of them are, nevertheless, really smart and some engage in quite sophisticated learning and teaching (think of vervet monkeys correcting incorrect signaling by young monkeys), sometimes achieving, over time, better and better approximate representations. One response is to treat their inductive and practical achievements as genuinely epistemic achievements of the same basic kind as human epistemic achievements—​as instances of approximate knowledge. That response is characteristic of many naturalistic approaches to epistemology. Antony (2000) nicely describes the ways in which naturalized epistemologists adopt a broad conception of epistemic categories. Naturalized epistemologists do not concern themselves with skeptical challenges. Nor are naturalized epistemologists much concerned with questions about what counts as “knowledge”, properly speaking. They do not worry if a bird’s natively specified program for star-​based navigation is “justified” for the bird, nor if the sub-​personal data structures and algorithms posited by cognitive psychologists can be properly counted as “beliefs”. The naturalized epistemologist is interested in the explanation of anything that even appears to be a cognitive achievement, whether or not it passes muster as “knowledge” in some preferred sense. 2000, 103 Naturalistic approaches of this sort have the advantage that, like the accommodationist conception of approximate representation, they could underwrite a plausible evolutionary continuity between knowledge in humans, with the full capacity for recursive language, and knowledge in those ancestral species where this capacity was not fully developed. The main challenge to approaches of this sort are, of course, variations on the idea that knowledge is justified approximately true belief. They have the advantage that they can emphasize the distinctive epistemic importance of the sort of rational demanding and giving of justifications that the representational capacities of human languages make possible. Nevertheless, the accommodationist conception of representation provides additional reasons to believe that we should think of cases of justified approximately true beliefs as special cases of a broader conception of knowledge along naturalist lines. Those reasons have to do with the ways in which merely insightful approximations in the sciences—​ones that certainly look like clear cases of approximate scientific knowledge—​don’t fit nicely with the idea of justified approximately true beliefs. Let’s go back to the sort of scientific scenario suggested by Willis. If his picture of the role of species-​ish terms and concepts in evolutionary theory turns out to be right, then the reasonable grasp we currently have of evolutionary phenomena is reflected in hundreds of thousands of publications deploying terms like ‘species’, ‘sub-​species’, ‘species complex’, ‘genus’, ‘variety’, etc.These publications will reflect the beliefs of their authors that particular populations do or do not belong together in various species, sub-​species, species complexes, genera, varieties, etc. None of these beliefs will be 355

Richard Boyd

approximately true in the ordinary sense, because each will lack (even modulo partial denotation) a subject matter. The uses of species-​ish terms (together, of course, with uses of lots of other terms, many of them better behaved semantically) will have contributed to biologists’ reasonable grasp—​that is to say approximate knowledge—​of key evolutionary phenomena but none of the evolutionary insights reflected in the uses of species-​ish terms will have a sentential home: an approximately true sentence that approximately expresses the insight. Those merely insightful approximations will come to have sentential homes only after new terms and concepts and theories are introduced that are much better accommodated to the underlying evolutionary phenomena. The investigations that lead to those new terms, concepts, and theories will not have been about species, sub-​species, species complexes, genera, etc. so much as they will have been about the real evolutionary phenomena to which the terms ‘species’, ‘sub-​species’, ‘species complex’, ‘genus’, etc. are so messily accommodated. Only after the acceptance of the new theories and terms will biologists be able to deploy the full resources of sentences with approximately determinate truth values to frame, and to justify or criticize, hypotheses about the relevant phenomena in ways that conform to ordinary conceptions of seeking justified approximately true beliefs. Whatever other conceptions of knowledge might be appropriate for other purposes, it’s hard to escape the idea that, on one important conception of knowledge at least, the reasonable grasp of the structure of evolution—​provided in part by these thousands of homeless sentences—​would constitute approximate scientific knowledge. Sententially homeless approximations like the ones entailed by Willis’s conception will occur whenever there are merely insightful approximations in science, and the capacity of scientists to subject them to ordinary standards of justification and criticism will depend on their eventually being reframed in theories, terms, and concepts that are much better behaved semantically. Nevertheless, the insights they provide will be instances of approximate knowledge. So, too, will the contributions to approximate knowledge made by other sentences whose acceptance doesn’t fit the ordinary model of justified approximately true beliefs. For example, there are lots of sentences in the literature in synthetic chemistry whose acceptance contributes to important knowledge, not because they are always, or even most often, approximately true, but instead because they are true often enough under the conditions where well-​trained chemists typically apply them, or because they are true ceteris paratis (true if you fiddle things right). Their contributions to approximate chemical knowledge depends on the particular inferential practices and judgments of practicing chemists. Many of those practices and judgments are only imperfectly articulable, given the current state of the field, and their explicit justifications would require theoretical and conceptual advances that have not yet been achieved. They correspond to the tacit dimensions of paradigms sensu Kuhn (1970) (Boyd and Boyd 2019). So, once again we have approximate representations that are clearly important examples of approximate chemical knowledge but that need not be even approximately true and whose systematic justification and criticism is not at present fully possible. Again we have approximate scientific knowledge that does not meet the standard of justified approximately true belief as that standard is ordinarily understood. None of these accommodationist considerations just rehearsed call into question the profound epistemic importance for the sciences (or other disciplines or everyday life) of having recursive languages whose sentences have something like determinate truth conditions that can underwrite semantically well behaved justifications and criticisms of theories and beliefs. Indeed, an absolutely central point of the accommodationist conception of reference and natural kinds—​and of most other conceptions as well—​is to explain the epistemic importance of that sort of representation. What’s true, however, is that accommodationism opens up the possibility of a somewhat broader conception of approximate scientific knowledge. The accommodationist conception does not—​at least as it’s presently developed—​entail any particular account of the sort of approximate scientific knowledge involved in cases like these, which are ubiquitous in science. Perhaps one should expand the conception of justification so that it applies to 356

The Accommodation Theory of Reference

approximate representations that are situated in the inferential architectures of human disciplines in the right way even though they lack anything like determinate truth conditions. Alternatively, perhaps one should see the justification and criticism of beliefs as special cases of kinds of learning and thinking in some of our non-​human animal friends. In any event representation in human languages is a special case of representation in signaling systems of other organisms and the idea that some of our animal friends are capable of knowledge is not far-​fetched (see Kornblith 2002). For the accommodationist, the role of theories of representation is to contribute to answering—​ about ourselves and other organisms—​a version of the central scientific question in epistemology originally formulated, imperfectly but strikingly, by Quine. For me then the problem of induction is the problem of how we, as we now are (by our present scientific lights), in a world we never made, should stand better than random or coin-​tossing chances of coming out right when we predict by inductions which are based on our innate scientifically unjustified similarity standard. Quine 1969

Further Reading Beebee and Sabbarton-​Leary, eds. (2010a) and Kendig, ed. (2016) are valuable collections of relevant papers. Boyd (2010) is a good introduction to accommodationism in the context of the philosophy of science literature. Ereshefsky and Reydon (2015) represents a different and important approach to many of the issues discussed here. Hacking (2007) is an important critique of the idea that natural kinds constitute a natural kind in philosophy. Slater (2016) is an extremely sophisticated discussion of the accommodationist literature including important critiques of some of the ideas presented in this essay.

Note 1 For much of this material see Boyd (2019).

References Antony, Louise. 2000, ‘Naturalized Epistemology, Morality and the Real World’. Canadian Journal of Philosophy 30 (Supplement), 103–​137. Beebee, Helen and Nigel Sabbarton-​ Leary, eds. 2010a. The Semantics and Metaphysics of Natural Kinds. New York: Routledge. Beebee, Helen and Nigel Sabbarton-​Leary. 2010b. ‘Introduction’, in Helen Beebee and Nigel Sabbarton-​Leary, eds. The Semantics and Metaphysics of Natural Kinds. New York: Routledge. Boyd, R. 1993. ‘Metaphor and Theory Change’ [second version], in A. Ortony, ed. Metaphor and Thought, 2nd edition. New York: Cambridge University Press. Boyd, R. 1999a. ‘Homeostasis, Species, and Higher Taxa’, in R. Wilson, ed. Species: New Interdisciplinary Essays. Cambridge, MA: MIT Press. Boyd, R. 1999b. ‘Kinds as the “Workmanship of Men”: Realism, Constructivism, and Natural Kinds’, in Julian Nida-​Rümelin, ed. Rationalität, Realismus, Revision: Proceedings of the Third International Congress, Gesellschaft für Analytische Philosophie. Berlin: de Gruyter. Boyd, R. 2001. ‘Truth through Thick and Thin’, in Richard Schantz, ed. What is Truth?. Berlin and New York: de Gruyter. Boyd, R. 2010. ‘Realism, Natural Kinds and Philosophical Methods’, in Helen Beebee and Nigel Sabbarton-​ Leary, eds. The Semantics and Metaphysics of Natural Kinds. New York: Routledge. Boyd, R. 2013. ‘Semantic Externalism and Knowing Our Own Minds:  Ignoring Twin-​ Earth and Doing Naturalistic Philosophy’. Theoria 79, 3, 204–​228. Boyd, R. 2019. ‘Rethinking Natural Kinds, Reference and Truth: Towards More Correspondence with Reality, not Less’. Synthese. https://​doi.org/​10.1007/​s11229-​019-​02138-​4. Boyd, W. C. and R. Boyd. 2019. ‘Natural Kinds and Ceteris Paratis Generalizations: in Praise of Hunches’. Hyle 25, 1, 21–​47. Ereshefsky, M. 1998. ‘Species Pluralism and Anti-​Realism’. Philosophy of Science, 65, 103–​120.

357

Richard Boyd Ereshefsky, M. 1999. ‘Species and the Linnaean Hierarchy’, in R. Wilson, ed. Species: New Interdisciplinary Essays. Cambridge, MA: MIT Press. Ereshefsky, M. 2010. ‘Darwin’s Solution to the Species Problem’. Synthese 15, 3, 405–​425. Ereshefsky, Marc and Thomas Reydon. 2015. ‘Scientific Kinds’. Philosophical Studies 172, 4, 969–​986. Fichte, Claudia and Peter M. Kappeler. 2011. ‘Variation in the Meaning of Alarm Calls in Verreaux’s and Coquerel’s Sifakas (Propithecus verreauxi, P. coquereli)’. International Journal of Primatology 32, 2, 346–​351. Field, H. 1973. ‘Theory Change and the Indeterminacy of Reference’. Journal of Philosophy 70, 462–​481. Griffiths, Paul. 1999. ‘Squaring the Circle: Natural Kinds with Historical Essences’, in R.Wilson, ed. Species: New Interdisciplinary Essays. Cambridge, MA: MIT Press. Hacking, Ian. 2007. ‘Natural Kinds: Rosy Dawn, Scholastic Twilight’. Royal Institute of Philosophy Supplement 82, 61, 203–​239. Hendry, A. P., S. M. Vamosi, S. J. Latham, J. C. Heilbuth, and T. Day. 2000. ‘Questioning Species Realities’. Conservation Genetics 1, 67–​76. Hey, J., R. S. Waples, M. L. Arnold, R. K. Butlin, and R. G. Harrison. 2003. ‘Understanding and Confronting Species Uncertainty in Biology and Conservation’. Trends in Ecology & Evolution 18, 597–​603. Kendig, Catherine, ed. 2016. Natural Kinds and Classification in Scientific Practice. London and New York: Routledge. Kitcher, Philip. 1984. ‘Species’. Philosophy of Science 51, 2, 308–​333. Kornblith, H. 2002. Knowledge and Its Place in Nature. Oxford: Oxford University Press. Kuhn, T. 1970. The Structure of Scientific Revolutions, 2nd edition. Chicago, IL: University of Chicago Press. Locke, J. 1689/​1975. An Essay Concerning Human Understanding. Oxford: Oxford University Press. Mishler, Brent. 1999. ‘Getting Rid of Species?’, in R. Wilson, ed. Species: New Interdisciplinary Essays. Cambridge, MA: MIT Press. Quine,W. V. O. 1969.‘Natural Kinds’, in W.V. O. Quine, Ontological Relativity and Other Essays. New York: Columbia University Press. Seyfarth, R. M. and D. Cheney. 1986. ‘Vocal Development in Vervet Monkeys’. Animal Behaviour 34, 1640–​1658. Sherman, Paul. 1977. ‘Nepotism and Evolution of Alarm Calls’. Science, 197, 1246–​1253. Sherman, Paul. 1985. ‘Alarm Calls of Belding’s Ground Squirrels to Aerial Predators:  Nepotism or Self-​ Preservation?’ Behavioral Ecology and Sociobiology 17, 4, 313–​323. Slater, Matthew. 2016. ‘Natural Kindness’. The British Journal for the Philosophy of Science 66, 2, 375–​411. Willis, Stuart. 2017. ‘One Species or Four? Yes!...and, No. Or, Arbitrary Assignment of Lineages to Species Obscures the Diversification Processes of Neotropical Fishes’. PLoS ONE 12, 2, e0172349. Wilson, R.A. 1999.‘Realism, Essence and Kinds: Resuscitating Species Essentialism’, in R.Wilson, ed. Species: New Interdisciplinary Essays. Cambridge, MA: MIT Press. Wilson, R. A., M. J. Barker and I. Brigandt. 2007. ‘When Traditional Essentialism Fails: Biological Natural Kinds’. Philosophical Topics, 35, 189–​215.

358

27 SCIENCE, THE VERNACULAR AND THE ‘QUA’ PROBLEM Robin Findlay Hendry

Introduction: Referential Continuity In the 1950s and 1960s, philosophy of science inherited from logical positivism a consensus that the meaning of a scientific term is determined by its theoretical role. Semantic holism suggested that this ‘theoretical role’ included the entire intellectual context within which scientific thought takes place: networks of beliefs and implicit assumptions including theoretical, experimental, semantic and ontological principles. Paul Feyerabend and Thomas Kuhn (Feyerabend 1962; Kuhn 1970, ch. X) then pointed out that this dependence of meaning on theoretical context opened up the possibility that the reference of a theoretical term might be very sensitive to even minor changes in that context. Even if successive generations of scientists use the same words (such as ‘gold’, ‘tiger’ or ‘water’), the historical development of science cannot be regarded as a cumulative process in which they gradually amass knowledge about gold, tigers or water. The retention of the words may mask the fact that their meaning has changed. This sensitivity might also preclude us from regarding past thinkers as being in genuine disagreement with us. Can we really say that Aristotle, for instance, believed falsely that water is an element? The alien intellectual context in which he worked should make us more careful about how we translate his texts. The work of Saul Kripke (1980) and Hilary Putnam (1975) provided a response: reference might be relatively independent of background belief because an expression attaches to its referent directly, like a proper name.1 Take water, or gold. For Putnam, samples of these substances would be picked out by ostension or description (e.g.‘the stuff that flows in rivers and falls from the sky as rain’, or ‘the stuff the King’s crown is made of ’), the descriptions understood indexically as picking out different natural kinds in different environments. ‘Water’ and ‘gold’ are now understood to apply to anything bearing the underlying microstructural features that are shared by the samples. The underlying assumption of this response is externalism, the idea that reference—​the fact that someone is thinking (and talking) about gold or water—​may be determined partly by facts of which they are unaware, facts which are external to their epistemic perspective. Externalism itself presupposes that the relevant facts exist independently of their being known. It is to that extent committed to some form of realism about them (see Putnam 1975, 235–​238).

359

Robin Findlay Hendry

Two Problems of Many Kinds In what follows I will examine two objections to Putnam’s account: the ‘qua’ problem (see Devitt and Sterelny 1999, ch. 5), and Joseph LaPorte’s charge of referential indeterminacy (2004, 2010). They are intimately related, both arising from the following facts: the users of a natural-​kind term are faced only with (a finite set of) particulars, and any such finite set of particulars may share more than one property. I will then examine two different responses to those objections. I argue against the idea that the problems of many kinds can be resolved by appeal to a priori criteria for what can count as a natural kind. Instead, I argue that we should consider the classificatory interests of scientists. As we shall see, this kind of response has a price: it is discipline specific because the classificatory interests of scientists may vary. This complicates the relationship between the realism that motivated Putnam’s account, classificatory pluralism, and our understanding of how science relates to the vernacular. I will not here address the relationship between essentialism and the ‘Kripke-​Putnam’ view of reference (on which see Mackie 2006, c­ hapter 10), or essentialism in chemistry (for recent discussions of which see Bursten 2014;Tahko 2015; Häggqvist and Wikforss 2018; Havstad 2018 and Hoefer and Martí 2019). The ‘qua’ problem concerns how, given these facts, a direct theory of reference can hope to show that some natural-​kind term refers to one kind rather than any of the others that might be instantiated by the particular examples surrounding its users. Take, for instance, a particular population of animals that live alongside some linguistic community that uses the term ‘tiger’ to refer to them. Which kind of animal should one regard ‘tiger’ as picking out? There will typically be many properties that the members of that population have in common: they may all be large, striped, carnivorous, four-​legged felines; they may share a particular evolutionary history, a certain amount of genetic information and so on. The ‘qua’ problem is the problem of identifying which of these properties, or which combination of them, could be taken to be the defining property of the relevant kind.The second objection, raised by LaPorte (2004, 2010), is that this very multiplicity of candidate kinds suggests that it is indeterminate which of a number of closely related kinds is the referent of a term. Such indeterminacy may only be resolved when the differences between the various kinds are discovered by the users: in the case above, members of the relevant linguistic community may come across large, non-​striped, carnivorous, four-​legged felines and face a decision as to whether they should count as tigers. This undermines the claim that, when scientists typically come to a view on what makes something gold or water, this can be regarded as a discovery. LaPorte argues that, because there is more than one way such a kind term may legitimately be made more precise, we must acknowledge that the KripkePutnam model does not entail that the essential properties of natural kinds are discovered. These two problems are forcefully illustrated by the phenomenon of isotopy: atoms of the same element that differ in respect of their atomic weight.2 The discovery of isotopy was occasioned by that of radioactivity. It wasn’t clear at first whether or not radioactivity involved the disintegration of atoms, or transmutation between chemical elements, but chemical analysis showed that the decay of thorium, for instance, gave rise to a metal that was chemically indistinguishable from lead, despite its having a different atomic weight. In 1910 Frederick Soddy proposed that in such cases the decay products should not be regarded as new elements, but instead should occupy the same place in the periodic table as the elements they resemble (hence the term ‘isotope’ for the co-​occupants). If it had been confined to a few heavy elements, the phenomenon of isotopy would have been a rare phenomenon, but the development of mass spectrography showed that it was widespread (see Bruzzaniti and Robotti 1989).The situation became clearer in 1913, when H. G. Moseley devised an X-​ray method for determining nuclear charge, the property that different isotopes share, but the issue of how to define the elements was not thereby settled (see van der Vet 1979; Kragh 2000). There was now a choice to be made between acknowledging different isotopes as different elements, which would have involved replacing Mendeleev’s periodic table with one containing many more elements (one might call this an abundant view), and keeping the table as it stood but changing the property (atomic weight) that defined each place in it (one might call this a sparse view). The sparse view won out: in 360

Science, the Vernacular, the ‘Qua’ Problem

1923 the International Committee on Chemical Elements, appointed by the International Union of Pure and Applied Chemistry (IUPAC), enshrined nuclear charge as determining the identity of the chemical elements (see Aston et al. 1923). The two ‘problems of many kinds’ arise in the case of isotopy in the following way: consider gold, which is unusual among the elements in being entirely composed of a single isotope, 197Au. The ‘qua’ problem arises because it is unclear what, in Putnam’s picture, could make it the case that past scientists used the word ‘gold’ to refer to populations of atoms with a nuclear charge of 79, rather than populations of atoms with an atomic weight of 197, or populations of atoms with both properties. LaPorte’s objection arises because if nothing makes it the case that past scientists used the word ‘gold’ to refer to populations of atoms with a nuclear charge of 79, with an atomic weight of 197, or with both properties, then it should be regarded as indeterminate whether exotic and short-​lived isotopes of gold, none of which would have been present in the naturally occurring populations available to scientists before the twentieth century, should be considered as falling within the extension of ‘gold’. The post-​isotopy decision was essentially stipulative, a choice between two possible ways to make element names more precise. In the face of these difficulties even friends of direct reference must admit that the theory, and Putnam’s original examples, were somewhat underdescribed. To defend the essentials of the view, and its utility as a response to the problem of conceptual change, one must identify considerations, missing from the original discussions, which privilege ‘the’ natural kinds of science, or rule out the alternatives to them. How might this be done? One way is to apply very general constraints and abstract on what can count as a natural kind, such as those posited by Brian Ellis as part of his characterisation of ‘scientific essentialism’ (Ellis 2002, ch. 2; 2009, ch. 3). Thus, for instance, a system of natural kinds might be required to be discrete, hierarchical and based on intrinsic properties (Ellis 2002, 26).3 The prospects of success for this kind of approach do not seem good, because there are core cases of natural kinds that violate them. For instance, it is actually far from clear that chemical kinds (which form Ellis’s paradigmatic example) are discrete in the way he thinks: ‘there is never a gradual transition from any one chemical kind to any other chemical kind’ (2002, 26). Even if the elements may be considered discrete, chemistry is not exhausted by the elements. Isomers are distinct compounds composed of the same elements in the same proportions, such as ethanol (CH3CH2OH) and dimethyl ether (CH3OCH3). The differences between isomers must be understood in terms of their different structures—​the ways in which the elements are combined together in the compound. But structure appears to be defined in terms of continuously varying quantities such as bond lengths and bond angles. Moreover, transformations between organic substances are explained in terms of reaction mechanisms which are precisely ‘gradual transitions’ between chemical kinds. William Goodwin (2011) distinguishes two notions of reaction mechanism: in what he calls the ‘thick’ sense, a mechanism traces the motions of the constituent atomic nuclei and electrons in a continuous path from the reagents’ molecular structure to that of the products; a mechanism in the ‘thin’ sense neglects some parts of the process in order to concentrate on a few well-​understood and explanatorily relevant stages. How the two conceptions fit together is a good question (Goodwin 2011, 310–​315), but to the extent that a mechanism in the thin sense focuses on salient stages of a process it does not fully describe, a mechanism in the thick sense seems more fundamental. If that is right, then chemistry itself sees chemical change as involving continuous transitions between chemical kinds. Chemical kinds cannot then be regarded as discrete. It is similarly debatable whether natural kinds more generally are hierarchical (see for instance Tobin 2010) or based solely on intrinsic properties (think of biological species considered as clades).4 One might defend Ellis’s a priori conditions for natural kinds by accepting the conclusion that chemical kinds are not natural kinds.5 Although he seems to be under the false impression that chemical kinds are discrete and hierarchical, Ellis himself rejects biological species as natural kinds on similar grounds (2009, 59). The general lesson I would draw is quite different: a priori speculation on what can or cannot count as a natural kind is a bad idea because it may fail to fit the classificatory 361

Robin Findlay Hendry

practices of real sciences. It is of course possible that the sciences work mostly with categories which turn out not to be natural kinds, but surely too much bullet-​biting of this sort will undermine the idea that metaphysics can help us to understand how the sciences work. I would not say that a priori theories of what counts as a natural kind are bound to fail. They do, however, seem to invite failure, arising from objections based on detailed examinations of the classificatory practices of the sciences. And is there a guarantee that such a priori criteria deliver a unique system of natural kinds? Without that the problems of many kinds go unanswered. Such claims and counter claims raise the question of who owns the notion of a natural kind. Ellis seems to assume that the answer to that question is metaphysics. Ian Hacking (2007b) argues that the idea of a natural kind originated in the philosophy of science, tracing it back to John Stuart Mill. Only its appropriation by metaphysics has spoiled it for the description of scientific practice. I think that Hacking gives up too quickly: let us reclaim the idea of a natural kind for the philosophy of science by starting with the sciences themselves, attempting to discern their classificatory interests from their past and present practice. In other papers (Hendry 2006, 2010a, 2010b, 2012, 2016), I have set out a discipline-​specific response to the ‘qua’ problem and LaPorte’s indeterminacy argument.6 The general idea is that the particular, and contingent, classificatory interests of a science may pick out which properties shared by exemplars of are definitive of the relevant kind. In the case of chemistry the claim is that, since around the mid-​eighteenth century, systematic thinking about the composition and behaviour of natural substances has been committed to three theoretical assumptions about elements as components of substances:  that elements are actually, rather than merely potentially, present in their compounds; that they survive the specific kinds of chemical change (combustion, calcination, acid-​base reactions) that were studied by the people among whom the discipline of chemistry emerged in the eighteenth century; and that the actual presence of an element is what explains (at least partially) the chemical and physical behaviour of its compounds (see Hendry 2019). Once this context is taken into account, I argue, there is no question which of the candidate kinds is being referred to by element names. The argument is as follows (for more detail see Hendry 2006, 2010a, 2010b, 2012, 2016): if elements survive in their compounds, their defining properties must be nuclear, for the electronic structure of atoms varies across different states of chemical combination. This gives us a choice between nuclear charge, mass or a combination of the two. Gold’s natural isotopic purity is unusual among the elements: the elements gathered by eighteenth-​century chemists, considered as atomic populations, would mostly have been diverse in respect of atomic weight. Tin, silver and mercury, all of which came to be regarded as elements in the eighteenth century, are more typical: their atomic populations are not dominated by any single isotope. Add in the fact that it is nuclear charge rather than mass that to a great extent determines (and therefore explains) the chemical behaviour of an element and its compounds, and nuclear charge emerges as the only serious candidate. An appeal to epistemic interests seems to provide a plausible answer both to the ‘qua’ problem and to LaPorte’s referential indeterminacy.

Classificatory Pluralism Putnam’s view presumes that what makes something gold is its composition (from atoms of a particular kind), which may be unknown both to historical scientists and to laypeople. But what makes it the case that atomic composition is what makes something gold, rather than (for instance) its appearance and behaviour? This is where a second assumption that Putnam makes comes in. ‘Gold’ and ‘water’ are used in a particular way that involves deference to experts:  he calls this a ‘socio-​ linguistic hypothesis’ (1975, 227). Someone who uses a natural-​kind term may have some idea of how to recognise examples, but need not possess the expertise to tell them apart from other things they resemble. In using the term they intend it to apply to things that experts would identify as gold. But science has determined that atomic composition (having atoms with a nuclear charge of 79 atomic 362

Science, the Vernacular, the ‘Qua’ Problem

units) is what makes something gold, so deference makes this the criterion for being gold even in a lay person’s usage. Applying this model in an obvious way to ‘water’, the word in its standard usage picks out samples of stuff which are predominantly composed of H2O molecules.7 The argument set out in the previous section focused on the classificatory interests of one particular science (chemistry), but what if laypeople have quite different classificatory interests? What if the classificatory interests of scientists themselves are diverse? Pluralism about classification is the thesis that there is more than one way to divide the world into kinds, and that, from a purely ontological point of view, none of these ways is privileged over the others. Putnam assumed that non-​ scientific users of a kind term intend it to refer to some objectively existing kind that scientific inquiry will, sooner or later, uncover.The scientific investigation may even be revisionary, in the sense that some of the things the non-​scientists had thought to be of one kind may be discovered to be of another kind. To take another biological example, on Putnam’s view one might say that whales were once thought of as fish, but were then discovered to be mammals, and therefore non-​fish. John Dupré opposes this assumption of deference (Dupré 1993), pointing out that there are many informal systems of classification associated with practical, non-​scientific activities such as cookery and gardening. These systems of classification are, he argues, autonomous:  they are governed by interests which are different than those that motivate scientific inquiry and should not be expected to defer to it. For instance, from a biological point of view tomatoes are fruit, whereas culinary classification would group them with vegetables, along with other salad ingredients. We cannot say that the culinary classification is incorrect merely because it is different from the scientific one. Interestingly, Dupré combines pluralism about classification with realism about kinds: he does not deny that there are real divisions in nature; rather he thinks that there are many, and that the practical or epistemic purposes of various crafts and sciences make different divisions salient. Clearly, scientific and non-​ scientific classification may become entangled, even though the autonomy of the practical categories makes deference to science ‘philosophically unmotivated’ (Dupré 1999, 461). In the past, whales counted as perfectly good fish, because ‘fish’ invoked something like a sea-​dwelling animal with a characteristic shape.Thus Moby Dick was both a fish and a whale, as was the great fish that swallowed Jonah. Educated people now typically exclude whales from the category of fish on the grounds that they are mammals, as reflected in the more restrictive Oxford English Dictionary definition of ‘fish’ as ‘cold-​blooded vertebrates with gills’ (Dupré 1999, 466). This exclusion does not arise from any discovery that whales are non-​fish: neither ‘whale’ nor ‘fish’ is the name of a particular biological species, and neither corresponds to a well-​formed higher taxon (1999, 462–​467). In short, biology does not really take a view on either category: ‘What a fish is is not the sort of thing a scientist (except, perhaps, a linguist) could find out’ (1999, 467). But if ‘folk once believed that whales were fish’ and now they do not, they must have been ‘duped into changing that belief for bad reasons’ (1999, 465). Moreover, Dupré argues that biology itself is not unified from a taxonomic point of view, with various parts of that science employing quite different species concepts based on morphology, phylogeny or membership of a breeding community (1993, ch. 2). This further undermines the idea that we are entitled to assume that one taxonomic system should be regarded as privileged.8 Now there are disanalogies between chemical substances and biological species. Firstly, in the chemical case the very same kind terms (‘water’ and ‘gold’) are used in both scientific and vernacular contexts. Secondly, chemistry is taxonomically more unified than biology: as we saw, in the early twentieth century IUPAC agreed on a definition of the element names in terms of nuclear charge, and in IUPAC’s current systematic nomenclature for chemical substances, molecular structure provides the sole basis for the naming of substances (see Thurlow 1998 and Hendry 2016). This suggests that our stance towards various classificatory terms might vary and should be sensitive to the particular facts about them. Nevertheless, the question arises whether there are folk classifications of chemical substances that cut across the scientific categories, and if so whether, like the folk-​biological categories, they should be regarded as being ‘on a par’ with them (Dupré 1999, 462). Clearly the same issues arise because the vernacular terms ‘water’ and ‘gold’ are also associated with practical interests, 363

Robin Findlay Hendry

chief among which are, presumably, the use of water for drinking and washing, and the use of gold for decoration and as a repository of value. However, with the IUPAC decision, scientists chose to define the extension of the elements in a way that includes pure deuterium oxide (2D2O) in the extension of ‘water’. Because drinking pure 2D2O will eventually kill you, this violates a commonplace expectation of water: that, when pure, it is safe to drink. Likewise, the IUPAC decision makes exotic and short-​lived gold isotopes fall within the extension of ‘gold’, even though their radioactivity and short half-​lives would make objects composed of them quite unsuitable as either jewellery or repositories of value. But by what right does science revise commonplace beliefs about water in this way? Why not reject the claims that 2D2O ever fell within the extension of ‘water’, and that exotic gold isotopes ever fell within the extension of ‘gold’? The response I sketched in the last section to the two ‘problems of many kinds’ depends on a contingent theoretical commitment on the part of a specific discipline, and its interest in explaining certain kinds of chemical change. As I have argued elsewhere (Hendry 2010a, 2010b, 2012, 2016), I  do not think that this contingency undermines a realist view of chemical classification in any interesting way. The interests that characterise a scientific discipline, or a craft or other activity such as cookery or gardening, are unlikely to be fickle. Moreover they will tend to focus on properties that are both robustly real and causally relevant to the behaviour of things. (These are defeasible assumptions.) However, the dependence does mean that the response applies only to users of a term who actually have the relevant interests and theoretical commitments (and those who defer to them). Take theoretical commitment first: it would be misleading to say simply that Aristotle thought that water is an element, and that we do not. Rather we must give a more complicated story that includes the wider context of Aristotle’s views on what it is to be an element. We need to be similarly careful when making the claims about the extension of the word ‘gold’ before the eighteenth century. Next consider interest-​dependence: it seems quite obvious that some kinds of substance are picked out by classificatory interests which are quite different from those of chemistry. For instance, some particular piece of artificial silk might be chemically identical to some particular piece of real silk, but the former is artificial silk because it did not originate in a silkworm, while the latter is real silk because it did. In this case, the causal history of the substance, and in particular its origin in a member of a particular biological species, seems more important than its chemical composition.9 Conversely, what counts as silk in virtue of its origin in a silkworm might well be heterogeneous from a chemical point of view. The same goes for wool, wood and various foodstuffs. Jade provides a different kind of example: it is well known to philosophers that ‘jade’ applies to two quite different chemical substances, jadeite and nephrite, the evolving application of the term being constrained by appearance and economic context as well as constitution (for details, see LaPorte 2004, 94–​100). But neither ‘jade’ nor ‘silk’ apply to chemical substances: neither term is used by the discipline of chemistry. What of terms, like ‘water’ and ‘gold’, which are used both inside and outside chemistry? If the vernacular uses are governed by interests quite different to those of science, then perhaps we should treat the different uses as homonyms. If so, then deference to science, and the idea that science discovers the natural kinds that underlie vernacular kind terms, will seem unmotivated.

Science and the Vernacular What does this all imply for the relationship between scientific and vernacular uses of the word ‘water’? To raise the question of what relationship there is between scientific and ordinary language presupposes that they are distinct, yet it is not immediately obvious why they should be: scientists are people, and their technical discourse mostly has its origin in the colloquial.Though scientific concepts are refined and often highly abstract, they are honed for the description of the very same world in which everyday life takes place. J. B. S. Haldane made just this assumption in his short educational sketches for the Daily Worker (collected in Haldane 1941). In one striking example (1941, 53–​55), he fills in some background to a recent explosion in an ice-​cream factory, explaining how compressed 364

Science, the Vernacular, the ‘Qua’ Problem

gases may cause explosions, also discussing ammonia the substance, its toxicity, its presence in the outer planets and its possible role as an alternative to water in an extraterrestrial chemistry of life. At one point he asks ‘What is ammonia?’ but continues ‘If you buy a bottle labelled “ammonia” at the chemist’s, you do not get a pure substance, but a solution of ammonia in water’ (1941, 53). Haldane does not actually say what ammonia is until later: it consists of ‘1 atom of nitrogen united into a molecule with 3 of hydrogen’ (1941, 54). So the assumption that motivates all Haldane’s sketches—​that science investigates the same world as that in which everyday life takes place, and that its discoveries can deepen our understanding of both the commonplace and the extraordinary—​is underwritten by a presumption that a practical category (ammonia, a cleaning fluid) and a scientific category (NH3) are linked, and that the scientific category determines the extension of the practical one. Containing ammonia the substance (NH3) is what makes ammonia the cleaning fluid the particular cleaning fluid that it is. In the last section we saw two models of how the relationship between science and the vernacular might work: Putnam’s deference, and the pluralist model, which one might call difference. Might linguistic research favour one of these models over the other? Barbara Malt (1994) and Noam Chomsky (1995) have argued that vernacular usage of ‘water’ does not track H2O content. Malt asked a sample of undergraduate students to judge the H2O content of a range of liquids that are called ‘water’ (e.g. tap water, bottled water etc.), and of a range of liquids which are not (e.g. tea, saliva, coffee etc.). Although the waters were judged on average to contain more H2O than the non-​waters, she found it significant that the students allowed waters to contain significant amounts of non-​H2O (up to a third), and judged many liquids with high H2O content (e.g. tea and coffee) to be non-​waters. She also asked them to judge how typical these liquids were as water: here again there was some correlation between typicality and judged H2O content, but also important were use (drinking water was the most typical), location (in a house, or perhaps a bottle) and source (coming from a tap). Malt concluded that her data were inconsistent with Putnam’s assumption of what she calls ‘psychological essentialism’ (1994, 64), which is roughly the idea that perceived chemical composition alone determines whether or not a particular liquid is judged to be water. The other factors—​function, location and source—​ should not be dismissed as ‘quick and dirty’ procedures for identification: rather they are ‘involved in the concept of water’ (1994, 65). There are a number of things one might say in defence of Putnam’s view. The first is that, although arresting, the title of Malt’s widely-​cited article (‘Water is not H2O’), is not really borne out by its contents.10 ‘Water is not H2O’ suggests that water is something else, whereas the following facts are quite clear from Malt’s results: being H2O is the only chemical requirement that is relevant to being water; it is the only requirement of any kind that is necessary to being water (Malt acknowledges this: 1994, 66); and the other relevant conditions she mentions (function, location and source) are fleeting, and apply only to particular types of water, suggesting that they are merely contingent features. Given these facts, being composed of H2O molecules seems to be uniquely relevant to being water. This is compatible with water being H2O even in vernacular usage, although some other factors may affect how closely a particular sample is judged to be stereotypical. This is perfectly consistent with Putnam’s analysis. A second response is that something doesn’t become water merely by being called water: perhaps we need instead to understand why some watery stuff is not judged to be water, and why some non-​watery stuff is called water (see Abbott 1997). For instance, substances that are mostly water may fall under a more salient kind (such as babies, who are mostly water). Conversely, substances that are called ‘water’ as a whole may contain significant amounts of other chemical substances (for instance dissolved impurities). This is not because impurities magically become water by being included among a far greater amount of H2O, but is rather because the sample as a whole is mostly water (i.e. H2O), and falls under no other more salient kind. Now these comments (although plausible, in my view) merely re-​interpret the linguistic phenomena in terms more congenial to Putnam’s account. They do not yet undermine Malt’s alternative view. 365

Robin Findlay Hendry

In the context of this debate it would be question-​begging to use science itself to undermine Malt’s view, for instance by arguing that her empirical results show only that the folk wrongly call some things ‘water’ which are not water, and that they should therefore be ignored. This presupposes the priority of the scientific usage, which is just what is at stake. Instead, we need a critical examination of the conception of water that is supposed to be embodied in vernacular usage of ‘water’. If it is confused or incoherent on its own terms, there would be an internal reason to defer to the scientific conception. There is a wider issue here about the limits of pluralism. Defenders of the difference model attempt to identify the extension of a folk category via linguistic evidence alone, contrast its extension with that of some scientific counterpart and conclude that the folk category is independent of the science. This makes two substantive assumptions: one is that the speakers of a natural language are authoritative about the extension of the kind terms they know how to use; the other is that kind terms in use have an extension.These assumptions are substantive because we cannot just assume that the competent speakers of a language collectively constitute Humpty Dumpty, whose words mean precisely what he chooses them to mean. The defender of Putnam’s view will reject the first assumption because on that view, speakers of a language use its kind terms to describe a world of which they have incomplete knowledge, and a complete account of classification and the meaning of natural-​kind terms must take this into account. Hence the argument against Putnam’s model based on the linguistic evidence alone begs the question of externalism. It must assume that ordinary speakers are authoritative, which is just to assume the falsity of externalism, to which Putnam is committed (see Chomsky 1995 for explicit argument against externalism).The second assumption, that the ‘folk concept’ of water has an extension, is false because the intuitions that Malt extracts from the linguistic evidence seem to be collectively inconsistent. If so, then one might say that the category has no extension. Consider two ways in which a substance name can be applied via ostension (e.g. ‘that is water’ and ‘that is gold’). The attribution may apply to a sample as a whole, or merely to a component of it. The first usage is in evidence in Malt’s study in what she interprets to be uses of the names of types of water (e.g. ‘bottled water’ or ‘river water’). One might see these rather as samples of water differentiated by their location or source, but let us leave that aside: the point is that the category ‘water’, she assumes, is applied to whole samples even when it is recognised that they are not 100 per cent H2O. It would seem that the impurities—​the non-​H2O parts—​are being called water. The second sense is in evidence in Malt’s study in assessing uses of the phrase ‘X is only partly water’ (Malt 1994, 62). In this sense we may ask how much water there is in the human body (it is widely known that the percentage is high), or how much water has been added to some ham (a lower percentage is evidence of higher quality ham).11 Interestingly, application of ‘water’ in this sense must be independent of its use, location and source. It is also independent of the display of the external characteristics of liquid water (e.g. transparency, potability etc.). The key point, however, is that in this usage, stuff that is not H2O does not become water merely by being mixed with H2O. But this is inconsistent with what is assumed in the other use of ‘water’. So if vernacular usage is not simply incoherent we are dealing with at least two folk categories: let us call them water-​as-​such and water-​in-​a-​mixture. I take it that we can dismiss water-​as-​such even as a practical category, for it cannot adequately meet the practical interests of everyday life. Even if drinking a very dilute solution of some heavy-​ metal ion would do me no noticeable harm over the short term, this does not mean that there is no toxicity lurking within it, ready to do me harm in more concentrated form, or if allowed to build up in my body over a long period of time. The toxic part no more becomes water than it becomes non-​toxic. So the category water-​as-​such is at the very best a vehicle for quick-​and-​dirty judgements about the immediately noticeable consequences of action. So let us turn instead to water-​in-​a-​ mixture: what is it that determines which parts of a mixture are or are not water? The only credible answer is whether or not they are composed of H2O molecules. Why? One might take application of ‘water’ to the parts of a mixture to reflect a potential for that mixture to yield water on separation

366

Science, the Vernacular, the ‘Qua’ Problem

of its components. But these dispositions are ungrounded unless explained by the presence of H2O molecules. Otherwise we lose the distinction between water that is actually present in a mixture (in the form of H2O molecules), and water that is merely potentially present because it would result from chemical reactions that accompany the separation process. Consider the question ‘how much water is present in hydrogen peroxide (H2O2)?’ This substance decays vigorously on heating: 2H2O2 → 2H2O + O2 Through spontaneous decay, some water is always present in a sample of H2O2, but a dispositional conception of the water component would misidentify the amount if it failed to distinguish the water that is actually present at some point in time from what would be generated by the decay of H2O2 as a result of the separation process. Appealing to the presence of H2O molecules is the only obvious way to make that distinction. Now the defender of difference might respond here that the usage of ‘water’ as a mixture-​component is a philosophically unmotivated change to a perfectly adequate folk category, mirroring Dupré’s commentary on the exclusion of whales from the category of fish. But this fails as a full defence of an independent folk category of ‘water’, because then the concentration of the folk concept on surface characteristics is insufficient to motivate all uses of ‘water’. It is one thing to regard deference to science as unmotivated (as Dupré does), and it is quite another to deny its existence. There are two further assumptions behind the project of identifying the classificatory concepts behind vernacular usage, which are more broadly questionable. One such assumption is that vernacular usage of kind terms embodies anything like a system of classification at all. Another is that, if it does embody such a system of classification, this system can be read directly off the linguistic evidence without any mediating selection or conceptual reconstruction. Why are these substantive assumptions? Because a system of classification is not merely a jumble of kind terms: presumably it aims at completeness for some domain, the extensions of its terms enter into set-​theoretical relationships with each other (such as inclusion, exclusion and overlap etc.), and its categories are connected by some kind of similarity relation which may either be theoretical (for instance, sameness of molecular structure) or phenomenological. Scientific systems of taxonomy—​those of chemistry, botany or zoology, for instance—​have been honed over centuries, more or less consciously, in order that their kind terms really do form a system. For instance, in the 1780s Lavoisier and others introduced into chemistry a binomial system of nomenclature that remains in use today (consider for instance the names ‘carbon dioxide’ and ‘copper sulphate’). This nomenclature was based on a project of identifying a complete list of the elements, in the sense that the system takes the names of the elements as the basis for constructing the names of compound substances. Such projects make substantive assumptions about their domains: in chemistry’s case the assumptions are that there are indeed substances (the elements) out of which other substances are composed, but which are themselves composed of no other substances, and that composition in this sense is important to understanding chemical behaviour (it is). Eventually this system grew into the periodic table, a system of classification par excellence in that it was shown in the twentieth century to be exhaustive (see Hendry 2019). So a system of scientific classification embodies a substantial amount of empirical knowledge and conceptual refinement, knowledge and refinement which it has acquired precisely with the aim of ensuring that the system is unified and coherent, in short that it makes some kind of sense of its domain. Now I don’t mean to say that these systems develop in the abstract, or on their own: clearly they are the work of generations of people who teach each other, influence each other, who set up institutions including scholarly societies, university departments and so on. This is what constitutes the reality of a scientific discipline as a historical entity. The same processes of knowledge-​gathering and refinement, perhaps less formally and explicitly, have gone on in the practical activities invoked by Dupré (gardening, cookery and so on), and are carried out by precisely analogous historical institutions. Now contrast this with the ‘concept

367

Robin Findlay Hendry

of water’ that one might attempt to extract from linguistic behaviour. Is there any reason to think that this evidence points to linguistic regularities that embody a coherent system of knowledge about the relevant domain (which in this case includes water)? No, and the mere noting of linguistic regularities fails to associate them with any particular set of knowledge or interests. This may not be enough to determine an extension.

Conclusion Many philosophers would argue that acknowledging the role of epistemic interests in reference must to some degree undermine a realist view of classification. In one way it does not: both Putnam’s defence of referential stability, and Dupré’s classificatory pluralism are compatible with, and indeed motivated by, a bare metaphysical realism. Where they part company is in how they see the relationship between science and the vernacular. While Putnam assumes that the latter must defer to the former, Dupré allows that ordinary usage might be autonomous. However, the recognition that epistemic interests play a role in reference enforces significant qualifications to each of these views. Putnam’s model makes two substantive assumptions; firstly that there are sufficient regularities in the usage of natural-​kind terms among speakers of a natural language to make sense of the idea of ‘the extension’ (at the level of the entire language) of words like ‘water’ and ‘gold’; secondly that there could be a scientific project which roughly corresponds to investigating the microstructural properties possessed by all the portions of matter that fall within these (whole-​language) extensions. We can now see that, taken together, these two assumptions are highly implausible. People learn to use words in all sorts of different contexts, and then use them in all sorts of idiosyncratic ways, in the pursuance of a great variety of epistemic, practical and (more broadly) creative projects. Why should we think that the many ways in which ‘water’ is used by English speakers collectively determine an extension? Why should we think that any single set of interests are at work across all the diverse contexts in which the terms ‘water’ and ‘gold’ are used? Dupré’s classificatory pluralism assumes that there are different systems of classification which should be taken to be ‘on a par’, but if this parity involves having a determinate extension then it assumes a unity and coherence that may fail to apply. Usage across a natural language may be disparate to the point of collective incoherence. Only more local kinds of usage, unified by their association with particular crafts or other kinds of activity and the interests that go with them, will fit the bill.

Notes 1 In what follows I will focus exclusively on Putnam. Kripke’s version of the direct theory of reference has little to do with science and targets Frege and Russell rather than Kuhn and Feyerabend. See Putnam (1990) and Hacking (2007a). 2 Isotopy is a natural focus for discussions of chemical classification: apart from LaPorte’s discussion see Weisberg (2006), Bird (2010), Massimi (2012) and Tahko (2015). 3 Ellis does not use his conception of a natural kind to address referential indeterminacy for natural kind-​terms, but the following remarks apply to any attempt to specify the notion of a natural kind through such a priori constraints. 4 These assumptions can be pervasive, even insidious: in a quite different setting Aronson, Harré and Way (1995) build an account of verisimilitude at the very heart of which is the assumption that systems of classification must be hierarchical. See Psillos 1999, ­chapter 11 for overview and criticism. 5 For further discussion of these points see Hendry (2016). 6 See Bird (2010) for a distinct though related response to LaPorte’s arguments, and LaPorte (2010) for a reply to both Bird and me. 7 To accommodate the fact that historical scientists may have lacked the expertise to distinguish gold from non-​ gold in every case, Putnam (1975, 235–​238) proposes that we interpret historical usage of ‘gold’ and ‘water’ in terms of current chemical theory. In effect, historical laypeople and scientists are assumed to defer to modern chemistry. 8 Species pluralism is, or course, controversial: for discussion see Ereshefsky (2017, section 3).

368

Science, the Vernacular, the ‘Qua’ Problem 9 See Biggs and Dosanjh (this volume) for relevant discussion. 10 Perhaps that title is irresistible when one wishes to argue against essentialism about chemical kind terms: see Weisberg (2006). 11 One may also ask what grade of gold a particular gold ring is: any grade below 24-​carat is, of course, a mixture with other metals. ‘This is gold’, said of a 9-​carat gold ring, will be said of an object which is mainly non-​gold.

References Abbott, B. 1997 ‘A note on the nature of “water”’ Mind 106, 311–​319. Aronson, Jerrold, Rom Harré and Eileen Cornell Way 1995 Realism Rescued: How Scientific Progress is Possible Chicago, IL: Open Court. Aston, F. W. et al. 1923 ‘Report of the International Committee on Chemical Elements’ Journal of the American Chemical Society 45, 866–​874. Bird, Alexander 2010 ‘Discovering the essences of natural kinds’ in Helen Beebee and Nigel Sabbarton-​Leary (eds.) The Semantics and Metaphysics of Natural Kinds New York: Routledge, 125–​136. Bruzzaniti G. and N. Robotti 1989 ‘The affirmation of the concept of isotopy and the birth of mass spectrography’ Archives Internationales d’Histoires des Sciences 39, 309–​334. Bursten, Julia 2014 ‘Microstructure without essentialism: a new perspective on chemical classification’ Philosophy of Science 81, 633–​653. Chomsky, Noam 1995 ‘Language and nature’ Mind 104, 1–​61. Devitt, Michael and Kim Sterelny 1999 Language and Reality Second Edition Oxford: Blackwell. Dupré, John 1993 The Disorder of Things Cambridge, MA: Harvard University Press. Dupré, John 1999 ‘Are whales fish?’ in D. L. Medin and S. Atran (eds.) Folkbiology, Cambridge, MA: MIT Press, 461–​476. Ellis, B. 2002 The Philosophy of Nature: A Guide to the New Essentialism Chesham: Acumen. Ellis, B. 2009 The Metaphysics of Scientific Realism Durham, UK: Acumen. Ereshefsky, M. 2017 ‘Species’ in E. Zalta (ed.) The Stanford Encyclopedia of Philosophy (Fall 2017 edition), http://​ plato.stanford.edu/​archives/​fall2017/​entries/​species/​. Feyerabend, Paul 1962 ‘Explanation, reduction and empiricism’ in Herbert Feigl and Grover Maxwell (eds.) Scientific Explanation, Space and Time Minneapolis, MN: University of Minneapolis Press, 28–​97. Goodwin,William 2011 ‘Mechanisms and chemical reaction’ in Robin Findlay Hendry, Paul Needham and Andrea I. Woody (eds.) Handbook of the Philosophy of Science, Volume 6: Philosophy of Chemistry. Amsterdam: Elsevier, 310–​327. Hacking, Ian 2007a ‘Putnam’s theory of natural kinds and their names is not the same as Kripke’s’ Principia 11,  1–​24. Hacking, Ian 2007b ‘Natural kinds:  rosy dawn, scholastic twilight’ Royal Institute of Philosophy Supplement 61, 203–​239. Haldane, J. B. S. 1941 Science and Everyday Life Harmondsworth: Penguin. Häggqvist, Sören and Åsa Wikforss 2018 ‘Natural kinds and natural kind terms: Myth and reality’ The British Journal for the Philosophy of Science, 69, 911–933. https://doi.org/10.1093/bjps/axw041 Havstad, Joyce 2018 ‘Messy chemical kinds’ British Journal for the Philosophy of Science 69, 719–​743. Hendry, Robin Findlay 2006 ‘Elements, compounds and other chemical kinds’ Philosophy of Science 73, 864–​875. Hendry, Robin Findlay 2010a ‘The elements and conceptual change’ in Helen Beebee and Nigel Sabbarton-​ Leary (eds.) The Semantics and Metaphysics of Natural Kinds London: Routledge, 137–​158. Hendry, Robin Findlay 2010b ‘Science and everyday life: “water” vs. “H2O”’ Insights 3 http://​www.dur.ac.uk/​ ias/​insights/​volume3/​article23/​. Hendry, Robin Findlay 2012 ‘Chemical substances and the limits of pluralism’ Foundations of Chemistry 14,  55–​68. Hendry, Robin Findlay 2016 ‘Natural kinds in chemistry’ in Grant Fisher and Eric Scerri (eds.) Essays in the Philosophy of Chemistry Oxford: Oxford University Press, 253–​275. Hendry, Robin Findlay 2019 ‘Elements and (first) principles in chemistry’ Synthese https://​doi.org/​10.1007/​ s11229-​019-​02312-​8. Hoefer, Carl and Genoveva Martí 2019 ‘Water has a microstructural essence after all’ European Journal for the Philosophy of Science 9(12). https://doi.org/10.1007/s13194-018-0236-2 Kragh, Helge 2000 ‘Conceptual changes in chemistry: the notion of a chemical element, ca. 1900–​1925’ Studies in History and Philosophy of Modern Physics 31B, 435–​450. Kripke, Saul 1980 Naming and Necessity Cambridge, MA: Harvard University Press. Kuhn, T. S. 1970 The Structure of Scientific Revolutions, Second Edition Chicago, IL: University of Chicago Press.

369

Robin Findlay Hendry LaPorte, Joseph 2004 Natural Kinds and Conceptual Change Cambridge: Cambridge University Press. LaPorte, Joseph 2010 ‘Theoretical identity statements, their truth and their discovery’ in Helen Beebee and Nigel Sabbarton-​Leary (eds.) The Semantics and Metaphysics of Natural Kinds London: Routledge, 104–​124. Mackie, Penelope 2006 How Things Might Have Been: Individuals, Kinds and Essential Properties Oxford: Oxford University Press. Malt, Barbara 1994 ‘Water is not H2O’ Cognitive Psychology 27, 41–​70. Massimi, Michela 2012 ‘Dwatery ocean’ Philosophy 87, 531–​555. Psillos, Stathis 1999 Scientific Realism: How Science Tracks Truth London: Routledge. Putnam, Hilary 1975 ‘The meaning of “meaning”’ in Mind Language and Reality:  Philosophical Papers, Volume 2. Cambridge: Cambridge University Press, 215–​271. Putnam, Hilary 1990 ‘Is water necessarily H2O?’ in James Conant (ed.) Realism with a Human Face Cambridge MA: Harvard University Press, 54–​79. Tahko, Tuomas 2015 ‘Natural kind essentialism revisited’ Mind 124, 795–​822. Thurlow, K. J. 1998 ‘IUPAC Nomenclature Part  1, Organic’ in K.J. Thurlow (ed.) Chemical Nomenclature. Dordrecht: Kluwer, 103–​126. Tobin, Emma 2010 ‘Crosscutting natural kinds and the hierarchy thesis’ in Helen Beebee and Nigel Sabbarton-​ Leary (eds.) The Semantics and Metaphysics of Natural Kinds London: Routledge, 179–​191. van der Vet, Paul 1979 ‘The debate between F.A. Paneth, G. von Hevesy and K. Fajans on the concept of chemical identity’ Janus 92, 285–​303. Weisberg, M. 2006 ‘Water is not H2O!’ in Davis Baird, Lee McIntyre and Eric Scerri (eds.) Philosophy of Chemistry: Synthesis of a New Discipline New York: Springer, 337–​345.

370

PART VII

The Empty Case

28 MILL AND THE MISSING REFERENTS David Braun

Empty names are proper names that fail to refer: they are proper names that are missing their referents. Plausible examples of empty names appear in fiction, myth, and mistaken scientific theories. Empty names raise difficult issues for nearly all semantic theories. They raise particularly difficult problems for Millianism, for Millianism says that there is nothing more to the meaning of a proper name than the object to which it refers.1

Millianism John Stuart Mill thought that proper names denote objects and do not connote attributes. He concluded that proper names have “strictly speaking, no signification” (Mill 1843, I.2.5). Modern Millians mostly agree with Mill.2 They agree that proper names denote, or refer to, objects. They agree that proper names do not connote, or semantically express, attributes. But modern Millians think that proper names do have signification, or meanings, of a certain kind, which they typically call contents or semantic contents. However, even here modern Millians follow Mill in spirit, for they claim that the semantic content of a proper name is merely its referent. Modern Millians also make claims that go beyond Mill. They say that declarative sentences have semantic contents, namely the propositions that they semantically express. Propositions are Russellian, that is, structured entities whose basic constituents are individuals and attributes. For example, the semantic content of ‘London is pretty’ is a structured proposition whose basic constituents are London and the property of being pretty. Propositions are the primary bearers of truth-​value; the truth-​value of a declarative sentence is identical with the truth-​value of the proposition that it semantically expresses. Propositions are also the objects of various attitudes and linguistic relations, such as assertion, denial, belief, and doubt. ‘That’-​clauses refer to the propositions that the sentences embedded within them semantically express. For example, ‘that London is pretty’ refers to the proposition that London is pretty. The belief ascription ‘Milena believes that London is pretty’ is true iff Milena stands in the binary relation expressed by ‘believes’ to the proposition that London is pretty.3 The above Millian theory goes by many names other than ‘Millianism’, including ‘the Naïve Theory’, ‘Russellianism’, ‘Naïve Russellianism’, ‘Neo-​Russellianism’, ‘Millian Russellianism’, and ‘Direct Reference’. I will be focusing on the theory’s claims about proper names, so I will here use the term ‘Millianism’ for it, even though it includes claims that Mill might have rejected.

373

David Braun

Problems that Empty Names Raise for Millianism Empty names raise significant problems for Millianism.4 One problem concerns the meaningfulness of empty proper names. Millianism entails that proper names that fail to refer have no semantic content. But if this is so, then Millianism seemingly entails that empty names are meaningless. For example, assume that ‘Santa Claus’ fails to refer.Then it has no semantic content. So, it is seemingly no more meaningful than any other expression that has no semantic content, such as ‘Kalakunkschmoo’. The preceding problem with empty names has implications for the semantics of sentences. Suppose empty proper names have no semantic content, and are meaningless. Then, plausibly, sentences that contain empty names lack semantic content, and fail to express propositions, and are meaningless. For example, assume again that ‘Santa Claus’ fails to refer.Then any sentence that contains it, such as ‘Santa Claus exists’, has no semantic content, and fails to express a proposition, and is no more meaningful than the “sentence” ‘Kalakunkschmoo exists’. ‘Santa Claus does not exist’ is similarly meaningless. Moreover, Millianism entails that a sentence that expresses no proposition has no truth-​value. So, if ‘Santa Claus exists’ and ‘Santa Claus does not exist’ do not express propositions, then they do not have truth-​values. Furthermore, if sentences that contain empty names fail to express propositions, then ‘that’-​clauses whose embedded sentences contain empty names fail to refer. For example, the ‘that’-​clause ‘that Santa Claus exists’ fails to refer. Therefore, if Millianism is true, the sentence ‘Obama believes that Santa Claus exists’ contains a non-​referring expression, and so cannot be true, and perhaps cannot be false either. Similarly, the sentence ‘Obama does not believe that Santa Claus exists’ is not true. The sentences ‘There is no such thing as the proposition that Santa Claus exists’ and ‘Some people believe that there is no such thing as the proposition that Santa Claus exists’ are truth-​value-​less. There are further, more disturbing, consequences. On Millian theories, it seems that the ‘that’-​ clause ‘that Santa Claus exists’ not only fails to refer, but also fails to have a semantic content.Therefore, ‘Obama believes that Santa Claus exists’ contains an expression that has no semantic content. So, plausibly, this belief ascription is not only truth-​value-​less, but also without semantic content, and therefore meaningless. Furthermore, ‘Obama does not believe that Santa Claus exists’ is meaningless. The sentences ‘There is no such thing as the proposition that Santa Claus exists’ and ‘Some people believe that there is no such thing as the proposition that Santa Claus exists’ not only fail to have truth-​values, but also fail to express propositions, and are meaningless. If ‘Santa Claus exists’ fails to express a proposition, then it is impossible for someone to believe the proposition that it expresses, for there is none. Not even someone who sincerely utters the sentence can believe the proposition it expresses. A parallel point holds for ‘Santa Claus does not exist’. So much for the problems that empty names raise for Millianism. We will continue to focus on Millianism, but parallel problems afflict views that are close cousins of it, including widely accepted versions of possible-​worlds semantics. For example, Worlds-​Millianism, or Willianism, says that proper names are rigid designators. That is, a proper name refers to, or designates, the same object at all possible worlds, if it designates any object at any world. The semantic content of a proper name (if it has one) is an intension, specifically a function on possible worlds whose value at any world is the proper name’s designation at that world (if any).Thus,Willianism entails that if a proper name fails to refer at the actual world, then it fails to refer at all worlds. Therefore, if such a name has an intension, it is the empty intension, that is, the function that has no value at any world. (If functions are sets of ordered pairs, then this function is the empty set.) An atomic sentence that contains an empty name has no truth-​value at any world. If it has any intension, it is also the empty intension.

Are There Any Empty Names? How should Millians reply to the objections from empty names? Perhaps they should follow Mill himself. Mill said, “All names are names of something, real or imaginary” (Mill 1843, I.2.2). If Mill 374

Mill and the Missing Referents

is right, then there are no empty names, and thus empty names do not raise problems for (modern) Millianism. But is he right? I said earlier that plausible examples of empty names come from fiction, myth, and mistaken scientific theories. Let us start with fiction, and try to determine whether each proper name from fiction refers to something. Both ordinary speakers and literary theorists are disposed to utter ‘Sherlock Holmes is a fictional character’. If that sentence is true, then ‘Sherlock Holmes’ refers to a fictional character, and so is not an empty name.The view that Sherlock Holmes is a fictional character has substantial intuitive appeal, but it also raises metaphysical questions. What is the nature of fictional characters? Are they mental entities? Physical entities? Abstract entities? What, if anything, do they have to do with the activities of storytellers? Fictional characters also raise meta-​semantic questions. How do we refer to fictional characters? Can distinct stories, storytellers, and readers refer to the same fictional characters? Do storytellers tell true stories about characters? Ordinary thought and talk about fictional characters raise further puzzles. Speakers and literary theorists alike are inclined to say ‘Sherlock Holmes does not exist’. Put that together with the claim that ‘Sherlock Holmes’ refers to something, and we get the result that ‘Sherlock Holmes’ refers to something that does not exist. That suggests, or entails, that Sherlock Holmes is a nonexistent entity.5 But many ordinary speakers and literary theorists are also inclined to think that Conan Doyle created Sherlock Holmes. Familiar creation processes produce existing things. Can an author create something that does not exist? Let us consider first a view of fictional characters on which they are existing objects.Van Inwagen (1977), Kripke (2013), Salmon (1998), Thomasson (1996, 1999), and Braun (2005) have endorsed views of this sort.Van Inwagen says that fictional characters are theoretical objects that exist in virtue of the activities of authors. Kripke, Thomasson, and Salmon agree, and add that fictional characters are artifacts. Kripke compares fictional characters to nations, which (he says) exist in virtue of the activities and attitudes of their members. Thomasson compares fictional characters to novels, human laws, and contracts. Braun follows them, and compares fictional characters to insurance policies and checking accounts. On this view, fictional characters are, in a certain sense, created by authors, for authors’ activities are sufficient for those objects to exist. The continued existence of fictional characters, like the continued existence of nations and checking accounts, is determined by (typically) long-​lasting events in which (many) people have (coordinated) thoughts, desires, and intentions. Thus, fictional characters (plausibly) cease to exist when no one remembers, or thinks about, them and the stories that (seemingly) refer to them. On this view, we could, very roughly speaking, think of familiar fictional characters as long-​lived events that typically involve mental events of many agents. (But not quite, for the existence conditions of fictional characters are different from these underlying events.) Let us say that the preceding view of fictional characters is the intentional artifact theory, since the view says that fictional characters exist in virtue of agents’ representational mental states and linguistic activities.6 Millians can claim that ‘Sherlock Holmes’ refers to the above sort of intentional artifact. If it does, then the name ‘Sherlock Holmes’ does not refer to a detective, for intentional artifacts are never detectives. However, intentional artifact theorists can reasonably say that according to certain stories written by Conan Doyle, Holmes is a detective. When ordinary speakers and literary critics utter the simple sentence ‘Holmes is a detective’, they typically wish to convey the proposition that, according to certain stories, Holmes is a detective.They may even offer such a paraphrase to hearers who do not realize that Holmes is a fictional character. Ordinary speakers and literary critics might also utter or write ‘Holmes does not exist’. On the intentional artifact theory, this sentence is false. But advocates can reasonably claim that when speakers utter that sentence, they pragmatically convey something that is true, such as that Holmes is (merely) a fictional character, or that no (real, non-​fictional) person has the properties that the stories seemingly attribute to Holmes. Kripke (2013) and Salmon (1998) extend a similar intentional artifact view to proper names from myth. The proper name ‘Santa Claus’ refers to a mythical object, whose existence is determined by the mental and linguistic activities of human beings. Santa Claus does not have a beard, or give toys 375

David Braun

to good little girls and boys, but according to certain myths, “he” does. Similarly, Zeus is a mythical object. Zeus does not throw thunderbolts, but certain stories and myths say that “he” does. Salmon (1998) extends a similar view to mistaken scientific theory. The nineteenth-​century astronomer Le Verrier incorrectly thought that a planet orbiting between Mercury and the Sun was perturbing Mercury’s orbit. He attempted to name that planet ‘Vulcan’. When he did so, he unintentionally created an intentional artifact, a hypothetical object. Kripke (2013), Salmon (1998), Caplan (2004), and Braun (2005) have proposed that perception, imagination, dreams, and hallucinations also create similar intentional artifacts. On the above intentional artifact theory, ‘Holmes’, ‘Santa Claus’, and ‘Vulcan’ are not empty names, and so present no problem for Millianism. On a strong version of this view, all such putatively empty names refer to intentional artifacts. So, Mill was more or less right when he asserted that all proper names name something, either real or imaginary. In fact, we might as well adopt the term ‘imaginary object’ for the class of all fictional, mythical, hypothetical, and other similar (putative) entities mentioned above. Artifactual Millianism combines Millianism with the view that at least some seemingly empty names from fiction, myth, and so on, refer to intentional artifacts. Strong Artifactual Millianism says that all such names refer to intentional artifacts. Artifactual Millianism is one option for Millians, but Millianism is consistent with other views about the nature of imaginary objects. Possibilist Millianism says that these objects are not artifacts, but instead possible, but non-​actual, objects. (Kaplan (1973) and Kripke (2013) discuss such a view.) ‘Sherlock Holmes’, for example, refers to a thing that does not actually exist, but which possibly exists. This merely possible object is not actually a detective, but is possibly a detective (is a detective in some possible world). It possibly has all of the properties that (roughly speaking) the stories ascribe to Holmes. Similarly,Vulcan does not actually exist, but possibly exists with all of the properties that Le Verrier’s theory ascribes to it. Abstract Object Millianism says that imaginary objects are abstract objects. Parsons (1980), for example, holds that ‘Sherlock Holmes’ refers to an abstract object that really is a detective, and lived in London, and so on. But this abstract object is a nonexistent object. So, ‘Holmes’ refers to something that does not exist. Zalta (2000) says that ‘Holmes’ refers to an abstract object that encodes the properties of being a detective and living in London, but this object does not exemplify those properties. Azzouni (this volume) says that ‘Sherlock Holmes’ refers to an object that neither exists nor has properties. On strong versions of the preceding views, all seemingly non-​referring proper names from fiction, myth, and so on in fact refer to merely possible objects, or to abstract objects, and Mill’s view is vindicated. Should Millians accept one of these strong views? All of them raise metaphysical issues. Everett (2005) and Brock (2010) say that there are no plausible identity and existence conditions for fictional characters, if they are intentional artifacts. For replies, see Caplan and Muller (2015). Heimir Geirsson (2013, 146–​ 150) raises related metaphysical problems for intentional artifacts. Various philosophers have argued that there are no merely possible objects (Yagisawa 2018). Thomasson (1996, 1999) objects to the distinctions and assumptions that abstract object theories need to avoid inconsistency and empirical refutation, such as Parson’s distinctions among nuclear, extra-​nuclear, and watered-​down nuclear properties, and Zalta’s distinction between an object’s exemplifying and encoding properties. But perhaps the more significant problems for these Millian views are meta-​semantic.

Yes, There are Some Empty Names Millianism says that the semantic content of a proper name is the object to which it refers. Millianism is consistent with the view that all putatively empty names from fiction, myth, and mistaken scientific theory refer to something. But there are meta-​semantic reasons, independent of Millianism itself, to think that not all such proper names refer. 376

Mill and the Missing Referents

Donnellan’s work (1970, 1974) and Kripke’s work (1980, 2013) suggest that there are two mechanisms for fixing the referents of proper names: causal chains and stipulative reference-​fixing by definite description.7 (An example of the latter is “Let ‘Bobby’ be a name of the current president of Brazil, whosoever that may be”. Kripkean reference-​fixing by definite description does not make the introduced name synonymous with the description.) Neither type of reference-​fixing fits well with the claim that ‘Holmes’ refers to a merely possible object. (See Kaplan 1973 and Kripke 2013.) There are no causal connections between merely possible objects and actual utterances of the name ‘Holmes’. So,‘Holmes’ cannot refer to a merely possible object in virtue of causal chains. Moreover, many distinct, merely possible objects have, in various possible worlds, the properties mentioned by sentences containing ‘Holmes’ in the stories. So, the definite description ‘the (non-​actual, merely possible) object that (possibly) has all of the properties ascribed by sentences containing “Holmes” in the Conan Doyle stories’ is non-​referring and cannot fix the reference of ‘Holmes’ on a single possible object. It is difficult to see how any definite description could be used to fix the reference of ‘Holmes’ on a single possible object. By contrast, it is relatively easy to see how proper names could refer to intentional artifacts, such as checking accounts and insurance policies. Checking accounts are caused to exist by humans who are consciously trying to create them. Checking accounts can also cause events, such as episodes of joy and grief in their owners. A speaker could exploit these causal connections to introduce a name for a particular checking account. Or a speaker could use a definite description to fix the referent of a proper name on a particular checking account. And, in fact, some numerals do serve as proper names of checking accounts. Thus, if fictional characters are intentional artifacts much like checking accounts and insurance policies, then ‘Sherlock Holmes’ could refer to a fictional character. It is also relatively easy to see how proper names could refer to abstract objects of the kind we considered earlier. Causal connections seem to be irrelevant, but there are definite descriptions that might refer to such objects, and that might be used to fix the reference of a name. For example, one might be able to fix the reference of ‘Holmes’ by declaring that it refers to the nonexistent abstract object that has exactly the (nuclear) properties that readers would commonly infer that Conan Doyle’s stories attribute to Holmes (Parsons 1980). One could instead say that ‘Holmes’ refers to the (existent) abstract object that encodes those properties (Zalta 2000). Thus, if imaginary objects are either intentional artifacts or abstract objects, then it is in principle possible for every proper name to refer to either a real or an imaginary object, just as Mill says. Nevertheless, there are examples that strongly suggest that some proper names fail to refer. More cautiously:  there are imaginary examples that suggest that it is possible for a proper name to refer to nothing. Salmon (1998) presents an example that is structurally like the following. Suppose that Maria is planning to go to the animal shelter nearest her home and adopt a dog. She wants to adopt a purebred, female Labrador, and she is rather confident that her local shelter has exactly one such dog. In anticipation of the adoption she says, “I hereby name the (existing) purebred female Labrador in the animal shelter nearest to my home ‘Laney’, if there is exactly one such dog. If there is no such dog, then ‘Laney’ will name nothing. I will now buy some dog food for Laney”. But suppose that there is no (existing) purebred, female Labrador at the animal shelter nearest to her. Then the name ‘Laney’ fails to refer in Maria’s language. After all, Maria consciously and explicitly stipulates that the name ‘Laney’ will refer to nothing if there is no object of the sort that she specifies. Speakers rarely use definite description to introduce names. They even more rarely explicitly stipulate that a name will fail to refer if the reference-​fixing description denotes nothing. Are there more common types of cases in which a proper name fails to refer? That is controversial. Parsons (1980) claims that ‘Vulcan’, in Le Verrier’s language, does not refer to a nonexistent object. Instead, it simply fails to refer. Kripke (2013) also denies that the name ‘Vulcan’ refers to something, in Le Verrier’s language. Braun (2005) agrees with Kripke, partly because he holds that Le Verrier implicitly had reference-​fixing intentions much like those Maria has: Le Verrier intended the name to fail to refer if there were no intra-​Mercurial planet. Salmon, by contrast, holds that ‘Vulcan’ referred in Le Verrier’s language to a hypothetical planet that Le Verrier (unintentionally) created. Salmon (1998) 377

David Braun

says that ‘Vulcan’ also refers to that hypothetical planet in our current language. Braun (2005) says that it is indeterminate whether ‘Vulcan’, in our current language, refers to a hypothetical planet or instead refers to nothing at all. There are also disagreements about proper names from fiction, such as ‘Sherlock Holmes’. Kripke (2013) thinks that the name, in Conan Doyle’s language as he wrote the (initial) story, referred to nothing. Later, speakers who talked about the stories used the name to refer to the fictional character that Doyle created. In our modern language, the name is ambiguous. On one disambiguation, it fails to refer, and on another, it refers to the fictional character. Salmon (1998) holds that Conan Doyle failed to establish a type of use on which ‘Holmes’ is a part of a language as he wrote the (initial) story, but he did thereby create a fictional character, and the name in our current language unambiguously refers to the character. Braun (2005) thinks that ‘Holmes’, in Conan Doyle’s language as he wrote the first story, may have referred to the fictional character that Conan Doyle was creating, or may have failed to refer, depending on his thoughts and intentions as he wrote the story. The name in our current language is unambiguous, but is indeterminate in reference and content. So, there are disagreements about how often proper names fail to refer. But nearly everyone agrees that they sometimes do. Or at least they think it is possible for some proper names to fail to refer.

A Millian Theory of Empty Names If there are empty names, or if it is possible for there to be empty names, then what should a Millian say about them? Consider ‘Laney’ again, and assume that ‘Laney’ refers to nothing, in Maria’s language. Any Millian theory worthy of the name must entail that ‘Laney’ lacks semantic content. But a Millian theory need not admit that sentences containing ‘Laney’ lack semantic content. Consider, for example, the sentence ‘Laney barks’. A  Millian can, in good conscience, follow a suggestion from David Kaplan (1989), and hold that the semantic content of this sentence, in Maria’s language, is a gappy proposition. (See Braun 1993, 2005; Adams et al. 1994, 1997; Salmon 1998; Adams and Fuller 2007. For a related view, see Taylor 2014.) A gappy proposition is a proposition that is missing something. Just as a table can have a missing leg, so a proposition can have a missing constituent.A proposition that is missing a constituent is a gappy proposition.The semantic content of ‘Laney barks’ is a proposition that has the same structure as the semantic content of ‘Fido barks’, where Fido is a real dog. But the semantic content of ‘Laney barks’ has a gap where the referent of ‘Laney’ would appear, if it had one. Millians typically use sets, or something like sets, to represent propositions. (They typically do not identify propositions with sets.) Such Millians might represent the semantic content of ‘Fido barks’ with . Those who accept the existence of gappy propositions can represent the semantic content of ‘Laney barks’ with < , barking> (Salmon 1998). Millians who wish to use only sets to represent propositions can represent the former proposition with and the latter proposition with . Alternatively, Millians can represent propositions with graphs or trees in which the ultimate constituents of propositions (individuals and attributes) are assigned to terminal nodes. Gappy propositions are then represented by graphs or trees that have nothing assigned to one or more of their terminal nodes. Gappy propositions appear to be good Millian candidates for the semantic contents of sentences with empty names. Consider ‘Laney barks’. The verb ‘barks’ has a semantic content. The Millian semantic content of ‘Laney barks’ should contain that semantic content, but also be distinct from it, for the sentence has a more complex syntactic structure than the verb. Gappy propositions capture the difference. For similar reasons, gappy propositions are good candidates for some objects of belief and other attitudes. Imagine that Maria’s friend Veena owns Fido and consciously thinks that Fido barks. 378

Mill and the Missing Referents

Plausibly, when Veena has this thought, an intrinsic mental event, an “event of thinking”, occurs in her mind. Plausibly, this event has a part that represents Fido and a part that represents barking. If so, then the structured proposition is a plausible candidate for being the content of her thinking event and the object of her thinking. Now suppose that the day after Maria completes her introduction of the name ‘Laney’, she utters ‘Laney barks’ while speaking to Veena.Veena knows nothing of the name-​introducing event, but thinks that Maria is a reliable informant and infers that ‘Laney’ refers to a dog.Veena sincerely utters ‘Laney barks’. Plausibly, a mental event occurs in Veena, part of which is suitable for representing an individual and part of which represents barking. But since the (genuine) part that is suitable for representing an individual has no referent or semantic content, the gappy proposition < , barking> is the content of Veena’s thinking event. The mental event occurring in her is structurally similar to the mental event that occurs in her when she thinks that Fido barks. Thus,Veena will be unaware that she is entertaining a gappy proposition as she utters ‘Laney barks’. Let us return to the objections to Millianism from empty names, and see whether gappy propositions can allow Millians to provide plausible replies. Some of the objections claim that empty names, and sentences containing them, are meaningless if Millianism is true. Millians must admit that empty names lack semantic content. But they can nevertheless insist on an important difference between an empty name like ‘Laney’, on the one hand, and a truly meaningless set of sounds or marks, such as ‘Kalakunkschmoo’, on the other. ‘Laney’ is part of a language (Maria’s and Veena’s language). Hearing utterances of sentences that contain ‘Laney’ causes hearers who are familiar with the name to have mental states that are structurally similar to those they have when they hear sentences that contain referring names. The mental states that cause them to utter sentences containing ‘Laney’ are also structurally similar to those that cause them to utter sentences containing referring names. That is a good reason to think that ‘Laney’ is not meaningless in the same way as ‘Kalakunkschmoo’, even though ‘Laney’ lacks semantic content in the Millian’s technical sense. Sentences that contain empty names are also meaningful in this same sense. Plus they have (gappy) Millian semantic contents. Some of the objections claimed that if Millianism is true, then sentences containing empty names lack truth-​value. Millians who use gappy propositions must say that a sentence containing an empty name has a truth-​value iff the gappy proposition that it expresses has one. But do gappy propositions have truth-​values? Let’s begin with atomic gappy propositions. An atomic structured proposition P is true iff: for some n, P has as its sole constituents an n-​place attribute, and n arguments, and the arguments exemplify the attribute. Thus atomic gappy propositions, such as < , barking>, are not true, because they do not contain the right number of arguments for their attributes. Whether they are false depends on what falsity is. Suppose falsity is a property possessed by P iff: for some n, P has as its sole constituents an n-​adic attribute, and n arguments, and the arguments fail to exemplify the attribute. If so, then atomic gappy propositions are not false, because they have insufficient arguments. However, suppose instead that falsity is a property that a proposition has iff it is not true. Then atomic gappy propositions are false. The first theory of falsity seems worse for Millianism, because according to it, atomic gappy propositions are neither true nor false. But how serious a problem would that really be for Millianism? Admittedly, ordinary thinkers sometimes attribute falsehood to sentences that contain non-​referring names. So, Millians who say that atomic gappy propositions are neither true nor false must say that those speakers are making errors. But Millians can hold that such errors are entirely reasonable, for two reasons. Firstly, it will not be evident to ordinary thinkers that sentences with empty names express gappy propositions, even after they become aware that a sentence contains a non-​referring name, for the mental states by which they entertain gappy propositions will be structurally like those by which they grasp non-​gappy propositions. Secondly, all untrue non-​gappy propositions are false. So, ordinary thinkers are unlikely even to consider whether sentences containing empty names 379

David Braun

are neither true nor false, rather than false. These same points hold for positive existential atomic sentences that express atomic gappy propositions, such as ‘Laney exists’. Some of the objections to Millianism claim that negative existentials lack truth-​value, on the Millian view. Whether this is so depends on various issues about negation and truth-​values. Rather than consider all of the views about negation and truth-​value that are consistent with Millianism, let us jump straight to the seemingly worst-​case scenario for Millians. Suppose that the gappy proposition expressed by ‘Laney does not exist’ is neither true nor false. Ordinary speakers, however, would say it is true. But we should not expect ordinary speakers to recognize that the sentence expresses a gappy proposition, even after they realize that the name fails to refer. Furthermore, closely related sentences do have truth-​values. The sentence ‘That Laney exists is not true’ is true, for this sentence attributes lack of truth to the gappy proposition that Laney exists. Moreover, the sentence ‘That Laney exists is not the case’ is also true, for that sentence attributes something like lack of truth to the same proposition. Therefore, ‘It is not the case that Laney exists’ is true. So, even if ‘Laney does not exist’ is neither true nor false, ‘It is not the case that Laney exists’ is true. Ordinary speakers understandably do not consider whether the two claims can differ in truth-​value, for this type of disparity in truth-​value does not occur with sentences containing referring names. A Millian who accepts gappy propositions can reasonably maintain that ‘that’-​clauses that contain empty names refer to gappy propositions. For example, ‘that Laney barks’ refers to the gappy proposition that Laney barks. Attitude ascriptions that contain such ‘that’-​clauses, for instance, ‘Maria believes that Laney barks’, express gappy propositions, and so have semantic contents (and are meaningful).These ascriptions, and the propositions that they express, can have truth-​values.The ascription ‘Veena believes that Laney barks’ is true. Even if ‘Santa Claus’ fails to refer, the sentence ‘There is no such thing as the proposition that Santa Claus exists’ is false. The sentence ‘Some people believe that there is no such thing as the proposition that Santa Claus exists’ may be true. Thus gappy propositions allow Millians to give reasonable replies to the earlier objections.

Objections to Millianism with Gappy Propositions, and Replies However, some philosophers may think that appeals to gappy propositions raise their own problems. Some philosophers may object to gappy propositions because they object to all structured propositions. Others may accept non-​gappy structured propositions, but deny that there are any gappy structured propositions. There is insufficient space here for me to address these metaphysical objections. But see King et al. (2014), Gilmore (forthcoming), and Caplan et al. (forthcoming) for various metaphysical theories of structured propositions. Caplan et al. explicitly argue that their theory is consistent with the existence of gappy propositions. Another objection to Millianism with gappy propositions concerns cognitive significance. Assume, for the sake of argument, that ‘Santa Claus’, ‘Sherlock Holmes’, and ‘Vulcan’ are empty names. Then the preceding gappy proposition theory entails that ‘Sherlock Holmes is Sherlock Holmes’ and ‘Sherlock Holmes is Santa Claus’ express the same gappy proposition. So do ‘Vulcan is a planet’ and ‘Sherlock Holmes is a planet’. Yet a speaker could understand both sentences in one of these pairs, and think that one is true and the other is false. For example, a misinformed student of astronomy might think that ‘Vulcan is a planet’ is true and ‘Sherlock Holmes is a planet’ is false. (See Braun (2005) for details of the objection, and Sawyer (2012) for discussion.) This objection strongly resembles objections from cognitive significance to Millianism that appeal to sentences containing referring names. Millianism entails (for example) that ‘Mark Twain is Mark Twain’ and ‘Mark Twain is Samuel Clemens’ express the same (non-​gappy) proposition. But a speaker could understand both, and so entertain the single proposition that they both express, and yet think that the sentences differ in truth-​value.

380

Mill and the Missing Referents

Millians have provided two different (compatible) replies to these older objections from cognitive significance. One appeals to pragmatics. It says that a speaker who utters the ‘Twain’ sentence may pragmatically convey a different descriptive proposition to hearers than she does when she utters the ‘Clemens’ sentence. Some hearers may believe one of these descriptive propositions without believing the other, and so think that one of the sentences is true and the other is false. (See Soames (2002). See Braun (2003) for limitations on a purely pragmatic explanation.) Another Millian reply emphasizes the use of propositional guises, which Millians sometimes take to be mental representations (Salmon 1986; Braun 2005). Someone who reads or hears the ‘Twain’ and ‘Clemens’ sentences entertains the proposition that they express under two different guises (for example, via two different mental representations). A hearer or reader can rationally believe the proposition that both express under one guise, while failing to believe that proposition under the other guise, and so think that one sentence is true and the other false.8 Both of these Millian theories of cognitive significance are compatible with the view that some sentences express gappy propositions. Millians can appeal to either view (or both) to explain differences in cognitive significance among sentences with empty names. A pragmatic Millian can hold that, though ‘Vulcan is a planet’ and ‘Sherlock Holmes is a planet’ semantically express the same gappy proposition, uttering the first sentence pragmatically conveys a different descriptive proposition than uttering the second sentence. A hearer may believe one of these descriptive propositions but not the other.9 A guise Millian can say that a hearer may entertain the gappy proposition that both express under one guise when hearing the first sentence, and under another guise when hearing the second. Such a hearer may rationally believe the gappy proposition under the first guise, but fail to believe it under the second (Braun 2005). Another objection to Millian gappy proposition theory appeals to attitude ascriptions. On the view given above, the sentences ‘Le Verrier believed that Vulcan was a planet’ and ‘Le Verrier believed that Sherlock Holmes was a planet’ express the same proposition, which attributes to Le Verrier belief in the same gappy proposition. Therefore, if one attribution is true, then so is the other. But it seems obvious that they can differ in truth-​value. Once again, similar issues arise for Millianism with belief attributions that contain referring proper names. ‘Smith believes that Twain is an author’ and ‘Smith believes that Clemens is an author’ express the same proposition on the above Millian theory, and so must have the same truth-​value.Yet it seems that they can differ in truth-​value. Millian replies to these objections again usually rely on pragmatics, or guises, or both.10 Some Millians (Salmon 1986; Soames 2002) say that utterances of these two belief attributions often pragmatically convey different propositions that can differ in truth-​value. Other Millians (Braun 2005) point out that those who understand the belief attributions usually entertain the proposition they express under two different guises, and they may believe the proposition under one guise, but fail to believe that proposition under the other guise. Both sorts of accounts can be extended to ascriptions that attribute belief in gappy propositions. Empty names do pose significant problems for Millianism. But Millians can reasonably hope to overcome those problems, with some help from gappy propositions.

Further Reading Azzouni, Jody. Forthcoming. “Singular Thoughts, Sentences and Propositions of that Which Doesn’t Exist”. In this volume. Braun, David. 2005. “Empty Names, Fictional Names, Mythical Names”. Noûs 39, pp. 596–​631. Brock, Stuart and Everett, Anthony. 2015. Fictional Objects. New York: Oxford University Press. García-​Carpintero, Manuel and Martí, Genoveva (Eds.). 2014. Empty Representations: Reference and Non-​Existence. Oxford: Oxford University Press. Kripke, Saul. 2013. Reference and Existence: The John Locke Lectures. New York: Oxford University Press. Parsons, Terence. 1980. Nonexistent Objects. New Haven, CT: Yale University Press. Salmon, Nathan. 1998. “Nonexistence”. Noûs 32, pp. 277–​319.

381

David Braun Sawyer, Sarah. 2012. “Empty Names”. In Gillian Russell and Delia Graff Fara (Eds.), The Routledge Companion to the Philosophy of Language, New York: Routledge, pp. 153–​162. Thomasson, Amie. 1999. Fiction and Metaphysics. Cambridge: Cambridge University Press.

Notes 1 I concentrate on empty proper names in this paper and ignore empty predicative expressions, such as ‘unicorn’ and ‘is a unicorn’. For the latter, see Kripke (2013), Salmon (1998), Braun (2015), and Sawyer (2015). 2 For modern Millian views, see Kaplan (1989), Salmon (1986, 1998), Soames (1987, 2002), Crimmins and Perry (1989), Richard (1990), Crimmins (1992), Braun (1993, 2005). 3 Some modern Millians, such as Crimmins and Perry (1989), Richard (1990), and Crimmins (1992), agree with the preceding Millian claims about proper names and simple sentences, but disagree with the preceding claims about belief ascriptions. More specifically, they disagree about the semantics of ‘believe’, or ‘that’-​ clauses, or both. See note 10 below. 4 For more detailed presentations of the following anti-​Millian arguments, see Braun (1993 and 2005) and Sawyer (2012). 5 Reimer (2001) argues that ordinary speakers think that fictional characters are nonexistent objects. Reimer does not endorse this view. Parsons (1980) and Azzouni (this volume) do. 6 Kripke (2013), Salmon (1998), and Thomasson (1999) say that fictional characters are abstract artifacts. I have not used their terminology because I am reserving the term ‘abstract object’ for other theories that I discuss below. 7 Perhaps there are other ways in which the reference of a proper name can be fixed, but I will not explore other options here. 8 See Gray (this volume) for more about cognitive significance, including objections to Millianism from cognitive significance, and Millian replies to those objections. 9 See Adams et al. (1994, 1997), and Adams and Fuller (2007). Reimer (2001), Everett (2005), and Braun (2005) criticize pragmatic explanations of cognitive significance for sentences containing empty names. See Sawyer (2015) for a summary. See Mousavian (2015) for further criticisms of pragmatic explanations. 10 Millians can also accept theories of belief ascription on which ‘Smith believes that Twain is an author’ and ‘Smith believes that Clemens is an author’ can express different propositions, and so can differ in truth-​value. See Crimmins and Perry (1989), Richard (1990), and Crimmins (1992) and note 3 above. Such views can be extended to belief ascriptions whose ‘that’-​clauses refer to gappy propositions.

References Adams, Fred and Fuller, Gary. 2007. “Empty Names and Pragmatic Implicatures”. Canadian Journal of Philosophy 37, pp. 449–​462. Adams, Fred, Fuller, Gary, and Stecker, Robert. 1994. “Vacuous Singular Terms”. Mind and Language 9, pp. 387–​401. Adams, Fred, Fuller, Gary, and Stecker, Robert. 1997. “The Semantics of Fictional Names”. Pacific Philosophical Quarterly 78, pp. 128–​148. Braun, David. 1993. “Empty Names”. Noûs 27, pp. 449–​469. Braun, David. 2003. “Review of Scott Soames’s Beyond Rigidity”. Linguistics and Philosophy 26, pp. 367–​379. Braun, David. 2005. “Empty Names, Fictional Names, Mythical Names”. Noûs 39, pp. 596–​631. Braun, David. 2015. “Wondering about Witches”. In Stuart Brock and Anthony Everett (Eds.), Fictional Objects. New York: Oxford University Press, pp. 71–​113. Brock, Stuart. 2010.“The Creationist Fiction: The Case against Creationism about Fictional Entities”. Philosophical Review 119, pp. 337–​364. Brock, Stuart and Everett, Anthony. 2015. Fictional Objects. New York: Oxford University Press. Caplan, Ben. 2004. “Creatures of Fiction, Myth, and Imagination”. American Philosophical Quarterly 41, pp. 331–​337. Caplan, Ben and Muller, Cathleen. 2015. In Stuart Brock and Anthony Everett (Eds.), Fictional Objects. New York: Oxford University Press, pp. 174–​207. Caplan, Ben, Tillman, Chris, and Nutting, Eileen. Forthcoming. “Hylomorphic Propositions”. In Chris Tillman (Ed.), Routledge Handbook of Propositions. New York: Routledge. Crimmins, Mark. 1992. Talk about Beliefs. Cambridge, MA: MIT Press. Crimmins, Mark and Perry, John. 1989. “The Prince and the Phone Booth: Reporting Puzzling Beliefs”. Journal of Philosophy 86, pp. 684–​711. Donnellan, Keith. 1970. “Proper Names and Identifying Descriptions”. Synthese 21, pp. 335–​358.

382

Mill and the Missing Referents Donnellan, Keith. 1974. “Speaking of Nothing”. Philosophical Review 83, pp. 3–​31. Everett, Anthony. 2003. “Empty Names and ‘Gappy’ Propositions”. Philosophical Studies 116, pp. 1–​36. Everett, Anthony. 2005. “Against Fictional Realism”. Journal of Philosophy 102, pp. 624–​649. García-​Carpintero, Manuel and Martí, Genoveva (Eds.). 2014. Empty Representations: Reference and Non-​Existence. Oxford: Oxford University Press. Geirsson, Heimir. 2013. Philosophy of Language and Webs of Information. New York: Routledge. Gilmore, Cody. Forthcoming. “Why 0-​adic Relations Have Truth Conditions”. In Chris Tillman (Ed.), Routledge Handbook of Propositions. New York: Routledge. Kaplan, David. 1973. “Bob and Carol and Ted and Alice”. In K. J. J. Hintikka, J. M. E. Moravcsik, and P. Suppes (Eds.), Approaches to Natural Language. Dordrecht: Springer, pp. 490–​518. Kaplan, David. 1989. “Demonstratives”. In Joseph Almog, John Perry, and Howard Wettstein (Eds.), Themes from Kaplan. New York: Oxford University Press, pp. 485–​563. King, Jeffrey, Soames, Scott, and Speaks, Jeff. 2014. New Thinking about Propositions. New  York:  Oxford University Press. Kripke, Saul. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press. Kripke, Saul. 2013. Reference and Existence: The John Locke Lectures. New York: Oxford University Press. Mill, John Stuart. 1843. A System of Logic. London: John W. Parker. Mousavian, Seyed. 2015. “Pragmatics of No Reference”. Mind & Language 30, pp. 95–​116. Parsons, Terence. 1980. Nonexistent Objects. New Haven, CT: Yale University Press. Reimer, Marga. 2001. “The Problem of Empty Names”. Australasian Journal of Philosophy 79, pp. 491–​506. Richard, Mark. 1990. Propositional Attitudes. Cambridge: Cambridge University Press. Salmon, Nathan. 1986. Frege’s Puzzle. Cambridge, MA: MIT Press. Salmon, Nathan. 1998. “Nonexistence”. Noûs 32, pp. 277–​319. Sawyer, Sarah. 2012. “Empty Names”. In Gillian Russell and Delia Graff Fara (Eds.), The Routledge Companion to the Philosophy of Language, New York: Routledge, pp. 153–​162. Sawyer, Sarah. 2015. “The Importance of Fictional Properties”. In Stuart Brock and Anthony Everett (Eds.), Fictional Objects. New York: Oxford University Press, pp. 208–​229. Soames, Scott. 1987. “Direct Reference, Propositional Attitudes, and Semantic Content”. Philosophical Topics 15, pp.  47–​87. Soames, Scott. 2002. Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. Oxford: Oxford University Press. Taylor, Kenneth A. 2014. “The Things We Do with Empty Names: Objectual Representations, Non-​Veridical Language Games, and Truth Similitude”. In Manuel García-​Carpintero and Genoveva Martí (Eds.), Empty Representations: Reference and Non-​Existence. Oxford: Oxford University Press, pp. 183–​214. Thomasson, Amie. 1996. “Fiction, Modality, and Dependent Abstracta”. Philosophical Studies 84, pp. 295–​320. Thomasson, Amie. 1999. Fiction and Metaphysics. Cambridge: Cambridge University Press. Tillman, Chris (Ed.). Forthcoming. Routledge Handbook of Propositions. New York: Routledge. Van Inwagen, Peter. 1977. “Creatures of Fiction”. American Philosophical Quarterly 14, pp. 299–​308. Yagisawa,Takashi. 2018. “Possible Objects”. In Edward Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Spring 2018 Edition), https://​plato.stanford.edu/​archives/​spr2018/​entries/​possible-​objects/​. Zalta, Edward. 2000. “Pretense Theory and Abstract Object Theory”. In Anthony Everett and Thomas Hofweber (Eds.), Empty Names, Fiction, and the Puzzles of Non-​Existence. Stanford, CA: CSLI Publications, pp. 117–​147.

383

29 FREGEAN THEORIES OF NAMES FROM FICTION Ben Caplan

1  Introduction J. K. Rowling introduced the name ‘Hermione Granger’ in the novel Harry Potter and the Sorcerer’s Stone and used it in a number of subsequent novels, including Harry Potter and the Prisoner of Azkaban.1 ‘Hermione Granger’ is a name from fiction.2 Gottlob Frege (1892) discusses names from fiction in “On Sense and Reference”. In this paper, I discuss views about names from fiction that are based on, or inspired by, what Frege says there. In sections 2 and 3, I discuss views on which names from fiction refer to numbers or properties. In sections 4 and 5, I discuss the view that names from fiction don’t refer to anything but express senses given by definite descriptions. We can distinguish three kinds of sentences (or three kinds of uses of sentences) that contain names from fiction. (1) Hermione Granger can attend multiple classes at the same time. (2) According to Harry Potter and the Prisoner of Azkaban, Hermione Granger can attend multiple classes at the same time. (3) Hermione Granger is a fictional character. (1) is fictive when it’s used in telling a story; (2) is metafictive when it’s used, from a perspective external to the story, to report what goes on in the story; and (3) is transfictive when it’s used from a perspective external to the story but not to report what goes on in the story.3 One difference between the metafictive (2)  and the transfictive (3)  is that prefixing (3)  with ‘According to Harry Potter and the Prisoner of Azkaban’ doesn’t yield a truth. (4) According to Harry Potter and the Prisoner of Azkaban, Hermione Granger is a fictional character. (4) is false. (In the novel, Hermione is a person, not a fictional character.)

2  Stipulation 2.1  The View According to stipulationism about names from fiction, ‘Hermione Granger’ is stipulated to refer to the number 0, as is any other name from fiction.4 384

Fregean Theories of Names from Fiction

Frege doesn’t discuss stipulationism about names from fiction, but he considers an analogous view about definite descriptions that don’t refer, like ‘the celestial body most distant from the earth’ or ‘the celestial body distinct from the earth’.5 Speaking of definite descriptions, he says, “an expression of the kind in question must actually always be assured of a referent, by means of a special stipulation, e.g. by the convention that 0 shall count as its referent, when the predicate applies to no object or to more than one”.6 (Frege doesn’t say who gets to stipulate what the referent is, or how the convention is established, or why the number that gets picked is 0 rather than, say, 1 or –​17.)

2.2  Fictive and Transfictive Sentences According to stipulationism about names from fiction, (1) is true if and only if (1S) is.7 (1) Hermione Granger can attend multiple classes at the same time. (1S) 0 can attend multiple classes at the same time. But, one might object, (1) is true, while (1S) is false. Stipulationists about names from fiction might reply that we don’t care about the truth-​values of fictive sentences like (1), so it doesn’t matter if stipulationism about names from fiction gets their truth-​values wrong. Frege (1892, 33) says, In hearing an epic poem, for instance, apart from the euphony of the language we are interested only in the sense of the sentences and the images and feelings thereby aroused. The question of truth would cause us to abandon aesthetic delight for an attitude of scientific investigation. But this reply might be at odds with the paradox of fiction, in which we care about what happens in a work of fiction even if we know that it’s just a story.8 Alternatively, stipulationists about names from fiction might replace fictive sentences with metafictive ones and say something like the following. “(1) is false. (So (1) and (1S) have the same truth-​value: they’re both false.) But that’s okay as long as (2) According to Harry Potter and the Prisoner of Azkaban, Hermione Granger can attend multiple classes at the same time is true”.9 According to stipulationism about names from fiction, (3) is true if and only if (3S) is. (3) Hermione Granger is a fictional character. (3S) 0 is a fictional character. But, one might object, (3) is true, while (3S) is false. Even if stipulationists about names from fiction can replace fictive sentences like (1)  with metafictive sentences like (2), it’s not as easy to replace transfictive sentences like (3) with metafictive sentences. Stipulationists about names from fiction can’t say, “(3) is false. (So (3) and (3S) have the same truth-​value: they’re both false.) But that’s okay as long as the straightforward metafictive counterpart of (3) is true”. For the straightforward metafictive counterparts of (3) is (4). (4) According to Harry Potter and the Prisoner of Azkaban, Hermione Granger is a fictional character. And (4) isn’t true. 385

Ben Caplan

2.3  Identities and Predications According to stipulationism about names from fiction, (5) and (6) are true if and only if (5S) and (6S) are. (5) (6) (5S) (6S)

Hermione Granger is identical with Luna Lovegood. Hermione Granger is a natural number. 0 is identical with 0. 0 is a natural number.

(‘Luna Lovegood’ is a name that J. K. Rowling introduced in Harry Potter and the Order of the Phoenix.10) But, one might object, (5) and (6) are false, while (5S) and (6S) are true. Stipulationists about names from fiction could reply that (sometimes) different names from fiction refer to different numbers. The modification would allow (5) to come out false, if it has the same truth-​value as (5S*). (5S*)

0 is identical with 1.

But the modification doesn’t address the problem with (6), which arises because, according to stipulationism about names from fiction, names from fiction refer to objects that have extraneous mathematical properties.

3  Reference to Sense 3.1  The View On Frege’s (1892, 26, 31) view, the name ‘J. K. Rowling’ refers to an object11 (namely, Joanne Rowling) and expresses a sense. This sense presents Rowling. Perhaps it’s given by ‘the author of a highly successful series of novels about a bespectacled wizard’ and is the property being the author of a highly successful series of novels about a bespectacled wizard. On one view, ‘Hermione Granger’ doesn’t refer to anything but expresses a sense. This sense doesn’t present anything. Perhaps it’s given by ‘the Muggle-​born Gryffindor who is best friends with Harry and Ron and often uses her dry humor, deft recall, and encyclopedic knowledge to save them’—​or, for short, ‘the person who saves Harry and Ron all the time’. I put off discussing that view until section 4; in this section, I discuss a different view that appeals to senses. On the reference-​to-​sense view, ‘Hermione Granger’ refers to the sense given by ‘the person who saves Harry and Ron all the time’, which is being the person who saves Harry and Ron all the time.12 And, similarly, other names from fiction refer to senses given by definite descriptions.

3.2  Fictive and Transfictive Sentences On the reference-​to-​sense view, (1) is true if and only if (1RS) is.13 (1) Hermione Granger can attend multiple classes at the same time. (1RS) Being the person who saves Harry and Ron all the time can attend multiple classes at the same time. But, one might object, (1) is true, while (1RS) is false. Proponents of the reference-​to-​sense view might replace fictive sentences with metafictive ones and say something like the following. “(1) is false. But that’s okay as long as

386

Fregean Theories of Names from Fiction

(2) According to Harry Potter and the Prisoner of Azkaban, Hermione Granger can attend multiple classes at the same time is true”.14 On the reference-​to-​sense view, (3) is true if and only if (3RS) is. (3) Hermione Granger is a fictional character. (3RS) Being the person who saves Harry and Ron all the time is a fictional character. And, one might object, (3) is true, while (3RS) is false. But proponents of the reference-​to-​sense view might reply as follows. “(3RS) is true. It’s absurd to think that a number is a fictional character. But it’s not absurd to think that a property is”.15

3.3 Identities According to Harry Potter and the Chamber of Secrets, ‘Lord Voldemort’ and ‘Tom Riddle’ refer to the same person.16 On the reference-​to-​sense view, ‘Lord Voldemort’ and ‘Tom Riddle’ might refer to different senses: for example, being Harry’s archnemesis and being the sympathetic guy Ginny talks to. In that case, on the reference-​to-​sense view, (7) is true if and only if (7RS) is. (7) Lord Voldemort is identical with Tom Riddle. (7RS) Being Harry’s archnemesis is identical with being the sympathetic guy Ginny talks to. But, one might object, (7) is true, while (7RS) is false.17 Proponents of the reference-​to-​sense view might reply that it’s okay if (7) is false, as long as (7) is true in the novel. On this reply, there are two villains—​Lord Voldemort and Tom Riddle—​that, in the novel, are the same person. But it might seem that, on the contrary, there’s one villain that, in the novel, goes by two names (‘Lord Voldemort’ and ‘Tom Riddle’).18

4  Sense without Reference 4.1  The View According to descriptivism about names from fiction, ‘Hermione Granger’ doesn’t refer to anything but expresses a sense given by ‘the person who saves Harry and Ron all the time’.19 And, similarly, other names from fiction don’t refer to anything but express senses given by definite descriptions. If ‘Hermione Granger’ expresses a sense given by ‘the person who saves Harry and Ron all the time’, then the definite description fixes the reference of ‘Hermione Granger’.20 Reference-​fixing: For any name n and any definite description d, d fixes the reference of n if and only if (i), for any object o, if n refers to o, then it’s because d refers to o that n refers to o; and (ii), for any object o, if n doesn’t refer to o, then it’s because d doesn’t refer to o that n doesn’t refer to o. According to descriptivism about names from fiction, ‘Hermione Granger’ doesn’t refer to anything, and it’s because ‘the person who saves Harry and Ron all the time’ doesn’t refer to anything that ‘Hermione Granger’ doesn’t refer to anything either.

387

Ben Caplan

4.2  Fictive and Transfictive Sentences According to descriptivism about names from fiction, (1) is true if and only if (1D) is.21 (1) Hermione Granger can attend multiple classes at the same time. (1D) The person who saves Harry and Ron all the time can attend multiple classes at the same time. But, one might object, (1) is true, while (1D) isn’t. (On Frege’s (1892, 32) view, (1D) is neither true nor false if ‘the person who saves Harry and Ron all the time’ doesn’t refer to anything.) Frege says that we don’t care whether the fictive (1) is true. If we do care, though, perhaps we can replace it with (2) According to Harry Potter and the Prisoner of Azkaban, Hermione Granger can attend multiple classes at the same time. And (2) is true.22 According to descriptivism about names from fiction, (3) is true if and only if (3D) is. (3) Hermione Granger is a fictional character. (3D) The person who saves Harry and Ron all the time is a fictional character. But, one might object, (3) is true, while (3D) isn’t (since ‘the person who saves Harry and Ron all the time’ doesn’t refer to anything). Descriptivists about names from fiction might say that (3) is true if and only if something more complicated like (3D*) is true. (3D*) The person who saves Harry and Ron all the time doesn’t exist, but there’s a fiction according to which the person who saves Harry and Ron all the time exists. (3D*) is at least partly metafictive; it contains an ‘according to the fiction’ operator. The strategy is to replace transfictive sentences like (3) with sentences that are at least partly metafictive. Descriptivists about names from fiction can then say that (3D*)—​and hence (3)—​is true.23 The first conjunct of (3D*) raises technical complications. On Frege’s view, atomic sentences that contain definite descriptions that don’t refer to anything are neither true nor false. So, unless descriptivists about names from fiction have a complicated story to tell about positive existentials like ‘The person who saves Harry and Ron all the time exists’, that sentence will come out neither true nor false; and, unless they have a complicated story to tell about negation, the negation of that sentence—​‘The person who saves Harry and Ron all the time doesn’t exist’—​will also come out neither true nor false. And, if the first conjunct of (3D*) is neither true nor false, then (3D*) isn’t true. But perhaps descriptivists about names from fiction could tell a complicated story about positive existentials.24 Or perhaps they could use exclusion rather than choice negation and get the negation of a truth-​valueless positive existential to come out true.25

4.3  More Transfictive Sentences According to descriptivism about names from fiction, (8)–​(11) are true if and only if (8D)–​(11D) are. (8) Hermione Granger appears in Harry Potter and the Prisoner of Azkaban. (9) Emma Watson portrayed Hermione Granger in eight films. (10) J. K. Rowling created Hermione Granger. 388

Fregean Theories of Names from Fiction

(11) Hermione Granger is J. K. Rowling’s favorite character. (8D) The person who saves Harry and Ron all the time appears in Harry Potter and the Prisoner of Azkaban. (9D) Emma Watson portrayed the person who saves Harry and Ron all the time in eight films. (10D) J. K. Rowling created the person who saves Harry and Ron all the time. (11D) The person who saves Harry and Ron all the time is J. K. Rowling’s favorite character. But, one might object, (8)–​(11) are (or could be) true, while (8D)–​(11D) aren’t.26 Descriptivists about names from fiction might say that (8)–​(10) are true if and only if more complicated sentences like (8D*)–​(10D*) are true. According to Harry Potter and the Prisoner of Azkaban, the person who saves Harry and Ron all the time exists. (9D*) Emma Watson is such that there are eight films according to each of which she is the person who saves Harry and Ron all the time. (10D*) There’s a fiction by J. K. Rowling according to which the person who saves Harry and Ron all the time exists and there’s no earlier fiction by anyone else according to which the person who saves Harry and Ron all the time exists. (8D*)

And, descriptivists about names from fiction might say, (8D*)–​(10D*) are true.27 It’s not obvious how to extend this piecemeal reply to other transfictive sentences, including (11), but a more systematic reply is available to descriptivists about names from fiction.28 According to realism about fictional characters, there are fictional characters.29 In “Fictionalism about Fictional Characters”, Stuart Brock (2002) uses an ‘according to realism about fictional characters’ operator to analyze certain sentences that contain names from fiction. Adopting Brock’s analysis, descriptivists about names from fiction can say that (8)–​(11) are true if and only if (8FD)–​(11FD) are.30 (8FD) According to realism about fictional characters, the person who saves Harry and Ron all the time appears in Harry Potter and the Prisoner of Azkaban. (9FD) According to realism about fictional characters, Emma Watson portrayed the person who saves Harry and Ron all the time in eight films. (10FD) According to realism about fictional characters, J. K. Rowling created the person who saves Harry and Ron all the time. (11FD) According to realism about fictional characters, the person who saves Harry and Ron all the time is J. K. Rowling’s favorite character. Descriptivists about names from fiction can then say that (8FD)–​(11FD)—​and hence (8)–​(11)—​are true.31

4.4 Alternatives According to descriptivism about names from fiction, senses are descriptive—​that is, they’re given by definite descriptions—​and object-​independent: that is, some senses don’t present anything. Frege (1892, 27 n2) suggests that the sense of ‘Aristotle’ is descriptive; in particular, he suggests that it’s given by ‘the pupil of Plato and teacher of Alexander the Great’ or ‘the teacher of Alexander the Great who was born in Stagira’. And he suggests that the sense of ‘Odysseus’ (from Homer’s Odyssey) is object-​independent, because the name ‘Odysseus’ expresses a sense even though it doesn’t refer to anything. Frege (1892, 32) says, 389

Ben Caplan

Is it possible that a sentence as a whole has only a sense, but no referent? At any rate, one might expect that such sentences occur, just as there are parts of sentences having sense but no referent. And sentences which contain proper names without referents will be of this kind. The sentence “Odysseus was set ashore at Ithaca while sound asleep” obviously has a sense. But since it is doubtful whether the name “Odysseus”, occurring therein, has a referent, it is also doubtful whether the whole sentence has one.32 But there are alternative views on which senses are non-​descriptive or object-​dependent. First, perhaps senses are non-​descriptive.33 Frege suggests as much.34 For example, in the Begriffsschrift, he says that the “mode of determination” associated with a name might determine a point “immediately through intuition”.35 A  sense that presents something in this immediate way might not be given by a definite description. And, in “The Thought”, Frege (1918, 66) says, “everyone is presented to himself in a special and primitive way, in which he is presented to no one else”. A sense that presents someone in this first-​person way isn’t given by a definite description like ‘the speaker’. For the sense given by ‘the speaker’ doesn’t seem to be primitive, and it presents everyone in the same way. A view on which senses are non-​descriptive doesn’t face Saul Kripke’s (2013) objections to descriptivism about names from fiction, which are discussed in the next section.36 But it still faces the objections discussed in sections 4.2 and 4.3. For any view on which ‘Hermione Granger’ doesn’t refer to anything faces objections about the truth-​values of transfictive sentences like (3) and (8)–​(11). Second, perhaps senses are object-​dependent.37 If senses are modes of presentation of objects or ways of thinking about them, then perhaps every sense presents something. For there might not be a mode of presenting an object if there’s no object there to be presented, just as there might not be a way of thinking about an object if there’s no object there to be thought about.38 But a view on which senses are object-​dependent still faces the objections discussed in sections 4.2 and 4.3. And, unlike descriptivists about names from fiction, proponents of such a view might not be able to adopt Brock’s analysis. For, if ‘Hermione Granger’ doesn’t express a sense and sentences that contain it don’t express thoughts, then it’s not clear whether there are any thoughts (or propositions) there to be true according to realism about fictional characters.

5  Kripke’s Objections 5.1 Background At the end of Lecture I in Reference and Existence, Kripke (2013, 26–​28) argues against descriptivism about names from fiction, which he calls “the Frege-​Russell theory”.39 In this section, I discuss two of Kripke’s objections.40 But first I sketch some background. In Naming and Necessity, Kripke (1980) argues against a number of theses, including the following four. (1) To every name or designating expression ‘X’, there corresponds a cluster of properties, namely the family of those properties φ such that A believes ‘φX’. (2) One of those properties, or some conjointly, are believed by A to pick out some individual uniquely. (3) If most, or a weighted most, of the φ’s are satisfied by one unique object y, then y is the referent of ‘X’. (4) If the vote yields no unique object, ‘X’ does not refer.41 Kripke uses a number of counterexamples to these theses. Here I  rehearse two pairs of counterexamples: ‘Cicero’ and ‘Feynman’, and ‘Gödel’ and ‘Peano’. 390

Fregean Theories of Names from Fiction

First, Kripke uses ‘Cicero’ and ‘Feynman’ as counterexamples to Theses (2) and (4). He says, Let’s see if Thesis (2) is true … In fact, most people, when they think of Cicero, just think of a famous Roman orator, without any pretension to think either that there was only one famous Roman orator or that one must know something else about Cicero to have a referent for the name. Consider Richard Feynman, to whom many of us are able to refer. He is a leading contemporary theoretical physicist. Everyone here (I’m sure!) can state the contents of Feynman’s theories so as to differentiate him from Gell–​Mann. However, the man in the street, not possessing these abilities, may still use the name ‘Feynman’. When asked he will say: well he’s a physicist or something. He may not think that this picks out anyone uniquely. I still think he uses the name ‘Feynman’ as a name for Feynman.42 Thesis (4): If the vote yields no unique object the name doesn’t refer. Really this case has been covered before—​has been covered in my previous examples … the vote may not yield a unique object, as in the case of Cicero or Feynman.43 Contrary to Thesis (2), speakers use ‘Cicero’ to refer to Cicero, and ‘Feynman’ to refer to Feynman, even if they don’t think that they could uniquely describe Cicero or Feynman. And, contrary to Thesis (4), ‘Cicero’ refers to Cicero, even though no unique object has being a famous Roman orator, just as ‘Feynman’ refers to Feynman, even though no unique object has being a leading theoretical physicist. Second, Kripke (1980, 83–​85) uses ‘Gödel’ and ‘Peano’ as counterexamples to Thesis (3). We are asked to imagine cases in which, contrary to Thesis (3), someone uniquely satisfies the description associated with the name, but the name doesn’t refer to that person: perhaps Schmidt is the person who discovered the incompleteness of arithmetic, but ‘Gödel’ doesn’t refer to Schmidt (since it refers to Gödel); or perhaps Dedekind is the person who discovered the Peano axioms, but ‘Peano’ doesn’t refer to Dedekind (since it refers to Peano).

5.2  Kripke’s First Objection Kripke uses ‘Sam Smith’ as a counterexample to the fictional analogs of Theses (2) and (4).44 Here are the fictional analogs of Theses (1)–​(4). (1*) To every name from fiction ‘X’, there corresponds a cluster of properties, namely the family of those properties φ such that according to the fiction ‘φX’. (2*) One of those properties, or some conjointly, are such that according to the fiction it or they pick out some individual uniquely. (3*) According to the fiction, if most, or a weighted most, of the φ’s are satisfied by one unique object y, then y is the referent of ‘X’. (4*) According to the fiction, if the vote yields no unique object, ‘X’ doesn’t refer. Kripke (2013, 26) says, One can see clearly now what an incorrect account of the facts about fiction the Frege-​ Russell theory gives, even supposing it is correct in a non-​fictional case. First, it says that the proper name means ‘the thing satisfying the properties in the story’. To affirm existence is to affirm that there is a unique thing satisfying the properties in the story. This is radically false. Why the uniqueness? Why should the story say anything that even putatively identifies an object uniquely? It talks about, very fleetingly, a certain tall man, Sam Smith, who accosted the hero on the corner of some street. Now does that mean that only one tall man ever accosted the hero of the story on some street? Or it may say just that Sam Smith is a tall man and that the hero knew him: it says ‘If 391

Ben Caplan

I were a tall man like Sam Smith, I would be such-​and-​such’. Does that really imply that Sam Smith is the unique tall man, or even the unique tall man about whom the hero ever so mused? It need not even mean that there is a unique tall man called ‘Sam Smith’ about whom the hero so mused. Perhaps there are several. Still, he was represented as musing about a particular tall man, Sam Smith, on this particular occasion. The story need not even putatively assert uniqueness.45 Contrary to Thesis (2*), there aren’t any properties corresponding to ‘Sam Smith’ that, according to the fiction, pick out some individual uniquely. And, contrary to Thesis (4*), it’s not the case that, according to the fiction, if the vote yields no unique object, then ‘Sam Smith’ doesn’t refer to anything. (According to the fiction the following is the case: the vote yields no unique object, but ‘Sam Smith’ refers to Sam Smith.) But, unlike ‘Cicero’ or ‘Feynman’, ‘Sam Smith’ isn’t a counterexample to Thesis (4). ‘Sam Smith’ doesn’t refer to anything (at least according to descriptivism about names from fiction). So, even if no unique object has being a tall man called ‘Sam Smith’ who the hero mused about,Thesis (4) is automatically true (since its consequent is true). My interest here isn’t in Theses (1*)–​(4*), the fictional analogs of Theses (1)–​(4). Rather my interest here is in descriptivism about names from fiction, understood as the view that names from fiction don’t refer to anything but express senses given by definite descriptions. And ‘Sam Smith’ isn’t a counterexample to descriptivism about names from fiction thus understood. Even if it’s a counterexample to Theses (2*) and (4*), ‘Sam Smith’ might express a sense given by ‘the tall man called “Sam Smith” who the hero mused about’. Descriptivists about names from fiction can say that ‘the tall man called “Sam Smith” who the hero mused about’ fixes the reference of ‘Sam Smith’: the definite description doesn’t refer to anything, and it’s because the definite description doesn’t refer to anything that ‘Sam Smith’ doesn’t refer to anything either.

5.3  Kripke’s Second Objection Kripke uses a case of accidental satisfaction as a counterexample to Thesis (3). He says, Second, it is held that if there is an object which uniquely satisfies the properties attributed to the object in a story, then it is not a story: the thing really exists, and the account is not fictional at all. But the common practice of authors is just the very reverse.They print at the beginning of their story: ‘The names used in this story are fictional, and any resemblance to characters living or dead is purely coincidental’. Suppose a person, believing himself to be possessed of a valid suit for invasion of privacy, sues the author of such a story, and proves in court that he uniquely satisfies the properties mentioned in that story. Then will the judge necessarily rule on behalf of the plaintiff? I think not. Suppose the author can show that he never heard of this man; that he definitely wasn’t writing about him; that it was indeed a coincidence, just as he said. Then a reasonable judge would rule against the plaintiff, against Frege, against Russell, and against Wittgenstein, and hold that the author had a valid defense, though this person uniquely fits the story.46 This case echoes the ‘Gödel’ and ‘Peano’ cases from Naming and Necessity. Here, we’re asked to imagine that someone accidentally satisfies the description associated with a name from fiction, but the name doesn’t refer to that person (not because it refers to anyone else, but rather because it doesn’t refer to anything at all). Suppose that there’s someone—​call her ‘HG’—​in the real world who J. K. Rowling has never heard of but who saves Harry and Ron (or the bespectacled wizard and his gangly friend) all the time. Contrary to Thesis (3), ‘Hermione Granger’ doesn’t refer to HG, even though HG uniquely satisfies 392

Fregean Theories of Names from Fiction

the description ‘person who saves Harry and Ron all the time’ (or ‘person who saves the bespectacled wizard and his gangly friend all the time’). This case is a counterexample to descriptivism about names from fiction. Descriptivists about names from fiction can’t say that ‘Hermione Granger’ expresses a sense given by ‘the person who saves Harry and Ron all the time’, nor can they say that the definite description fixes the reference of ‘Hermione Granger’. Since ‘Hermione Granger’ doesn’t refer to anything, it doesn’t refer to HG. But it’s not because ‘the person who saves Harry and Ron all the time’ doesn’t refer to HG that ‘Hermione Granger’ doesn’t refer to HG. For, in this case, ‘the person who saves Harry and Ron all the time’ does refer to HG.47

Acknowledgments For discussion, thanks to students at Kansas, Manitoba, Ohio State, and Umeå over the years, especially in the spring of 2015; to participants at McKinseyfest at Wayne State and at a panel at the CPA in 2019; and to Stephen Biggs, David Friedell, Jenn Gleason, Eric Hiddleston, Michael McKinsey, Erin Mercurio, Eileen Nutting, Murali Ramachandran, Joshua Spencer, and Jamie Tappenden.Thanks especially to David Braun for last-​minute comments and discussion.

Notes 1 See Rowling (1998, 1999b). Harry Potter and the Sorcerer’s Stone was originally published in the UK as Harry Potter and the Philosopher’s Stone. See Rowling (1997). 2 Not every name that occurs in a work of fiction is a name from fiction in the relevant sense. For example, ‘King’s Cross’ occurs in Harry Potter and the Sorcerer’s Stone, but it’s not a name from fiction in the relevant sense. (It was used to refer to a train station in London prior to the novel.) For a related distinction at the level of fictional characters, see Parsons (1980, 51–​52). 3 See Currie (1990, 11, 158, 171). For ease of exposition, I talk about sentences, rather than uses of them, in what follows. 4 Brock (2002, 2) mentions but doesn’t endorse the view. 5 Frege (1892, 28) uses ‘the celestial body most distant from the earth’ as an example of a definite description that doesn’t refer to anything. 6 Frege (1892, 41 n9). Frege doesn’t endorse this view in “On Sense and Reference”. But he later endorses a related view. In the first volume of The Basic Laws of Arithmetic, Frege (1893, §11) says that a definite description that otherwise wouldn’t refer—​because no object, or more than one object, satisfies its predicate—​ instead refers to the extension of the predicate.This entails something like the view that a definite description that otherwise wouldn’t refer because no object satisfies its predicate instead refers to the empty set. 7 ‘S’ is for ‘stipulationism’. 8 See Radford (1975). 9 See Parsons (1982, 83). Whether (2) is true depends on what thought (1) expresses, which depends on what sense ‘Hermione Granger’ expresses; and stipulationism about names from fiction doesn’t say what sense ‘Hermione Granger’ expresses. (On senses, see sections 3 and 4.) I ignore these complications in the text and assume that stipulationists about names from fiction can get (2) to come out true. As discussed below in the text, stipulationism about names from fiction faces further objections. Thanks to David Braun here. 10 See Rowling (2003). ‘Hermione Granger’ and ‘Luna Lovegood’ aren’t supposed to name the same person or fictional character. 11 People are objects. See Frege (1891, 17). 12 Parsons (1982) develops but doesn’t endorse the view. Lamarque (1996, 31–​34) endorses the view in Fictional Points of View but rejects it elsewhere. See note 19. 13 ‘RS’ is for ‘reference-​to-​sense’. 14 Whether (2) is true depends on what sense ‘Hermione Granger’ expresses in (1). I continue to ignore this complication in the text. See note 9. 15 For example, on Wolterstorff ’s (1980, 134–​149) view, fictional characters are kinds, which are associated with properties. See also Wolterstorff (1980, 47). 16 See Rowling (1999a). 17 See Parsons (1982, 84–​88). 18 See Parsons (1982, 85).

393

Ben Caplan 19 Lamarque and Olson (1994, 78–​84) endorse the view. I  am supposing that the definite description that, according to descriptivism about names from fiction, gives the sense of ‘Hermione Granger’ itself contains names from fiction: namely, ‘Harry’ and ‘Ron’. This can give rise to circularity problems, if for example the definite description that gives the sense of ‘Harry’ contains ‘Hermione Granger’. I ignore these problems in the text. For discussion, see Kripke (1980, 81–​82). 20 See Kripke (1980, 59). 21 ‘D’ is for ‘descriptivism’. 22 If ‘the person who saves Harry and Ron all the time’ gives the sense of ‘Hermione Granger’, then the definite description gives the meaning of the name. (See Kripke 1980, 59.) In that case, (1) and (1D) express the same thought. And that thought is true according to the fiction, so (2) is true. See notes 9 and 14. 23 See McKinsey (2020, 133). Alternatively, descriptivists about names from fiction could use Brock’s (2002) ‘according to realism about fictional characters’ operator and replace (3D) with ‘According to realism about fictional characters, the person who saves Harry and Ron all the time is a fictional character’. See note 31. 24 Frege might hold that ‘The person who saves Harry and Ron all the time exists’ is necessarily equivalent to ‘ “The person who saves Harry and Ron all the time” refers to something’. He distinguishes two cases in which “we speak of existence” and says that in one of them “the question is whether a proper name designates, names, something”. (See Frege 1895, 104.) And he says, “People certainly say that Odysseus is not an historical person, and mean by this contradictory expression that the name ‘Odysseus’ designates nothing”. (See Frege 1906, 161.) Elsewhere, though, Frege ([pre-​1884?], 60) seems to suggest that ‘Sachse exists’ doesn’t mean “The word ‘Sachse’ is not an empty sound, but designates something”, since “this is not a new premise, but the presupposition of all our words”. (On reference being presupposed, see also Frege 1892, 31, 40.) See Salmon (1998, 282–​285). 25 The choice negation of a proposition that’s neither true nor false is itself neither true nor false, whereas the exclusion negation of a proposition that’s neither true nor false is true. (The distinction comes from Bočvar 1939.) See Salmon (1998, 280). 26 (11) isn’t true, but it could be. (Harry Potter is Rowling’s favorite character. See Flood 2011.) 27 See McKinsey (2020, 134–​135). One might worry that, even understood de re, (9D*) is false. Alternatively, descriptivists about names from fiction might adopt the fictionalist strategy discussed below in the text. Thanks to David Friedell and Erin Mercurio for discussion here. 28 Thanks to Joshua Spencer for suggesting this reply to me. 29 See van Inwagen (1977), Parsons (1980, 49–​60, 175–​211), Wolterstorff (1980, 134–​149), Salmon (1998, 293–​ 304), Thomasson (1999), Braun (2005, 609–​614), Kripke (2011, 2013, 61–​83). 30 ‘F’ is for ‘fictionalism’. 31 For (8FD)–​(11FD) to be true requires a version of fictional realism on which Hermione Granger is a concrete object instantiating being the person who saves Harry and Ron all the time. (See Brock 2002, 11–​14.) Alternatively, descriptivists about names from fiction could replace ‘the person who saves Harry and Ron all the time’ with something like ‘there’s an object such that, according to some fiction, it saves Harry and Ron all the time’ (and make corresponding adjustments). 32 See also Frege ([1891–​1895?], 133; 1897, 141). 33 See McDowell (1977, 172–​175). 34 See Evans (1982, 14–​22). 35 Frege (1879, §8). I am assuming that “modes of determination” (Bestimmungsweisen) from the Begriffsschrift are precursors of “modes of presentation” (Art des Gegebenseins)—​or senses—​from “On Sense and Reference”. See Frege (1892, 25–​26). 36 Thanks to Murali Ramachandran for discussion here. 37 See McDowell (1977, 172–​175), Evans (1982, 10–​14, 22–​30). 38 See Evans (1982, 22). 39 Kripke (2013, 26). The view, he says, analyzes “names in fiction according to the same paradigm that Frege used for ‘Aristotle’, the name of a historical figure—​that is as the unique person (or other object) fulfilling the story”. See Kripke (2013, 10 n12). 40 I don’t discuss Kripke’s (2013, 27–​28) third objection, based on a case in which ‘Napoleon’ occurs in a work of fiction. ‘Napoleon’ isn’t a name from fiction in the relevant sense (see note 2), so the case isn’t a counterexample to descriptivism about names from fiction. (The case is a counterexample to a thesis “subsidiary” to Thesis (4): namely, “it is apriori true for the speaker that, if not enough of the φ’s are satisfied, then X does not exist”. See Kripke (1980, 66; italics in original). On Thesis (4), see below in the text. The case echoes Kripke’s (1980, 67) ‘Jonah’ case.) 41 Kripke (1980, 71; italics in original). Cf. Kripke (1980, 64, 65). 42 Kripke (1980, 80–​81; italics in original). 43 Kripke (1980, 86; italics in original).

394

Fregean Theories of Names from Fiction 44 I owe the observation about the fictional analog of Thesis (4) to Joshua Spencer. 45 Italics in original. 46 Kripke (2013, 26–​27). See also Kripke (1980, 157–​158; 2011, 56–​57). 47 Descriptivists about names from fiction might say that the sense of ‘Hermione Granger’ is given by a more complicated definite description, one that’s immune to the possibility of accidental satisfaction. For example, McKinsey (2020, 131) suggests that the sense of ‘Hermione Granger’ is given by ‘the person who the fictional author of Harry Potter and the Prisoner of Azkaban uses “Hermione Granger” to refer to’. (Currie (1990, 76) appeals to what he calls a “fictional author”—​“that fictional character constructed within our make-​believe whom we take to be telling us the story as known fact”—​to provide an account of truth in fiction: roughly, what’s true in the fiction is what it’s reasonable for us to infer that the fictional author believes. See Currie (1990, 92), McKinsey (2020, 130).) But, as McKinsey (2020, 131) notes, one difference between actual authors and fictional authors is that, whereas actual authors actually exist, fictional authors don’t: “the fictional author …, being fictional, does not exist”. So, unlike ‘the actual author of Harry Potter and the Prisoner of Azkaban’, which refers to Rowling, ‘the fictional author of Harry Potter and the Prisoner of Azkaban’ doesn’t refer to anything. And so, unlike ‘the person who the actual author of Harry Potter and the Prisoner of Azkaban uses “Hermione Granger” to refer to’, which might accidentally refer to someone (if Rowling uses ‘Hermione Granger’ to refer to someone), ‘the person who the fictional author of Harry Potter and the Prisoner of Azkaban uses “Hermione Granger” to refer to’ can’t ever refer to anything.

References Bočvar, D. A. 1939. “Ob odnom tréhznačnom isčislénii i égo priménénii k analizu paradoksov klassičeskogo rasširénnogo funkcional’nogo isčisléniá”. Matématičéskij sbornik 4:  287–​308. Trans. by Merrie Bergmann as “On a Three-​valued Logical Calculus and Its Application to the Analysis of the Paradoxes of the Classical Extended Functional Calculus”. History and Philosophy of Logic 2 (1981): 87–​112. Braun, David. 2005. “Empty Names, Fictional Names, Mythical Names”. Noûs 39: 596–​631. Brock, Stuart. 2002. “Fictionalism about Fictional Characters”. Noûs 36: 1–​21. Currie, Gregory. 1990. The Nature of Fiction. Cambridge: Cambridge University Press. Evans, Gareth. 1982. The Varieties of Reference. Oxford: Clarendon. Flood, Alison. 2011. “JK Rowling Reveals Her Favourite Harry Potter Character”. The Guardian (16 May). Available at www.theguardian.com/​childrens-​books-​site/​2011/​may/​16/​jk-​rowling-​favourite-​harry-​potter-​ character. Accessed 2 July 2019. Frege, Gottlob. 1879. Begriffsschrift, eine der arithmetischen nachgebildete Formelsprache des reinen Denkens. Halle: Nebert. Excerpts trans. by Michael Beaney as “Begriffsschrift: A Formula Language of Pure Thought Modelled on that of Arithmetic” in Frege 1997: 47–​78. Frege, Gottlob. [pre-​1884?]. “[Dialog mit Pünjer über Existenz]” in Frege 1969: 60–​75. Trans. by Peter Long and Roger White as “[Dialogue with Pünjer on Existence]” in Frege 1979: 53–​67. Frege, Gottlob. 1891. Function und Begriff:  Vortrag, gehalten in der Sitzung vom 9.  Januar 1891 der Jenaischen Gesellschaft für Medicin und Naturwissenschaft. Jena: Pohle. Trans. by P. T. Geach as “Function and Concept” in Frege 1980: 21–​41. Reprinted in Frege 1997: 130–​148. Frege, Gottlob. [1891–​1895?]. “[Ausführungen über Sinn und Bedeutung]” in Frege 1969: 128–​136. Trans. by Peter Long and Roger White as “[Comments on Sense and Meaning]” in Frege 1979: 118–​125. Reprinted as “[Comments on Sinn and Bedeutung]” in Frege 1997: 172–​180. Frege, Gottlob. 1892. “Über Sinn und Bedeutung”. Zeitschrift für Philosophie und philosophische Kritik 100: 25–​50. Trans. by Max Black as “Sense and Reference”. Philosophical Review 57 (1948): 209–​230. Reprinted as “On Sense and Meaning” in Frege 1980: 56–​78. Reprinted as “On Sinn and Bedeutung” in Frege 1997: 151–​171. Frege, Gottlob. 1893. Grundgesetze der Arithmetik:  Begriffsschriftlich abgeleitet. Volume 1. Jena:  Pohle. Trans. by Philip A. Ebert and Marcus Rossberg in Basic Laws of Arithmetic: Derived using Concept-​Script. Oxford: Oxford University Press, 2013. Frege, Gottlob. 1895. “Kritische Beleuchtung einiger Punkte in E.  Schröders Vorlesungen über die Algebra der Logik”. Archiv für systematische Philosophie 1: 433–​456. Trans. by P. T. Geach as “A Critical Elucidation of Some Points in E. Schroeder’s Vorlesungen ueber die Algebra der Logik” in Frege 1980: 86–​106. Frege, Gottlob. 1897. “Logik” in Frege 1969: 137–​163.Trans. by Peter Long and Roger White as “Logic” in Frege 1979: 126–​151. Reprinted in Frege 1997: 227–​250. Frege, Gottlob. 1906. “Einleitung in die Logik” in Frege 1969: 201–​212. Trans. by Peter Long and Roger White as “Introduction to Logic” in Frege 1979: 185–​196. Excerpt reprinted in Frege 1997: 293–​298. Frege, Gottlob. 1918. “Der Gedanke”. Beiträge zur Philosophie des Deutschen Idealismus 1: 58–​77. Trans. by Peter Geach and R. H. Stoothoff as “Thoughts” in Collected Papers on Mathematics, Logic, and Philosophy, ed. Brian McGuinness. Oxford: Blackwell, 1984, 351–​372. Reprinted as “Thought” in Frege 1997: 325–​345.

395

Ben Caplan Frege, Gottlob. 1969. Nachgelassene Schriften und Wissenschaftlicher Briefwechsel. Volume 1 (Nachgelassene Schriften). Ed. Hans Hermes, Friedrich Kambartel, and Friedrich Kaulbach. Hamburg: Meiner. Frege, Gottlob. 1979. Posthumous Writings. Chicago, IL: University of Chicago Press. Frege, Gottlob. 1980. Translations from the Philosophical Writings of Gottlob Frege. Ed. Peter Geach and Max Black. 1952. 3rd ed. Oxford: Blackwell. Frege, Gottlob. 1997. The Frege Reader. Ed. Michael Beaney. Oxford: Blackwell. Kripke, Saul A. 1980. Naming and Necessity. 1972. Cambridge, MA: Harvard University Press. Kripke, Saul A. 2011. “Vacuous Names and Fictional Entities”. 1973. In Philosophical Troubles:  Collected Papers. Volume 1. New York: Oxford University Press, 52–​74. Kripke, Saul A. 2013. Reference and Existence: The John Locke Lectures. 1973. Oxford: Oxford University Press. McDowell, John. 1977. “On the Sense and Reference of a Proper Name”. Mind 86: 159–​185. McKinsey, Michael. 2020. Consequences of Reference Failure. New York: Routledge. Lamarque, Peter. 1996. Fictional Points of View. Ithaca, NY: Cornell University Press. Lamarque, Peter and Stein Haugom Olsen. 1994. Truth, Fiction, and Literature:  A Philosophical Perspective. Oxford: Clarendon. Parsons, Terence. 1980. Nonexistent Objects. New Haven, CT: Yale University Press. Parsons, Terence. 1982. “Fregean Theories of Fictional Objects”. Topoi 1: 81–​87. Radford, Colin. 1975. “How Can We Be Moved by the Fate of Anna Karenina?” Proceedings of the Aristotelian Society, Supplementary Volumes 49: 67–​93. Rowling, J. K. 1997. Harry Potter and the Philosopher’s Stone. London: Bloomsbury. Rowling, J. K. 1998. Harry Potter and the Sorcerer’s Stone. New York: Scholastic. Rowling, J. K. 1999a. Harry Potter and the Chamber of Secrets. New York: Scholastic. Rowling, J. K. 1999b. Harry Potter and the Prisoner of Azkaban. New York: Scholastic. Rowling, J. K. 2003. Harry Potter and the Order of the Phoenix. New York: Scholastic. Salmon, Nathan. 1998. “Nonexistence”. Noûs 32: 277–​319. Thomasson, Amie L. 1999. Fiction and Metaphysics. Cambridge: Cambridge University Press. van Inwagen, Peter. 1977. “Creatures of Fiction”. American Philosophical Quarterly 14: 299–​308. Wolterstorff, Nicholas. 1980. Works and Worlds of Art. Oxford: Clarendon.

396

PART VIII

Singular (De Re) Thoughts

30 REFERENCE AND SINGULAR THOUGHT François Recanati

1  Reference and Singular Propositions: The Russellian Legacy According to a widespread view, originating from Russell and championed by Donnellan, Kripke, Kaplan, Evans and many other leading philosophers of language of the twentieth century, (genuine or direct) reference, illustrated in (1), contrasts with (mere) description, illustrated in (2). (1) That man is dangerous. (2) The man who did this is dangerous. When there is (genuine or direct) reference to some object a, as in (1), the proposition that is expressed is about a, in the sense that its correctness depends upon whether or not a satisfies what is predicated of it in the proposition. The proposition is singular (with respect to a). In such a case, as Peacocke (1975) pointed out, there is a particular object such that the proposition is true if and only if that object satisfies certain conditions (e.g. being a dangerous man). In contrast, when an object is only given through a definite description ‘the F’, as in (2), there isn’t reference in the full-​blown sense of the term. The proposition expressed is about the description’s satisfier (the man who did this), and that may be turn out to be a, but the proposition is not directly about a. Evaluated with respect to a circumstance or ‘possible world’ w in which some object x ≠ a is the satisfier, its true value will depend upon x’s condition in w rather than upon a’s. So it is not the case that there is a particular object such that (whatever the circumstance of evaluation) the proposition is true just in case that object satisfies certain conditions.1 The foregoing amounts to saying that referring expressions are rigid, while descriptions are not (Kripke 1980). Or at least, descriptions are not ‘rigid de jure’, as referring expressions are. A description will only be rigid if, like mathematical descriptions, it has the same satisfier in every possible world (rigidity de facto). But the reference of a genuine referring expression is determined before the encounter with the circumstance of evaluation, so it cannot vary according to the circumstance of evaluation: it is circumstance-​independent (Kaplan 1989). Within a structured proposition framework, it is customary to phrase the reference/​description contrast in terms of distinct types of propositional constituent. The constituent of the structured proposition which corresponds to a definite description is not an object but what early Russell referred to as a denoting concept (Russell 1903). The constituent of the proposition corresponding 399

François Recanati

to a genuine referring expression is supposed to be (directly) an object, namely the object which has to be mentioned in specifying the truth-​conditions of the proposition.

2  Modes of Presentation and Cognitive Content: The Fregean Legacy The idea that the semantic contribution of a referring expression—​its content—​is directly an object raises a fundamental objection, due to Frege. If the content of a referring expression is its reference, then two coreferential expressions must carry the same content. Yet a rational subject can entertain contradictory attitudes towards two sentences that only differ by the substitution of one of two coreferential expressions for the other. Thus a rational subject may endorse ‘Cicero was Roman’ while rejecting ‘Tully was Roman’, i.e. they may believe the content of one sentence while disbelieving the content of the other. Or they may assent to ‘That man is dangerous’ (pointing to one man), and dissent from ‘That man is dangerous’ (said while pointing to what the speaker wrongly takes to be another man). How is that possible if the content of the two sentences is the same, i.e., if they express the same singular proposition? And how could it not be the same, on the assumption that the content of a referring expression is its reference? After all, the two sentences in each of the pairs (‘Cicero was Roman’ /​‘Tully was Roman’, ‘That man is dangerous’ /​‘That man is dangerous’) only differ by the substitution of one coreferential expression for another. Faced with that objection, we do not have to give up the distinction between those expressions that are directly referential (and contribute an object to semantic content) and those that aren’t. We only have to draw another distinction, between two aspects of content, or two perspectives on content (McGinn 1982). Thus far we have characterized the content of an utterance—​the proposition it expresses—​in terms of its truth-​conditions. Let us start with the following ‘criterion of difference’ for content: (TC) If there are circumstances with respect to which a sentence S is true while another sentence S′ is not, then S and S′ differ in content. If we strengthen this into a biconditional, we get a characterization of one type of content: truth-​ conditional content. Two sentences S and S′ differ in truth-​conditional content—​express different propositions—​just in case there are circumstances with respect to which S is true while S′ is not. A referring expression and a definite description make different types of contribution to content thus understood: a referring expression contributes an object, while a definite description contributes an identifying property, and this results in differences in possible-​worlds truth-​conditions. The proposition expressed by a sentence in which a referring expression occurs is said to be singular because its truth-​condition is singular: there is an object, namely the referent of that expression, such that the sentence is true just in case that object satisfies certain conditions. When we individuate content by truth-​conditions in this manner, we have to say that ‘Cicero was Roman’ and ‘Tully was Roman’ have the same content or express the same proposition: they ascribe the same property to the same object. That is what raises the Fregean objection. What the objection actually shows is that content can and should also be individuated in epistemological or cognitive terms. Truth-​conditions are not enough. Even if two sentences express the same proposition in the truth-​conditional sense, they may still carry distinct contents, in virtue of a second ‘criterion of difference’: (CC) If it is possible for a rational and linguistically competent subject to accept a sentence S as true while taking a different stance (e.g. rejection) towards another sentence S′, then S and S′ differ in content.

400

Reference and Singular Thought

If we strengthen this into a biconditional, we get a characterization of another type of content: cognitive content. Two sentences S and S′ differ in cognitive content—​express different thoughts—​just in case a rational and linguistically competent subject may adopt different epistemic attitudes towards them (e.g. accept S as true while rejecting S′). The difference between referring expressions and definite descriptions is a difference at the level of truth-​conditional content: these two types of expression make different types of truth-​conditional contribution—​one contributes an object, the other an individual concept. Two genuine referring expressions that designate the same object make the same contribution to truth-​conditional content (viz. the object itself); but that does not mean that they carry the same content simpliciter. For there is more to content simpliciter than truth-​conditional content: there is also cognitive content. In so-​called ‘Frege cases’, such as the Cicero/​Tully example, two sentences which express the same proposition in the truth-​conditional sense nevertheless carry distinct cognitive contents.2 Since the sentences only differ by the substitution of one coreferential expression for another, the difference in cognitive content has to be traced to these expressions themselves. According to Frege, referring expressions like ‘Cicero’ and ‘Tully’ refer to the same individual, but carry distinct ‘senses’: the individual is presented differently in the two cases—​as Tully, or as Cicero. What makes it possible for a rational subject to take conflicting attitudes towards what is in effect the same content, truth-​ conditionally individuated, is the fact that the content in question is presented differently. The modes of presentation associated with the names ‘Cicero’ and ‘Tully’ therefore affect the utterance’s cognitive content, while the names contribute their reference (the same individual in both cases) to truth-​ conditional content. In other words:  The sentences ‘Cicero was Roman’ and ‘Tully was Roman’ express the same proposition in the truth-​conditional sense, and that proposition is singular, as we have seen; but they express different thoughts. The thought expressed by an utterance involves more than the objects, properties etc. which are constituents of the singular proposition that captures its truth-​conditional content. The thought involves, in addition, particular ways of thinking of these constituents. To clarify the nature of singular thought, one needs to clarify the nature of these modes of presentation.

3  Modes of Presentation as Descriptions So, what are modes of presentation? Whatever they are, they have to satisfy what Schiffer calls ‘Frege’s Constraint’: Necessarily, if m is a mode of presentation under which a minimally rational person x believes a thing y to be F, then it is not the case that x also believes y not to be F under m. In other words, if x believes y to be F and also believes y not to be F, then there are distinct modes of presentation m and m′ such that x believes y to be F under m and disbelieves y to be F under m′. Let us call this Frege’s Constraint; it is a constraint which any candidate must satisfy if it is to qualify as a mode of presentation. Schiffer 1978, 180 This very minimal characterization of modes of presentation leaves considerable latitude to the theorist, however. Frege himself thought of modes of presentation as individual concepts, that is, as descriptions of the referent. When we think about an object, we think of it as ‘the F’, where F is some uniquely-​ identifying property of the referent. A linguistic expression is thus associated with a ‘sense’ which mediates between the expression and its reference. The sense is a set of conditions which an item in the world has to satisfy in order to count as the reference of the expression. Equivalently, the sense of an expression is a collection of things known about the reference—​the reference being the entity which fits the body of

401

François Recanati

knowledge in question. Competent language users know what the conditions are, even if they do not know which entity in the world fits those conditions. Appearances notwithstanding, thinking of modes of presentation as descriptions in this way is compatible with the idea that referring expressions contrast with definite descriptions by directly contributing an object, rather than an individual concept, to truth-​conditional content. For there may still be a role for individual concepts on the direct-​reference story: the object which is the contribution of a referring expression to truth-​conditional content may itself be determined via satisfaction of an individual concept, even though it is the object rather than the individual concept that is the expression’s contribution to truth-​conditional content (Kaplan 1978). In other words, in the case of referring expressions and singular thought, the individual concept is confined to the level of cognitive content, and does not impinge upon truth-​conditional content. It serves merely to ‘fix the reference’, as Kripke (1980) says.3 Individual concepts in general are, or can be modelled as, functions from situations to individuals.The individual concept contributed by a definite description is a function that applies to the circumstance of evaluation, returning an individual that depends upon the circumstance in question. In the case of referring expressions, however, the individual concept applies to the situation of utterance (the ‘context’), not to the situation of evaluation (the ‘circumstance’). It returns an individual which goes into the proposition and is, therefore, fixed before the encounter with the circumstance of evaluation (Kaplan 1989). There are major objections to that descriptivist construal of modes of presentation, however. The most important one is the buck-​passing objection. If the reference of an expression is determined by a set of conditions mentally represented by the linguistically competent subject, and constituting its sense, then the relevant mental representations in the subject’s mind will have to refer to the conditions in question. But if the reference of these mental representations is determined in the same way, via a set of conditions, then an infinite regress is launched. ‘Sooner or later’, Pylyshyn says, ‘the regress of specifying concepts in terms of other concepts has to bottom out’ (Pylyshyn 2001, 129). As Devitt puts it, There must be some representations whose referential properties are not parasitic on those of others, else language as a whole is cut loose from the world. Description theories pass the referential buck, but the buck must stop somewhere. It stops with theories … that explain reference in terms of direct relations to reality. Devitt 2014, 477 In addition to that general objection, there are a number of specific arguments against Frege’s descriptivist approach. Let me mention three of them: • The argument from perception. When we perceive an object and have a thought about it, the object the thought is about is the object the perception is about; and that, arguably, is not determined by properties the subject takes the referent to have. (Pylyshyn 2007).4 • The argument from indexicals. Indexicals like ‘I’, ‘here’ and ‘now’ present their reference in a manner that is essentially perspectival and cannot be captured by means of descriptions, unless the descriptions themselves involve indexicals. For any indexical α and non-​indexical description ‘the F’, it is always possible for the subject to doubt, or to wonder, whether α is the F. (Castañeda 1999; Perry 2000). • Kripke’s argument from ignorance and error. A linguistically competent subject may not possess the sort of identifying knowledge of the reference which that conception takes to be required to grasp the ‘sense’ of a proper name; instead, she may possess only ‘unsteady and confused notions’, as Locke puts it (Locke 1690, IV, X, §4). Such notions are insufficient to determine the reference. Or, worse, the subject may be radically mistaken concerning the nature and properties of the reference. The possibility of such mistakes about the reference shows that the reference of a proper name cannot be what fits the subject’s conception (otherwise the subject could not make mistakes). 402

Reference and Singular Thought

The three arguments point in the same direction. On the descriptivist picture, reference is determined satisfactionally: the referent is what satisfies descriptions in the subject’s mind. But in the three types of case we talked about (perceptual reference, indexical reference, and reference via names) what seems to be crucial are the subject’s relations to the objects of thought: reference is determined relationally, not satisfactionally (Bach 1987, 12). Demonstrative thought rests on a perceptual/​attentional link to the object of thought, while indexical reference rests on various contextual relations—​the relation each individual bears to himself or herself in the case of first person thought, the relation to the place one occupies in the case of here-​thoughts, and the relation to the present time in the case of now-​ thoughts. For proper names, what matters is the historical chain of communication through which the subject is related to the reference of the name. In all three cases, what fixes the reference is the relation, not the satisfaction of descriptions in the subject’s mind.

4  Mental Files If the modes of presentation of objects involved in singular thoughts are not individual concepts, what are they? So far we know two things: (i) they have to satisfy Frege’s Constraint; (ii) they are ‘nondescriptive’ in the sense that their reference is determined relationally rather than satisfactionally. Mental files are a candidate that (arguably) fits that bill. The mental file idea was first introduced in a descriptivist framework. Because the knowledge of the reference associated with a proper name typically involves a rich body of information rather than a single identifying description, a variant of the Fregean approach has emphasized the fact that the descriptions associated with proper names come in clusters. The ‘cluster’ idea is to be found in the work of Wittgenstein and Searle, but also in that of Strawson and Grice. A cluster of descriptions can be thought of as a mental file or dossier about an individual, involving many distinct pieces of information. The mental file idea does not, by itself, entail descriptivism; it is entirely compatible with the relational stance characteristic of the anti-​descriptivist movement. Instead of saying, with Searle and Grice, that the reference of a dossier is the item ‘which satisfies the majority of, or each member of a specially favoured subset of, the descriptions in the dossier’ (Grice 1969, 142), one can say the following. A mental file refers not to what satisfies the descriptions in the file (since the subject may be severely mistaken) but to what stands in the right relation to the file or, equivalently, to the subject who deploys the file. The relevant relation is a relation serving as information channel: a relation to the object which makes it possible for the subject to gain information from that object, and thus to feed the file based on that relation (Recanati 2012). For example, a demonstrative file is based on the attentional relation, a relation that makes it possible to gain perceptual information from the object one is attending to. The information thus gained goes into the file. (A demonstrative file, thus understood, is a short-​term file, which exists only as long as one is paying attention to the object. But long-​term files too are based on relations to the reference, e.g. the ‘familiarity relation’ which holds whenever ‘multiple exposure to [an] object has created and maintained in the subject a disposition to recognize that object’ (Recanati 2012, 71).) Because they enable the subject to gain information from the object, the relations on which mental files are based, and which determine their reference, are epistemically rewarding (ER) relations. They are also called acquaintance relations because they put the subject in epistemic contact with the object, by making information about it available (information which it is the function of the file to store). Although very different from the attentional relation which underlies the deployment of demonstrative concepts, or the recognitional relation which underlies the deployment of recognitional concepts, the subject’s relation to himself or herself, namely identity, counts as an ER relation, because it makes it possible for the subject to gain information about herself through a particular information channel, namely ‘from inside’ (through proprioception, introspection, episodic memory, etc.). It makes that possible in the following sense: one can get information about 403

François Recanati

an individual in this manner (viz. from inside) only if one is that individual. The information thus gained goes into the subject’s SELF-​file, a special file based on the identity relation. Acknowledging the role of ER relations therefore provides a solution to the ‘problem of the essential indexical’ (Perry 2000): just like demonstrative concepts (‘that thing’), indexical concepts such as the concept of oneself, the concept of the present time (‘now’) or the concept of the place one occupies (‘here’) should be construed not as descriptive concepts (‘the x such that Fx’) but as mental files based on particular ER relations. Only a subject in the right relation to the referent can think a thought involving such a concept.This captures the ‘limited accessibility’ feature supposedly characteristic of indexical thought. An important aspect of the framework is that, in order to play their part, mental files need to stand in relation to the entities they refer to, via the ER relation which the subject deploying the file stands in to these entities. To be so related mental files have to be construed as entities on their own right, that is, as particulars—​‘concrete mental particulars’, as Crimmins and Perry (1989) say. Construing mental files as particulars in this way neatly solves the problem which ignorance of the reference raises for Fregean descriptivism. Kit Fine formulates the problem as follows: Surely one may learn something different upon being told ‘Cicero = Tully’ and upon being told ‘Cicero = Cicero’ … The main problem with the Fregean position is to say, in particular cases, what the difference in the meaning or sense of the names might plausibly be taken to be. Although there appear to be good theoretical reasons for thinking that there must be a difference, it seems hard to say in particular cases what it is. For as Kripke (1980) has pointed out, it seems possible for a speaker, or for speakers, to associate the same beliefs or information with two names, such as “Cicero” and “Tully”. And if the information or beliefs are the same, then how can the sense be different? Fine 2007, 35 A descriptivist might respond that the ‘Cicero’ file contains the metalinguistic piece of information ‘called Cicero’ while the Tully file contains ‘called Tully’. This is enough of a difference in descriptive content between the two files. But, Pryor recently argued, there can be Frege cases even in the absence of any such metalinguistic difference (Pryor 2016; see Gray 2016 for discussion).5 Let us assume this is right, and ask Fine’s question again: ‘if the information or beliefs are the same, then how can the sense be different?’ Construing mental files as particulars provides an obvious answer. If there are two distinct files, then there are two distinct modes of presentation, even if the information in the two files is exactly the same (‘a Roman orator’).

5  Are Modes of Presentation Aspects of Content? In the above citation from Fine I omitted one sentence. With the sentence restored (and italicized), the beginning of the passage reads as follows: Surely one may learn something different upon being told ‘Cicero = Tully’ and upon being told ‘Cicero = Cicero’ … It is hard to see how to account for this possible cognitive difference except in terms of a semantic difference. The main problem with the Fregean position is to say, in particular cases, what the difference in the meaning or sense of the names might plausibly be taken to be. Fine 2007, 35 I answered Fine’s question, ‘if the information or beliefs are the same, then how can the sense be different?’ by saying that mental files, which play the role of Frege’s senses, are particulars. It follows that, if there are two numerically distinct files (possibly containing the same information), then there 404

Reference and Singular Thought

are two distinct senses. But, Fine might retort, are we still talking about senses? Is the (purely numerical) difference between the two files construed as particulars a semantic difference? Clearly not. The difference is ‘syntactic’:  it is a difference between the mental representations that are respectively deployed, rather than a difference in the content of the representations. The representations have the same content—​they refer to the same individual and carry the same information about it—​but they are distinct representations. Arguably, a purely syntactic difference such as this is sufficient to generate Frege cases, so it was a mistake on Frege’s part to argue from the possibility of Frege cases to the necessity of adding an extra layer of content, that of ‘sense’, on top of the reference of expressions. On this type of approach, defended by Jerry Fodor (1998) and Sainsbury and Tye (2012), what plays the role of mode of presentation is the mental representation itself, qua syntactic entity. The subject who believes that Cicero was Roman but that Tully was not entertains conflicting attitudes towards the same truth-​conditional content because that content is apprehended via distinct mental representations (involving two distinct mental files: a ‘Cicero’ file and a ‘Tully’ file). So, to account for Frege cases, we need only two things (the representation and its reference) rather than three (the representation, its sense, and its reference). Is the same move available for linguistic expressions? Can we say that we need only two things: the reference of the expression (its semantic content), and the expression itself serving as mode of presentation and accounting for Frege cases? This is what Mates’s cases may seem to suggest: there being two different words (e.g. ‘psychiatrist’/​‘alienist’, or ‘Greek’/​‘Hellene’) is sufficient to make Frege cases possible (Mates 1950). Note, however, that the existence of distinct words is not necessary to generate Frege cases. Even if there is a single word in the language, e.g. ‘Paderewski’ as the proper name of the Polish citizen who was well-​known both as a politician and as a pianist, Frege cases will still be possible if the subject associates distinct mental files with that name (thinking there are two distinct Paderewskis, Paderewski the musician and Paderewski the politician).6 This shows that it is the associated mental representation, not the linguistic expression itself, which matters. Indeed, the existence of distinct words is not even sufficient to generate Frege cases, contrary to what Mates’s cases superficially suggest: the subject may treat ‘Cicero’ and ‘Tully’ as two names of the same individual, and associate both of them with the same mental file for Cicero/​Tully. No Frege case can be generated in these circumstances.7 What I have just said suggests that, for linguistic expressions, we do need three things rather than two. But the three things are not, as for Frege, the expression, its sense, and its reference. The three things now are: the linguistic expression, the associated mental representation, and the reference of the mental representation, which the linguistic representation inherits. On this view, what plays the role of mode of presentation for a linguistic expression is neither a sense construed as an aspect of its semantic content, nor the linguistic expression itself, but the associated mental representation. Thus a singular term such as the name ‘Cicero’ has a reference (Cicero), which it inherits from the mental file it is associated with (the ‘Cicero’ file). The file itself refers to Cicero in virtue of informational connections to Cicero. Mental files thus construed are ‘singular terms in the language of thought’ (Recanati 2012). They are syntactic entities. Still, the idea of an additional level or aspect of content, namely cognitive content, does not have to be given up. For mental files themselves are typed by the sort of ER relation they exploit: different types of file—​e.g. SELF-​files, demonstrative files, recognitional files, deferential files, or what have you—​exploit different types of contextual relation to the reference. On this view (i) the type of a file corresponds to its function or role: exploiting a given ER relation, and (ii) tokening a file of a given type presupposes that the subject deploying the file is standing in the right relation to the reference. That presupposition corresponds to a layer of content that comes in addition to the truth-​conditional content of the thought in which the file is deployed. It is that layer of content—​cognitive content as opposed to truth-​conditional content—​that accounts for the subject’s behaviour, and in terms of which the subject’s rationality is to be assessed. For example, if the subject hallucinates a desirable object in front of her, the demonstrative file she deploys about ‘it’ is empty8 and fails to contribute to 405

François Recanati

truth-​conditional content (so the thought she entertains about the hallucinated object is neither true nor false) but the subject’s attempt at grasping the object can be rationally accounted for in terms of the subject’s presupposition that she is standing in front of a desirable object and gaining information from it in perception. Likewise, the subject’s saying ‘That is desirable’ can be rationally accounted for in terms of the content of the subject’s presupposition, even if the utterance itself fails to express a singular proposition (because the referring expression is empty). Cognitive content is what is shared by two persons who both think e.g. ‘My pants are on fire’ (Kaplan 1989, 533). The truth-​conditional contents of their respective thoughts are different (since A thinks that A’s pants are on fire, while B thinks that B’s pants are on fire), but they both deploy a SELF-​file, and the sort of behaviour their thought causes is appropriate to the situation of someone whose pants are on fire. So, despite the difference in truth-​conditional content, the cognitive content is the same. Frege would of course deny that two thoughts can be the same while carrying different truth-​ conditions. For him, thoughts simultaneously obey the two criteria of difference mentioned in section 2: the truth-​conditional criterion TC and the cognitive criterion CC. But this is a terminological matter. Cognitive contents as I have defined them are individuated in purely cognitive terms and abstract from truth-​conditional content.They are thoughts in the narrow sense (thoughtsN), in contrast to Fregean thoughts, which are thoughts in the broad sense (thoughtsB). Two subjects both thinking ‘My pants are on fire’ entertain the same thoughtN, despite the difference in truth-​conditional content. This is of course compatible with Frege’s claim that these subjects entertain different thoughtsB, in virtue of the first criterion of difference (the truth-​conditional criterion).9 Be that as it may, Frege cases work in the opposite direction. The subject who looks at himself in the mirror and, not recognizing himself, thinks ‘His pants are on fire’, entertains a thought whose truth-​conditional content is the same as that of the thought ‘My pants are on fire’, if he were to entertain it; but the subject in that situation dissents from the latter (he does not accept ‘My pants are on fire’) while he assents to the former. So the cognitive content of the two thoughts, i.e. that which accounts for the subject’s behaviour (including his assent/​dissent behaviour), is different. In that situation the subject could say or think, without irrationality: ‘His pants are on fire, but mine aren’t’.The thoughts are different, whether understood narrowly or broadly.

Further Reading Only recent works are listed. For a more comprehensive bibliography see the ‘Singular Thought’ entry by M. Murez, F. Recanati and F. Carvalho in Oxford Bibliographies Online, 2014. Dickie, I. (2015) Fixing Reference. Oxford: Oxford University Press. Goodman, R., Genone, J. and Kroll, N. (eds.) (2020) Singular Thought and Mental Files. Oxford:  Oxford University Press. Hawthorne, J. and Manley, D. (2012) The Reference Book. Oxford: Oxford University Press. Jeshion, R. (ed.) (2010) New Essays on Singular Thought. Oxford: Oxford University Press. Millikan, R. (2018) Beyond Concepts: Unicepts, Language and Natural Information. Oxford: Oxford University Press. Murez, M. and Recanati, F. (eds.) (2016) Mental Files. Special issue, The Review of Philosophy and Psychology, 7. Perry, J. (2012) Reference and Reflexivity, 2nd ed. Stanford, CA: CSLI Publications. Raftopoulos, A. and Machamer, P. (eds.) (2012) Perception, Realism, and the Problem of Reference. Cambridge: Cambridge University Press.

Notes 1 As Donnellan (1966) emphasizes, definite descriptions can be used referentially, to talk about some object the speaker has in mind. On that use they work like genuine referring expressions. But this is a matter of pragmatics: definite descriptions are not intrinsically referential, in contrast to genuine referring expressions. 2 The expression ‘Frege case’ was introduced in the literature by Jerry Fodor (1994). 3 In Recanati (1993) modes of presentation are said to be ‘truth-​conditionally irrelevant’.

406

Reference and Singular Thought 4 Often, we are unable to properly describe the object that is given to us in experience: we don’t know what it is, yet that does not prevent us from referring to it directly (without conceptual mediation) and e.g. wondering what it can be (Dretske 1988, 73). 5 This is actually controversial. Even if one accepts the idea of ​qualitatively identical but numerically distinct files, one can doubt that the existence of a pair of such files, as in Pagin’s ‘moth’ example (Pagin 2013, 140), could give rise to Frege cases. Indeed, as long as the files are indistinguishable, it is far from clear that the subject can deploy one (rather than the other) in a distinctive way to think—​for example—​of one of the alleged moths (rather than the other). In the absence of any distinctive element in the content of the files, it seems that the existence of two separate files only allows the subject to represent the cardinality of the set of moths he takes himself to perceive, that is, the fact that there are two, but not to relate distinctively to one of the alleged moths in thought. If this is true, then one cannot have a Frege case involving two numerically distinct but indistinguishable files. 6 The ‘Paderewski’ example is discussed in Kripke (1979). 7 Consider the following example, involving anaphora (from Recanati 2016): ‘I saw Johni the other day. The bastardi did not greet me’. The mental file associated with the proper name ‘John’ in the first sentence is redeployed in association with the anaphoric description ‘the bastard’ in the subsequent sentence. As a result no Frege case is possible:  a rational and linguistic competent subject cannot ascribe contradictory properties to John and to ‘the bastard’, since, simply in virtue of understanding the discourse, she knows they are one and the same individual. 8 The file is empty in the sense that it does not refer to anything (since there is no object at the other end of the ER relation); but, of course, it may still contain putative information about the object the subject takes herself to be perceiving. 9 Likewise, when I said that, if the subject hallucinates an object in front of her, the demonstrative thought she entertains about the hallucinated object is neither true nor false, I used ‘thought’ in the narrow sense. A neo-​ Fregean like Evans or McDowell, unwilling to individuate thoughts narrowly, would rather say that there is no singular thoughtB, in this type of case, but only an illusion of singular thoughtB (Evans 1982; McDowell 2011, part II). (For there to be a singular thoughtB, there has to be an object such that the thought is true iff that object satisfies certain conditions.)

Bibliography Bach, K. (1987) Thought and Reference. Oxford: Clarendon Press. Castañeda, H.-​N. (1999) The Phenomeno-​Logic of the I:  Essays on Self-​Consciousness (edited by J. Hart and T. Kapitan). Bloomington, IN: Indiana University Press. Crimmins, M. and Perry, J. (1989) The Prince and the Phone Booth. Journal of Philosophy 86: 685–​711. Reprinted in Perry (1993), 249–​278. Devitt, M. (2014) Lest Auld Acquaintance Be Forgot. Mind & Language 29: 475–​484. Donnellan, K. (1966) Reference and Definite Descriptions. Philosophical Review 75: 281–​304. Donnellan, K. (2012) Essays on Reference, Language, and Mind (edited by J. Almog and P. Leonardi). Oxford: Oxford University Press. Dretske, F. (1988) Explaining Behavior. Cambridge, MA: MIT Press. Evans, G. (1982) The Varieties of Reference (edited by J. McDowell). Oxford: Clarendon Press. Evans, G. (1985) Collected Papers. Oxford: Clarendon Press. Fine, K. (2007) Semantic Relationism. Oxford: Blackwell. Fodor, J. (1994) The Elm and the Expert. Cambridge, MA: MIT Press/​Bradford Books. Fodor, J. (1998) Concepts: Where Cognitive Science Went Wrong. Oxford: Clarendon Press. Frege, G. (1967) Kleine Schriften (edited by I. Angelelli). Hildesheim: Olms. Gray, A. (2016) Minimal Descriptivism. Review of Philosophy and Psychology 7: 343–​364. Grice, P. (1969) Vacuous Names. In D. Davidson and J. Hintikka (eds.) Words and Objections, pp. 118–​145. Dordrecht: Reidel. Kaplan, D. (1978) DTHAT. Syntax and Semantics 9: 221–​243. Kaplan, D. (1989) Demonstratives. In J. Almog, J. Perry and H. Wettstein (eds.) Themes from Kaplan, pp. 481–​563. New York: Oxford University Press. Kripke, S. (1979) A Puzzle about Belief. In A. Margalit (ed.) Meaning and Use, pp. 239–​283. Dordrecht: Reidel. Kripke, S. (1980) Naming and Necessity. Oxford: Blackwell. Locke, J. (1690) An Essay Concerning Human Understanding. London: Thomas Basset. McDowell, J. (2011) Meaning, Knowledge and Reality. Cambridge, MA: Harvard University Press. McGinn, C. (1982) The Structure of Content. In A. Woodfield (ed.) Thought and Object, pp. 207–​258. Oxford: Clarendon Press. Mates, B. (1950) Synonymity. University of California Publications in Philosophy 25: 201–​226.

407

François Recanati Pagin, P. (2013) The Cognitive Significance of Mental Files. Disputatio 5/​36: 133–​145. Peacocke, C. (1975) Proper Names, Reference, and Rigid Designation. In S. Blackburn (ed.) Meaning, Reference and Necessity, pp. 109–​132. Cambridge: Cambridge University Press. Perry, J. (2000) The Problem of the Essential Indexical and Other Essays, expanded edition. Stanford, CA: CSLI. Pryor, J. (2016) Mental Graphs. Review of Philosophy and Psychology 7: 309–​341. Pylyshyn, Z. (2001) Visual Indexes, Preconceptual Objects, and Situated Vision. Cognition 80: 127–​158. Pylyshyn, Z. (2007) Things and Places:  How the Mind Connects to the World. Cambridge, MA:  MIT Press/​ Bradford Books. Recanati, F. (1993) Direct Reference: From Language to Thought. Oxford: Blackwell. Recanati, F. (2012) Mental Files. Oxford: Oxford University Press. Recanati, F. (2016) Mental Files in Flux. Oxford: Oxford University Press. Russell, B. (1903) The Principles of Mathematics. Cambridge: Cambridge University Press. Sainsbury, M. and Tye, M. (2012) Seven Puzzles of Thought. Oxford: Oxford University Press. Schiffer, S. (1978) The Basis of Reference. Erkenntnis 13: 171–​206. Searle, J. (1958) Proper Names. Mind 67: 166–​173. Strawson, P. (1974) Subject and Predicate in Logic and Grammar. London: Methuen. Wittgenstein, L. (1953) Philosophical Investigations, trans. E. Anscombe. Oxford: Blackwell.

408

31 SINGULAR THOUGHTS, SENTENCES AND PROPOSITIONS OF THAT WHICH DOES NOT EXIST Jody Azzouni

I Introduction “Singular” thoughts about nonexistent things, “singular” sentences with nonreferring terms—​ ones nevertheless that have (varying) truth values depending on their content—​these strike many as oxymoronic.1 The oxymoron is more complex than first appears because of several assumptions—​although these are widely shared enough to be tacit in the work of some and “obvious” to others. Before starting on this, I’ll note the unhappy nomenclature in this area. “De re”, builds “objectness” into the label for the phenomena; this doesn’t fit with singularity’s indifference to ontology. “Singular” is awkward too, since it can be plural (Azzouni 2011) as when a speaker demonstrates several people and says, “They’re hungry”. “Object-​directed” is probably best (although it has also been co-​opted, by Hawthorne and Manley (2012), for object-​dependent reference). Singular is entrenched so I’ll continue to use it here. “General”, “fully general” and “quantificational” are infelicitous: they badly fit uniquely used definite descriptions (unless Russell’s theory of descriptions is presupposed). Descriptive is best, even though singular thoughts and sentences often have descriptive content; what makes them singular, as opposed to descriptive, isn’t their descriptive content but that content’s role (see section IV). Lastly, I’ll use phrases like “empty singular thought” and “empty reference” to describe singular thought, etc., about the nonexistent. My aim is modest. I don’t summarize or evaluate the evidence and arguments for the contested robustness of the singular/​descriptive distinction; see elsewhere in this handbook for that. I’ll only show that all the linguistic/​psychological data supporting singular thoughts and sentences (controversial as they are) don’t differ with respect to empty and nonempty singularity. This is, in part, to illustrate how families of arguments and thought experiments already present in the literature for singularity have easy extensions to empty cases (section V). Exposing the assumptions that prima facie bear against the very possibility of empty singularity comes first, however—​in the next three sections.

II  Fregeanism and Russellianism One assumption about the metaphysics of propositions or thoughts is that there are only two choices, Russellianism and Fregeanism.These labels describe families of positions some members of which bear only indirect relations to Russell (1910), Russell (1912) and Frege (1892). Consider, (1) Bertrand Russell was born in 1872. (2) Anyone who invented the theory of descriptions was born in 1872. 409

Jody Azzouni

Assume, as many do, a propositional semantics for natural-​language sentences; that is (relative to a context), propositions are contents of sentences (they’re “expressed” by sentences); they’re true or false, have modal properties, are necessary or contingent, and are targets of propositional attitudes. Utterances and sentences inherit these properties from their propositional contents. Consider the terms “Bertrand Russell” and “anyone who invented the theory of descriptions”. On a pure “Fregean” view, these terms refer via Fregean senses—​descriptive conditions. “Anyone who invented the theory of descriptions” refers uniquely to Bertrand Russell via the senses of “invented”, “theory”, etc. For the Fregean, “Bertrand Russell” also has a sense—​a descriptive condition—​that holds of Russell. These descriptive conditions occur in the structured propositions these sentences express:  structured propositions aren’t sets of descriptive conditions or ordered lists but are contoured in ways highly dependent on the syntactic/​semantic properties of natural-​language sentences. On a Russellian view, the referents of (some) terms appearing in sentences aren’t mediated by descriptive conditions:  they’re directly given. That is, according to Russellianism, propositions are structural combinations of items, some of which are the referents of terms in the sentences that express them.2 (More weakly, the individuation of Russellian propositions depends on the referents of their terms, not on something more fine-​g rained.) Russellians differ on which terms in sentences unmediatedly refer. For Russell, at one point, only the terms in a person’s present-​tense idiolect for his sense data, “I”, and certain universals, were such. Later “neo-​Russellians” include proper names and indexicals, predicates and (some) definite descriptions. According to certain Russellians, that is, the proposition expressed by, (3) Mark Twain is Samuel Clemens, is structurally composed of Mark Twain and the is-​relation, where Mark Twain occurs twice, associated once with “Mark Twain” and once with “Samuel Clemens”. On the pure Fregean view, the proposition (3) expresses is instead structurally composed of the senses, Mark Twain, Samuel Clemens, and the sense of the is-​relation. Although “Mark Twain” and “Samuel Clemens” co-​refer, their senses are different. For Frege and others, this explains why (3) is informative and “Mark Twain is Mark Twain” isn’t.

III  The Metaphysics of Existence and Nonexistence Another widely shared assumption, crucial to seeing “singular” sentences and thoughts about nonexistent entities as oxymoronic, is metaphysical/​linguistic: the univocality of “exist” and the simple nontological status of what doesn’t exist. Univocality is that our notion of “existence”—​the concept corresponding to the word—​demands that items either exist or they don’t. Simple nontology,3 in addition, demands that if “something” doesn’t exist, “it” isn’t anything at all: “it” has no other ontological status (“being”, “subsistence”, etc.) nor does “it” have properties of any sort. This is because what doesn’t exist isn’t anything and so there isn’t anything to have such a status or properties. (The pedantic quotes around “something” and “it” indicate the pleonastic grammatical role of these words—​nothing referentially corresponds to them.) A simple nontologist can reject the univocality of “exist”, by believing in different kinds of existence—​perhaps spatio/​temporal-​location and atemporal/​aspatial-​presence are two examples. Some Meinongians, though, accept the univocality of “exist” but reject simple nontology; they think there is only one kind of existence but many kinds of nonexistence—​subsistence perhaps, or being. Property-​attributing Meinongians instead (and like simple nontologists) deny different ontological statuses to what doesn’t exist—​they deny there are different kinds of nonbeing—​but persist (against simple nontologists) in thinking nonexistent entities differ from one another in their properties. Lastly, a philosopher may reject both the univocality of “exist” and simple nontology. All these positions are occupied—​especially in the twentieth and twenty-​first centuries. See Chisholm (1972) for an informed discussion of Meinong’s original position; McDaniel (2017) has a kinds-​of-​existence view, one he attributes to Heidegger; Crane (2013), 410

That Which Doesn’t Exist

Priest (2005) and Routley (1980) are property-​attributing Meinongians; Azzouni (2010) is a simple nontologist committed to the univocality of “exist”.4 Simple nontology isn’t distinctive and important because of its metaphysical claims about the nonexistent—​the position only requires that there’s nothing to say about this. What’s important is a concomitant assumption about natural language: there are terms—​“Superman”, “Vulcan” and “The present king of France”, are candidates—​which the simple nontologist denies reference to. A corollary, thus, is that semantic theories for sentences containing these terms can’t introduce the relata of these terms as truth-​makers. It’s simple nontology—​not the claim “exist” is univocal—​that’s in tension with there being empty singular propositions. A simple nontological attitude towards Mickey Mouse is that there is no Mickey Mouse, nothing with such properties, nothing with that sort of being. It seems, therefore, that (4) Mickey Mouse has black ears, (5) Mickey Mouse was invented by Walt Disney, can’t express singular propositions. A  singular proposition corresponding to (4)  has a slot where Mickey Mouse, the very thing, would have appeared if there had been one—​for example, the first slot in .5 But the singular propositional schema corresponding to (4), even if analyzed (or stipulated) to have a truth value, must nevertheless still have the same truth value as (6) Dumbo has black ears, because (6) expresses exactly the same singular propositional schema as (5).6 Meinongianism circumvents this: Dumbo and Mickey Mouse are different nonexistent beings with different properties.Thus (4) expresses , and (5) expresses . Given the foregoing assumptions, the following seems to exhaust the possibilities7 (compare Jeshion 2010a, 15–​16): (i) Despite appearances, (4) and (6) are different from (1). The propositions expressed by sentences like (4)  and (6)  have additional semantic/​pragmatic content—​a fictional/​pretense/​intentional operator, or the like, that “seals off ” ontological implication—​that the propositions expressed by sentences like (1) don’t have (Contessa 2012; Everett 2013; García-​Carpintero 2010; Azzouni 2012 and Collins 2019, for criticism). (ii) (4) and (6)—​despite appearances—​have the same truth values, modal statuses, etc., because they don’t express propositions, or they express the same propositions or propositional schemas (Bach 2010; Donnellan 1966; Recanati 2014; Sainsbury 2010).The appearance of different truth values, etc., is handled pragmatically or by an error theory. (iii) There are no singular sentences or thoughts:  a descriptivism (e.g., causal descriptivism, e.g., Lewis 1984, 58–​60) that’s sophisticated is needed that can handle empty thoughts and propositions (Kroon 2010). (iv) Meinongianism is right:  different Meinongian objects support the different truth values and modal statuses that (4) and (6) have (Salmon 1998; Crane 2013). I return to this in section VI, and give a fifth option on which simple nontology is compatible with empty singularity.

IV Acquaintance A third widespread assumption makes it seem empty singularity is impossible all on its own. This is that singularity is governed by an acquaintance constraint that requires the something (acquainted 411

Jody Azzouni

with) to exist.8 Russell’s restricted acquaintance is broadened by subsequent neo-​Russellians in two directions, cognitively/​epistemically and metaphysically (hybrid views are also possible). Epistemic broadenings allow acquaintance in other than epistemic states of the pure manifestation of an object to an agent. Consider what Evans (1982) calls Russell’s principle (Hawthorne and Manley 2012, 71, n1 for references; their ­chapter 3 for criticism): To have a singular thought about an object, one must know which object one is thinking about. This principle can be interpreted more or less strictly depending on conditions for the successful discrimination of objects. Regardless of how it’s understood, it’s widely thought successful discrimination suffices to recognize the acquainted item exists. Cognitive/​epistemic broadenings of acquaintance must be sharply distinguished from metaphysical broadenings. The agent is aware of significant aspects of the mechanisms of acquaintance in the first set of cases. Metaphysical broadenings of acquaintance don’t require this; they turn on the agent being in relations to items singularly referred to, or thought of, where being in these relations needn’t place a cognitive condition—​awareness of the relations—​on the agent. Some philosophers claim that an appropriate sort of causation explains the singular use of names and demonstratives (Bach 2010; Recanati 2010; for criticism, Hawthorne and Manley 2012, 17–​18; Jeshion 2014, 76–​77). Broad acquaintance relations that aren’t strictly causal have also been suggested: agents can singularly refer to here and now, there and then, via a metaphysical relation of temporal or spatial locality—​not by causal relations with time slices and spatial locations. Cognitive grasping of the existence of what’s singularly referred to isn’t necessarily required, as I noted; that what’s acquainted with exist, however, is widely required.9 Acquaintance-​nihilists reject acquaintance altogether. Hawthorne and Manley (2012) reject it along with a rejection of the singular/​descriptive distinction itself; Jeshion (2014) rejects acquaintance, and via that, rejects the mutual exclusion of the singular/​descriptive distinction. Others reject acquaintance, but retain the mutual exclusion of the distinction by grounding it psychologically, and denying its semantic status (Azzouni 2011; Crane 2011).10 Jeshion (2010b) argues that emotional commitments determine whether a name-​use is singular; Azzouni (2011) argues that the singularity/​ description distinction turns on whether the agent is focused on the cognitive relation to the relata or instead on the relata itself, that is, whether the agent thinks of the relatum via her relationship to it or not. A relatum’s existence isn’t required on his view (see section VI for further discussion). For Crane (2011, 2013), the distinction turns on whether the agent’s mental file must be of one item or can be of more than one. Mental file models are popular;11 these attribute files or “dossiers” to agents that contain descriptive contents on (real or unreal) referents.12 An agent’s mental file on “the moon”, for example, contains various descriptions that the agent attributes to the moon. Philosophical uses of mental files are oblique to current cognitive/​psychological results in vision science because they go far beyond perceptual cases; thus many different versions of them are being developed by philosophers. Jeshion (2014) claims that such files are opened if an individual (real or unreal) is “significant”; Crane distinguishes a singular file from a descriptive file by the requirement that singular files can only be of one thing—​no such requirement is placed on general files (Crane 2011, 2013; Azzouni 2011, for criticism). Recanati (2013, 2014) treats files as Fregean senses, their contents being singular requires objects; Azzouni (2011) requires a “mental file”—​which is just a list of beliefs—​to be singular if all those beliefs are defeasible: they can be deleted or changed without changing what the beliefs are taken to refer to (real or unreal). Otherwise the list of beliefs is descriptive. A simple nontologist who accepts empty singular thought must rule out acquaintance—​even broadened acquaintance—​on programmatic grounds alone. If the singular/​descriptive distinction is retained, another ground for it is needed—​as in Azzouni (2011). 412

That Which Doesn’t Exist

V  Empty and Nonempty Singularity are Semantically/​Psychologically the Same I’ve discussed theoretical constraints that philosophers have argued should (or shouldn’t) be placed on reference, in language and in thought, to distinguish two categories—​singular and descriptive.The debate over whether there can be empty singular thought or language, however, arises (or should arise) at the evidential level. Theoretical constraints are applied to and tested against apparent facts of usage and the phenomenological impressions of agential thought while using various linguistic devices. These are the data supporting the theories in this area. It’s here, at the evidential level that many theorists have ruled out empty singularity; they’ve done so by (implicitly) denying referential neutrality: There are no psychological/​semantic/​syntactic markers that distinguish reference to what exists and to what doesn’t (other than pragmatically-​expressible ontological commitments or denials undertaken by agents). This principle is a broad empirical claim about natural languages and the psychological facts about agents (us) who speak them. One corollary of it is Chomskian internalism (Chomsky 2000; Collins 2009; Pietroski 2003). A  second corollary is pragmatic ontological commitment:  all expressions of ontological commitment and denial are pragmatic (Azzouni 2010, 2017). A  third corollary is the falsity of disjunctivism (Snowdon 1981 for the classic statement and defense of the position; Azzouni 2010 for opposition). Rather, psychologically and perceptually, identically experienced percepts accompanied by (or “constituted of ”—​depending on one’s view) intentional states of various sorts are identical, regardless of whether what’s experienced is real or not. More strikingly, one’s knowledge that one’s hallucinating leaves completely untouched one’s object-​directed thinking: the semantics and truth values of one’s thought is the same—​apart from sheer ontological claims. This is the truth lurking in the traditional argument from hallucination. A fourth corollary is the maximally simple-​fiction hypothesis: there is nothing that semantically distinguishes fictive discourse from non-​fictive discourse (Collins 2019).13 Yet another corollary is explored in this article: that apparent examples of singularity, either in thought or in diction, run parallel, regardless of whether what’s singularly thought or spoken of exists or not. I’ll start with considerations about truth values. Consider the following sentences:14 (7) James Bond is the most famous sociopathic killer of all time. (8) Charles Manson is the most famous sociopathic killer of all time. (9) Daffy Duck is the most famous sociopathic killer of all time. Most of us have seen numerous films and pictures (and documentaries) about these characters; we’ve talked and speculated about them extensively (many of us have strong emotional attachments or aversions to them).We’ve demonstrated these referents, using pictures or films, to others,“That’s James Bond”, “That’s Charles Manson”, “That’s Daffy Duck”, we’ve pointed and said. (Few of us, though, have seen Charles Manson in person.) Our use of the names, “James Bond”, “Charles Manson” and “Daffy Duck”, therefore, are singular—​if we use any names at all singularly. Furthermore, (7) or (8) is true, not both (I have no idea which—​it may be a close call). Regardless, (9) is false. It’s implausible to treat (7) and (9) as having a different syntactic/​semantic structure from (8). It’s also implausible to treat (7) and (9) as gappy or as not expressing propositions (Kaplan 1977; Braun 1993)—​they’d then have the same truth values, unless some other pragmatic/​intentional apparatus is brought to bear that doesn’t bear in the case of (8) (Azzouni 2010; Collins 2019). I turn now to the modal and epistemic considerations that many philosophers have used to make their case for singularity. Kripke (1972, 1980) famously invented a number of thought experiments 413

Jody Azzouni

that—​on the face of it—​show that proper names operate differently from definite descriptions, epistemically and modally. Some philosophers have taken this as evidence for the singularity of many uses of proper names—​although Kripke has never committed himself on this. I’ll illustrate first the epistemic considerations that seem to separate (most) names from (most) descriptions. With respect to (8), we recognize that, unbeknownst to us, Charles Manson may never have done anything he’s famous for.15 Consider the possibility (some Americans believe this!) that the police and others organized a cover-​up. I omit spelling out details—​anyone can do it easily (having heard many conspiracy stories). Any such story is implausible but possible. Thus, any description of Charles Manson, e.g., “the man who masterminded several horrible murders on August 9, 1969 in a Benedict Canyon mansion in California”, is open to disconfirmation on epistemic grounds: for all we know, it may be that Charles Manson isn’t the man who masterminded several horrible murders on August 9, 1969 in a Benedict Canyon mansion in California. It’s not, however, that for all we know, Charles Manson isn’t Charles Manson. Fictional characters aren’t different. It’s also possible, but implausible, that how we describe James Bond is wrong. Perhaps manuscripts of Ian Fleming exist in which there is an intricate backstory (deliberately designed by Fleming to fool all his readers, whom he despised) in which the real James Bond is an innocent helpless fool, like Johnny English, while the man who uses his name (with the connivance of British Intelligence) is a brilliant (sociopathic) German linguist turned spy named Schmidt who speaks flawless Edinburgh-​sounding British English. With only a bit of ingenuity, examples can be constructed to show how the same sorts of epistemic fallibility we enjoy vis-​à-​vis what’s real we also enjoy vis-​à-​vis what’s unreal (Azzouni 2010). Isn’t there still a disanalogy that undercuts the semantic parallel between referring and nonreferring names? After all, Ian Fleming couldn’t have been wrong about James Bond; he got to stipulate (some of) the truths about Bond. Not quite. Fleming could have forgotten what he’d written, or he could have misconstrued it (Horowitz 1983 on why stipulation doesn’t confer epistemic privileges). It’s relatively obvious that James Bond is a sociopath; it’s relatively obvious Fleming didn’t think this. And couldn’t it be that Fleming didn’t write these stories—​someone else did—​although he thinks he did? Any attempt to find descriptions that hold “a priori” of fictional objects but not real ones—​and that can be used to semantically distinguish them—​ is hopeless. This is despite the fact that, obviously, the properties attributed to fictional beings are stipulated by someone or other! Thus, although these fictional names (usually) originate in circumstances of sheer stipulation, there are no “stipulative conventions” governing them that distinguish them semantically from nonempty names. Let’s turn to names and natural-​ language modality.16 Kripkean possible-​ worlds thought experiments that purport to show the rigidity of proper names do so regardless of the reality of what’s designated. Just as there are (nearby) possible worlds in which Charles Manson, a resentful drifter, masterminded no one’s murder, there are possible worlds in which James Bond isn’t the famous suave murderous British spy he is in our world. There are possible worlds in which the Bond books were published, but weren’t popular; there are possible worlds in which James Bond doesn’t spy or isn’t murderous. In none of these worlds can we say that James Bond isn’t James Bond.17 I turn to demonstratives, to illustrate the same lessons about the real and the unreal.The ideal, perhaps paradigmatic case of singular reference is perceptual: I point at someone I see, and I say, (10) She’s wearing red shoes. She may be an acquaintance of mine (in the ordinary sense), a close friend, or a stranger on the street. Regardless, it does seem I’m singularly thinking of her when I say (10), which has a truth value based on what she’s wearing. If she’s, instead, wearing blue shoes, I’ve spoken falsely—​either deliberately or not. 414

That Which Doesn’t Exist

Under experientially identical circumstances, I may say (10) although I know I’m hallucinating. I may be communicating information to vision scientists in an experiment, where I’ve taken a certain drug. In another sort of case, I’m wearing virtual-​reality glasses, which perceptually blend things that don’t exist—​generated from various computer programs—​seamlessly with things that do (for example, I may see two people who are perfectly placed in the environment, dancing side by side, although only one of them is real).18 Then I can also speak truly or falsely when I utter (10) (perhaps the unreal person is wearing blue shoes, not red). In this case, notice, I’m not speaking of the sensory experience—​I’m not saying that my experience has red shoes. I’m also not speaking of the hallucination—​if only because hallucinations don’t wear shoes either. I’m speaking of nothing at all (there’s nothing there). Nevertheless (10) is true, informative and singular (Azzouni 2010; Evans 1982 denies (10) expresses a proposition in these circumstances: instead it’s pretended by me that I’ve expressed a proposition.) To further illustrate, notice that Donnellan’s (1966, specifically 9–​10) attributive/​referential distinction—​see Bezuidenhout, this volume—​can be replicated in unreal circumstances; thus whatever lessons the examples illustrating that distinction impart about singularity are lessons about empty singularity as well. I’m wearing virtual-​reality glasses, and I’m at a party (with both real and unreal people). I say, “Who is the man drinking a martini?” (He’s Gandalf the wizard—​someone unreal, as it turns out—​and when I subsequently converse with him, I discover he’s drinking water.) That’s referential. A virtual-​reality experimenter, part of the team monitoring my experiences from afar, hears from an associate that a man at the party (real or not real isn’t said) is drinking a martini. “Who is the man drinking a martini”, she asks. Not singular—​just as in Donnellan’s original example. Consider this version of a thought experiment due to Kaplan (1977, 512–​513)—​one that’s used to establish the singularity of certain demonstrative thoughts (or statements). David points at Paul, who is to his right, and says (at time t), (11) He lives in New Jersey. (11) is distinct from the proposition, (12) The person on David’s immediate right (at time t) lives in New Jersey, for there is a possible world, presumably, in which the person to David’s right doesn’t live in New Jersey (because, in this world, although Paul still lives in New Jersey, the person to David’s immediate right is Aoife O’Ahem, who lives in Dublin). (11) is true in this world although (12) isn’t. A case like this works just as well for hallucinations (or for those wearing virtual-​reality glasses). Consider the following alternative thought experiment. David knows he’s hallucinating, and so does his audience. He often hallucinates his imaginary good friend, a large rabbit named Harvey (Chase 1944). He points to his right and says, (13) He’s a giant rabbit who lives in New Jersey. This is distinct from the statement, (14) The nonexistent being to David’s immediate right is a giant rabbit who lives in New Jersey. For, in a nearby possible world, David is, instead, hallucinating the Cat in the Hat (Seuss 1957) who doesn’t live in New Jersey, but in Massachusetts. (In the actual world, David sometimes hallucinates the Cat in the Hat, but rarely.) In this possible world (13) is true but (14) isn’t. A variant: it’s been such a long time since David has seen the Cat in the Hat that, in both worlds, when he senses the 415

Jody Azzouni

hallucinated being to his right, he doesn’t look when he utters (13), only nodding his head slightly in the being’s direction. “Are you sure?” the vision scientists ask, noticing he hasn’t looked. In both cases, he turns to his right and in our world says, “Yes, that’s him”, but in the other world, he says instead, “Oops, that’s the Cat in the Hat. (pause) Hi, how have you been doing?”19 Some philosophers use phenomenological facts to support their claims about singularity—​by indicating how, often, our thinking about something isn’t “conceptual” but instead of the object in some “direct” way. Perception is the paradigm case—​but hallucination or virtual reality replicates this phenomenology perfectly, as I’ve indicated. I’ll say a little more about this at the end of the next section.

VI  The Metaphysics of Singular Propositions (Again) Let’s return to the issue raised at the end of section III. Given the foregoing, how do we make sense of the application of structured propositions, if we want to use them as our background tool for characterizing singularity (and other semantic phenomena)? Semantic hypocrisy is the view that the logical devices—​e.g., quantifiers—​in a metalanguage housing the semantics of an object language must differ in their ontological commitments from those in the object language that the semantics target. Semantic hypocrisy makes simple nontology and the metaphysics of propositions appear incompatible. If we reject semantic hypocrisy, we treat the language of the semantics of languages as ontologically noncommitting just as we take the object languages to be—​in particular we take the logical devices used in both as equally ontologically committing or not. Suppose (for example) that the background mathematics for structured propositions is set theory. Set theory, let’s say, uses ordinary first-​order quantifiers.20 In that case, the quantifiers of set theory can “range over” the nonexistent, and indeed, can be designed so that there are terms of the set theory for nonsets, “MICKEY MOUSE”, “DUMBO”, etc., that are stipulated to co-​refer with object-​language terms, “Mickey Mouse”, “Dumbo”, etc. Then, the structured propositions that (4) and (6), “Mickey Mouse has black ears” and “Dumbo has black ears”, respectively express, can be taken, in the metalanguage, to be and . The proponent of structured singular propositions intent on using structured propositions to indicate the ways in which the truth values of those propositions depend on what they’re about, can’t do this on the picture of semantics I’m giving here. For treating empty singularity on a par with nonempty singularity requires giving up another background assumption: truth-​correspondence (see David 2018). One can, instead, adopt a neutral view of truth (Azzouni 2018); the other purposes semantics is traditionally put to, e.g., indicating the scope of compositionality in natural languages, characterizing the ways that semantic constraints operate along with syntactic constraints, etc., remain. One may worry, however, that with singular reference (empty or otherwise), we aren’t conceptualizing the object in order to refer to it, and we aren’t conceptualizing our means of referring to the object either. But then, mustn’t the object itself play a role in the process by which we refer to it? For how else is reference to work if we don’t conceptualize the object or our means of referring to it? And if so, isn’t empty singular thought ruled out, to begin with? (This has already been responded to in this article, but I’m using the answer to this question to summarize the results.) No. First, as mentioned in section IV, it isn’t required an agent conceptualize a referential relationship to be in it. Second, although “relationship” implies that two items are involved, all the relevant “relationships” that reference is constituted of (or supervenient on) don’t need objects as relata (they can be “one-​legged”). We can perceive something real; we can just as well perceive something unreal; we can hear about something real, or unreal; we can think about something real or unreal, and either case forms emotional attachments (of any sort) to it; we can use demonstratives or proper names or (referentially used) definite descriptions on what’s real or unreal. More generally described, we can 416

That Which Doesn’t Exist

be involved in a rich causal relationship, where (normally) that relationship is to something, but that’s actually to nothing at all. Third, the phenomenological experience of experiencing or thinking about something real requires “object-​directed” psychological states; but these states are the same whether someone is (knowingly or unknowingly) entertaining fictions or contemplating something real (Azzouni 2014). Any intimate psychological state directed towards something real can as easily be directed towards something unreal—​whether realized or not. This is the fundamental truth behind singular thought and reference being the same, whether empty or not.

Acknowledgments I’m grateful to John Collins for helpful comments on an earlier version of this; I’m similarly grateful to the editors.

Notes 1 Singular reference is often distinguished from descriptive reference by requiring objects for one but not the other. McDowell (1982, 204) defines a singular thought as “a thought that would not be available to be thought or expressed if the relevant object, or objects, did not exist”. Evans (1982) agrees. Donnellan (1974, 101) writes: “If a child says, ‘Santa Claus will come tonight’, he cannot have spoken the truth, although, for various reasons, I think it better to say that he has not even expressed a proposition”.The Stanford Encyclopedia of Philosophy article on the topic, Fitch and Nelson (2018), doesn’t mention the possibility of empty singularity—​ even in passing. Hawthorne and Manley’s (2012) book-​length discussion routinely presumes ontological commitment of singularity.They write (2012, 28, n88), “Meinongians aside, we can all agree that only existing things can serve as relata to the reference relation”. They (2012, 17, n47) do mention the possibility of empty singular thought, and cite Jeshion (2002). Jeshion (2002, 73, n1) writes: “To have a de re thought, it is not, I think, necessary for a singular proposition to be the content of the thought. I believe that there may be exceptions in instances in which an agent thinks with an empty name. If there is no object that is the object of the de re thought, then the agent cannot have a singular proposition as the thought content. (I am not sure whether “gappy” singular propositions will do.) I will not deal with this problem in this paper, but do so in ‘Farewell Acquaintance’. There I attempt to argue that we ought to abandon not only acquaintance, but also the traditional existence condition on de re thought”. This paper hasn’t appeared; Jeshion (2014, 79–​81) uses empty names—​presumed singular—​against acquaintance. (Surprisingly, the earlier Jeshion (2009) describes the functions of names as if there are no fictional/​mythic uses of them.) Recanati (2014) denies empty sentences and thoughts have contents or truth values (they are singular only by virtue of conveying singular “vehicles”). Bach (2010, 56) allows empty singular thought, but claims “the singular thought has no object, and has no hope of being true”. Sainsbury (2010) argues similarly. Azzouni (2010) takes empty singular sentences to have truth values varying according to their content and to not semantically differ from nonempty singular sentences; he treats empty singular thought similarly. 2 Propositional “metaphysics” is mired in controversy—​there are (currently) no convincing fully developed characterizations of “structured propositions”. This is, in part, because structured propositions are theoretical posits based on the apparent syntactic and semantic properties of utterances/​sentences—​which is itself mired in controversy. “Structured propositions” are commonly treated as abstracta—​often as set-​theoretic constructions. Requiring the relata of “Russellian terms”, proper names for example, to be simple constituents of these propositions while simultaneously honoring the structural requirements induced by the sentences they’re derived from, is nontrivial. See King (2007) for an attempt presupposing Chomsky’s minimalist program; see Hawthorne and Manley (2012, 14–​15) for a succinct description of (only some of) the serious problems structured propositions face. See Deutsche (2008) for a vigorous (perhaps angry) presentation of apparently overlooked objections to structured propositions, and in particular, to King’s approach. 3 Nomenclature is challenging. I hope “simple nontology” is a new phrase; the other natural labels, “noneism” in particular, are already co-​opted. Routley (1980) used “noneism” for his version of Meinongianism. Others, e.g., Lewis (1990) and Priest (2005), follow Routley. 4 Meinongianisms easily translate into univocality denials, and vice versa, by translating different Meinongian ontological statuses into different kinds of existence, and/​or translating items that don’t exist (but have properties) into ones that exist in particular ways. Given these structure-​preserving translations, hereon, for simplicity, I’ll describe Meinongianism as the alternative to simple nontology, leaving implicit the more complicated set of alternatives.

417

Jody Azzouni 5 This notation is deliberately incomplete: the singular proposition can’t be an ordered triple; rather, what are indicated are the singular constituents in the structured proposition, ordered as in the sentence that expresses that proposition; the real structure isn’t given. Recall note 2. 6 A view of singular propositions as individuated by referents (rather than as containing them) is also apparently incompatible with simple nontology. 7 Overlap is possible, e.g., as with Kroon (2010) and Taylor (2010), who use multiple strategies. 8 There are many acquaintance theorists. See Jeshion (2009, 387, n22) for a longish list. See Jeshion (2009, 388, n23) for a shorter list of “semantic instrumentalists”—​those she describes as rejecting acquaintance as a constraint on singular thought. (Hawthorne and Manley 2012 use instead “liberalism” for this position.) Kaplan (1977, 536) is an interesting case; according to him, we “manipulate the conceptual apparatus of direct reference” to generate singular thought: singularity originates by the manipulation of linguistic tools. Kaplan subsequently changed his view. 9 García-​Carpintero (2014, 8)  indicates a different faultline in the singularity debate:  whether “descriptive information might nonetheless figure in a full account of [the] semantic nature [of singularity]”. 10 Call this the nonsemantic singularity hypothesis. Donnellan’s (1966) attributive/​referential distinction supports it; Kripke’s (1977) reconstrual of these examples retains the support. Variants on Donnellan’s examples confirm it with respect to demonstratives, and the distinction between descriptive and singularly-​ used names also suggests it’s the speaker’s psychological state, and not semantics, that determines singularity vs. descriptivity. 11 Crane (2011, 37) and Hawthorne and Manley (2012, 16, n44) give some history. Jeshion (2009), and elsewhere, tries to base the wide-​ranging philosophical notion in contemporary vision theory. Recanati (2013) is dedicated to the topic; see Recanati, this volume, as well. 12 Mental-​file models are compatible, of course, with various acquaintance views. But I’m describing their use when rejecting or minimizing the role of acquaintance. 13 Collins (2019) writes: “(i) Fictional expressions as linguistic types or utterance tokens have no essential linguistic signature … such as uniform covert intensionality or a pretense or make-​believe mode of speech; (ii) the truth values of fiction-​involving statements depend on how the world/​reality is without recourse to any ontology (ficta) one would otherwise not endorse; (iii) fiction-​involving statements, both in and out and their copredication, can be variously true or false”. (Copredication will be discussed in a moment.) This thesis is compatible with understanding theater or fiction as pretenses within which all the sentences uttered (or written) fall—​where these pretenses aren’t to be characterized semantically. 14 These sentences illustrate the copredication phenomenon, studied in Collins (2017, 2019), where in-​fictional-​ text attributions are copredicated with metafictional ones. 15 Kripke (1972, 1980) considers an Aristotle who never became a philosopher and a Gödel who never showed incompleteness. 16 See Everett (2005) for critical and detailed discussion of fairly recent rejoinders to Kripke’s “modal arguments” by Nelson, Sosa and Stanley. It’s worth pointing out in this regard that there are deep parallels between how names operate within the scope of natural-​language modalities and how they operate given epistemic possibilities. This is important because some responses to Kripke’s “modal argument” won’t work against his “epistemic” argument. 17 Are these possible world Bonds the same as ours? We certainly talk as if they are—​just as we do with real objects. And similar issues arise in both cases about identity across possible worlds (Kripke 1972, 1980; Lewis 1986; Azzouni 2010). 18 We are, perhaps, only years—​not decades—​away from devices like this. 19 We can be misinformed about the content of our own hallucinations or dreams, etc. (Azzouni 2010). We can think we were dreaming of (or hallucinating) dwarves, but when we describe the experience later, we’re corrected: No, those were hobbits, not dwarves. After all you said they had hairy feet, didn’t have axes, like mushrooms, weren’t grumpy … (Tolkien 1965). 20 There are other options, of course, but the points made here generalize to the other cases.

References The asterisked items are especially concerned with empty singular reference. Azzouni, Jody. 2010. Talking about nothing: Numbers, hallucinations and fictions. Oxford: Oxford University Press.* Azzouni, Jody. 2011. Singular thoughts (object-​directed thoughts). Proceedings of the Aristotelian Society Supplementary Volume, 85: 45–​61.* Azzouni, Jody. 2014. Freeing talk of nothing from the cognitive illusion of aboutness. The Monist 97(4): 443–​459. Azzouni, Jody. 2012. Responses to Gabriele Contessa, Erin Easker, and Nikk Effingham. Analysis 72(2): 366–​379.

418

That Which Doesn’t Exist Azzouni, Jody. 2017. Ontology without borders. Oxford: Oxford University Press. Azzouni, Jody. 2018. Deflationist truth. In Glanzberg, ed., 2018, 477–​502. Bach, Kent. 2010. Getting a thing into a thought. In Jeshion 2010c, 39–​63.* Braun, David. 1993. Empty names. Noûs 27: 449–​469. Chase, Mary. 1944. Harvey. www.snowballpublishing.com (2014). Chisholm, Roderick M. 1972. Beyond being and nonbeing. In Peter van Inwagen and Dean W. Zimmerman, eds. Metaphysics: The big questions (2008), 40–​50. Oxford: Blackwell Publishing. Chomsky, Noam. 2000. New horizons in the study of language and mind. Cambridge: Cambridge University Press. Collins, John. 2009. Methodology not metaphysics: Against semantic externalism. Aristotelian Society 83: 53–​69. Collins, John. 2017.The copredication argument. Inquiry 60(7): 675–​702. DOI: 10.1080/​0020174X.2017.1321500. Collins, John. 2019. The diversity of fiction and copredication:  An accommodation problem. Erknenntnis. DOI: 10.1007/​s10670-​019-​00150-​1. Contessa, Gabriele. 2012. Sweet nothings. Analysis 72: 354–​366. Crane,Tim. 2011.The singularity of singular thought. Proceedings of the Aristotelian Society Supplementary Volume, 85: 21–​43.* Crane, Tim. 2013. The objects of thought. Oxford: Oxford University Press. David, Marian. 2018. The correspondence theory of truth. In Glanzberg, ed., 2018, 238–​258. Deutsche, Harry. 2008. Review of “The nature and structure of content”. Notre Dame Philosophical Reviews (May 24). Donnellan, Keith. 1966. Reference and definite descriptions. Reprinted in Joseph Almog and Paolo Leonardi, eds. Essays on reference, language, and mind. Oxford: Oxford University Press, 3–​30. Donnellan, Keith. 1974. Speaking of nothing. Reprinted in Joseph Almog and Paolo Leonardi, eds. Essays on reference, language, and mind. Oxford: Oxford University Press, 81–​114.* Evans, Garth. 1982. The varieties of reference. Oxford: Oxford University Press.* Everett, Anthony. 2005. Recent defenses of descriptivism. Mind & Language 20(1): 103–​139. Everett, Anthony. 2013. The nonexistent. Oxford: Oxford University Press. Fitch, Greg and Michael Nelson, Singular propositions. The Stanford Encyclopedia of Philosophy (Spring 2018 Edition), Edward N. Zalta (ed.), URL = https://​plato.stanford.edu/​archives/​spr2018/​entries/​propositions-​ singular/​. Frege, Gottlob. 1892. On Sinn and Bedeutung. (Translated by Max Black.) Reprinted in Michael Beaney, ed. The Frege reader. Malden, MA: Blackwell Publishers, 151–​171. García-​Carpintero, Manuel. 2010. Fictional singular imaginings. In Jeshion 2010c, 273–​299.* García-​Carpintero, Manuel. 2014. Introduction: The problem of empty representations. In García-​Carpintero and Martí, eds. 2014, 1–​22. García-​Carpintero, Manuel and Genoveva Martí, eds. 2014. Empty representations:  Reference and non-​existence. Oxford: Oxford University Press.* Glanzberg, Michael, ed. 2018. The Oxford handbook of truth. Oxford: Oxford University Press. Hawthorne, John and David Manley. 2012. The reference book. Oxford: Oxford University Press. Horowitz, Tamara. 1983. Stipulation and epistemological privilege. Philosophical Studies 44: 305–​318. Jeshion, Robin. 2002. Acquaintanceless de re belief. In Joseph Keim Campbell, Michael O’Rourke, and David Shier, eds. Meaning and truth: Investigations in philosophical semantics. New York: Seven Bridges Press,  53–​74. Jeshion, Robin. 2009. The significance of names. Mind & Language 24(4): 370–​403. Jeshion, Robin. 2010a. Introduction to New essays on singular thought. In Jeshion 2010c, 1–​35. Jeshion, Robin. 2010b. Singular thought: Acquaintance, semantic instrumentalism, and cognitivism. In Jeshion 2010c, 105–​140. Jeshion, Robin, ed. 2010c. New essays on singular thought. Oxford: Oxford University Press. Jeshion, Robin. 2014. Two dogmas of Russellianism. In García-​Carpintero and Marti, ed., 2014, 67–​90. Kaplan, David. 1977. Demonstratives. In J. Almog, H. Wettstein, and J. Perry, eds., Themes from Kaplan (1989). Oxford: Oxford University Press, 481–​563. King, Jeffrey C. 2007. The nature and structure of content. Oxford: Oxford University Press. Kripke, Saul A. 1972, 1980. Naming and necessity. Cambridge, MA: Harvard University Press. Kripke, Saul A. 1977. Speaker reference and semantic reference. In P. French, T. Uehling, and H. Wettstein, eds. Contemporary perspectives in the philosophy of language. Minneapolis, MN: University of Minnesota Press, 6–​27. Kroon, Frederick. 2010. Content relativism and the problem of empty names. In García-​Carpintero and Marti, eds., 2014, 142–​164. Lewis, David. 1984. Putnam’s paradox. In Papers in metaphysics and epistemology (1999). Cambridge: Cambridge University Press, 56–​77. Lewis, David. 1986. On the plurality of worlds. Oxford: Basil Blackwell.

419

Jody Azzouni Lewis, David. 1990 . Noneism or allism? In Papers in metaphysics and epistemology (1999). New York: Cambridge University Press, 152–​163. McDaniel, Kris. 2017. The fragmentation of being. Oxford: Oxford University Press. McDowell, John. 1982. Truth-​value gaps. Reprinted in Meaning, knowledge and reality (1998). Cambridge, MA: Harvard University Press, 199–​213. Pietroski, Paul. 2003. The character of natural language semantics. In A. Barber, ed., Epistemology of language. Oxford: Oxford University Press, 217–​256. Priest, Graham. 2005. Towards non-​being: The logic and metaphysics of intentionality. Oxford: Oxford University Press. Recanati, Franҫois. 2010. Singular thought: In defense of acquaintance. In Jeshion 2010c, 141–​189. Recanati, Franҫois. 2013. Mental files. Oxford: Oxford University Press. Recanati, Franҫois. 2014. Empty singular terms in the mental file framework. In García-​Carpintero and Marti, eds., 2014, 165–​182. Routley, Richard. 1980. Exploring Meinong’s jungle and beyond: An investigation of noneism and the theory of items, inter med. Philosophy department monograph 3. Canberra: Research School of Social Sciences, Australian National University. Russell, Bertrand. 1910. Knowledge by acquaintance and knowledge by description. Proceedings of the Aristotelian Society, 11: 108–​128. Russell, Bertrand. 1912. The problems of philosophy. Oxford: Oxford University Press. Sainsbury, R. M. 2010. Intentionality without exotica. In Jeshion 2010c, 300–​318.* Salmon, Nathan. 1998. Nonexistence. Noûs 32(3): 277–​319. Seuss, Dr. 1957. The cat in the hat. New York: Random House. Snowdon, Paul. 1981. Perception, vision, and causation. Proceedings of the Aristotelian Society 81: 175–​192. Taylor, Kenneth A. 2010.The things we do with empty names: Objectual representations, non-​veridical language games, and truth similitude. In García-​Carpintero and Marti, eds., 2014, 183–​214. Tolkien, J. R. R. 1965. The lord of the rings, Part one: The fellowship of the ring. New York: Ballantine Books.

420

32 NAMES AND SINGULAR THOUGHT Rachel Goodman

1  Names, Singular Thought and Name-​Based Singular Thought Influential work on proper names, most centrally associated with Kripke (1980), has had a significant influence in the literature on singular (or ‘de re’) thought.To many, work on names has seemed to entail, contra early theorists of singular thought like Russell (1910), that there are indeed singular thoughts about ordinary external objects like tables and chairs. It has also seemed to entail that the range of singular thoughts about ordinary external objects is much broader than many post-​Russellian theorists assumed it to be. For example, many of those theorists thought there were singular thoughts about external objects, but only those about which the thinker had discriminating knowledge, or towards which she bore some relatively demanding epistemic relation.1 In contrast, a dominant position among contemporary theorists of singular thought, whose views diverge in other respects, is that we are able to think singular thoughts about any object we can refer to by name. Firstly, if we assume (as I will for current purposes) that having a singular thought that is about an object o requires entertaining a singular content that contains o,2 we can see how Kripke’s view of names would seem to displace Russell’s view that we cannot have singular thoughts about ordinary external objects.3 Kripke is meant to have taught us that the content of a sentence containing an ordinary proper name is a singular proposition containing the referent of the name.4 So, it seems that, when I sincerely assert a belief about my mother with the sentence, ‘Luciana Goodman is Italian’, my belief has a singular proposition containing Luciana, as its content. Secondly, it seems we can refer by name to many objects besides the ones we know well (like our mothers!) or even know at all. This includes very distant (in space or time) objects of testimony, and even objects with which we may lack a testimonial connection at all (but for which we have coined a name).5 This fact about the range of name-​use has shaped the singular thought literature over the last several decades: the idea that any ‘acquaintance’ constraint on singular thought must be very permissive is now truly dominant;6 the view that there is no acquaintance requirement on singular thought is now fairly popular.7 Broadly, though, we can ask whether the facts about the semantics of names in fact entail much at all about the conditions under which we can have singular thoughts about external objects. This chapter can be viewed as an opinionated overview focused on this question. It outlines the reasoning and presumptions of theorists who do think important entailments of this kind exist. It also aims to illustrate a certain set of reasons to resist this idea. The piece is distinctive in approaching these 421

Rachel Goodman

questions by focusing on the function of names and what this function entails (or doesn’t) about the kinds of thoughts (singular or descriptive) we have when we use names to communicate.

2  The Name-​based Singular Thought Thesis The influence of the Kripkean account of names on views about singular thought is most clearly evident in the widespread acceptance of a view, which I have called the name-​based singular thought thesis (NBT). NBT holds that taking part in communication involving a proper name entails the ability to entertain singular thoughts about the name’s referent.8 The thinking behind NBT is natural enough, but worth being clear about. If proper names are Millian (if their semantic contribution in context to the content of a sentence containing them is their bearer rather than a description of their bearer), then utterances of sentences containing names semantically express singular contents.9 This seems to mean that understanding an utterance of a sentence containing a name involves having a singular thought (on the assumption that having a singular thought is entertaining a singular content). On the assumption that communication involves understanding, this generates NBT. A very common extension of NBT is the claim that name-​use can enable one to think otherwise unavailable singular thoughts about the name’s referent. This view—​call it extended-​NBT—​is generated via the assumption that there are cases where we successfully communicate with utterances of sentences containing names even though we have no antecedent ability to think a singular thought about the referent of the name (for example, cases in which one receives testimony that employs a name, about a thing or person one hasn’t encountered and doesn’t already know about). It’s worth noting the difference between simple NBT and extended-​NBT because theorists are not always clear about their relation to extended-​NBT, and some proponents of NBT, who are motivated by the range of name-​use to extend or reject the acquaintance constraint on singular thought, claim not to think that communication with a public name is itself what explains the ability to think singular thoughts about the name’s referent.10 Before saying anything about the plausibility of NBT and its extension, it’s worth briefly documenting the fact that influential thinkers, whose views about singular thought otherwise differ, have held these views to be true. In the context of defending a view on which linguistic competence can precede and explain semantic knowledge, Soames (1989) focuses on name-​use and claims that one can acquire the ability to think singular thoughts about Pluto on the basis of exposure to the name ‘Pluto’: In short, the explanation of my belief that Pluto is a distant planet involves the fact that, (i) I accept the sentence ‘Pluto is a distant planet’; (ii) the sentence expresses the proposition that Pluto is a distant planet; and (iii) I  am a competent speaker, and thereby understand the sentence. Moreover, my understanding of the sentence is not a matter of using it to express descriptive propositions that I might have come to believe on independent grounds.11 Similarly, Salmon (2004), makes the following claim about the case in which one ‘encounters someone’s name for the first time without being adequately introduced … say by looking at a new class enrollment list or a luggage identification tag’: If one then uses the name to state something about the person so named, while still having no idea whose name it is, one makes a de re assertion and expresses a de re belief. (‘This belongs to one Byron Mallone. Mr Mallone, whoever he is, has traveled to Israel and was very recently around someone who smokes cigars’.)12 422

Names and Singular Thought

NBT proponents vary in whether they emphasize cases in which name-​use comes at the end of a causal, testimonial chain leading back to the name’s referent, cases in which names are brought into the language through descriptive reference-​fixing, or both.13 For example, Kaplan (1989b) famously held a version of NBT that extends to names whose referents are established by acts of descriptive reference-​fixing: The introduction of a new proper name by means of dubbing in terms of description and the active contemplation of characters involving dthat-​terms—​two mechanisms for providing direct reference to the denotation of an arbitrary definite description—​constitute a form of cognitive restructuring; they broaden our range of thought.14 On Kaplan’s view, the sense in which thought is ‘broadened’ is that one is enabled to ‘apprehend singular propositions’ containing objects previously known only by description.15 Bach, who holds a nominal description theory of names (on which the meaning of a name is a metalinguistic description) nonetheless also endorses NBT. He claims that name-​based singular thoughts are a species of communication-​based singular thoughts, thus emphasizing cases where names come with a causal, representational connections to their referents. He claims that names are ‘made to order’ for the role of ‘enabling others to have de re thoughts about unfamiliar things’.16 Hearing an unfamiliar proper name puts one in a position to think singular thoughts about its referent: Now if the speaker is thinking of something by name, he is entertaining a mental token of that name; and the audience, upon hearing that token, forms a mental token of the same name, which he then retains in memory. Since the hearer’s mental token of the name ‘inherits’ the same objects as the speaker’s, the object of the hearer’s thought is determined relationally, not satisfactionally.17 In fact, resistance to NBT is very rare among singularists.18 However Evans (1982, 74)  arguably rejects NBT. In discussing name-​use that comes at the end of a causal, testimonial chain leading back to the referent, he claims that cases in which one uses a name but lacks discriminating knowledge of its reference are ‘straightforward instances, in the field of singular reference, of the gap that opens up between what a speaker says and what thoughts he may have in his mind’. In discussing names introduced by acts of descriptive reference-​fixing, he follows Grice (1969, 140) in claiming that we cannot produce new thoughts simply by the ‘stroke of a pen’.19

3  Assumptions behind NBT As I’ve stressed in previous work (Goodman 2018), although the thinking behind NBT and its extension seem compelling, it is important and under-​appreciated that both claims are far from obligatory. They are based on optional assumptions.20 The first assumption, required to generate the claim that understanding an utterance of a sentence containing a (non-​empty) name involves having a singular thought, is that understanding an utterance made with a sentence, s, requires grasping the semantic content expressed by s. I call this semantic content accessibility.21 This assumption is admittedly intuitive. A  function of language is to express our thoughts, so it seems natural that a successful linguistic exchange—​which I will take to require understanding—​involves the speaker using a sentence whose semantic content in context is the content of the thought she intends to convey, and a hearer grasping that content.22 From another perspective, though, it would be unsurprising if semantic content accessibility had exceptions. Languages are social tools, or social artifacts. Their features—​including the semantic contents of their sentences in context—​are not determined by any particular user’s relation to them, but by complex facts about the way they function in contexts of that kind for a community of users. 423

Rachel Goodman

Given this, it would hardly be surprising if different individuals could (successfully) use the social tool by bearing a range of different relations to it. Perhaps speakers and hearers can make successful use of a sentence—​can communicate using it—​by grasping slightly different information. If we think there is such a thing as the semantic content of a sentence in context, and that this content is determined by facts about its function for the community of speakers, then it seems unsurprising that successful use of the sentence to communicate might involve grasp by language users of some content that is different to the semantic content (but related to it in some systematic way, the story about which we need to tell). The second assumption frequently made by NBT proponents is crucial for extended-​NBT—​that is, for the conclusion that name-​use enables (previously unavailable) singular thoughts. This assumption is that there are cases where we successfully communicate with utterances of sentences containing names even though we have no antecedent ability to think a singular thought about the referent of the name. I call these uses pure testimony cases. The assumption is that there are pure testimony uses in which communication succeeds. In a pure testimony use of a name, the hearer lacks any independent means of having a singular thought about the name’s referent, which could be brought to bear in forming a singular thought about the referent of the name. For example, if you have an acquaintance I have never met or heard of before, and about whom I know nothing, and you communicate about her by uttering (1), (1) Manuela DeNicola is from Rome this is a pure testimony case. If, however, I have some memory-​based singular thoughts about Manuela, based originally on perceptual demonstrative thoughts I have had about her, and you tell me, (2) Manuela DeNicola, the Italian girl you met last Tuesday, is from Rome this will not be a pure testimony case, since I can bring to bear my independent ability to think singular thoughts about Manuela, to form a singular thought about Manuela DeNicola as such.23 If hearers in pure testimony cases grasp the singular content semantically expressed, then it seems that hearing the name enabled them to have otherwise unavailable singular thoughts. Extended-​NBT therefore relies on the claim that communication can be successful in pure testimony cases, because these are the cases that isolate the effect of name-​use on a thinker’s ability to think singular thoughts about the referent of the name.24 If both the assumption about pure testimony cases and semantic content accessibility hold, extended-​ NBT follows.25 In previous work (Goodman 2018), I allowed that there are cases of successful communication involving pure testimony uses of names, and resisted both NBT theses by arguing for an expanded conception of understanding, which involved rejecting semantic content accessibility. I tried to illustrate how successful communication with a name could take place (in pure testimony cases) without the hearer grasping a singular content containing the referent:  it could take place in virtue of the hearer grasping a metalinguistic or meta-​communicative descriptive content that is pragmatically conveyed by the utterance.26 I  claimed that a picture of communication rejecting semantic content accessibility in fact better reflects the two-​tiered structure of name-​using practices than one on which grasping the semantic content of a sentence containing a name is required for understanding.27 The suggestions outlined here are consistent with that picture and in fact intended to flesh it out from another angle. It is worth reflecting on the fact that intuitions about the success of communication in pure testimony cases are, at best, shaky. We can take this observation as a starting point and, by thinking about the functional role of names, ask why, or in what sense, pure testimony cases seem deficient. This 424

Names and Singular Thought

discussion will suggest that the functional role of names does motivate the claim that the semantic content expressed by a sentence containing a name is a singular proposition containing the name’s referent, but does not motivate the idea that the thoughts of speakers in such contexts are singular thoughts about the name’s referent. In this way, thinking, first, about the felicity of pure testimony cases in relation to the functional role of names, and next, about the functional role of names more generally, sheds light on NBT, extended-​NBT and semantic content accessibility from a new direction.The discussion is meant as an overview and starting point. It will not constitute a full argument against NBT or extended-​NBT, but rather an illustration of an outlook that rejects both NBT theses (sections 4 and 7). It will also provide an illustration of one way to go about defending them (section 5), as well as a discussion of the challenges facing it (section 6).

4  Pure Testimony Cases and the Function of Names Given a Millian picture of names, a natural enough idea is that the function of a proper name is to allow for a specifically referential form of communication—​that is, a form of communication about an object that does not go by way of a description that singles it out. And it might seem that pure testimony cases would simply project the possibility of referential communication into contexts where some or all participants don’t have antecedent familiarity with the object being communicated about. However, the situation is not so simple. Firstly, many have noted that pure testimony cases are often infelicitous. For example, Evans (1982, 310) points out, it is ‘impolite to use the name of something right off if one does not expect one’s audience to be able to identify the referent’.28 Setting aside Evans’s stringent theoretical sense of ‘identify the referent’,29 there does seem to be some sort of norm against using a name in cases where one knows or believes one’s interlocutor has no familiarity with its referent as such. And it’s natural to think that this felt infelicity, impoliteness or abnormality suggests that referential communication in such cases fails. If this is correct, then it blocks extended-​NBT. But why would it be that communication in pure testimony cases fails? Answering this question requires us to further spell out the function of names. What is it that names are for? What is their communicative function? As Strawson (1974) and Evans (1982) emphasize, they seem to solve a problem that would otherwise make it hard for us to communicate about particular individuals when they’re not present.30 We cannot rely on enough natural overlap in the descriptive information about an individual possessed by different speakers to make describing that individual a reliable way to communicate about it. If I know that an individual o is the Φ, I cannot rely on your also knowing that o is the Φ (you may have other ways of describing o, but not this one) so communication by way of description is unreliable. Names help to solve this problem by creating ‘an arbitrary distinguishing feature which everyone learns’ so that, in using that feature of o (its name) to communicate about it, I can (more) reliably expect that my interlocutor will know which thing I intend.31 Thus, it is the institution of a name-​using practice connecting an ‘arbitrary distinguishing mark’ to a particular individual, which makes communication go smoothly. So, what names function to allow for is coordination on an object o as the object of interest, in cases in which interlocutors each have banks or stores of information about o that differ (or are likely to differ, or cannot be relied upon not to differ, etc.). Notice, though, there is only a problem for names to solve if what we’re aiming at is coordination, in the sense that interlocutors already have some body of information about an object o—​call it a ‘mental file’ if you like, but this popular terminology is inessential to the basic point—​and the aim is to coordinate those existent bodies of information. If the aim were simply to communicate about an object o, without necessarily coordinating existing bodies of information, then a description would do just fine: when I say, ‘I had lunch with the Φ’ and you form the belief that I had lunch with the Φ (without realizing that the Φ is the Italian girl you met last Tuesday) this would count as communicative success. 425

Rachel Goodman

So the kind of communication it is the central function of names to facilitate is the kind in which speaker and hearer coordinate their existing bodies of belief about an object o. And names function to facilitate this by allowing us to bypass descriptions of the object.32 I’ll call cases like this, in which names perform their primary communicative function, name-​based referential communication. The communication in such cases is referential in that it does not work by way of expressing a descriptive content about the object of communication; it is name-​based in that it is the kind of case in which names perform their primary function.33 But if we view the central function of names as the facilitation of name-​based referential communication, then this does suggest that pure testimony cases are deficient. They are cases in which a name cannot perform its primary function because one party doesn’t yet have a bank or body of information about o such that the name can function to ‘call it up’ and achieve coordination. This would explain the infelicity of pure testimony uses of names: in such cases one is using a linguistic device in a context in which it cannot fulfill its function. When you use a name, you’re presuming that your interlocutor has a body of information about its referent that is associated with that name. When she doesn’t, you violate norms of use.34 On this way of seeing things, pure testimony cases are a bit like cases in which someone uses a demonstrative expression when their interlocutor is blindfolded. You can do this and, when you do, it is true that some information will indeed be conveyed, but the expression won’t fulfill its central communicative function.The function of a demonstrative is to allow for communication about a particular individual by exploiting contextual cues to make clear which individual is intended. If, when you are blindfolded, I point at a woman in the vicinity and say, ‘that woman is from Rome’, I cannot expect my utterance to perform its primary function. At best, I manage to convey the information that some perceptually salient woman is from Rome. Similarly, the function of a name is to coordinate existing bodies of belief about its referent. If I utter (1) when you have no existing body of belief about Manuela DeNicola as such, I at best manage to convey the information that some person named ‘Manuela DeNicola’ is from Rome.35 This view of pure testimony cases as defective suggests that extended-​NBT is false because extended-​ NBT only gets off the ground in virtue of there being pure testimony cases where communication is successful. If pure testimony cases are defective, then extended-​NBT isn’t generated. Having said this, it is worth reflecting again on the initial intuition—​that pure testimony cases are infelicitous, impolite or uncooperative—​in order to see that it too is somewhat shaky and is therefore a starting point at best. We might note that, while discourse initial uses of names in pure testimony contexts are infelicitous, the infelicity is easily resolved by adding supplementary descriptive information to the name. Me beginning an exchange with (1) may be problematic if you don’t know Manuela DeNicola, but beginning with (3) or (4) seems fine: (3) My brother’s friend, Manuela DeNicola, is from Rome (4) This girl I know, Manuela DeNicola, is from Rome Since these felicity-​making descriptive supplementations could fall short of providing an individuating description for the name’s referent, you might think pure testimony cases still force extended-​ NBT on us.36 But, even if so, the suggestion that pure testimony cases are defective qua cases of name-​use, and therefore fail to support extended-​NBT, is not immediately defeated. Firstly, one might claim that the infelicity of pure testimony uses was just a starting point, which ultimately served to uncover a point about the function of names that does not rely on it. Secondly, perhaps even grammatically felicitous pure testimony uses are defective qua cases of name-​use. Perhaps, for example, the felicity-​making descriptive supplementations like those in (3) and (4) in fact function to highlight and mitigate the

426

Names and Singular Thought

fact that a name is being used without the expectation that it can perform its primary communicative function. The additional descriptive information provides some context for the name-​use, and thereby makes the communication go more smoothly, despite the fact that the name cannot fulfill the primary function of a name. More discussion of the range of cases is necessary.

5  Referential Communication, Significant Objects and the Learning of Names Section 4 illustrated the shape that one kind of argument against extended-​NBT might take: it might start with the idea that the function of names is to allow for name-​based referential communication, and be based on the claim that pure testimony cases are cases in which a name cannot perform its primary function. But, this leaves out part of the story about the function of names. And this part forms the beginning of a story about how name-​use might be said to generate singular thought. The story about the function of names already discussed is that the role of a name is, to use Evans’s terms, to provide ‘an arbitrary distinguishing mark’, the use of which allows speakers to coordinate their existing, differing but coreferential, bodies of information. The part left out so far is that this role can only be fulfilled if the arbitrary distinguishing mark is one that everyone in the community (or relevant part of the community, or relevant contexts) knows. This means that it is required by the function of names that there is a norm that one should learn the names of people and things one takes oneself (and others) to have cause to communicate about across a range of different contexts.37 What does this mean for generally competent speakers of a language? If they come across an unfamiliar name being used—​that is, if they find themselves playing the role of hearers in a pure testimony case—​then this is an indication of the significance of the name’s referent for some group: it is an indication that others in the community take themselves to have cause to communicate about this object across a range of contexts.38 Thus, as a competent speaker, if you come across a name being used in some salient corner of your community, this is an indication that you ought to learn it. That is, an indication that you ought to form a body of beliefs and attach the name to it—​or, to use the optional terminology, you ought to open a mental file and label it with the name.39 What is it for a body of beliefs (or a file) to be ‘labeled’ with the name? All this comes to is that the name will play what I will call a governing role with respect to the way information is handled for that body of information (or file).40 When new information arises, the question of whether this information should be included in this body of information depends on the answer to the question, ‘is this information about the object called “NN”?’. Only information for which the answer is ‘yes’ belongs. In other words, encountering a new name in some salient corner of your community comes with a norm that you should form a body of information that is then kept in the kind of order that would make name-​based referential communication using that name possible on future occasions. It might be suggested that this gives us a way to motivate extended-​NBT despite the points made in section 4. Perhaps it is true that the function of names is to facilitate name-​based referential communication but, since this function relies on a norm that new names should be learned, pure testimony cases are cases in which competent speakers of the language can be converted into speakers who are in a position, on future occasions, to take part in name-​based referential communication about a name’s referent? So (assuming that taking part in name-​based referential communication requires singular thought about the name’s referent) name-​use does generate previously unavailable singular thoughts about the name’s referent after all.41 I do think it is basically reasonable to think that, in this way, pure testimony cases play a role in converting language users into speakers who can participate in name-​based referential communication with a particular name. And, it’s worth noting that this suggestion marks a choice-​point for extended-​ NBT proponents: do they think that discourse initial uses of names in pure testimony contexts constitute successful name-​based communication? Or are they happy with the kind of story suggested here about how name-​use might generate singular thoughts about the name’s referent?

427

Rachel Goodman

If the latter, it remains true that discourse initial uses of names in pure testimony contexts are cases in which names cannot perform their primary function—​that is, even if they are part of the story about one way they can come to perform this function in the future. However, it is important that while the suggestion in this section points in the direction of a defense of extended-​NBT, it falls short of amounting to such a defense. This is for a basic reason to be discussed next: even if name-​use in pure testimony cases puts speakers in a position, in the future, to take part in successful name-​based referential communication, it is not to be taken for granted that name-​based referential communication requires singular thoughts of its participants. That is, even in cases in which names do perform their primary function, we cannot simply assume that it is required that participants think singular thoughts about the name’s referent. Another way to put this: setting aside extended-​NBT, NBT itself—​even when restricted to paradigm cases of name-​ use—​is disputable. And, it might be suggested that thinking about the function of names and the nature of name-​based referential communication in fact shows that the connection between name-​use and singular thought is not what NBT assumes it to be. This is the subject of the next section.

6  Name-​Based Referential Communication and Singular Thought As things have been presented here, extended-​NBT depends, unsurprisingly, on whether NBT is true. I’ll start with a very flat-​footed version of the worry that might be pressed upon NBT proponents:  our reflections on the function of names yields no obvious sense in which name-​ based referential communication requires singular thoughts about the referent of the name on the part of its participants. The function of names is to allow for coordination of coreferential bodies of information—​or mental files, if you prefer—​which cannot be relied upon to contain overlapping descriptive information. This coordination is achieved when speakers associate the same ‘arbitrary distinguishing mark’ with their bodies of information. On the face of things, it seems consistent with this that the thoughts of participants are descriptive in nature rather than singular. In other words, it seems reasonable to ask: why can’t the bodies of information being coordinated in a case of name-​ based referential communication be based on descriptive ways of thinking about their object, even if they are also associated with a name?42 As long as there are interlocutors with bodies of information about a particular individual, the aim is to coordinate those bodies of information, and the same name is associated with those bodies of information, then that name can perform its primary function. That which it is the function of names to facilitate seems to have nothing in particular to do with singular thought. Setting aside a broader commitment to semantic content accessibility for now (we’ll address this issue in the next section), the commitment to NBT seems to rely on the assumption that thoughts employing bodies of information—​mental files, to use the popular language—​must be singular. In this sense, the approach we’ve taken, which focuses on the function of names to ask about their connection with singular thought, finds a natural meeting place with the focus of the singular thought literature in recent years. Those familiar with the literature will know that recent leading theorists do indeed make the assumption that file-​thinking must be singular. They hold what I have called the mental files conception of singular thought (MFC): that all and only file-​based thought is singular.43 I myself have rejected this view elsewhere,44 but it is a way to connect name-​use to singular thought, which respects the points made above about the function of proper names. We can therefore note that MFC is one way to defend NBT.45 Building a case for or against MFC is not our aim here, but it’s worth giving a sense both of the motivation for it and the reasons it can be rejected.46 One reason people have held MFC is that they’ve thought that the constitutive features of file-​ hood are inconsistent with descriptivism about any file-​based thoughts: mental files are said not to

428

Names and Singular Thought

be individuated by the descriptive information stored in them and their semantic content is also said not to be identified with this descriptive information. The same file can persist while descriptive information is discarded and added. Despite variation of descriptive information in a file over time, beliefs employing the file are said to be continued beliefs.47 In addition, mistaken information about the referent, which is contained in a file, is said not to defeat referential success. Thus, the semantic content of thoughts employing the file cannot be equated with the entirety of the information stored in it. Since the semantic content of a file-​based thought is not identified with the descriptive information stored in the file, the story goes, the semantic content must be singular rather than descriptive.48 However, as Goodman (2016a) points out, it is consistent with the idea that files are not individuated by the entirety of the descriptive information stored in them and their semantic content is not given by a conjunction of this information, that some files are descriptive, and that the semantic content of thoughts employing these files is descriptive. In a ‘descriptive file’, individuation conditions and reference determination for the file are not determined by the entirety of the descriptive information stored in the file, but by a privileged piece of descriptive information (or subset of the descriptive information). Short of a mere stipulation that what we mean by ‘singular thought’ is file-​based thought, it certainly doesn’t follow from a thought’s being file-​based that it is singular.49 Another motivation for MFC comes from the suggestion that mental files give us an empirically respectable and purportedly cognitively real way of understanding the admittedly murky category of singular thought. And, indeed, MFC proponents have tended to appeal (albeit somewhat loosely) to cognitive scientific work on mental files to motivate their view. Again, a full discussion of the issues is beyond the scope of this piece, but it is far from clear (and I’ll venture to say, unlikely) that the philosophers’ notion of a mental file—​essentially, a body of coordinated beliefs allowing for a certain characteristic kind of justified inference—​picks out the same thing as the notions on offer in the cognitive scientific literature. Most centrally, the cognitive scientific files philosophers most frequently appeal to are posited to explain specifically perceptual phenomena.50 Even if the cognitive scientific work that posits ‘files’ demonstrates that certain perceptual states are singular, there is no reason to think these points extend to the non-​perceptual or ‘long term’ files philosophers posit as used in name-​based communication.51 One way to see how optional the view is that the thoughts involved in name-​based referential communication are singular, is to step back from the terminology of ‘mental files’ and the common (in my view mistaken) attendant associations with singularity, and focus instead simply on the fact that names function to allow speakers to coordinate coreferential bodies or banks of information. Seen this way, it seems arbitrary to insist that the thoughts involved should be singular.Why should all thoughts attached to bodies of information be singular? What about the very idea of a collection of information implies the singularity of the thoughts attached to such a collection? Why would the fact of information being clustered or organized as a body that is assumed to corefer imply singular rather than descriptive thought?52 Despite the suggestion above that it seems that names can perform their primary function without singular thoughts on the part of users, the idea that thoughts attached to bodies of information or mental files are always singular has had a strong influence on thinking about the connection between names and singular thought even in the work of those who have been clearest about the functional role of names. In recent literature, the connection has played an important role in Robin Jeshion’s work on the function of names and the nature of singular thought (Jeshion 2002, 2009, 2010). In earlier literature, Evans (1982) combines the insights about the function of names, which are the basis for my points in sections 4 and 5, with the claim that names invoke ‘information-​based thoughts’. These are thoughts that are ‘governed’ by an information-​relation (like perception or testimony) to the object they are about, and are thereby singular not descriptive thoughts.53 However, the line suggested by denying NBT is that the claim that names invoke specifically information-​based thoughts is not entailed, or even really suggested, by the communicative function of names.

429

Rachel Goodman

7  Singular Linguistic Content and Singular Thought: Denying Content Accessibility In the previous section we saw the way that NBT can be connected with the mental files conception of singular thought (MFC), thereby pointing in the direction of one way to defend a modest version of extended-​NBT. I  also outlined an opposing suggestion:  that there is no sense in which paradigmatic name-​based referential communication requires singular thoughts of its participants. But this suggestion may sound odd to some. A way to bring this out is to voice some questions it will raise. First, if name-​based referential communication does not require singular thoughts of participants, in what sense is it referential—​that is, why should we think the semantic content expressed in such communication is singular? Is rejecting NBT consistent with Millianism? Second, if the semantic content expressed in such cases is singular, then doesn’t the suggestion imply that participants need not understand the utterances involved in these communications? Discussion of these questions allows us to elaborate the outlook that rejects NBT, and to see how it preserves Millianism by rejecting semantic content accessibility. This approach can make space—​on at least one way of spelling it out—​for an important connection between proper names and singular thought. But it will not be the connection usually assumed to hold. First, on the picture being outlined, in what sense is name-​based referential communication indeed referential—​that is, in what sense, if any, is the communication singular? The obvious sense in which the communication is singular is that it bypasses the need to express a descriptive content that singles out the object of interest. It does this by exploiting an ‘arbitrary distinguishing mark’, rather than a description the object uniquely satisfies.54 But why should we think that the semantic content expressed in such cases is singular? The answer is related to the fact that we cannot assume sufficient overlap in the descriptive information possessed by different individuals to make describing an individual a reliable way to communicate about it. Imagine all the different ways that different individuals might have at their disposal to describe an object o, and note that what these descriptions have in common is that they are descriptions of the same object: o. A natural theoretical move is to say that the semantic content of the expression—​which does not have the content of any of the particular descriptions associated with it but whose function it is to coordinate—​is o. A similar idea is in fact stressed by a Millian like Soames (2002, ch. 3) who (setting aside his endorsement of NBT) holds that the semantic content of a name can be identified with the informational content that is common across the range of descriptive contents associated with the name by different speakers. Soames claims that these differing descriptive contents may be conveyed by uses of the name, but only what is common between them makes its way into the semantic content expressed. Second, isn’t this view one on which participants in successful name-​based referential communication need not understand the utterances involved? Goodman (2018) denies this implication by suggesting that understanding an utterance need not involve grasping the singular content semantically expressed by it. The proposal is thereby that semantic content accessibility does not hold for paradigm cases of communication involving names. This may seem to some like a bullet to bite, but there are at least two ways to defend this move that make it seem not only acceptable, but motivated. Again, an initial elaboration, rather than a full defense, is what this chapter offers. The first way is the simpler one. It involves pointing out that the falsity of semantic content accessibility is in fact unsurprising for basically the simple reason stated in section 3, and elaborated by the discussion of the function of names in sections 4 and 5. Names are a social tool, used (paradigmatically) to coordinate (potentially) differing bodies of information that nonetheless have in common that they are about the same thing. Their function makes it the case that the content semantically expressed by uses of sentences containing them is singular. It is the common object of interest, rather than the semantic content of the attending thoughts of speakers, that is marked by the semantic content of the expression. And therefore grasping that singular content is not what’s required to successfully use the social tool to perform its function.This means that one can count as understanding a use 430

Names and Singular Thought

of a name—​even in a case of name-​based referential communication—​when the semantic content of one’s thought differs from the semantic content expressed. A second way to defend the potential gap that opens up between the semantic content expressed by a use of a name and the contents of the thoughts of speakers who understand these uses, is inspired by Evans’s (1982, ch. 11) account of name-​using practices, and was laid out in Goodman (2018).55 This way to motivate rejecting semantic content accessibility ties the semantic content expressed by proper names more straightforwardly to facts about the contents of the thoughts of (some) users, but claims that this semantic content aligns only with the thought contents of a group of users who play a privileged role in the name-​using practice. Evans’s idea was that name-​using practices have an asymmetric two-​tier structure. There are two ways to participate successfully in such practices.The possibility of successful name-​based referential communication is supported by the existence of systematic practices whereby ‘Manuela’ is used to refer to Manuela, ‘Luciana’ is used to refer to Luciana, ‘Felicia’ is used to refer to Felicia, and so on. Producers in a particular name-​using practice are users of the name who have the ability to think singular thoughts about the name’s referent as such. They are speakers who can think true singular thoughts that might be expressed with, ‘that is Manuela’ or ‘she is Luciana’. They have an independent ability to think singular thoughts about the referent of ‘NN’, and also knowledge that the object of those thoughts is NN. According to Evans, the role of producers in a practice is to anchor the name to its referent.That is, the fact that Manuela is the individual who producers judge to be named ‘Manuela DeNicola’ is what makes it the case that the name ‘Manuela DeNicola’ refers to the individual to whom it refers. One of Evans’s central contributions was to make this claim about the role of producers in a name-​ using practice plausible, by pointing out that cases in which the judgments of producers shift (such that a different object is now identified as the bearer of ‘NN’) are cases of reference change.56 What is particularly crucial for the rejection of NBT, however, is that a producer’s independent ability to entertain singular thoughts about the referent of a name, in addition to her knowledge of the referent as the bearer of the name, is what accounts for her ability to grasp the singular content expressed by a use of a name, rather than the other way around. Her participation in name-​based referential communication—​whether in a pure testimony case or after initiation into a name-​using practice through a pure testimony case—​does not generate the ability to think singular thoughts about the name’s referent. Rather, this ability serves to anchor the name’s semantic content without necessarily being available to all participants in the name-​using practice. However, being a producer is not the only way to successfully participate in a name-​using practice. A consumer is a speaker who can successfully communicate using a name but is not a producer. Consumers are participants in the practice who do not have independent knowledge of the referent of the name as such. For example, a speaker who becomes able to take part in name-​based referential communication with a particular name through initiation into the name-​using practice in a pure testimony case will be a consumer in the practice. Such users cannot grasp the singular semantic content expressed by uses of sentences containing the name, but this does not mean they cannot successfully participate in the practice and communicate using the name. This provides an explanation of how a speaker could understand a use of a sentence containing a name without grasping the semantic content expressed. The first of these two options is a simpler route to denying semantic content accessibility for communication involving names. The second has the benefit (or flaw, depending on your point of view) of tying the semantic content expressed by names more straightforwardly to the semantic content of the thoughts of speakers (in particular, producers). A related way to adjudicate between the two options is to ask whether a name-​using practice could exist for which there were no producers, and never have been any. The Evans-​inspired route to rejecting semantic content accessibility rests the semantic content of a name on the judgments of producers.57 If we think that a practice that supports successful name-​based referential communication could be built and sustained on the use of a name that is brought into the language through an act of descriptive reference-​fixing, then perhaps this speaks against the 431

Rachel Goodman

Evans-​inspired view.This makes the connection between names and singular thought weaker still. But, either way, neither option presented here requires of participants in successful name-​based referential communication that they have singular thoughts about the name’s referent, and both options hold that the semantic content expressed by uses of names is singular. Both are thereby consistent with the suggestion that thinking about the function of names speaks against NBT and thereby extended-​NBT. An upshot of the perspective that rejects NBT and its extension is worth mentioning in conclusion. If NBT has been taken as a reason to think there are singular thoughts about external objects, acquaintance-​less singular thoughts, a wide range of singular thoughts, and so forth, rejecting NBT highlights the need to give arguments, from within the philosophy of mind, for these now-​popular views. But, given the murkiness of the category of singular thought, the stalemate with descriptivists about why we should think there are such thoughts, and the lack of clarity about their role in a broader theory of mind, this invitation to singularists who reject NBT should arguably be a welcome one.

Notes 1 See Strawson (1959) and Evans (1982). 2 This could be a requirement on having a singular thought that is about an object o even if there are cases of singular thoughts that are not mental states with singular content, because there could be a class of thoughts that are singular, despite not being about an object o. For example, see Azzouni’s contribution to this volume for discussion of putative cases of singular thoughts that lack singular content. The possibility of empty singular thoughts does not make a difference to the points made in this contribution concerning the relationship between proper names and singular thoughts. 3 For Russell, this view seems to go hand in hand with the view that ordinary names are disguised descriptions (but Russell (1910) contains some conflicting passages). 4 I’m leaving predicative uses of names (as in, ‘most Lucianas are Italian’) aside for current purposes. 5 There is a question to be asked about the conditions under which it’s possible to (successfully) coin a name for an object one knows only by description, but I’ll leave this issue aside. See Kaplan (1989a, 1989b), Strawson (1974), Jeshion (2009) for discussion. 6 See Jeshion’s (2010), ‘Standard Standard on Acquaintance’. 7 Jeshion (2002), Hawthorne and Manley (2012), Borg (2007). 8 See Goodman (2018), though note the difference in formulation. 9 Note, though, that even predicativists about names (who think names have metalinguistic descriptive meaning) tend to think that the semantic content expressed in context by referential uses of names is singular, so their view might also seem to entail NBT. 10 A good example is Jeshion (2010) who holds that the significance of an object known by description explains both the generation of singular thoughts about it and the ability to successfully introduce a (linguistic) name for it. She thinks generation of a mental name explains introduction of a public name, not vice versa (Jeshion 2010, section 4). Insofar as Jeshion is right that Kaplan also holds mental name production to be prior to public name production, Kaplan’s ‘instrumentalism’ would fall in a similar category (but see the quotation from Kaplan below). See section 5 for a way of thinking about the connection between this view and extended-​NBT. 11 Soames (1989, 588). 12 Salmon (2004). 13 See Goodman (2018) for further elaboration. 14 Kaplan (1989b). 15 Ibid. 16 Bach (1987, 33). Here, we have a clear commitment to extended-​NBT, although some of the above theorists arguably hold it too. 17 Bach (1987, 32). Bach is unusual in that he tries to give a story about how name use enables singular thought. See also Jeshion (2002), who emphasizes the need to do this. 18 Goodman (2018) explicit rejects it, as does Geirsson (2013, 2018). See also Sullivan (2010) for problematization of NBT. Raven (2008) claims there are challenges to the related idea of testimonial acquaintance, but presupposes that such acquaintance must be possible and thereby, I suspect, endorses NBT. 19 Evans (1982, 50).Those who hold NBT but not its extension may dispute Evans’s claims about cases, but may claim to agree that new thoughts are not produced merely ‘by the stroke of a pen’ (Jeshion 2002, 63). 20 See Goodman (2018) for further discussion of both assumptions.

432

Names and Singular Thought 21 This principle is related to several that are widely held. The most famous is Kripke’s (1979) Disquotation Principle (p. 248): ‘If an individual i who understands a sentence S sincerely asserts S, then i believes (and therefore grasps) the proposition expressed by S’). Others are Jeshion’s (2001) Accessibility of Content, Hawthorne and Manley’s (2012) Anti-​Latitude and Recanati’s (1993) Congruence Principle. See also Jeshion (2010, 111–​112). 22 I’m talking here about the Russellian content of thoughts, and leaving aside whatever further fineness of grain one may think is part of our best account of thought content. 23 Cases in which testimony causes one to think a singular thought about an object, but in which one uses independent resources to form that thought are also not pure testimony cases in the relevant sense. E.g., in a case in which you use a demonstrative utterance to draw my attention to an object o in my visual field which I have not previously thought about (‘look at that!, pointing at o), I previously had no singular thoughts about o but they were nonetheless there for the taking.Your utterance caused me to have a singular thought about o but does not explain my ability to have singular thoughts about o. Intuitively, the fact that I perceive o is what explains that. 24 I’m aware I’ve spotted myself the assumption that successful communication requires understanding. I’ll assume that a view that doubted this could just as well be stated as a view that doubted semantic content accessibility. 25 Interestingly, some thinkers (Jeshion, e.g.) who claim that (linguistic) name-​use itself does not enable singular thought do not deny that there are pure testimony cases where hearers grasp the singular content expressed. This suggests they do not simply reject extended-​NBT, but rather adopt a nuanced version of it. See sections 5 and 6 for elaboration. 26 The pragmatic mechanism by which content is generated will be non-​Gricean, since the view holds that speakers cannot grasp the singular content semantically expressed. 27 This picture relied on Evans’s (1982) distinction between ‘producers’ and ‘consumers’ in a name-​using practice, and on the idea that both kinds of participants in the practice can successfully communicate using the name, although only producers can think singularly about the name’s referent as such. See section 7 of this chapter for discussion of the relation of this picture to the current proposals. 28 He claims, ‘In such a situation, politeness demands that one say, not “I had dinner with NN”, but, “I had dinner with someone called ‘NN’ ” ’. If true, this would accord well with the claims made here, but my sense is that something along the lines of, ‘I had dinner with my friend (my sister/​this guy, etc.) NN’ is just as good. See the end of this section for further discussion. 29 See Evans (1982, ch. 4). 30 That is, in cases where contextual cues and tools like demonstrations and perceptual salience do not allow us to use ‘one off ’, context sensitive, referential devices to communicate about them (see Evans 1982, ch.11). 31 Evans (1982, 380). 32 It is true that, along with the attachment of an ‘arbitrary distinguishing mark’ to an object—​that is, a name—​ comes the possibility of using a metalinguistic description (‘person called “NN” ’) to think about an object and in communication, but this is consistent with my claims about the function of names. Names solve a coordination problem through the institution of a practice by which an individual is associated with an arbitrary mark. That this connection can be used to generate a metalinguistic description does not cause problems for this claim. See also note 54. 33 That is, not every case in which a name is used is one in which a name performs its primary function. Not all cases of referential communication use names or are name-​based. 34 You may think that pure testimony uses of names cause hearers to initiate new files—​or banks of information—​ on the referent of the name. More on this in section 5. 35 Note, a person could form this belief and initiate a body of information about the (salient) person called ‘Manuela DeNicola’, such that they would add information to the body when it was believed to be information about the (salient) person called ‘Manuela DeNicola’. See section 5 for discussion of the relevance of this fact to extended-​NBT. 36 Goodman (2018, 4). (4) works despite the existence of many girls called ‘Manuela DeNicola’. Does (3) work even if my brother has two friends so named? I’m not sure, but perhaps it does. 37 It may be important to be able to communicate about an object in this context but, if we don’t expect this need to project into various contexts, a one-​off device (like a demonstrative), or a description, is more likely to be adequate. 38 See Jeshion (2009) for work that takes a Strawsonian starting point, but emphasizes the connection between name-​use and the significance of the object named. Also note: I’m claiming the norm to learn a name is contingent on the object being one that an individual takes themselves (and others) to have cause to communicate across a range of contexts (the norm is not that one should learn all the names). This is different to Jeshion’s significance requirement, but is not unrelated to it. 39 Of course, there are files or bodies of information not associated with names, but the point of one that is attached to a name is (partly) to facilitate name-​based referential communication.

433

Rachel Goodman 40 See Goodman (2016a) for more on governance of mental files. Here, I advert to what’s there called, ‘the gatekeeping role’. Note also that the story told there about governance and gatekeeping for files bears some resemblance to Geirsson’s account (in this volume) of the role that information in a file can play in generating anti-​substitution intuitions for coreferential names. 41 Setting aside some important details of her view, this suggestion does bear similarity to Jeshion’s (2002, 2010) outlook, in that it connects names to singular thought via a concept of significance (albeit through a somewhat different conception of significance than the one proposed by Jeshion). Jeshion’s view is essentially that significance is a condition on, and also produces, both name-​use and singular thought because it triggers the formation and use of a mental file. The suggestion I’m considering is that singular thoughts are required to understand name-​based referential communication, that name-​based referential communication requires interlocutors to coordinate their coreferential mental files, and that the significance that is indicated by coming across a new name causes one to open a name-​governed mental file. Insofar as this is similar to Jeshion’s view, this brings out the sense in which it might be fair to count Jeshion as a proponent of extended-​ NBT (not just NBT) even if she might not classify herself as such. 42 Note, that this is so for several different conceptions of what it means for a thought to be based on a descriptive way of thinking of an object: the thought can have its object determined by description, it can involve ‘description-​centered cognition’ and so forth. A plausible case in which the thoughts on which the bodies of information involved in name-​based referential communication are descriptive is the one discussed in section 5, where the body of information was itself formed on the basis of a pure testimony case, such that governing description is metalinguistic: ‘the person called, NN’. 43 The view is often simply assumed, but the most well-​known proponents (whose work is valuable in part because they attempt to argue for MFC) are Recanati (e.g. 2012, 2015) and Jeshion (e.g. 2002, 2010). 44 See Goodman (2016a, 2016b, 2018 and ms) for a case against MFC. 45 I take this to be one dimension of Jeshion’s strategy, and it is part of the value of her work that it has made this connection clear. 46 It’s implied by the above that one motivation is to adopt a theory of singular thought that maintains semantic content accessibility. And, indeed, it’s fair to say this is also one motivation for Jeshion. 47 See Perry (1980). 48 See, e.g., Grice (1969), Evans (1985), Donnellan (1978), Perry (1980), Recanati (2012, 2015) for elaboration of the file-​picture. See Goodman and Gray (forthcoming) for critical discussion of its commitments. 49 Wisely, central proponents of MFC speak against this kind of stipulation (see Jeshion 2010). 50 In particular, success in multiple object tracking tasks (see, Storm and Pylyshyn 1988) and object specific preview benefits (see, Kahneman, Treisman and Gibbs 1992). The files linguists appeal to are different, but it’s less clear why such file-​based thought should be construed as singular. 51 The proposal that they do is described by Murez, Smortchova and Strickland (2020) as ‘The Projection Argument’. They claim current empirical research does not support that proposal. See Geirsson (2018) for a similar suggestion. 52 This raises a final possible motivation for the idea that all thoughts involved in name-​based referential communication are singular. This is found in the storied but somewhat nebulous connection between singular thought and de jure coreference, or inferences that ‘trade on identity’ (Campbell 1987). It takes me too far afield to discuss this here but the motivation is examined and argued against in Goodman (ms). See Taylor (2010) and especially Dickie (2015) for attempts to clarify what this connection is meant to be. Although I disagree with this aspect of her view, Dickie’s is the most promising version of the suggestion. Skating over many details, her approach is to say that beliefs that are part of bodies of beliefs are formed and bundled through a particular kind of information-​management activity that gives rise to cognitive focus. The result of cognitive focus is to bring a particular object before the mind in a relation of robust or genuine ‘aboutness’. And this, for her, is what singular, but not descriptive thought, involves. 53 A thought is governed by an information relation when the thinker’s assessment of its truth or falsity is determined by information gained through that relation. See, e.g. Evans (1982, 121–​122). 54 Note, it is consistent with this that the existence of a practice whereby the arbitrary distinguishing mark, ‘NN’, is associated with a particular individual makes it the case that the individual bears a property which could be used to describe it: the property of being called ‘NN’. See also Hawthorne and Manley (2012, 227–​ 233) for a distinction between describing uses and calling uses of names. 55 See also Geirsson (2013, ch. 5; 2018) for what I understand to be a rejection of semantic content accessibility. 56 The point that semantic reference is determined by and tracks the judgments of users was first made, without the distinction between producers and consumers that makes its way into the more sophisticated version of the picture, in Evans (1985). 57 The way in which it does will have to be spelled out carefully, since we want to allow that there are name-​ using practices for which the names have determinate semantic content despite the fact that all producers have died or lost their status as such.

434

Names and Singular Thought

References Bach, K. (1987) Thought and Reference. Oxford: Oxford University Press. Borg, E. (2007) Minimal Semantics. Oxford: Oxford University Press. Campbell, John (1987) ‘Is Sense Transparent?’, Proceedings of the Aristotelian Society, 88: 273–​292. Dickie, I. (2015) Fixing Reference. Oxford: Oxford University Press. Donnellan, K. (1978) ‘Speaker Reference, Descriptions and Anaphora’, Syntax and Semantics, 9, 47–​68. Evans, G. (1982) The Varieties of Reference. J. McDowell (ed.) Oxford: Oxford University Press. Evans, G. (1985) ‘The Causal Theory of Names’, in Collected Papers. Oxford: Oxford University Press. Geirsson, H. (2013) Philosophy of Language and Webs of Information. New York: Routledge. Geirsson, H. (2018) ‘Singular Thought, Cognitivism, and Conscious Attention’, Erkenntnis, 83, 613–​626. Goodman, R. (2016a) ‘Against the Mental Files Conception of Singular Thought’, Review of Philosophy and Psychology, 7(2), 437–​461. Goodman, R. (2016b) ‘Cognitivism, Significance and Singular Thought’, The Philosophical Quarterly, 66(263), 236–​260. Goodman, R. (2018) ‘On the Supposed Connection between Proper Names and Singular Thought’, Synthese, 195(1), 197–​223. Goodman, R. (ms) ‘Singular Thought and Trading on Identity’. Goodman, R. and Gray, A. (forthcoming) ‘Mental Filing’, Noûs. Grice, H. P. (1969) ‘Vacuous Names’, in Words and Objections, Donald Davidson and Jaako Hintikka (eds). Dordrecht: Reidel, 118–​145. Hawthorne, J. and Manley, D. (2012) The Reference Book. Oxford: Oxford University Press. Jeshion, Robin (2001) ‘Donnellan on Neptune’, Philosophy and Phenomenological Research, 63(1): 111–​135. Jeshion, Robin (2002) ‘Acquaintanceless de re Belief ’, in Meaning and Truth: Investigations in Philosophical Semantics, J. Campbell, M. O’Rourke, and D. Shier (eds). New York: Seven Bridges Press. Jeshion, Robin (2009) ‘The Significance of Names’, Mind and Language, 24(4): 372–​405. Jeshion, Robin (2010) ‘Singular Thought: Acquaintance, Semantic Instrumentalism, and Cognitivism’, in New Essays on Singular Thought, R. Jeshion (ed.), Oxford: Oxford University Press. Kahneman, D., Treisman, A. and Gibbs, B. J. (1992) ‘The Reviewing of Object Files: Object-​Specific Integration of Information’, Cognitive Psychology, 24: 175–​219. Kaplan, D. (1989a) ‘Demonstratives’, in Themes from Kaplan, J. Almog and H. Wettstein (eds). Oxford: Oxford University Press, 481–​564. Kaplan, D. (1989b) ‘Afterthoughts’, in Themes from Kaplan, J. Almog and H. Wettstein (eds). Oxford:  Oxford University Press, 565–​614. Kripke, S. A. (1979) ‘A Puzzle about Belief ’, in Meaning and Use, A. Margalit (ed.). Dordrecht: Springer, 239–​283. Kripke, S. (1980) Naming and Necessity. Cambridge, MA: Harvard University Press. Murez, M., Smortchova, J. and Strickland, B. (2020) ‘The Mental Files Theory of Singular Thought:  A Psychological Perspective’, in Mental Files and Singular Thought, Rachel Goodman, James Genone and Nick Kroll (eds). Oxford: Oxford University Press. Perry, J. (1980) ‘A Problem about Continued Belief ’, Pacific Philosophical Quarterly, 61(4): 317–​332. Raven, M. J. (2008) ‘Problems for Testimonial Acquaintance’, Noûs, 42(4), 727–​745. Recanati, F. (1993) Direct Reference: From Language to Thought. Oxford: Blackwell. Recanati, F. (2012) Mental Files. Oxford: Oxford University Press. Recanati, F. (2015) Mental Files in Flux. Oxford: Oxford University Press. Russell, B. (1910) ‘Knowledge by Acquaintance and Knowledge by Description’, Proceedings of the Aristotelian Society, 11, 108–​128. Salmon, N. (2004) ‘The Good Bad and the Ugly’, in Descriptions and Beyond, A. Reimer and A. Bezuidenhout (eds). Oxford: Oxford University Press, 230–​260. Soames, S. (1989) ‘Semantics and Semantic Competence’, Philosophical Perspectives, 3, 575–​596. Soames, S. (2002) Beyond Rigidity: The Unfinished Semantic Agenda of Naming and Necessity. New York: Oxford University Press. Storm, R. and Pylyshyn, Z. (1988) ‘Tracking Multiple Independent Targets:  Evidence for a Parallel Tracking Mechanism’, Spatial Vision, 3(3), 179–​197. Strawson, P. S. (1959) Individuals. London: Routledge. Strawson, P. S. (1974) Subject and Predicate in Logic and Grammar. London: Routledge. Sullivan, A. (2010) ‘Millian Externalism’, in New Essays on Singular Thought, R. Jeshion (ed.). Oxford: Oxford University Press. Taylor, K. (2010) ‘On Singularity’, in New Essays on Singular Thought, R. Jeshion (ed.). Oxford:  Oxford University Press.

435

PART IX

Indexicals

33 HOW DEMONSTRATIVES AND INDEXICALS REALLY WORK Stephen Neale and Stephen Schiffer

What determines the referent of a demonstrative or indexical expression on a particular occasion of use? Our answer isn’t new, but it’s insufficiently appreciated: the referent of a demonstrative or indexical on a particular occasion of use is determined by the speaker’s referential intentions as constrained, in a way to be explained, by the expression’s meaning. This requires us to say what we take these meanings to be, which in turn requires us to say what we take meanings of expressions of any kind to be. We’ll take the meaning of an expression to be the fixed condition that something must satisfy in order to be the content of that expression on a given occasion of use. We take the content of a declarative sentence σ on a given occasion of use to be a proposition, and we take the content of a part of σ on that occasion to be its contribution to the determination of the content of σ on that occasion. For most of this paper we’ll operate as if (i) the propositions in question are Russellian, structured entities built up from objects and properties (as opposed to “modes of presentation” of them), and (ii) the content of a demonstrative or indexical on a particular occasion of use is its referent, so that, for example, the content of ‘She is French’ on a given occasion of use is a proposition that can be represented as the ordered pair ⟨x, being French⟩, where x is the referent of ‘she’ on that occasion of use.1 Therefore, relative to this Russellian assumption, we can say that the meaning of a demonstrative or indexical is the fixed condition that something must satisfy in order to be its referent (on an occasion of use).2 We address this explicitly in section VI.3 We take the meaning of a demonstrative or indexical δ to be a condition of the form: (A) x is the referent of δ on an occasion of use o iff (1) Cx (2) the speaker refers to x with δ on o. (So we take speaker reference to be a more basic notion than expression reference.) It follows from (A) that we take a difference in meaning between two demonstratives or indexicals to be simply a difference in condition Cx, which in the limit may be null. Two examples that will concern us later: (B) x is the referent of demonstrative ‘she’ on an occasion of use o iff (1) x is female; (2) the speaker refers to x with ‘she’ on o.

439

Stephen Neale and Stephen Schiffer

(C) x is the referent of ‘I’ on an occasion of use o iff (1) x = the speaker; (2) the speaker refers to x with ‘I’ on o. Now we need to explain the notion of speaker reference to which we’ve appealed; and to do that we need first to explain the notion of speaker meaning.

I  Speaker Meaning Speaker meaning is the most general kind of illocutionary act, the genus of which all other kinds of illocutionary acts are species. There are interrogative and imperatival acts of speaker meaning, but for ease and economy of exposition we’ll identify speaker meaning with assertoric speaker meaning (henceforth simply speaker meaning), i.e. a speaker’s meaning p, for some proposition p. For the Gricean, speaker meaning is defined wholly in terms of acting with certain audience-​ directed intentions. There is no official Gricean definition of speaker meaning, and Griceans may disagree among themselves about the exact details, but the following toy conditions illustrate the kind of account favored by many Griceans: (D) For any person S, proposition p, and utterance u,4 S meant p in uttering u iff for some person A and feature φ, S intended it to be common ground between S and A that u has φ and at least partly on that basis that S uttered u intending her utterance of u to result in A’s actively believing p. (Here we borrow Stalnaker’s (2002) conception of common ground:  a proposition p is common ground between two people iff they mutually believe p (or mutually assume p for immediate purposes). Two people mutually believe p iff both believe p, both believe that both believe p, both believe that both believe that both believe p, and so on.5) For the Gricean it is no requirement that φ be u’s meaning, or even that u have a meaning; but the Gricean is of course aware that normally when S means something in uttering u, u is a sentence and φ is its meaning; and, the Gricean explains, this is because when S means something by uttering a sentence σ, it’s by virtue of σ’s having the meaning it has that uttering σ is standardly an optimal way for S to make known to A the meaning-​constituting intentions with which S uttered σ (if you want to inform your child that Rome is the capital of Italy you can hardly do better than to utter ‘Rome is the capital of Italy’). We, however, do not wish to commit to speaker meaning’s being definable wholly in terms of a speaker’s non-​semantic propositional attitudes. We prefer to commit only to the claim that: (E) S meant p in uttering u only if, for some person A and feature φ, S uttered u intending it to be common ground between S and A that u has φ and at least partly on that basis common ground between them that in uttering u, S meant p.

II  Speaker Reference We take the foundational case of speaker reference, which we’ll call primary speaker reference (speaker referencep) to be a species of speaker meaning: (F) S referredp to x in (the course of) uttering u iff in uttering u, S meant an x-​dependent proposition. An x-​dependent proposition is a proposition that is individuated partly in terms of x and that wouldn’t exist if x didn’t exist. Relative to the assumption that the propositions speakers mean are Russellian, an x-​dependent proposition is a proposition that contains x (a so-​called singular proposition). For 440

How Demonstratives and Indexicals Work

example, we may represent the proposition that Romeo loves Juliet as the ordered pair ⟨⟨Romeo, Juliet⟩, loving⟩, so that the proposition that Romeo loves Juliet is both Romeo-​and Juliet-​dependent. Let’s call the notion of speaker referencep just defined referringp-​in. Referringp-​in needs to be distinguished from a closely related notion that we’ll call referringp-​with. So consider the following two scenarios: 1

During Jack’s cocktail party two guests, Rudy and Trudy, make their way to the upstairs drawing room to steal Jack’s Degas statuette. Rudy is the lookout while Trudy is charged with hiding the statuette in her handbag. At one point, Rudy whispers ‘He’s coming’, thereby meaning that Jack is coming, and in so doing refers to Jack. 2 Same set up, only this time it’s in coughing that Rudy means that Jack is coming, a cough being the pre-​arranged signal that Jack is coming, and in so doing Rudy refers to Jack. In both scenarios Rudy refers to Jack in meaning that Jack is coming, but there is the following important difference: in (1) Rudy refers to Jack in uttering ‘He’s coming’ and there is a part of the sentence uttered—​viz. ‘he’—​with which he refers to Jack; but in (2), although Rudy refers to Jack in coughing, there is no part of that coughing (or that cough) with which he refers to Jack.Thus, the other notion of speaker referencep we need at this point, referringp-​with, is defined as follows, where e is an expression that occurs at least once in u: (G) In uttering u, S referredp to x with e, relative to its i-​th occurrence in u, iff for some person A and property φ, S intended it to be common ground between S and A that the i-​th occurrence of e has φ and, at least partly on that basis, that S referredp to x in uttering u. φ can vary with kinds of expression. For example, if e is a proper name of x, φ may be that property

that e has just in case it’s common knowledge in the community to which speaker and hearer belong that there is a practice of referring to x with e. (This may look circular, but it’s not.) If e is a demonstrative or indexical … well, we’re coming to that.

III  Non-​Primary Speaker Reference Consider utterances of the following sentences and suppose that in producing them S meant that Odile was French (i.e. meant ⟨Odile, being French⟩): (1) That woman is French. (2) That woman next to that boy is French. (3) That woman next to that boy patting that dog is French. Then we would say that: • in uttering (1), S referredp to Odile with ‘that woman’; • in uttering (2), S referredp to Odile with ‘that woman next to that boy’; • in uttering (3), S referredp to Odile with ‘that woman next to that boy patting that dog’. In primary speaker reference, S refers to a thing in order to mean something about it. In non-​primary speaker reference S refers to a thing not in order to mean something about it, but in order to identify the thing to which S is making a primary reference. Primary speaker reference is first-​order speaker reference, and non-​primary speaker reference is speaker reference of any order > 1. In (2) there is a boy y to whom S makes a secondary (= second-​order) reference with ‘that boy’ by virtue of S’s referring to

441

Stephen Neale and Stephen Schiffer

y in order to identify the thing to which S is making a primary reference as a woman standing next y. In (3) there is a dog z to which S makes a tertiary reference with ‘that dog’ by virtue of S’s referring to z in order to identify the thing to which S is making a secondary reference as a boy patting z. But in every case of non-​primary speaker reference (henceforth speaker referencen), no matter of what order, there is an occurrence of a singular term δ′ within the occurrence of a singular term δ that doesn’t itself occur within the occurrence of any other singular term, and for some x′ S intends it to be common ground between S and A that S referredn to x′ with δ′ in order that, for some x, it be common ground that S referredp to x with δ. (H) S referredn to x′ in uttering u iff for some person A, thing x and relation R, S utters u intending it to be common ground between S and A that R(x′, x) and primarily on that basis common ground between them that S referredp to x in uttering u. This defines referringn-​in—​i.e. referring-​in for non-​primary speaker reference. We can now define referringn-​with: (J) In uttering u, S referredn to x with e, relative to its i-​th occurrence in u, iff for some person A and property φ, S intended it to be common ground between S and A that the i-​th occurrence of e has φ and, at least partly on that basis, that S referredn to x in uttering u. The general notions of referring-​in and referring-​with may now be defined as follows: (K) In uttering u, S referred to x iff in uttering u, S referredp or referredn to x. (L) In uttering u, S referred to x with e, relative to its i-​th occurrence in u, iff in uttering u, S referredp or referredn to x with e relative to its i-​th occurrence in u.

IV  The Semantics of Indexicals and Demonstratives Earlier we said that the meaning of a demonstrative is the fixed condition a thing must satisfy in order to be its referent on a given occasion of use. We illustrated this with two examples, the first of which was: (B) x is the referent of the demonstrative ‘she’ on an occasion of use o iff (1) x is female; (2) the speaker refers to x with ‘she’ on o. For present purposes, we’ll assume there are demonstratives ‘she’ and ‘her’ that are distinct from any other (e.g. bound) ‘she’ and ‘her’.6 Suppose the meaning of demonstrative ‘she’ is given by (B). Suppose, too, that one is informed that some speaker S, using demonstrative ‘she’, produced an utterance of the sentence ‘She is famous’ (unembedded) in conformity with its meaning. Then, just by knowing that, one will also know that, unless S was deluded, there was some female x such that S meant that x was famous. So suppose Sid said to Ava ‘She is famous’ and meant thereby that Mary was famous. Then Sid intended Ava to recognize that in uttering that sentence he meant that Mary was famous, and since (roughly speaking) one can’t intend one’s act to bring about a certain result unless one expects it to bring about that result, Sid also expected Ava to recognize that in uttering ‘She is famous’ he meant that Mary was famous.7 How might it be reasonable for Sid to have that expectation? Well, even before he spoke, he and Ava occupied an enormous area of common ground (that is to say, myriad propositions were already mutually believed by them), and included in that common ground was the proposition that if a speaker, using ‘she’ as a demonstrative, produces an utterance of 442

How Demonstratives and Indexicals Work

the (unembedded) sentence ‘She is famous’ in conformity with its meaning, then (delusion aside) there is some female x such that the speaker means that x is famous. Perhaps as he uttered ‘she’, Sid also nodded in the direction of Mary. Then it shouldn’t take much imagination to appreciate that the common ground occupied by Sid and Ava before Sid spoke, now updated to include the fact that he produced an utterance of the (unembedded) sentence ‘She is famous’ while nodding in the direction of Mary, enabled Ava to recognize that in uttering ‘She is famous’ Sid meant that Mary was famous, just as Sid intended and expected. So, although what we have proposed as the meaning of ‘she’ may seem sparse, it’s enough to do the job that ‘she’ needs to do.8 We especially want to emphasize that only confusion about the nature and formation of genuine intentions would lead one to suppose that condition (1) in (B) (the meaning of ‘she’) should invoke salience. The heart of such a confusion would be overlooking the fact that a rational person can’t perform any act in order to bring about a certain result unless she expects her performing that act to bring it about. The application of this truism to speaker reference is that a rational speaker can’t just do anything intending thereby to refer to a particular thing x unless she expects that her hearer will recognize that it’s x to which she is referring; “the expectation of such recognition itself entails that the speaker takes the referent to have an appropriate salience” (Schiffer 2005, 1141); consequently, unless the speaker takes it to be common ground between her and her hearer that x is uniquely relevantly salient in a certain way, she can’t reasonably expect her hearer to recognize that it’s x to which she is referring. Salience belongs with intention, not meaning.9 To suppose salience must be mentioned in specifying the meaning of any demonstrative or indexical is to conflate something that plays an epistemic role in identifying content with something that plays a role in constituting content (to conflate determining in the sense of ascertaining or identifying with determining in the sense of constituting).10 Our account of the meaning of ‘I’, you may recall, is that: (C) x is the referent of ‘I’ on an occasion of use o iff (1) x = the speaker; (2) the speaker refers to x with ‘I’ on o. So, according to us, the only difference between the meaning of demonstrative ‘she’ and the meaning of ‘I’ is in their respective conditions (1). Both definitions give the speaker of ‘she’ or ‘I’ freedom to refer to anything, provided it satisfies the constraint imposed by condition (1) of the pronoun’s meaning. The only difference between the two pronouns is that, whereas the meaning of ‘she’ only constrains the literal speaker11 to refer to a female, the meaning of ‘I’ constrains her to refer to herself. Because of that difference, it may seem that clause (2) in the characterization of the meaning of ‘I’ is redundant and hence that the meaning of ‘I’ should simply determine the content of ‘I’ on an occasion of use simply to be the speaker of ‘I’ on the occasion, whatever the speaker’s intention might be; there is no need for a further condition requiring that the speaker refer to herself with ‘I’. We disagree. Suppose Harold begins singing “You’re a pink toothbrush, I’m a blue toothbrush”. Only a stubborn attachment to a philosophical theory would lead a philosopher to suppose that on that occasion of utterance ‘I’ refers to Harold. Linguistic meaning is a means to an end: it’s a conventional device for performing acts of speaker meaning. For example, one primary function of language is to impart knowledge; in the normal case, when a speaker acquires knowledge from a speaker’s utterance, that knowledge derives from knowledge of what the speaker meant in producing the utterance. The importance of sentence meaning is its role in making known what the speaker meant. Turning to expressions of the form ‘that F’, we assume for present purposes that, as with ‘she’, there is a demonstrative ‘that F’ distinct from any other (e.g. bound) ‘that F’.12 We take the meaning of demonstrative ‘that female’ to be the same as the meaning of demonstrative ‘she’; that is to say: (M) x is the referent of demonstrative ‘that female’ on an occasion of use o iff (1) x is female; 443

Stephen Neale and Stephen Schiffer

(2) the speaker refers to x with ‘that female’ on o. Mutatis mutandis demonstrative ‘this female’. This treatment of demonstrative ‘she’, ‘that female’, and ‘this female’ as having the same meaning is consistent with there being other, non-​semantic conventions pertaining to their use (cf. tu and vous in French). And for the meaning of demonstrative ‘that F’ generally we propose: (N) x is the referent of demonstrative ‘that F’ on an occasion of use o iff (1) Fx; (2) the speaker refers to x with ‘that F’ on o. When we first displayed the schema (A) x is the referent of δ on an occasion of use o iff (1) Cx (2) the speaker refers to x with δ on o, we remarked that for some expressions condition Cx might be null. We take that to be the case with demonstratives ‘this’, ‘that’, and ‘it’. So for us: (O) x is the referent of demonstrative ‘that’ [‘this’, ‘it’] on occasion o iff the speaker referred to x with ‘that’ [‘this’, ‘it’] on o.

V  An Alternative Conception of the Semantics of Indexicals and Demonstratives The semantics of indexicals and demonstratives we have presented makes no use of any notion of context. In this respect it stands opposed to an alternative conception of the semantics of indexicals and demonstratives that we will now discuss. According to Stalnaker: What we want from our semantic theory is a mechanism that takes as its input a sentence with a certain meaning together with a context and delivers, as its output, a proposition—​the content that is asserted or expressed with some other force. 2014, 22; emphasis added It’s our impression that virtually everyone working on the semantics of demonstratives and indexicals will accept Stalnaker’s declaration; but there are divisions over what contexts and propositions should be taken to be. We will continue pretending that the propositions speakers assert are Russellian, so that won’t be an issue for now.There are two dominant positions about what “contexts” are.The first, associated with Kaplan’s work (1989a, 1989b), construes contexts as packages (ordered n-​tuples) of all those things (other than the meanings of expressions) that determine the contents of demonstratives and indexicals relative to occasions of their use. So, for example, according to this view, the context modeling a given occasion of use must include a speaker, a time, and a place in order to determine the contents of ‘I’,‘here’, and ‘now’ on that occasion. On such accounts, the meaning (i.e. the meaning as determined by linguistic convention) of a demonstrative or indexical δ is a function—​called a character—​that maps every context for which it is defined onto the content of δ relative to that context. For example, the character of ‘I’ would be taken to be a function that maps every context onto the speaker in that context. This might work well for ‘I’ and ‘me’, but it does not extend in any obvious way to other so-​called “pure indexicals” such as ‘here’ and ‘now’; nor does it extend in 444

How Demonstratives and Indexicals Work

any remotely plausible way to any demonstrative. As regards ‘here’ and ‘now’, the Kaplanian theory requires every context to contain a definite place and a definite time to serve as their contents relative to that context, with the condition defining the location coordinate being the same for all contexts. Consequently, the theory has no way to accommodate the fact that on different occasions in which the sentence ‘Jane’s here’ is uttered, the content of ‘here’ can vary widely—​the room the speaker is in, the building the speaker is in, the city the speaker is in, and so on (mutatis mutandis for ‘now’ and temporal locations).13 But it is the core idea that is wrong: to suppose context must be mentioned in specifying the meaning of any demonstrative or indexical is, again, to conflate something that plays an epistemic role in identifying content with something that plays a role in constituting content.14 In Kaplan’s original theory, the content of a demonstrative on a given occasion was determined by an “associated demonstration” which was “typically, though not invariably, a (visual) presentation of a local object discriminated by a pointing” (1989a, 490). The caveat is required because Kaplan acknowledges that for some utterances there is only an “implicit demonstration”, which is presumably what he would say about Evans’s example of an utterance of ‘He’s had enough’, when exactly one soldier in a company standing to attention has just collapsed, thereby rendering himself salient and so an easy referent for ‘he’ (1982, 312–​313).15 But later, Kaplan came to appreciate the hopelessness of the idea of a demonstrative having its reference fixed by an accompanying demonstration and shifted to the view that the reference of a demonstrative, relative to a context, is determined by the speaker’s “directing” intention (1989b, 582). However, he failed to realize that this in effect ruined his conception of characters as functions from contexts to contents. For the only way of keeping [contexts] along with the fact that reference is determined by the speaker’s referential intentions is to take the referential intentions as a component of the [context] for ‘she’ et al. The trouble is that then the notion of [a context] is rendered superfluous. Schiffer 2003, 112 In other words, it’s trivial to devise a formal notion of  “context” that incorporates referential intentions; the problem is that when you frame your semantics that way you’ve concocted a completely gratuitous Rube Goldberg (or Heath Robinson) device. It is rather puzzling that the theory presented in a paper called “Demonstratives” does not work for demonstratives. So, our objection to this first version of the view that “context” determines the contents of demonstratives and indexicals is that once it’s fixed to get it to work it collapses into our position. The second of the dominant conceptions of context is Stalnaker’s notion of context as common ground (see above, p. 440). According to Stalnaker, it’s common ground that determines the referent of a demonstrative or indexical. The problem with this is that the function of common ground is not and cannot be to determine the referent; its function is to provide an epistemological basis for recognizing what the speaker is referring to with a demonstrative on an occasion of use. In other words, the proper role of common ground is the one we assigned to it in the definitions we gave above. For example, suppose Sid and Ava are walking in the park and pass a man on a bench who is covered in pigeons. Mistakenly thinking that Ava was aware of the man on the bench, Sid says, ‘He must crave attention’, thereby referring to the man on the bench with ‘he’. On our account, it follows from this that the man on the bench is the referent of ‘he’ on this occasion. Evidently, Stalnaker must either deny that the referent of ‘he’ is the man on the bench or explain how the man on the bench can be the referent of ‘he’ when there is no proposition about that man in the common ground. So our objection to this second version of the view that “context” determines the contents of demonstratives and indexicals is that it mistakenly takes the role of common ground to be the constitutive one of determining the contents of demonstratives and indexicals relative to occasions of use, when in fact its role is the epistemic one of enabling speakers to intend hearers to recognize their referential (and other communicative) intentions. 445

Stephen Neale and Stephen Schiffer

VI  Dropping the Russellian Assumption The theory presented up to this point has been given relative to the simplifying idealization that the semantic content of a declarative sentence relative to an occasion of use is a Russellian proposition, and, further, that if the sentence contains a primary occurrence of an indexical or demonstrative δ, then the sentence’s content relative to an occasion of use o, if it has one, is a singular Russellian proposition containing the referent of δ relative to o. We regard the assumption that Russellian propositions are contents as false for the familiar reason that it’s possible for a rational person to believe both the proposition expressed in a literal utterance of ‘He flies’, when the speaker in uttering ‘he’ indicates the guy in the monogrammed caped spandex outfit, and the proposition expressed in a literal utterance of ‘He doesn’t fly’, when the speaker in uttering ‘he’ indicates the nerdy, bespectacled reporter named Clark Kent. But while the assumption has been a useful simplifying device, it’s time now for us to drop it and to say how that will affect what we have so far said. Happily, the view of propositions we favor leaves most of what we have said pretty much intact. We will explain this favored view, and its effect on what we have so far said, with respect to a literal and unembedded utterance of ‘She is French’ relative to an occasion o in which ‘she’ refers to Odile, and in which S is the speaker and A the hearer. Our view is that: 1 The proposition—​call it Q—​that is the semantic content of ‘She is French’ relative to o is an Odile-​and being-​French-​dependent proposition that has the same modal profile, and thus the same possible-​worlds truth conditions, as the singular Russellian proposition ⟨Odile, being French⟩. 2 As well as being partly individuated by Odile and being French, Q is also partly individuated by constraints on how one must think of Odile and of being French in order to entertain Q. Perhaps to entertain Q one must think of Odile as the only female at the time of o visible to both S and A and one must think of being French as the property expressed by the predicate ‘is French’.Yet while these constraints partially individuate Q, they are also truth-​conditionally irrelevant in that Q has the same possible-​worlds truth conditions as the Russellian proposition ⟨Odile, being French⟩. Call Q a quasi-singular proposition;16 it may be represented by the ordered pair , where w and w are the truth-conditionally irrelevant ways of thinking of Odile and being French, respectively. These truth-conditionally-irrelevant propositional constituents are also what John Perry (1986) would call unarticulated constituents of the quasi-singular propositions containing them, which is to say they are constituents of the content of the sentence ‘She is French’ relative to our imagined occasion of utterance, but they are not the contents of any of the sentence’s uttered constituents. 3 Consequently, we may continue to say that the content of ‘she’ relative to o is simply Odile and that the content of ‘is French’ is simply the property of being French. There remain, of course, difficult questions about how exactly unarticulated constituents get into the semantic contents of sentences relative to contexts. These questions are discussed in Neale (2017) and Schiffer (2017), but remain projects for another day. 4 Notice, too, that the definitions of speaker meaning and speaker reference may also stay the same, but with the proviso that, often when, for some thing x and property ψ, the speaker means an x- and ψ-dependent proposition, it will to some degree be indeterminate how she intends her hearer to think of x and ψ, and thus to some degree indeterminate what she means in uttering the sentence she uttered, and thus to some degree indeterminate what proposition is the sentence’s semantic content relative to the occasion of utterance. Yet if such indeterminacy in speaker meaning is confined to the constraints on how one must think of the objects and properties that determine a proposition’s truth conditions, then it’s unlikely even to be noticed, since all the indeterminacy will obtain among truth-conditionally equivalent propositions, and thus won’t affect the truth conditions the uttered sentence has relative to the occasion of use. 446

How Demonstratives and Indexicals Work

Notes 1 A Russellian proposition may be represented as an ordered pair of the form ⟨⟨x1, …, xn⟩, Rn⟩, where ⟨x1, …, xn⟩ is an n-​ary sequence of things of any kind and Rn is an n-​ary relation, and where ⟨⟨x1, …, xn⟩, Rn⟩ is true iff ⟨x1, …, xn⟩ instantiates Rn, false otherwise (it’s customary to drop the brackets for one-​member sequences). 2 This assumption allows us to go along with David Kaplan (1989a) as far as propositions and content are concerned. In Kaplan’s theory, the role of what we are calling meanings is played by what he calls characters. As will become clear, we see Kaplan’s take on the representation of these “characters” as quite ill-​suited to natural language semantics. 3 The theory we offer will also be presented relative to other idealizations that we won’t be dropping in this article. Perhaps the most important is one that is made in the presentation of virtually every semantic or metasemantic theory: just as Galileo ignored friction in his theories of motion, so, in presenting our theory, we ignore the vagueness of vague expressions and the vagueness of vague speech acts. Whether any semantic or metasemantic theory can be revised (as opposed to abandoned) to accommodate vagueness, is a question we will leave open here. For discussion, see Schiffer (2019a, 2019b). There are three further idealizations that we also won’t be dropping in this paper: (i) we ignore all substantive issues concerning the possibility of aphonic demonstratives and indexicals (for discussion, see Neale 2016 and Schiffer 2016); (ii) we make no attempt to show how our talk of “occasions of use” can be extended to answering machine messages and occurrences of expressions in publications; (iii) our definitions of speaker meaning and speaker reference are intended to capture paradigm cases, so we ignore attenuated cases such as those involving the cross-​examination of witnesses, giving examination answers, or shouting ‘help’ even though one doubts one will be heard. 4 Here, following Grice (1957), we use ‘utterance’ and its cognates as a technical term that applies to both linguistic and non-​linguistic items and behavior. 5 See e.g. Lewis (1969, 1975) and Schiffer (1972). Lewis defined common knowledge but had no expression for mutual belief. 6 On any particular occasion, the speaker’s intentions determine whether or not the occurrences of ‘she’ and ‘her’ in, for example,‘Ava hopes she wins’ and ‘Ava lost her dog’ are bound by ‘Ava’ (though most speakers will, at best, conceptualize the bound anaphora (an asymmetric notion) in such cases as something like sameness of reference (a symmetric notion) rather than, say, binding). Hearers’ interpretive processes are geared to making the right call, which is not to say they always do so.We do not need to speculate here about the nature of those processes or the information to which they are sensitive: we are addressing the constitutive question of what determines the referent of ‘she’ or ‘her’ on a given occasion, not the epistemic question of how hearers identify referents. 7 As Donnellan observed half a century ago, for a speaker to have “a certain complex kind of intention involving recognition on the part of his audience of his intention … may depend upon what expectations he has about his audience and their ability to grasp his intention … and the existence of an established practice may be usually required for speakers to have the right expectations” (1968, 212). It is a mystery why these points are so often overlooked—​in legal and literary theory, as well as in philosophy and linguistics. They are articulated in weaker and stronger forms by Grice (1969, 168), Grice (1971, 266), Schiffer (1972, 43 and 69), Schiffer (1995, 115), Schiffer (2003, 122), Schiffer (2005, 1141), Audi (1973), Pears (1985), Neale (1992, 552), Neale (2005, 181), and Neale (2016, 278–​281). 8 As Evans observes, “All that the conventions governing the referring expression ‘he’ insist upon, in any given context, is that the object referred to should be male” (1982, 312). And “There is no linguistic rule which determines that a ‘he’ or a ‘that man’ refers to x rather than y in the vicinity, or that it refers to someone who has just left rather than someone who has been recently mentioned” (1980, 348–​349). 9 See also Schiffer (1995, 115), Schiffer (2003, 122), and Neale (2016, 272–​273, 301–​302). 10 What goes for salience also goes, mutatis mutandis, for prominence, relevance, topic, discourse structure, and other things of that ilk. For further discussion, see Neale (2007a, 359 n7) and Neale (2016, 269–​273). 11 The “literal speaker” is one who in producing an utterance of the unembedded sentence means a proposition that “fits” the sentence’s meaning, as, for example, the proposition that Mary is famous fits the meaning of ‘She is famous’ but the proposition that the sun is setting doesn’t fit the meaning of ‘Withers once more the old blue flower of day’. 12 On bound and various other expressions of the form ‘that F’, see Neale (2007b). 13 To deny that context plays a role in natural language semantics is not to deny that a formal notion of context might play a role in a formal system designed to capture validity in formal languages containing analogues of natural language indexicals and demonstratives. 14 As Bach observes, “It is one thing for something to be determined by context in the sense of being ascertained on the basis of contextual information and quite another for it to be determined by context in the sense of being fixed by contextual factors” (2000, 271). 15 Having collapsed and having been pointed at are two among myriad ways of being salient (though the latter has a conventional aspect for many communities), as is having been mentioned previously in a conversation

447

Stephen Neale and Stephen Schiffer or utterance. This suggests to Kripke (1977) and Lewis (1979) that sometimes a seemingly anaphoric pronoun is plausibly a demonstrative pronoun that refers to something the speaker has made salient by uttering the pronoun’s seeming antecedent. Suppose Sid mistakenly thinks that the man talking in French to Odile is Odile’s husband. He says to Ava, ‘Oh, Odile’s husband is French’, and Ava replies, ‘He isn’t her husband, but he is French’. If Kripke and Lewis are right, then we say ‘he’ and ‘her’ have on this occasion the meanings our theory assigns to demonstrative ‘he’ and demonstrative ‘her’. A demonstrative ‘him’ seems to be required to capture the anadeictic reading of ‘Odile got married; she met him online’ as there is no antecedent noun phrase to serve as the antecedent of an anaphoric ‘him’. 16 This is what they were called when introduced by Schiffer (1978) and used by Récanati (1993).

References Audi, R. 1973. Intending. Journal of Philosophy 70: 387–​403. Bach, K. 2000. Quantification, Qualification and Context: A Reply to Stanley and Szabó. Mind and Language 15, 2 & 3: 262–​283 Donnellan, K. 1968. Putting Humpty Dumpty Together Again. Philosophical Review 77: 203–​215. Evans, G. 1980. Pronouns. Linguistic Inquiry 11: 337–​362. Evans, G. 1982. The Varieties of Reference. Oxford: Clarendon Press. Grice, P. 1957. Meaning. Philosophical Review 66: 377–​388. Grice, P. 1969. Utterer’s Meaning and Intentions. Philosophical Review 78: 147–​177. Grice, P. 1971. Intention and Uncertainty. Proceedings of the British Academy 57: 263–​279. Kaplan, D. 1989a. Demonstratives. In Themes from Kaplan, ed. J. Almog, J. Perry, and H. Wettstein, New York: Oxford University Press, pp. 221–​243 Kaplan, D. 1989b.Afterthoughts. In Themes from Kaplan, ed. J.Almog, J. Perry, and H.Wettstein, New York: Oxford University Press, pp. 481–​563. Kripke, S. 1977. Speaker Reference and Semantic Reference. Midwest Studies in Philosophy 2, 1: 255–​276. Lewis, D. 1969. Convention: A Philosophical Study, Cambridge, MA: Harvard University Press. Lewis, D. 1975. Languages and Language. In Minnesota Studies in the Philosophy of Science, Volume VII, ed. Keith Gunderson, Minneapolis, MN: University of Minnesota Press, pp. 3–​35. Lewis, D. 1979. Scorekeeping in a Language Game. Journal of Philosophical Logic 8, 1: 339–​359. Neale, S. 1992. Paul Grice and the Philosophy of Language. Linguistics and Philosophy 15: 509–​559. Neale, S. 2005. Pragmatism and Binding. In Semantics Versus Pragmatics, ed. Z. Szabo, Oxford: Oxford University Press, pp. 165–​285. Neale, S. 2007a. On Location. In Situating Semantics: Essays on the Philosophy of John Perry, ed. M. O’Rourke and C. Washington. Cambridge, MA: MIT Press, pp. 251–​395. Neale, S. 2007b. Heavy Hands, Magic, and Scene-​reading Traps. European Journal of Analytic Philosophy 3, 2: 77–​132. Neale, S. 2016. Silent Reference. In Meanings and Other Things: Essays in Honor of Stephen Schiffer, ed. G. Ostertag, Oxford: Oxford University Press, pp. 229–​344. Pears, D. 1985. Intention and Belief. In Essays on Davidson: Actions and Events, ed. B.Vermazen and M. Hintikka. Oxford: Clarendon Press, 75–​88. Perry, J. 1986. Thought without Representation. Proceedings of the Aristotelian Society, Supplementary Volume, 60: 137–151. Récanati, F. 1993. Direct Reference: From Language to Thought. Blackwell. Schiffer, S. 1972. Meaning. Oxford: Clarendon Press. Schiffer, S. 1978. The Basis of Reference. Erkenntnis 13: 171–206. Schiffer, S. 1995. Descriptions, Indexicals, and Belief Reports: Some Dilemmas (But Not the Ones You Expect). Mind 104: 107–​131. Schiffer, S. 2003. The Things We Mean. Oxford: Oxford University Press. Schiffer, S. 2005. Russell’s Theory of Descriptions. Mind 114: 1135–​1183. Schiffer, S. 2016. Reply to Avramides and Neale. In Meanings and Other Things: Essays in Honor of Stephen Schiffer, ed. G. Ostertag, Oxford: Oxford University Press, pp. 494–​512. Schiffer, S. 2019a. Vague Speaker Meaning. In Further Advances in Pragmatics and Philosophy, Part II, ed. A. Capone , M. Carapezza, and F. Lo Piparo. Cham: Springer, 3–​23. Schiffer, S. 2019b. Expression Meaning and Vagueness. In Sensations, Thoughts, Language: Essays in Honor of Brian Loar, ed. A. Sullivan, London: Routledge, pp. 78–​112. Stalnaker, R. 2002. Common Ground. Linguistics and Philosophy 25: 701–​721. Stalnaker, R. 2014. Three Notions of Context. In Context. Oxford: Oxford University Press, pp. 13–​34.

448

34 DEMONSTRATIVE REFERENCE TO THE UNREAL The Case of Hallucinations Marga Reimer

Is this a dagger which I see before me, The handle toward my hand? Come, let me clutch thee. I have thee not, and yet I see thee still. Art though not, fatal vision, sensible To feeling as to sight? Or art though but A dagger of the mind, a false creation, Proceeding from the heat oppressed brain?

1  Macbeth’s Dagger In the well-​known passage that opens the present paper, Shakespeare’s Macbeth hallucinates a dagger about which he goes on to soliloquize, asking himself whether what he is seeing is a real dagger or merely a “dagger of the mind”. Macbeth’s hallucination would appear to be one of delirium, “proceeding”, as it does, from his “heat oppressed brain”. There are, of course, other kinds of hallucinations as common (or very nearly as common) as hallucinations of delirium. There are, for example, the auditory hallucinations of paranoid schizophrenia, the visual hallucinations of Charles Bonnet Syndrome, the multi-​sensory hallucinations of drug or alcohol intoxication, and the benign hallucinations of hypnagogia. Because of the sheer pervasiveness of hallucinations, both pathological and benign, demonstrative reference to such phenomena is no doubt quite common; indeed, it is likely as common, or very nearly as common, as the hallucinations themselves. After all, who experiencing such wondrous (if sometimes frightening) phenomena would not refer to them, whether in talking with others or in talking (or thinking) to oneself, as in the case of Macbeth? “What’s that?” or “That’s amazing!” are natural responses to hallucinatory phenomena, responses that exploit, and make essential use of, devices of demonstrative reference. The questions to be addressed in the present paper concern demonstrative reference to hallucinations. How is such reference achieved? What (in other words) are the “mechanisms” behind it? And (no less importantly) how does the phenomenon of demonstrative reference to hallucinations inform our understanding of related issues, whether in the philosophy of language or in the philosophy of perception? Despite the pervasiveness of hallucination, the phenomenon of demonstrative reference to such phenomena has received scant attention in the philosophy of language literature.1 This lack of 449

Marga Reimer

attention is unfortunate but hardly surprising as demonstrative reference to hallucinations is paradigmatically non-​paradigmatic. There is nothing (no thing) “out there” in public space that is being gestured at (“clutched”, perhaps, as with Macbeth) while simultaneously being referred to. There is also, as with Macbeth’s soliloquy, no audience apart from the speaker themselves. Such soliloquies are therefore doubly non-​paradigmatic. It might thus be assumed, naturally if erroneously, that there is nothing to be learned about the nature of demonstrative reference from such patently non-​standard cases. Yet there are good reasons for thinking otherwise. First, demonstrative reference to hallucinations is likely pervasive, however non-​standard. It should therefore be attended to by theorists of reference. Indeed, reference to hallucinations might provide the theorist with helpful test cases for a more general theory of demonstrative reference. Can their favored theory accommodate such cases? Second, exploration of demonstrative reference to hallucinatory phenomena might result in a more comprehensive understanding of, inter alia: (i) the competing roles of ostensive gestures and speaker intentions in the determination of such reference, (ii) the importance, if any, of the speaker’s knowledge of the general kind of thing to which they are referring, and (iii) the variety of other unrealities, such as numbers and fictional characters, amenable to (deferred2) demonstrative reference. On behalf of theoretical attention to such non-​standard cases, it is instructive to recall Frege (1892) and his focus on identity statements between co-​referring singular terms—​or Russell (1905), and his focus on non-​denoting descriptions. These atypical cases provided the impetus for truly revolutionary thinking about both proper names and definite descriptions. There is thus much to be said for attending to paradigmatically non-​paradigmatic cases—​in the philosophy of language no less than in other disciplines, including (no doubt) the natural sciences and disciplines involving a blend of natural science and philosophy. Cognitive neuropsychiatry is a case in point insofar as its advocates seek to understand the nature of the “normal” (human) mind by reflecting on “abnormal” cases (as in mental disorder), cases where something has gone awry with the mind. Here, it is the psychiatric case that is patently non-​standard—​and yet potentially enormously illuminating with regard to the more “standard” cases (of thought, belief, perception, and affect). My general intentions in what follows are three-​fold: to sketch an account of demonstrative reference to hallucinations, to situate that account within a more general theory of demonstrative reference, and to then connect it to other issues in contemporary analytic philosophy—​both philosophy of language and philosophy of perception. Issues to be examined include: (i) the grounding of names in their objects, (ii) the puzzle of true negative existentials, and (iii) the argument from hallucination against naïve realism (and in favor of sense data). The format of this paper is as follows. I begin by introducing some terminology after which I discuss the phenomenon of hypnagogia and benign hallucinations more generally. I then present a puzzle raised by demonstrative reference to such hallucinations after which I  adumbrate an account (a Kripkean “picture” vs. full-​fledged theory) of such reference (Kripke 1980). I go on to consider two possible solutions to the aforementioned puzzle, one of which I reject and one of which I tentatively endorse. This is followed by a discussion of the proposed account’s relevance to a more general theory of demonstrative reference. In the paper’s penultimate section, I discuss the account’s relevance to other issues in the philosophy of language and the philosophy of perception. Finally, I conclude with some methodological remarks concerning the importance of “odd ball” (paradigmatically non-​paradigmatic) cases to the “fine-​tuning” of theories, with an emphasis on theories in the philosophy of language. Let us begin, then, by introducing some terminology with the aim of providing the linguistic and conceptual resources necessary for the articulation, motivation, and defense of the proposed account of demonstrative reference to hallucinations.

2  Terms and Terminology There are three notions particularly in need of characterizing; these are the very notions invoked in this paper’s title: demonstrative reference, the unreal, and hallucinations. First, however, it is important to 450

Demonstrative Reference to the Unreal

distinguish between terms or (in other words) expressions of ordinary language and terminology or (in other words) expressions of theory. I will then attempt to make use of terms from ordinary language in devising terminology appropriate to the theoretical tasks at hand. This is a familiar strategy, not only in the sciences (think, for example, of the theoretical notions of life, charge, force), but also in philosophy, particularly analytic philosophy. Indeed, the literature abounds with examples from the philosophy of language (meaning, reference, truth), from the philosophy of mind (mind, thought, consciousness), from epistemology (knowledge, evidence, belief), from metaphysics (existence, causality, space and time), and even from logic (validity, entailment, soundness). What we have in all these cases are terms from ordinary language that have been adopted and precisified by theorists and thereby transformed into theoretical terms whose meanings are anchored in those of their ordinary language counterparts. I like this way of doing things because it grounds theoretical terms in real and familiar phenomena. It is thus intuitive, enhancing a theory’s comprehensibility to the extent that its terminology draws on what is already familiar to the theorist’s audience. Let us begin, then, with the all-​important notion of demonstrative reference.

(i)   Demonstrative Reference Let us analyze this notion by considering its individual constituents, beginning with that of reference. The ordinary everyday notion of reference (adopted herein) involves the idea of a speaker’s drawing the audience’s attention to someone or something, typically with the intention of saying something about that person or thing. Reference is thus something that a speaker does and that her words do, if at all, only derivatively. It is thus the speaker who refers to her cat by way of the name “Mieze”; the name itself refers to the cat only because (and only insofar as) the speaker is using that name to refer to her cat. Demonstrative, as used in the present paper, is a technical notion from contemporary philosophy of language, originating in the works of David Kaplan (1989). For Kaplan, demonstratives represent a special case of indexical (or context-​sensitive) expressions. What distinguishes demonstratives from other indexical expressions is the requirement of some sort of accompanying “demonstration” (paradigmatically, a pointing gesture). Demonstrations, in this technical sense (which need not involve any overt action on the speaker’s part), enable the audience to identify to whom or to what the speaker is referring and about which she is speaking. The Kaplanian notion of demonstrative reference would thus appear to be grounded in notions captured in terms from ordinary language: “reference” and “demonstrative”. In this way, Kaplan’s notion of demonstrative reference (adopted herein) is an eminently accessible one. So much for the notion of demonstrative reference, what now of the phenomenon itself? What does demonstrative reference involve when it occurs in ordinary everyday conversations (or soliloquies)? Paradigm cases of demonstrative reference involve visually perceived objects or individuals; such reference (as noted above) is often accompanied by some sort of “demonstration” on the part of the speaker, such as a pointing gesture. Demonstrative reference can also, of course, involve auditory (or other sensory) phenomena, although in such cases demonstration would not typically involve any sort of overt ostension. Thus, one might utter (either to oneself or to another) “That was loud”, immediately after a nearby explosion has “presented” itself though its “sound effects”. In “presenting” itself in this manner, the explosion makes itself available as a potential item of demonstrative reference. No additional action on the part of the speaker would be required or even in order.

(ii)   The Unreal The unreal, as I use that notion, incorporates (and is perhaps exhausted by) things that do not “really” exist but that can nevertheless be thought about and spoken of with ease. So conceived, unreal things are not unlike the unreal objects of Alexius Meinong (1904).Yet the unreal (as used herein) is a broader notion than Meinong’s insofar as it incorporates certain mental phenomena, such as hallucinations or 451

Marga Reimer

(perhaps more accurately3) their “objects”. The unreal is a notion whose meaning is naturally and helpfully elucidated by way of examples, which are remarkably easy to come by. There are (for instance) hallucinatory objects, fictional characters, mythical creatures and legendary heroes, the imaginary friends of childhood, and the frightening incubi and succubae of sleep paralysis. What of expressions like “the round square cupola on Berkeley College”—​to use a famous example of Russell’s? That is perhaps a question to be settled by stipulation (pace Meinong and Russell). Indeed, I suspect that the number of theorists who believe that thought and talk about such things is possible, is matched by the number of theorists who think that it is not. At any rate, there is no need to attempt to settle the issue here. As used here, the expression “the unreal” (or “unreality”) is likely a Wittgensteinian “family resemblance” term (Wittgenstein 1953) and so it would be pointless to attempt a characterization of it in the philosopher’s language of “necessary and sufficient conditions”.The admittedly and, indeed, deliberately vague notion of things that do not really exist, coupled with a denotative characterization of the sort given above, provides sufficient clarity for present purposes insofar as it successfully captures uncontroversial cases of the phenomenon in question, demonstrative reference to which I am concerned to explore.4

(iii)   Hallucinations Hallucinations are often characterized as “false perceptions”: perceptions of things that do not “really” exist, although they are generally (though not invariably) believed to exist by the percipient. Importantly, while the “objects” of hallucination do not “really” exist insofar as they are mind-​dependent, the hallucinations themselves do exist. Thus, while the dagger hallucinated by Macbeth does not really exist, Macbeth’s dagger-​hallucination certainly does. I prefer to think of hallucinations, not as “false perceptions”, but as “pseudo-​perceptions” or, alternatively, as “perceptual conceptions”. After all, they do not involve genuine or (in other words) sensory perception: there is no external object detected or identified by any of the five senses: vision, hearing, touch, etc. Hallucinations stereotypically involve a kind of internal image or sound, a phenomenon that is not located outside the percipient’s body—​or, speaking metaphorically,“outside” the percipient’s mind. Perceptions can occur in any one (or more) of five or more modalities: visual, auditory, olfactory, tactile, or gustatory. A possible sixth modality involves “felt” presences: presences that are felt but not in the ordinary tactile sense of that notion. They occur, paradigmatically, when someone “feels” as though they are being watched or that someone or something (unseen and unheard) is in the room with them. “False perceptions” is arguably a better characterization of illusions—​such as optical illusions, including (for example) the oases “seen” by those dying of thirst in the desert. Something real is indeed seen but not (alas) a real oasis. Such illusions are “false” insofar as they involve misleading or otherwise inaccurate perceptions of real phenomena. Illusions are thus importantly different from hallucinations in two overlapping respects: they involve genuine perceptions of real things—​rather than pseudo perceptions of unreal things. Below, I focus on the benign visual hallucinations of hypnagogia. I have chosen to do so for three reasons. First, such hallucinations are quite common in the general population, especially among young children. Second, in contrast to the visual hallucinations of (e.g.) Charles Bonnet Syndrome, those of hypnagogia cannot be dismissed on the grounds of a pathological etiology. Third, in contrast to auditory hallucinations (whether benign or pathological), the “objects” of visual hypnagogia can (seemingly) be pointed to or otherwise gestured at. This fact enables fruitful discussion of the role of Kaplanian “demonstrations” in demonstrative reference to hallucinations. As we will see, such discussion has implications for the role of such demonstrations in demonstrative reference more generally.

3  Hypnagogia Hypnagogia is a conscious state occurring at the “borderlands” of wakefulness and sleep: either just before the onset of sleep or, somewhat less frequently, just before waking. Such states are (medically) benign if occasionally frightening; they are generally visual or auditory. Common 452

Demonstrative Reference to the Unreal

hypnagogic hallucinations involve “seeing” figures or landscapes just before the onset of sleep, or “hearing” one’s name called out just before (or upon) waking. Famous (if atypical) examples would include the angels a young William Blake “saw” in a tree and the “ring” of benzene “seen” by August Kekule in a dream or dream-​like state. Such phenomena would appear to involve what philosopher of psychiatry Mike Jackson (2007) calls “benign psychosis”. I do not like the term “psychosis” used in connection with hypnagogia on account of its association with mental illness. There is absolutely nothing psychiatric or otherwise pathological about hypnagogia, which has been the source of genuinely original thinking—​as with Kekule’s famous hypnagogic discovery of benzene’s “ring”. I will thus construe the phenomenon for what it appears to be: a kind of hallucinatory experience, common, medically benign, generally visual and occasionally auditory, at times genuinely creative, and typically occurring just before the onset of sleep or just upon wakening. Despite my focus on specifically visual hypnagogia, I attempt to generalize what is said about demonstrative reference to such phenomena to hallucinations more generally—​including those that are pathological and/​or auditory (or otherwise non-​visual) in nature. This, in turn, facilitates the incorporation of the emerging view of demonstrative reference within an account of demonstrative reference more generally—​which, of course, paradigmatically involves reference to real phenomena really perceived and really demonstrated by the speaker. Below, I attempt to motivate an account of demonstrative reference to hallucinations by considering an obvious question and an equally obvious, if problematic, response to that question.

4  An Obvious Question and an Equally Obvious (if Problematic) Response The obvious question posed by demonstrative reference to hallucinations is:  What exactly is being demonstratively referred to in such cases? The equally obvious answer is: The hallucination, of course! But this response is problematic, its “obviousness” notwithstanding. After all, hallucinations are unambiguously mental phenomena and so they cannot possibly be demonstrated in the usual way—​through some sort of ostensive gesture.5 How, then, are such things demonstrated by the speaker, thereby rendering them suitable objects for demonstrative reference? Perhaps such hallucinations admit of a kind of inward and internal demonstration, as when the philosophically inclined speaker, cognizant of the fact that they are hallucinating, says to themselves, while attending to their hallucination, “This is all in my head”. Their focused inward attention to the “object” of their hallucination arguably amounts to a kind of Kaplanian demonstration, thereby rendering the pseudo-​object amenable to demonstrative reference. Yet such cases are, of course, extraordinarily unusual. They involve philosophers practicing their craft by demonstratively referring to their own conscious experiences, while simultaneously reflecting on those experiences. In the far more typical cases considered herein, the speaker takes themselves to be demonstrating a very real and publicly accessible entity, thereby rendering it amenable to (successful) demonstrative reference. What are we to say about a case where, for example, the young William Blake, while pointing in the direction of a silver birch, excitedly declares, “That is a real angel”? One possible response, developed below, is that Blake’s utterance involves a kind of deferred demonstrative reference. In demonstrating the apparent location of the “object”, of the angel-​hallucination, Blake successfully refers to that object—​a mental phenomenon—​and says something false about it. After all, if angels (vs. angel-​hallucinations) really do exist, they are not mind-​dependent creatures; they are (presumably) heavenly and thus mind-​independent creatures. This sort of account has some obvious virtues insofar as it captures some pretty strong (pre-​theoretical) intuitions, including the intuition that Blake demonstrated the apparent (to him) location of the object of his hallucination, the intuition that he thereby demonstratively referred to that same mental entity, and the intuition that he said something false about it: that it was an angel, a real angel. This sort of account has the additional virtue of drawing on a well-​known and quite common phenomenon: deferred demonstrative reference. Thus, if I show a friend an old photograph of my father’s 453

Marga Reimer

family and, pointing at the image of my father, say “This is my dad”, I have surely referred to my father by way of the demonstrative pronoun “this” and have done so by demonstrating a portion of an old photograph. Just as my father is connected to the demonstrated portion of an old photograph at which I am pointing, Blake’s angel-​hallucination is connected to a particular silver birch, in whose direction he is pointing. In both cases, something is demonstratively referred to by way of the demonstration of something very different from—​but nonetheless clearly connected to—​that very thing. Yet despite these clear similarities, there is surely a difference, of potential theoretical significance, between deferred demonstrative reference to the “objects” of hallucination and deferred demonstrative reference to things that really do exist. This concern is well-​taken and is accordingly kept in mind when attempting to motivate the proposed view of (deferred) demonstrative reference to hallucinations. Such motivation involves attention to cases that are more clearly analogous to those involving deferred demonstrative reference to the objects of hallucination. It is to such cases that I now turn.

5  Numbers, Fictional Characters, and Auditory Hallucinations The analogous cases that I am interested in can be divided into two groups: those involving deferred demonstrative reference to abstract entities and those involving deferred demonstrative reference to mental phenomena. Both sorts of case are amenable to Kaplanian demonstration, initial appearances to the contrary. The first group includes demonstrative reference to numbers and fictional characters and the second group includes demonstrative reference to the benign auditory hallucinations of hypnagogia. Let us consider these two groups in turn, beginning with the first.

(i)   Numbers and Fictional Characters In cases of visual hypnagogia, the speaker is demonstratively referring (albeit unwittingly) to a mental phenomenon, not an abstract one. But demonstrative reference to abstracta, deferred demonstrative reference to such “entities”, is not only possible but in fact quite common. Because deferred demonstrative reference to hallucinations turns out to have much in common with deferred demonstrative reference to abstracta, by making a case for the latter, I hope to strengthen my case (however indirectly) for the former. Let us begin with deferred demonstrative reference to numbers. Such reference arguably (if not obviously) occurs when one refers to a number by demonstrating a token of the numeral of which the number in question is the denotation. Thus, one might declare, while demonstrating tokens of numerals on a chalk board, “That number is even and that number is odd”. Interestingly, such cases would appear to mirror those involving hallucinations insofar as they give rise to certain types of ontological confusions. Just as persons might confuse hallucinatory phenomena with their “real” counterparts, persons might confuse numerals with numbers.Yet in neither case does such confusion preclude successful (if deferred) demonstrative reference. Deferred demonstrative reference to abstracta also arguably occurs when (for example) one refers to a fictional character by demonstrating the actor portraying that character in the performance of a play. Thus, imagine that someone says, while pointing in the direction of the actor playing the part of King Lear’s Goneril, “That woman is evil”. Surely, it is plausible to suppose that they have thereby referred to the Shakespearean character and that they have done so by demonstrating the actor portraying that character. Thus, while abstracta themselves cannot, of course, be demonstratively referred in the usual “direct” way, they can arguably be referred to through a kind of deferred demonstrative reference. To make the point vivid, just imagine that the actor demonstrated is not playing Goneril but rather Cordelia. Imagine, further, that the actor playing Cordelia is in fact evil. That would not affect the truth value of the speaker’s utterance of “That woman is evil” which, in this particular scenario, is unequivocally false, as Cordelia is one of the play’s more sympathetic characters. 454

Demonstrative Reference to the Unreal

There are, of course, significant differences between deferred demonstrative reference to abstracta and deferred demonstrative reference to mental phenomena, like Blake’s angel in the tree. There is, after all, a difference in kind between the types of objects demonstratively referred to. Fictional characters are abstract entities and hallucinations are mental phenomena.6 However, there are no such differences between deferred demonstrative reference to visual hallucinations and deferred demonstrative reference to auditory hallucinations. Moreover, even in the latter case one can speak, just as in the former case, of demonstrating something physical as a means of demonstratively referring to something non-​physical. This brings us to the phenomenon of auditory hypnagogia.

(ii)   Auditory Hypnagogia Although auditory hypnagogia is as varied and fascinating as its visual counterpart, here I focus on a rather humdrum case involving something that many people have experienced. One is asleep in bed when, all of a sudden, one sits bolt upright and, convinced that someone has just called out their name, says to themselves (something like) “Who was that?” or (perhaps) “That sounded just like my mom calling my name”. In such cases, one demonstratively refers to an “object” of auditory hypnagogia. A familiar “voice” is referred to deferentially by demonstrating its apparent (to the speaker) location. The demonstration is effected by means of one’s (whole) body. This sort of phenomenon arguably occurs in non-​ hallucinatory cases as well as when one says, after hearing a loud explosion, “That was a loud noise”, effectively using one’s whole body to demonstrate the “loud noise” occurring (literally) “within earshot”. Such cases are incredibly common. My point (aside from that one) is that they arguably involve deferred demonstrative reference to the objects of auditory hypnagogia, where such reference is facilitated by Kaplanian demonstrations, demonstrations of the apparent location of the voice-​hallucination. Such demonstrations (as noted above) are not achieved by any sort of ostensive gesture, but instead by the speaker’s whole body, metaphorically a kind of pointing finger. These demonstrations can arguably be seen as pragmatically equivalent to more ordinary ostensive gestures used to demonstrate the approximate location of some real (non-​hallucinatory) auditory phenomenon.

6  A Couple of Semantically Irrelevant Distinctions Before exploring some of the connections between demonstrative reference to hallucinations and other philosophical issues, I  would like to examine a couple of distinctions that are arguably not semantically significant, initial appearances to the contrary. The intended effect of this examination will be to emphasize the uniformity of the proposed picture, thereby lending it some additional credence. The distinctions in question, broadly construed, concern (i) conversation vs. soliloquy and (ii) ignorance vs. knowledge. Let us consider these two distinctions in turn, beginning with the former.

(i)   Conversation vs. Soliloquy In paradigm cases of demonstrative reference, regardless of whether the intended object of reference is real or hallucinatory, the speaker is speaking to someone other than themselves: they are not (in other words) engaging in soliloquy but in conversation. In cases where the speaker is attempting, albeit unwittingly, to direct the attention of the audience to some “private” mental object, it might appear as though they are not referring to anything at all.Their referential intentions notwithstanding, reference has failed. Or so it might appear. This perspective is captured in the audience’s likely response to the speaker’s utterance: But there’s nothing there! Or (perhaps):  What are you taking about? That’s not an angel; it’s just a tree! Suppose, 455

Marga Reimer

however, that the audience were to discover that the speaker was hallucinating and that the object of their hallucination was, in fact, an angel. In that case, the audience’s natural response to the speaker’s utterance might be (something like), “That’s not a real angel you are referring to; it’s just a hallucination”. Or perhaps, “There’s no angel in that tree; it’s just in your head”. So which of these two sets of responses should be given credence? I would suggest the latter set be given credence as it reflects greater knowledge of the relevant facts—​in particular, of the fact that the angel “seen” by the speaker is a mental phenomenon, something that is, indeed, “all in their head”. What now of cases where the speaker is talking to themselves and thereby functioning as their own audience? Here, it would be natural to say that they have unwittingly referred to a hallucination and that they have done so “deferentially”: that is, by demonstrating the location where the mentally projected “object” of hallucination (an angel, perhaps) appears, to the speaker, to be. The proposed picture thus gains some additional credibility insofar as it is simpler than a picture on which the speaker says different things depending on whether she is engaging in conversation or soliloquy. In either case, the speaker is referring, demonstratively and deferentially, to a private mental “object”. Indeed, the proposed account is not only simpler but considerably more intuitive insofar as what a speaker actually says (vs. communicates) wouldn’t seem to depend on whether they are talking to themselves or talking to another person—​at least not generally. What are we to say about the semantic content and truth value of the speaker’s utterance in such cases, whether involving conversation or soliloquy? The utterance’s semantic content would presumably involve the hallucination itself (as logical subject) as well as the attributed property (as logical predicate). Because the attributed property (angelhood) is not true of the angel-​hallucination (a mind-​ dependent phenomenon), the utterance would emerge as false. Because this would appear to be the case regardless of whether the speaker is engaged in conversation or soliloquy, that distinction emerges as semantically insignificant. Such insignificance amounts to a kind of theoretical uniformity not present in accounts on which the distinction in question is of semantic significance.

(ii)   Ignorance vs. Knowledge In paradigm cases of demonstrative reference to hallucinations, the speaker is ignorant of the fact that they are hallucinating; they are unaware (for example) that there is no angel “out there” to which they are attempting to demonstratively refer.7 But now suppose that they do know that they are hallucinating: they know, for example, they are experiencing a hypnagogic hallucination, the object of which is an angel. What should we say about cases of this sort? Surely, the natural thing to say is that the speaker is referring deferentially (and knowingly) to a hallucination by demonstrating the apparent location of its object—​where it is projected by their mind/​brain. To appreciate the plausibility of this approach, let us alter the speaker’s utterance to “That’s not a real angel; it’s just a hallucination of an angel”. In such a case, the speaker’s utterance would clearly be true. The speaker would have referred, deferentially and demonstratively, to a mental phenomenon; they would, moreover, have said something true (perhaps even necessarily true8) of that phenomenon. Regardless of whether or not they know they are hallucinating, the “angel” the speaker “sees” is the intentional object of their angel-​hallucination. It is the “object” about which they are thinking and speaking. Indeed, even knowing (believing, suspecting) that they are hallucinating, the speaker might nevertheless point to, or otherwise gesture in the direction of, the apparent location of the “angel”. This would be of a piece with the soliloquizing Macbeth, whose unsuccessful attempt to grasp the handle of the “dagger” he “sees” before him only confirms his suspicion that it is but a “dagger of the mind”—​a “false creation proceeding from the heat oppressed brain”. The point that emerges from reflection on the foregoing scenarios is that speakers may demonstratively and deferentially refer to the intentional objects of their hallucinations regardless of whether or not they know (believe, suspect) that they are hallucinating. In this way, the proposed picture of deferred demonstrative reference can be seen to exhibit a kind of uniformity; the speaker refers to the 456

Demonstrative Reference to the Unreal

object of their hallucination when they assume that object to be real—​and when they believe or suspect that it is not.

7  A Provisional Picture Extended In this section, I attempt to extend the proposed picture so as to incorporate the auditory hallucinations of paranoid schizophrenia.This expanded picture is then discussed in terms of the twin (and arguably competing) roles played by private communicative intentions and public context. Suppose that the speaker, suffering from the auditory hallucinations characteristic of paranoid schizophrenia, utters, with absolute conviction, “That’s an invisible spy”.What (if anything) have they referred to and what (if anything) have they said? If we attempt to extend the proposed picture to this sort of case, we might say that the speaker has demonstrated (in Kaplan’s sense) the general area of the hallucinatory spy’s mentally projected “voice”; they have done so by way of their own body, which is itself situated in the general location of the spy-​hallucination and their “voice”. In doing so, the speaker has deferentially and demonstratively referred to the auditory “object” of their hallucinatory experience. We might also say that the speaker has said falsely of the successfully referred to auditory spy-​hallucination that it is a spy: a real, if invisible, spy.The fact that it is so easy to extend the proposed picture in this way, lends that picture credibility—​considerable credibility insofar as the results of that extension are intuitively plausible. What do these brief observations suggest about the twin roles of “private” communicative intentions and “public” demonstrations with respect to demonstrative reference? They suggest a couple of things. First, they suggest that the object (if any) to which the speaker’s demonstration directs the audience’s attention does not always emerge as the semantic referent of the demonstrative expression used. If it did, the patent falsity of the speaker’s utterance (“That is an invisible spy”) would be due to the fact that such things as general areas do not have properties like invisibility or spyhood. But that just seems wrong; the falsity of that utterance appears to be due to the fact that the intentional objects of auditory hallucination are unreal and so do not (and, indeed, cannot) have the particular properties attributed to them: properties attributable only to what are real. Unreal intentional objects are the wrong kind of object for the attributed properties. In this way, the cases in question undermine the idea that demonstrations have semantic significance. For again, if they did, the falsity of our paranoid speaker’s utterance would be grounded in the fact that general locations do not have properties like invisibility or spyhood. Second, the foregoing observations suggest that speaker intentions, when appropriately disambiguated, do determine the actual (semantic) referent of the demonstrative expression. The speaker thus refers successfully to a hallucinatory “object”, though they mistake that object for a physical rather than mental object: they mistake it (in other words) for a real rather than unreal thing. They achieve such reference indirectly by demonstrating (with the appropriate sort of communicative intentions) the general location of the hallucination’s mentally projected “object” and they do so by way of their own (whole) body. The observations and reflections of this section lead naturally to the issue of their relevance to a more general theory of demonstrative reference—​a theory that incorporates demonstrative reference to the “real” as well as to the “unreal”. Let us now turn to this issue.

8  Situation within a General Theory of Demonstrative Reference In cases of demonstrative reference to hallucinations, we often have something amenable to demonstration—​just not the object of hallucination. After all, such unrealities cannot actually be demonstrated or otherwise ostended, as they have no location in space. But (as suggested above) such unrealities can be referred to deferentially: that is, by demonstrating or otherwise ostending such “real” 457

Marga Reimer

spatial phenomena as numerals, actors, or general areas (such as the apparent locations of hallucinatory angels or spies). Let us now consider how these phenomena are related to more ordinary cases of demonstrative reference, cases involving “real” objects demonstratively referred to in the usual ways. How do we connect the proposed picture of demonstrative reference to hallucinatory objects or individuals, to demonstrative reference to real objects or individuals? There are any number of ways of addressing this issue. I will consider it from the perspective of the different, and potentially competing, roles played by speaker intentions (private or “internal” factors) and contextual features (public or “external” factors). Let us begin with the latter and then turn to the former. I suspect (as suggested above) that demonstrations are pragmatically but not semantically relevant. That is, they provide, and are intended to provide, the audience with hints as to the speaker’s referential intentions. Thus, when the hallucinating speaker says, while pointing squarely in the direction of a silver birch, “That’s a real angel” they do not in fact demonstratively refer to the tree in question (or even to a portion thereof) and say falsely of it that it is an angel.The reason for this is simple: they have no intention of referring to any such tree even if they respond affirmatively to the question, “Are you talking about what you are actually pointing to?”; their affirmative response would be based on false assumptions including, and in particular, the assumption that they are ostending a real angel (situated in the tree in question), one that is as perceptible to the audience as to the speaker herself. Once informed of the relevant facts, the speaker might well say (something like), “I wasn’t actually pointing to the tree; I was pointing in its direction because that’s where I saw, or at least I thought I saw, a real angel—​an angel that turned out to be all in my head”. These ideas would appear to have important implications for the truth value of what the speaker says. Suppose, again, that the speaker says, while pointing in the direction of a particular silver birch, “That’s a real angel”. When the hearer responds with, e.g., “That’s false; that is not an angel but a tree”, they are partly but not entirely mistaken. Although the speaker has indeed said something false, they have not said something false about a tree; they have said something false about an angel-​hallucination. The same considerations apply, mutatis mutandis, to cases of demonstrative reference to auditory hallucinations. Thus, suppose that the speaker, suffering from the auditory hallucinations of paranoid schizophrenia, says, “That’s an invisible spy”. If the audience responds with, e.g., “That’s false; that’s not an invisible spy; it’s just the wind blowing”, they are partly but not entirely mistaken. Although the speaker has indeed said something false, they have not said something false about the wind; they have instead said something false about an auditory hallucination—​they have said that it is the “real” voice of a “real”, if invisible, spy.

9  Related Philosophical Problems There are no doubt a variety of distinctively philosophical problems related, in some way or another, to the problem of demonstrative reference to hallucinations. I  will consider three such problems, two from the philosophy of language and one from the philosophy of perception. Let us consider these problems in turn, while simultaneously making note of any connections that might emerge between them.

(i)   The Qua-​Problem The basic issue here concerns the reference-​fixing (or “grounding”) of names in their bearers. The central point (made by Devitt and Sterelny, 1999) is that the grounding of such terms fails in cases where the speaker is radically mistaken about the nature (the kind) of the entity in whom or in which they are attempting to ground the name. Suppose, for example, I  hear a dog barking whenever 458

Demonstrative Reference to the Unreal

I  approach my neighbor’s door. His bark reminds me of the bark of my childhood pet Witz and so I decide to call him “Witz”, saying such things as “Witz is a great watchdog” or “I bet Witz is a schnauzer”. When I ask my neighbors about the breed of their dog, they respond with, “We don’t own a dog; you must be hearing the barking alarm that goes off whenever anyone approaches our front door”. I say to myself, “I can’t believe it; Witz is just an alarm”. Despite my new knowledge, I might continue to say such things as “Witz is at it again; someone must be at the neighbor’s place”. Or, perhaps, “I wonder why I  never hear Witz barking anymore; maybe his batteries died or the neighbors complained to the neighborhood association about his incessant barking”. The obvious meaningfulness and coherence of such utterances suggest that Devitt and Sterelny are mistaken.9 It is quite possible to name and to subsequently say true/​false things about something about whose nature one was, at the time of the naming, radically mistaken. Let us now consider how these observations and reflections might be extended to the grounding of names in the intentional objects of hallucination mistaken by the grounder for “real” (i.e., for mind-​independent) objects. Thus, consider a young girl who experiences recurrent episodes of hypnagogia. Such occurrences regularly feature a large winged angel. The parochial school child, having just learned about archangels, decides to name the hypnagogic figure “Archie”. According to Devitt and Sterelny, the child’s attempt to name the angel-​hallucination will necessarily fail, as she is radically mistaken about its nature insofar as she mistakes a hallucinatory object for a real object. Intuitively, this just seems wrong. To make the point vivid, imagine that the child is told that the angel she saw was just a hallucination. She might then say to herself (or to her young friends), “Archie isn’t a real angel; he’s just a hallucination”. Has she not used the name “Archie” to refer to the (object of) her angel-​hallucination, thereby lending credence to the view that she did indeed successfully ground the name “Archie” in her angel-​hallucination? The proposed view has the conceptual resources to capture and explain these questions. The basic idea is that hallucinations really do exist; they exist no less so than the individuals who have them—​just ask a psychiatrist. Names can accordingly be grounded in them (or their “objects”), as when a patient talks to her psychiatrist about the threatening voices she regularly hears from an evil individual she has named (appropriately) “Abaddon”. She might say, for example, “Abaddon has been threatening my family again”. How can she possibly talk this way, coherently if mistakenly, unless she did in fact successfully name a hallucination (or its “object”), an “object” to which she was then able to refer and about which she was then able to speak?10

(ii)   Negative Existentials Discussion of the qua-​problem leads naturally to problem of (true) negative existentials, a problem easily illustrated by appeal to hallucinations mistaken for the “real thing”. Suppose a child with a history of hallucinating angels, the most impressive of which she names “Archie”, is now a grown woman majoring in philosophy. She is also a staunch atheist, having been convinced of God’s non-​ existence by the “problem of evil”. Having done the relevant research, she is now convinced that “Archie” was a hypnagogic hallucination. Suppose that she says (either to herself or to someone else): “Archie doesn’t really exist—​in fact, he never did”. Intuitively, if not obviously, the woman’s utterance is true. But how can that be? After all, if the woman actually refers to Archie (as she certainly seems to), Archie must exist. Echoing Quine (1948) on Pegasus, “Archie must in some sense be, otherwise what is it that is not?” Thus, the apparently true utterance must actually be false. The proposed view has a simple and intuitive response to this conundrum—​one that involves neither disguised Russellian descriptions nor ontologically suspect Meinongian entities. The woman does in fact refer to Archie, a hallucination of years past, and she says of that hallucination, or rather of its putative “object”, that it does not “really” exist. Surely, she says something true. Archie, after all, is (or was) a mind-​dependent phenomenon. If the speaker then goes on to say (something like): “In fact, Archie does not exist in any sense”, then she arguably says something false, as the 459

Marga Reimer

objects of hallucination do indeed exist in at least a logical (or perhaps “intentional”) sense, much like the unicorn in a unicorn tapestry. But they are no more ontologically independent of the hallucination itself than the unicorn in a unicorn tapestry is ontologically independent of the tapestry itself.11 These reflections lead naturally to a well-​known argument from the philosophy of perception: the argument from hallucination. Let us therefore turn to that argument against naïve realism.

(iii)   The Argument from Hallucination This famous argument might (following Jeff Speaks12) be articulated as follows: 1 In every experience, there is an object of the subject’s awareness. 2 In the case of hallucination, the object of the subject’s awareness cannot be a material object. 3 The objects of awareness are the same in the case of hallucinatory and veridical experience. C1 Material things are never the objects of experience. C2 Every perceptual experience has something other than a material thing as its object. While Speaks claims that the argument is valid and that the second premise is “not open to dispute”, I would question both of these claims. In fact, I think all three premises of this invalid argument are open to dispute. However, here I will discuss only the argument’s validity and its third premise. Let us begin with the question of validity. One might reasonably claim that the argument (formulated as above) is invalidated by an equivocation on the phrase “object of x”. “Object of x” can mean (refer to) something independent of x—​or not. In either case, however, it can function as the intentional object of an agent’s thought: that about which they are thinking. Thus, the object of my current experience is my computer screen, something that exists independently of me, independently of my mind. The object of the child’s hallucinatory experience, an angel, is not something that exists independently of the child, independently of her mind. In both cases, however, there is an intentional object but only in the veridical case is the intentional object mind-​independent. The point can be made vivid by a comparison of paintings and portraits. Imagine the famous late medieval “Unicorn in Captivity” tapestry (a “painting” of sorts) alongside a not so famous painting, a portrait, of the famous racehorse Seabiscuit. In the first case, the tapestry is of a unicorn in captivity; the unicorn is the intentional object of the tapestry—​it is what the tapestry is “about”. In the second case, the painting, a portrait, is of Seabiscuit; Seabiscuit is the intentional object of the portrait—​he is what the portrait is “about”. Summarizing, the tapestry is of a unicorn; the unicorn is its intentional object. The portrait is of Seabiscuit; Seabiscuit is its intentional object. And yet the “objects” of the two works of art are importantly and, indeed, fundamentally different; they are ontologically different. Thus, while there is no unicorn apart from the unicorn tapestry, there is indeed a horse, Seabiscuit, apart from the horse-​painting. That painting is, after all, a portrait of the famous racehorse. The unicorn featured in the famous tapestry did not exist independently of that tapestry (or the tapestry-​maker’s mind); the latter is not, after all, a portrait of the former. The object of hallucination, a hypnagogic angel (say), is importantly similar to the unicorn in the tapestry and importantly different from the horse in the portrait of Seabiscuit. It is thus misleading to say that the “object” of the child’s hallucination is an angel in the same way that the “object” of my current perception is a computer screen. Although both the angel and the computer screen are indeed intentional objects (as is anything that can be thought of), only the latter is an “object” in the ordinary everyday sense of that word. The argument from hallucination, as articulated above, would thus seem to be vitiated by an equivocation on the word “object”—​which is used equivocally to mean both intentional object and material object—​where the latter is understood (roughly) as a Cartesian res extensa or (in other words) a denizen of the first of Frege’s three realms. 460

Demonstrative Reference to the Unreal

Indeed, talk of the “objects” of thought is arguably what Gilbert Ryle refers to as a “systematically misleading way” of speaking (Ryle 1951). But we nevertheless engage it—​whether because of ontological confusion, convenience, or (most likely) a mix of the two. In the computer case, there is the (intentional) object of my experience as well as the experience itself; in the hallucination case, the (intentional) object of the child’s experience just is the experience—​the hallucinatory experience. There is thus no real “object” of the child’s hallucination any more than there is a real “object” of the unicorn tapestry. There is only the angel-​hallucination whose intentional object is an unreal angel; similarly, there is only the unicorn tapestry whose intentional object is an unreal unicorn. One might also question the argument’s third premise and do so on “Shakespearian” grounds. As Macbeth remarks in the lines following those that open the present paper: Mine eyes are made fools o’ the other senses Or else worth all the rest. The obvious response is that, indeed, Macbeth’s eyes are “made fools” of by his other senses, particularly his sense of touch: Come let me clutch thee. I have thee not, and yet I see thee still. It thus seems doubtful (pace Speaks) that the “objects of awareness are the same in the case of hallucinatory and veridical experience”. They are phenomenologically different, indeed dramatically so. In hallucinatory experiences, like Macbeth’s, the senses do conflict with one another; in veridical experiences, they do not (at least not generally). Indeed, they tend to corroborate one another. Thus, Macbeth’s eyes are indeed “made fools of ” by the other senses which, collectively, disconfirm the presence of a real dagger “before” him; Macbeth’s eyes are thus not “worth all the rest”. Rather, “all the rest”, taken collectively, are worth more than his eyes. As these observations and reflections suggest, veridical experience is phenomenologically—​ and not just ontologically—​different from hallucinatory experience insofar as the senses do not, in veridical experiences, conflict with one another, as they so obviously do in Macbeth’s case. The connection between these reflections and the equivocation on the notion of an object (or on the word “object”) should now be clear. When intentional objects are objects of hallucination, whether visual (“angels”) or auditory (“voices”), the operative sense modalities are indeed “made fools of ” by “all the rest”. Moreover, “all the rest” are “worth more” insofar as, collectively, they corroborate the unreality of the hallucinated intentional objects—​objects that are perhaps “seen” but not heard or “heard” but not seen—​or, in the case of Macbeth’s dagger, “seen” but not felt. How does all this tie in with the proposed view of demonstrative reference to hallucinations? The connection is simple: the proposed view respects, while the argument from hallucination violates, Occam’s Razor. More specifically, the former treats as one and the same the so-​called “object” of hallucination and the hallucination itself. It is a single mental phenomenon—​logically of a piece with the unicorn in the famous unicorn tapestry of which it is literally a part.Thus, the hallucinatory angel is one and the same as the angel-​hallucination. By rejecting the equivocation in question, the proposed view is able to escape the argument from hallucination.13

10  The Theoretical Importance of Paradigmatically Non-​paradigmatic Cases This paper has been concerned largely with a decidedly atypical (yet quite common) case of demonstrative reference:  demonstrative reference to hallucinations (or to their intentional “objects”). As noted above, such cases are paradigmatically non-​paradigmatic. For this reason, they might naturally be dismissed or, more likely, simply disregarded by theorists of demonstrative reference. There is in fact something to be said for this sort of attitude insofar as it is natural to begin one’s theorizing about 461

Marga Reimer

demonstrative reference by reflecting on paradigm (clear-​cut, uncontroversial) cases of such reference. This is true of any phenomenon about which one intends to theorize. In the case of demonstrative reference, paradigm cases involve reference to real (physical) objects and individuals and are frequently accompanied by some sort of overt demonstration intended to assist the audience in determining the speaker’s intended referent. Indeed, this is just what theorists generally do—​and why shouldn’t they? This is where theorizing about any given phenomenon should begin—​with the “obvious”, or at least relatively uncontroversial, cases of that phenomenon. Otherwise, the theory will be controversial from the get-​go, its subject matter being ill-​defined. However, after the motivation, development, and defense of a given theory, fine-​tuning is in order. It is at this point that attention to atypical cases is not only apt but arguably essential. Importantly, such fine-​tuning is, or at least can be, considerably more than the proverbial “tying up” of “loose ends”. It might, for example, result in a deeper and simultaneously broader understanding of the atypical phenomenon being investigated. Fine-​tuning in light of atypical cases might also promote a more nuanced understanding of certain distinctions the theoretical importance (or unimportance) of which is made particularly vivid by those cases. Such fine-​tuning might also facilitate an appreciation of the connection between the atypical cases in question and philosophical issues not evident when focusing on the more typical cases. I like to think that this paper can be seen as an illustration of these three points. First, we have recognized that (deferred) demonstrative reference to unrealities is not only common but quite varied insofar as not only hallucinations but numbers and fictional characters appear amenable to such reference. Second, we have gained some insight into the competing roles of speaker intentions and public “demonstrations”. The atypical cases in question seem to suggest that speaker intentions trump public demonstrations when it comes to securing the referent of a demonstrative expression. But not all operative distinctions are significant in this way—​that is, semantically significant. Thus, in terms of what gets said (or referred to), it does not seem to matter whether the speaker is engaging in conversation or soliloquy; nor does it seem to matter whether they are unaware or aware of the fact that the “object” to which they intend to refer is an unreal object. Third, we have a better appreciation of the connection between the “problem” of demonstrative reference to hallucinations and other philosophical problems including, and in particular, the qua-​problem, the problem of true negative existentials, and the argument from hallucination. There is at least one more reason for the philosopher of language (or any other theorist) to attend to the atypical cases of a given phenomenon—​being atypical, they are intrinsically more interesting than the typical cases—​at least for this theorist.

Notes 1 Gareth Evans (1982) is a notable exception. See his provocative discussion of demonstrative reference to “little green men”. See also Reimer (2015) and (forthcoming). 2 See discussion of deferred demonstrative reference in sections 4 and 5 below. 3 For an explanation of the importance of reference to the “objects” of hallucination, see section 8 below. 4 It might be thought that, for present purposes, the unreal might be defined as “not concrete”. However, such a proposal would simply raise further questions, such as whether non-​hallucinatory sensory phenomena, such as sounds and smells, are unreal. I would think not, despite their not being “concrete”—​at least not in any intuitive sense of concrete. 5 I suppose someone, knowingly in the midst of an angel-​hallucination, could point to their head while saying “That’s not a real angel; it’s just a hallucination”. But does this really show that hallucinations can be demonstrated “in the usual way”? Only if the speaker can meaningfully be said to have demonstrated (“in the usual way”) the angel “in his head”. 6 It is also true that while there is an isomorphism between numerals and numbers, one that arguably figures in the explanation of how, in demonstrating the former, one can demonstratively refer to the latter, there is no such isomorphism between locations and the “objects” of visual hallucinations. However, this disanalogy

462

Demonstrative Reference to the Unreal does not undermine the idea that the two cases are analogous insofar as both involve deferred demonstrative reference to non-physical entities (numbers, visual hallucinations) that cannot be demonstrated in the usual sorts of ways, as with ostensive gestures. 7 In this respect, paradigm cases of deferred demonstrative reference to hallucinations differ from paradigm cases of deferred demonstrative reference more generally. In paradigm cases of the latter sort, the speaker knows what it is that they are demonstrating—​such as a photograph of a person who is deferentially referred to by way of that demonstration. 8 If being unreal is an essential property of the object of (deferred) demonstrative reference, then saying of that object that it is unreal would presumably involve a necessary truth—​as would saying of something real (e.g., the keyboard on which I am now typing) that it is real. 9 For a defense of the view that they are in fact mistaken, see Reimer (forthcoming). 10 As a reviewer pointed out to me, a central goal of semantic externalism would appear to lend credence to my case here. As they explain (quoting): “Putnam was motivated partly by a desire to reply to Kuhn’s incommensurability argument.The idea was roughly this: if the meaning of some term is fixed by the causes of relevant uses/​representations, then speakers can associate quite different properties with the expression and yet still refer to the same entity. Insofar as one constrains this view, e.g., by holding that speakers must at least be right about the nature of the referent, one loses some of the ability to respond to Kuhn. So, for example, if theory T treats the referent of some common expression E as having nature N and theory T* treats it as having distinct nature N*, then T and T* are incommensurable according to Devitt but not according to [me]”. 11 This reply is very similar to McX’s reply to Quine (1948). 12 Speaks (2007). 13 Of course, the proposed view posits two kinds of objects of perception (material objects and mental objects) whereas the sense data theorist can posit one.

References Devitt, M. and K. Sterelny (1999) Language and Reality: An Introduction to the Philosophy of Language, 2nd edition. Cambridge, MA: MIT Press. Evans, G. (1982) The Varieties of Reference. Oxford: Oxford University Press. Frege, G. (1892/​1952) “On Sense and Reference”. In M. Black and P. Geach (eds.) Translations from the Philosophical Writings of Gottlob Frege. Oxford: Basil Blackwell. Jackson, M. (2007) “The Clinician’s Illusion and Benign Psychosis”. In M. Chung, K. Fulford and G. Graham (eds.) Reconceiving Schizophrenia. Oxford: Oxford University Press. Kaplan, D. (1989) “Demonstratives: An Essay on the Semantics, Metaphysics and Epistemology of Demonstratives and Other Indexicals”. In Joseph Almog, John Perry and Howard Wettstein (eds.) Themes from Kaplan. Oxford: Oxford University Press. Kripke, S. (1980) Naming and Necessity. Cambridge, MA: Harvard University Press. Meinong, A. (1904) Untersuchungen zur Gegenstandstheorie und Psychologie. Leipzig: J. A. Barth. Quine, W.V. O. (1948) “On What There Is”. Review of Metaphysics, 2, 5: 21–​38. Reimer, M. (2015) “Drawing, Seeing, Referring: Reflections on Macbeth’s Dagger”. In Andrea Bianchi (ed.) On Reference. Oxford: Oxford University Press. Reimer, M. (forthcoming) “The Qua-​Problem for Names (Dissolved)”. In Andrea Bianchi (ed.) Language and Reality from a Naturalistic Perspective: Themes from Michael Devitt. Cham: Springer. Russell, B. (1905) “On Denoting”. Mind 14: 479–​493. Ryle, G. (1951) “Systematically Misleading Expressions”. Proceedings of the Aristotelian Society, New Series, 32: 139–​170. Shakespeare, W. (1623) Macbeth. First Folio. Speaks, J. (2007) “The arguments from illusion and Hallucination”. www3.nd.edu/​~jspeaks/​courses/​2007–​8/​ 43904/​_​HANDOUTS/​illusion-​sense-​data.pdf. Wittgenstein, L. (1953) Philosophical Investigations, G. E. M. Anscombe (trans). Oxford: Basil Blackwell.

463

35 WHAT IS SPECIAL ABOUT DE SE ATTITUDES? Stephan Torre and Clas Weber

1  Introduction De se attitudes seem to play a special role in action and cognition. This raises a challenge to the traditional way in which mental attitudes have been understood. In this chapter, we review the case for thinking that de se attitudes require special theoretical treatment and discuss various ways in which the traditional theory can be modified to accommodate de se attitudes. In section 2, we give a tentative characterization of indexical and de se attitudes, outline the traditional theory, and then present the main theoretical accounts of de se attitudes. In section 3, we discuss the question of whether de se attitudes play a special role in cognition and analyze their relation to Frege’s puzzle. In section 4, we discuss the question of whether de se attitudes play a special role in explaining action. In section 5, we consider the claim that de se judgments are immune to error through misidentification.

2  Indexical Attitudes, De Se Attitudes, and the Traditional Theory There is an intuitive contrast between beliefs which represent the world from a god’s-​eye perspective and beliefs that concern our place within the world, i.e. beliefs about who we are, what we are like, and where we are spatio-​temporally located. For instance, you may believe that quarks exist, that metals conduct electricity, or that 2+2=4. These beliefs do not concern your place in the larger order of things. On the other hand, you may think that you are Napoleon, that today is Monday, or that you are currently in Helsinki. Indexical beliefs are beliefs of the latter type. The distinction also holds for other types of attitudes, e.g. you may fear that metals conduct electricity, or hope that you are currently in Helsinki. It is difficult to give a more explicit definition of indexical attitudes which is also theory-​neutral. Indexical attitudes are often characterized in relation to linguistic indexicality, as attitudes whose expression or ascription involves indexical sentences. Linguistic indexicality (in a broad sense) can in turn be defined in a straightforward way: an expression e is indexical (or context-​sensitive) iff its extension depends on the context, i.e. iff there are contexts c, c* such that ⟦e⟧c ≠ ⟦e⟧c* (Kaplan 1989; Zimmermann 1991).1 This characterization gets many cases right, but it is not entirely adequate. Sometimes, indexical attitudes are expressed in a non-​indexical manner. For instance, LeBron James might express his 464

What is Special about De Se Attitudes?

indexical belief I do what is best for me with the sentence: “LeBron James does what is best for LeBron James”. Similarly, not all attitudes ascribed by an indexical sentence are indexical. Pointing to LeBron James, I might report Aren’s non-​indexical belief LeBron James is the greatest player of all time with the indexical sentence: “Aren believes that he is the greatest player of all time”. One might try to give a definition of mental indexicality that mirrors the above definition of linguistic indexicality: a belief b is indexical iff its truth-​value depends on the context, i.e. iff there are contexts c, c* such that ⟦b⟧c ≠ ⟦b⟧c*. But this approach also faces difficulties. The identification of linguistic expressions across contexts is relatively uncontroversial, but this is not the case for attitudes. Many think that judging on both Monday and Tuesday that Today is Monday involves different beliefs, rather than the same belief in different contexts. The best option may be to give an ostensive definition (Ninan 2016), i.e. point to clear examples and define indexical attitudes as attitudes of the same kind. As sample cases we can use famous examples from the literature, such as Perry’s messy shopper, or we might rely on the phenomenology of inner speech and focus on cases where we explicitly make internal indexical judgments, such as Oh dear, I am making a mess right now. Below, we will also consider the main accounts of indexical attitudes and corresponding theory-​internal definitions. Assume that we have a satisfactory understanding of indexical attitudes.What, then, are de se attitudes? There are two options. According to the dominant view, de se attitudes are a special subset of the set of indexical attitudes, where we represent ourselves in a special, first-​personal manner. Relatedly, de nunc attitudes are attitudes in which we represent the present time as now. On a different view, there is no interesting contrast between de se/​de nunc attitudes and other indexical attitudes.2 The motivation for this understanding of de se attitudes is that indexical attitudes seem in general a priori equivalent to complex I-​now attitudes. For instance, attitudes about here are equivalent to attitudes about the place where I am now; attitudes about that thing are equivalent to attitudes about the thing I am now pointing to, and so on.3 Further, indexical attitudes exhibit many of the features that make de se attitudes theoretically interesting. We can distinguish the main accounts of de se attitudes in terms of how they diverge from what we will call the Traditional Theory of Attitudes. This theory has the following commitments:

The Traditional Theory of Attitudes Binary Relationism: Attitudes are binary relations between subjects and contents. Absolutism: Attitude contents vary in truth-​value at most between different possible worlds. Shareability: It is possible for different subjects to be related to the same content. The Traditional Theory treats attitudes as binary relations between subjects and absolute and non-​ private contents. Contents are abstract entities postulated for two main explanatory tasks: first, they should account for the role of attitudes in rationalizing and explaining behavior; second, they should account for the cognitive significance of attitudes, i.e. their representational content and rational connections to other attitudes. Accounts of contents that are absolute and shareable have been given in terms of sets of possible worlds or structured Russellian propositions. In the late 1970s, John Perry (1977, 1979) and David Lewis (1979) convinced many that the Traditional Theory is inadequate because it fails to accommodate indexical attitudes. Several rival accounts of attitudes have been proposed which abandon different commitments of the theory. While  differing superficially, on a deeper level they have more in common than is ordinarily acknowledged. Below we outline different ways in which the Traditional Theory can be modified to accommodate indexical attitudes. Relativism  Relativists, such as Lewis (1979), abandon Absolutism. They take contents to be not true or false absolutely, but only relative to subjects or times. Relativistic contents can be identified 465

Stephan Torre and Clas Weber

with sets of centered possible worlds (Lewis 1979).  A  centered world can be represented as a pair of a possible world and a designated center within the world. Centers can in turn be modeled as spatio-​temporal locations, or as pairs of a spatio-​temporal location and an individual.4 In principle, Relativists could treat indexical attitudes as relations between subjects and relativistic contents, and non-​indexical attitudes as relations to absolute contents. However, that would lead to a doubling of attitudinal relations: for each attitude we would need two different relations, one linking to relativistic contents, the other to absolute ones. Lewis’s simpler proposal is to treat indexical and non-​indexical attitudes alike as binary relations between subjects and relativistic contents.5 Even though all attitudes are relations to relativistic contents, i.e. sets of centered worlds, some sets of centered worlds are interesting, others are boring (Egan 2007). Boring contents map all centers within a possible world to the same truth-​value. Interesting contents map some centers within the same world to different truth values. Indexical attitudes can then be defined as attitudes with interesting contents; non-​indexical attitudes, on the other hand, have boring contents. Relativists may try to define de se attitudes as indexical attitudes that directly impose a condition on the center-​individual. For instance, the content of the de se belief I am happy is the set of centered worlds, such that the center-​individual is happy (at the time of the center). But it is unclear whether this can be made more precise and the approach is most naturally combined with the view that there is no deep distinction between de se/​de nunc attitudes and indexical attitudes more generally. Relativism can salvage many of the advantages of the Traditional Theory. Maintaining Binary Relationism and Shareability enables the comparison of attitudes across people and times, as well as the explanation of behavioral and cognitive patterns in terms of content. Relativism is the dominant position within the quantitative approach to attitudes, as centered propositions can easily be incorporated into the Bayesian set-​theoretic framework (Chalmers 2011; Titelbaum 2016). A seeming downside is that relativistic accounts of belief-​updating (Meacham 2010; Schwarz 2012) and communication (Gibbard 2012; Weber 2013) are more complicated than standard approaches. But the standard approaches are founded on the simplistic Traditional Theory; compared to other non-​traditional approaches, Relativism does not fare worse. Guises Absolutism  The main alternative to Relativism is Guises Absolutism. Officially, Guises Absolutists abandon Binary Relationism and leave the traditional conception of content untouched. Attitudes are treated as ternary relations between subjects, traditional propositions, and propositional guises. Guises Absolutism was proposed by Perry (1979) and is endorsed by Neo-​Russellians (Salmon 1986; Soames 1987; Braun 2002).6 The mental-​file approach of Recanati (2012b, 2016) also belongs to this category. Two questions arise with respect to guises: First, what exactly are they? Second, how do guises manage to fulfill their explanatory roles? For instance, Perry (1979) treats guises as entities similar to Kaplanian characters. This aims to provide an answer to both questions. However, others treat guises as primitive entities (e.g. Frege 1892) leaving both questions unanswered. On this view, indexical attitudes are relations between subjects, contents/​propositions, and indexical guises. De se attitudes are attitudes that involve a first-​personal guise. (The account needs to also provide an explanation of what makes a guise indexical or first-​personal.) For instance, Claudia’s de se belief I like fairies is analyzed as a relation between (i) Claudia, (ii) the content that Claudia likes fairies, and (iii) a first-​personal guise. Guises Absolutists acknowledge that traditional contents cannot explain attitudes’ behavioral and cognitive role all by themselves—​hence the need to supplement contents with guises. This suggests a different way of looking at things (Weber 2016; Ninan 2016; Torre and Weber 2019). Above, we have introduced the technical term content by its two central explanatory roles. Since the entities that play these roles are not traditional contents alone, but rather traditional contents in combination with guises, it seems more appropriate for Guises Absolutists to identify contents with pairs of the form: . According to this interpretation, the approach can be seen as abandoning 466

What is Special about De Se Attitudes?

Shareability rather than Binary Relationism. Claudia’s de se belief is a binary relation between her and the ordered content pair .7 Only Claudia can be related to this pair, since only she can have beliefs about Claudia under the first-​personal-​guise. Russellian Absolutism  It is possible to give an Absolutist account of indexical attitudes that avoids guises. Call this view Russellian Absolutism. Russellian Absolutism maintains Binary Relationism and explicitly abandons Shareability. The view encompasses four key claims (Weber 2016). First, de se and de nunc attitudes are relations to singular contents concerning the subject S or the present time t. Second, all non-​indexical attitudes are relations to qualitative contents. Third, only S can entertain singular contents about S, and only at t can we grasp singular contents about t. Fourth, every subject S at time t knows a priori and with certainty that they are S and that the present time is t.8 Here, S and t can play the role that centers play in Relativism. For instance, Relativists model purely temporal ignorance as involving several differently located centers within a possible world; Russellian Absolutists instead employ several qualitatively identical worlds that differ merely in t’s temporal location. Russellian Absolutists can characterize a subject S’s de se attitudes as attitudes with singular content about S (de nunc attitudes are singular attitudes about t). Other indexical attitudes can then be reduced to more complex de se/​de nunc attitudes, in the way suggested above (though note the difficulties pointed out in note 3). For instance, beliefs about yesterday are beliefs about the day before the day that includes t; beliefs about that this thing are beliefs about the thing that S is pointing to at t, etc. Like Relativism, the account is most naturally combined with the view that there is no principled distinction between de se attitudes and indexical attitudes more generally. Pluralism  Pluralists associate attitudes with several contents, which play different theoretical roles. Its main proponents are Perry (2001) and Chalmers (2002, 2011). According to Chalmers  (2011),  beliefs can be associated with (at least) two contents:  a primary intension and a secondary intension. Primary intensions resemble relativistic contents; secondary intensions resemble traditional absolute contents. We can also associate beliefs with enriched intensions which are complexes of primary and secondary intensions. Enriched intensions can do the combined explanatory work of primary and secondary intensions. Pluralism thus understood sacrifices Shareability—​only Claudia can be related to the enriched intension .9 Pluralists can define indexical attitudes in a similar way as Relativists, as attitudes whose primary intension varies in truth-​value within the same world. And, as suggested above, they might then try to define de se attitudes as indexical attitudes that impose a direct constraint on the center-​individual. Alternatively, they may attempt to define de se attitudes as indexical attitudes whose secondary intension is a singular content about the subject. However, it is doubtful whether this will work, as it seems possible to entertain indexical attitudes about oneself in a non-​de se way. Once more, this suggests that there may not be a deep distinction between indexical and de se attitudes. The different accounts of de se attitudes agree in important respects. All acknowledge that contents, i.e. the entities which play the relevant theoretical roles, cannot both be absolute and shareable at the same time. All accept that reference is often mediated. All maintain that certain referential relations, e.g. to the subject or the present time, have to be accepted as fundamental (either explicitly as Russellian Absolutists, or implicitly as Relativists). And all assign de se attitudes a special theoretical status in that they require a modification of the Traditional Theory. There are two main lines of argument for the last claim, corresponding to the two above-​mentioned theoretical roles. The first relies on the role of de se attitudes in cognition and rational inference; the second relies on their role in action. We will consider them in turn in the following sections.

467

Stephan Torre and Clas Weber

3  The Cognitive Role of De Se Beliefs 3.1  Three Arguments from De Se Attitudes against the Traditional Theory De se beliefs seem to play a special role in cognition.They seem to convey a distinctive, indexical type of information, irreducible to the information encoded in non-​indexical, objective beliefs. Perhaps the most famous case for this claim is Lewis’s story of the two gods (Lewis 1979); Lewis uses the term “proposition” to refer to traditional (i.e. absolute and shareable) contents. [The two gods] inhabit a certain possible world, and they know exactly which world it is. Therefore they know every proposition that is true at their world. Insofar as knowledge is a propositional attitude, they are omniscient. Still I can imagine them to suffer ignorance … Neither one knows whether he lives on the tallest mountain or on the coldest mountain. … Surely, their predicament is possible. … But if it is possible to lack knowledge and not lack any propositional knowledge, then the lacked knowledge must not be propositional. Lewis 1979, 520–​521 The scenario Lewis describes seems conceivable and hence possible. But its possibility is incompatible with the Traditional Theory of Attitudes.Therefore, the theory is inadequate.We can turn Lewis’s story into an explicit argument against the Traditional Theory: Lewis’s Indexical Knowledge Argument 1 2 3 4 5

Conceivability: It is conceivable to be omniscient with respect to every proposition true at the world while having de se ignorance (i.e. ignorance about who or where one is). Conceivability—​Possibility: If it is conceivable to be omniscient with respect to every proposition true at the world while having de se ignorance, then this is possible. Possibility: Hence, it is possible to be omniscient with respect to every proposition true at the world while having de se ignorance. Traditional Theory and Omniscience: The Traditional Theory of Attitudes entails that it is not possible to be omniscient with respect to every proposition true at the world while having de se ignorance. Conclusion: The Traditional Theory of Attitudes is flawed.

The argument resembles Jackson’s knowledge argument (Jackson 1982).10 Both rely on the idea that complete knowledge in one domain (physical facts vs. objective facts) is compatible with ignorance in another domain (phenomenal facts vs. indexical facts). According to the Traditional Theory, all attitudes, including de se attitudes, are binary relations between subjects and propositions (i.e. traditional contents). So, possessing de se knowledge is to know that the relevant proposition is true. The theory therefore implies that a subject who knows all true propositions also has complete de se knowledge. While some commentators have questioned the coherence of the scenario (Stalnaker 2008; Kwon 2010),11 the main complaint has been that the argument is question-​begging (e.g. Stalnaker 1981; Cappelen and Dever 2013). In its present form, the criticism seems apt. The problem is that Lewis formulates his case using the technical term “proposition”. This hampers a pre-​theoretical assessment of the argument and seems to prejudge the question of whether indexical knowledge is propositional knowledge. A way to remedy this issue is to put the argument in more neutral terminology, replacing “knowledge with respect to propositions” with “objective knowledge”. The Indexical Knowledge Argument 1 2

Conceivability: It is conceivable to have all objective knowledge while having de se ignorance. Conceivability—​Possibility: If it is conceivable to have all objective knowledge while having de se ignorance, then this is possible. 468

What is Special about De Se Attitudes?

3 4 5

Possibility: Hence, it is possible to have all objective knowledge while having de se ignorance. Traditional Theory and Omniscience: The Traditional Theory of Attitudes entails that it is not possible to have all objective knowledge while having de se ignorance. Conclusion: The Traditional Theory of Attitudes is flawed.

To assess this argument, we need an explanation of objective knowledge.The central idea is that objective knowledge is the possession of information which is shareable and standpoint-​independent, i.e. not bound to a certain context or individual. The reformulated Conceivability premise which no longer uses the technical term “proposition” is both pre-​theoretically plausible and no longer question-​ begging. We can imagine the gods having a “god’s-​eye view on the world”, i.e. an all-​encompassing perspective that doesn’t privilege any specific viewpoint. We can further imagine that their view of the world is perfect—​they know, in this objective, non-​perspectival manner, everything that is going on in their world.12 In the original formulation, Traditional Theory and Omniscience was true by definition. This is no longer the case for the modified argument. Still, the premise seems compelling. Traditional contents are absolute and shareable, i.e. their truth does not depend on who considers them when and they can be grasped by any subject with the appropriate cognitive resources. There should therefore be no obstacle to acquiring such contents in an objective manner. Hence, someone who knows all contents which are knowable in an objective way should know all knowable contents whatsoever. A defender of the Traditional Theory may resist the conclusion by insisting that de se ignorance is failure to know a traditional content and that the scenario is therefore impossible, or by claiming that the notion of objective knowledge is either tendentious or unclear.We can sidestep these concerns by modifying Lewis’s case (Weber 2015, 2016). Assume that the gods have resolved their de se ignorance and are now completely omniscient, both with respect to objective truths and de se truths. According to the Traditional Theory they are now belief-​related to all and only true contents. (We can further assume that the gods also share all other attitudes towards traditional contents.) But according to the Traditional Theory, subjects who share all attitudinal relations to traditional contents cannot differ cognitively.Yet, the god’s perspectives are not the same: e.g. the god on the highest mountain believes I am on the highest mountain, while the god on the coldest mountain believes I am on the coldest mountain. These perspectival differences seem to be cognitive: they correspond to differences in representational contents (at least in one sense), differences in rational inferences, and differences in behavioral dispositions. We can summarize this argument as follows: The Perspectives Argument 1 2 3 4 5

Traditional Theory and Cognitive Sameness: The Traditional Theory of Attitudes entails that subjects who stand in the same attitudinal relations to absolute contents are cognitively indistinguishable. Different Perspectives: It is possible for subjects who stand in the same attitudinal relations to absolute contents to differ in de se attitudes. Perspectival Difference—​Cognitive Difference: Subjects who differ in de se attitudes differ cognitively. Cognitive Difference: Hence, there are subjects who stand in the same attitudinal relations to absolute contents but differ cognitively. Conclusion: The Traditional Theory of Attitudes is flawed.

The final argument concerning the special cognitive role of de se beliefs that we will consider concerns how de se knowledge can be acquired.The Traditional Theory seems to imply that one might in principle acquire de se knowledge in an objective manner. Using a somewhat simplistic metaphor, we may imagine that there is a Book of the World that contains all truths about the world, i.e. it contains all true sentences which express absolute contents. According to Absolutism, one should be able to learn every fact from reading the book. But this seems impossible for de se truths. Consider two subjects, Ani and Aren. Assume they start from a state of absolute ignorance. Then they are handed the Book 469

Stephan Torre and Clas Weber

of the World. Can they learn who they are just from reading the book? It doesn’t seem so. Since both have the exact same informational resources and are both starting from a state of absolute ignorance, it would be odd if they arrived at different conclusions. It would seem unjustified for Aren to infer I am Aren and for Ani to infer the different conclusion I am Ani. What could be the rational ground for these differences, if all they had to go on is what’s written in the book?13 These considerations can be summarized in the following argument: The Learning Argument 1 2 3 4

Indexical Learning: It is not possible to learn all graspable de se truths merely from learning which absolute contents are true. Impossible Learning: Hence, it is not possible to learn all graspable truths merely from learning which absolute contents are true. Traditional Theory and Learning: The Traditional Theory of Attitudes entails that it is possible to learn all graspable truths merely from learning which absolute contents are true. Conclusion: The Traditional Theory of Attitudes is flawed.

To be compelling, each argument would need further elaboration. Still, together they form a strong negative case: de se attitudes cannot be accounted for in the confines of the Traditional Theory. But they do not tell us which alternative account to choose instead.14 First, Relativists maintain that the gods are lacking knowledge of relativistic contents. Such knowledge is not available in an objective, context-​independent manner. Only the god on the highest mountain can know the relativist content I am the god on the highest mountain, since only relative to his point of evaluation is this content true. Second, Guises Absolutists hold that belief and knowledge involve contents and guises. One may come to know the same absolute content under an objective, third-​personal guise or under a first-​personal guise. The gods’ objective omniscience consists in their knowledge of all contents under objective guises. Their de se ignorance consists in their failure to know some contents under a first-​personal guise. Pluralists can give a similar response, using the first element of content to do the job of guises. Russellian Absolutists deny that it is possible for the gods to know all absolute contents. Most singular contents are out of the reach even of divine beings. The gods’ objective omniscience corresponds to omniscience concerning the qualitative character of their world.Their de se ignorance is haecceitistic ignorance regarding which of two qualitatively identical worlds they inhabit.15 Parallel responses regarding the other two arguments are available to each account.

3.2  De Se Attitudes and Frege’s Puzzle The main opposition against assigning de se attitudes a special theoretical status is based on the idea that the puzzle of the de se is simply an indexical variant of Frege’s puzzle (Boer and Lycan 1980; Stalnaker 1981; Spencer 2007; Devitt 2013; Cappelen and Dever 2013; Magidor 2015). We can, so the idea goes, simply subsume the puzzle under the more general Frege puzzle and solve both with the same resources. Frege’s original puzzle is about language (Frege 1892). But underneath the linguistic puzzle there is a puzzle about the mind. Instances of the puzzle involve rational subjects who ascribe a certain property to an object presented in one way, but fail to ascribe the same property to that object when it appears in a different way. For instance, Ani may rationally have the belief Hesperus is a planet and fail to have the belief Phosphorus is a planet. Her rationality is explained by the fact that she is unaware of the fact that Hesperus = Phosphorus. Such ignorance of identities is the defining mark of Frege puzzle cases. Famous instances of the de se puzzle, such as Lewis’s two gods or Perry’s messy shopper, appear to have the same structure, appear to involve ignorance of identities, and therefore support the analogy: the god on the highest mountain does not know I = the god on the highest mountain, Perry’s messy shopper fails to realize I = the messy shopper.16 470

What is Special about De Se Attitudes?

Frege’s puzzle shows that reference to objects is often mediated by what Frege called “sense”—​ we often refer to objects via their properties. Thus understood, it seems possible to solve the puzzle within the traditional framework by replacing singular contents with contents involving the relevant properties. For instance, we no longer conceive of the belief that Hesperus is a planet as having a singular content involving Hesperus. Rather, we ascribe it a qualitative content involving the property of being the brightest celestial object in the evening sky. The fact that Frege’s puzzle is prima facie compatible with the Traditional Theory, whereas the de se puzzle is not, suggests that the two puzzles are distinct. But things are not that simple. What complicates the situation is that many senses include an indexical element (e.g. Lewis 1997). In a sense, it is therefore correct to say that both puzzles can be solved with the same resources. However, the reason is not that de se attitudes fail to pose a unique theoretical challenge. Rather, the reason is that the de se thoroughly permeates Frege’s puzzle as well; whatever is needed for an adequate treatment of de se attitudes is also required for a solution of Frege’s puzzle. In the remainder of this section, we will survey additional reasons for thinking that the de se puzzle cannot simply be subsumed under Frege’s puzzle.17 Resolution  An important difference between de se puzzle cases and Frege puzzle cases concerns the manner in which such cases can be resolved. Assuming that senses are non-​indexical, Frege puzzle cases can be resolved using objective information alone, but de se puzzle cases cannot be so resolved. To resolve a puzzle case is to remove the subject’s underlying ignorance of identities. According to the above analysis, Ani’s ignorance that Hesperus = Phosphorus consists in her failing to realize that one and the same object is both the brightest celestial object in the evening sky and the brightest celestial object in the morning sky. When she learns that both properties are co-​instantiated, her ignorance is removed. The relevant information is objective, and only concerns the instantiation of properties. Assuming that the arguments in section 3.1 are essentially correct, indexical ignorance cannot be resolved with purely objective information. Shareability  Another important difference between de se puzzle cases and Frege puzzle cases is that Frege puzzle cases are shareable in a way in which de se puzzle cases are not. Consider the case in which Ani believes that Hesperus is a planet and believes that Phosphorus is not a planet. This case involving a Frege puzzle is shareable in the following way: another individual, Aren, could also believe the same contents as Ani, and thereby find himself in the same Frege Puzzle case. Furthermore, just like Ani, Aren gains new information when he learns that Hesperus is Phosphorus. Ani and Aren can be just alike with respect to their content before realizing Hesperus is Phosphorus, as well as just alike after realizing that Hesperus is Phosphorus: both gain new information when they learn that Hesperus is Phosphorus and this change in information is reflected in a change in content. However, de se puzzle cases are not shareable in the same way. Consider the case in which Perry, prior to realizing that he himself is making a mess, believes of the messy shopper that he is making a mess. Call whatever content Perry believes prior to his realization ‘C1’. A fellow shopper Susan can presumably have a belief with the same content ‘C1’ and thereby believe of Perry that he is making a mess.18 However, when Perry realizes that he himself is making a mess, he gains new information and presumably comes to believe a new content ‘C2’. However, it does not seem that Susan can share this change in belief content: there does not seem to be any plausible candidate for C2 such that C2 is the same content that Perry believes after his realization, and it involves a change in belief for Susan.19 Non-​Fregean de se puzzle cases  Resolution and Shareability suggest that even if all de se puzzle cases were structurally similar to Frege puzzle cases, their treatment would still require additional theoretical resources. But there is a further problem: some de se puzzle cases do not have a Fregean structure. Frege puzzle cases involve ignorance of identities, but not all de se puzzle cases exhibit such ignorance. The most prominent example of a non-​Fregean de se puzzle case is Perry’s (1977) bear-​attack case (see section 4); for other examples see (Perry 2006; Weber 2015, 2016; Stalnaker 2016; Torre and Weber 2019). A skeptic may suspect that we have perhaps overlooked a 471

Stephan Torre and Clas Weber

hidden piece of identity-​ignorance. The modified two gods case in section 3.1 proves that this is not so: it is a de se puzzle case, but there can be no ignorance of identities involved, since the two gods are indexically and non-​indexically omniscient.

4  De Se Attitudes and Action Explanation Much of the literature on de se attitudes focuses either explicitly or implicitly on the role of such attitudes in action explanation. Many of the well-​known examples involve an action that seems to require de se attitudes in order to explain why the action occurred. Consider perhaps the best-​known such example, Perry’s messy shopper: I believed at the outset that the shopper with a torn sack was making a mess. And I was right. But I didn’t believe that I was making a mess. That seems to be something I came to believe. And when I came to believe that, I stopped following the trail around the counter, and rearranged the torn sack in my cart. My change in beliefs seems to explain my change in behavior. Perry 1979, 5, our emphasis Perry’s claim is that an indexical belief is necessary in order to explain why Perry changed his behavior: why he finally bent down to adjust the leaking bag of sugar in his shopping cart. Some have questioned the extent that examples like the messy shopper establish that de se attitudes are required for action explanation (e.g. Spencer 2007; Cappelen and Dever 2013, ch. 3; Magidor 2015). In recent years, however, substantial work has been done to more fully articulate such arguments.20 Before considering such arguments, it is worth spending some time clarifying exactly what thesis about the connection between de se attitudes and action explanation they aim to establish. Frequently in the literature on de se attitudes and action explanation, the thesis under dispute seems to be the following. (AE1) All successful action explanations necessarily require an appeal to de se attitudes.21 However such a thesis is far too strong. It is not difficult to come up with merely possible models of action that do not require an appeal to indexical attitudes in explaining action. Cappelen and Dever (ms) provide several, one of which is an occasionalist world in which all events are brought about by an occasionalist god:  “When God sees that Jones wants a beer, and that Jones believes there are beers in the refrigerator, God causes Jones to go to the refrigerator” (Cappelen and Dever ms, 3). Maintaining that de se attitudes play a distinctive role in action explanation does not require maintaining that de se attitudes are needed in the explanation of all possible actions including in those distant possibilities in which actions might be produced by different mechanisms. So what distinctive role do de se attitudes play in action explanation? The most developed arguments for the need for de se attitudes in action explanation all involve cases of a certain sort. In one such case, Al is being chased by a bear and Betty is not (Perry 1977; Ninan this volume). Al curls up into a ball to avoid being mauled, and Betty waves her arms to frighten away the bear. In another case, David’s pants are on fire, and Susan’s are not (Torre 2018). David stops, drops, and rolls, and Susan runs to get the fire extinguisher. Cases such as these involve two agents who are in distinct predicaments (chased by a bear/​not chased by a bear; pants on fire/​pants not on fire). Furthermore, the agents are fully aware of their own predicament and the other agent’s predicament; there is no identity confusion or confusion with respect to whose pants are on fire or who is being chased by a bear. Another feature of such cases is that awareness of being in distinct predicaments motivates the agents to perform actions of distinct action-​types: Al is motivated to curl up in a ball, whereas Betty is motivated to wave her arms; David is motivated to stop, drop, and roll, whereas Susan is motivated 472

What is Special about De Se Attitudes?

to get the fire extinguisher. Let us call such cases Known Distinct Predicament (KDP) cases. KDP cases are cases in which there are agents in distinct predicaments in the sense that one has a particular feature that the other lacks, the agents are aware of their own predicament and that of the other agent, and this awareness of being in distinct predicaments motivates the agents to perform actions of distinct action-​types. We maintain that considering KDP cases provides support for the following claim: (AE2) For many ordinary actions, the best explanation of those actions appeals to de se attitudes. Note that (AE2) is weaker than (AE1). (AE2) is compatible with the existence of actions that can be explained without appeal to de se attitudes. Nonetheless, (AE2) maintains that de se attitudes are required in our best theory of action explanation.22 KDP cases provide compelling arguments for (AE2) and establishing this principle vindicates the claim that de se attitudes play a significant role in action explanation. The argument for (AE2) involves recognizing that the Traditional Theory of Attitudes presented in section 2, when combined with a plausible thesis connecting mental attitudes and action explanation, provides the wrong predictions in KDP cases. In order to provide the correct prediction in such cases, it is argued further that the Traditional Theory of Attitudes must be amended in such a way as to introduce de se attitudes. It is natural to appeal to an agent’s beliefs and desires in explaining why she performed a particular action. Why did Ani get milk from the fridge? Because she desired milk, believed that there was milk in the fridge, etc. Furthermore, if appealing to these beliefs and desires successfully explains why Ani performs this action, anyone with the same relevant beliefs and desires as her will act similarly, other things being equal. If Aren has the same relevant beliefs and desires as Ani, other things being equal, Aren will also get milk from the fridge. Of course, the plausibility of this claim depends very much on the ‘other things being equal’ clause. If Aren is paralyzed, has a fear of fridges, or is suffering from vivid hallucinations, then sharing Ani’s relevant beliefs and desires will not necessarily result in Aren performing the same action.23 These plausible claims about the connection between shared attitudes and action explanation can be stated in the following thesis: Explanation: If the explanation for why S performed an action of action-​type α is the fact that S has attitudes a1, … an, then if T shares the attitudes a1, … an then, other things being equal, T will also perform an action of action-​type α. Let us refer to the Traditional Theory of Attitudes introduced in section 2 together with Explanation as the Traditional Theory of Attitudes + Explanation or TTA+E. What is it for two subjects to share attitudes according to TTA+E? Given the Traditional Theory’s commitment to Binary Relationism and Shareability, two subjects share an attitude just in case they stand in the same attitudinal relation to one and the same content. For example, Ani and Aren share the belief that there is milk in the fridge just in case they both stand in the belief relation to the content that there is milk in the fridge. The argument for the claim that TTA+E provides the wrong prediction in KDP cases can be summarized as follows:24 Al and Betty are out for a walk in the woods and a bear begins to chase Al (and let us suppose that the bear is not chasing Betty). Al believes that he is being chased by a bear, he desires not to be mauled, and he believes that if he curls up in a ball, he will not be mauled. Al curls up in a ball and avoids being mauled. Suppose that Betty shares Al’s attitudes. Given Absolutism, if Betty believes what Al believes when Al believes that he is being chased by a bear, Betty believes that Al is being chased by a bear.25 Sharing Al’s belief cannot involve believing that Betty is being chased by a bear since this belief is false and, by Absolutism, Al’s belief is true for every subject that shares it. Similarly, in sharing his attitudes, Betty desires that Al not be mauled and she believes that if Al curls up, he will avoid being mauled. If Betty shares all Al’s relevant attitudes, then, by Explanation, Betty will also curl up. However, this is the wrong result: given her attitudes, Betty would presumably 473

Stephan Torre and Clas Weber

perform an action of a different action-​type such as trying to frighten away or distract the bear, using her bear mace, calling for help, etc. So, TTA+E predicts that Betty would perform an action of the same action-​type as Al: curl up in a ball. But this is the wrong result and so TTA+E must be false. We do not yet have an argument for the conclusion that de se attitudes are required in explaining Al’s action. Rather we have an argument for the conclusion that one of the theses of TTA+E must be rejected. In section 2 we outlined how rejecting one of the theses of the Traditional Theory involves adopting an account of indexical attitudes (either Relativism, Guises Absolutism, Russellian Absolutism, or Pluralism). The Relativist response to the above argument would be that the content of Al’s belief that he would express by saying ‘I am being chased by a bear’ is an interesting content true for all and only those individuals that are being chased by a bear. So, for Betty to share this content would be for Betty to falsely believe that she herself is being chased by a bear. Explanation could then be upheld since it is plausible to claim that given such shared attitudes, Betty would also curl up in a ball. If the Guises Absolutist reserves the term ‘content’ for traditional contents that are absolute and shareable, she will deny that sameness of content entails sameness of action. Al and Betty may share the same traditional contents (both believe that Al is being chased by a bear, both desire that Al not be mauled), but they do not perform actions of the same action-​type because they believe and desire these contents under different guises. Likely, the Guises Absolutist will accept a version of Explanation that appeals to guises instead of traditional contents: if two agents share the relevant guises, then they will perform actions of the same action-​type. In this sense, the version of Explanation that the Guises Absolutist adopts will be similar to that adopted by the Relativist except that it will be formulated in terms of shared guises rather than shared contents. The Russellian Absolutist response to the above argument would be to reject Shareability and maintain that Betty cannot believe what Al believes when he believes that he is being chased by a bear. In believing that he himself is being chased by a bear, there is some absolute, singular proposition that Al believes that is unshareable: no one distinct from Al can believe it. Although this response manages to uphold Explanation, it does so by claiming that in cases like the bear chase, it is vacuously true since the explanation of Al’s action involves unshareable contents. So although Explanation is upheld, this response renders it idle in cases involving de se attitudes. If we are inclined to think that Explanation provides an informative thesis about the connection between mental attitudes and action, this seems to provide reason to favor Relativism or Guises Absolutism over accounts like Russellian Absolutism that reject Shareability. So what is special about de se attitudes when it comes to action explanation? It is not plausible to maintain that all successful action explanations necessarily require an appeal to de se attitudes. The proponent of de se attitudes should grant that there are some actions that can be successfully explained without appeal to de se attitudes, especially in merely possible cases in which actions might be produced by different causal mechanisms. Nonetheless, granting this doesn’t mean denying that de se attitudes play a significant role in action explanation. As KDP cases demonstrate, our best theory for explaining actions seems to require an appeal to de se attitudes.

5  Are De Se Beliefs Epistemically Special? Many maintain that de se beliefs have distinctive epistemic properties. Most commonly it is claimed that they are immune from a certain kind of error, specifically that they are immune from error through misidentification. Suppose Bryan sees a woman in the distance walking a terrier and he judges “Aunt Lillian is walking a terrier”. There are two ways in which Bryan’s judgment might be mistaken. It might be mistaken in virtue of ascribing a property to Aunt Lillian that she doesn’t in fact have; for example, she may be walking a retriever. Or Bryan might be mistaken, not in the property he ascribes, but in who or what he is ascribing it to: it might not be Aunt Lillian, but Aunt Lillian’s neighbor who is walking a terrier. The latter type of mistake has been called an error through misidentification. 474

What is Special about De Se Attitudes?

Many have noted that first-​person judgments seem to be immune to this sort of mistake. When Bryan judges that he himself has his legs crossed, he might be mistaken about the position of his legs, but it seems impossible for him to be mistaken regarding whom he is ascribing the property to. This purported immunity to error through misidentification (henceforth ‘IEM’) seems to grant a special status to first-​person or de se judgments: they seem to be immune from error in a way in which judgments about others are not. It is not a straightforward matter to articulate exactly what IEM is. If Bryan acquires the judgment that his legs are crossed through the usual means, by proprioception, it seems plausible that the judgment is immune from error through misidentification:26 there is a possibility that the belief is false, for example if his proprioceptive system is malfunctioning, but it is difficult to see how it could be false because he mistakes himself for a distinct individual that has crossed legs. If, on the other hand, Bryan forms the belief by looking into a mirror during yoga class while anesthetized from the waist down, it does seem possible that his judgment could be false by mistaking himself for a different individual that has crossed legs, “I thought it was me whose legs were crossed, but it’s that guy!” This suggests that whether a judgment exhibits IEM depends on the way in which the judgment is acquired. An attempt at characterizing IEM can be given as follows: (IEM) S’s judgment that ‘a is F’ arrived at through way w is immune to error through misidentification just in case there are no possibilities in which S acquires the judgment that ‘a is F’ through way w, the judgment is false, and it is false in virtue of mistaking a for something that is truly F.27 In investigating whether de se judgments are epistemically special in virtue of exhibiting IEM, we will focus on the following two questions: (1) Do some de se judgments exhibit IEM? (2) Assuming that some de se judgments exhibit IEM, is this indicative of a special feature of de se judgments? One who wishes to maintain that exhibiting IEM is a special epistemic feature of de se judgments will maintain, not only that de se judgments exhibit IEM, but that this highlights something distinctive about such judgments. One who denies that IEM accounts for a special epistemic feature of de se judgments could either maintain that they do not in fact exhibit IEM (deny (1)) or that they do, but this is not indicative of anything special about de se judgments; rather it is a feature that holds of a broader class of singular judgments (accept (1), but deny (2)). We address both questions (1) and (2) below.

5.1  Do Some De Se Judgments Exhibit IEM? Most in the literature agree that there are some cases of de se judgments that are immune to error through misidentification. A classic example of a de se judgment that exhibits IEM is (P) I feel pain. Sydney Shoemaker writes: The statement “I feel pain” is not subject to error through misidentification:  it cannot happen that I am mistaken in saying “I feel pain” because, although I do know of someone that feels pain, I am mistaken in thinking that person to be myself. 1968, 557 475

Stephan Torre and Clas Weber

Cappelen and Dever (2013) are among the few who claim that no de se judgments exhibit IEM. In fact, they claim that no judgments whatsoever exhibit IEM. In arguing against the claim that de se judgments exhibit IEM, they consider a case in which Bryan is wired up in such a way that he sometimes receives proprioceptive information about John’s legs. Being aware of this, it makes sense for Bryan to wonder whether it is John’s legs that are crossed rather than his own. “So even if [Bryan] has not been deviantly wired, he is not immune to error, because until the possibility of deviant wiring has been eliminated, the possibility of error due to misidentification cannot be eliminated” (Cappelen and Dever 2013, 131). They further claim that this counterexample provides a “general recipe” for showing that de se judgments in general fail to exhibit IEM “for any of [Bryan’s] beliefs about himself and any source of evidence for that belief, we can imagine him being such that he sometimes receives evidence of that sort from someone else’s state” (Cappelen and Dever 2013, 131). Whether or not one thinks that this deviant wiring case succeeds in showing that Bryan’s judgment fails to exhibit IEM, it isn’t clear that the general recipe they propose succeeds in demonstrating the strong claim that they are arguing for: that no de se judgments, including judgments like (P), exhibit IEM.28 Being deviantly wired up to John in such a way that Bryan feels pain when John steps on a tack would presumably not make Bryan’s judgment ‘I feel pain’ false. It is difficult to see how we would construct the deviant wiring case in order to introduce the possibility of error through misidentification in Bryan’s judgment ‘I feel pain’ and similarly for other de se judgments involving phenomenal experiences like ‘I am experiencing redness’ or ‘I feel as though my legs are crossed’. So even if one grants Cappelen and Dever’s argument for the claim that judgments of ‘My legs are crossed’ gained through proprioception fail to exhibit IEM, they haven’t provided a sufficiently general argument for the claim that no de se judgments, including relatively uncontroversial ones like ‘I feel pain’, exhibit IEM.

5.2  Assuming that Some De Se Judgments Exhibit IEM, is this Indicative of a Special Feature of De Se Judgments? Let us assume that there are genuine cases of de se judgments that exhibit IEM. It seems implausible to maintain that exhibiting IEM is a distinctive feature of de se judgments in the sense that all and only de se judgments exhibit IEM. There are relatively uncontroversial cases both of de se judgments that fail to exhibit IEM and of non-​de se judgments that exhibit IEM. For example, Bryan could form the de se judgment that he himself has Groat’s disease from reading through his medical records. Given the way he acquired the de se judgment, it is not immune from error through misidentification because there is a possibility in which his belief is false in virtue of mistaking himself for another patient that truly has Groat’s disease (for example, if he reads another patient’s records believing that they are his own). Many claim that demonstrative judgments like that is a lemon, based on vision, exhibit IEM: it seems possible that the demonstrated object lacks the property ascribed, but it does not seem possible for the subject to misidentify the demonstrated object with another that truly has the property. However, the fact that IEM is not unique to de se judgments does not mean that exhibiting IEM is not indicative of a special epistemic feature of de se judgments. It may well be that the IEM exhibited by many de se judgments nonetheless results from certain distinctive features of de se attitudes. The IEM exhibited by demonstrative judgments may be derived from, or of a different sort from, that exhibited by de se judgments. One such account is given by François Recanati (2012a).29 Recanati maintains that de se judgments exhibit IEM when they are directly based on experience. There is a mode associated with certain experiences that makes them about the subject who has them. When Bryan forms the judgment ‘I have crossed legs’ based on proprioception, the proprioceptive mode ensures that the content is about the subject, and the content of his judgment is a property:  the

476

What is Special about De Se Attitudes?

property of having crossed legs. The content of the judgment contains no identification of the subject and so no misidentification is possible: the IEM exhibited by the judgment results from the fact that the subject-​less content having crossed legs of the judgment is associated with an experiential mode that must concern the subject of the experience. Other theorists, for example, Daniel Morgan (2012), Jason Stanley (2011), and Crispin Wright (2012), follow Gareth Evans (1982) in providing a more general account of when singular judgments exhibit IEM that denies that this epistemic feature highlights anything distinctive about de se judgments. They endorse what has come to be called ‘the Simple Explanation’: that judgments exhibit IEM just in case they are not based on30 an identity statement.31 The Simple Explanation has the implication that any singular judgment that is not based on an identity will be IEM. Wright writes, According to [the Simple Explanation], IEM is a phenomenon of singular judgment in which no significant identification, associating one singular mode of presentation with another, features as part of the grounds. It is thus, in particular, a feature of basic singular thought. Moreover, a proponent of the Simple [Explanation] may continue, there is absolutely nothing here that is peculiar to the first person. IEM I-​thoughts are simply one kind of basic singular thought—​basic singular thoughts which concern oneself. Wright 2012, 255 So embracing the Simple Explanation seems to involve granting that exhibiting IEM is not indicative of a special feature of de se judgments, but rather stems from a more general fact about the basis of singular judgments. However it is far from clear that embracing the Simple Explanation requires one to accept that the IEM exhibited by de se judgments fails to be indicative of a special feature of such judgments. It may well be that there is a distinctive mechanism that produces de se judgments and the mechanism ensures that such judgments are not based on identity statements. This is consistent with the claim that there are other non-​de se judgments that exhibit IEM produced in a different way. Recanati’s account provides one such case in point:  the nature of the experiential mode and the content associated with certain de se attitudes precludes an identification component as part of the subject’s grounds for the judgment, and so the Simple Explanation seems to support, rather than undermine, such an account. The disagreement in the literature about what sorts of judgments exhibit IEM, as well as the necessary background conditions required for a judgment to exhibit IEM, suggest that IEM is not a unitary epistemic property possessed by some judgments and lacked by others. Rather there are different kinds of IEM perhaps resulting from differences in the way in which the judgment was acquired. Nonetheless, some de se judgments, such as those involving the self-​ascription of phenomenal states like (P), seem to exhibit the strongest form of IEM: logical IEM, and so the question of whether this is indicative of a special epistemic feature of such judgments deserves further investigation.

6  Conclusion In this chapter we have attempted to highlight the distinctive role that de se attitudes play in cognition and action.We have also considered the question of whether de se judgments possess a distinctive epistemic property, namely immunity to error through misidentification. We take the case against the Traditional Theory of Attitudes to be a strong one: the role that contents play in cognition and action explanation requires us to adopt contents that are either non-​absolute or unshareable.We have sketched different ways in which the Traditional Theory can be modified so as to accommodate de se attitudes. We conclude that there are compelling reasons for giving de se attitudes a prominent place in our best theory of mind.

477

Stephan Torre and Clas Weber

Further Reading Cappelen, H. and J. Dever (2013). The Inessential Indexical: On the Philosophical Insignificance of Perspective and the First Person. Oxford: Oxford University Press. Feit, N. (2008). Belief about the Self: A Defense of the Property Theory of Content. Oxford: Oxford University Press. Lewis, D. (1979). Attitudes de dicto and de se. The Philosophical Review 88, 513–​543. McGlynn, A. (2016). Immunity to error through misidentification and the epistemology of de se thought. In M. García-​Carpintero and S. Torre (Eds.), About Oneself:  De Se Thought and Communication, pp. 25–​55. Oxford: Oxford University Press. Ninan, D . (2016). What is the problem of de se attitudes? In M. García-​Carpintero and S. Torre (Eds.), About Oneself: De Se Thought and Communication, pp. 86–​120. Oxford: Oxford University Press. Perry, J. (1979). The problem of the essential indexical. Noûs 13, 26–​49. Schwarz, W. (2017). Diachronic norms for self-​locating beliefs. Ergo 4, 709–​738. Torre, S. and C. Weber (2019). De se puzzles and Frege puzzles. Inquiry. https://​doi.org/​10.1080/​ 0020174X.2019.1698455.

Notes 1 If one does not want to count contingency as a form of indexicality, one should take the quantifier to range over contexts within the same world. 2 In the following, we will mostly just focus on de se attitudes. 3 See (Perry 1979) for this suggestion; see (Chalmers 2011) for the claim that this reduction fails and that we need primitive experiential demonstratives (or at least an orientation) to deal with ‘two tubes cases’ (Austin 1990). 4 Chalmers (2011), in relation to the concern from note 3, maintains that centers need to include further parameters. 5 Lewis (1979) explicitly mentions the relation corresponding to belief, i.e. that of self-​ascribing a relativistic content. Relations corresponding to other types of attitudes can be understood analogously. For instance, the relation of hoping is that relation which holds between a subject and the set of centered worlds, such that the center-​individual is the way the subject hopes to be. 6 Perry uses the term ‘belief state’ for what we are here calling ‘guises’. 7 Here, we have indicated propositional guises by corresponding English sentences, and traditional contents by a that-​clause. It is now doubtful whether contents thus construed still count as absolute (Weber 2016). 8 As the label indicates, an account along these lines has been proposed by Russell (1912); and a similar approach has been advocated by Stalnaker (1999). However, neither Russell nor Stalnaker explicitly endorse all four claims. 9 The indexical that-​clause represents the belief ’s relativistic primary intension, the non-​indexical that-​clause represents its absolute secondary intension. 10 This similarity has been noted by Stalnaker (2008) and Kwon (2017). 11 Stalnaker (2008) has questioned whether the god’s ignorance is compatible with their role as agents. But it is not essential for the argument that the gods are agents. All the argument really requires is that the gods have a spatio-​temporal location, e.g. by possessing a physical body. 12 Lewis writes: “perhaps … they have an equally perfect view of their world, and hence cannot identify the perspective from which they view it” (Lewis 1979, 521). 13 Compare the scrutability argument in Torre and Weber (2019). 14 For instance, Lewis takes the two gods case to be compatible with both Relativism and Guises Absolutism; he opts for Relativism because of its greater simplicity (Lewis 1979, 537–​538). 15 Lewis (1979, 522–​524) has argued that haecceitistic information would not allow the gods to resolves their ignorance. However, given the assumptions of Russellian Absolutism, in particular the gods’ a priori acquaintance with their haecceity, such information would indeed enable them to know their location. 16 Guises Absolutists, such as Salmon (1986), describe the situation differently; on their view, the subject is aware of the identity, just not under the relevant guise. 17 See Torre and Weber (2019) for further details. 18 Here we assume that initially Perry and Susan share the de re belief of Perry that he is making a mess. Both may express this belief as “That guy is making a mess”. Alternatively, we could assume that initially Perry and Susan share the de dicto belief that Perry is making a mess and then Perry comes to believe a new content when he realizes that he himself is making a mess. 19 For a discussion of various candidate contents for C2 and why they are implausible, see Torre and Weber (2019, section 3).

478

What is Special about De Se Attitudes? 20 Examples include García-​Carpintero (2017), Lima (2018), Ninan (2016, this volume), Torre (2018), Valente (2018). 21 Cappelen and Dever take the thesis at issue to be “Impersonal action rationalizations (IAR) are necessarily incomplete because of a missing indexical component”. Cappelen and Dever (2013, 30) provide a number of quotes that are plausibly interpreted as endorsing (AE1). 22 García-​Carpintero (2017) provides compelling arguments for a similar conclusion, arguing that cases like the bear attack and the messy shopper provide evidence for a “de se effect” resulting from an application of Mill’s Method of Difference to KDP cases: why did Al curl up into a ball and Betty wave her arms? De se attitudes are part of the folk-​psychological explanation for the difference in behavior. García-​Carpintero’s conclusion is also compatible with the existence of possible action explanations that do not appeal to de se attitudes. 23 As Ninan (2016, 102) notes, it is “notoriously difficult to explain what it is for ‘other things to be equal’ ”; however, Ninan offers a partial list of conditions under which other things are not equal. 24 A more thorough and nuanced version of the argument can be found in (Ninan, this volume). 25 Or at least a proposition that is truth-​conditionally equivalent to the proposition that Al is being chased by a bear. 26 Although we consider a challenge to this claim below. 27 Of course how ‘ways’ are understood in this characterization of IEM will be important.There is disagreement over this and our use of this term is meant to remain neutral with respect to different conditions under which a judgment exhibits IEM (see Coliva 2006; Wright 2012; Recanati 2012a). 28 In response to such deviant wiring cases, many in the literature follow Shoemaker in distinguishing between de facto IEM and logical IEM. IEM is de facto if it is due to contingent features of the circumstances. One could respond to such deviant wiring cases by claiming that judgments arrived at through proprioception are de facto IEM rather than logical IEM. Unless more is said to spell out what de facto IEM amounts to, the distinction runs the risk of undermining the epistemic authority of IEM. ‘Immunity’ is obviously a modal claim: that one will not arrive at a certain kind of error in a range of possibilities. The objector to IEM denies this by pointing to a possibility in which the subject will arrive at an error of misidentification. What principled restriction on possibilities rules out such deviate wiring cases as relevant for assessing IEM yet upholds IEM as an epistemically significant property of judgments? See McGlynn (2016) for one attempt at providing a principled distinction between logical and non-​logical IEM in terms of safety from error. 29 Others who claim that the exhibiting of IEM by de se judgments derives from distinctive features of the account of de se attitudes include Higginbotham (2003) and García-​Carpintero (2018), both of whom claim that IEM of de se judgments results from their token-​reflexive nature. 30 Our used of ‘based on’ is purposefully vague.There is disagreement about what it is for a singular judgment to be based on an identity statement: some claim that it ought to be part of the subject’s grounds whereas others claim it is sufficient for an identity statement to be presupposed by the judgment. Coliva (2006) claims that this distinction motivates the classification of two different varieties of IEM. 31 A challenge to the Simple Explanation is another type of error through misidentification pointed out by James Pryor (1999): Which-​misidentification. Errors of Which-​misidentification don’t seem to be based on identity statements, but rather based on existential statements (“There is a skunk in my backyard”) that are falsely witnessed (“That animal is a skunk”). However, the challenge to the Simple Explanation raised by the phenomenon of Which-​misidentification seems to be merely classificatory: either one can deny that errors of Which-​misidentification are the same sort of phenomenon as errors of classic de re-​misidentification and insist that the Simple Explanation explains the latter only. Or one could grant that Pryor has highlighted a different way in which the same phenomenon can arise and expand the Simple Explanation to claim that judgments exhibit IEM just in case they fail to be based on an identity statement or existential statements.

References Austin, D. F. (1990). What’s the Meaning of ‘This’? Ithaca, NY: Cornell University Press. Boër, S. and W. Lycan (1980). Who, me? The Philosophical Review 89, 427–​466. Braun, D. (2002). Cognitive significance, attitude ascriptions, and ways of believing propositions. Philosophical Studies 1–​2(108),  65–​81. Cappelen, H. and J. Dever (2013). The Inessential Indexical: On the Philosophical Insignificance of Perspective and the First Person. Oxford: Oxford University Press. Cappelen, H. and J. Dever (ms). Action without the first person perspective. Unpublished Manuscript, 1–​27. Chalmers, D. (2002). The components of content. In D. Chalmers (Ed.), Philosophy of Mind:  Classical and Contemporary Readings, pp. 608–​633. Oxford: Oxford University Press. Chalmers, D. (2011). Frege’s puzzle and the objects of credence. Mind 120(4), 587–​635. Coliva, A. 2006. Error through misidentification: Some varieties. Journal of Philosophy 103: 403–​425.

479

Stephan Torre and Clas Weber Devitt, M. (2013). The myth of the problematic de se. In A. Capone and N. Feit (Eds.), Attitudes De Se: Linguistics, Epistemology, Metaphysics. Stanford, CA: CSLI Publications. Egan, A. (2007). Epistemic modals, relativism and assertion. Philosophical Studies 133, 1–​22. Evans, G. (1982). Varieties of Reference. Oxford: Oxford University Press. Frege, G. (1892). Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik 100, 25–​50. García-​Carpintero, M. (2017). The philosophical significance of the de se. Inquiry 60(3), 253–​276. García-​ Carpintero, M. (2018). De se thoughts and immunity to error through misidentification. Synthese 195(8): 3311–​3333. Gibbard, A. (2012). Meaning and Normativity. Oxford: Oxford University Press. Higginbotham, J. (2003). Remembering, imagining, and the first person. In A. Barber (Ed.), Epistemology of Language, pp. 496–​533. Oxford: Oxford University Press. Jackson, F. (1982). Epiphenomenal qualia. Philosophical Quarterly 32(127), 127–​136. Kaplan, D. (1989). Demonstratives. In J. Almog, J. Perry, and H.Wettstein (Eds.), Themes from Kaplan, pp. 481–​614. New York: Oxford University Press. Kwon, H. (2010). Self-​Identification and Self-​Knowledge. Phd thesis, MIT. Kwon, H. (2017). Mary and the two gods:  Trying out an ability hypothesis. The Philosophical Review 126(2), 191–​217. Lewis, D. (1979). Attitudes de dicto and de se. The Philosophical Review 88, 513–​543. Lewis, D. (1997). Naming the colours. Australasian Journal of Philosophy 75, 325–​342. Lima, J. (2018). Indexicality and action: Why we need indexical beliefs to motivate intentional actions. Inquiry. DOI: 10.1080/​0020174X.2018.1487881. Magidor, O. (2015). The myth of the de se. Philosophical Perspective 29, 249–​283. McGlynn, A. (2016). Immunity to error through misidentification and the epistemology of de se thought. In M. García-​Carpintero and S. Torre (Eds.), About Oneself:  De Se Thought and Communication, pp. 25–​55. Oxford: Oxford University Press. Meacham, C. (2010). Unravelling the tangled web:  Continuity, internalism, non-​uniqueness and self-​locating beliefs. In T. Szabó-​Gendler and J. Hawthorne (Eds.), Oxford Studies in Epistemology, volume 3, pp. 86–​125. Oxford: Oxford University Press. Morgan, D. (2012). Immunity to error through misidentification:  What does it tell us about the de se. In S. Prosser and F. Recanati (Eds.), Immunity to Error through Misidentification:  New Essays, pp. 104–​123. Cambridge: Cambridge University Press. Ninan, D. (2016). What is the problem of de se attitudes? In M. García-​Carpintero and S. Torre (Eds.), About Oneself: De Se Attitudes and Communication, pp. 86–​120. Oxford: Oxford University Press. Perry, J. (1977). Frege on demonstratives. The Philosophical Review 86(4), 474–​497. Perry, J. (1979). The problem of the essential indexical. Noûs 13, 26–​49. Perry, J. (2001). Reference and Reflexivity. Stanford, CA: CSLI Publications. Perry, J. (2006). Stalnaker on indexical belief. In J. J. Thomson and A. Byrne (Eds.), Content and Modality: Themes from the Philosophy of Robert Stalnaker, pp. 204–​221. New York: Oxford University Press. Pryor, J. (1999). Immunity to error through misidentification. Philosophical Topics 26: 271–​304. Recanati, F. (2012a). Immunity to error through misidentification:  What it is and where it comes from. In S. Prosser and F. Recanati (Eds.), Immunity to Error through Misidentification:  New Essays, pp. 180–​201. Cambridge: Cambridge University Press. Recanati, F. (2012b). Mental Files. Oxford: Oxford University Press. Recanati, F. (2016). Mental Files in Flux. Oxford: Oxford University Press. Russell, B. (1912). The Problems of Philosophy. London: Williams and Norgate. 9th impression, with appendix, published Oxford: Oxford University Press, 1980. Salmon, N. (1986). Frege’s Puzzle. Cambridge, MA: MIT Press. Schwarz, W. (2012). Changing minds in a changing world. Philosophical Studies 159(2), 219–​239. Shoemaker, S. (1968). Self-​knowledge and self-​awareness. Journal of Philosophy, 65, 555–​567. Soames, S. (1987). Direct reference, propositional attitudes, and semantic content. Philosophical Topics 15, 47–​87. Spencer, C. (2007). Is there a problem of the essential indexical? In M. O’Rourke and C. Washington (Eds.), Situating Semantics: Essays on the Philosophy of John Perry, pp. 179–​197. Cambridge: MIT Press. Stalnaker, R. C. (1981). Indexical belief. Synthese 49, 129–​151. Stalnaker, R. C. (1999). Context and Content: Essays on Intentionality in Speech and Thought. Oxford: Oxford University Press. Stalnaker, R. C. (2008). Our Knowledge of the Internal World. New York: Oxford University Press. Stalnaker, R. C. (2016). Modelling a perspective on the world. In M. García-​Carpintero and S.Torre (Eds.), About Oneself: De Se Content and Communication, pp. 121–​139. Oxford: Oxford University Press. Stanley, J. (2011). Know How. Oxford: Oxford University Press.

480

What is Special about De Se Attitudes? Titelbaum, M. G. (2016). Self-​locating credences. In A. Hájek and C. Hitchcock (Eds.), The Oxford Handbook of Probability and Philosophy, pp. 666–​680. Oxford: Oxford University Press. Torre, S. (2018). In defense of de se content. Philosophy and Phenomenological Research 97(1), 172–​189. Torre, S. and C. Weber (2019). De se puzzles and Frege puzzles. Inquiry. https://​doi.org/​10.1080/​ 0020174X.2019.1698455. Valente, M. (2018) What is special about indexical attitudes? Inquiry, 61(7), 692–​ 712. DOI:  10.1080/​ 0020174X.2017.1402701. Weber, C. (2013). Centered communication. Philosophical Studies 166(S), 205–​223. Weber, C. (2015). Indexical beliefs and communication: Against Stalnaker on self-​location. Philosophical and Phenomenological Research 90(3), 640–​663. Weber, C. (2016). Being at the centre: Self-​locating in thought and language. In M. García-​Carpintero and S. Torre (Eds.), About Oneself: De Se Thought and Communication. Oxford: Oxford University Press. Wright, C. (2012). Reflections on François Recanati’s ‘immunity to error through misidentification: What it is and where it comes from’. In S. Prosser and F. Recanati (Eds.), Immunity to Error through Misidentification: New Essays, pp. 247–​280. Cambridge: Cambridge University Press. Zimmermann, T. E. (1991). Kontextabhängigkeit. In A. von Stechow and D. Wunderlich (Eds.), Semantik/​ Semantics: Ein internationales Handbuch der zeitgenössischen Forschung, pp. 156–​229. Berlin/​New York: Walter de Gruyter.

481

36 DE SE ATTITUDES AND ACTION Dilip Ninan

1  Introduction Actions can often be explained by beliefs and desires. Why is Mary flapping her arms like that? Because she wants Sam to laugh, and she believes that he will laugh if she flaps her arms like that.Very often such explanations appeal, as this one does, to a de se belief, a belief that one would naturally express using the first-​person pronoun. Mary, for example, would presumably express her belief by saying, Sam will laugh if I flap my arms like this. A persistent theme in the literature on de se attitudes is that such attitudes enjoy a special connection to action. Is this right? If so, what is the nature of this special connection? Before we get to that, we should first say what a de se attitude is. Let’s start with the more general category of indexical attitudes. To a first approximation, an indexical attitude is one that could be expressed or reported using an appropriate sentence containing an indexical such as I, here, or now. This is not intended as a definition, but merely as a gesture towards the class of attitudes that constitutes our object of study. My belief that I’m over five-​feet tall is an indexical belief, as is my belief that it is now raining. My desire that I avoid harm is an indexical desire, as is my desire that it stop raining in the next few minutes. The indexical attitudes that tend to receive the most attention are of two kinds:  (i) de se or self-​locating attitudes, which are attitudes that could be expressed or reported using an appropriate sentence containing a first-​person pronoun such as I, me, or my; and (ii) de nunc or temporally self-​locating attitudes, which are attitudes that could be expressed or reported using an appropriate sentence containing a temporal adverbial such as now. For simplicity, I shall focus mostly on de se (i.e. first-​person) attitudes in what follows, though much of what I say may well apply, mutatis mutandis, to the temporal case. In what follows, I consider three proposals for what the special connection between action and de se attitudes is. In section 2, I consider the claim that de se attitudes are, in a sense to be made precise, essential to the explanation of action. In sections 3–​4, I consider one of Perry’s classic examples, along with his claim that such examples pose a problem for the doctrine of propositions, a traditional way of thinking about attitudes and their contents. For reasons I shall discuss, I don’t think that either of these approaches reveals the distinctive connection between actions and de se attitudes. In sections 5–​8, I consider a third proposal. According to this proposal, the doctrine of propositions runs into trouble in connection with cases in which two agents can, in a certain sense, be said to “have all the same attitudes”, but in which they are rationally motivated to perform different actions. Section 9 482

De Se Attitudes and Action

closes with a brief discussion of how two prominent theories (Lewis’s and Perry’s) accommodate the feature of de se attitudes that gives rise to this problem.

2  Essentiality Consider the idea that de se attitudes are essential to the explanation of action.This may be understood as implying that, for any action A, there is some de se attitude that figures in the explanation of A. You can’t explain an action without appealing to a de se attitude.1 This might seem like an overly-​strong claim, but one way to try to argue for it would be to argue that, according to the best regimentation of such explanations, they always advert to a de se attitude. For example, perhaps the canonical form of such explanations corresponds to the following schema of practical reason: If I ϕ, then p. I desire that p. So, I should ϕ. The crucial idea here is that such explanations always rely on a belief that the agent could express by means of a ‘means-​end’ conditional, If I ϕ, then p. And perhaps it is the nature of such beliefs that they are de se beliefs, the agent conceiving of the action as one that she might perform. This broad idea strikes me as interesting, and not wholly implausible, but I won’t pursue it further here. The reason is that the special connection between actions and de se attitudes is often thought to tell us something interesting about the nature of attitude content: about the objects of belief and desire. After all, it is questions about the contents of attitudes that dominate the literature on the de se (Frege 1956; Perry 1977, 1979; Lewis 1979; Evans 1981). But consider the claim that for any action A, a de se attitude figures in the explanation of A. Suppose it is true. What, if anything, follows concerning the nature of the objects of belief and desire? Would it show that the objects of de se attitudes are private, as Frege thought? Or would it show that the contents of such attitudes are properties, as Lewis thought? The answer to these questions is unclear, and the same goes if we suppose instead that the ‘essentiality thesis’ is false. There is simply no obvious connection between this essentiality thesis and the questions about content that dominate the literature on the de se.Whether or not a certain type of attitude always figures in a certain type of explanatory context does not obviously imply anything about the contents of such attitudes. Note, for example, that if such folk-​psychological explanations of action are best regimented in the manner described above, then they also always include a conditional belief (If I  ϕ, then p). But reflecting on that fact would seem to tell us little about the contents of conditional beliefs.

3  The Doctrine of Propositions The idea that there is a special connection between de se attitudes and action that reveals something important about the nature of attitude content is often associated with John Perry’s seminal work on de se attitudes, or self-​locating attitudes, as he styled them (Perry 1977, 1979). Perry (1979) took such attitudes to pose a problem for what he called the doctrine of propositions, a series of claims about attitudes and their contents that he associated with traditional theories of attitudes.2 Perry’s doctrine consists of three theses, which I render as follows. Thesis (I): Attitude relations, such as the relation of believing, are two-​place relations between an agent and an abstract object called a proposition.3 Given an attitude relation R and a proposition p, we have the property of bearing R to p. Such properties are attitudes, i.e. attitude types. So, for example, the property of bearing the relation of believing to the proposition that it is raining—​the property of believing that it is raining—​is an attitude; in particular, it is a belief, a belief type. If B is the property of believing p, then we say that p is the content of B.4 483

Dilip Ninan

Thesis (II): Propositions are absolute, i.e. one and the same proposition cannot be true for one person and false for another. In this respect, propositional truth differs from sentential truth. The sentence I am Italian might be true as uttered by Laura, false as uttered by Lakshmi. According to the doctrine of propositions, propositions are not like that; if proposition p is true for person x, then it is true for all other persons as well. Thesis (III): Attitudes are fine-​grained. This idea is familiar, though it is not easy to express it in a precise and general way. But I take the following to be a consequence of this idea: If a rational agent x understands sentences ϕ and ϕ′ and accepts ϕ and rejects ϕ′, then the belief that (were she to have it) x could express by uttering ϕ is distinct from the belief that (were she to have it) x could express by uttering ϕ′.5 Since a rational agent can understand the sentences Mark Twain was a writer and Samuel Clemens was a writer, and accept the former while rejecting the latter, it follows that from this thesis that the belief B he could express by uttering the first sentence is distinct from the belief B′ he could express by uttering the second (even though Mark Twain is Samuel Clemens). Note that when you put this part of the doctrine together with the first part of the doctrine you get the result that beliefs B and B′ differ in their content. For if B and B′ are distinct beliefs, and if beliefs are individuated (solely) by their content, then B and B′ differ in their content. Thus, the doctrine of propositions is ‘Fregean’, in a broad sense of this term. Those are the three components of the doctrine of propositions, as I shall understand it. Note that my version of the doctrine of propositions has been conspicuously silent on issues concerning the proper analysis of attitude ascriptions. This is because our focus is on the attitudes—​the mental states—​ themselves, not on how we attribute them to ourselves and others. But of course in order to pursue our topic of study, we will need a way of picking out attitudes and so we cannot avoid the topic of attitude ascriptions entirely. But we can skirt some controversial questions by ascribing attitudes not by using ordinary attitude ascriptions, but by using sentences like this: (1) Al has a belief that he could express by saying, I am being chased by a bear. (2) Betty has a desire that she could report by saying, I want to visit Mark Twain’s grave. Adopting this somewhat indirect manner of speaking allows us to avoid some of the controversies surrounding the semantics and pragmatics of attitude ascriptions. Furthermore, if what an ordinary attitude ascription conveys about a particular subject’s state of mind often depends on the context in which it is used, it may aid clarity to describe attitudes using sentences like the ones displayed above, rather than by using ordinary attitude ascriptions.

4  The Specification Problem What problem do de se attitudes pose for the doctrine of propositions? Much of the discussion in Perry (1979) concerns cases like the following (I paraphrase here, rather than quote): The messy shopper Perry once followed a trail of sugar along a supermarket floor, looking for the shopper with the torn sack to tell him that he was making a mess.With each trip around the store, the trail became thicker, but there was no sign of the messy shopper. Finally, Perry realized that he was the shopper with the torn sack that he was trying to catch. Having realized this, Perry of course stopped following the trail and turned the torn sack upright. Perry 1979, 33 484

De Se Attitudes and Action

The case reveals some sort of connection between action and de se attitudes, for it is Perry’s coming to the de se realization that he is the one making the mess that leads to his change in behavior. But what problem does the case pose for the doctrine of propositions? According to Perry, the trouble arises because the doctrine of propositions is committed to saying that Perry’s coming to believe that he is the one making the mess is his coming to believe some absolute proposition. But it turns out to be rather difficult to say just which absolute proposition this is. And if it is difficult to say just which absolute proposition Perry comes to believe when he comes to believe that he is the one making a mess, that might seem to cast doubt on the idea, central to the doctrine of propositions, that beliefs are solely individuated by absolute propositions. Why is it so difficult to say just which absolute proposition it is that Perry comes to believe when he comes to believe that he is the one making the mess? Well, as Perry observes, there seems to be no purely qualitative property F such that Perry’s believing that he is the one making the mess is his believing that the F is the one making the mess. For any choice of F, we can imagine Perry believing that he is the one making the mess without believing that the F is the one making the mess, since Perry may not realize that he is the F. Thus, the change in Perry’s behavior cannot be explained by saying that he came to believe that the F is the one making the mess, for some qualitative property F. Furthermore, it is not clear that we can characterize Perry’s realization by saying that he came to believe the ‘singular proposition’ that John Perry is the one making the mess. For given standard accounts concerning what it takes to believe a singular proposition, Perry could come to believe this proposition without having the de se belief to the effect that he is the one making a mess. Perry might, for example, see a reflection of the messy shopper in a mirror, fail to realize that he is seeing himself, and come to believe that (as he would put) that man is making a mess. He would thereby believe the singular proposition that John Perry was making a mess (on standard accounts), but would still fail to believe de se that he was making a mess. So what Perry learned cannot simply be characterized by that singular proposition (according to this line of argument). Perry can be seen as issuing a specification challenge to the advocate of the doctrine of propositions. If what it is for Perry to believe that he is the one making the mess is for him to believe an absolute proposition, then the advocate of the doctrine of propositions ought be in position to specify that proposition, i.e. to tell us which proposition it is that Perry comes to believe. Now, a somewhat murky question is this: what must one do in order to count as meeting the specification challenge? For suppose the advocate of the doctrine of propositions just said this: “The proposition we are looking for is simply that absolute proposition that Perry could have expressed by saying, I am the one making the mess”. Isn’t that a perfectly good way to specify a proposition?6 If it is, then isn’t it trivial to meet the specification challenge? If, on the other hand, this way of specifying a proposition is in some sense illegitimate, then what criteria must a description of a proposition meet in order to count as adequate answer to the specification challenge? Fortunately, we can sidestep these difficult questions, and focus on a simpler observation, one that a number philosophers have made.7 The observation is that, although Perry’s example and argument concern a de se belief, the de se-​ness of that belief seems to play no essential role in the argument. Consider, for example, Cappelen and Dever’s structurally similar case of ‘messy Superman’: Messy Superman Pushing my cart down the aisle I was looking for Clark Kent to tell him he was making a mess. I kept passing by Superman, but couldn’t find Clark Kent. Finally, I realized, Superman was Clark Kent. I believed at the outset that Clark Kent was making a mess … But I didn’t believe that Superman was making a mess. That seems to be something that I  came to believe. And when I came to believe that, I stopped looking around and I told Superman to clean up after himself. My change in beliefs seems to explain my change in behavior. Cappelen and Dever 2013, 33 485

Dilip Ninan

The similarity between this case and Perry’s de se case makes one wonder what problem the de se case raises that the Superman case does not. One can bolster this thought by noting that one can use the Superman case to construct an argument against the doctrine of propositions that parallels Perry’s argument (Ninan 2016, 95).Thus, the case of messy Superman suggests that de dicto attitudes also raise a specification challenge for the doctrine of propositions, a challenge does not appear to be substantially different from the one raised by de se attitudes.8 We will have to look elsewhere if we want to find a special connection between action and de se attitudes.

5  Expanding the Doctrine of Propositions Perry (1977) briefly discusses another type of case: When you and I both apprehend the thought that I am about to be attacked by a bear, we behave differently. I roll up in a ball, you run to get help. Perry 1977, 23 Perry uses this case not to press a problem for the doctrine of propositions, but as a means of illustrating his preferred theory. But I think that such cases do pose a problem for the doctrine of propositions, at least if two other plausible theses about the nature of attitudes are accepted.9 And I think that the resulting problem is indeed particular to the de se. I shall try to substantiate these claims in what follows, and then try to say what it is about de se attitudes that gives rise to this problem.10 Let’s start with the two additional theses about the nature of attitudes. The first thesis concerns the role of attitudes in the explanation and prediction of action. Let me build up to the first thesis by discussing a version of Perry’s ‘bear attack’ case. Imagine that Al and Betty are hiking in the woods when a bear begins to chase Al. Al and Betty both realize that this is what is happening, and they both want the bear to leave. Neither suffers from any sort of identity confusion; the case is not a ‘Frege case’. Let’s focus on Al’s attitudes for the moment, and let us suppose that Al has the following attitudes: I’m being chased by a bear. If I roll up in a ball, the bear will leave.  B I want the bear to leave.  D Here, we pick out Al’s attitudes via the sentences he would use to express or report them. We use B to denote Al’s de se belief that if he rolls up in a ball, the bear will leave, and we use D to denote Al’s desire that the bear leave. Now, given that Al has these attitudes, what would we expect Al to be motivated to do? Well, I take it that we’d expect Al to be motivated to roll up in a ball, all else being equal. Furthermore, I take it that it is Al’s having B and D that would motivate Al to do this. It isn’t simply that Al happens to have these attitudes and Al happens to be motivated to roll up in a ball, with no connection between these facts. It’s his having these attitudes that would motivate him to act in that way. Note that our prediction that this pattern of attitudes will motivate Al to roll up in a ball (if all else is equal) does not reflect any special knowledge we have of Al, for we have no such special knowledge; Al is essentially an arbitrary name for someone who has a certain pattern of attitudes.Thus, our prediction about what Al would do is presumably based on a generalization that links having certain attitudes to performing certain actions. Since our prediction is based on the fact that Al has belief B and desire D, the relevant generalization could presumably be expressed as follows: (⋆) Necessarily, for all individuals y, if y has B and D, then, if all else is equal, y’s having B and D will motivate y to roll up in a ball. 486

De Se Attitudes and Action

Our first additional thesis generalizes this point: Thesis (IV): Law-​like generalizations that link attitudes with actions play a central role in the explanation and prediction of rational action. For simplicity, I shall focus on the role of such generalizations in predicting rational action. And I shall interpret Thesis (IV) as implying the following: The normal way of arriving at the prediction that agent x will perform action ϕ is as follows: • The predictor knows a particular claim: (a) Agent x has attitudes A1, … , An. • The predictor knows a law-​like generalization: (b) Necessarily, for all y, if y has attitudes A1, … , An, then, if all else is equal, y’s having attitudes A1, … , An will motivate y to perform action ϕ. • And from these two things, the predictor infers: (c) If all else is equal, x’s having attitudes A1, … , An will motivate x to perform ϕ. The law-​like generalizations in question contain a wide-​scope necessity operator. The type of necessity is something like nomological necessity; the operator may be understood as quantifying over all the psychologically possible worlds. Note also that these law-​like generalizations are ‘ceteris paribus laws’: the consequent says that if all else is equal, then the agent’s attitudes will motivate her to perform the action in question. It is notoriously difficult to specify the content of these clauses in an informative way. For our purposes, it will suffice to note that all else is not equal if: the agent is suffering from some form of practical or theoretical irrationality; or the agent is weak-​willed; or an alternative action that would achieve the same end is more suitable for the agent in question. So Thesis (IV), interpreted in the way just indicated, is my first additional thesis. Now for the second. Thesis (V): Propositions are shareable or public in the following sense: if an agent x can bear attitudes towards a proposition p, then, generally speaking, so can any other agent y. The “generally speaking” qualification is needed for the following sorts of cases. Suppose p is a singular proposition about z. Then perhaps in order to entertain p, one needs to stand in a certain relation of acquaintance to z. If x stands in that relation to z, but y does not, then x may be able to entertain p while y cannot. Another type of case is where, in order to entertain p, one must possess a certain concept c. Then if x possesses c but y does not, x may be able to entertain p, while y cannot. Neither of these qualifications will be particularly relevant in what follows.

6  The Problem Let the expanded doctrine refer to the conjunction of the doctrine of propositions with these two additional theses, i.e. the expanded doctrine consists of Theses (I)–​(V). I claim that de se attitudes pose a distinctive problem for the expanded doctrine. The problem takes the form of a reductio against the expanded doctrine. Return to our bear attack scenario. Assume, as before, that Al has attitudes B and D: Al’s attitudes If I roll up in a ball, the bear will leave.  B I want the bear to leave.  D 487

Dilip Ninan

According to the Thesis (I) of the expanded doctrine, belief B is the property of believing p, for some proposition p. According to Thesis (V), proposition p is shareable, which means Betty can believe it (assume Betty satisfies the requisite acquaintance and concept-​possession requirements). But if Betty can believe p, she can have belief B, since belief B is just the property of believing p. The same goes for desire D: Betty can have desire D. Let us suppose that she has belief B and desire D. What did we just suppose? What is it for Betty to have these attitudes? According to Thesis (II), propositions are absolute, i.e. propositions are true or false simpliciter. This means that we can also say that beliefs are true or false simpliciter: a belief is true iff its content is true. And we can say that desires are satisfied or not satisfied simpliciter: a desire is satisfied iff its content is true. Since B is a belief that Al can express by saying, If I curl up into a ball, the bear will leave, B is presumably true iff: if Al curls up into a ball, the bear will leave. Thus, for Betty to have belief B is for Betty to have a belief that is true iff: if Al curls up into a ball, the bear will leave. Parallel reasoning shows that for Betty to have desire D is for Betty to have a desire that is satisfied iff the bear leaves. But knowing the conditions under which belief B is true doesn’t tell us exactly what it is for Betty to have belief B. To see this, we can consider some beliefs that Betty could have that have these truth-​conditions.We pick out the beliefs via the sentences Betty could use to express them (were she to have them): (i) If you curl up into a ball, the bear will leave. (speaking to Al) (ii) If Al curls up into a ball, the bear will leave. (iii) If the actual man in the red hat curls up into a ball, the bear will leave. … Given Thesis (III) (fine-​grainedness) of the expanded doctrine, these are all distinct beliefs. Betty could have one without having the others. So which one is B? From the point of view of the expanded doctrine, I  think the most plausible answer to this  question is that what it is for Betty to have B is for Betty to have belief (i)  on the above list, the you-​belief. I will start off by assuming this, but my argument does not actually depend on this assumption. And the fact that it doesn’t depend on this assumption turns out to be an extremely important fact, one that reveals something distinctive about de se attitudes and their relationship to action. I return to this point below, but for the moment, let us proceed with the assumption that what it is for Betty to have B is for Betty to have belief (i) on the above list, the you-​belief. I will also assume that for Betty to have D is for her to have a desire she could report by saying, I want the bear to leave. So we assume that Betty has attitudes B and D: Betty’s attitudes If you curl up into a ball, the bear will leave.  B I want the bear to leave.  D Note that here we’ve described B and D via the sentences Betty would use to express and report them. We introduced B and D above by describing them via the sentences that Al would use to express and report them. Now recall our generalization (⋆), the generalization we used to arrive at the prediction that, if all else is equal, Al’s attitudes will motivate him to roll up in a ball: (⋆) Necessarily, for all individuals y, if y has B and D, then, if all else is equal, y’s having B and D will motivate y to roll up into a ball. 488

De Se Attitudes and Action

I take it that the expanded doctrine is committed to (⋆). For we were able to predict, based on the fact that Al has attitudes B and D that if all else is equal, Al’s having B and D would motivate him to roll up in a ball. According to Thesis (IV), such predictions are standardly arrived at by inferring them from a particular fact about the agent’s attitudes and a law-​like generalization linking attitudes to actions. And given the nature of the particular fact (Al has attitudes B and D), what could that generalization be other than (⋆)? So I assume that the expanded doctrine is committed to (⋆), given the facts about the bear attack case. But (⋆) is false. For (⋆) implies that in every world compatible with our description of the bear attack scenario, the following holds: If Betty has belief B and desire D, then if all else is equal, Betty’s having B and D will motivate her to roll up in a ball. But this is implausible. For given what it is for Betty to have B and D, the above claim is equivalent to the following: If Betty has a belief she could express to Al by saying, If you roll up in a ball, the bear will leave, and Betty wants the bear to leave, then if all else is equal, Betty’s having this belief and this desire will motivate her to roll up in a ball. But it is easy to imagine the bear attack scenario in such a way that Betty has this pattern of attitudes, all else is equal, but in which Betty’s having these attitudes does not motivate her to roll up in a ball. Indeed, in any normal version of the bear attack scenario, Betty’s having this pattern of attitudes will not motivate her to behave in that way. Why would she roll up in a ball simply because she thinks that if Al rolls up in a ball, the bear will leave? That would be bizarre given that she knows that she is not Al. Since the expanded doctrine is committed to (⋆), and since (⋆) is false, the expanded doctrine is likewise false. Before moving on, let me return to the assumption I made earlier about what it is for Betty to have belief B. I assumed above that for Betty to have B is for Betty to have a belief that she could express by saying to Al, If you roll up in a ball, the bear will leave. But, as I noted above, the expanded doctrine per se isn’t committed to that; it is only committed to the claim that what it is for Betty to have B is for Betty to have a belief that is true iff: if Al rolls up in a ball, the bear will leave. And, as I noted above, there are many beliefs like that: (i) If you roll up in a ball, the bear will leave. (speaking to Al) (ii) If Al rolls up in a ball, the bear will leave. (iii) If the actual man in the red hat rolls up in a ball, the bear will leave. … But although I  was assuming that what it is for Betty to have B is for her to have the first of these beliefs, the foregoing argument against the expanded doctrine doesn’t actually depend on that assumption. To see this, pick any belief on this list—​pick any belief that Betty could have that is true iff: if Al rolls up in a ball, the bear will leave. If you combine the fact that Betty has that belief with the fact that Betty wants the bear to leave, we’ll run into the same problem. For that pattern of attitudes would not normally motivate Betty to roll up in a ball, and so we will still have our counterexample to (⋆). Betty’s believing that if Al rolls up in a ball, the bear will leave will not, in conjunction with her desire that the bear leave, motivate her to roll up in a ball. Betty’s believing that if the actual man in the red hat rolls up in a ball, the bear will leave will not, in conjunction with her desire, motivate her to roll up in a ball. And so on and so forth.11 So our argument against the expanded doctrine does not depend on any highly specific assumption about what it is for Betty to have belief B. It relies only on the assumption that for Betty to have B 489

Dilip Ninan

is for Betty to have a belief that is true iff: if Al rolls up in a ball, the bear will leave. As I said, this fact about our argument turns out to be quite important, as we shall see below.

7  Is the Problem Specific to the De Se? My argument against the expanded doctrine relied on claims about Al’s de se belief that if he rolled up in a ball, the bear would leave. But is the de se-​ness of his belief essential to this argument? Could the same problem be raised without appealing to a de se attitude at all? Let us approach these issues by imagining a dispute between two characters: the de se skeptic thinks that essentially the same problem can be raised without appealing to de se attitudes; the de se exceptionalist denies this. The de se exceptionalist thinks the de se-​ness of Al’s belief is essential to the role it plays in the above argument. It is not straightforward to argue for the skeptical claim that essentially the same problem can be raised without appealing to de se attitudes.This is because it is difficult to come up with an uncontroversial case in which an agent’s action is explained solely by her de dicto attitudes (this is connected to the issues discussed in section 2). So we shall have to proceed somewhat indirectly. Let us imagine that the dispute between the skeptic and the exceptionalist plays out as follows. In response to the above problem, the de se exceptionalist proposes an alternative theory, which I shall call the revised theory. According to the exceptionalist, the revised theory is the minimal way of revising the expanded doctrine so that it avoids the problem posed by de se attitudes, the problem revealed by the Al-​Betty case. The skeptic then tries to argue that she can run essentially the same argument against the revised theory that we ran above against the expanded doctrine. If the skeptic is right about this, then the exceptionalist should concede that the problem raised by the Al-​Betty case was not specific to the de se after all. For by her own lights, the revised theory solves whatever problem in this vicinity is raised by de se attitudes per se. So if the revised theory faces essentially the same problem, it must be de dicto attitudes that are causing the trouble this time. And in that case, the general problem here is presumably not specific to the de se. The exceptionalist’s revised theory consists of all the theses of the expanded doctrine, save for Thesis (II), the claim that the contents of attitudes are absolute. In place of Thesis (II), it opts for the following: Thesis (II′): The contents of de dicto attitudes are absolute, but the contents of de se attitudes are relative, things that vary in truth value across individuals. The revised theory is thus closely related to the property view of attitudes, the view according to which the content of an attitude is a property or a relative proposition (Loar 1976; Lewis 1979; Chisholm 1981).12 Take Al’s de se belief that he is being chased by a bear. The content of this belief, according to the revised theory, is the property of being chased by a bear. To believe this property is to have a belief that one could express by saying, I am being chased by a bear. If we say that a property is true at an individual just in case the individual has the property, then we can say that properties are things that (potentially) vary in truth value across individuals. The property of being chased by a bear is true for some people (those being chased by a bear), false for others (those not being chased). Of course, not all properties have this feature which means that the contents of de dicto attitudes may be identified with properties that do not vary in truth value across individuals. For example, take Al’s de dicto desire that the bear leave. The content of this desire may be identified with the property of being an x such that the bear leaves. This property is absolute: if Al has it, then so does Betty, and so does everyone else. Properties like this do not vary in truth value over individuals. Here is how the revised theory avoids the problem for the expanded doctrine discussed above. The problem took the form of a reductio, where one of the premises of the reductio was the claim that 490

De Se Attitudes and Action

what it is for Betty to have belief B is for her to have a belief that is true iff: if Al rolls up in a ball, the bear will leave.The revised theory blocks the reductio by rejecting this premise, for this premise is false according to the revised theory. Belief B is, by definition, the belief Al can express by saying, If I roll up in a ball, the bear will leave. According to the revised theory, the content of B is the property p of being an x such that if x rolls up in a ball, the bear will leave. For an agent x to believe this property is for x to have a belief she could express by saying, If I roll up in a ball, the bear will leave. Given Thesis (V) (shareability), Betty can believe p too, and so can have belief B. But what it is for Betty to have belief B is for her to believe that if she rolls up in a ball, the bear will leave. In that case, if we suppose that Betty has belief B and desire D, it is not at all implausible to think that, if all else is equal, that belief and desire will motivate her to roll up in a ball. Thus, the argument to the effect that (⋆) is false is blocked. So the fact that the revised theory is committed to (⋆) is no knock against it. Note that, according to the revised theory, belief types cannot be said to be true or false simpliciter. To see this, suppose otherwise, and consider the property of believing q, where q is the property of being chased by a bear. Suppose I believe q, and I am being chased by a bear. Suppose you believe q, and you are not being chased by a bear. Let B* be the property of believing q. Is B* true or false? Well, I believe that I am being chased by a bear, and I am being chased by a bear. So my belief is true. If “my belief ” picks out B*, then B* is true. But you believed that you are being chased by a bear, but you are not being chased by a bear. So your belief is false. If “your belief ” picks out B*, then B* is false, and so not true. Contradiction. What has gone wrong? According to the expanded doctrine, a belief type is true iff its content is true. This claim makes sense in the context of the expanded doctrine since contents, according to that doctrine, are absolute, true or false simpliciter. But that claim no longer make sense in the context of the revised theory since the revised theory holds that contents are relative, true or false relative to individuals. Since q, the property of being chased by a bear, is true for me, the belief type B* is true for me. Since q is false for you, B* is false for you. The contradiction is avoided by relativizing the truth-​values of belief types. Of course, we should still like to make sense of the idea that ‘my belief ’ is true and ‘your belief ’ is not. According to the revised theory, we can make sense of this by thinking of these descriptions as picking out belief tokens rather than belief types. What is a belief token? Consider the token event of my believing that I’m being chased by a bear, my believing q. We can identify token beliefs with token events of this sort. We might further identify token events like this with a pair consisting of a believer and a belief type, e.g. ⟨DN, the property of believing q⟩.13 And we could say that a token belief ⟨x, the property of believing p⟩ is true iff x has p. Thus, in our example above, my token belief is true and your token belief is false, even though their content is the same. For that content is true for me (I have property q), but false for you (you lack property q). So the exceptionalist claims, quite plausibly, that the revised theory solves the problem posed by de se attitudes, the problem revealed by the Al-​Betty case. (This is not, of course, to say that the revised theory is true, only that if it is false, it is not because of this problem.) Thus, if the skeptic could show that de dicto attitudes pose essentially the same problem for the revised theory that de se attitudes posed for the expanded doctrine, that would seem to show that the underlying problem here is not specific to the de se. Let’s consider the prospects of such an argument Start by again imagining two agents, Carl and Diane. Carl has the following attitudes: Carl’s attitudes Mark Twain is buried in Elmira.  B′ I want to visit Mark Twain’s grave.  D′ As before, we pick out Carl’s attitudes via the sentences he would use to express or report them. 491

Dilip Ninan

Now according to the revised theory, Diane can have belief B′ and desire D′. That Diane can have belief B′ follows from Theses (I) and (V) of the revised theory. From Thesis (I), it follows that there is a proposition p′ such that one has belief B′ just in case one believes p′. From Thesis (V)—​the shareability of propositions—​it follows that Diane can believe p′ (assume Diane satisfies the requisite acquaintance and concept-​possession requirements). And from that it follows that she can have belief B′. A parallel argument shows that Diane can have desire D′. But what is it for Diane to have belief B′ and desire D′? Start with the latter. Desire D′ is a de se desire. Thus, its content is a property—​presumably it is the property of visiting Mark Twain’s grave. What it is for an agent x to have D′ is for x to have a desire that x could report by saying, I want to visit Mark Twain’s grave. Thus, what it is for Diane to have D′ is for her to have a desire that she could report by saying, I want to visit Mark Twain’s grave. Belief B′, on the other hand, is a de dicto attitude, and so, according to the revised theory, its content is absolute. In that case, it makes sense to treat the belief type B′ as true or false simpliciter, and not merely true or false relative to an individual. Given that B′ is the belief Carl could express by saying, Mark Twain is buried in Elmira, B′ is presumably true iff Mark Twain is buried in Elmira. So, according to the revised theory, what it is for Diane to have B′ is for her to have a belief that is true iff Mark Twain is buried in Elmira. But, as before, knowing the conditions under which belief B′ is true doesn’t tell us exactly what it is for Diane to have belief B′. For there are many beliefs that Diane could have that are true iff Mark Twain is buried in Elmira: (i′) Samuel Clemens is buried in Elmira. (ii′) Mark Twain is buried in Elmira. (iii′) The actual author of Huckleberry Finn is buried in Elmira. ... Here we pick out the relevant beliefs via the sentences Diane could use to express those beliefs, were she to have them. Note that, given Thesis (III) (fine-​g rainedness), these are all distinct beliefs, since Diane could have one without having the others. So it is consistent with the revised theory that what it is for Diane to have B′ is for her to have belief (i′) on the above list, a belief she could express by saying, Samuel Clemens is buried in Elmira. For this belief has the appropriate truth-​conditions: it is true iff Mark Twain is buried in Elmira. So we may assume that this is what it is for Diane have B′. (We shall question the legitimacy of this assumption below, but let us carry on with it for the moment.) Thus, Diane has the following attitudes: Diane’s attitudes Samuel Clemens is buried in Elmira.  B′ I want to visit Mark Twain’s grave.  D′ Here we pick out the attitudes via the sentences Diane would use to express or report them. Let’s return to Carl. What will Carl be motivated to do, given what we’ve said about his beliefs and desires? Well, if all else is equal, he will be motivated to visit Elmira. For Carl believes that Mark Twain is buried in Elmira, and he wants to visit Mark Twain’s grave. Now, how did we arrive at that prediction? Given Thesis (IV), we must have inferred it from the particular claim that Carl has belief B′ and desire D′, together with the following generalization (†): (†) Necessarily, for all y, if y has belief B′ and desire D′, then if all else is equal, y’s having B′ and D′ will motivate y to visit Elmira. 492

De Se Attitudes and Action

I take it that, given Thesis (IV) and the facts of the case, the revised theory is committed to (†). But (†) is false. For (†) implies that in any possible situation in which Diane has belief B′ and desire D′, and in which all else is equal, Diane’s having those attitudes will motivate her to visit Elmira. But this is false. To see this, just imagine a scenario in which Diane has the following attitudes: Samuel Clemens is buried in Elmira.  B′ I want to visit Mark Twain’s grave.  D′ Mark Twain is buried in Poughkeepsie. If the case is a normal one, having these attitudes will motivate Diane to visit Poughkeepsie, not Elmira. If this is right, then (†) is false. Since the revised theory is committed to (†), the revised theory is false. Furthermore, since Carl’s de dicto belief poses essentially the same problem for the revised theory that Al’s de se belief posed for the expanded doctrine, it seems the general problem here is not specific to the de se. That is the skeptic’s argument. I think this argument fails, and it will be instructive to see just where it fails. It will also be instructive to see why our earlier argument against the expanded doctrine does not fail in the same way. Where does this argument against the revised theory go wrong? The problem is with what the argument assumes about what it is for Diane to have belief B′. Belief B′, recall, was introduced as a belief that Carl could express by saying, Mark Twain is buried in Elmira. The argument assumes that what it is for Diane to have B′ is for her to have a belief she could express by saying, Samuel Clemens is buried in Elmira. It is true that this assumption appears, at least as first glance, to be consistent with the revised theory, but it is hardly entailed by the revised theory. What the revised theory is committed to is the claim that there is a belief (n′) on the following list such that for Diane to have belief B′ is for her to believe (n′): (i′) Samuel Clemens is buried in Elmira. (ii′) Mark Twain is buried in Elmira. (iii′) The actual author of Huckleberry Finn is buried in Elmira. … So it is open to the advocate of the revised theory to say that what it is for Diane to have B′ is for her to believe (ii′), the belief she could express by saying, Mark Twain is buried in Elmira. And this move blocks the argument against the revised theory. For together with (†), these claims imply that having belief B′ and desire D′ will motivate Diane to visit Elmira, all else being equal. But this consequence is not implausible. For on this version of the revised theory, what it is for Diane to have B′ and D′ is for her to believe that Mark Twain is buried in Elmira, and to desire to see Mark Twain’s grave. It is not at all implausible to suppose that that pattern of attitudes will motivate Diane to visit Elmira, all else being equal. So the defender of the revised theory has a fairly straightforward reply to this argument. In contrast—​and this is important—​the defender of the expanded doctrine has no parallel reply to our earlier argument against that view. When we initially presented that argument, we assumed that what it was for Betty to have belief B is for her to have a belief she could express by uttering sentence (i) on the following list (repeated from above): (i) If you roll up in a ball, the bear will leave. (speaking to Al) (ii) If Al rolls up in a ball, the bear will leave. (iii) If the actual man in the red hat rolls up in a ball, the bear will leave. … 493

Dilip Ninan

We then noted that if this is what it is for Betty to have belief B, then Betty’s having belief B and desire D will not motivate her to roll up in a ball, if all else is equal. And this contradicts (⋆), a claim to which the expanded doctrine is committed. But unlike the skeptic’s argument against the revised theory, one cannot defeat this argument by simply making an alternative assumption about what it is for Betty to have belief B. For the expanded doctrine is committed to the claim that there is a belief (n) on the above list such that what it is for Betty to have belief B is for her to have belief (n). But as we noted before, there is simply no belief (n) on the above list such that Betty’s having belief (n) will combine with desire D to motivate her to roll up in ball, if all else is equal.

8  Truth and Motivation Thus, the skeptic’s attempt to show that the problem facing the expanded doctrine isn’t specific to the de se fails. The problem really does seem to reveal a difference between de se and de dicto attitudes. Let us try to zero in on what that difference consists in. It will help to introduce some terminology at this point. Let us say that a pair of token beliefs b, b′ are truth-​conditionally equivalent just in case, necessarily, b is true iff b′ is true. And let us say that a pair of token beliefs b, b′ are motivationally equivalent just in case: necessarily, for any attitude type A, action ϕ, and agents x and x′: x’s having b and A will motivate x to perform ϕ (all else being equal) iff x′’s having b′ and A will motivate x′ to perform ϕ (all else being equal). Let b1 be Al’s token de se belief that he could express by saying, If I roll up in a ball, the bear will leave. If all else is equal, then Al’s having belief b1 and desire D will motivate him to roll up in a ball. Let b2 be Betty’s token belief that she could express by saying to Al, If you curl up in a ball, the bear will leave. Note that Betty’s having belief b2 and desire D will not motivate her to roll up in a ball, if all else is equal. Thus, although b1 and b2 are truth-​conditionally equivalent—​both are true iff: if Al rolls up in a ball, the bear will leave—​they are motivationally distinct. Beliefs b1 and b2 differ in their capacity for motivating action in conjunction with other attitudes. What’s interesting is that the point generalizes beyond Betty’s belief b2. To see this, take any token belief b* that anyone distinct from Al could possess. If b* is truth-​conditionally equivalent with Al’s token belief b1, b* will not be motivationally equivalent to b1. To see this, let x be any agent distinct from Al, and let b* be any token belief that x could possess that is truth-​conditionally equivalent to b1. Then b* will be a token belief that, were he to have it, x could express by uttering one of the sentences on the following list: (i) If you roll up in a ball, the bear will leave. (speaking to Al) (ii) If Al rolls up in a ball, the bear will leave. (iii) If the actual man in the red hat rolls up in a ball, the bear will leave. … But no token belief (n) on that list is such that x’s having token belief (n) and desire D will motivate x to roll up in a ball, if all else is equal.Thus, b* is motivationally distinct from b1. Since this holds for an arbitrary token belief that could be possessed by someone distinct from Al, it holds for them all.There is no token belief that someone distinct from Al could possess which is both truth-​conditionally and motivationally equivalent to Al’s token de se belief. Contrast this with the de dicto case. Let b′1 be Carl’s token de dicto belief that he could express by saying, Mark Twain is buried in Elmira. In this case, it does seem plausible to suppose that there is a token 494

De Se Attitudes and Action

belief b′2 that someone other than Carl could have that is both truth-​conditionally and motivationally equivalent to Carl’s token belief b′1. Just take b′2 to be the token belief Diane could express by saying, Mark Twain is buried in Elmira. Just as having b′1 and desire D′ will motivate Carl to visit Elmira (all else being equal), having b′2 and D′ would likewise motivate Diane to visit Elmira (all else being equal). And b′1 and b′2 have the same truth-​conditions: both are true just in case Mark Twain is buried in Elmira. So there is a token belief that someone distinct from Carl could possess which is both truth-​ conditionally and motivationally equivalent to Carl’s token de dicto belief. This feature of Carl’s token de dicto belief seems to be a general feature of de dicto beliefs. It seems that for any token de dicto belief b held by agent x, we can find a token belief that someone x′ distinct from x can possess which is both truth-​conditionally and motivationally equivalent to b. Here’s a recipe that will work, at least for token beliefs that can be expressed in a natural language. Find a sentence S that x could use to express her token belief b. The token belief b′ that x′ can express by uttering S is very likely to be both truth-​conditionally and motivationally equivalent to b. For example: take Carl’s token belief that he could express by saying, Mark Twain is buried in Elmira. We found a token belief that Diane could possess which is both truth-​conditionally and motivationally equivalent to this belief simply by considering the token belief that (where she to have it) Diane could express by saying, Mark Twain is buried in Elmira. So we may hypothesize the following difference between de se attitudes and de dicto attitudes. For a great many ordinary token de se beliefs b, if x is the holder of b, there is no token belief b′ that someone other than x can possess that is both truth-​conditionally and motivationally equivalent to b. But for all de dicto beliefs b, if x is the holder of b, then there will be a token belief b′ that someone other than x can possess that is both truth-​conditionally and motivationally equivalent to b. Although the arguments of this chapter could not be said to have established this hypothesis, they have, I hope, rendered it plausible. And the hypothesis itself may serve as a useful guide for further inquiry into the special connection between action and de se attitudes.

9  Conclusion One way a pair of token beliefs can be similar is in respect of their truth-​conditions. Another way a pair of token beliefs can be similar is in respect of their capacity to motivate action in conjunction with other attitudes.What de se attitudes reveal is that these respects of similarity come apart in a particular way. As we saw, no token belief (held by someone distinct from Al) is both truth-​conditionally and motivationally equivalent to Al’s token de se belief. And this would seem to hold not just for Al’s token de se belief, but for many other ordinary de se beliefs. The practice of individuating beliefs via propositions permits us to capture certain similarities between token beliefs. On the orthodox view, propositions have their truth-​values absolutely. This means that if a pair of token beliefs are of the same type, then they have the same truth-​conditions. But given the shareability of propositions, this has the consequence that certain token de se beliefs b are treated as type-​equivalent to certain token de dicto beliefs b′, even though b and b′ have very different motivational profiles. And this, in turn, leads to difficulties in formulating law-​like generalizations that link attitudes to actions, as we have seen. Two responses to this situation dominate the literature. One response—​most closely associated with Lewis—​is that of the “revised theory” discussed above. This denies the traditional assumption that if a pair of token beliefs are of the same type, then they have the same truth-​conditions. Beliefs are still individuated by contents, but contents are no longer absolute propositions. Instead, contents are properties or relative propositions, things that vary in truth value across individuals.The principal advantage of this move, as I see it, is that it allows us to preserve Thesis (IV), the idea that law-​like generalizations linking attitudes and actions have a central role to play in our practice of predicting and explaining action. For example, on this approach our generalization (⋆) would be equivalent to the following plausible-​sounding generalization: 495

Dilip Ninan

Necessarily, for all individuals y, if y believes de se that if he rolls up in a ball, the bear will leave, and y wants the bear to leave, then, if all else is equal, y’s having these attitudes will motivate y to roll up in a ball. A second response—​most closely associated with Perry—​maintains the traditional assumption that propositions are absolute. As I  understand Perry’s view, a belief type is a property that consists in believing an absolute proposition. So if a pair of token beliefs are type-​equivalent, then they have the same truth-​conditions. Perry then introduces another kind of psychological property to play the role traditionally thought to be played by attitudes in the explanation and prediction of action. Let us call such properties attitudes*, beliefs*, etc. Like Lewis’s beliefs, Perry’s beliefs* are individuated by properties (or what Perry variously calls senses or relativized propositions), and so do not have absolute truth-​ conditions. It is beliefs*, rather than beliefs, that figure in the explanation of action. Although Perry doesn’t put the point this way, we might say that folk-​psychological generalizations link attitudes*, rather than attitudes, to actions. Perry (1977) puts the point in Fregean terminology: We use senses [i.e. properties] to individuate psychological states, in explaining and predicting action. It is the sense entertained and not the thought [i.e. absolute proposition] apprehended that is tied to human action. When you and I entertain the sense of “A bear is about to attack me”, we behave similarly. We both roll up in a ball and try to be as still as possible. Different thoughts apprehended, same sense entertained, same behavior. When you and I both apprehend the thought that I am about to be attacked by a bear, we behave differently. I roll up in a ball, you run to get help. Same thought apprehended, different sense entertained, different behavior. Perry 1977, 23 Both Lewis and Perry can be seen as agreeing that certain psychological properties have a central role to play in the explanation and prediction of action. Furthermore, both agree that the psychological properties in question do not consist in standing in a relation to an absolute proposition. Lewis takes the psychological properties in question to be attitudes, and so denies that attitudes are individuated by absolute propositions. Perry does not identify the psychological properties in question with ordinary attitudes—​belief, desires, and the like. Instead, he identifies the psychological properties in question with what I  called attitudes* above. This leaves him free to maintain that beliefs and desires are individuated by absolute propositions. How to choose between these closely related approaches is a question we must leave as a matter for future inquiry.14

Acknowledgments For helpful comments on earlier versions of this material, thanks to David Braun, Mike Caie, Jeremy Goodman, and Jeff King.

Further Reading Cappelen, H. and Dever, J. (2013). The Inessential Indexical. Oxford University Press, Oxford. Lewis, D. K. (1979). Attitudes de dicto and de se. Philosophical Review, 88(4), 513–​543. Reprinted in Lewis 1983b, 133–​159. Magidor, O. (2015). The myth of the de se. Philosophical Perspectives, 29(1), 249–​283. Ninan, D. (2016). What is the problem of de se attitudes? In M. Garcia-​Carpintero and S. Torre (eds.), About Oneself: De Se Attitudes and Communication. Oxford University Press, Oxford. Perry, J. (1977). Frege on demonstratives. Philosophical Review, 86(4), 474–​497. Reprinted in Perry 1993, 3–​32. Perry, J. (1979). The problem of the essential indexical. Noûs, 13(1), 3–​21. Reprinted in Perry 1993, 33–​52. Stalnaker, R. (1981). Indexical belief. Synthese, 49(1), 129–​149. Reprinted in Stalnaker 1999, 130–​149. Torre, S. (2018). In defense of de se content. Philosophy and Phenomenological Research, 97(1), 172–​189.

496

De Se Attitudes and Action

Notes 1 For discussion of this thesis, see Cappelen and Dever (2013, ch. 3), Bermúdez (2017), and Morgan (2019). 2 In particular, Perry associated the doctrine of propositions with Frege (1892). 3 Perhaps what we often think of as two-​place relations between x and y are really three-​place relations between x, y, and a time or four-​place relations between x, y, a time, and a possible world. I gloss over this subtlety for the most part in what follows. 4 I introduce attitude tokens later in the chapter. When I speak of an attitude or a belief or a desire without qualification, I usually mean to speak of types or properties, things that different individuals may have at different times. 5 Thanks to David Braun for suggesting a principle similar to this one. The principle should be interpreted as holding at an arbitrary fixed time. 6 Perry considers this possibility at one point (Perry 1979, 45–​46), but assumes that it requires us to say that the proposition Perry comes to believe is one that only he can believe, and he goes on to argue against the idea of ‘private propositions’. But it is unclear why this view would require us to say that the proposition Perry comes to believe is one that only he can believe. A fellow shopper might come to believe that same proposition, and might express the corresponding belief by saying to Perry, You are the one making the mess. 7 See, for example, Boer and Lycan (1980, 450–​453),Tiffany (2000, 38–​41), Spencer (2007, 183–​184), Cappelen and Dever (2013, section 3.1), Magidor (2015), and Ninan (2016). 8 By a de dicto attitude, I simply mean an attitude that is not de se. 9 Stalnaker (1999, 19–​21) seems to have been the first to observe that what is distinctive about the de se is revealed most clearly by cases with the structure of Perry’s ‘bear attack’ case. 10 For discussion of arguments similar to the one presented here, see Ninan (2016) and Torre (2018). 11 Recall our stipulation that, in the bear attack case, Betty is not suffering from any form of identity confusion. So she knows that she is not Al, she knows that she is not the actual man in the red hat, etc. 12 See also Lewis (1983a, 1986). Unlike Lewis, I continue to use the term proposition interchangeably with content. So according to the revised theory, propositions are properties. 13 If we were attending to temporal matters, we would include times in our token beliefs. See Kim (1976) for a more general theory of token events along these lines. 14 A more comprehensive discussion would also consider the view—​closely associated with Frege and his followers—​that the contents of de se beliefs are private. On this approach, no one other than Al can believe the proposition that serves as the content of his de se belief that he is being chased by a bear. See Frege (1956), Evans (1981), Peacocke (1981), McDowell (1984), Forbes (1987), Heck (2002), Kripke (2008), Stanley (2011, ch. 3), and Ninan (2016) for further discussion.

References Bermúdez, J. L. (2017).Yes, essential indexicals really are essential. Analysis, 77(4), 690–​694. Boer, S. E. and Lycan, W. G. (1980). Who, me? Philosophical Review, 89(3), 427–​466. Cappelen, H. and Dever, J. (2013). The Inessential Indexical. Oxford University Press, Oxford. Chisholm, R. (1981). The First Person. The Harvester Press, Brighton. Evans, G. (1981). Understanding demonstratives. In H. Parret and J. Bouveresse (eds.), Meaning and Understanding. Walter de Gruyter, Berlin. Reprinted in Evans 1985, 291–​321. Page references are to the 1985 reprint. Evans, G. (1985). Collected Papers. Clarendon Press, Oxford. Forbes, G. (1987). Indexicals and intensionality: A Fregean perspective. Philosophical Review, 96(1), 3–​31. Frege, G. (1892). Uber sinn und bedeutung. Zeitschrift fur Philosophie und philosophische Kritik, C, 25–​50. English Translation: “On Sense and Meaning”, in B. McGuinness (ed.), Frege: Collected Works, Basil Blackwell, Oxford, 157–​177. Frege, G. (1918/​1956). The thought: A logical inquiry. Mind, 65(259), 287–​311. Heck, R. (2002). Do demonstratives have senses? Philosophers’ Imprint, 2(2), 1–​33. Kim, J. (1976). Events as property exemplifications. In M. Brand and D. Walton (eds.), Action Theory. Reidel, Dordrecht, 159–​177. Kripke, S. (2008). Frege’s theory of sense and reference: Some exegetical notes. Theoria, 74, 181–​218. Reprinted in Kripke 2011, 254–​291. Kripke, S. (2011). Philosophical Troubles, volume 1. Oxford University Press, New York. Lewis, D. K. (1979). Attitudes de dicto and de se. Philosophical Review, 88(4), 513–​543. Reprinted in Lewis 1983b, 133–​159. Lewis, D. K. (1983a). Individuation by acquaintance and by stipulation. Philosophical Review, 92(1), 3–​32. Reprinted in Lewis 1999, 373–​402.

497

Dilip Ninan Lewis, D. K. (1983b). Philosophical Papers, volume 1. Oxford University Press, New York. Lewis, D. K. (1986). On the Plurality of Worlds. Blackwell, Oxford. Lewis, D. K. (1999). Papers in Epistemology and Metaphysics. Cambridge University Press, Cambridge. Loar, B. (1976). The semantics of singular terms. Philosophical Studies, 30(6), 353–​377. Magidor, O. (2015). The myth of the de se. Philosophical Perspectives, 29(1), 249–​283. McDowell, J. (1984). De Re senses. The Philosophical Quarterly, 34(136), 283–​294. Morgan, D. (2019). Temporal indexicals are essential. Analysis, 79(3), 452–​461. Ninan, D. (2016). What is the problem of de se attitudes? In M. Garcia-​Carpintero and S. Torre (eds.), About Oneself: De Se Attitudes and Communication. Oxford University Press, Oxford. Peacocke, C. (1981). Demonstrative thought and psychological explanation. Synthese, 49(2), 187–​217. Perry, J. (1977). Frege on demonstratives. Philosophical Review, 86(4), 474–​497. Reprinted in Perry 1993, 3–​32. Perry, J. (1979). The problem of the essential indexical. Noûs, 13(1), 3–​21. Reprinted in Perry 1993, 33–​52. Perry, J. (1993). The Problem of the Essential Indexical and Other Essays. Oxford University Press, New York. Spencer, C. (2007). Is there a problem of the essential indexical? In M. O’Rourke and C. Washington (eds.), Situating Semantics: Essays on the Philosophy of John Perry. MIT Press, Cambridge, MA. Stalnaker, R. (1999). Context and Content. Oxford University Press, New York. Stanley, J. (2011). Know How. Oxford University Press, Oxford. Tiffany, E. (2000). What’s essential about indexicals? Philosophical Studies, 100(1), 35–​50. Torre, S. (2018). In defense of de se content. Philosophy and Phenomenological Research, 97(1), 172–​189.

498

37 ACTING WITHOUT ME Corporate Agency and the First Person Perspective Herman Cappelen and Josh Dever

The work of Castañeda, Lewis, and Perry convinced many philosophers that so-​called de se attitudes (and the first person perspective in particular) are the source of deep and wide-​reaching philosophical insights. The de se, it is claimed, plays an essential role in giving an account of thought and agency. Others claim that the de se is essential to understanding morality, perception, or epistemic phenomena such as immunity to error through misidentification.1 The de se deflationists,2 on the other hand, have argued that the phenomena Lewis and Perry described tell us nothing of philosophical significance. According to the deflationists, Lewis and Perry used new and confusing terminology to describe old and familiar phenomena. This paper is a new defense of the deflationary view. Our focus is on the connections between the de se and agency. We have two central goals: To provide a new argument that de se attitudes play no essential (or explanatorily important) role in understanding actions. Our argument turns on the nature of corporate agency and is, in outline, very simple: corporations, such as Apple and BMW, act, yet we don’t and shouldn’t attribute to them de se states, so de se states aren’t necessary to explain at least one common and important type of agency. 2 To show that corporate and human agency are more similar than is typically assumed: corporate agency can serve as a faithful model of human agency. 1

We proceed as follows. • In the first section, we give some representative examples of corporations and their actions, with the goals of (a) making plausible that corporate entities can undertake genuine action and (b) showing some ways in which those actions can float free of the intentions and actions of individuals composing and associated with the corporate entities. • In the second section, we set out a number of considerations suggesting that corporations do not need any sort of proprietary indexical states in order to act. We show that several lines of thought that people have taken to show that action is impossible without de se thoughts are not only uncompelling in the corporate case, but turn out to have their grip in the individual case weakened by virtue of seeing why they fail in the corporate case. At the end of the second section,

499

Herman Cappelen and Josh Dever

then, the minimal goal is that the reader will be convinced that corporate action does not require de se states. • In the third section, we turn to considering responses that attempt to preserve an essentiality thesis by arguing that corporate action is in some important sense different from “normal human action”, and is best understood as a kind of pseudo-​action. We argue that all attempts to differentiate corporate from human action are dialectically ineffective, as they turn out to beg exactly the question at issue in the discussion of essentiality theses. • In the last section we consider an attempt at a transcendental refutation of our argument. Before we get to all of that, a brief outline of the history of this debate.The tradition we’re responding to can be traced back to influential work by John Perry and David Lewis. Perry’s most famous example—​the famous messy shopper (Perry 1979)—​has been interpreted as showing that there’s a deep connection between indexicality and action. In that example, Perry is described as changing his messy behavior only after thinking the thought that would be expressed in words by ‘I’m making a mess’. Recognizing, in a first-​personal way, that he was the shopper making a mess, is, according to Perry, essential for the change in action. Perry says: “When we replace (‘I’) with other designations of me, we no longer have an explanation of my behavior and so, it seems, no longer an attribution of the same belief. It seems to be an essential indexical” (1979, 3). This line of thought has been extremely influential throughout philosophy. Here are two illustrations of how the assumption is now appealed to more or less as common ground in contemporary philosophy: [P]‌ractical guidance is, in Perry’s phrase, essentially indexical, in the sense that its function depends not only on which of many propositions it expresses but also on how that proposition is determined by the context—​specifically, on its being determined in the same way as the reference of indexical expressions such as “I”, “you”, “here”, and “now”. Velleman 2005, 78 It is widely agreed that agents need information in an egocentric form: they must think of places as “here” and “there”, times as “now” and “then” if they are to be able to act on what they know. Owens 2011, 267 Our goal in what follows is to sketch a view of human and corporate agency according to which they are fundamentally similar and not tied to the first person or other indexical attitudes.

1  Introducing Corporate Agency In this first section, we explain what we mean by ‘corporate agency’ and ‘corporate action’ and describe some of its relevant features. By ‘corporate actions’ we don’t mean simply ‘collective’ action. Collective actions paradigmatically involve multiple agents deliberately coordinating on the performance of a single task, in such a way that there is a clear sense in which each individual agent is, via the collective action, involved with and intentionally undertaking that task. Think, for example, of multiple agents cooperating to lift a table. Collective actions, however, are not the case we will focus on. Rather, we will consider cases of what we will broadly call corporate action. One particular kind of corporate action is action undertaken by traditional corporations. Apple, for example, chooses to announce a new model of iPhone at a conference.This is an intentional action—​it results from deliberation on Apple’s part, as a strategy for satisfying Apple’s desire to win more market share, a strategy

500

Corporate Agency and the First Person

selected in light of Apple’s beliefs about what announcement strategy will maximize sales.3 But Apple’s action differs in some important ways from the action of the agents collectively lifting a table: • When Apple announces a new model of iPhone, relatively few members of Apple may act. The announcement may be made solely by Tim Cook, while thousands of other employees4 go about their usual business. • Even the deliberative process leading up to the announcement may have been undertaken only by a few members. So there is no helpful sense in which, when Apple announces a new iPhone, all members of Apple are involved with and intentionally undertaking the announcing. • Indeed, it could be that no one does any announcing. One person might write part of an announcement script, someone else another part, a third person might edit it, yet a fourth person might forward the final version to the newspapers, and so on. Broadly, then, there are many organizations and social structures—​entities which presumably in some broad sense are grounded in features of individual people (more or less—​see one case to the contrary below)—​which can be thought of as independent loci of deliberation and action.5 It is these sorts of broadly corporate agents, and the actions undertaken by them, on which we will focus. For some of our subsequent purposes, it is worth attending to some features of the expansive boundaries of the category of corporate action. Note the following: • A corporation can perform action A, as guided by an intention to A, resulting from a desire D and a belief B, even when every single member of the corporation is attempting not to do A (or even attempting to block A), as guided by an intention that not A resulting from a desire that not D and a belief that not B. Apple, for example, acts to cut prices on the iPhone, wanting to increase profits and thinking that lower prices will lead to higher sales and more profits. But it turns out that every member of the board actually wants the iPhone price to go up, because they want to sabotage Apple profits (perhaps to trigger a golden parachute clause in their contracts). But each board member thinks (incorrectly) that they are hated by the other board members, so that if they speak out in favor of and vote for the price cut, the other board members will vote in the opposite way. As a result, the vote comes out unanimously in favor of the price cut, and Apple so acts. The moral to draw from this is that the determination relations between the mental states and actions of corporate members and the mental states and actions of the corporation can be more or less arbitrarily complex and convoluted. A corporation can perform an action even at times when the corporation has no members at all: • Even if every member of Apple were tragically killed in an iPhone presentation disaster, the corporation could continue to act. Perhaps this is because there are authorized agents—​not members of Apple—​who can trigger further actions. (Lawyers hired by Apple can still bring it about that Apple defends its patent rights.) Or perhaps it is because there are mechanisms already put in place to trigger action in the right circumstances—​stock purchasing algorithms, designed and authorized by Apple (although not necessarily by any member of Apple) bring it about that Apple performs a hostile takeover of Yahoo when Yahoo’s stock drops below a crucial threshold. The moral to draw from this is that the determination basis for corporate action needs to be temporally extended, depending on the actions and intentions of corporate members across a series of times before (and in suitable cases after) the time of the action, and agentially extended, depending on actions and intentions of agents who are not part of the corporation.

501

Herman Cappelen and Josh Dever

What action a corporation performs can be underdetermined even by the totality of facts about all its members’ mental states. Consider the following rather contrived example: • Apple has two people on the payroll as “possible spokespeople”. Each day, a coin is flipped by a coin-​flipping machine in a sealed vault. If the coin lands heads, person P1 is the actual spokesperson for that day, and statements made by that person (and not by P2) count as declarations by Apple. If the coin lands tails, P2 is instead the actual spokesperson for the day, and it is P2’s statements that count as declarations by Apple. On Wednesday, P1 makes a public statement that Apple is raising the price of the iPhone, and P2 makes a public statement that Apple is lowering the price of the iPhone. If the coin (unseen by anyone) has landed heads that morning, then Apple has raised the price of the iPhone, while if it has landed tails, Apple has lowered the price.6 So Apple’s action can be altered simply by changing a non-​intentional fact: the orientation of the coin. It is not people’s beliefs about the coin that determine what Apple has done, but only how the coin actually (again, unbeknownst to anyone) lies. The brief moral: the determination basis for corporate action can extend beyond the (human) intentional. Of course, in this way corporate actions are actually like human actions, since human actions can be grounded, partially or fully, in non-​intentional facts, given that brute physiological features may in some cases help determine what action a person performs.

2  There is no De Se Requirement on Corporate Action Even if, when concentrating on ordinary human action, you find the thought that agency is impossible without mental states of some special de se kind compelling, you shouldn’t find it compelling when you pay attention to corporate action. The burden of this section will be to examine a cluster of morals that derives from demonstrating that and considering why there is no plausible case for a de se requirement on corporate action. We show that there is simply nothing that would even count as an interesting de se kind of content for corporate entities in the first place.

2.1  Corporate Agency and the De Se We are targeting the claim that no action is possible without the possession of some de se mental states, and those de se states are involved in the explanation of the action.We approach this question by asking: when Apple decides to announce the new iPhone, does it need to be thinking about itself in some special way? We answer this question by showing that specific theories of de se content, especially ones which are ambitious in making such content particularly unusual or distinctive, are going to look implausible when we move to the corporate case. Consider the following four theories of de se content: 1

Egocentric space. Suppose, for example, that one has a roughly Evans-​inspired thought that first-​ personal thought is thought that organizes the world around a spatial representational system that is egocentrically organized, so that to think of the world first-​personally is to think of things in terms of where they are in relation to you.7 Then it’s going to look wildly implausible to think that Apple has such states. Apple doesn’t organize its beliefs about the world by locating objects in “egocentric space”. It isn’t even clear that Apple has a location, and even if it does, there is no reason why Apple should systematically structure its representation of the world in terms of the relation of other objects to Apple. (Apple, for example, doesn’t typically gain information about the world by a faculty of perception which delivers “centered information”.8) So given that Apple does act, Apple better not need de se states of this sort as an essential prerequisite for action.

502

Corporate Agency and the First Person

2

Locating the agent’s body relative to external objects. Suppose you think that de se states play a special role in getting action locally connected to the body (i.e. that de se attitudes are involved in locating the objects we act on relative to ourselves). Here is an example from Perry: Consider a transaction with a fax machine.To press certain buttons on it, I have to move my fingers a certain distance and direction from me. It isn’t enough to know where the buttons were relative to one another, or where the fax machine was in the building or room. I had to know where these things were relative to me. Perry 1998, 87

If the role of the de se states (and the source of their essentiality) is to get the targeted external objects hooked up with the body of the acting agent in the right way, then it’s not clear that it even makes sense to talk about corporations having states of this sort, much less that corporations must have such states in order to act. Apple doesn’t have a body in any normal sense, so it has no need for a special category of mental states that track a body or help coordinate the body with the external world. Apple’s lack of a body is in part a further consequence of the fact that Apple lacks a clear physical location. But it is also a consequence of the fact that Apple doesn’t have dedicated and stable causal connections to specific parts of the world. Jones’s hand is part of Jones’s body, one might think, because Jones reliably has causal control over the state of the hand. The body/​non-​body distinction for Jones is drawn by the distinction between the things Jones can reliably control and the things he cannot. But Apple needn’t have any such distinction—​the things Apple can control might vary constantly, and there might be a spectrum of degrees of control.9 3

4

The de se as implicit/​unarticulated representation. A  third view has it that de se states are identified by their special mode of mental representation (see for example Recanati 2013). Perhaps de se states characteristically don’t include “explicit representation” of the self, but do include explicit representation of other objects. When Jones believes de se that he is in Athens, on this view, were we to look inside Jones’s “belief box”, we would see a mental representation of Athens (a token of a Mentalese term for Athens), but we would not see a mental representation of Jones himself. Again, it is unclear what it would even mean for corporations to have states of this sort. How could we distinguish between what corporations “explicitly represent” and what they merely implicitly represent? It’s already unclear how to cash out the “belief box” metaphor with ordinary people, but here it’s at least possible that an adequate computational neurobiology might reveal syntactic representation systems that allow us to show that there is no “self-​representation” in certain beliefs. But it isn’t even clear how to get started in looking for something “belief box”-​like in the case of Apple. Epistemic characterization of the de se. Suppose you think that de se states are epistemically characterized. Perhaps de se states are states that originate in internally directed epistemic capacities, such as proprioception and introspection.10 (To think of things in this way is to tie de se states to the epistemic phenomenon of immunity to error through misidentification—​states are de se because there is no chance of epistemological error via misidentifying who the state is about, because the state comes about via a faculty that can only deliver information about the self.) Now, Apple might have such internally directed epistemic capacities. (Although given the greater flexibility of corporate structure than human biological structure, any such capacities are less likely to be essentially internally directed.) Apple might find out what its profits are in a very different way from the way it finds out what Google’s profits are. But it’s hard to see why any such difference should play an essential role in Apple’s actions. Even if Apple did have to learn its own profits by reading the Wall Street Journal, it would still seem quite capable of acting on the resulting information.

503

Herman Cappelen and Josh Dever

Obviously we can’t exhaustively consider every possible substantive way of characterizing de se mental states, and argue that corporations can act without states of that sort (either because corporations can’t even have states of that sort, or because corporate action can proceed without such states, even when corporations can have them). But we think that minimally a serious challenge has been offered to the fan of a de se essentiality view: give a substantive account of de se states that is compatible with (a) those states being required for action and (b) corporations being able to act. We think the considerable flexibility in the internal causal structure of corporations makes it very hard to believe that that challenge can be met. Let us finish this section with two final points about the challenge. First, an additional reason for thinking that the challenge is hard to meet is that we can imagine corporations which are set up so as to be simply incapable of self-​beliefs of any sort. Perhaps Apple by explicit corporate policy never tracks information about Apple. Apple can know what raw materials are needed to produce iPhones, and can want iPhones to sell well. On the basis of these states, Apple can act to purchase materials at low prices. But in doing so, it does not and cannot represent itself as acting. Second, it’s sometimes tempting to meet this kind of challenge by deflating the de se: by treating it as evidence that the de se must be something “thinner” than we might pre-​theoretically have suspected.We aren’t trying to resist deflationary strategies here. One version of a deflationary strategy is a broadly functionalist approach. One cheap way to get an essentiality thesis is to simply define de se states to be those states which are efficacious in producing action. If that’s all that’s meant by de se states, we of course agree that corporations, like people, have de se states, and that such states are essential to action. But it’s obvious that there is no philosophical interest in that deflationary conclusion. In endorsing it, we haven’t learned anything substantive about the nature of content, the nature of mental representation, or the relation between mental content and action.

3  Corporations Can Act We have been developing a very simple argument against de se essentiality views: (Premise 1) Corporations do not have de se mental states. (Premise 2) Corporations can act. (Conclusion) De se mental states are not essential for action. The burden of the previous section has been to defend the first premise (modulo the thought that it is Inflationary, rather than Deflationary, de se states that corporations do not have). In this section, we turn to defending the second premise. A foreseeable reaction to the considerations we set out in the previous section is to say: “Of course corporations can’t have genuine de se mental states. That’s unsurprising, because corporations can’t undertake genuine actions”. The critic agrees, of course, that Apple does things. Apple releases the new iPhone, fires employees, and carries out hostile takeovers of competitors. But the critic makes a distinction between mere doings and genuine actions, and holds that everything Apple does falls into the first category. Our argument, then, depends on rejecting this distinction by the critic. We will briefly make some comments in favor of the view that corporate doings are genuine actions. But we don’t intend these comments to do more than create a pro tanto case in favor of our view, to set the initial burden of proof. The bulk of this section will then be devoted to rejecting arguments against the view that corporate doings are genuine actions. Why should we think that corporate doings are actions? Two brief considerations. First, our linguistic practice with respect to corporate doings looks indistinguishable from our linguistic practice with respect to ordinary actions by individuals. We describe individuals as asserting, planning, lying, attacking, being insincere, and defending—​we equally describe corporations as asserting, planning, lying, attacking, being insincere, and defending. We find headlines such as “Apple acted as a bully in 504

Corporate Agency and the First Person

the ebook market” and “Apple acted unfairly in stolen iPhone incident”, which explicitly use the language of acting with corporations. Moreover, these intentional descriptions of both individuals and corporations are distinguishable from mere “as if” intentional loose talk (“My new iPhone hates me!”) by being part of a rich practice of using the intentional states in explanation, prediction, and evaluation.We describe corporations just as much as individuals as doing things on the basis of reasons and considerations, and as subject to rational evaluability and praise-​and-​blame for the things they do. Our ordinary practice, then, suggests a continuity between the individual and the corporate case.11 The second point is that there had better not be a deep distinction between individual “genuine action” and corporate “mere doings”, because for all we could a priori tell, we might have turned out to be corporations. Of course we know this to be false, but that’s an empirical discovery. We could have discovered that each of us is inhabited by a host of corporate employees, and that we each act by way of an elaborate corporate authority structure.We move our hands not by initiating a sequence of neural firings leading to muscle contractions, but rather by issuing an order to the “hand division” of Me Incorporated,12 leading to various employees running around inside us pulling various cords and tendons. Now, one might take this as reason to think that it’s an empirical discovery that we genuinely act, but we think it’s greatly preferable to take “genuine action” just to be the kind of thing we do, and thus to think that such action must be compatible with corporate-​style realization.13 So much for the pro tanto positive case. What reason, then, is there to doubt that corporate doings are genuine actions? The general form of the discussion for the remainder of this section will to be consider specific markers of genuine action that are putatively missing in the corporate case. We will consider three types of markers: • Epistemological markers: The thought here is that genuine actions are distinguished by a special epistemological relation to what is done, as in the Anscombian view that in genuine action, the acting agent has non-​observational knowledge of what they are doing (and why they are doing it). • Structural markers: Alternatively, some think that genuine actions are structured by the agent’s exercise of “direct guidance” or “control” over what is done, and that actions are built up out of a vocabulary of “basic actions”. • Phenomenological markers: A third view is that a genuine action is characterized by its seeming a distinctive way to the acting agent. We will along the way express some skepticism about whether these various markers are reasonable constraints to place on a category of genuine action, but our primary strategy will be to hold that there is no non-​question-​begging case to be made that corporate doings lack any of these features. Let’s consider epistemological markers first. It is a philosophical commonplace that action carries with it certain epistemic privileges. George Wilson and Samual Shpall’s Stanford Encyclopedia of Philosophy article on action observes that “it is frequently noted that the agent has some sort of immediate awareness of his physical activity and of the goals that the activity is aimed at realizing”. As a more specific spelling-​out of the epistemic privilege, Wilson and Shpall note: David Velleman [1989] describes knowledge of one’s present and incipient actions as ‘spontaneous’ (knowledge that the agent has achieved without deriving it from evidence adequate to warrant it), and as ‘self-​fulfilling’ (expectations of acting that tend to produce actions of the kind expected). Wilson and Shpall 2016, §1.1 Perhaps, then, corporate doings are mere doings because they lack this epistemic privilege. But is there any good reason to think so? Suppose Apple releases a new iPhone. Does Apple have “immediate awareness” of its having done so? Is that knowledge “spontaneous” in Velleman’s sense, and does Apple’s expectation of releasing a new iPhone “tend to produce” releasings of 505

Herman Cappelen and Josh Dever

iPhones by Apple? At least in normal cases, affirmative answers to all of these questions seem reasonable. In normal cases, the mere fact that Apple has released a new iPhone is taken as sufficient reason to say that Apple knows that it has released a new iPhone. Much as in the individual case, there is something peculiar about the question of whether Apple knows that it has done so (except in certain specifically deviant cases we’ll discuss soon.) We normally take that knowledge to be immediate, spontaneous, and non-​observational in much the same ways we do in the individual case. Apple might know that Samsung is releasing a new phone by attending Samsung’s press conference, or by receiving reports from corporate spies in Samsung factories—​but this isn’t (typically) how Apple knows that Apple is releasing a new phone. Of course, individual employees at Apple will know that Apple is releasing a new iPhone in a wide variety of ways. But Apple’s way of knowing need not be the same as the way of knowing used by the Apple employees, even in cases in which Apple’s knowing is in some important sense constituted by the knowing of individual Apple employees. And, to stress again points made above, we can easily give cases in which Apple knows that it has released a new iPhone without, or constitutively independently of, anyone at Apple knowing this. In such cases, we may well feel that Apple’s knowledge of the release is simply constituted by the release. The skeptic about corporate action may, of course, feel that these are not genuine states of knowledge on Apple’s part. On the skeptic’s view, corporations engage in mere mock action, distinguished from real action by the lack of (for example) non-​observational knowledge by the acting party of what is done.The skeptic then responds to these considerations by holding that we have only provided more mockery—​we have backed up mock action with mock knowledge.Well, perhaps. But we’re far from convinced. Now we need a reason to think that the corporate epistemic states are mock knowings, rather than real knowings.14 Since the whole game by the skeptic was to provide a positive reason to distinguish corporations from individuals, we don’t think much progress has been made. Perhaps, though, convincing cases can be described in which Apple releases a new iPhone without thereby suitably knowing that it has done so. The iPhone is ready and in the factories, but no corporate decision to release has been made. But a rogue spokesperson at an industry event announces the new phone. This constitutes a release (because the spokesperson has suitable authority within Apple), but Apple doesn’t know that the release has happened, because it didn’t know that the spokesperson would make this announcement. It’s tricky deciding what to make of such cases, but here are three thoughts: (i) Perhaps the spokesperson knowing what he has announced is a way of Apple’s knowing what he has announced. (ii) Even when construed most favorably to the skeptic, such cases would only show that some things we call corporate actions aren’t genuine actions. (Still, that would serve to undermine some of our initial pro tanto case.) (iii) Finally, and most importantly, we think that these cases just cast doubt on the plausibility of epistemic markers for action—​because we think that similar cases can be given in the case of individual action. Individuals can (in, for example, a state of distraction) do things (e.g. push a button) that ought to count as real actions, but about which they lack non-​observational knowledge, because the relevant executive epistemic systems haven’t been monitoring what the individual is doing. Turning now to structural markers, consider another characterization of the acting/​doing distinction given by Wilson and Shpall: It is also important to the concept of ‘goal directed action’ that agents normally implement a kind of direct control or guidance over their own behavior. An agent may guide her paralyzed left arm along a certain path by using her active right arm to shove it through the 506

Corporate Agency and the First Person

relevant trajectory. The moving of her right arm, activated as it is by the normal exercise of her system of motor control, is a genuine action, but the movement of her left arm is not. That movement is merely the causal upshot of her guiding action, just as the onset of illumination in the light bulb is the mere effect of her action when she turned on the light. Wilson and Shpall 2016, §1.2 Perhaps corporate doings don’t allow a suitable distinction between what is directly controlled and guided and what is merely a causal upshot of the guidance, and hence don’t qualify as action. But why would we think that? On the face of it, a suitable distinction is easily available. In the above example, the difference between the way the right arm moves and the way the left arm moves is spelled out in part in terms of normal systems. But corporations, too, have both normal and abnormal ways of doing. Apple has a normal way to move iPhones from China to the United States, in the same way that we have a normal way to raise our left arm. In Apple’s case, the normal way is to put the iPhones in a Boeing 747 and fly them across the Pacific. But Apple could move the iPhones in an abnormal way, in the same way that we can raise our left arm in an abnormal way. If the Boeing 747 was unable to fly (was ‘paralyzed’), Apple could use other planes to lift that plane and carry it, with its iPhones, across the ocean. Maybe “normal systems” talk isn’t the right way of cashing out the direct control/​guidance structural thought. Again, we can’t survey every possible way of running the skeptical strategy, so we primarily hope here to raise a challenge to the skeptic—​they need to give a convincing way of distinguishing individual genuine action from corporate mere doings. We’ll consider two options: (i) Appeal to basic actions construed representationally: Perhaps a better way of capturing “guidance” is via a notion of a basic action, where we can initially think of these as the point at which representation of the means of carrying out the action gives out. The thought is roughly that there are certain things that we are just directly able to do (perhaps typically things like moving our hands), and then other more distally characterized things (like answering the phone) count as actions just in case those doings are brought about in the right way by sequences of basic actions, in the sense introduced by Danto (1973). The challenge, then, would be to give a suitable sense of basic actions for corporations. But why would this be difficult? Of course, when Apple does something, like releasing a new iPhone, it is typically exploiting underlying action capacities of other individuals (sometimes employees of Apples, sometimes not). The releasing may get done by various people speaking and moving in various ways. But that does not show that Apple’s action is not basic qua action by Apple, any more than your moving your hand is not basic because you exploit various nerves and muscles doing things in performing your basic action. If basic actions mark the point at which representation of the means of carrying out the action gives out—​if, as Lavin (2013, 4) says, “With basic action, thought about how to realize the end in question gives out; one acts immediately, directly or ‘just like that’”, then corporations seem rife with basic actions. If there is a worry here, it is perhaps that corporations end up with too many basic actions. Basic actions for us, on standard ways of thinking about it, are limited to movements of bits of the body. But basic actions for Apple might include just about anything—​Apple might basically design a new MacBook, or open a plant in Singapore, or overtake Google in the search engine market. (Of course, Apple could have structured schemes for the doings of any of these, but it also looks possible that at the level of Apple’s acting, that’s all there is to the action, and any further scheming is devolved to (e.g.) Apple employees, in the same way that muscular triggering sequences in hand movements are devolved to the non-​intentional anatomy of an acting individual.) (ii) Appeal to basic actions construed as the domain of autonomous actions: Basic actions are called on to play a dual role. They are representationally basic, in that the agent need not further intentionally articulate how to carry things out. But they are also often taken to represent a domain of autonomy. When we seek, for example, to answer the phone, we may succeed or fail, depending on how things go in the 507

Herman Cappelen and Josh Dever

world around us. But when we seek to perform a basic action, we have a kind of guarantee of success. And perhaps the thought is that that kind of guarantee can’t be spread as widely as the apparent range of corporate basic actions seeks to spread it. If this is the thought, we think it encapsulates a familiar kind of mistake. To look for a domain of autonomy in this sense is one more instance of trying to find a “cognitive home” in which the possibility of error and failure is eliminated. But there is no cognitive home—​failure is always a possibility. Corporations may fail in the exercise of their putative “basic actions”, but that had better not be an argument against their basicness, because we too can fail in the exercise of our basic actions. Finally, consider phenomenological markers. Perhaps actions are distinguished from mere doings by being accompanied by a characteristic phenomenal feel to the acting individual.What might the phenomenal feel be? Elisabeth Pacherie in “The Phenomenology of Action” gives: A non-​exhaustive list of proposed distinctions includes awareness of a goal, awareness of an intention to act, awareness of initiation of action, awareness of movements, sense of activity, sense of mental effort, sense of physical effort, sense of control, experience of authorship, experience of intentionality, experience of purposiveness, experience of freedom, and experience of mental causation. 2008, 180 We might helpfully appeal here to phenomena like alien limb syndrome, in which there is a loss of felt ownership of a part of the body, undermining our ability to act with (rather than merely do with) that body part. By now the structure of the dialectic should be clear. The skeptic proposes that corporate doings are mere doings because they are not accompanied by the characteristic phenomenology of action. We then ask why we are supposed to think that.When Apple releases a new iPhone, is it not aware of its intention to do so? Does it not have an experience of authority, intentionality, and purposiveness? Does it not, even, have an experience of mental (corporate?) causation? If the skeptic has any leverage here, it presumably comes from a general skepticism about whether corporations have phenomenal states of any sort. We admit we’re not entirely sure ourselves what phenomenal states are and whether we ourselves have them (have them in general, and even more so have them in the specific forms listed above as characteristic of the experience of action). So one response to which we are friendly is a general deflationism about the phenomenal constraints on action.15 That said, it also worth noting that phenomenal vocabulary is freely used in discussing corporations. We are told that Apple “feels the need to defend Steve Jobs”, “feels the pressure with the new iPhone launch”, “fears Android”, and “hates open-​source software”. The skeptic, of course, will suggest that this is mere mock phenomenology, backing up mock action. We then again suggest that the question is being begged. We are told that corporations are importantly different from individuals, but every attempt to cash out that important difference ends up resting on an undefended putative prior difference.16 More generally: we have been happy to assume throughout that corporations have genuine mental states (beliefs, desires, intentions). However, that assumption is more than is needed for our central argument. Our de se deflationism requires only that corporations act. If they can act without having mental states, they thereby provide an even more direct counterexample to the claim that action requires de se representation. The unconvinced skeptic, of course, may want to link action and mental states, deny that corporations have mental states, thereby deny that corporations act, and then retain an inflationary de se requirement on action. But such a skeptic inherits an explanatory burden on their account of the de se. Take whatever feature F corporations lack that prevents their states from being

508

Corporate Agency and the First Person

mental states. (For example, F might be phenomenal character.) Corporations can, even according to the skeptic, do things without feature F, so the presence of feature F is crucial for converting mere doings into actions. Since it’s actions that impose a requirement for de se representation, feature F had better also explain the appearance on the scene of those de se representations. It’s thus incumbent on this skeptic to say enough about what de se representation is to explain why feature F contributes to giving rise to it. This is a non-​trivial requirement–​many existing accounts of the de se, for example, fail to link de se representations distinctively to phenomenal character.

3.1  Objection: Corporations Only Act Because they Consist of People with De Se States A natural response to the line of thought above is the following: Suppose that corporations do act without de se states. This doesn’t show that de se states are inessential for action. A corporation is composed of people, and the corporation acts because the people composing the corporation act. But for the familiar Lewis-​Perry reasons, the individual people can’t act without having de se states. Thus de se states are essential for the action of the corporation—​just not de se states of the corporation. Note first that this objection begins by conceding our case against the claim that agency require de se states. The objection grants that corporations act without de se mental states, so it isn’t true that “agents need information in egocentric form … to be able to act on what they know”. The objector wants to propose a new essentiality thesis to replace the defeated thesis. The exact form of the new essentiality thesis isn’t completely clear; perhaps it’s something like “agents need something in the full causal explanation of their actions to involve someone’s having de se mental states”.Whether such a thesis, if true, would give the de se a philosophically interesting position in our understanding of agency and personhood is far from clear. But our immediate concern is whether there is any reason to believe the revised essentiality thesis. We outline two reasons for skepticism:

First Reply If corporations can act without de se mental states by having their actions determined by the actions of individual members of the corporations, presumably we too can act without de se mental states in the same way. As noted above, we could have been corporation-​like: our actions could be the result of the actions of individual agents, call them gnomes, who are members of our corpora. It would work like this: Humans acting with the assistance of internal gnomes: Suppose A wants button B to be pushed. Inside A there are finger gnomes and arm gnomes. They implement A’s desire by moving the arm and finger in an appropriate way so that button B is pushed. So A pushed the button with the help of the arm and finger gnomes (similar to how Apple acts with the help of employees in various departments). It’s hard to resist the idea that this is a genuine action by A: A often whips cream using an electric whipper. The gnomes are just internal versions of the electric whipper. Of course, no humans we know actually have internal gnomes, but it’s not unlikely that technological advances in the relatively near future can make available electronic versions of such gnomes. What we in the first instance take from this example is that if corporations can act without having de se states, so can we: we could be organized much as corporations (with the internal gnomes as

509

Herman Cappelen and Josh Dever

the analogue of employees). The revised essentiality thesis offers no objection to that conclusion—​it merely observes that our ability to act without having de se states ultimately rests on the de se states of the gnomes. Now ask: why should we take seriously the thought that the gnomes (that by hypothesis make our actions possible) need some special de se mental state in order to act? The original motivation for heading down the de se path was specific cases that looked like they called for specific kinds of action explanation. When we are told about the messy shopper’s sudden change in behavior, our immediate reaction is to think that the change in behavior calls for (i) a specific kind of belief (ii) in a specific person: Perry. What we’re supposed to learn from the example is that the shopper himself comes to believe that he himself is making a mess. That compelling thought then leads to the philosophical puzzle about what the nature of that specific belief could be. But if Perry had ended the messy shopper case by announcing that clearly someone else (e.g., one of the gnomes inhabiting him) had a new belief about where he was, or about how many fingers he had, we would have found the announced conclusion completely baffling. The messy shopper case just doesn’t provide any reason to think that those sorts of beliefs are required. So once the fan of the de se has conceded that the de se states might be in our gnomes rather than in us, and might be about features of the gnomes’ environment quite unrelated to the actions we are performing, the original philosophical impetus to propose a new kind of mental state goes away. Absent any positive reason to incorporate de se mental states in the gnomes, we’ll then propose that the gnomes in fact act without the de se. They could, for example, be little computers. Now we have a counterexample to the revised essentiality thesis as well: when we act, there aren’t any de se mental states anywhere. And once we reach that point, we can drop the appeal to the gnomes, and just think that the story about action begins and ends with us, and doesn’t involve the de se anywhere.

Second Reply A second point in response to the revised essentiality thesis: we were supposed to think that corporate action involved the de se somewhere because the corporate actions depended on the actions of the people making up the corporation. But perhaps we can have corporations not made up of people. We’ve already noted that corporations can act when they have no members at all, if the corporation is acting on instructions or via algorithms set in place when there were members. And we could imagine a corporation whose members were entirely robots or computer programs. The thought that corporate actions need to be underwritten by the actions of corporate members would then be met with the suggestion that corporate actions could be underwritten by the doings of the corporate members, whether or not those doings rose to some standard of counting as actions. Would such non-​human corporations threaten our confidence that the corporations were genuinely acting? Consider again us and our hypothetical gnomes. If we’ve already been convinced that we could be acting as we do as a consequence of the actions of gnomes who make us up, would it really be a threat to our status as actors to suggest further that the component gnomes are more like robots than people; that the gnomes do rather than act? That change in the gnomes won’t entail any change in our ability to deliberate and subsequently intend on the basis of rationally weighed reasons—​why would it entail a change in our ability to act? We might then suggest that the gnomes have a familiar name: neurons.

4  An Objection We have now finished the core of our argument: we have argued that corporate action is genuine action, but that in no interesting sense are de se mental states required for corporations to act. Hence, 510

Corporate Agency and the First Person

de se states cannot in general be a requirement for action, so any strong de se essentiality principle must be false. In this final section, we consider an attempt at a transcendental refutation of our argument. The central thought is simple. Our argument, holds the transcendental refuter, requires two contradictory things. (i) On the one hand, we need corporate action to be real, and not merely mock, action. We thus need there to be no trait that the skeptic can use to argue for a crucial difference between corporate and individual action. (ii) On the other hand, we need corporate action to be distinguished by the absence of distinctive and interesting de se mental states (or by the inessentiality of such states, if they do occur). The goal is then to use that difference to convince the de se theorist that de se states aren’t essential to action—​they might be a local feature of the way action systems are implemented in beings like us, but consideration of corporations shows that action could be implemented in other non-​de-​se-​requiring  ways. But (i) and (ii) pull in different directions: • On the one hand, to establish (ii) we need to show that corporations don’t have or require de se mental states. In order to do that, we inadvertently make the case for an important difference between individual and corporate action (and so undermine (i)). • On the other hand, the more convincingly we argue in favor of (i), i.e. that individual and corporate action are the same, the more we inadvertently make the case that corporations do have and require de se states. We can’t have it both ways—​either there is a crucial and philosophically interesting distinction here, or there isn’t. The situation here is one which confronts many broadly deflationary strategies in philosophy.The classic instance is Berkeleyian idealism. Berkeley wants to argue for two things: • First, the nature of reality differs sharply from what we ordinarily believe—​the existence of objects is, so to speak, much more metaphysically “lightweight” than we take it to be. • Second, the “lightweight” nature of reality is deeply indistinguishable from the “heavyweight” reality we took there, in our pre-​Berkeleyian naïveté, to be. The problem that Berkeley confronts is then that the success of the second point threatens the plausibility (and even the sensibility) of the first point. If objects being metaphysically lightweight is a perfect simulation of objects being metaphysically heavyweight, then it’s unclear what the actual difference is between things being light or heavy. Thus the threat that by “external objects”, all we ever meant was “idea in the mind of God”, so that we all agreed with Berkeley the whole time, and he made no progress toward convincing us that reality was not heavyweight. Similarly in many other areas. The ethical noncognitivist might try to leave everything in our moral talk and practice unchanged, showing that he can make noncognitivist sense of ascriptions of truth, belief, and facts to moral matters. But if his success is complete, we are at risk of losing track of the difference between cognitivism and noncognitivism, and the deflation that the noncognitivist wants becomes indistinguishable from inflation. And if the success is incomplete, then the cognitivist is left with a mismatch between noncognitivist theory and ordinary practice that can be leveraged into an argument for cognitivism. The truth relativist might try to leave everything in our ordinary talk unchanged, by showing how a monadic truth predicate can be constructed out of relativistically acceptable resources. But if his success is complete, we are at risk of being able to say things like “It is true simpliciter, and not merely true relative to my standards of taste, that the joke is funny”, and 511

Herman Cappelen and Josh Dever

thus of losing track of the difference between the relativist and the absolutist. Deflationary perfect simulation of inflationary phenomena looks an awful lot like genuine inflation. Thus the challenge for us is this: if corporate action is a perfect simulation of genuine action, then in what sense have we deflated the conditions for real action, rather than inflated the understanding of corporate action? We are not convinced that we are genuinely subject to a perfect simulation challenge. We deny one prong of that challenge, because we deny that the more convincingly we argue that individual and corporate action are the same, the more we inadvertently make the case that corporations do have and require de se states. We think individual actions don’t require de se states, so we don’t think bringing out similarities between corporate and individual actions will create pressure to treat corporate actions as de se involving. Furthermore, we don’t think that response is question-​begging in context. It’s not that we need already to assume de se deflationism in order to avoid the perfect simulation argument. Rather, we think that there’s just a real similarity between corporate and individual actions which can be spotted by people independent of their commitments on de se matters, and then reflection on that perceived similarity can be leveraged into evidence against the essentiality of the de se. Perhaps we are wrong about that. If so, we do indeed face a perfect simulation challenge to our ability to draw a meaningful distinction between the deflationary and the inflationary account of action. We can then at least take comfort in knowing that our company is good. As we’ve noted, a broad range of deflationary projects face challenges of this sort. We think there are promising routes for responding to perfect simulation arguments where they do have bite, but this is not the place to pursue a project of that size. For now we rest content with the possibility that the de se deflationist does not outright win the day, but merely makes the case that there is no substantive difference between the presence and absence of de se mental states.

Notes 1 The seminal texts are by Perry (1977, 1979), and Lewis (1979, 1986). They both acknowledge the work of Castañeda (e.g. 1966)  as an important precursor. More recent fans of the de se include Bermúdez (2017), Prosser (2015), Ninan (2010), Torre (2010), while work that puts the de se to work in diverse areas of philosophy includes McGlynn (2016), Paul (2017), and Titelbaum (2017). 2 For example Boer and Lycan (1980), Millikan (1990), Spencer (2007), Cappelen and Dever (2013), Magidor (2015). See Ninan (2016) for more references, some quotes, and a useful framing of the debate pro and con. 3 There is of course room for skepticism both about whether Apple genuinely performs actions and about whether, if it does, those actions are underwritten by mental states like belief and desire. For some skepticism about the skepticism, see section 3 below, especially the initial pro tanto case for corporate action and the closing comments on corporations and the mental. 4 There’s a technical sense of ‘corporation’ where Apple’s employees don’t count as members of the Apple corporation. That’s not how we use the term. 5 We start with the paradigm case of traditional corporations, but the category that interests us is not limited to such cases.Traditional corporations have a clear member/​nonmember distinction, but we do not insist on this feature in all cases. Traditional corporations come with a hierarchical structure that may include designated decision-​makers and dedicated spokespersons authorized to speak on behalf of the corporation, but we do not insist on this feature. 6 We can separate the action of raising/​lowering the price (an action making a normative change in what customers should be charged) from the action of charging customers differently.This second action might also be underdetermined by any fully mental supervenience base, if customer charges are automatically linked to the combination of a coin flip outcome and P1 and P2’s statements. 7 Cappelen and Dever (2013) note that there’s a harmless deflationary version of this proposal, in which you think of objects in terms of their relation to you in the same, non-​indexical, way that you might think of objects in terms of where they are in relation to the Eiffel Tower.We are targeting the non-​deflationary version in the text. 8 How then does Apple gain information about the world? It’s not our goal here to develop an account of corporate epistemic faculties. Perhaps Apple has faculties that are in some broad sense perceptual; perhaps

512

Corporate Agency and the First Person obtaining contingent information about the world need not route through even broadly perceptual faculties. Whatever the final story is, though, there is no reason to think that it involves distinctively egocentric information. 9 Should Apple employees count as part of “Apple’s body”? Perhaps some employees can be controlled by Apple, and others can’t. Researchers might have research autonomy included in their contracts, so that their research activity can’t be controlled by Apple, while salespeople might be contractually obligated to follow company policy. And employees might frequently and unpredictably move from category to category. Where there is control, it could be idiosyncratic in pattern. Perhaps the spokesperson can be told what policy to announce, but not what specific words to use in announcing it (except perhaps for the adjective “vibrant”) or what clothes to wear when announcing. Should Apple’s ad agency count as part of “Apple’s body”? The ad agency isn’t a corporate part of Apple, but it can in certain ways and under certain circumstances be directed by Apple. 10 “Internally directed” in that they represent things as internal, or that they normally represent the internal. This is not meant to deny that things other than the body can be represented in proprioception. 11 Moreover, these “act” and “action” uses are clearly distinguishable from non-​agential uses such as “The refrigerator acted strangely last night”. Actions of both agents and corporations can be modified by various intentional adverbs (“John/​Apple acted capriciously/​unfairly/​compassionately in firing Bob”) while actions of artifacts and other objects cannot (“The refrigerator/​cold front acted capriciously/​unfairly/​compassionately in lowering the temperature”). Actions of both agents and corporations are performed for reasons, while actions of refrigerators and cold fronts are not. (Thanks to Matthew McKeever for discussion on this point.) 12 The ’hand division’ could consist of little finger gnomes that move the fingers around—​as suggested to us by Michael Murez. Note the difference between this scenario and Block’s (1978) Homunculi-​Headed Robots. We allow the possibility that you, as “CEO” have complex mental states, including intentions to act. It’s only the implementation in action of these intentions that are then handed off to the corporate underlings.There’s no aspiration here to give a generic functional recreation of a person in terms of homunculi with a “low level of intelligence”. 13 That is, we take most philosophical opposition to genuine corporate action to be driven by a priori considerations about the nature of action, not by an empirical discovery, through investigation of our own natures, that action must be realized thusly. It can’t be simultaneously (a) epistemically possible that we are corporations, (b) epistemically necessary that we act, and (c) epistemically necessary that corporations don’t act. 14 For an argument that corporations have genuine intentional states, see Tollefsen (2002). 15 This line, in particular as applying to the corporate case, is supported both by philosophical argument and empirical data. Tollefsen (2002) assumes that corporations don’t have phenomenal states, as does List (2018). Interestingly, so do the folk: Knobe and Prinz (2008) reports the results of a study showing people are less likely to attribute phenomenal states to corporations than they are to attribute attentional states to them. 16 Again, this route is also supported by some literature. Schwitzgebel (2015) argues on the basis of cases with some similarities to our corporate examples (the Sirian supersquids and Antarean antheads) that we should take seriously the possibility that full-​fledged phenomenal consciousness is much more common than we ordinarily take it to be. (For example, that it might be a feature of nation-​states.) While we aren’t opposed to pushing these considerations that far, for current purposes all we need is that corporations are phenomenal enough to meet any phenomenal requirements that there are on action.

References Bermúdez, José Luis (2017).Yes, Essential Indexicals Really are Essential. Analysis 77 (4): 690–​694. Block, Ned (1978). Troubles with Functionalism. Midwest Studies in the Philosophy of Science 9: 261–​325. Boer, Steven E. and Lycan, William G. (1980). Who, Me? Philosophical Review 89 (3): 427–​466. Cappelen, Herman and Dever, Josh (2013). The Inessential Indexical: On the Philosophical Insignificance of Perspective and the First Person. Oxford: Oxford University Press. Castañeda, Héctor-​Neri (1966). ‘He’: A Study in the Logic of Self-​Consciousness. Ratio 8: 130–​157. Danto, Arthur C. (1973). Analytical Philosophy of Action. Cambridge: Cambridge University Press. Knobe, Joshua and Prinz, Jesse J. (2008). Intuitions about Consciousness: Experimental Studies. Phenomenology and the Cognitive Sciences 7 (1): 67–​83. Lavin, Douglas (2013). Must there be Basic Action? Noûs 47 (2): 273–​301. Lewis, David K. (1979). Attitudes De Dicto and De Se. Philosophical Review 88: 513–​543. Lewis, David K. (1986). On the Plurality of Worlds. Oxford: Blackwell. List, Christian (2018). What is it Like to be a Group Agent? Noûs 52 (2): 295–​319. Magidor, Ofra (2015). The Myth of the De Se. Philosophical Perspectives 29 (1): 249–​283.

513

Herman Cappelen and Josh Dever McGlynn, Aidan (2016). Immunity to Error through Misidentification and the Epistemology of De Se Thought. In Manuel Garcia-​Carpintero and Stephan Torre (eds.), About Oneself:  De Se Thought and Communication. Oxford: Oxford University Press, 25–​55. Millikan, Ruth Garrett (1990). The Myth of the Essential Indexical. Noûs 24: 723–​734. Ninan, Dilip (2010). De Se Attitudes: Ascription and Communication. Philosophy Compass 5: 551–​567 Ninan, Dilip (2016). What is the Problem of De Se Attitudes? In Manuel Garcia-​Carpintero and Stephan Torre (eds.), About Oneself: De Se Thought and Communication. Oxford: Oxford University Press, 86–​120. Owens, David (2011). Deliberation and the First Person. In Anthony Hatzimoysis (ed.), Self Knowledge. Oxford: Oxford University Press, 261–​277. Pacherie, Elisabeth (2008).The Phenomenology of Action: A Conceptual Framework. Cognition 107 (1): 179–​217. Paul, L. A. (2017). First Personal Modes of Presentation and the Structure of Empathy. Inquiry: An Interdisciplinary Journal of Philosophy 60 (3): 189–​207. Perry, John (1977). Frege on Demonstratives. Philosophical Review 86: 474–​497. Perry, John (1979). The Problem of the Essential Indexical. Noûs 13: 3–​21. Perry, John (1998). Broadening the Mind. Philosophical and Phenomenological Research 58 (1): 223–​231. Prosser, Simon (2015). Why are Indexicals Essential? Proceedings of the Aristotelian Society 115 (3): 211–​233. Recanati, François (2013). Mental Files. Oxford: Oxford University Press. Schwitzgebel, Eric (2015). If Materialism is True, the United States is Probably Conscious. Philosophical Studies 172 (7): 1697–​1721. Spencer, C. (2007). Is there a Problem of the Essential Indexical? In M. O’Rourke and C. Washington (eds.), Situating Semantics: Essays on the Philosophy of John Perry. Cambridge, MA: MIT Press. Titelbaum, Michael G. (2017). One’s Own Reasoning. Inquiry:  An Interdisciplinary Journal of Philosophy 60 (3): 208–​232. Tollefsen, Deborah (2002). Organizations as True Believers. Journal of Social Philosophy 33 (3): 395–​410. Torre, Stephan (2010). Centered Assertion. Philosophical Studies 150: 97–​114. Velleman, J. David (2005). Self to Self: Selected Essays. New York: Cambridge University Press. Wilson, George and Shpall, Samuel (2016). Action. The Stanford Encyclopedia of Philosophy (Winter 2016 Edition), Edward N. Zalta (ed.), URL = https://​plato.stanford.edu/​archives/​win2016/​entries/​action/​.

514

38 SEMANTIC MONSTERS Brian Rabern

If I were to say to you, “I love you”, it would be true just in case the author loves the reader. But imagine we devised a clever way of speaking such that when we preface a statement with the expression “swap”, the contextual interpretation of speaker and hearer are switched. Under this supposition, my utterance of “swap: I love you”, would be true just in case the reader loves the author, i.e. just in case you love me. Are natural languages populated by linguistic devices such as “swap”? Or are such devices exceptions to the semantic norms? In the 1970s, David Kaplan famously bracketed off such devices as linguistic freaks or what he called semantic monsters (Kaplan 1989a, 510). Roughly speaking, a semantic monster is an expression which when affixed to a sentence requires that the sentence be re-​interpreted as if it were uttered in a different context. Of course, we can in theory entertain artificial languages with such operators, but Kaplan claimed that natural language doesn’t have such means of expression.1 (It is contested whether this is supposed to be a contingent fact about natural languages or a non-​contingent fact grounded in the nature of context and content.) In this chapter, I will provide a general overview of the issues surrounding semantic monsters. In section 1, I will outline the basics of Kaplan’s framework and spell out how and why the topic of “monsters” arises within that framework. In section 2, I will distinguish four notions of a monster that are discussed in the literature, and show why, although they can pull apart in different frameworks or with different assumptions, they all coincide within Kaplan’s framework. In section 3, I will discuss one notion that has spun off into the linguistics literature, namely “indexical shift”. In section 4, I will emphasize the connection between monsters and the compositionality of asserted content in Kaplan’s original discussion. Section 5 discusses monsters and the more general idea of reinterpretation or meaning-​shift. Section 6 closes with a brief survey of where monsters may dwell, and pointers to avenues for future research.

1  Kaplan on Monsters Kaplan was initially interested in the semantics of utterances of sentences such as ‘He is suspicious’, where the utterance is accompanied by the speaker pointing at something (Kaplan 1978). Reflection on this case prompts various questions: what does the sentence mean? What is said by the utterance?

515

Brian Rabern

What does the pronoun contribute to the truth-​conditions of the utterance? How does the context of utterance determine what is said? What role does the accompanying pointing play in the semantics? More generally, Kaplan was interested, not just in demonstratives such as ‘he’, ‘she’ and ‘that’, but in a broader class of expression that he called “indexical” expressions. What is common to the words or usages in which I am interested is that the referent is dependent on the context of use and that the meaning of the word provides a rule which determines the referent in terms of certain aspects of the context. Kaplan 1989a, 490 This included ‘I’, ‘you’, ‘he’, ‘she’, ‘this’, ‘that’, ‘here‘, ‘now’, ‘actual’, among others. Kaplan put forward two key principles about indexicals: Principle 1. The referent of an indexical depends on the context of utterance. Principle 2. Indexicals are directly referential, where a term is directly referential just in case the contribution it makes to the content of sentences that contains it is exhausted by its referent (in a context).2 Taken together these principles entail that the contribution that an indexical makes to the content of utterances of sentences that contain it is fixed by the context of utterance. That is, no matter what operators an indexical is embedded under, or no matter what linguistic environment it occurs in, the contribution it makes to the content remains its referent in the context of utterance.3 Principle 1 seems harmless enough: e.g. if you and I each utter ‘I’ we refer to different people, or if you and I each utter ‘that’ while pointing at different mountains we refer to different mountains, etc. The referent of an indexical is sensitive to various features of the context of utterance, and thus the referent of an indexical can vary across different contexts. Kaplan insists that Principle 2 is also obvious, but it has controversial and contested corollaries (in fact, one of the corollaries is the prohibition of monsters). Nevertheless here are some of the obvious sorts of things that Kaplan points to in order to motivate Principle 2—​the principle of direct reference. Consider the following dialogue: David: I love you. Richard: I love you. David and Richard have each uttered the same sentence and the sentence isn’t ambiguous. That is, there is a single English sentence “I love you” with a univocal meaning. Yet, it is an important fact about David and Richard’s communicative exchange that they have each said different things. If they said the same thing, then there has not been an expression of mutual love. (Contrast with Richard uttering ‘You love me’.) But there has been such an exchange, so they said different things. So its not just the referent of an indexical that varies across contexts, it’s also the contribution that an indexical makes to the content of utterances of sentences that contain it. Furthermore there seems to be an important contrast between indexicals and other expressions in terms of their contribution to content. In particular, unlike other expressions such as definite descriptions, the contribution of an indexical never seems affected by a linguistic embedding. Trump is both the referent of ‘the US president’ and the referent of the personal pronoun in the mouth of the current US president, but ‘the US president’ and ‘I’ in the mouth of the current president differ in how they contribute to content. Trump is not always the contribution that ‘the US president’ makes to the content of sentences that contain it. If it was then both (1) and (2) would be true, but (2) is false (on its most natural reading). 516

Semantic Monsters

(1) The US president is Trump. (2) Five years ago, the US president was Trump. Whereas, in line with Principle 2, indexicals seem to contribute their referent even when embedded under shifty operators. Consider Trump’s utterances of the following: (3) I am Trump. (4) Five years ago, I was Trump. Indexicals seem to “cling tightly” to the context of utterance, and Kaplan took this to be a general semantic feature of such expressions. Accordingly, Kaplan insisted that theorizing about indexical languages calls for a distinction between two aspects of meaning. There is what is said by an utterance—​its content in a context—​but there is also the semantic rule that encodes what content an expression has across contexts—​this is the character of the expression. A sentence like “I love you” has a univocal character, and this is the level of meaning that is known by competent language users. This character yields different contents depending on the context, e.g. depending on who utters it. Kaplan models this idea in terms of intensional semantics as follows: a sentence is assigned a character. A character is a function from contexts of utterance to contents. The content of a sentence is itself a function from possible circumstances to truth-​values. Thus, we get Kaplan’s familiar two-​step semantic procedure:

Although often conflated there are actually two independent reasons sometimes cited for Kaplan’s two-​step procedure:4 • a motivation stemming from the compositional interaction of intensional operators and indexicals (i.e. double or multiple indexing); • a motivation stemming from the notion of assertoric content (“what is said”) and its broader role in communication. Let’s consider each in turn. First double indexing: the motivation from the compositional interaction of intensional operators and indexicals doesn’t actually motivate the character/​content distinction (cf. Lewis 1980).5 Instead it calls for points of reference to be “doubly indexed”, in the sense that the semantic index must contain two parameters of the same type, e.g. two time parameters, two world parameters, etc.That a semantics for languages with indexicals embedded under intensional operators requires double indexing was first pointed out by Kamp (1971) with regard to tense logic.6 Consider sentences such as this: (5) Everyone now alive will be dead. What is required to handle such cases is a time parameter t that ‘now’ is sensitive to and an additional time parameter t′ that is shifted by ‘will’. Although double indexing can handle the problematic cases of indexicals embedded under intensional operators, Kaplan insisted that “mere double indexing, 517

Brian Rabern

without a clear conceptual understanding of what each index stands for, is still not enough to avoid all pitfalls” (Kaplan 1989a, 510). The other, more fundamental reason, for Kaplan’s insistence on the two-​step procedure concerns the relationship between semantics and the contents of assertion. Essential to Kaplan’s two-​step picture is a particular view about the division of theoretical labor between the parameters that make up the semantic index. There’s a context, and there’s a circumstance. We need both, according to Kaplan, because we need our semantic theories to be able to capture the two different ways in which, when somebody says (for example), “I love you”, the truth of their utterance depends on the context in which they say it—​roughly, that the situation in which they say it influences both what was said, and whether whatever was said is true. Contexts play a content-​generating role—​resolving context-​ dependence in order to determine what’s said—​and circumstances play a content-​evaluating role—​ they’re the things of which what’s said is either true or false. index = ⟨content-​generating parameters, content-​evaluating parameters⟩ In its content-​generating role, the context provides all the various contextual parameters for the resolution of indexicals and other context-​ sensitive expressions—​ different people say different things, depending on who’s speaking and who’s being spoken to, with “I love you”. In its content-​ evaluating role, the circumstance provides various parameters appealed to in the semantics of intensional constructions. For example, in the evaluation of an utterance of “Necessarily, I love you” the modal operator “Necessarily” checks whether, in every circumstance, things are as that particular utterance of “I love you” represents things as being. If there is a circumstance of evaluation differing with respect to whether things are as the utterance, in context, represents things as being, then the utterance of the modalized sentence is false (and it’s true otherwise). According to Kaplan the two-​ step procedure is crucial, since a central task of a semantic theory is to tell us what sentences say in various contexts—​what propositions or pieces of information do they express in a given context.7 Kaplan takes the motivations for double indexing to be in harmony with the division of parameters into content-​generating and content-​evaluating. Indexicals are sensitive to the first sort of parameter, while intensional operators shift the second sort. The role of context is to generate content. The role of circumstance is to be the sort of object that content is true or false of. When a shifty operator O is applied to a sentence χ the character of χ first uses the context to generate the content of χ in context, and then the shifty operator looks at the profile of the content of χ across various circumstances. Given this sort of picture it could never be that the content of ϕ in context c is the same as the content of ψ in context c, while for an operator O, O(ϕ) and O(ψ) have different contents in context c.8 For example, assume that in context c, Richard says to David “I love you”. In this case, “I love you” and “Richard loves David” have the same content in c. Now consider embedding those sentences under the “swap” operator: (6) swap: I love you. (7) swap: Richard loves David. If the prefix “swap” worked so that the contextual roles of speaker and addressee were swapped, then (6) would be true in c just in case David loves Richard, while (7) would be true in c just in case Richard loves David. So (6) and (7) would have different truth-​values, and thus different contents in c, in spite of their embedded sentences (‘I love you’ and ‘Richard loves David’) having the same content in c. Notice that it follows from this that an operator such as “swap” has to be sensitive to the character of its embedded sentence, not merely its content. Such an operator doesn’t work by shifting a content-​ evaluating parameter, instead it works by shifting a content-​generating parameter—​that is, it shifts 518

Semantic Monsters

the context. This situation would destroy the elegant division of parameters into content-​generating and content-​evaluating. Most importantly, such operators would violate Kaplan’s Principle 2: since the contribution that ‘I’ makes to the content of (6) is not the referent of ‘I’ in c (namely, Richard), indexicals would thereby not be directly referential. For these reasons Kaplan bans such operators from his system:  “My liberality with respect to operators on content … does not extend to operators which attempt to operate on character” (Kaplan 1989a, 510). In spite of Kaplan’s ban, theorists have pointed to various potentially monstrous constructions of English. Consider some examples: (8) Never put off until tomorrow what you can do today. [See Kaplan 1989a, n34]9 (9) Said by a condemned prisoner: I am traditionally allowed to order whatever I like for my last meal. [See Nunberg 1993, 20]10 (10) Said by an amnesiac who thinks he’s a Scottish philosopher: I might be Hume. [See Santorio 2012 for related examples] (11) Only I ordered a drink I liked (no one else did). [See Partee 1989, n3]11 Focus on the last example. Assume three people walk into a pub (A, B, and C).They each order a pint of the local beer. A likes the beer, but B and C don’t. A then says “Only I ordered a drink I liked”. On one reading he’d be saying “I’m the only x such that x ordered a drink that I liked”. (11a)  ∀x(x ordered a drink A liked → x = A)

That reading is false.They all order the local beer, which A liked, so B and C also ordered a drink that A liked. But, in the situation, the naturally intended reading is “I’m the only x such that x ordered a drink that x liked”. (11b)  ∀x(x ordered a drink x liked → x = A)

And this is true, since A liked what he ordered but no one else liked what they ordered. Importantly, on this reading the second occurrence of ‘I’ in (11) does not refer to the speaker of the context. It seems that Principle 2 has been violated; it seems that there is a monster present. But it depends on how the construction is analyzed, and also on background issues concerning the operative notion of “monster”.

2  Different Notions of Monsters Kaplan describes monsters in various ways, sometimes focusing on their formal properties, sometimes on the sorts of effects they’d have on indexicals, context, or content. There are at least four notions of a monster found in Kaplan that are discussed in the literature. Even though these notions all coincide within Kaplan’s particular framework, they can pull apart within different frameworks or with different background assumptions.12 Initially the banned operators are glossed as character-​operators, i.e. operators that are sensitive to the character instead of the content of their embedded sentence. • “… operators which attempt to operate on character” (1989a, 510). • “Operators … which attempt to meddle with character” (1989a, 511). Consider ‘swap: I love you’ uttered in context c. The ‘swap’ operator takes the character of ‘I love you’ and evaluates it at c*, where the speaker and hearer are swapped. Importantly, to work in the intended 519

Brian Rabern

way ‘swap’ must operate on character not content—​its object must be the level of meaning that is unsaturated by context.13 A second pass glosses them as context-​shifting operators: • “such operators as ‘In some contexts it is true that’ …” (1989a, 510). In the example above ‘swap’ shifts the context c to the context c*. Just as standard intensional operators “shift” various parameters such as world and time, monstrous operators shift context parameters. Monsters are also described as indexical-​shifting devices in the sense of operators that shift the referent of “indexicals” and therefore violate the principle of direct reference. • “… indexicals always take primary scope” (1989a, 510). • monsters violate Principle 2 (1989a, 510). • “… no operator can control the character of the indexicals within its scope” (1989a, 510). Given that indexicals are understood as expressions whose content varies with context, then if monsters shift the context they can thereby shift the content (and thus referent) of indexicals. In the presence of monsters, the contribution an indexical makes to the content of sentences that contain it is not exhausted by its referent (in a context). Thus monsters violate Principle 2: the principle of direct reference. Finally, monsters are described as devices that would alter the content of their embedded sentence: • “[an operator], which when prefixed to a sentence yields a truth if and only if in some context the contained sentence … expresses a content that is true in the circumstances of that context?” (1989a, 510). If there was such an operator O, then what a sentence ϕ contributes to what is said by an utterance of O(ϕ) in a context c might not be ϕ’s content at c. Instead the contribution to the content of O(ϕ) in c may be the content of ϕ in some shifted context c*. Monsters then are operators that can shift the contribution that an expression makes to what is said—​in other words, a monster is a shifter of a content-​fixing parameter. Thus we have the following four notions of a monster: • • • •

Characterological operator: An operator that operates in the character of its operand. Context-​shifting operator: An operator that shifts a parameter of the context. Indexical-​shifting operator: An operator that shifts the referent of indexicals in its scope. Content-​shifting operator: An operator that shifts a content-​fixing parameter.

Clearly these notions all coincide in Kaplan’s framework given that the key terms are inter-​ defined: Character is a function from contexts to contents; a context is a sequence of content-​fixing parameters; and indexicals are expressions whose content (and thus referent) varies across contexts. So an operator shifts the referent of an indexical in its scope iff it shifts a parameter of context iff it operates on the character of its operand14 iff it shifts a content-​fixing parameter.15

3  Indexical Shift In connection with the notion of an indexical-​shifting operator the phenomenon of “indexical shift” has become a topic of its own in empirical linguistics. The primary focus in this literature is on variations in the interpretation of certain indexicals embedded under certain attitude verbs.16 520

Semantic Monsters

Indexicals are paradigmatically understood as expressions such as ‘I’, ‘you’, ‘here’, and ‘now’, whose interpretation depends on the context of utterance. A paradigm case of “indexical shift” would be a case when the first personal pronoun is embedded under an attitude verb and can be interpreted as referring to the subject of the attitude verb instead of to the speaker of the context (for early discussion see Anderson and Keenan 1985, 304 and Partee 1989, n2). Consider a case where a man named Hesen says that he himself is rich, that is he utters the sentence “I am rich”. In English, if we wished to report what Hesen said we can’t do so by uttering (12), since it can’t be interpreted as (12b)—​it can only be interpreted as (12a). (12)

Hesen said that I am rich. a. Hesenj said that I (the speakeri) am rich. b. # Hesenj said that he (himselfj) is rich.

But some languages have been reported to differ on this score. That is, in some languages the counterpart of (12) could be used to report what Hesen said. For example, it has been claimed that such readings are permissible in the language Zazaki—​an Indo-​European language spoken primarily in Eastern Turkey (Anand and Nevins 2004). Consider (13), a Zazaki counterpart to (12) after transliteration. (Note that ‘εz’ is the personal pronoun.) (13)

Hεseni va kε εz dεwletia. a. Hesenj said that I (the speakeri) am rich. b. Hesenj said that he (himselfj) is rich.

Anand and Nevins maintain that (13) can be read as (13b), and in this case the referent of the personal pronoun ‘εz’ is shifted (by the attitude verb ‘vano’) from the speaker to the subject of the report.17 Moreover, this sort of optional shiftability in speech reports applies to indexical expressions more generally. All indexical expressions in Zazaki are in principle shiftable.That is, the Zazaki counterparts to English I, you, here, and yesterday all have the option of shifting when within the scope of the verb vano (meaning ‘say’). Anand and Nevins 2004 This sort of phenomenon of indexical shift has been documented in various languages around the world. Deal says that “the phenomenon has been reported for languages spanning five continents and at least ten language families” (Deal 2020, 3–​4). Just to get a sense of how widespread this is it includes the following languages: Amharic, Navajo, Slave, Tamil, Korean, Japanese, Farsi, Zazaki, Nez Perce, Turkish, Uyghur, and many more. (See Deal 2020, n3, for references.) There is by now a library full of empirical data concerning indexical shift. This data doesn’t just involve a list of languages where shifting is observed; there is also detailed descriptions of the sorts of variation found. For example: which indexicals can be shifted in a given language, which attitude verbs can induce the shift, whether the shifting is optional or required, and what constraints are involved (e.g. if one indexical in a given clause is shifted is every other indexical also shifted, see Anand and Nevins (2004) on Shift Together). A goal here would be to systematize the wide array of cross-​linguistic data and to then put forward an overarching syntactic/​semantic package that best covers the data. See, e.g. Deal’s new manuscript A Theory of Indexical Shift (2020). The issue of indexical shift is clearly relevant to the broader discussion of semantic monsters. But sometimes it is unclear what the theoretical upshot is, given that the relevant frameworks don’t always align with Kaplan’s original framework. For example: are the questions being asked under the assumption of a character and content distinction? Is it assumed that the role of the context parameter 521

Brian Rabern

is to be a package of content-​fixing parameters? Is it assumed that an indexical is the sort of expression that has different contents in different contexts? Given that the indexical shift literature makes different assumptions on these sorts of questions there may be some misalignment between their claims and Kaplan’s original ban.18 Here is one sort of tangle one can get in here: if monsters are understood as operators that shift the reference of indexicals or context-​sensitive expressions in particular, one might then think that the definition of a monster is something like this: Contextual reference shift. For a construction Σϕα, which contains a context-​sensitive expression α, Σ is a monster iff Σ semantically shifts parameter z and the extension of α varies across different values for z. This seems to line up well with what theorists in the literature say, and is in line with Kaplan’s discussion. But the definition prompts a further question: what is a context-​sensitive expression? One idea would be the following:19 Type 1 context-​sensitivity. α is context-​sensitive iff there are contexts c and c′ such that ⟦α⟧c,ic ≠ ⟦α⟧c′,ic′. Given this definition, context-​sensitivity is a matter of extension varying with context.The definition of contextual reference shift construed along these lines seems to be in accord with what Schlenker calls the Fixity Thesis: the referent of an indexical is fixed solely by the context, and cannot be affected by any logical operators (Schlenker 2003, 29). To define the class of “indexicals” Schlenker appeals to a definition in terms of context-​dependency: At this point one might be tempted to plead terminological ambiguity. In what sense is [an expression] an ‘indexical’ in Kaplan’s sense? Let us use context-​dependency as a Definition: an expression qualifies as indexical if its semantic value is determined by some feature of a context of utterance. Schlenker 2003, 31 If by “semantic value” Schlenker means referent then his notion of indexicality is in accord with type 1 context-​sensitivity. But then too many expressions will count as “indexicals” (e.g. non-​rigid expressions such as “the president” or contingent sentences such as “Obama is a Democrat” vary in extension across contexts; cf. Lewis who says “contingency is a kind of indexicality” (Lewis 1980, 82)). And it is uncontroversial that there are operators that shift the referent of expressions that are type 1 context-​sensitive.20 Context-​sensitivity is usually understood to involve more than mere variability of reference—​it is usually understood as a claim about the variability of the content of an expression across contexts. (So perhaps by “semantic value” Schlenker intends content.) A more standard definition would go as follows (where |α|c provides the content of α in context c).21 Type 2 context-​sensitivity. α is context-​sensitive iff there are contexts c and c′ such that |α|c ≠ |α|c′. This definition is good as far as it goes, but since the definition of contextual reference shift employs a notion of context-​sensitivity which appeals to “content”, the definition must build in certain assumptions about the operative notion of content (see Rabern and Ball 2019). Are contents supposed to be objects of the attitudes? Or are they just sets of indices? Is there a character/​content distinction? 522

Semantic Monsters

Those who wish to avoid issues concerning character and content and the definition of indexicals might just list a certain class of expressions: {I, you, …}. Then replace mention of context-​sensitivity in the definition with an appeal to expressions in that class. It seems that the linguistics literature on indexical shift leans in this direction. “Indexical” reference shift. For a construction Σϕα, which contains an expression α ∈{I, you, …}, Σ is a monster iff Σ semantically shifts parameter z and the extension of α varies across different values for z. But the set {I, you, …} is stipulative.Why is it interesting that there are operators that shift the extension of expressions in that class? We need to be told something about the expressions in that class in order for the ban on monsters to have any substance. One way to flesh out the definition is to say that the interesting class of expressions are the ones whose content varies with context, but this again raises the specter of content. Another idea would be to just insist that pronouns are clearly a natural class of context-​ sensitive expression. And while third person pronouns are easy to bind there is a subset of the pronouns (i.e. the first and second person pronouns) that are notable in that they are difficult to bind in English.22 If in certain languages attitude verbs can easily shift first and second person pronouns that is certainly interesting and important. While the issues here lead to various difficulties everyone can at least agree with the following: if there are cases where a genuine first personal pronoun is being used in a sentence and it doesn’t refer to the speaker of the context, that would seem to be a “monster”, if anything is.

4  Monsters and Content Compositionality In Kaplan’s original discussion the existence of monsters and the compositionality of content are intimately related. If the language is compositional at the level of content, then there are no operators that shift content-​fixing parameters, and thus there are no monsters (cf. Rabern 2013 and Westerståhl 2012). Kaplan is explicit that he endorses the compositionality of content:  “The Content of the whole is a function of the Content of the parts” (Kaplan 1989a, 507). Monsters are operators that shift the parameters upon which semantic interpretation depends in a way that alters the assertoric content of their embedded clauses. They require reinterpretation of their operands, in the sense that they dissociate the context of interpretation from the context of utterance: this is the feature that Kaplan found so objectionable. But this relies on a certain picture of semantic metatheory. On Kaplan’s picture contents, i.e. sets of circumstances, play two roles: in addition to being “what is said”, the content of a sentence (in context) also plays the compositional role of being the object of shifty operators; contents are constrained depending on the operators of the language, to be the type of semantic entities that enter into compositional relations with those operators. Rabern (2013) argues that there is an inherent tension in Kaplan’s system stemming from the semantics of pronoun binding or variable binding (see also Rabern and Ball 2019).23 Briefly: Consider (14) which embeds under a quantifier to form (15). (14) Hei is mortal. (15) Every mani is such that hei is mortal. The Tarskian semantics for quantifiers treats them as an assignment-​shifting operator as follows:24 ⟦Every mani is such that ϕ⟧g,w = 1 iff ∀x ∈man, ⟦ϕ⟧g[i/​x],w = 1 Since the content of (14) in a context c is,

|Hei is mortal|c = {w: ⟦Hei is mortal⟧gc,w = 1} 523

Brian Rabern

the quantifier, which shifts the assignment, shifts a parameter upon which the content of its complement depends. The semantic value of “hei is mortal” relevant for the compositional semantics of (15) is not the proposition it expresses at c, and the content of “hei” at c doesn’t enter into the content of (15). Thus, quantificational phrases such as “every man”, or the binders associated with them, are monstrous. This conclusion relies on the assumption that the assignment of values to variables is a content-​ fixing parameter. This is explicitly endorsed by Kaplan (1989a; see p.  546). Since for Kaplan the distinction between content-​fixing and content-​evaluable parameters just is the distinction between context and circumstance, Kaplan suggests that the assignment be understood as a parameter of context (see Kaplan 1989b, 591). Those wishing to avoid the conclusion while retaining Kaplan’s alignment between the objects of assertion with compositional values could instead abandon direct reference. That is, one could insist that the assignment is not a content-​fixing parameter, so that the contents of assertion are sets of world-​assignment pairs (cf. Zimmermann 2012). Instead one could follow Lewis (1980) and insist on a distinction between the objects of assertion and compositional values. Such a framework could have it that some content-​fixing parameters live in the index. And this opens up space for expressions whose content varies with these shiftable parameters: their content varies across index parameters.25 In such a framework, given that the context is no longer the package of content-​fixing parameters quantifiers are not context-​shifting operators but nevertheless come out as monsters in the sense of a content-​shifting operator (see Rabern 2013). Yet once we respect the distinction between the objects of assertion and compositional values, and consider the roles they play with our linguistic theories, this feature of language appears absolutely mundane and expected.26 It is only when one is in the grip of the Kaplanian picture, where “what is said” has a privileged compositional role, that operators on character would appear to be exceptions to the semantic norms.

5  Meaning-​Shifting Operators Monsters are understood to require “reinterpretation” of their embedded clause, and thus they are sometimes described as meaning-​changing operators. For example, Predelli describes what he calls “character shifters”: Because character is semantically primitive, the idea of (non-​trivial) operators on character informally provokes a sort of ‘meaning change’, one unobjectionably expressible in English only with the appeal of pure quotation. Predelli 2014, 392 The devices Predelli has in mind are ones that might change the linguistic meaning of their complement, e.g. an operator that results in ‘dog’ meaning what ‘cat’ means. Predelli detects this notion in Kaplan, but when Kaplan talks about “operations on character” he seems to just mean a context-​ shifting operator. Kaplan contrasts an operator on character with an operator on content. The notion of a “character operator” here is clearly of one that is sensitive to character and looks at its profile across various contexts—​on analogy with a content operator that takes the intension of its complement and looks at its profile across circumstances—​not an operation that reassigns the complement a different character.27 The notion of a “meaning-​shifting” operator, however, is interesting and is at least suggested in connection with diagonal operators of other two-​dimensional frameworks (cf. Stalnaker 1978). Monsters are at least content-​shifters and the connection between monsters and the more extreme “meaning shifters” has certainly been made before (see, e.g., Israel and Perry 1996). In setting out his prohibition of monsters Kaplan seems to also ban all hyperintensional operators. Given that he says “all operators that can be given an English reading are ‘at most’ intensional”, it seems he would 524

Semantic Monsters

have viewed all object-​language hyperintensional devices as exceptions to the semantic norms. But there are various hyperintensional devices that are not context-​shifting devices, such as quotation, or perhaps attitude reports. When the monster prohibition is understood in this more general way, a “meaning-​shifting” operator, would certainly be a violation of that prohibition. This accords with the standard view that the expressions of a language are associated with a fixed linguistic meaning. According to this view, there is long-​term meaning change but the linguistic meaning of a word doesn’t shift as the result of some meaning-​shifting operator. This view, however, makes it difficult to accommodate various linguistic phenomena. For example, given that ‘eye doctor’ and ‘optometrist’ are synonymous how can it be that the following sentences might differ in truth-​value? (16) Rebecca believes that all eye doctors are optometrists. (17) Rebecca believes that all optometrists are optometrists. Assuming compositionality it follows that sentences (16) and (17) must always have the same truth-​value. A natural explanation for why Rebecca doesn’t believe that all eye doctors are optometrists is that she has a false belief about the meaning of ‘optometrist’ (or ‘eye doctor’). If the meaning of ‘optometrist’ could be shifted when it was within the belief context, this would provide a mechanism by which sentences (16) and (17) could differ in truth-​value. In this way, meaning-​shifting operators might provide useful resources for the semantics of various constructions of epistemic discourse (e.g. indicative conditionals, epistemic modals, and attitude reports).28 There are already treatments of attitude verbs that treat them as operators that shift the assignment of values to variables (e.g. Cumming 2008; Ninan 2018; Santorio 2012; Rabern forthcoming). Shifting the assignment is tantamount to shifting the “interpretation” of the variables, and thus assignment shifters are a close, though perhaps more innocent, cousin to meaning-​shifting devices.29 Given that these proposals only work for singular terms, a natural generalization would be to have epistemic contexts shift the interpretation of general terms such as ‘eye doctor’ and ‘optometrist’ as well. Here there is likely to be an interesting convergence of the resulting picture and certain aspects of two-​ dimensional semantics (cf. Stalnaker 1978; Chalmers 2004). Or consider cases of meaning negotiation.30 In a dispute about what should be in the extension of ‘athlete’ someone who is arguing that it only applies to humans might utter the following conditional (see Ludlow 2008). (18) If Secretariat is an athlete, then the greatest athlete of all time was a horse. The semantics of this conditional most naturally involves evaluation of the consequent relative to a shifted (or more discerning) interpretation. There are also cases that concern a more explicit form of meaning-​shift.These involve constructions of semantic stipulation, where one coins new terms or stipulates new meanings for old terms. (19) Let’s call horse appendages legs. Then, how many legs does a horse have?31 It might seem reasonable to answer ‘five’ (the lower limbs plus the appendage of ligaments and long hairs beginning at the coccygeal vertebrae) but if the meaning of ‘leg’ is stable, then the answer should be ‘four’. These types of stipulations are common in both everyday conversations and in theoretical writing, e.g. when we introduce names for objects or concepts. (20) Let us use ‘Julius’ to refer to whoever invented the zipper. (21) The function f has two roots—​call them ‘r1’ and ‘r2’. (Dever 1998, section 2.3.2.4.1.1) 525

Brian Rabern

These constructions of semantic stipulation result in a rearrangement of the relations between words and meanings in the language system, or at least alter the “linguistic scoreboard”. Shifts of this kind are highly reminiscent of certain commands in imperative programming languages, where one might have a command like this: Let ‘a’ = 5. The result would be that the string ‘a’ gets assigned to the value 5. But such an assignment isn’t permanent, since later within the same program there might be the command: Let ‘a’ = 7. A semantic framework built around the semantics of interpretation-​ shifting imperatives (and “mutable variables”) might provide a systemic account of natural language constructions of semantic stipulation (and meaning negotiation).32

6  Where Monsters Dwell The topic of monsters has reared its head in a wide variety of philosophical and linguistic debates. In an important sense the issue concerning monsters is at the very foundations of semantic theory and has been implicit in many early discussions of semantic metatheory. In the early tense logical work on the interactions of temporal operators and indexicals stemming from the work of A. N. Prior, which led to the development of multiply-​indexed semantics, the issue of monsters is just below the surface (see Prior 1968; Kamp 1971; and Vlach 1973).33 Likewise, when Evans (1985) asks whether tense logic rests on a mistake, the final theory he considers, T3, is arguably a monstrous semantic theory, and he notices that it employs what he calls “a hitherto unknown form of embedding”.34 And, of course, Kaplan’s monster prohibition was in part a reaction to the early formulations of semantic theory, and what Kaplan viewed as a misguided attempt to “assimilate the role of context to that of circumstance” (Kaplan 1989a, 509).That is, in so far as a point of reference was to do the work of a context of utterance it is arguable that the frameworks developed in Montague (1968), Scott (1970), and Lewis (1970b) were monstrous. In the intervening literature the issue has come up most explicitly with respect to the semantics of discourse about thought and speech, especially with respect to reports of indexical attitudes (e.g. Israel and Perry 1996 and Schlenker 2003), including reports involving de se attitudes, self-​locating belief, or logophoricity (e.g. Castañeda 1966; Chierchia 1989).35 As mentioned above there has been much discussion in empirical linguistics of indexicals that can shift when embedded in attitude reports, e.g. some much discussed examples include Amharic (Schlenker 2003), Zazaki (Anand and Nevins 2004), and Nez Perce (Deal 2014). For a comprehensive overview see Deal (2020). In theorizing about mental and epistemic discourse monsters have been employed in treatments of various puzzles of substitution in opaque contexts (e.g. Recanati 2000 and Cumming 2008). For similar reasons, linguistic devices that make implicit reference to attitudes, the common ground, or information states, e.g. the semantics of knowledge claims, epistemic modals, and indicative conditionals, have been given a monstrous semantics (e.g. Israel and Perry 1996; Weatherson 2001; and Santorio 2010).36 For example, consider an epistemic use of “I might not be here now”. And the threat of monsters has also been noted with respect to the semantics of embedded knowledge claims and embedded epistemic modals in the literature on relativistic semantics (e.g. Egan et al. 2005 and Ninan 2010). Monsters are also featured in the semantics of metalinguistic discourse, including the semantics of quotation, speech reports, talk about fiction, and ascriptions of indexical validity.37 A key area of the philosophical literature where the topic of monsters (or quasi-​monstrous operators) arose was in theorizing about attributions of apriority with respect to analyses of the Kripkean contingent apriori and necessary aposteriori (see Stalnaker 1978; Evans 1979; and Davies and Humberstone 1980).38 These two-​ dimensional analyses often employ monstrous or quasi-​ monstrous two-​dimensional modal operators. For example, Davies and Humberstone’s fixedly operator F and Stalnaker’s “copy up” operator † (Davies and Humberstone 1980; Stalnaker 1978). In this connection various puzzles arise when epistemic and non-​ epistemic modalities are combined with indexicals. One such puzzle is Fritz’s nesting puzzle for temporal-​epistemic logics 526

Semantic Monsters

(Fritz 2013, 1768). Assume a multi-​modal propositional language with both temporal and epistemic operators. So the language has standard temporal operators such as P (“It has at some time been the case that …”), F (“It will at some time be the case that …”), H (“It has always been the case that …”), and G (“It will always be the case that …”). But also assume it has N, the indexical “now” operator (Kamp 1971). The language also has an epistemic operator, K, for “the agent knows that …”. We assume perfectly rational agents, so that the agent always knows every validity. Or one might insist that we are concerned with the implicit knowledge of a rational agent—​what is compatible with the information available to the agent. In any case, we assume the following logical omniscience principle: eternal logical omniscience: If ⊨ϕ, then ⊨GKϕ

Knowledge is factive, so that if ϕ is known then ϕ, but moreover it seems that factivity holds at all times. That is, it is always the case that if ϕ is known, then ϕ. Call this eternal factivity: eternal factivity: ⊨G(Kϕ → ϕ) While both principles seem plausible, along with common assumptions they entail the absurd principle that if ϕ, then it will always be the case that ϕ. That is, ϕ → Gϕ comes out as valid. A quick derivation is as follows: 1. Np → p 2. GK(Np → p) 3. G(K(Np → p) → (Np → p)) 4. G(Np → p) 5.  p → GNp 6.  p → Gp

[theorem] [1, logical omniscience] [eternal factivity] [2, 3, dist, mp] [theorem] [4, 5, dist, chain]

But of course we don’t want that result. The fact that it is raining certainly doesn’t entail that it is always going to rain. One way out of the puzzle is to reject eternal factivity. For indexical languages the validities are the sentences that are true in every context. To say that the agent knows all the validities, then is to say that if ϕ is true in every context then, the agent knows ϕ. This motivates the idea that K should be interpreted as a quantifier over all the contexts compatible with the information available to the agent.39 On this interpretation K is monstrous. Another way out of the puzzle is to deny the logical omniscience principle. Given that Np → p is valid, a perfectly rational agent should know it, thus K(Np → p) ought to hold as well. But since G(Np → p) is false (let’s assume) we should expect GK(Np → p) to be false as well. For example, assume it’s raining today but not tomorrow, then although “if its raining now, it is raining” is valid (and known), it won’t be true tomorrow—​so then why think it will be known tomorrow? The puzzle here is an instance of a more general puzzle.The nesting problem arises in a multi-​modal logic with operators for metaphysical necessity, □, and epistemic necessity (apriority), ■. Assuming □ distributes, the following are inconsistent (1) ■ϕ ∧¬□ϕ (2) ■ϕ → □■ϕ (3) □(■ϕ → ϕ) Thus, in light of the contingent priori we face a dilemma: Either (i) what is a priori is a contingent matter or (ii) it is possible that something is a priori but false. Neither options looks especially appealing.40 527

Brian Rabern

Acknowledgments For comments and discussion thanks to Derek Ball, Eliot Michaelson, Bryan Pickel, Anders Schoubye, and Wes Holliday.

Further Reading Anand, P. and Nevins, A.: 2004, Shifty operators in changing contexts, Proceedings of SALT 14, 20–​37. Israel, D. and Perry, J.: 1996, Where monsters dwell, in J. Seligman and D. Westerståhl (eds), Logic, Language and Computation, volume 1, Stanford, CA: CSLI Publications, pp. 303–​316. Rabern, B. and Ball, D.:  2019, Monsters and the theoretical role of context, Philosophy and Phenomenological Research, 98(2), 392–​416. Recanati, F.: 2010, Indexicality and context-​shift, URL: https://​jeannicod.ccsd.cnrs.fr/​ijn_​00000552/​document. Santorio, P.: 2019, Context-​free semantics, in E. LePore and D. Sosa (eds), Oxford Studies in Philosophy of Language, volume 1, Oxford: Oxford University Press, pp. 208–​239.

Notes 1 Strictly speaking he said this: “I am not saying we could not construct a language with such operators, just that English is not one. And such operators could not be added to it” (Kaplan 1989a, 510). Although the claim is in the first case only about English, the modal claim suggests the more general thesis that natural languages don’t have such operators. 2 The intuitive idea of “direct reference” is that the term refers directly without mediation by a Fregean sense. Importantly, though, the idea is not that nothing mediates the relation between such an expression and its referent. Note that Kaplan thinks that the relation between an expression like ‘I’ and its referent is mediated by a (descriptive) rule: find the speaker of the context. The key claim is that whatever mediation there may be is “off the record” with respect to the resulting content. As Kaplan says “The ‘direct’ of ‘direct reference’ means unmediated by any propositional component, not unmediated simpliciter” (see Kaplan 1989b, 568–​582). Kaplan is primarily thinking about this in terms of structured propositions, so if a term is directly referential then propositions expressed by sentences containing the term would “contain” the referents directly instead of a proxy or mode of presentation. To be neutral between structured and unstructured conceptions of content I put it in terms of the contribution an indexical makes to the content of sentences that contain it. 3 Quotation is acknowledged to be special and set aside as a harmless exception to this general claim. We can obviously make claims that require evaluating expressions in other contexts by mentioning rather than using the expression. For example, when we say “In all contexts ‘I am here now’ is true”. Or consider the operator V, which is such that ⌜V(ϕ)⌝ is true iff ⌜ϕ⌝ is valid (see Deutsch 1989 on adding an indexical validity operator to Kaplan’s logic). In the natural language setting there is much discussion of the connection between monsters and quotation—​any alleged monster must be inspected to see whether or not its “sneaking in a quotation device” (Kaplan 1989a, 511). See, e.g., Maier (2007) and Maier (2016). 4 See Stalnaker (2018) for a nice discussion of this point. Cf. Rabern (2012a). 5 See Stalnaker (2018), Santorio (2019), and Rabern (2012b, ch. 0). 6 The ancestor to the 1971 paper on ‘now’ is Kamp (1967) “The treatment of ‘now’ as a 1-​place sentential operator”, 1967.This is stored in the Prior Archives at the Bodleian Library Oxford (Box 15). A copy is available here: http://​semanticsarchive.net/​Archive/​Tk3ZmEyN/​. 7 “The idea of Content—​the what-​is-​said on a particular occasion—​is central to my account” (Kaplan 1989b, 568). 8 What is assumed by this picture is that content in a context is compositional (cf. Kaplan 1989a, 507). See section 4, for discussion of the connection between compositionality of content and the monster prohibition. See also Rabern (2012a) and Westerståhl (2012). 9 From Lewis Carroll we find a related example: “You couldn’t have it if you did want it”, the Queen said. “The rule is, jam tomorrow and jam yesterday—​but never jam today”. “It must come sometimes to ‘jam today’ ”, Alice objected. See Through the Looking-​Glass, and What Alice Found There, 1871, Chapter 5. 10 Other examples form Nunberg (1993) include “Tomorrow is always the biggest party night of the year”. 11 Also from Oscar Wilde’s poem Easter Day we find the following: “Foxes have holes, and every bird its nest, I, only I, must wander wearily, And bruise My feet, and drink wine salt with tears”. 12 See Rabern (2012b, ch. 6, “Semantic monsters”), Rabern and Ball (2019), and Yalcin (forthcoming).

528

Semantic Monsters 13 Monstrous constructions are character-​compositional but not content-​compositional. See Rabern (2012a), Westerståhl (2012), and McCullagh (2017). In certain frameworks, this is encoded in the compositional rule called monstrous functional application. 14 Some common assumptions are in the background here. For example, of course one could construe negation as an operation on character, but we want to say that genuine character operators are the ones that could only be specified as taking characters as argument. One might also worry about strange character operators such as Fav, where Fav(ϕ) is true iff the character of ϕ is the speaker’s favorite character. On a first pass this doesn’t look like a context-​shifter at all, but with some ingenuity it can fit the mold. 15 Predelli (2014) claims that Kaplan fails to “provide a univocal definition of a monster and vacillates between at least three non-​equivalent alternatives” (p. 389). See McCullagh (2017) for critique of Predelli’s claim.The first two notions Predelli discusses—​“context shifters” and “global shifters”—​fall under the definitions of a context-​shifting operator provided here. Global shifters are just a certain kind of context-​shifting operator that also shifts the circumstance—​such operators are monstrous in virtue of their context-​shifting aspect.The third alternative Predelli detects is what he calls “character shifters”.These are operators that change the character. These are not what I’ve called characterological operators—​“character shifters” are something more extreme, and I don’t see them in Kaplan. But see section 5 for discussion. 16 It would seem to be an empirical hypothesis that indexical shift can only occur in speech and attitude reports, but some in this sub-​literature seem to take it as definitional that indexical shift could only occur in speech and attitude reports. For example, I  once put forward the example “Only I  got a question I understood (no one else did)” as a clear example of indexical shift, and a linguist present protested that it can’t be such a case given that there is no attitude verb involved! Of course Kaplan’s concern with monsters was not restricted to embedding under attitude verbs, cf. his original “In some contexts it is true that I am not tired now” (Kaplan 1989a, 510). Theorists are of course free to use terminology as they like, as long as they are explicit about it. 17 There are tests that are supposed to rule out the hypothesis that these cases involve tacit quotation. See Deal (2020, section 2.1.1); but see Maier (2016) for disagreement. 18 Von Stechow (2004) expresses some worries along these lines.The sort of issue presented here mostly follows Rabern and Ball (2019). 19 In what follows I use the bracket notation ⟦.⟧c,i to indicate the denotation of an expression at a context c and index i (cf. Scott 1970). I also use ic for the index of the context, the index that has coordinates that match the appropriate features of the context (cf. Lewis 1980). 20 Von Stechow (2004) complains that Schlenker’s (2003) semantics is not actually monstrous since “attitudes quantify over triples that are composed of the same components as context, but they are not contexts of utterances but rather indices” (2004, 479–​480). On Schlenker’s “monster-​friendly” semantics (see Schlenker 2003, 104ff) the attitude operators never shift the context parameter c. Instead they modify the assignment function s, and the personal pronoun is essentially treated as a variable that is sensitive to the assignment.Thus, von Stechow’s complaint is that Schlenker’s semantics shifts indices, and not contexts: “IAmharic” is sensitive to the index and “says” is an index-​shifting operator. Of course, Schlenker’s semantics has operators that shift the referent of the personal pronoun “IAmharic” (or its analog in the formal system “agent(ci)”), so this is why Schlenker insists that it is monstrous. But since Schlenker says nothing about how to define assertoric content in his system it isn’t clear whether the attitude operators in Schlenker’s system merely shift type 1 expressions or whether they also shift type 2 expressions. 21 The distinction here between type 1 and type 2 context sensitivity tracks the distinction between nonindexical and indexical context-​sensitivity (see MacFarlane 2009). If α is type 1 but not type 2 context-​sensitive, it is nonindexically context-​sensitive, if α is type 2 context-​sensitive, then it is indexically context-​sensitive. 22 Perhaps this is how Santorio (2019) sees things. See also Yalcin (forthcoming). 23 For related worries see Zimmerman (1991), Dever (2004), and Yli-​Vakkuri (2013). 24 Of course, standardly the roles of quantification and variable-​binding are separated, such that strictly speaking variable binding is done by a covert abstraction operator, λ (see Lewis 1970b, 45 and Heim and Kratzer 1998, 186). On this construal (15) is actually composed out of the following constituents: “every man” + λ1 + “he1 is mortal”, and λ1 first composes with “he1 is mortal” to provide a predicate meaning to compose with “every man”.To simplify I treat [every man λ1] as an atom which composes with sentence types.This doesn’t impact the key point—​what we say about the variable-​binding operator “every man1 is such that” can instead be made by focusing on the lambda-​binder alone (see Rabern 2013, section 4). 25 See Lewis (1980). Ninan (2010) discusses this under the label “shiftable contextualism”, and Yalcin (forthcoming) under the label “parametric context-​sensitivity”. 26 See Lewis (1980), and for further discussion Ninan (2010) and Rabern (2012a). 27 Given that Stalnaker’s 1978 propositional concepts track variations in linguistic facts, operators that shift the Stalnakerian “context” could result in ‘dog’ meaning what ‘cat’ means—​it is a contingent fact what words

529

Brian Rabern mean. But Kaplan’s characters are importantly different—​they capture the way the content of an expression-​ of-​English (with its fixed linguistic meaning) varies with context, they are not concerned with the variation of metalinguistic (or presemantic) facts and the contingent connections between expressions and linguistic meanings. 28 Note that the literature on the ‘logicality of language’ or ‘natural logic’ often makes use of interpretation-​ shifting devices in the logical form. These are sometimes called “modulation” functions. See, e.g., Del Pinal (2017) and Chierchia (ms). 29 The operators of early modal logic in the style of Carnap were like this. Even the semantics for quantification is sometimes given in terms shifting the interpretation (see Bostock 1997). In this connection see Wehmeier (2018) and Pickel and Rabern (ms). 30 See the literature on metalinguistic negotiation, e.g. Plunkett and Sundell (forthcoming). 31 See Stalnaker (1978, 317) who says, “If you call a horse’s tail a leg how many legs does a horse have? The answer, of course is four, since calling a tail a leg doesn’t make it one. But one can see a different way to take the question”. 32 All forms of semantic stipulation, however, are not this explicit. There are many forms of implicit semantic shifts. For example, if some conversational participants don’t remember the name of a certain bike part, they might spontaneously begin referring to it as the “whatchamacallit”. Here there would be no explicit stipulation but it is clear that “whatchamacallit” would come to have a meaning in this conversation via the cooperation of the interlocutors. This points to a dynamic conversational phenomenon of shifts in the lexicon throughout a discourse, which calls for a logical framework that can model the processes of “on the fly” meaning shift. For discussion of the dynamic lexicon along these lines see Ludlow (2014). 33 In fact, Vlach (1973) introduced the “index operator” K, whose function is to fix the context to which the “then-​operator” refers—​and this is technically a monstrous operator as it shifts the temporal parameter that represents the time of utterance. The issue even comes up in connection to disambiguations of Anslem’s ontological argument (see Lewis 1970a, 185–​186). 34 Evans compares the situation to the following: “Suppose that there is a language exactly like English, save that it possesses two additional operators, ‘To the right’, and ‘To the left’, which can be prefixed to sentences in the first person. A sentence like ‘To the left (I am hot)’ as uttered by a speaker x at t is true iff there is at t on x’s left someone moderately near who is hot” (Evans 1985, 357–​358). The discussion in Evans appears to be independent of Kaplan’s work but the unknown form of embedding Evans is considering is clearly monstrous embedding. See also Evans (2004), which is a letter dated July 14, 1979, written to Martin Davies in response to a draft of “Two notions of necessity”. 35 It is also arguable that the issue is just below the surface in discussions of indexical belief revision and the semantics of credential attributions in relation to Bayesian probability theory; see e.g. Elga (2000) and Chalmers (2011). 36 See also Chalmers (1998) and Nolan (2003). 37 For monsters in the semantics of quotation see e.g. Maier (2007, 2016), Cappelen and Lepore (2003), Geurts and Maier (2005), and Cappelen and Lepore (2007), in speech reports see e.g. Schlenker (2003) and Anand and Nevins (2004), in talk about fiction see Predelli (2008), and on indexical validity see Deutsch (1989). A further fundamental topic where monsters arise is with various phenomena whereby the context of interpretation is dissociated from the context of utterance: the historical present, answering machine cases, talk about fiction, shotgun assertions, free indirect discourse, imaginary contexts, and pretense. See e.g. Predelli (1996, 1998), and Egan (2009) for discussion of the nature of context, shotgun assertions, and delayed assertions. See Predelli (2008) for a monstrous treatment of discourse about fiction and pretense. 38 For a survey of various two-​dimensional approaches to the Kripkean contingent apriori and necessary aposteriori see Chalmers (2006). 39 That is, one could give K this sort of interpretation: ⟦Kϕ⟧w,t,t′ = 1 iff for all contexts c compatible with the information available to the agent (in w relative to ⟨t,t′⟩), ⟦ϕ⟧wc,tc,tc = 1. This is to give up eternal factivity. Assume it’s raining today (tc) but not at some other time t, then G(Nr → r) is false—​Nr is true at t (relative to tc) but r is false at t (relative to tc).Yet, even at t, K(Nr → r) is true since its the diagonalized proposition that is relevant to the K operator. 40 See Dever (2007), Chalmers and Rabern (2014), and Fritz (2013).

References Anand, P. and Nevins, A.: 2004, Shifty operators in changing contexts, Proceedings of SALT 14, 20–​37. Anderson, S. and Keenan, E.: 1985, Deixis, in T. Shopen (ed.), Language Typology and Syntactic Description, volume 3, Cambridge: Cambridge University Press, pp. 259–​308. Bostock, D.: 1997, Intermediate Logic, Oxford: Oxford University Press.

530

Semantic Monsters Cappelen, H. and Lepore, E.: 2003,Varieties of quotation revisited, Belgian Journal of Linguistics 17(1), 51–​75. Cappelen, H. and Lepore, E.: 2007, Language Turned on Itself: The Semantics and Pragmatics of Metalinguistic Discourse, Oxford: Oxford University Press. Castañeda, H.: 1966, ‘He’: A study in the logic of self-​consciousness, Ratio 8, 130–​157. Chalmers, D. J.: 1998, The tyranny of the subjunctive. Unpublished manuscript: http://​consc.net/​papers/​tyranny.html. Chalmers, D. J.: 2004, Epistemic two-​dimensional semantics, Philosophical Studies 118(1), 153–​226. Chalmers, D. J.:  2006, The foundations of two-​dimensional semantics, in M. Garcia-​Carpintero and J. Macia (eds), Two-​Dimensional Semantics: Foundations and Applications, Oxford: Oxford University Press, pp. 55–​140. Chalmers, D. J.: 2011, Frege’s puzzle and the objects of credence, Mind 120(479), 587–​635. Chalmers, D. J. and Rabern, B.:  2014, Two-​dimensional semantics and the nesting problem, Analysis 74(2), 210–​224. Chierchia, G.: 1989, Anaphora and attitudes de se, in R. Bartsch, J. van Benthem and P. van Emde Boas (eds), Semantics and Contextual Expression, Dordrecht: Foris, pp. 1–​31. Chierchia, G.: manuscript, On being trivial: Grammar vs. logic. Cumming, S.: 2008,Variablism, Philosophical Review 117(4), 605–​631. Davies, M. and Humberstone, L.: 1980, Two notions of necessity, Philosophical Studies 38(1), 1–​31. Deal, A. R.: 2014, Nez perce embedded indexicals, Proceedings of SULA 7, 23–​40. Deal, A.  R.:  2020, A Theory of Indexical Shift:  Meaning, Grammar, and Crosslinguistic Variation, Cambridge, MA: MIT Press. Del Pinal, G.: 2017, The logicality of language: A new take on triviality, “ungrammaticality”, and logical form, Noûs. https://​doi.org/​10.1111/​nous.12235. Deutsch, H.:  1989, On direct reference, in J. Almog, J. Perry and H. Wettstein (eds), Themes from Kaplan, New York: Oxford University Press, pp. 167–​195. Dever, J.: 1998, Variables, PhD thesis, University of California, Berkeley. Dever, J.: 2004, Binding into character, Canadian Journal of Philosophy 34, 29–​80. Dever, J.: 2007, Low-​grade two-​dimensionalism, Philosophical Books 48(1), 1–​16. Egan, A.: 2009, Billboards, bombs and shotgun weddings, Synthese 166(2), 251–​279. Egan, A., Hawthorne, J. and Weatherson, B.: 2005, Epistemic modals in context, in G. Preyer and G. Peter (eds), Contextualism in Philosophy: Knowledge, Meaning, and Truth, Oxford: Oxford University Press, pp. 131–​170. Elga, A.: 2000, Self-​locating belief and the sleeping beauty problem, Analysis 60(266), 143–​147. Evans, G.: 1979, Reference and contingency, The Monist 62, 161–​189. Evans, G.: 1985, Does tense logic rest upon a mistake, in Collected Papers, Oxford: Clarendon Press, pp. 343–​363. Evans, G.: 2004, Comment on ‘Two Notions of Necessity’, Philosophical Studies 118 11–​16. Fritz, P.: 2013, A logic for epistemic two-​dimensional semantics, Synthese 190(10), 1753–​1770. Geurts, B. and Maier, E.: 2005, Quotation in context, Belgian Journal of Linguistics 17(1), 109–​128. Heim, I. and Kratzer, A.: 1998, Semantics in Generative Grammar, Malden, MA: Blackwell Publishers. Israel, D. and Perry, J.: 1996, Where monsters dwell, in J. Seligman and D. Westerståhl (eds), Logic, Language and Computation, volume 1, Stanford, CA: CSLI Publications, pp. 303–​316. Kamp, H.: 1967, The treatment of ‘now’ as a 1-​place sentential operator, multilith, University of California in Los Angeles. Kamp, H.: 1971, Formal properties of ‘now’, Theoria 37, 227–​274. Kaplan, D.: 1978, Dthat, Syntax and Semantics 9, 221–​243. Kaplan, D.:  1989a, Demonstratives, in J. Almog, J. Perry and H. Wettstein (eds), Themes from Kaplan, New York: Oxford University Press, pp. 481–​563. Kaplan, D.:  1989b, Afterthoughts, in J. Almog, J. Perry and H. Wettstein (eds), Themes from Kaplan, New York: Oxford University Press, pp. 565–​614. Lewis, D.: 1970a, Anselm and actuality, Noûs 4(2), 175–​188. Lewis, D.: 1970b, General semantics, Synthese 22(1), 18–​67. Lewis, D.:  1980, Index, context and content, in S. Kanger and S. Ohman (eds), Philosophy and Grammar, Amsterdam: Reidel, pp. 79–​100. Ludlow, P.: 2008, Cheap contextualism, Philosophical Issues 18(1), 104–​129. Ludlow, P.:  2014, Living Words:  Meaning Underdetermination and the Dynamic Lexicon, Oxford:  Oxford University Press. MacFarlane, J.: 2009, Nonindexical contextualism, Synthese 166(2), 231–​250. Maier, E.: 2007, Quotation marks as monsters, or the other way around?, in M. Aloni , P. Dekker and F. Roelofsen (eds), Proceedings of the Sixteenth Amsterdam Colloquium, Amsterdam: ILLC, pp. 145–​150. Maier, E.: 2016, A plea against monsters, Grazer Philosophische Studien 93(3), 363–​395. McCullagh, M.: 2017, Kinds of monsters and kinds of compositionality, Analysis 78(4), 657–​666.

531

Brian Rabern Montague, R.:  1968, Pragmatics, in R. Klibansky (ed.), Contemporary Philosophy:  A Survey, volume 1, Florence: La Nuova Italia Editrice, pp. 102–​122. Reprinted in Formal Philosophy: Selected Papers of Richard Montague, 1974. Ninan, D.: 2010, Semantics and the objects of assertion, Linguistics and Philosophy 33(5), 335–​380. Ninan, D.: 2018, Quantification and epistemic modality, Philosophical Review 127(4), 433–​485. Nolan, D.:  2003, Defending a possible-​worlds account of indicative conditionals, Philosophical Studies 116(3), 215–​269. Nunberg, G.: 1993, Indexicality and deixis, Linguistics and Philosophy 16(1), 1–​43. Partee, B.: 1989, Binding implicit variables in quantified contexts, in Proceedings of the Chicago Linguistics Society, volume 25, Chicago, IL: University of Chicago Press, pp. 342–​365. Pickel, B. and Rabern, B.: manuscript, Against Fregean quantification. Plunkett, D. and Sundell, T.:  forthcoming, Metalinguistic negotiation and speaker error, Inquiry:  An Interdisciplinary Journal of Philosophy. Predelli, S.: 1996, Never put off until tomorrow what you can do today, Analysis 56(2), 85–​91. Predelli, S.: 1998, I am not here now, Analysis 58(2), 107–​115. Predelli, S.: 2008, Modal monsters and talk about fiction, Journal of Philosophical Logic 37(3), 227–​297. Predelli, S.: 2014, Kaplan’s three monsters, Analysis 74(3), 389–​393. Prior, A.: 1968, Now, Noûs 2(2), 101–​119. Rabern, B.:  2012a, Against the identification of assertoric content with compositional value, Synthese 189(1),  75–​96. Rabern, B.: 2012b, Monsters and Communication: The Semantics of Contextual Shifting and Sensitivity, PhD thesis, The Australian National University. Rabern, B.: 2013, Monsters in Kaplan’s Logic of Demonstratives, Philosophical Studies 164(2), 393–​404. Rabern, B.: forthcoming, Binding bound variables in epistemic contexts, Inquiry: An Interdisciplinary Journal of Philosophy. Rabern, B. and Ball, D.:  2019, Monsters and the theoretical role of context, Philosophy and Phenomenological Research 98(2), 392–​416. Recanati, F.: 2000, Oratio Obliqua, Oratio Recta: An Essay on Metarepresentation, Cambridge, MA: MIT Press. Santorio, P.: 2010, Modals are monsters: On indexical shift in English, Proceedings of SALT 20, 289–​308. Santorio, P.: 2012, Reference and monstrosity, Philosophical Review 121(3), 359–​406. Santorio, P.: 2019, Context-​free semantics, in E. LePore and D. Sosa (eds), Oxford Studies in Philosophy of Language, volume 1, Oxford: Oxford University Press, pp. 208–​239. Schlenker, P.: 2003, A plea for monsters, Linguistics and Philosophy 26(1), 29–​120. Scott, D.: 1970,Advice on modal logic, in K. Lambert (ed.), Philosophical Problems in Logic: Some Recent Developments, Dordrecht: Reidel, pp. 143–​173. Stalnaker, R.: 1978, Assertion, Syntax and Semantics 9, 315–​332. Stalnaker, R.: 2018, David Lewis on context, in D. Ball and B. Rabern (eds), The Science of Meaning, Oxford: Oxford Scholarship Online. Vlach, F.: 1973, ‘Now’ and ‘Then’: A Formal Study in the Logic of Tense Anaphora, PhD thesis, UCLA. von Stechow, A.: 2004, Binding by verbs: Tense, person and mood under attitudes, in H. Lohnstein and S.Trissler (eds), The Syntax and Semantics of the Left Periphery, Berlin-​New York: Mouton de Gruyter, pp. 431–​488. Weatherson, B.: 2001, Indicative and subjunctive conditionals, The Philosophical Quarterly 51(203), 200–​216. Wehmeier, K.: 2018,The proper treatment of variables in predicate logic, Linguistics and Philosophy 41(2), 209–​249. Westerståhl, D.: 2012, Compositionality in Kaplan style semantics, in M. Werning, W. Hinzen and E. Machery (eds), The Oxford Handbook of Compositionality, Oxford: Oxford University Press, pp. 192–​219. Yalcin, S.: forthcoming, Parametric context-​sensitivity, in E. LePore and D. Sosa (eds), Oxford Studies in Philosophy of Language, Oxford: Oxford University Press. Yli-​Vakkuri, J.: 2013, Propositions and compositionality, Philosophical Perspectives 27(1), 526–​563. Zimmerman, T. E.:  1991, Kontextabhängigkeit, in A. von Stechow and D. Wunderlich (eds), Semantik:  ein internationales Handbuch der zeitgenössischen Forschung, Berlin: De Gruyter, pp. 151–​229. Zimmermann, T.  E.:  2012, Context dependence, in C. Maienborn, K. von Heusinger and P. Portner (eds), Handbook of Semantics, volume 3, Berlin: De Gruyter, pp. 2360–​2406.

532

PART X

Epistemology of Reference

39 CROSS-​CULTURAL SEMANTICS AT 15 Edouard Machery

Do people interpret proper names similarly across languages and cultures? A small, but thriving area in experimental philosophy has examined this question in great detail.1 The goal of this chapter is to review the experimental results in experimental semantics, the controversies they have led to, and the theoretical, experimental, and methodological advances that have resulted. In section 1, I review the original findings reported in Machery, Mallon, Nichols, and Stich (2004) and the efforts to replicate and extend them. In section 2, I discuss the contrast between people’s judgments about the reference of a proper name and their judgments about what a speaker intends to refer to when using a proper name. In section 3, I review the literature contrasting the use of proper names with judgments about what proper names refer to. Finally, in section 4, I examine a philosophical implication of experimental semantics.

1  Experimental Results 1.1  Descriptivism vs. Causal-​Historical Approach In the philosophy of language, theories about the reference fixation of proper names belong to one of two broad theoretical approaches:  descriptivism and the causal-​historical approach. According to descriptivism (e.g., Frege 1892; Searle 1958; Lewis 1970), each proper name is associated with a description; when this description is satisfied by a single individual, the proper name refers to this individual. By contrast, according to causal-​historical theories of reference (e.g., Kripke 1980; Devitt 1981), a proper name may be associated with a description, but, at least after its introduction in a linguistic community, the description plays no role in determining the reference of the proper name; rather, the proper name refers to the individual it is associated with by a historical, causal chain of uses. Descriptivism and the causal-​historical approach are not precise theories; rather, to use Kripke’s apt expression, they are “pictures” of how a proper name gets its reference. Second, as they have been formulated here, descriptivism and the causal-​historical approach are not theories of the meaning of proper names: roughly, they don’t characterize the contribution proper names make to the propositions expressed by the sentences that include them. Rather, they are about reference determination, i.e., about the facts in virtue of which a proper name refers to what it refers to, and thus they belong to metasemantics rather than to semantics.

535

Edouard Machery

Many arguments have been put forward to assess these two approaches to the reference fixation of proper names (more on this in section 4). Of particular significance for this chapter is the appeal to thought experiments, which describe the use of a proper name in a fictional situation. The reader is then meant to assess what the proper name would refer to in this situation.The famous Gödel case in Kripke’s Naming and Necessity illustrates such thought experiments (1980, 83–​84): Suppose that Gödel was not in fact the author of [Gödel’s] theorem. A man called “Schmidt” … actually did the work in question. His friend Gödel somehow got hold of the manuscript and it was thereafter attributed to Gödel. On the [descriptivist] view in question, then, when our ordinary man uses the name ‘Gödel’, he really means to refer to Schmidt, because Schmidt is the unique person satisfying the description “the man who discovered the incompleteness of arithmetic” … But it seems we are not. We simply are not. Judgments about the reference of proper names in these fictional situations are often called “semantic intuitions”, but the label “intuition” may be misleading (Machery 2017). Thought experiments are also found in Evans (1973), Devitt (1981), and others. Kripke’s masterstroke was to devise thought experiments that elicited judgments about reference that were at odds with the simplest forms of descriptivism, although more sophisticated versions can accommodate them (e.g., Jackson 1998, this volume). In the Gödel case, “Gödel” is associated with a description—​“the man who discovered the incompleteness of arithmetic”—​that is fulfilled by the man originally called “Schmidt”. Nonetheless, Kripke and nearly all philosophers of language understood “Gödel” as referring to the man originally called “Gödel”, which happens to be historically connected to contemporary uses of “Gödel”. The verdict about the case seems at odds with at least naïve versions of descriptivism.

1.2  Machery et al. (2004) Machery and colleagues were inspired by Nisbett and colleagues’ work in cultural psychology, showing that people in the West, particularly Americans, and in East Asia tend to differ systematically on a range of psychological measures (e.g., Nisbett 2003). In particular, causal relations seem to be more salient among Westerners than East Asians when people are asked to classify and describe situations. From this, Machery and colleagues hypothesized that Westerners may be more likely than East Asians to make judgments about reference along the lines of the causal-​historical approach (henceforth, “Kripkean judgments”). We tested this hypothesis by presenting participants in Hong Kong (undergraduates at the University of Hong Kong) and in the USA (undergraduates at Rutgers) with four cases: two Gödel cases and two Jonah cases. The first Gödel case closely followed Kripke’s own Gödel case; the second one was structurally similar, but used a Chinese name and character “Tsu Ch’ung Chih”. A Jonah case describes a situation where a proper name is associated with a description that is not fulfilled by anything, including the individual with whom this proper name is historically associated. Machery and colleagues reported finding differences between their two samples for the Gödel cases, but not for the Jonah cases (Figure 1).

1.3  Replicating and Extending Machery et al.’s (2004) Results: The Gödel Case The results reported in Machery and colleagues (2004) have been extended in various respects, some of which will be discussed at greater length in the remainder of this chapter. Beebe and Undercoffer’s (2016) Study 1 successfully replicated the findings of Machery and colleagues (2004) (“Gödel 1 Beebe and Undercoffer 2016” in Figure  1), as have Machery, Sytsma, and Deutsch’s (2015) study (Figure 1). Beebe and Undercoffer’s Study 2 successfully replicated these findings after modifying Machery and colleagues’ two Gödel cases (“Gödel 3 Beebe and Undercoffer 2016” and “Gödel 4 Beebe and Undercoffer 2016” in Figure 1). Beebe and Undercoffer (2015) noted that Machery 536

Cross-Cultural Semantics at 15

74

44

41

65

58

48

43

48

43

37

51

60

48 45 48 37

ild re n

ul ts

ch 8

01

al .2

öd

el

öd

el

Li

Li

et

et

D

al .2

on

ge r

01 8

n

ad

et

et n ge

on D

Va n

G

el

1 G

öd

al .

al .

16

16

20

co er

Va n 1 G

G

öd

el

an e eb

el

4

Be

ffe r

20

16

nd d

U

nd

er

co

ffe r

20

16 ffe r co U d e

an

d eb Be 3

öd G

öd

el

2

Be

eb

e

an

an

d

U

nd

er

co

ffe r

20 er

co ffe r U

nd

er nd e el

Americans

20

15

5

5

01

et a a

m

d an

öd G

G

öd

el

1

Be

eb

e

el el

Be

eb

öd öd

U

ts 2

Sy

ts Sy

1 el G G

l. 2

01

5 et a

a m

ry he

2

öd G

l. 2

01

5 et a

l. 2

01

0 01

l. 2

ry M

ac

he el

1 G

öd

el öd G

67

63 56

39

et a

00

l. 2

et a ry M

ac

he ac

M el

68

55

30

4

4 00

l. 2

al .2

69

56

32

ry he öd G

60

39

et a

et ry

ac

he

M

ac M

hi h

el

g

C

öd G

h’ un C u Ts

62.2

55

G

100 90 80 70 58 60 50 40 29 30 20 10 0

Chinese

Japanese

Dutch

Figure 1  Proportion of Kripkean Answers with Gödel Cases

and colleagues’ (2004) Gödel scenario involves a moral element, a possible issue since moral factors influence many kinds of judgments (e.g., about the intentionality of actions, causal judgment, etc.). In a series of experiments, Beebe and Undercoffer manipulated the moral element of Machery and colleagues’ (2004) original vignettes. For instance, the “Good Gödel” case used in Beebe and Undercoffer’s first study read as follows (2015, 450): Suppose that John has learned2 in college that Gödel is the man who proved an important mathematical theorem, called the incompleteness of arithmetic. John is quite good at mathematics and he can give an accurate statement of the incompleteness theorem, which he attributes to Gödel as the discoverer. But this is the only thing that he has heard about Gödel. Now suppose that Gödel was not the author of this theorem. A man called “Schmidt” actually did the work in question. However, because Gödel was widely recognized at the time of the discovery to be the greatest mathematician in the world and because Schmidt was relatively unknown, many journalists and historians have wrongly attributed the discovery to Gödel. Most people who have heard the name “Gödel” are like John; the claim that Gödel discovered the incompleteness theorem is the only thing they have ever heard about Gödel. When John uses the name “Gödel”, is he talking about: (A) the person who really discovered the incompleteness of arithmetic? or (B) the person to whom the discovery is often wrongly attributed? In the Good Gödel case, the man originally called “Gödel” does nothing wrong. As can be seen in Figure  1, Beebe and Undercoffer’s (2015) Study 1 successfully replicated the pattern found by Machery and colleagues (2004). Sytsma and Livengood (2011) examined a possible ambiguity in the question at the end of vignettes (henceforth “the probe question”): it can ask what the speaker is talking about from the perspective 537

Edouard Machery

of the speaker or from the perspective of the narrator. People across cultures may disambiguate the probe question differently, resulting in a misleading appearance of cross-​cultural diversity. To support their point, they disambiguated the question in various ways, and showed that among Westerners the proportion of Kripkean judgments increases when the probe question is disambiguated. The following disambiguation has been widely used in the literature (the “disambiguated perspective” probe question; italics added): “When John uses the name ‘Gödel’, regardless of whom he might intend to be talking about, he is actually talking about”. Cross-​cultural research has found that the cross-​cultural difference remains when the original probe question is replaced with the disambiguated perspective probe question. Using the disambiguated perspective probe question, Beebe and Undercoffer’s (2016) Study 1C observed the usual cultural difference between Chinese and Americans (“Gödel 2 Beebe and Undercoffer 2016” in Figure 1). Furthermore, Sytsma, Livengood, Sato, and Oguchi (2015) reported finding a similar difference between Japanese and Americans (Figure 1), providing some empirical support to Machery and colleagues’ (2004) original hypothesis drawn from cultural psychology (see also Machery, Sytsma, and Deutsch 2015). Izumi, Kasaki, Zhou, and Oda (2018) noted that English differs from Mandarin and Japanese in the way definite descriptions are lexicalized. Mandarin and Japanese apparently rely on “bare noun phrases”, i.e., noun phrases that are not preceded by an overt determiner. As is the case in other languages (e.g., German), bare noun phrases are ambiguous between various readings, including an existential reading, a definite description reading, an anaphoric reading, and a generic reading. Izumi and colleagues hypothesized that the difference between English, which requires the use of a definite description operator, and Mandarin and Japanese, which don’t, may explain the cross-​cultural difference in judgments about reference. To support this hypothesis, in Study 1 they reformulated the probe question in the vignette used by Sytsma and colleagues (2015) in two different ways: first, they used the so-​called “intermediate” demonstrative sono-​phrases; second, they used a version of the clarified perspective question probe. A sono-​phrase is similar, though not identical, to a definite description in English. They found that these changes increase the proportion of Kripkean answers from 31% to 50%. However, far from supporting Izumi and colleagues’ contention, their own results suggest that the cross-​cultural difference is not due to the role of bare noun phrases in Mandarin and Japanese. First, Izumi and colleagues’ Study 1 is confounded by two changes, and it isn’t clear whether the increased proportion of Kripkean answers is due to the use of a “sono”-​phrase or to the change to the probe question. Second, while Izumi and colleagues didn’t compare Westerners and East Asians, the 50% proportion of Kripkean answers was lower than most proportions found with Western samples. The result of Izumi and colleagues’ Study 2 confirms that their own hypothesis is erroneous since using a sono-​phrase alone had no impact on the proportion of Kripkean judgments. In contrast to the successful replications by, among others, Beebe and Undercoffer and Machery and colleagues (Figure 1), Lam (2010) reported a failed replication. Lam noted that Machery and colleagues (2004) had presented the vignettes in English to both participants in Hong Kong and in the USA. While Hong Kong participants are fluent, they are not native speakers, and the difference in linguistic fluency could confound Machery and colleagues’ (2004) results. To test this hypothesis, he presented a new version of the Gödel case (about “Shakespeare”) in Cantonese to Chinese immigrants in Los Angeles who were native speakers of Cantonese or in English to native speakers of English. Importantly, the probe question of this case differed from the probe question used elsewhere in the literature: When these people use the name “Shakespeare” in a conversation, is their use of the name to talk about A. Shakespeare B. Spencer? Lam reported that both samples made Kripkean judgments, his native speakers of Cantonese more so than his native speakers of English. 538

Cross-Cultural Semantics at 15

Machery and colleagues (2010) have responded to Lam’s criticism in detail. First, they showed that most Hong Kong participants report descriptivist answers when the original vignette is presented in Cantonese to them (Figure 1). As a result, the original results of Machery and colleagues (2004) are not due to participants being more or less competent in English.3 Furthermore, when presented with Lam’s vignette in English, Hong Kong participants responded as they had in Lam’s study, which was presented in Cantonese. Machery and colleagues concluded that the difference between Lam’s (2010) and their results was due to the formulation of the probe question in each study. They further argued that Lam’s formulation of the probe question biases people to give apparently Kripkean answers by both quoting and using the name: it primes people to answer “Shakespeare” when they are asked a question about whom “Shakespeare” may refer to. A more challenging replication failure was recently published by Cova and colleagues (forthcoming). As part of the large replication study of experimental philosophy, Van Dongen, Colombo, Romero, and Sprenger attempted to replicate Machery and colleagues (2004) (https://​osf.io/​ qdekc/​).Van Dongen and colleagues failed to find any difference between their samples of Chinese participants studying in the Netherlands and of Dutch participants (Figure 1). While it is not clear what to make of this result in light of the many other successful replications and in light of the additional successful replications reported by Van Dongen, Colombo, Romero, and Sprenger in their unpublished meta-​analysis (Van Dongen, Colombo, Romero, and Sprenger, ms; https://​osf.io/​et86f/​), the low proportion of Kripkean judgments among their Dutch participants compared to American participants contributes to this result. All the studies discussed so far examined adults’ judgments. Assuming that the difference is cross-​ culturally robust, how does this cross-​cultural difference emerge? Li, Liu, Chalmers, and Snedeker (2018) developed an age-​appropriate Gödel vignette (Super Dog Race), and presented it to children and adults: Long ago, there was a race called the Super Dog Race. Max, Pickles and Blaze participated in the race. Max crossed the finish line first, winning the race, but he got too excited and ran all the way to the North Pole. Pickles crossed the finish line second. He stopped and watched Max run away. The race announcer mistakenly thought that Pickles won the race. He told every newspaper in the world that Pickles won. He also told them that another dog, Blaze, ran very fast despite his short legs. Since then, everyone learned that Pickles won the race. They don’t know anything else about Pickles. Tom and Emily learned at school that Pickles won the Super Dog Race. This is the only thing they know about the dog race and Pickles. They don’t know anything about Max. That night, their dad asked: Do you know who won the Super Dog Race? Tom replied: Blaze was the dog that won the Super Dog Race. Emily said: Pickles was the dog that won the Super Dog Race. Participants had to assess the truth-​value of both statements. The statement about Tom is a control condition, the statement about Emily the target condition. Li and colleagues (2018) reported finding the expected cross-​cultural difference with adults and with 6 to 8 year-​old children (Figure 1). This finding suggests that whatever is the source of the observed cross-​cultural difference is already in place very early on in life.

1.4  Replicating and Extending Machery et al.’s (2004) Results: The Jonah Case Machery and colleagues reported a negative result with the Jonah case, and speculated about the possible sources of this negative result. Beebe and Undercoffer’s (2016) Study 1 obtained different results from Machery and colleagues (2004). Aggregating across the two Jonah cases, they found that most participants gave descriptivist answers (in contrast to the participants in Machery et al. 2004), and they also reported the cultural difference in the expected direction (Figure 2). 539

Edouard Machery 100 90 80 70 60 50 40 30

47 39

39 26

34 25

20 10 0

Jonah 1 Beebe and Undercoffer 2016

Jonah 2 Beebe and Undercoffer 2016

Americans

Study 2 aggregated results (two Jonah cases) Beebe and Undercoffer 2016

Chinese

Figure 2  Proportion of Kripkean Answers with the Jonah Case from Beebe and Undercoffer (2016)

They additionally found a cultural difference in the expected direction in Study 2 (aggregated results with two Jonah cases reported in Figure  2), with two vignettes that modified the probe question of Machery and colleagues’ (2004) Jonah cases. Machery and colleagues used the following probe question: When a contemporary German high-​school student says “Attila was the king who drove the Romans from Germany”, is he actually talking about the wise and gentle nobleman, Raditra, who is the original source of the Attila legend, or is he talking about a fictional person, someone who does not really exist? (A) He is talking about Raditra. (B) He is talking about a fictional person who does not really exist. Beebe and Undercoffer (2016) used instead the following probe question: When a contemporary German high-​school student says “Attila was the king who drove the Romans from Germany”, is he actually talking about the wise and gentle nobleman, Raditra? (A) He is talking about Raditra. (B) He is not talking about Raditra. Further work is needed to determine whether judgments about reference vary cross-​culturally when they are elicited by a Jonah case, and how answers to the Jonah and Gödel cases are related.

2  Speaker’s Reference and Semantic Reference 2.1  Criticism of Machery et al. (2004) Kripke (1977) drew an important distinction between speaker’s reference—​what a speaker intends to refer to when she uses a proper name—​and semantic reference—​what a proper name refers to according to the rules of the language it belongs to. In many circumstances, the two coincide: the speaker intends to be talking about the person who is the semantic referent of the proper name 540

Cross-Cultural Semantics at 15

she uses. But in some circumstances the two can come apart: for instance, in situations of mistaken identities, one can use a proper name to refer to someone who is not the conventional reference of this name. Critics of Machery and colleagues’ results have often argued that the probe question is ambiguous between these two forms of reference (e.g., Ludwig 2007; Deutsch 2009; Vignolo and Domaneschi 2017; Heck 2018): the probe question could be asking about the semantic reference of “Gödel” or about the speaker’s reference of the particular occurrence of “Gödel”. It is unclear how participants disambiguate the question, and it thus unclear whether their answers should be taken to report their judgments about semantic reference. People across cultures may even disambiguate the probe question differently, which would explain the results reported in section 1.

2.2  Theoretical Responses Machery and colleagues have expressed doubts about the plausibility of this response in various ways (e.g., Machery 2011, 2014). Machery and Stich (2012) noted that it only makes sense to ask about the speaker’s reference of an occurrence of a proper name: in a given context, a speaker intends to refer to a particular individual with the utterance of a proper name. By contrast, questions about the semantic reference make sense about proper names qua types. Now, in the Gödel case there is no information about the context of the occurrence of “Gödel” since no occurrence is even mentioned. As a result, the reader could not interpret the question as being about speaker’s reference: the speaker’s reference of which utterance in which context? Heck (2018) has responded to this argument, arguing that in light of the vignette, it is natural for the reader to imagine the vignette’s character making some utterance such as “Gödel must have studied really hard” and to determine that when the character utters that sentence, she intends to be talking about the man who proved the theorem instead of the man who took it (see also Vignolo and Domaneschi 2017).

2.3  Empirical Responses Machery, Sytsma, and Deutsch (2015) have tried to address the concern based on the distinction between speaker’s reference and semantic reference empirically. Studies 2 and 3 attempted to make it clear that the probe question is not about speaker’s reference. In both studies, the probe question read as follows (in English in Study 2 and Chinese in Study 3): When John uses the name “Gödel”, regardless of whom he might intend to be talking about, he is actually talking about: (A) the person who really discovered the incompleteness of arithmetic; (B) the person who got hold of the manuscript and claimed credit for the work. In Study 2, the usual pattern was observed:  Chinese participants tended to report descriptivist judgments, Americans Kripkean judgments (“Gödel 1 Machery et al. 2015” in Figure 1). Study 3 was not comparative, but the proportion of descriptivists among Chinese participants was even larger. Study 4 followed another strategy: it made it clear that the speaker has the intention to refer to the man who stole the theorem. If a participant reports that the speaker is talking about the man who proved that arithmetic is incomplete, then, she must be reporting a judgment about semantic reference.To do this, Machery and colleagues added the following paragraph to the end of the Gödel Case, before the probe question: One night, John is sitting in his room, reviewing for his mathematics exam by going over the proof of the incompleteness theorem. After a while, he says to his roommate, “Gödel 541

Edouard Machery

probably got a huge number of awards from mathematical societies!”When John utters that sentence, he is talking about: Machery and colleagues again found a difference between American and Chinese participants (“Gödel 2 Machery et al. 2015” in Figure 1). In response, Heck (2018) noted that the character in the vignette could intend to refer to the man who proved the theorem since the character believes that the man who got all the awards also proved the theorem. While it is indeed possible for a reader of the vignette to assign this referential intention to the character in the vignette, it seems unlikely: the probe question primes the reader to interpret the character in the vignette as intending to refer to the man who won the awards from mathematical societies, that is to the man originally called “Gödel”. Heck is aware of this response but dismisses it by referring to Sytsma and Livengood’s discussion of the ambiguity between the perspective of the narrator and the character, which was presented earlier. However, Sytsma and Livengood also showed that by using the disambiguated perspective question probe (see above), as was done in the study under discussion, this ambiguity is dispelled, and participants are unlikely to focus on the character’s perspective. In any case, this is testable: we could ask participants to describe who the speaker intends to refer to. I predict they will answer the man who received all the awards for the work.

3  How to Test Theories of Reference? 3.1  Martí’s Criticism In an influential article, Martí (2009) drew a distinction between “two kinds of data”: the use of proper names and people’s thoughts about what a given proper name refers to (see also Martí 2012, 2014). Martí further argued that semantic theories are tested by means of the former, while Machery and colleagues (2004) and, following this article, much of experimental semantics only provide evidence about the latter. The following quotation captures her argument (Martí 2009, 44, my emphasis): The question … is inadequate, for it does not test the right kind of intuitions. It does not test the intuitions that could allow us to tell whether or not the participants in the experiment use names descriptively; rather the question tests their opinions as regards which theory of reference determination they think is correct. Presenting the participants in the experiment with the Gödel/​Schmidt scenario and asking them whether John talks about (i.e. refers) to one or another of two people, described under alternatives A and B, when he uses “Gödel”, invites the participants to reflect on how, in their opinion, reference is determined, and so the question MMNS pose tests opinions as regards which way of thinking, i.e. theorizing, about reference determination is the correct one. Martí’s apparently simple distinction glosses over several distinctions. First, there is a contrast between the use of a proper name and the metalinguistic judgments about this proper name: the latter mention the proper name. Second, metalinguistic judgments could be about various properties: the actual or counterfactual use of this name, its reference, the truth-​value of sentences that contain it, etc. Third, judgments and use may be more or less spontaneous and more or less reflective. It is also unclear why use should be the relevant evidence to assess semantic theories. First, just like syntactic theories, semantic theories are not about use; they are about the semantic properties of words, and use is not a direct reflection of those since it is influenced by many different factors (mistakes, pragmatics, etc.). Second, Martí’s methodological point is at odds with the methods used in every other area of linguistics (Machery 2014): Elsewhere in linguistics, metalinguistic judgments 542

Cross-Cultural Semantics at 15

are used to assess linguistic theories. Synonymy, antonymy, judgments, and coreferentiality judgments, etc., illustrate this use in semantics, acceptability judgments in syntax, and so on. Martí owes us an explanation of why theories of reference are different.

3.2 

Truth-​Value Judgments

Machery, Olivola, and de Blanc (2009) responded to Martí’s challenge by comparing truth-​value and reference judgments. Participants in Mongolia, France, and India were presented with the version of the Gödel case with the Chinese name “Tsu Ch’ung Chih”, and were asked one of two questions (see Machery and Stich 2012, for additional data about the USA): Reference Question (A) the person who (unbeknownst to Ivy) really determined the solstice times? Or (B) the person who is widely believed to have discovered the solstice times, but actually stole this discovery and claimed credit for it? Truth-​value Question Having read the above story and accepting that it is true, when Ivy says, “Tsu Ch’ung Chih was a great astronomer”, do you think that her claim is: (A) true or (B) false? Machery and colleagues found that the proportion of descriptivist answers was roughly the same when presented with either question (Figure 3). Similarly, Li and colleagues asked adults and children whether a statement involving a proper name is true. As noted above, using this dependent variable instead of asking a question about the reference of a proper name, they successfully replicated the basic finding in Machery and colleagues (2004; see Figure 1). These findings put further pressure on Martí since truth-​value judgments are used throughout semantics. Furthermore, Machery and colleagues argued that truth-​value judgments are indicative of how one would use the relevant proper name (for discussion, see Vignolo and Domaneschi 2018; Devitt and Porot 2018; Domaneschi and Vignolo, in press): if one judges that it is true that Tsu Ch’ung Chih was a great astronomer, then plausibly one would use “Tsu Ch’ung Chih” in a descriptivist manner too. It is true that in some circumstances (e.g., when sentences contain 100 90 80 70

67

66 58

60

56

56 48

50 40 30 20 10 0 India

France Truth-value judgment

Mongolia

Reference judgment

Figure 3  Truth-​Value and Referential Judgments (Proportion of Kripkean Judgments)

543

Edouard Machery

indexicals or when speaker meaning differs from sentence meaning) truth-​value judgments do not line up with use, but the sentences used in Machery and colleagues (2009) don’t seem to fit any of these circumstances. Vignolo and Domaneschi (2018) have, however, questioned the connection between use and truth-​value judgments by appealing to the distinction between speaker’s and semantic reference: if the reader takes the user of a proper name to intend to refer to a particular individual that the reader herself does not take to be the semantic reference of this name, she may judge a sentence involving that name to be true, although she would not be disposed to use it in a corresponding way. Furthermore, in an experimental study, participants were given a version of the Gödel case and asked the following questions: TASK 1 (truth-​value judgment) Having read the above story and accepting that it is true, when Mary says, “Gödel was a great logician”, do you think that her claim is (A) true or (B) false? TASK 2 (comprehension) If Mary said to you, “Write Gödel’s biography”, whose biography would you write? (i) The person who really demonstrated the theorem of the incompleteness of arithmetic; or (ii) The person who stole the demonstration of the theorem. TASK 3 (forced-​choice) Suppose you wrote the biography and you send it to the Editor for printing. Which title would you choose for the biography? (iii) “The biography of Gödel”; or (iv) “The biography of Schmidt”. Most participants who answered “true” in task 1 answered (ii) and (iii) in tasks 2 and 3, suggesting that despite answering that the speaker’s sentence, “Gödel was a great logician”, is true, they were not disposed to use “Gödel” in a descriptivist manner. Vignolo and Domaneschi’s (2018) strategy is ingenious, and it shows that Machery and colleagues (2009) were wrong to assume that in their experiments when people assert that the sentence uttered by the vignette’s character is true, they would be disposed to use the relevant name descriptively. That said, in their study 17% of participants who said that the sentence was true also agreed to use names descriptively, which is quite unexpected if the causal-​historical theory describes how competent speakers use proper names. Domaneschi and Vignolo (in press) have also argued that “true” is ambiguous: it can be used to refer to what is true from the narrator’s epistemic perspective or from Ivy’s epistemic perspective. Their three studies used the version of the Gödel case developed by Li and colleagues (see above), but changed the probe questions: Experiment 1 Question 1 That night, their dad asked:  Do you know who won the Super Dog Race? Emily said: Pickles was the dog that won the Super Dog Race. Do you think that her claim is (A) true or (B) false? Question 2 (only for participants who answered “true” to Question 1) You think Emily’s claim is true because A. Pickles did not win the Super Dog Race, but Emily believes that Pickles won the race because her teacher told it to her. B. What Emily believes is a true description of Max and she uses the name “Pickles” to talk about the dog that really (unbeknown to Tom and Emily’s teacher) won the Super Dog Race.

544

Cross-Cultural Semantics at 15 Proportion of Kripkean judgments 100

89

90 74

80 70 60

58

50 40 30 20 10 0 Question 1

Question 2

Question 3

Figure 4  Proportion of Kripkean Answers in Domaneschi and Vignolo (in press)

Experiment 2 That night, their dad asked: Do you know who won the Super Dog Race? Tom said: Blaze was the dog that won the Super Dog Race. Emily said: Pickles was the dog that won the Super Dog Race. Who tells the truth about the real (unbeknown to Tom and Emily’s teacher) winner of the Super Dog Race? A. Tom B. Emily C. Neither Experiment 3 That night, their dad asked: Do you know who won the Super Dog Race? Tom said: Blaze was the dog that won the Super Dog Race. Emily said: Pickles was the dog that won the Super Dog Race. Who reports correctly the story of the Super Dog Race that the teacher told to Tom and Emily? A. Tom B. Emily C. Neither Figure 4 reports the results. Unfortunately, Experiment 1 suffers from severe demand characteristics: asking participants Question 2 invites them to rethink their answer to Question 1.  So, even if participants had descriptivist leanings when answering Question 1, they would be pushed to give a causal-​ historical answer when answering Question 2. Despite being led to give a Kripkean answer, a small proportion of Western (Italian) participants still make a descriptivist judgment. Experiments 2 and 3 tell little about how people understand “true”. The probe question of Experiment 2 is obviously leading participants to give answer C since Max is the dog that Tom and Emily’s teacher are not aware of. Experiment 3 merely asks participants which statement Emily’s teacher could have said.

3.3  Testing Use Directly While Machery and colleagues (2009) tried to examine use indirectly—​by looking at truth-​value—​ experimental semanticists have recently attempted to study the use of proper names directly. In Study 1 of Domaneschi,Vignolo, and Di Paola (2017), participants were presented with fifteen versions of the Gödel case (involving fictitious names of people) such as the following:

545

Edouard Machery

Suppose you are a detective. The people you work for ask you to track down the author of the famous painting The Woman on the Beach, the young Mr. Marco Salvi. While you are conducting your inquiries, you find out that the real painter of The Woman on the Beach is not Marco Salvi, but the old Mr. Eugenio Marini. Participants were then asked, “If you were asked to track down the painter of The Woman on the Beach, Mr. Marco Salvi, who would you look for?”, and they had to choose a picture to answer this question. They found that a majority (58%) chose the Kripkean answer. Study 2 attempted to control for the possibility that apparently descriptivist answers resulted from participants focusing on speaker’s reference rather than semantic reference; it also examined names of people and of locations. Participants were presented with sixteen cases such as the following one: In Sweden the city of Yalmo is known as the place of the 99-​days battle.You find out that, actually, the 99-​days battle was not fought in Yalmo but in the small close island of Grund. They were then asked two questions: If you were asked to go to Yalmo, the place of the 99-​days battle, where would you go? Suppose you visited the place you chose. You would say you have been to: (i) Yalmo or (ii) Grund. For the proper names of people, Domaneschi and colleagues found that a majority gave a Kripkean answer (57%), and that the majority who happened to give a descriptivist answer to the first question used the name historically associated with the reference. For the proper names of locations, they found that only 32% gave a Kripkean answer, but that many who gave an apparently descriptivist answer to the first question used the name historically associated with the reference. They concluded that apparently descriptivist answers in fact express participants’ judgments about speaker’s reference. Domaneschi and colleagues’ (2017) studies fail to establish this conclusion, however. While the vignette specifies a description (“the place of the 99-​days battle”) associated with a proper name (“Yalmo”), it does not make it clear that it is the only description associated with it. The reader can naturally assume that other descriptions are true of the original bearer of the name (the place originally called “Yalmo”); if so, then, even a descriptivist would judge that the description refers to the original bearer of the name, and not to the individual or place that satisfies the description explicitly associated with the proper name in the vignette. In Devitt and Porot (2018), American participants were asked to express their opinion about the character in several vignettes, including “Tsu Ch’ung Chih” in the Tsu Ch’ung Chih version of the Gödel case. For instance, they were presented with a simplified version of this case, and then asked a question eliciting the use of the proper name, two truth-​value judgments, and a referential judgment: 1 2

3

4

EP: Having read the above story and accepting that it is true, what is your opinion of Tsu Ch’ung Chih? TVJ-​D: Having read the above story and accepting that it is true, please indicate below whether you think the following statement is true or false: Tsu Ch’ung Chih was a great astronomer. True /​ False TVJ-​CH: Having read the above story and accepting that it is true, please indicate below whether you think the following statement is true or false: Tsu Ch’ung Chih was a thief and a liar. True /​ False RI: Having read the above story and accepting that it is true, who does the name “Tsu Ch’ung Chih” refer to? a. The man who discovered the summer and winter solstices. b. The man who stole the discovery and took credit for it. 546

Cross-Cultural Semantics at 15 Proportion of Kripkean answers 100

95

95

90

81

80

TVJ-CH

RI

80

89

90

EP

TVJ-RI

70 60 50 40 30 20 10 0

EP

TVJ-D

Study 1

Study 2

Figure 5  Proportion of Kripkean Answers in Studies 1 and 2 of Devitt and Porot (2018)

Devitt and Porot then classified the answers to the EP (for “elicited production”) question as referring descriptively or causally-​historically. They found that in both Jonah and Gödel cases, people use the proper names causally-​historically, make truth-​value judgments in a Kripkean manner (answers to TVJ-​D and TVJ-​CH), and even make Kripkean referential judgments (Figure 5). A problem with this study is that the reader has more information than the character: even if she uses names descriptively to refer to characters in the fictional situation, the reader will update her description when told the story about Gödel and Schmidt. Devitt and Porot are aware of this problem, which they call “the New Meaning objection”, and address it in a follow-​up study, asking participants to “give their opinions of Tsu Ch’ung Chih as if they were speaking to a Hong Konger”. Because a Honk Konger in the fictional situation would not have access to the information given by the narrator of the vignette, Devitt and Porot assume that a descriptivist reader would only associate the description “the man who first determined the precise time of the summer and winter solstices” with “Tsu Ch’ung Chih”. They again report Kripkean use and truth-​value judgments (Figure  5). Their assumption is, however, questionable:  in contrast to the speaker’s intention to refer (which determines the speaker’s reference), the description semantically associated with a proper name may not change as a function of the information available to a speaker’s interlocutors.

4  Philosophical Implication: Assessing Theories of Reference Machery and colleagues (2004) claimed that thought experiments, including the Gödel and Jonah cases, play a central role in assessing theories of reference (an instance of the method of cases discussed in Machery (2017)). On their view, a theory of reference is supported to the extent that it assigns a reference to proper names that is consistent with the reference speakers tend to assign to them in everyday, actual or in fictional situations. The Gödel case, for instance, undermines descriptivism because it reveals a discrepancy between the simpler versions of descriptivism and the reference assigned by Kripke and most, if not all, philosophers of language to “Gödel” in the fictional situation described by the Gödel case. If replicable, the cultural variation in the judgments elicited by the Gödel case and, possibly, the Jonah case suggests, in my view, either that proper names are assigned reference differently in different cultures or that, if proper names refer identically everywhere, the usual methodology of assessing theories of reference in the philosophy of language should be revised (see also Mallon, Machery, Nichols, and Stich 2009). 547

Edouard Machery

Philosophers of language have resisted this conclusion in many ways.4 Some have dismissed the relevance of experimental semantics on the grounds that the judgments of experts (presumably, linguists and philosophers of language) are more evidentially relevant than lay people’s (Ludwig 2007; Devitt 2011, 2012a, 2012b), an instance of the expertise defense (Machery 2017, ch. 5). This alleged role of expert judgments in assessing theories of reference would distinguish metasemantics from the rest of linguistics, where lay people’s judgments are taken to carry evidential weight, although the place of expert judgment in linguistics is complicated. Experts’ linguistic judgments can also be influenced by their theoretical commitments and thus be misleading (Machery and Stich 2012). Some evidence tentatively suggests that this might be the case for referential judgments (Machery 2012a, 2012b). Other philosophers of language have also argued that Machery and colleagues exaggerated the significance of the Gödel case in Kripke’s refutation of descriptivism (Ichikawa, Maitra, and Weatherson 2012) or, more generally, the significance of fictional cases in metasemantics (Devitt 2011; Domaneschi and Vignolo in press).These responses underestimate the importance of judgments about fictional cases to develop a theory of reference (Mallon, Machery, Nichols, and Stich 2013). The crucial point is that Kripke’s causal-​historical theory of reference is, by his own admission, not a theory, but a picture. To turn it into a theory and to distinguish it from competing theories, many fictional cases are likely to be needed, and the findings discussed in this article suggest that such cases are bound to elicit different judgments across demographic groups.

Conclusion There is little doubt that Gödel cases, and perhaps other cases, elicit different answers among East Asians and Westerners. The source of this variation remains, however, debated. Several explanatory hypotheses have been undermined by the growing body of data, but others are still under consideration. Lay people could fail to distinguish speaker’s and semantic reference, and assign speaker’s reference differently across cultures. The role of use to assess theories of reference, and its relation to truth-​value and reference judgments remains controversial. Methodological progress has been made to study the use of proper names rather than simply metalinguistic judgments. Future research will cast more light on these and other outstanding questions. Much is at stake for scientists interested in language since our basic understanding of how reference works depends on answering these questions.

Notes 1 While this chapter focuses on whether proper names are ascribed a reference differently across languages and cultures, experimental semanticists have also been examining how predicates, particularly natural kind terms, are assigned an extension (e.g., Jylkkä, Railo, and Haukioja 2009; Genone and Lombrozo 2012; Nichols, Pinillos, and Mallon 2016; Tobia, Newman, and Knobe, in press; Machery et al., ms). 2 The use of “learn” in this vignette comes from Machery and colleagues (2004). Some have objected that since learning is factive, the whole vignette is contradictory. Whether or not “learn” is factive, similar results are obtained with the Gödel case adapted to China, which does not use “learn”. 3 Incidentally, this result is an additional successful replication of Machery and colleagues 2004. 4 See also Jackman (2009), Ostertag (2013), Cohnitz and Haukioja (2015), Kallestrup (2016), Nado and Johnson (2016).

References Beebe, J. R., and Undercoffer, R. J. (2015). Moral valence and semantic intuitions. Erkenntnis, 80, 445–​466. Beebe, J. R., and Undercoffer, R. (2016). Individual and cross-​cultural differences in semantic intuitions: New experimental findings. Journal of Cognition and Culture, 16(3–​4), 322–​357.

548

Cross-Cultural Semantics at 15 Cohnitz, D., and Haukioja, J. (2015). Intuitions in philosophical semantics. Erkenntnis, 80, 617–​641. Cova, F., et al. (forthcoming). Estimating the reproducibility of experimental philosophy. Review of Philosophy and Psychology. Deutsch, M. (2009). Experimental philosophy and the theory of reference. Mind & Language, 24, 445–​466. Devitt, M. (1981). Designation. New York: Columbia University Press. Devitt, M. (2011). Experimental semantics. Philosophy and Phenomenological Research, 82(2), 418–​435. Devitt, M. (2012a). Whither experimental semantics? Theoria, 27(1), 5–​36. Devitt, M. (2012b). Semantic epistemology: Response to Machery. Theoria, 27(2), 229–​233. Devitt, M., and Porot, N. (2018). The reference of proper names: Testing usage and intuitions. Cognitive Science, 42, 1552–​1585. Domaneschi, F., and Vignolo, M. (in press). Reference and the ambiguity of truth-​value judgments. Mind & Language. Domaneschi, F.,Vignolo, M., and Di Paola, S. (2017). Testing the causal theory of reference. Cognition, 161, 1–​9. Evans, G. (1973). The causal theory of names. Supplementary Proceedings of the Aristotelian Society, 47, 187–​208. Frege, G. (1892). Über Sinn und Bedeutung. Zeitschrift für Philosophie und philosophische Kritik, 100, 25 –​50. Genone, J., and Lombrozo, T. (2012). Concept possession, experimental semantics, and hybrid theories of reference. Philosophical Psychology, 25, 717–​742. Heck, R. G. (2018). Speaker’s reference, semantic reference, and intuition. Review of Philosophy and Psychology, 9, 251–​269. Ichikawa, J., Maitra, I., and Weatherson, B. (2012). In defense of a Kripkean dogma. Philosophy and Phenomenological Research, 85, 56–​68. Izumi, Y., Kasaki, M., Zhou, Y., and Oda, S. (2018). Definite descriptions and the alleged east–​west variation in judgments about reference. Philosophical Studies, 175, 1183–​1205. Jackman, H. (2009). Semantic intuitions, conceptual analysis, and cross-​cultural variation. Philosophical Studies, 146, 159–​177. Jackson, F. (1998). Reference and description revisited. Philosophical Perspectives, 12, 201–​218. Jylkkä, J., Railo, H., and Haukioja, J. (2009). Psychological essentialism and semantic externalism: Evidence for externalism in lay speakers’ language use. Philosophical Psychology, 22, 37–​60. Kallestrup, J. (2016). Counteractuals, counterfactuals and semantic intuitions. Review of Philosophy and Psychology, 7,  35–​54. Kripke, S. (1977). Speaker’s reference and semantic reference. Midwest Studies in Philosophy, 2, 255–​276. Kripke, S. (1980). Naming and Necessity. Cambridge, MA: Harvard University Press. Lam, B. (2010). Are Cantonese speakers really descriptivists? Revisiting cross-​cultural Semantics. Cognition, 115, 320–​329. Lewis, D. (1970). How to define theoretical terms. Journal of Philosophy, 67, 427–​446. Li, J., Liu, L., Chalmers, E., and Snedeker, J. (2018). What is in a name:  The development of cross-​cultural differences in referential intuitions. Cognition, 171, 108–​111. Ludwig, K. (2007). The epistemology of thought experiments: First-​person approach vs. third-​person approach. Midwest Studies in Philosophy, 31, 128–​159. Machery, E. (2011).Variation in intuitions about reference and ontological disagreements. In S. D. Hales (Ed.), A companion to relativism (pp. 118–​136). Oxford: Wiley-​Blackwell. Machery, E. (2012a). Expertise and intuitions about reference. Theoria, 27, 37–​54. Machery, E. (2012b). Semantic epistemology: A brief response to Devitt. Theoria, 27, 223–​227. Machery, E. (2014). What is the significance of the demographic variation in semantic intuitions? In E. Machery and E. O’Neil (Eds.), Current controversies in experimental philosophy (pp. 3–​16). London: Routledge. Machery, E. (2017). Philosophy within its proper bounds. Oxford: Oxford University Press. Machery, E., Deutsch, M., Mallon, R., Nichols, S., Sytsma, J., and Stich, S. P. (2010). Semantic intuitions: Reply to Lam. Cognition, 117, 361–​366. Machery, E., Mallon, R., Nichols, S., and Stich, S. P. (2004). Semantics, cross-​cultural style. Cognition, 92, B1–​12. Machery, E., Mallon, R., Nichols, S., and Stich, S. P. (2013). If folk intuitions vary, then what?. Philosophy and Phenomenological Research, 86(3), 618–​635. Machery, E., Olivola, C.Y., and de Blanc, M. (2009). Linguistic and metalinguistic intuitions in the philosophy of language. Analysis, 69, 689–​694. Machery, E., Olivola, C.Y., Cheon, H., Kurniawan, I. T., Mauro, C., Struchiner, N., and Susianto, H. (ms). Is folk essentialism a fundamental feature of human cognition? Machery, E., and Stich, S. P. (2012). Experimental philosophy of language. In G. Russell and D. G. Fara (Eds.), Routledge companion to the philosophy of language (pp. 495–​512). New York: Routledge. Machery, E., Sytsma, J., and Deutsch, M. (2015). Speaker’s reference and cross-​cultural semantics. In A. Bianchi (Ed.), On reference (pp. 62–​76). Oxford: Oxford University Press

549

Edouard Machery Mallon, R., Machery, R., Nichols, S., and Stich, S. P. (2009). Against arguments from reference. Philosophy and Phenomenological Research, 79, 332–​356. Martí, G. (2009). Against semantic multi-​culturalism. Analysis, 69, 42–​48. Martí, G. (2012). Empirical data and the theory of reference. In W. P. Kabasenche, M. O’Rourke, and M. H. Slate (Eds.), Reference and referring: Topics in contemporary philosophy. Cambridge, MA: MIT Press. Martí, G. (2014). Reference and experimental semantics. In E. Machery and E. O’Neil (Eds.), Current controversies in experimental philosophy (pp. 17–​26). London: Routledge. Nado, J., and Johnson, M. (2016). Intuitions and the theory of reference. In J. Nado (Ed.), Advances in experimental philosophy and philosophical methodology (pp. 125–​154). London: Bloomsbury Academic. Nichols, S., Pinillos, N. A., and Mallon, R. (2016). Ambiguous reference. Mind, 125, 145–​175. Nisbett, R. E. (2003). The Geography of thought. New York: The Free Press. Ostertag, G. (2013). The “Gödel” effect. Philosophical Studies, 166, 65–​82. Searle, J. (1958). Proper names. Mind, 67, 166–​173. Sytsma, J. and Livengood, J. (2011).A new perspective concerning experiments on semantic intuitions. Australasian Journal of Philosophy, 89, 315–​332. Sytsma, J., Livengood, J., Stato, R. and Oguchi, M. (2015). Reference in the land of the rising sun: A cross-​cultural study on the reference of proper names. Review of Philosophy and Psychology, 6, 213–​230. Tobia, K. P., Newman, G. E., and Knobe, J. (in press). Water is and is not H2O. Mind & Language. van Dongen, N., Colombo, M., Romero, F., and Sprenger, J. (ms). Intuitions about the reference of proper names: A meta-​analysis. Vignolo, M., and Domaneschi, F. (2018). Referential intuitions are still problematic. Analysis, 78(3), 472–​483. Domaneschi, F., & Vignolo, M.  (in press). Reference and the ambiguity of truth-​value judgments.  Mind & Language.

550

40 REFERENCE AND INTUITIONS Daniel Cohnitz and Jussi Haukioja

Introduction Our intuitions about reference are often seen to be a crucial source of evidence for theories of reference. For example, Mallon et al. (2009) argue that in finding the correct theory of reference, philosophers are in fact committed to the method of cases: The method of cases: The correct theory of reference for a class of terms T is the theory which is best supported by the intuitions competent users of T have about the reference of members of T across actual and possible cases. Mallon et al. 2009, 338 They even go as far as suggesting that if the method of cases would prove to be unreliable, the most prudent consequence to draw would be “to give up on the idea that the search for a substantive theory of reference is a viable enterprise” (Mallon et al. 2009, 342) and instead endorse some form of deflationism about reference. This is certainly an overdramatic characterization of the role that intuitions play in philosophical theorizing, particularly in the philosophy of language. While Mallon et al. claim that they “have no idea what other considerations philosophers of language might appeal to” (Mallon et al. 2009, 343)  when justifying their theories, we should like to remind them that they themselves make use of such other considerations in that same paper just a couple of pages later, when they argue that a pluralistic view on reference would predict an implausible amount of verbal disagreements. Theories of reference are typically supposed to be able to explain aspects of linguistic understanding, communication and representation. Whether a given theory can in fact explain these aspects and whether it can explain these better than its competitors is a rich source of such “other considerations” that philosophers of language use. However, it is indeed widely agreed that intuitions play some evidential role in the philosophy of language. What is not widely agreed upon is why they play this role, whose intuitions should matter, and which kind of intuitions these are. For example, Mallon et al. 2009 seem to think that philosophers use intuitions as evidence because there is no alternative. If there is no way to independently train or tutor your intuitions, then maybe it doesn’t matter much whose intuitions you

551

Daniel Cohnitz and Jussi Haukioja

consult, and if the intuitions of laypersons vary dramatically then this speaks against the method of cases in general. Michael Devitt (2015, 2006), on the other hand, does believe that, besides intuitions, also other considerations matter. But if there is more than just untutored intuitions to go by, then maybe there can be experts about reference that are better informed by better previous theorizing or training and consequently have better intuitions.These are then the intuitions that should matter most as evidence for and against theories of reference. If it is found that the intuitions of lay people vary a lot, then this doesn’t by itself speak against the reliability of the intuitions of experts. This latter argument is one version of the so-​called “Expertise Defence” against the experimentalist challenge that Mallon et al. (2009) pose for the method of cases. A further disagreement among philosophers in this debate concerns the relationship between the intuitive judgements on the one hand and the facts these judgements are supposed to be about on the other. Some philosophers (again, for example, Michael Devitt) see the method of cases in the philosophy of language to be in this respect very similar to the method of cases in other areas of philosophy or even in the natural sciences. There is a certain objective fact, for instance, whether a certain fossil is a pig’s jawbone (to use one of Devitt’s examples), and an expert palaeontologist may make a reliable judgement about whether the fossil she is looking at is indeed a pig’s jawbone. Experience with identifying jawbones and having more knowledge about them can make you better in getting these facts right. That’s why we may rely on an experienced palaeontologist’s intuitive judgement in such a case over the intuitive judgement of a random layperson. Likewise, if we think that philosophers have reliable intuitions about reference, then that too is due to the fact that they have knowledge of relevant theories and a lot of experience with the metalinguistic task of identifying what words of certain expression types refer to. This contrasts with the view (implicit in Mallon et al. 2009, explicit in Martí 2009) that the source of the (purported) reliability of our intuitions about what terms in an actual or hypothetical case refer to is not our experience with metalinguistic tasks or our knowledge of metalinguistic theories, but our linguistic competence (which we, presumably, fully acquired at a relatively early age). At the same time, the facts that the intuitive judgements are about, e.g. what a term of a certain class in fact refers to in a certain case, are themselves arguably grounded in that same linguistic competence. On this view, intuitive judgements in philosophy of language enjoy an authority that intuitive judgements in other areas maybe lack. At the same time, experts do not necessarily make better such judgements simply because they know more—​what is required for the reliability of the judgement is primarily linguistic competence and this is shared by laypersons and experts. Thus, the relation between intuitions and reference is not straightforward. Which of these views is right? Are facts about reference indeed grounded in individual linguistic competence? How does that square with the widely held view that reference is externally determined and that what an expression refers to may be fully independent from what even very competent speakers believe about the reference of their terms? Furthermore, even if it is the case that facts about reference are grounded in individual linguistic competence, why should we think that this will result in reliable judgements? In the next sections we will take these questions up in the following order: in section 1 we will contrast two views about the relationship between individual competence and facts about reference. On—​what we call—​the meta-​externalist view, the two may be quite independent; on the meta-​ internalist view, by contrast, facts about reference are determined by individual psychological facts. In section 2 we will then explore how, from a meta-​externalist point of view, the relationship between intuitions and reference would look. In section 3 we will do the same from a meta-​internalist point of view. After clarifying the relationship between intuitive competences and intuitive judgements, we conclude tentatively that, on the meta-​internalist view, intuitive judgements in the philosophy of language may enjoy a more privileged status than intuitions in other areas. 552

Reference and Intuitions

1  Meta-​Externalism and Meta-​Internalism How are the facts about reference related to facts about linguistic competence, or facts about the individual psychology of speakers more generally? Here, a distinction can be made between two fundamentally different conceptions. This distinction is often not explicitly drawn, but it has far-​ reaching implications on the methodology of theories of reference, and on the proper role (if any) of intuitions in it. In Cohnitz and Haukioja (2013), we label the two conceptions meta-​internalism and meta-​externalism. Just as the familiar distinction between semantic internalism and semantic externalism, the meta-​ level distinction has to do with the question of whether factors internal to a speaker are sufficient for determining some (meta)semantic phenomenon. The distinction between semantic internalism and externalism, extensively discussed since the 1970s, concerns the question of whether the extension of a linguistic expression, as used by a speaker, is determined by that speaker’s (narrow) individual psychology; according to internalism, it is, according to externalism, it is not. Let us call this distinction the first-​level distinction, and the opposing views first-​order internalism and externalism: Putnam’s natural kind externalism and Burge’s social externalism are familiar examples of first-​order externalism, while traditional descriptivism would be an instance of first-​order internalism. The meta-​level distinction concerns the following question: what makes it the case that a given (first-​order internalist or externalist) theory is true of particular expressions that a speaker uses? Again, an internalist and an externalist answer is possible:  a meta-​internalist will hold that the question of which theory of reference holds of a given expression, as used by a speaker (and thereby the truth of first-​order internalism or externalism concerning that expression), is determined by the speaker’s (narrow) individual psychology, while a meta-​externalist will argue that this is, at least in part, determined by external factors. While the first-​order distinction is concerned with what determines the extension of a term, the meta-​level distinction is concerned with how that extension is determined. These two issues are often not distinguished from each other with sufficient care, and it is therefore not always apparent whether even much-​discussed views in meta-​semantics are committed to meta-​internalism or meta-​ externalism (but we will discuss what we take to be fairly clear cases below). Both internalism/​externalism distinctions can be illustrated in terms of what is or is not possible about internal duplicates. First-​order internalism claims, and first-​order externalism denies, that the referents of expressions uttered by internal duplicates always coincide; meta-​internalism claims, and meta-​externalism denies, that the referents of expressions uttered by internal duplicates are always determined in the same way, and hence that the same theories of reference are true of the relevant expressions. It is crucial to notice that the two distinctions are logically separate. In particular, first-​order externalism does not entail or presuppose meta-​externalism: indeed, in Cohnitz and Haukioja (2013) we argue for a combination of meta-​internalism and first-​order externalism (see also Biggs and Dosanjh, this volume). On this combination of views, first-​order externalism is true of at least some of the referring expressions we use, but the fact that it is true of a given expression, as used by a particular speaker at a particular time, is determined by the individual psychology of the speaker in question. As will become apparent below, on a meta-​internalist view the truth or falsity of first-​order views is an empirical question concerning systematic patterns of dispositions among actual speakers; what is crucial to note here is that there is no logical entailment from the meta-​level views to the first-​level views.

2  Meta-​Externalism and Intuitions As we said, meta-​externalism denies that the (narrow) psychological states of a speaker determine which theory of reference is true of the expressions used by that speaker: external factors also play a role in determining this. In this section we will look at some candidates for what these external 553

Daniel Cohnitz and Jussi Haukioja

factors could be and discuss in what way intuitions still might play a role in investigating which theory of reference is correct. The candidates we look at are “ideal-​types” in the sense that they are relatively schematic, while “real-​life” theories may be more subtle and sophisticated. This is fine for our purposes here: we do not here try to show that meta-​externalism is mistaken, we are merely providing an overview of the complex role that intuitions can play in investigating theories of reference.

2.1 Platonism The first candidate we want to consider is what may be called reference Platonism. According to such a view, the relation between expression types (say, natural kind terms) and object types (say, natural kinds) is, as a relation between abstract objects, itself an abstract object and thus independent of any empirical matters. The independence runs both ways, of course: the nature of the reference relation is independent of speaker dispositions, empirical facts about use, etc., and also causally isolated from affecting those dispositions. This view is meta-​externalist, since it locates the facts that determine which theory of reference is correct outside the narrow psychological states of speakers, it places them instead in a third, platonic realm. The proper way to investigate semantics is then an a priori investigation of these abstract metaphysical facts.This is a non-​naturalist conception of the philosophy of language, as envisaged by, for example, Jerrold Katz (1990). On such a non-​naturalist conception, philosophy is after synthetic a priori truths (such as the truth about the nature of the reference relation) with the help of rational intuitions. When and why rational intuitions are a reliable guide to synthetic a priori truths is then to be sorted out in a non-​naturalist epistemology. Intuition is often understood in parallel to perception. According to such a view, intuition allows us to “see” with the mind’s eye into that third, platonic, realm. Such a non-​naturalist epistemology needs to explain how we can distinguish between intuiting (in the relevant way) that something is the case and merely believing it. Distinguishing features of rational intuitions can be sought in the phenomenology of these intuitions. For example, intuitions may present their content as (necessarily true). The method of cases is then a systematic way of eliciting these intuitions, as part of a more general non-​naturalist methodology in philosophy. While the role of intuitions for the study of reference is clear on this conception (in fact, on this conception there is no other way to study the reference relation, given the construal of it as a non-​ natural subject matter), the epistemology that would justify this approach is, we believe it is fair to say, less than clear. Katz himself, who argues for the view by arguing against the tenability of the naturalist alternative, explicitly admits that he lacks a positive account (Katz 1990, 313).

2.2  Reference as a Natural Phenomenon A meta-​externalist account doesn’t have to be anti-​naturalist, of course. On a naturalist, meta-​ externalist view (exemplified by Devitt 2006, 2015 and elsewhere) the reference relation is a relation between physical events (utterance tokens, patterns of use, etc.) and objects in the world, to be studied empirically like all natural phenomena. Intuitions may still have a role to play in the study of reference, at least in the early stages of the inquiry (similarly as in naturalized epistemology, cf. Kornblith 2002). However, “intuitions” are now not the rational intuitions of the non-​naturalist, but rather ordinary “central-​processor” judgements that are laden with background knowledge and experience (in that sense they are not sui generis experiences). The more experience the intuiter has, and the better her internalized background knowledge is, the more reliable these intuitive judgements are likely to be. Thus, the expert’s intuitions—​that is, the expert’s judgements concerning what a given referential expression refers to in a given situation—​are more trustworthy than the non-​expert’s corresponding intuitions. However, a serious study of reference should eventually move beyond the reliance on intuitions (even the experts’), and study reference more directly by empirical means. 554

Reference and Intuitions

In contrast to the Platonist conception, intuitions are here not the only evidence that we can use for assessing theories of reference: they are not even the best evidence to consult. After all, the exact provenance of our intuitions is opaque, we can only hope that experience and background knowledge have shaped them into a reliable indicator of the reference relation. Unless we have independently confirmed that our intuitive central-​processor judgements tend to be right, we should rather rely on other ways of finding evidence about reference.

2.3 Inscrutability Both conceptions that we have considered so far, the Platonist and the naturalist, hold that there is a stable phenomenon, reference, that can be systematically studied. Meta-​meta-​semantic considerations (that is, considerations about what determines the meta-​semantics of terms) might lead one to abandon that view, though. On Herman Cappelen’s meta-​meta-​semantic view, reference is determined by—​what is probably—​a complex net of linguistic interactions (Cappelen 2018, 63). Which these are, however, is too complex to pin down. The view is meta-​externalist, since it locates the facts that determine how the reference relation operates outside of the speaker, in the linguistic community as a whole. Speaker intuitions are therefore also not authoritative for how expressions refer. In principle, we can be vastly mistaken about how the reference relation operates. Cappelen writes: Most or even all speakers of the language can believe that a predicate F applies to an object, o, but be wrong. They can all want o to be in the extension of F, but wanting o to be F doesn’t make it so. They can all be disposed to apply F to o even though o isn’t F. Cappelen 2018, 63 The only reliable “intuitive” claims about reference are purely disquotational observations such as that “cat” refers to cats (Cappelen 2018, 83). On this view, intuitions about reference have absolutely no evidential value for theories of reference. On the other hand, there isn’t much to find out either: according to Cappelen, reference is inscrutable and too messy to fit in a theory. At best, there is one role that intuitions can play: namely show us that more substantial theories of reference are mistaken. Cappelen’s deflationist view is, after all, itself based on thought experiments by Kripke, Putnam, Burge and Williamson. But these are all thought experiments that present counterexamples to substantial theories rather than establish positive substantial theories of their own.

3  Meta-​Internalism and Intuitions On the meta-​internalist picture, many of the considerations mentioned above can also apply. In particular, the meta-​internalist can agree with the naturalist meta-​externalist that experience can lead to expertise that allows for making reliable judgements about how words in fact refer, only disagreeing about what ultimately constitutes the subject matter of these judgements. Moreover, a meta-​internalist, may, just as Cappelen, end up doubting that there is enough systematicity in the reference-​determining facts for there to be a stable phenomenon of reference in the first place, only disagreeing about where we should look for such reference-​determining facts. However, meta-​internalism also holds that it is our individual linguistic competence that grounds the reference relation. To understand how that allows intuitions to play another role in the investigation of reference, we need to clarify what we can mean by “intuitions” here.

3.1  Different Kinds of “Intuitions” The method of cases is supposed to elicit intuitive judgements about what words mean in certain hypothetical utterance situations. Typically, a hypothetical scenario is described by stipulating certain 555

Daniel Cohnitz and Jussi Haukioja

facts about the utterance situation and (perhaps) about the speaker. If the method is applied in order to test theories of reference, the intuitive judgement elicited is typically supposed to be about what an expression refers to, or what expression would be correct to use in the hypothetical situation described. That is how the familiar examples by Kripke (e.g. the Gödel/​Schmidt-​case) and Putnam (e.g. H2O on Earth vs XYZ on Twin Earth) work. However, what this task actually involves can be understood in various different ways, and some of the conflicts about the role and value of intuitions for assessing theories of reference seem to us to be due to an insufficient appreciation of these differences. To clarify this, we shall in the next two subsections distinguish between different notions of “intuition”, and then return (in section 3.2) to how we think the role of intuitions in such thought experiments should be understood.

3.1.1  Intuitions and Intuition Reports The metaphilosophical discussion of intuitions typically focuses on “intuitive judgements”, that is, on a certain type of propositional attitude the content of which can linguistically be expressed by a that-​clause. But this runs together the outputs of an intuitive capacity with the linguistic report of it. It is helpful to first distinguish intuitive capacities and the outputs of these capacities. We can do a lot of things intuitively: we can intuitively produce and interpret utterances, we can intuitively estimate the distance of an approaching vehicle, we can intuitively predict the trajectory of a thrown baseball. These are intuitive capacities in the sense that we can exercise them without reasoning or conscious reflection: they are spontaneous, effortless and “automatic”. The outputs of intuitive capacities can be of a wide variety: they can be expectations and beliefs, but they can also be feelings or certain motor-​responses, for example when a native speaker of English says the word “yes” as a spontaneous expression of agreement with her interlocutor in a conversation in English. As this list shows, not all outputs of intuitive capacities are judgements or even have (partial) propositional form. But that doesn’t mean that we can’t report them with the help of a that-​clause. Let us consider the intuitive capacity to catch a ball. Reed et al. characterize that capacity as follows: [The skill] gives rise to a phenomenal sense of intuition. If you know how to catch and a ball is thrown towards you, you get an immediate feeling that you should run backwards or forwards to catch it.You do not do any conscious computation. Reed et al. 2010, 64 Here, the output of the intuitive capacity is a feeling that is triggered by certain cues. Of course, when presented with those cues (or shortly afterwards) one can try to report what that feeling was with a that-​clause (e.g. ‘I have the intuition that I have to run forwards to catch the ball’). Even if that intuitive capacity is reliable (people catch balls when provided with those cues), it is not thereby guaranteed that the reports of the outputs of that capacity are accurate. As Reed et al. (2010) show, in the case of the intuitive capacity of ball-​catching people are relatively bad at reporting the outputs of their capacity. This might be due to the fact that the relevant outputs are in part sensorimotor reactions that are not accessible to consciousness. Arguably, the outputs of our intuitive capacity to interpret expressions in utterances is more reliable, since these are accessible to consciousness (for example, upon hearing and understanding an utterance you can consciously reason about its content parts, e.g. upon hearing “Jim is in Barcelona”, you can infer that Jim is in Spain). Thus, we can distinguish between feelings, seemings and various kinds of propositional attitudes as the outputs of an intuitive capacity and the reports of these outputs (and there is an open question about the reliability of the reports concerning the outputs which is independent of the relevant reliability or quality of the outputs). Data about the outputs of intuitive capacities may be relevant for theories of reference, even if it turns out that we are not reliable in reporting them: in this case they should be studied by other means. 556

Reference and Intuitions

3.1.2  First-​and Second-​Level Intuitions A second important distinction concerns the subject matter of the intuition in relation to the subject of study. Let’s assume that the subject of study is itself an intuitive capacity—​for example when you study how people intuitively catch balls or how people intuitively ascribe beliefs and desires to others, etc. In this case, intuitions—​as the outputs of the capacity that you are trying to understand—​are constitutive of the subject matter of the theory you are trying to develop. In line with the terminology that we developed in Cohnitz and Haukioja (2015) we call that the “constitutive role” of intuitions. The intuitions that play that constitutive role, we will call “first-​level intuitions”—​these are the outputs of the intuitive capacity under study. Now, in addition to these first-​level intuitions, one can also have second-​level intuitions, i.e. intuitions about what the outputs of an intuitive capacity are. Thus, second-​level intuitions are intuitions about intuitions. For example, a researcher developing a theory of how people intuitively ascribe mental states to others may have intuitions about how they do that. These intuitions may be her starting points for theory building, but they are different from the intuitions that constitute the subject matter of that theory, namely the intuitive ascriptions of mental states themselves. Our researcher will also have intuitions of the latter kind. Our researcher can use her first-​level intuitions as evidence for her theory, if she has reason to believe that the intuitive capacity is universally shared (and that she possesses the capacity), she can use her second-​level intuitions as evidence to the extent that she has reason to believe that her intuitions are reliable.

3.2  First-​Level Intuitions and Meta-​Internalism As we saw earlier, meta-​internalism holds that the facts about reference—​which theory of reference is true of a given referring expression—​are determined by the internal states of the relevant speaker at the time of utterance. Meta-​internalism, as such, is not committed to any specific view regarding which internal states of a speaker determine the facts about reference. The most natural view about this question, and the one that we have defended at length in Cohnitz and Haukioja (2013 and 2015), is that it is the speaker’s dispositions to apply and interpret expressions of the relevant type, as well as her dispositions to revise her application and interpretation in response to relevant new information, that determine how the expression in question refers (among other things, whether a first-​order internalist or externalist theory of reference is correct of that expression). Such dispositions to apply and interpret linguistic expressions, as well as to revise one’s application and interpretation, are “intuitive” in the sense discussed above: they are spontaneous, and not the outcomes of a conscious reasoning process. Adopting the terminology introduced in the previous subsection, we can then say that, on this view, the speaker’s first-​level intuitions constitute the facts about reference: theories of reference are really theories about (presumably) relatively stable and systematic patterns of first-​level dispositions possessed by the relevant speakers, to apply and interpret the relevant referring expression. The relevant first-​level intuitions here include dispositions to use one name rather than another to convey information about a given individual, dispositions to interpret a general term, when applied to an object, as attributing one property rather than another to it, and so on. However, the set of relevant first-​level intuitions is not limited to such application and interpretation dispositions, but also includes the speaker’s dispositions to revise her usage in response to new information. The distinction between first-​order externalist and internalist theories arises within meta-​internalism as a distinction between different kinds of dependence relations holding between the relevant speaker’s first-​level intuitions and her beliefs concerning external matters of fact. For example, to say that (first-​order) social externalism is true of an expression, as used by a speaker, is to say that that speaker’s application and interpretation of the term is, in complex but systematic ways, dependent on her beliefs regarding how other members of her speech community use the 557

Daniel Cohnitz and Jussi Haukioja

same term. (Meta-​internalism does not require that ordinary speakers, in order to use a term with its publicly determined extension, have beliefs about how, say, experts use a given term—​indeed, often they have no definite beliefs, or even mistaken ones, about this. What meta-​internalism does require is that such speakers are sensitive to new information about this; that they would revise their own use as a consequence of acquiring or revising their beliefs about the experts’ use.) To say that (first-​order) internalism is true of an expression, as used by a speaker, is to say that her application and interpretation is independent of the speaker’s beliefs concerning such external factors (including, but not limited to, her beliefs about other speakers). And so on, for any first-​order internalist or externalist theory of reference: on this understanding of what intuitions are, there is a very direct, constitutive, relationship between intuitions and the facts about reference. If intuitions are seen as constitutive of the facts about reference, in the above sense, the use of both armchair methods and experimental methods in theorizing about reference can be defended. The relevance of experimental results falls directly out of the meta-​internalist view: if theories of reference are really theories about systematic patterns of intuitive responses among speakers who use the expression in question, empirical evidence about what kinds of patterns of intuitions ordinary speakers in fact have is obviously relevant. At the same time, the kinds of thought experiments typically used in theorizing about reference can be seen as vehicles for eliciting first-​level intuitions. That is, the armchair theorist is not, at least primarily, evaluating hypothetical cases on the basis of empirical generalizations about linguistic behaviour, but rather on the basis of her own linguistic competence. For example, someone contemplating the Twin Earth thought experiment is, in our view, really engaged in a kind of mental simulation: she is imagining herself, or any of her linguistic peers, as coming across a watery liquid on Twin Earth, and activating the relevant parts of her own linguistic competence. On first contact with Twin water, she would be likely to apply the term “water” to it, but—​if her first-​level intuitions line up with Putnam’s—​she would also be disposed to retract her usage of “water” to Twin water upon finding out the relevant chemical facts. The thought experiment, if successful, points our direction at precisely those first-​level intuitions which, according to meta-​internalism, make first-​order externalism (more precisely, natural kind externalism, or physical externalism) true. It is true that we, as theorists, often find it natural to report the outcomes of our thought experiments in metalinguistic terms (“The term ‘water’ does not refer to XYZ”; “The Twin English term ‘water’ does not mean water”, etc.), but these statements are generalizations over a wide range of first-​level intuitions: the primary purpose of thought experiments like Twin Earth is to probe our own first-​level intuitions, and point our attention at systematic patterns in these intuitions. The theorist’s own judgements about hypothetical cases will then be admissible as data for theories of reference, under two conditions. First, the theorists’ own first-​level intuitions must be representative of the linguistic community for which the theory is supposed to hold. Second, the theorist’s reports of her own first-​level intuitions must be accurate. There may be reason to think that experienced theorists in fact do satisfy both conditions: there may even be reason to think that they do so to a greater extent than non-​experts: a version of the expertise defence is, arguably, available to the meta-​ internalist. However, we should keep in mind that both conditions are empirical hypotheses, and as such always subject to empirical repudiation. Finally, it should be noted that meta-​internalism is centrally motivated by considerations having to do with linguistic communication, rather than general claims about intuitions and facts: the arguments for meta-​internalism cannot be translated into arguments for analogous views about the constitution of, say, moral or epistemological facts. Thus, the constitutive relationship between intuitions and facts accepted by meta-​internalists may give intuitions a privileged status as data in philosophy of language, compared to other areas of philosophy. At the very least, analogous positions in other areas of philosophy would require an independent argument, and are likely to be perceived as far more radical than meta-​internalism about reference is.

558

Reference and Intuitions

References Cappelen, Herman (2018): Fixing Language, Oxford: Oxford University Press. Cohnitz, Daniel and Haukioja, Jussi (2013): Meta-​Externalism vs Meta-​Internalism in the Study of Reference, Australasian Journal of Philosophy, 91, 475–​500. Cohnitz, Daniel and Haukioja, Jussi (2015): Intuitions in Philosophical Semantics, Erkenntnis, 80, 617–​641. Devitt, Michael (2006): Ignorance of Language, Oxford: Oxford University Press. Devitt, Michael (2015):  Testing Theories of Reference, in:  J. Haukioja, Advances in Experimental Philosophy of Language, London: Bloomsbury, 31–​63. Katz, Jerrold (1990): The Metaphysics of Meaning, Cambridge, MA: MIT Press. Kornblith, Hilary (2002): Knowledge and Its Place in Nature, Oxford: Oxford University Press. Mallon, R., Machery, E., Nichols, S., and Stich, S. P. (2009): Against Arguments from Reference, Philosophy and Phenomenological Review, 79, 332–​356. Martí, G. (2009): Against Semantic Multi-​culturalism, Analysis, 69, 42–​48. Reed, N., McLeod, P., and Dienes, Z. (2010): Implicit Knowledge and Motor Skill: What People Who Know How to Catch Don’t Know, Consciousness and Cognition, 19, 63–​76.

559

41 THE MYTH OF QUICK AND EASY INTUITIONS John Bengson

This chapter discusses the epistemology of reference, focusing on a range of justified beliefs and knowledge about reference whose status has been the subject of vast philosophical controversy. After identifying an array of data that an adequate epistemological theory of the relevant range must handle, I distinguish inferentialist and intuition-​based theories. I then proceed to develop and defend a version of the latter, by showing that it can both handle pertinent data and respond to two prominent challenges. One arises from experimental research taken to show that intuitions are unreliable, while the other questions the idea that philosophers rely on intuitions. Both criticisms, I argue, are generated by serious misconceptions about intuition. I conclude by identifying a persistent and popular error—​the Myth of Quick and Easy Intuitions—​lurking behind both sets of criticisms.

1  The Epistemology of Reference: Data and Theories Knowledge of reference is often based on testimony or sense experience. Our parents and teachers tell us that ‘Mesopotamia’ refers to a region of southwest Asia between the Tigris and Euphrates rivers, and that ‘yak’ refers to a large bovid native to the Himalayas.When a friend exclaims “That’s beautiful” or “I love this song” we see the painting or hear the music to which the friend’s use of ‘that’ or ‘this song’ refers. We also learn about reference by engaging in complex mathematical reasoning, medical study, scientific inquiry, and philosophical investigation—​candidates include firsthand knowledge or justified belief that ‘3’ refers to the sole prime number that is a divisor of 27, that ‘arthritis’ refers to inflammation of the joints, that the chemical composition of the clear, odorless liquid to which ‘water’ refers is H2O, and that ‘person’ refers to whatever possesses the capacity for certain sorts of self-​conscious thoughts. In other cases, however, our knowledge of reference seems to be based on something rather different. That ‘yak’ (as it is standardly used) may refer to a person’s favorite animal, but not her favorite fruit, is something many thinkers immediately perceive to be true, in the absence of any conscious reasoning. That ‘3’ refers to a number that is equal to the sum of all the natural numbers less than it—​this is equally obvious. Similarly, when imagining a member of your linguistic community who states, erroneously, that arthritis might occur in the thigh (where no bones meet), it may seem obvious that such an individual’s use of ‘arthritis’ would still refer to an ailment of the joints.1 Or contemplate a duplicate of you or me who inhabits an environment in which the clear, odorless liquid that flows through the lakes, rivers, and faucets, although superficially quite similar 560

The Myth of Quick and Easy Intuitions

to water, is chemically composed not of H2O but XYZ.2 When asked whether this twin earthling’s use of ‘water’ would refer to something chemically composed of H2O, as it does for us earthlings, it may strike us as plain that the answer is negative. Yet, when asked whether the twin earthling’s use of ‘person’ would continue to have the same reference as it does for us earthlings, it may seem clear that the answer is affirmative—​even though twin earthlings are composed not of H2O but XYZ, and even if they are not members of the species homo sapiens (because they evolved differently).3 In all these cases, there is no telling, seeing, or hearing. But there is knowledge, or at least justified belief, about reference. An adequate theory of a given domain must handle the data regarding that domain. When it comes to the epistemology of reference, these data include the following observations about the relevant cases:4 D1 In a wide range of such cases, it strikes us that the reference of the term under consideration is such-​and-​such. D2 These responses are conscious states or events. D3 These responses occur straightaway, that is, in the absence of any consciously mediated transition from the presumed truth of one set of claims to the truth of the verdict about reference that strikes us as true. D4 These responses may prompt us to form beliefs about the reference of the term, though we may instead withhold judgment. D5 These responses are sometimes about a highly unusual or fantastical case, far removed from any actual situation that we’ve faced. D6 These responses sometimes conflict (directly or indirectly) with our antecedent views. D7 These responses are regularly taken to serve as justificatory bases for the beliefs they prompt, provide evidence for those beliefs’ contents, and put their subjects in a position to know those contents. D8 In many cases, we rightly possess a very high level of confidence in these beliefs. D9 We are often unable to identify a set of considerations, independent of our responses, that provide a correspondingly strong rationale for these beliefs. D10 Justified beliefs and knowledge about reference are sometimes achieved with little effort, but they are also sometimes the result of substantial reflection or analysis. This list is not meant to be exhaustive. Nevertheless, it identifies a number of important data that an adequate theory of our justified beliefs and knowledge about reference in the relevant cases must accommodate and explain.5 One theory of such cases is inferentialist: we justifiably believe or know the indicated facts about reference by inferring them from background beliefs we are justified in holding, where inference is a certain kind of rationally evaluable mental transition. There are numerous ways to develop an inferentialist theory. Proponents may choose among diverse approaches to the character of the inference (e.g., deductive, inductive, abductive), the nature of the background beliefs (e.g., hypotheses, elements of the common ground), and the epistemic status of those beliefs (e.g., empiricists maintain that the beliefs are justified or known in a way that is dependent on sense experience, or a posteriori, while rationalists hold that the beliefs are justified or known a priori, and pragmatists claim that the beliefs play certain practical or dialogical roles). It is an open question whether inferentialism is positioned to handle all of the data enumerated above (and other pertinent data). It is not simply that inferentialism is silent about the responses described in the first six data, D1–​D6, which the view thereby fails to either accommodate or explain. Its central thesis is also difficult to square with several of these data: the idea that our verdicts about reference in the relevant cases are inferred from our background beliefs appears to point in the opposite direction from the observations that those verdicts are not consciously mediated (per 561

John Bengson

D3) and are sometimes novel (per D5) or revolutionary (per D6). Inferentialism also looks ill-​suited to handle the observations about justification, evidence, and knowledge in the remaining data. For example, inductive and abductive inferences do not typically support the high level of confidence described in D8.The point is far from decisive. But it becomes concerning when combined with D7, which tells us that our conscious responses, rather than any inferences, are regularly treated as sufficient, epistemically speaking, and D9, regarding our inability to identify independent considerations that might provide the indicated support. Of course, it is open to inferentialists to supplement their theory with additional claims that handle the data. Alternatively, inferentialists may provide reasons to think that some or all of the putative data I’ve enumerated are not really data, or needn’t be accommodated and explained—​ in which case inferentialism’s failure to handle them matters not a whit. Suffice it to say that while I welcome efforts to take these steps, I do not think it likely that either strategy will prove successful. I am more optimistic about an intuition-​based theory of such cases.6 According to this theory, we justifiably believe or know the indicated facts about reference via intuition, where an intuition is a conscious non-​sensory mental state or event in which it strikes one that things are a certain way when one reflects on the matter. The remainder of this chapter develops and defends an intuition-​ based theory.

2  An Intuition-​Based Theory There are numerous ways to develop an intuition-​based theory. Proponents may choose among diverse approaches to the nature of intuition (e.g., phenomenological, etiological, content-​oriented, etc.) and sundry accounts of intuition’s epistemic powers (e.g., dogmatist, reliabilist, virtue-​theoretic, etc.). These powers include the capacity to justify beliefs to varying extents, the ability to serve as a source of knowledge, and the potential to be or supply evidence. My preferred version of an intuition-​based theory combines four main theses. The first tells us what an intuition is: Quasi-​Perceptualism: Intuitions are a specific type of presentational state, a state that presents its content as true. Such a state has several characteristic features: it is conscious, contentful, non-​factive (it is not success-​entailing), gradable (it can be more or less clear, vivid, hazy, or obscure), baseless (it is not consciously formed, by a subject, on the basis of any other mental states), fundamentally non-​ voluntary (it just happens), and compelling (it inclines one to belief). The next three theses are not specific to intuition, but apply to all presentational states. The first delivers an account of intuitive justification: Presentationalism: A thinker is justified in believing that p on the basis of a presentational state—​ such as intuition—​with content p because in having that state, it is presented to the thinker that p.  Such justification is both defeasible and gradable:  it is greater to the extent that the corresponding presentation is clear and vivid, and it is lesser to the extent that the presentation is hazy and obscure. Another addresses intuitive evidence: Content is Evidence: The content of a presentational state—​such as intuition—​belongs to a thinker’s body of evidence because it is the content of a presentational state. Such evidential status is both defeasible and gradable: it is greater to the extent that the corresponding presentation is clear and vivid, and it is lesser to the extent that the presentation is hazy and obscure.7 562

The Myth of Quick and Easy Intuitions

And the last explains the possibility of knowledge via intuition: Naïve Realism: A presentational state—​such as intuition—​with content p is non-​accidentally correct, hence able to serve as a source of knowledge that p, because it is constituted by the (possibly mind-​independent) fact that p. Elsewhere I endeavor to clarify and substantiate each of these theses, along with several other claims about the nature and epistemic status of intuition (concerning, e.g., intuition’s role as a source of understanding, relation to epistemically fecund social practices, place in disagreement, and difference from a seeming).8 What I  wish to emphasize here is that these four theses are well-​positioned to accommodate and explain all of the data enumerated above. To appreciate this, notice that the first thesis, Quasi-​Perceptualism, not only renders it likely that the initial six data are true; it also helps to explain why those data hold. For suppose that in the relevant cases we have intuitions that are presentational in the sense described by the Quasi-​Perceptualist thesis.This would make sense of the fact that subjects experience conscious strikings (per D1 and D2) that, while occurring without conscious mediation (per D3), could but need not prompt beliefs (per D4), perhaps even beliefs that are novel (per D5) or revolutionary (per D6). The remaining three theses—​ Presentationalism, Content is Evidence, and Naïve Realism—​ accommodate and explain the data about justification, evidence, and knowledge in D7 and D8. Together, the four theses help to make sense of D9, regarding our inability to identify independent considerations that provide the requisite support for our verdicts. And when the four theses are supplemented with a plausible claim about the etiology of presentational states, they also handle D10. The claim is this: effort  While some presentations—​ including intuitions—​ require very little preparation, others (perhaps those with certain contents, and particularly those that are clear and vivid) may occur only in the wake of considerable mental effort, including sustained attention to and careful analysis of various questions or scenarios.9 This etiological claim is important, and will play a pivotal role below. I have formulated the main theses of my preferred version of an intuition-​based theory, and shown how they handle the data enumerated above. I will not pause to apply the theory to each of the cases of knowledge or justified belief about reference mentioned in the second paragraph of section 1.10 Rather, I dedicate the remainder of this chapter to critical discussion of two recently popular sources of skepticism about intuition.

3  Experimental Philosophy Recent empirical research at the intersection of psychology and philosophy includes experiments claimed to show that a variety of intuitions, including those about reference, are unreliable, and therefore epistemically suspect. The challenges are varied.11 Some experiments are said to reveal a worrying sort of cultural relativity in our intuitions about reference. Other experiments allegedly show that such intuitions display alarming intra-​and inter-​personal diversity. Still others purportedly show that intuitions are vulnerable to order or wording effects: depending on how a case is described, or the order in which cases are presented, subjects may have varying intuitions about the scenario in question. In all of these instances, the underlying worry is the same: intuitions have an epistemically problematic feature, being formed on the basis of influences that have nothing to do with, but often point away from, the truth of the matters that subjects are being asked to consider. The experiments relied upon to yield such a pessimistic conclusion share a similar structure. Subjects are presented with a question about a particular vignette, and are then instructed to provide 563

John Bengson

an answer to the question. Notice that in order to secure the relevance of such studies to questions about the status of intuitions, it is necessary to make an inference from an observation about how it is with subjects’ prompted answers to a conclusion about how it is with subjects’ intuitions. Such an inference assumes that the former express the latter. I call this claim the answers express intuitions thesis, or ‘aei’: aei   Subjects’ prompted answers express subjects’ intuitions. Again, the conclusion of the experimentalist’s critique is that intuitions possess a certain epistemically problematic feature. We can represent the argument for this conclusion schematically, as follows: 1 Subjects’ prompted answers have feature F. 2 F is an epistemically problematic feature. 3 Subjects’ prompted answers express subjects’ intuitions. (aei) So, 4 Subjects’ intuitions have an epistemically problematic feature. (from 1, 2, 3) 5 Subjects’ intuitions are representative. So, 6 Intuitions have an epistemically problematic feature. (from 4, 5) This formulation covers experimentalist challenges across a wide variety of philosophical subfields. It also enables a neat catalog of existing responses to these challenges. In the case of reference, some critics contest the empirical claim (premise 1),12 while others reject the epistemological claim (premise 2)13 or deny the generalization (premise 5).14 I agree with such critics that each of these premises is challengeable. But my aim here is to briefly summarize my doubts about aei (premise 3), the key principle connecting the experiments to intuitions.15 An important step in assessing aei draws attention to the great many cases in which a thinker’s answer to a question does not express her intuition. How she responds does not express how things strike her. In some cases, a thinker gives an answer to a question even though she does not have any intuition one way or another about it—​call such answers ‘plain’. At other times, a thinker’s answers run contrary to her intuitions—​call such answers ‘stray’. Examples of plain answers are guesses offered in response to obscure questions in Trivial Pursuit, and quick hypotheses or inferences about people’s whereabouts (e.g., from darkened windows to the assessment that nobody is home). Examples of stray answers are those that express one’s group allegiances but contradict one’s own intuitions, and emotion-​based condemnations of deviant behaviors that one’s intuitions, by contrast, deem permissible (e.g., human cannibalism). Answers do not always express intuitions, and intuitions are not always expressed in answers. In both directions we find a gap between intuitions and answers. Consequently, aei is false: subjects’ prompted answers do not always express subjects’ intuitions. But how often do we see such a gap? An experimentalist might reply to this criticism by weakening aei, as follows: aei*  Subjects’ prompted answers typically or usually (normally, etc.) express subjects’ intuitions. One might worry that this weakened principle does not suffice to construct a valid or cogent experimental attack on intuition.16 But let us simply grant that it does.The main problem for the envisioned reply is that we have no reason to think that plain or stray answers are rare or anomalous, and so we 564

The Myth of Quick and Easy Intuitions

cannot simply assume, without argument, that subjects’ prompted answers typically or usually express subjects’ intuitions, as asserted by aei*.17 Jonathan Weinberg and Jonathan Alexander (2014, 200ff.) have argued that “dual process” or “two systems” theories in cognitive psychology support the idea that subjects’ answers express System 1’s “fast, apparently effortless cognition, based more on similarities and associations” than “slower, effortful, rule-​based cognition”. That is, they contend that s1 Subjects’ prompted answers express System 1 responses. They also affirm that intuitions are among the products of System 1: s1-​int Some System 1 responses are intuitions. Weinberg and Alexander conclude that aei**  Some of the subjects’ prompted answers express subjects’ intuitions. This defense merits attention. But it is problematic, for (at least) two reasons. The first concerns the inference from s1 and s1-​int to aei*. Even if we were to grant both of the former principles, it would not follow that aei** is true. Nor, I submit, do the former justify aei**, at least not in the manner that’s required to vindicate the experimentalist critique of intuitions. To appreciate that the conjunction neither entails nor adequately supports aei**, recall the point that prompted answers easily express guesses, hypotheses, inferences, allegiances, and emotional reactions. Such states are among the myriad products of System 1; there are of course many others. That subjects’ prompted answers easily could express only such non-​intuitional products of System 1 blocks the entailment. When allied with the observation that we lack reason to think that any of the subjects’ prompted answers differ from the non-​intuitional products, it also implies that we lack reason to believe that any of those answers express intuitions.18 Now for the second reason, which questions s1-​int. Weinberg and Alexander do not argue that proponents of intuition-​based theories are committed to this principle. And it is not clear why proponents of such theories should accept it, in light of a pair of interrelated thoughts. The first is that the goal of a general intuition-​based theory (such as mine) is to identify a type of mental state, distinct from states such as guesses, hypotheses, inferences, allegiances, and emotional reactions, that is able to deliver justified belief and knowledge—​not just about reference, but also modality, mathematics, logic, essence, normativity, and other matters, often highly abstract and difficult ones, of considerable intellectual interest. The second is that the claim I’ve labeled effort (in section 2) is true of this type of mental state: its instances regularly occur only in the wake of considerable mental effort. The various states pinpointed by cognitive scientific research on System 1’s responses—​at least those that are “fast, apparently effortless cognition, based more on similarities and associations”—​are poorly suited to do the work registered by the first thought, and lack the feature highlighted by the second.19 Indeed, such states differ from the type of mental state invoked by intuition-​based theories precisely in ways that problematize an inference to any claims about that state. The point bears emphasis. The belief that the issues that separate proponents of intuition-​based theories and their opponents can be addressed by investigation of System 1’s “heuristic” or “hot” cognition manifests a serious misconception about intuition—​one that is an instance of what I dub the Myth of Quick and Easy Intuitions (more on which below). The criticisms I have been pressing against various attempts to bridge the gap between subjects’ prompted answers and intuitions have not relied on the thesis that I’ve labeled Quasi-​Perceptualism, according to which intuitions are a specific type of presentational state. But the criticisms do flow naturally from this thesis, insofar as that thesis distinguishes intuitions from guesses, hypotheses, 565

John Bengson

inferences, allegiances, and emotional reactions, and so makes good sense of plain and stray answers. At the same time, basically all theories of intuition recognize that answers are sometimes plain or stray.20 It is this general recognition, not endorsement of the specific Quasi-​Perceptualist thesis, that fuels my criticism of the principle aei and variants thereof. The failure of these principles does not simply undermine the experimental argument against intuition. It also provides the basis for a simple argument that the empirical research in question cannot warrant any negative conclusions about intuitions: 7 Prompted answers may (typically, usually, normally, etc.) express a variety of mental states, many of which are not intuitions. 8 If so, then an empirical study that elicits prompted answers cannot warrant any negative conclusions about intuitions, unless: [*]‌ The study employs experimental controls or is accompanied by supplemental argumentation that affords assurance that subjects’ prompted answers are best understood as indicating how the cases struck the subjects—​what intuitions the subjects had, if any. 9 The empirical studies in question do not satisfy [*]‌. So, 10 The empirical studies in question cannot warrant any negative conclusions about intuitions. (from 7, 8, 9) Premise 7 follows from the failure of aei and variants thereof. Premise 8 is a near-​platitude: it simply observes that if there is no reason to think that subjects’ prompted answers in a given empirical study express intuitions rather than other mental states, then that study cannot warrant any negative conclusions about intuitions rather than the other mental states. I believe that premise 9 is secure; in fact, I have argued elsewhere for the stronger claim that there are plausible explanations of the experimental results that attribute plain or stray answers to experimental subjects.21 In the present context, however, it suffices to spotlight the following challenge: experimental philosophers endeavoring to support a negative conclusion about intuitions must ensure that the studies to which they appeal satisfy  [*]‌.22 While there is little doubt that intuitions are fallible, intuitional errors are regularly both detectable and correctable.23 Against the claim that various empirical results challenge intuition’s epistemic credentials, I’ve argued that once we attend to the fact that the relevant studies simply elicit subjects’ answers to questions, the idea that such studies probe the epistemology of intuition emerges as an article of faith. Absent some reason to think that subjects’ answers are neither plain nor stray, we’re rationally compelled to view extant empirical research as unable to support skepticism about intuition.

4  Centrality and “the Myth of the Intuitive” Philosophy in the analytic tradition has witnessed a series of attacks on intuitions and their place in philosophical inquiry. First came the positivists, rejecting as nonsense the putative verdicts of intuition, which they denounced as mysterious.24 Next came the Quineans, rejecting the indefeasible a priority and foundationalism on which intuitions’ claim to epistemic fame was widely thought to rest. More recent are the experimental philosophers, whose criticism we just considered. In the latest wave, however, are those who deny that intuition plays a central role in contemporary philosophy. A leading proponent of this last criticism, Herman Cappelen, labels the target of these critics’ ire 566

The Myth of Quick and Easy Intuitions

Centrality: “Contemporary analytic philosophers rely on intuitions as evidence (or sources of evidence) for [or against] philosophical [claims]”.25 According to Cappelen’s ally, Max Deutsch, “the idea that philosophy relies on intuitions as evidence is a myth, an enduring and fairly widely held, yet entirely false, belief about the methods of philosophy”.26 Deutsch calls this “the myth of the intuitive”. Although the intuition-​based theory that I  presented in section 2 is not committed to the Centrality thesis, the theory and the thesis are intimately connected. For among the beliefs about reference whose epistemic status the theory seeks to illuminate are those possessed by philosophers.27 If Centrality were false, that would suggest that those philosophers do not rely on intuitions; consequently, a theory that claims that those beliefs enjoy a positive epistemic status because they are based on intuition would be mistaken. I discern two arguments against Centrality in recent literature. One begins with the observation that ‘intuition’ displays a state/​content ambiguity. Distinguish state  one’s having an intuition, which is a psychological matter, from content  the content of one’s intuition, which is not.28 For example, when I have the intuition that a twin earthlings’ use of ‘water’ does not refer to H2O, there is my having the intuition that a twin earthlings’ use of ‘water’ does not refer to H2O, on the one hand, and the content of this intuition, namely, that a twin earthlings’ use of ‘water’ does not refer to H2O, on the other. Importantly, as Deutsch emphasizes, mere reliance on the latter (content) as evidence is not at issue when thinking about Centrality.29 The reason is that a thinker may easily rely on a given content as evidence without treating any particular mental state, such as an intuition, that has that content as in any way relevant to the content’s being evidence. For example, suppose a thinker has an intuition with the content that a twin earthlings’ use of ‘water’ does not refer to H2O, and she constructs an argument whose conclusion is the selfsame content. She might rely on that content as evidence solely because it is the conclusion of the argument. Accordingly, the argument against Centrality begins with the premise that 1 Centrality is true only if philosophers rely on the fact that they have such-​and-​such intuition as evidence (i.e., only if they rely on state). But, the argument continues, 2

Philosophers do not rely on the fact that they have such-​and-​such intuition as evidence.

So, 3 Centrality is false. The intuition-​ based theory articulated above, in section 2, exposes the misstep in this argument:  premise 1 is false. According to the thesis I’ve called Content is Evidence, the contents of intuitions are evidence in virtue of their being the contents of a certain type of mental state. The evidence is not the thinker’s mental state. Nor, however, is it the content of that mental state, in itself, independent of any relation to the thinker’s mental state, that is the evidence. Rather, the content is evidence because it is the content of intuition, a presentational state. This is an epistemic power 567

John Bengson

intuitions possess: they make it the case that their contents belong to one’s body of evidence. But it is not a power possessed by a range of other contentful states—​such as hopes, wishes, desires, imaginings, guesses, hunches, hypotheses, astrological speculations, or putative divinations.30 Hoping or wishing that a twin earthling’s use of ‘water’ does not refer to H2O does not make it the case that the content a twin earthling’s use of ‘water’ does not refer to H2O belongs to one’s body of evidence. But, according to Content is Evidence, having the intuition that this is so does. Content is Evidence thus implies that a thinker needn’t treat the fact that they have such-​and-​ such intuition as evidence in order to rely on intuition as evidence, in a way that goes beyond merely relying on the content of intuition as evidence. Such reliance also goes beyond merely treating intuition as the psychological origin of a content, or as the mere cause of one’s possessing a given piece of evidence. For it involves treating one’s intuition as the epistemic basis of one’s evidence. Put differently, Content as Evidence implies that what is at issue is neither a state nor a mere content, but rather e-​basis  the content of one’s intuition, qua content of intuition. To rely on intuitions as evidence, from the perspective of Content is Evidence, is to treat the contents of intuitions as evidence because they are the contents of intuitions. (This explicates what I take reliance on e-​basis as evidence to be, by contrast with both reliance on state as evidence and mere reliance on content as evidence.) If philosophers do this, then Centrality is true. It follows that premise 1 in the above argument against Centrality is false. A second type of argument against Centrality consists in close analysis of important philosophical texts. Focusing on texts that proponents of Centrality have alleged to display reliance on intuitions as evidence, opponents of Centrality claim to fail to find evidence of reliance on intuitions, but instead to succeed in finding arguments. Specifically, the opponents advance a series of claims about the relevant texts, among which the following three are prominent: no refer  Philosophers do not explicitly reference intuitions as such when they make claims about thought experiments and cases. qualify  Philosophers use ‘intuitive’, ‘intuitively’, and similar expressions to qualify assertions or convey some other information, rather than to designate a type of mental state. argue  Philosophers argue for the claims they make about thought experiments and cases. Although opponents of Centrality are careful to avoid asserting that these (and various other similar) claims entail that Centrality is false, the opponents do suggest that these claims provide evidence against Centrality. In reply, I will very briefly offer some evidence against no refer.31 I’ll then argue that, even if no refer is true, it does not cast doubt on the claim that philosophers rely on intuitions as evidence in the epistemic-​basis sense, specified above, that is made possible by the Content is Evidence thesis. I’ll then argue for parallel theses in regard to qualify and argue. Evidence against no refer is manifest in three of the most influential works in the philosophy of reference in the twentieth century. First, in the original discussion of twin earth cases, Hilary Putnam explicitly references intuition as such: [This is] a topic which deals, after all, with matters which are in everyone’s experience, matters concerning which we all have more data than we know what to do with, matters concerning which we have, if we shed preconceptions, pretty clear intuitions.32 Second, in his original discussion of the arthritis thought experiment (briefly sketched in section 1), Tyler Burge not only explicitly references intuition as such; he also offers a rationale for focusing on the intuitions that he does: 568

The Myth of Quick and Easy Intuitions

I have presented the experiment as appealing to ordinary intuition.33 And I  wanted to encourage you, dear reader, to imagine actual cases of incomplete understanding in your own linguistic community. Ordinary intuitions in the domestic case are perhaps less subject to premature warping in the interests of theory.34 Third, in Naming and Necessity, Saul Kripke explicitly references intuition as such when discussing his evidence for the text’s primary thesis: My main remark, then, is that we have a direct intuition of the rigidity of names, exhibited in our understanding of the truth conditions of particular sentences.35 Additional examples are easy to find. That evidence against no refer is not simply available but is abundant emerges once it is recognized that explicitly referencing a given epistemic basis (intuition) as such does not require using one, and only one, specific English word (the term ‘intuition’ or one of its cognates). But I  do not wish to place much stock in this contention, and not simply because doing so might require quoting a large number of texts and defending a face-​value construal of the relevant passages against Centrality-​unfriendly re-​interpretation strategies.36 The main reason is that I am not convinced that textual analysis is probative in the present context. More precisely, my view is that even if no refer were true, it would fail to support opposition to the claim that philosophers treat intuition as an epistemic basis. To appreciate this, notice that this last claim is an instance of the following general schema: schema  A range of thinkers treat the contents of a certain type of mental state as evidence because they are contents of states of that type. This schema delivers the right verdict about a wide range of domains. Consider perception: percept  Perceivers treat the contents of their perceptions as evidence because they are the contents of perceptions. For instance, I treat the content there is a cardinal at the birdfeeder as belonging to my body of evidence about the external world because I have a perception with this content. I treat my perception as evidence. Of course, perceivers do not typically explicitly reference their perceptions—​their epistemic bases—​ as such when they make claims about the external world. So we should endorse no referP  Perceivers do not explicitly reference their perceptions as such when they make claims about the external world. But no referP is not evidence against percept. Likewise, then, for no refer with respect to the corresponding claim in the case of intuition: intuit  Philosophers treat the contents of their intuitions as evidence because they are the contents of intuitions. That is, no refer is not evidence against the claim that philosophers rely on intuitions as evidence. Here is a second reason to think that no refer does not impugn intuit. Consider the truism that premise  Philosophers rely on sets of premises as evidence.

569

John Bengson

And now consider: no referPR  Philosophers do not explicitly reference premises as such when they make claims about thought experiments and cases. Suppose we found widespread evidence for this claim. After all, the term ‘premise’ does not appear in Putnam’s text. Nor does Kripke apply the term ‘premise’ to any of the considerations to which he adverts in Naming and Necessity. I trust that opponents of Centrality would agree that this does not provide evidence against the truism (premise). But if so, then the absence of explicit reference to intuitions as such, as asserted by no refer, is not evidence against the Centrality-​friendly view that philosophers rely on intuitions as evidence (intuit). Let us now consider whether qualify is evidence for the denial of this view. There are good reasons to think that qualify is true. Expressions such as ‘intuitive’ and ‘intuitively’ are regularly used as hedges, functioning to weaken the speaker’s commitment to the embedded sentence. They are also sometimes used to “signal that one is not going to provide any further arguments for [a given] claim”, or to avoid “a pure and chilly way of writing philosophy in which premises and conclusions are baldly asserted”.37 In still other cases, they’re used to indicate that something is familiar or easy in some relevant way: for instance, an idea is intuitive when it is familiar or easy to understand; a device is intuitive when it is familiar or easy to operate; or a location is intuitive when it is familiar or easy to find. That philosophers undoubtedly use expressions such as ‘intuitive’ and ‘intuitively’ in these ways—​ not in order to designate a mental state upon which they’re relying as evidence, but rather to play some other communicative role—​appears to undermine a simple argument from the appearance of these expressions in philosophical texts to the conclusion that Centrality is true. But that simple argument is too simple, and to my knowledge, it has not been advanced by any proponent of Centrality. Our question is whether qualify is evidence against the claim that philosophers treat intuition as an epistemic basis. The answer to that question is negative. The fact that members of a group use expressions such as ‘intuitive’ and ‘intuitively’ in ways that do not designate any type of mental state is not evidence that they do not rely on intuition as evidence. Compare perceptual terms, regarding which the following is true: qualifyP  Perceivers use perceptual terms (‘perceive’, ‘see’, ‘hear’, etc.) to do things other than designate perceptions. Such a claim, however, does not cast doubt on percept, the claim that perceivers rely on perceptions as evidence. Similarly, that philosophers use intuitional terms to do things that do not designate intuitions is not evidence against intuit, the claim that philosophers rely on intuitions as evidence.38 We turn, finally, to argue, which asserts that philosophers argue for the claims they make about thought experiments and cases. Once again, I do not think that this is evidence against the claim that philosophers treat intuition as an epistemic basis.We often collect multiple pieces of evidence for one and the same claim; we treat all of them as evidence. Suppose, for example, that you testify that there is a bug in my soup.That I look and see for myself is not evidence that I am not treating your testimony as evidence. Or suppose I seem to see a small grin now appear on your face. I also quickly reason as follows: you just heard the same pun I did, and you like such puns, so it is only to be expected that you are now grinning.That I reason in this way—​constructing an argument in my head for the belief that you are grinning—​is not evidence that I am not treating my own visual experience as evidence for the selfsame claim. The point is not simply that relying on one thing, such as the provision of testimony or argument, is compatible with treating something else, whether perception or intuition, as an epistemic basis. Rather, it is that the provision of argument (in a given context) provides no evidence 570

The Myth of Quick and Easy Intuitions

that one is not relying on intuition as evidence (in that context).39 If this is correct, then argue is not evidence against the Centrality-​friendly thesis that philosophers treat intuition as an epistemic basis (intuit). To be clear, my intention is not to deny argue. Interpreted correctly, I affirm this claim, alongside several other theses about philosophical practice emphasized by opponents of Centrality. For example, I agree with Timothy Williamson’s observation that even in philosophical investigation of apparently simple matters “hard theoretical work is needed”.40 I am also amenable to Cappelen’s suggestion that “Reflection on cases is a hyper-​rational, epistemically hyper-​demanding context” in which philosophers exercise a “tendency to question everything”.41 And I recognize that, as Deutsch writes, philosophy is “difficult business”, requiring “a considerable amount of ingenuity, careful thought”, “time, reflection, and many difficult inferences”.42 These claims are of a piece with the observation labeled D10 (identified as a datum in section 2). And all of them are compatible with my preferred intuition-​based theory. By endorsing the general claim about presentations that I’ve called effort (again, in section 2), this theory avows that some intuitive presentations may occur only in the wake of considerable mental effort. This is particularly true of those intuitions regarding matters of philosophical importance that are clear and vivid, and so able to provide the level of justification and evidential support required for knowledge of such matters. So, if the intuition-​based theory I’ve articulated is correct, it is not surprising that philosophers seeking knowledge about reference engage in the sort of analytic work described by Williamson, Cappelen, and Deutsch.

5  The Myth of Quick and Easy Intuitions I conclude that no refer, qualify, and argue do not provide evidence against Centrality. The idea that philosophers treat intuitions as evidence cannot be unmasked as a myth on the grounds that these claims, or others like them, are true. The real myth in this neighborhood is the thought that makes it look like Centrality is squeezed out by giving proper weight to the role of argument in philosophical texts, and to the importance of ingenuity, careful thought, and complex reasoning in philosophical practice.This myth figures centrally in opposition to Centrality, in the shape of the idea that intuitions aren’t on the scene when philosophers are working hard, engaging in deep and sustained reflection to craft and assess thought experiments and cases. This, to give it a label, is the Myth of Quick and Easy Intuitions.43 The Myth of Quick and Easy Intuitions is familiar. It is displayed in philosophical activity whenever disputants simply toss around appeals to ‘intuitions’—​as if intuitions are, well, quick and easy. It is equally there in many perfunctory dismissals of intuition, which fail to appreciate just how arduous it can be to develop intuitions (once again, as stated in effort). Similarly, the Myth informs the charge that intuition-​based theories are guilty of “empty self-​congratulation”,44 and underlies various quips about intuition, such as this one by Wittgenstein: It is just as if somebody claimed to have knowledge of human anatomy by intuition; and we say: “We don’t doubt it; but if you want to be a doctor, you must pass all the examinations like anybody else”.45 The Myth also shows up in the experimental challenge discussed above, in section 3, in the guise of the assumption that we can conclude that experimental subjects had intuitions merely on the basis of the fact that the subjects answered the questions they were given (recall aei and variations thereof). Similarly, the Myth fuels the assumption, also identified in section 3, that empirical research in cognitive psychology on “heuristic” or “hot” cognition speaks straightforwardly to philosophical debates about intuition. Despite their differences, then, the two challenges that we’ve been discussing are properly diagnosed as two expressions of one and the same error, that of failing to appreciate the 571

John Bengson

viability of an intuition-​based theory, such as the one described in section 2, that accepts effort—​ and, concomitantly, of the mistake of falling into the Myth of Quick and Easy Intuitions.

Acknowledgments I am grateful to an audience at, and the other participants in, a symposium on intuition at the 2018 Pacific APA, and especially to Max Deutsch, Jonathan Livengood, and Stephen Biggs, for helpful discussion and comments on the material in this paper.

Further Reading Bealer, George. 2008. “Intuition and Modal Error”. In Q. Smith, ed. Epistemology:  New Essays. Oxford University Press. Bengson, John. 2015. “The Intellectual Given”. Mind, 495: 707–​760. Hansen, Nat. 2015. “Experimental Philosophy of Language”. Oxford Handbooks Online. Book Symposium in Philosophical Studies (vol. 171, no. 3, 2014) on Herman Cappelen’s Philosophy without Intuitions.

Notes 1 Burge (1979). Burge maintains that this is just one instance of a very general phenomenon. 2 Putnam (1975). Although I  focus on water, Putnam takes his discussion to apply to all natural kinds; his examples include gold, tiger, and lemon. 3 Bealer (1987, 296–​298). Similarly for terms such as ‘proposition’, ‘line’, ‘plus’, ‘red’, ‘conscious’, ‘friendship’, and ‘morally right’, which thus appear to differ from natural kind terms such as ‘water’. See, e.g., Bealer (1996, 134–​135 and 2008, §6.2) on “semantic stability” and Chalmers (2006) on “semantic neutrality” and “non-​ twin-​earthability”. For so-​called moral twin earth scenarios, see Horgan and Timmons (1992). 4 By ‘relevant cases’, I mean cases like the ones described in the previous paragraph. 5 Or explain why it needn’t do these things. For a fuller discussion of philosophical data and the requirement that a theory of a domain handle the data in that domain, see Bengson, Cuneo, and Shafer-​Landau (forthcoming-​a). 6 Intuition-​based theories in the case of reference are endorsed by, e.g., Bealer (1987), Korman (2006, 505ff.), and Boghossian (2014). See also the quotations from Putnam, Burge, and Kripke in section 4. 7 This thesis is inspired by Bealer (1992, 130 n7): “When we say that an intuition counts as prima facie evidence, we of course mean that the content of the intuition counts as prima facie evidence”. Content is Evidence is an alternative to the view that intuition itself—​the mental state—​is evidence (see, e.g., Weatherson 2003; Alexander and Weinberg 2007; Goldman 2007; Alexander 2010; and Mizrahi 2012 and 2013; cp. Moffett forthcoming). The latter view problematically “psychologizes evidence”, to borrow Williamson’s (2007, 211) apt phrase. More on these issues below, in section 4. 8 See, e.g., Bengson (2015a, 2015b, 2015c) and Bengson, Cuneo, and Shafer-​Landau (forthcoming-​b). 9 See Bengson (2010), as well as Koksvik (2013) and Chudnoff (2019), for recent examples and discussion of the mental effort involved in some cases of intuition. But the claim itself is old. Plato required ten years of mathematical training, plus more experience and education besides (see Burnyeat 2000 for an excellent discussion of this requirement). Descartes formulated some of his own lofty demands in Rules for the Direction of the Mind. And recall Frege’s (1884/​1950, vii) comment in the Grundlagen on intellectual vision: “Often it is only after immense intellectual effort, which may have continued over centuries, that humanity at last succeeds in … stripping off the irrelevant accretions which veil [a topic] from the eyes of the mind”. 10 Korman (2006, 505–​506) helpfully discusses intuitions’ role in the case of twin earth (and so-​called dry earth). His treatment arguably extends to the other cases mentioned in section 1. One caveat is that as is standard, Korman’s discussion sometimes emphasizes not intuitions about reference (e.g., what ‘water’ does or does not refer to), but rather intuitions about referents (e.g., what is or is not water). One explanation links x’s being F to ‘F’ referring to x. But another thought is that intuitions about reference are sometimes transparent: in some cases, when we seek to consult our intuitions about whether ‘F’ refers to x, we end up just considering whether x is F. (Thanks to Dan Korman for this suggestion.) 11 See, e.g., Braisby et al. (1996), Machery et al. (2004, 2009, 2015), Jylkkä et al. (2009), Mallon et al. (2009), Machery (2012), Sytsma et al. (2015), Beebe and Undercoffer (2016), and Devitt and Porot (2018). Hansen (2015, §§3–​4) provides an overview of several of the challenges.

572

The Myth of Quick and Easy Intuitions 12 See, e.g., Deutsch (2009), Cullen (2010, §2), Sytsma and Livengood (2011), Ichikawa et al. (2012, 59–​60), Haggqvist and Wikforss (2015), and Baz (2017). 13 See, e.g., Jackman (2008) and Wikforss (2017). 14 See, e.g., Ludwig (2007, 2010), Martí (2009, 2015), and Devitt (2011, 2012). 15 See Bengson (2013) for a much fuller discussion. 16 To illustrate the sort of worry I  have in mind here, consider an alternative principle that Weinberg and Alexander (2014, 201) have proposed in response to my critique of aei; they label it ‘aeie’: “The undesirable effects [epistemically problematic features] are not themselves completely localized to the seeming-​less, ‘blind’ answers, but also manifest in seeming-​ful, intuition-​based answers”. However, replacing premise 3 with aeie does not yield a deductively valid argument. Nor does it generate a cogent argument, one that provides inductively strong support for its conclusion: after all, since aeie simply asserts that subjects’ intuitions possess epistemically problematic features, it’s not suitable as a premise in any rationally persuasive argument for that conclusion. I discuss another strand in Weinberg and Alexander’s response in the main text. 17 Nor can we simply help ourselves, without argument, to the claims that subjects’ prompted answers are indirect evidence of subjects’ intuitions, or that the probability of the hypothesis that subjects’ prompted answers express their intuitions is higher than the probability of the hypothesis that they express other mental states (cp. Sytsma and Livengood 2015, 96 n11 and §4.2.3). 18 Here is another argument for the same conclusion: the conjunction of s1 and s1-​int does not itself justify aei** over various rival principles, such as Some of the subjects’ prompted answers could express subjects’ intuitions—​ though it’s also possible that few if any do so in fact. But such a principle fails to vindicate the experimentalist critique. It follows that the conjunction of s1 and s1-​int do not justify aei** in the manner required to vindicate the experimentalist critique of intuitions. (Here I assume, plausibly, that if q is a premise in an argument, A, and A is not rationally persuasive when premise q is replaced with another premise, r, then: if p does not justify q over r, then p does not justify q to the extent that is required to vindicate A.) 19 Weinberg and Alexander point out that cognitive scientists are happy to refer to the products of System 1 using the language of ‘intuition’ and ‘seeming’. But this does not support the relevant inference. That there are diverse uses of such language, and in particular that there are numerous differences between the referents of cognitive scientists’ and philosophers’ use of such language, is well-​known (see, e.g., Chisholm 1957, 44ff.; Bealer 1992, 130 n10; Osbeck 1999 and 2001; Cappelen 2012, pt. I; Bengson 2014, §§2–​3 and 2015a, §2). Nor is a simple appeal to parsimony capable of justifying dismissal of these differences, especially when alternative models that respect these differences while better explaining the full range of cognitive-​scientific data are available (cp. Chudnoff 2019, §3). 20 Cf. Deutsch’s (2015b, 28–​32) “no-​theory” characterization of intuitions, which fails to recognize such answers. Ironically, although Deutsch is a critic of experimental philosophy, he endorses several of its core commitments, including aei; he writes, “We can assume that a subject’s assent or dissent to a survey question expresses an intuition about the cases the questions concern” (2015b, 24). Cp. Machery’s (2017, ch. 1) “minimalist characterization of the method of cases”, which is allied with an experimental attack on intuitions that is criticizable for the reasons I discussed in section 3. 21 Bengson (2013, §5). There I show that such explanations are better than, or at least just as good as, those that attribute intuitions. In effect, my original discussion already anticipates and responds to Deutsch’s (2015b, 166) objection that the gap between intuitions and answers that I highlight is an idle possibility. 22 I should note that even if this challenge were met, there would still be more work to do. For example, experimentalists would also need to ensure that the studies elicit intuitive presentations that are not extremely hazy or obscure. After all, the conjunction of Quasi-​ Perceptualism, Presentationalism, and Content is Evidence already implies that extremely hazy or obscure presentations possess extremely little justificatory and evidential significance. 23 See, e.g., Bealer (2008) and Bengson, Cuneo, and Shafer-​Landau (forthcoming-​b, §5.2). 24 See, e.g., Ayer (1946, 108). 25 Cappelen (2012, 3). Other critics of Centrality include Williamson (2007, ch. 7), Deutsch (2009, 2010, 2015a, and especially 2015b, xv), Dorr (2010), and Cappelen and Dever (2018, §9.9). I will elide ‘contemporary analytic’ and use ‘as evidence’ as shorthand for ‘as evidence (or sources of evidence)’. 26 Deutsch (2015b, xiii). 27 At least some. I myself favor an intuition-​based theory that endorses a variant on Centrality that takes reliance on intuitions as evidence to be part of the essence of philosophy. This complicates the thesis’s descriptive purport. After all, it may belong to the essence of a heart to pump blood (rather than not), even though there are numerous hearts that fail to pump blood. To simplify the discussion and make direct contact with the extant debate, I focus on Cappelen’s formulation of Centrality. 28 The distinction between state and content in the case of intuition can be traced at least to Lycan (1986, 88) and is stressed by Bealer (see note 7 above) and Pust (2001). For ease of exposition, I will often refer to the former using the ideology of facts, as in ‘the fact that one has such-​and-​such intuition’.

573

John Bengson 29 Deutsch (2015b, 36–​37) invokes the state/​content ambiguity to distinguish two distinct epistemic claims about intuition: EC1 concerns reliance on state as evidence, whereas EC2 concerns reliance on content as evidence. Although Deutsch rejects EC1, he describes EC2 as “unassailable”. 30 Or so I allow. 31 For a fuller discussion of no refer, see Bengson (2014). 32 Putnam (1975, 193). 33 Burge (1979, 88). 34 Burge (1979, 84). 35 Kripke (1980, 14; cp. 42). 36 Of the sort discussed by Cappelen (2012, pt. I). 37 Dorr (2010). The other functions mentioned in this paragraph are discussed by Cappelen (2012, pt. I). 38 Nor does qualify do this when conjoined with no refer, no more than qualifyP does this when conjoined with no referP. Incidentally, recall that I am using ‘as evidence’ as shorthand for ‘as evidence (or source of evidence)’. 39 Cf. Cappelen (2012, 162)  and Deutsch (2015b, §2.3 and passim). I  describe four other ways in which arguments and intuitions are complementary in Bengson (2014, §5.2). 40 Williamson (2007, 219) and Cappelen (2012, 189–​190). 41 Cappelen (2012, 189–​190). 42 Deutsch (2015b, 98–​99). 43 Here I riff on McDowell (2007, 350). 44 Wright (2004, 157). 45 Note from October 18, 1937, reprinted in Wittgenstein (1976, 419).

References Alexander, Joshua. 2010. “Is Experimental Philosophy Philosophically Significant?” Philosophical Psychology, 23: 377–​389. Alexander, Joshua and Jonathan Weinberg. 2007. “Analytic Epistemology and Experimental Philosophy”. Philosophy Compass, 2: 56–​80. Ayer, A. J. 1946. Language,Truth, and Logic, second edition. Penguin. Baz, Avner. 2017. The Crisis of Method in Contemporary Analytic Philosophy. Oxford University Press. Bealer, George. 1987. “The Philosophical Limits of Scientific Essentialism”. Philosophical Perspectives, 1: 289–​365. Bealer, George. 1992. “The Incoherence of Empiricism”. Proceedings of the Aristotelian Society, 66: 99–​138. Bealer, George. 1996. “A Priori Knowledge and the Scope of Philosophy”. Philosophical Studies, 81: 121–​142. Bealer, George. 2008. “Intuition and Modal Error”. In Q. Smith, ed. Epistemology:  New Essays. Oxford University Press. Beebe, James R. and Ryan Undercoffer. 2016. “Individual and Cross-​ Cultural Differences in Semantic Intuitions: New Experimental Findings”. Journal of Cognition and Culture, 16: 322–​357. Bengson, John. 2010. The Intellectual Given. Ph.D. Thesis, University of Texas at Austin. Bengson, John. 2013. “Experimental Attacks on Intuitions and Answers”. Philosophy and Phenomenological Research, 86: 495–​532. Bengson, John. 2014. “How Philosophers Use Intuition and ‘Intuition’”. Philosophical Studies, 171: 555–​576. Bengson, John. 2015a. “The Intellectual Given”. Mind, 495: 707–​760. Bengson, John. 2015b. “Grasping the Third Realm”. Oxford Studies in Epistemology, 5: 1–​38. Bengson, John. 2015c. “A Noetic Theory of Understanding and Intuition as Sense-​Maker”. Inquiry, 58: 633–​668. Bengson, John, Terence Cuneo, and Russ Shafer-​Landau. Forthcoming-​a. Philosophical Methodology: From Data to Theory. Oxford University Press. Bengson, John, Terence Cuneo, and Russ Shafer-​Landau. Forthcoming-​b. “Trusting Moral Intuitions”. Noûs. Boghossian, Paul. 2014. “Philosophy without Intuitions? A Reply to Cappelen”. Analytic Philosophy, 55: 368–​381. Braisby, N., B. Franks, and J. Hampton. 1996. “Essentialism, Word Use, and Concepts”. Cognition, 59: 247–​274. Burge, Tyler. 1979. “Individualism and the Mental”. Midwest Studies in Philosophy, 4: 73–​122. Burnyeat, Myles. 2000. “Plato on Why Mathematics is Good for the Soul”. In T. Smiley, ed. Mathematics and Necessity: Essays in the History of Philosophy. Oxford University Press. Cappelen, Herman. 2012. Philosophy without Intuitions. Oxford University Press. Cappelen, Herman and Joshua Dever. 2018. Puzzles of Reference. Oxford University Press. Chalmers, David. 2006. “The Foundations of Two-​Dimensional Semantics”. In M. Garcia-​Carpintero and J. Macia, eds. Two-​Dimensional Semantics: Foundations and Applications. Oxford University Press. Chisholm, Roderick. 1957. Perceiving: A Philosophical Study. Cornell University Press.

574

The Myth of Quick and Easy Intuitions Chudnoff, Elijah. 2019. “In Search of Intuition”. Australasian Journal of Philosophy, DOI:  10.1080/​ 00048402.2019.1658121. Cullen, Simon. 2010. “Survey-​driven Romanticism”. Review of Philosophy and Psychology, 1: 275–​296. Deutsch, Max. 2009. “Experimental Philosophy and the Theory of Reference”. Mind & Language, 24: 445–​466. Deutsch, Max. 2010. “Intuitions, Counter-​Examples, and Experimental Philosophy”. Review of Philosophy and Psychology, 1: 447–​460. Deutsch, Max. 2015a. “Kripke’s Gödel Case”. In J. Haukioja, ed. Advances in Experimental Philosophy of Language. Bloomsbury. Deutsch, Max. 2015b. The Myth of the Intuitive. MIT Press. Devitt, Michael. 2011. “Experimental Semantics”. Philosophy and Phenomenological Research, 82: 418–​435. Devitt, Michael. 2012. “Whither Experimental Semantics?” Theoria, 27: 5–​36. Devitt, Michael and Nicolas Porot. 2018. “The Reference of Proper Names:  Testing Usage and Intuitions”. Cognitive Science, 42: 1552–​1585. Dorr, Cian. 2010. “Review of Everything Must Go”. Notre Dame Philosophical Reviews. Frege, Gottlob. 1884/​1950. The Foundations of Arithmetic. Trans. J. L. Austin. Northwestern University Press. Goldman, Alvin. 2007. “Philosophical Intuitions: Their Target, Their Source, and Their Epistemic Status”. Grazer Philosophische Studien, 74: 1–​26. Haggqvist, Soren and Asa Wikforss. 2015. “Experimental Semantics:  The Case of Natural Kind Terms”. In J. Haukioja, ed. Advances in Experimental Philosophy of Language. Bloomsbury. Hansen, Nat. 2015. “Experimental Philosophy of Language”. Oxford Handbooks Online. Horgan, Terence and Mark Timmons. 1992. “Troubles for New Wave Moral Semantics: The ‘Open Question Argument’ Revived”. Philosophical Papers, 21: 153–​175. Ichikawa, Jonathan, Ishani Maitra, and Brian Weatherson. 2012. “In Defense of a Kripkean Dogma”. Philosophy and Phenomenological Research, 85: 56–​68. Jackman, Henry. 2008. “Semantic Intuitions, Conceptual Analysis, and Cross-​Cultural Variation”. Philosophical Studies, 146: 159–​177. Jylkkä, J., H. Railo, and J. Haukioja. 2009. “Psychological Essentialism and Semantic Externalism: Evidence for Externalism in Lay Speakers’ Language Use”. Philosophical Psychology, 22: 37–​60. Koksvik, Ole. 2013. “Intuition and Conscious Reasoning”. Philosophical Quarterly, 63: 709–​715. Korman, Daniel Z. 2006.“What Externalists Should Say about Dry Earth”. The Journal of Philosophy, 103: 503–​520. Kripke, Saul. 1980. Naming and Necessity. Harvard University Press. Ludwig, Kirk. 2007. “The Epistemology of Thought Experiments:  First versus Third Person Approaches”. Midwest Studies in Philosophy, 31: 128–​159. Ludwig, Kirk. 2010. “Intuitions and Relativity”. Philosophical Psychology, 23: 427–​445. Lycan, William. 1986. “Moral Facts and Moral Knowledge”. The Southern Journal of Philosophy, 24: 79–​94. Machery, Edouard. 2012. “Expertise and Intuitions about Reference”. Theoria, 73, 37–​54. Machery, Edouard. 2017. Philosophy within its Proper Bounds. Oxford University Press. Machery, Edouard, Max Deutsch, and Justin Sytsma. 2015. “Speaker’s Reference and Cross-​Cultural Semantics”. In A. Bianchi, ed. On Reference. Oxford University Press. Machery, Edouard, Ron Mallon, Shaun Nichols, and Stephen P. Stich. 2004. “Semantics, Cross-​Cultural Style”. Cognition, 92: 1–​12. Machery, E., C.Y. Olivola, and M. de Blanc. 2009. “Linguistic and Metalinguistic Intuitions in the Philosophy of Language”. Analysis, 69: 689–​694. Mallon, Ron, Edouard Machery, Shaun Nichols, and Stephen Stich. 2009. “Against arguments from reference”. Philosophy and Phenomenological Research, 79: 332–​356. Martí, G. 2009. “Against Semantic Multi-​Culturalism”. Analysis, 69: 42–​48. Martí, G. 2015. “General Terms, Hybrid Theories and Ambiguity: A Discussion of Some Experimental Results”. In J. Haukioja, ed. Advances in Experimental Philosophy of Language. Bloomsbury. McDowell, J. 2007. “What Myth?” Inquiry, 50: 338–​351. Mizrahi, Moti. 2012. “Intuition Mongering”. The Reasoner, 6: 169–​170. Mizrahi, Moti. 2013. “More Intuition Mongering”. The Reasoner, 7: 5–​6. Moffett, Marc A. Forthcoming. “Intuitions as Evidence”. In M. Lasonen-​Aarnio and C. Littlejohn, eds. The Routledge Handbook of the Philosophy of Evidence. Routledge. Osbeck, Lisa M. 1999. “Conceptual Problems in the Development of a Psychological Notion of ‘Intuition’”. Journal for the Theory of Social Behaviour, 29: 229–​250. Osbeck, Lisa M. 2001. “Direct Apprehension and Social Construction: Revisiting the Concept of Intuition”. Journal of Theoretical and Philosophical Psychology, 212: 118–​131. Pust, Joel. 2001. “Against Explanationist Skepticism regarding Philosophical Intuitions”. Philosophical Studies, 106: 227–​258.

575

John Bengson Putnam, H. 1975. “The Meaning of ‘Meaning’”. In Mind, Language and Reality: Philosophical Papers, volume 2. Cambridge University Press. Sytsma, Justin and Jonathan Livengood. 2011. “A New Perspective Concerning Experiments on Semantic Intuitions”. Australasian Journal of Philosophy, 89: 315–​332. Sytsma, Justin and Jonathan Livengood. 2015. The Theory and Practice of Experimental Philosophy. Broadview Press. Sytsma, J., J. Livengood, R. Sato, and M. Oguchi. 2015. “Reference in the Land of the Rising Sun: A Cross-​ Cultural Study on the Reference of Proper Names”. Review of Philosophy and Psychology 6: 213–​230. Weatherson, Brian. 2003. “What Good are Counterexamples?” Philosophical Studies, 115: 1–​31. Weinberg, Jonathan and Joshua Alexander. 2014. “The Challenge of Sticking with Intuitions through Thick and Thin”. In A. R. Booth and D. P. Rowbottom, eds. Intuitions. Oxford University Press. Wikforss, Asa. 2017. “Semantic Intuitions and the Theory of Reference”. Teorema, 36: 95–​116. Williamson, Timothy. 2007. The Philosophy of Philosophy. Blackwell. Wittgenstein, Ludwig. 1976. “Cause and Effect: Intuitive Awareness”. Trans. P. Winch. Philosophia, 6: 409–​425. Wright, Crispin. 2004. “Intuition, Entitlement, and the Epistemology of Logical Laws”. Dialectica, 58: 155–​175.

576

INDEX

17th and 18th-​century philosophy 22–​26 19th-​century philosophy  26–​28 A-​intension 7, 174, 177n18, 229, 231, 234, 235, 244, 245, 246, 250, 251, 252n19, 261, 262, 263, 264, 266nn19&21 a posteriori 7, 41, 237, 238, 241, 245, 246, 247, 251, 251n2, 252nn15,17&20, 257, 260, 263, 269, 291, 315, 318, 325, 326, 327, 328, 330, 331, 349, 526, 530n38, 561 a priori 9, 41, 78, 83, 90, 108, 170, 174, 175, 177n15, 239, 240, 245, 246–​247, 250, 251n2, 252n24, 253n26, 257, 258, 259, 260, 261, 262, 263, 265, 265n1, 269, 291, 317, 318, 320, 322n19, 324, 325, 326, 327, 328, 329, 330, 331, 332n5, 360, 361, 362, 368n3, 414, 465, 467, 478n15, 505, 513n13, 527, 554, 561 A-​proposition 245–​246, 247 A-​worlds 244, 246, 247, 252n20 Abelard, Peter 19 absolutism 465, 469, 473; see also Guises Absolutism; Russellian Absolutism abstract objects 377 accidental satisfaction 392–​393 acquaintance 21, 411–​412; acquaintance-​nihilists  412 actions, de se attitudes and 482–​498 additive implicatures 181 affections of the soul 18 Agricola, Rudolph 22 Alexander, Jonathan 565, 572nn7&16, 573n19 Alston, William 179, 180, 185 ancient Greek philosophy 17–​19 answers express intuitions thesis (AEI) 564–​566 antecedently important kinds 311, 316 appellatio 19 Appiah, Anthony 1

Apple 499–​509 arbitrary distinguishing mark 427 Aristotelian essentialism 240 Aristotle 18, 19, 21, 22, 23, 32, 38n12, 42, 43–​44, 47, 50, 83–​84, 100, 109, 248, 283–​284, 285, 315, 339, 359, 364, 389, 394n39, 418n15 Arnauld, Antoine 24–​25 artifacts 292, 295, 296, 298, 300, 322n16, 376, 377, 382n6, 423, 513n11; fictional characters as 375 ascriptions 108, 109, 114, 115, 116, 117, 118, 120n7, 125, 126, 127, 129, 130, 130nn7&8, 136, 137, 138, 139, 164n3, 188, 382n10, 511, 526, 557; attitude ascriptions 111–​112, 190, 380, 381, 381n3, 484 assertoric speaker meaning 440 Astuti, R. 296, 298 Atran, S. 296, 298 attributive/​referential distinction  415 auditory hypnagogia 455 Augustine of Hippo 19 Augustino Nifo 22 Austin, John 7, 179, 180–​181, 184, 185, 186–​190, 193n45 Azzouni, Jody 10, 376, 382n5, 409, 411, 412, 413, 414, 415, 416, 417, 417n1, 418nn17&19, 432n2 B-​conditionals 314, 320 Bach, Kent 180, 181, 182, 183, 184, 185, 190, 191n15, 192n34, 199, 208n4, 403, 412, 417n1, 423, 432nn16&17, 447n14 ‘bachelor’ example 257–​258, 266n14, 290, 311, 317, 321n15 Bacon, Francis 22–​23 Bailenson, J.N. 296 Ball, Brian 180–​181 baptism 5, 45, 78, 99, 101, 102n1, 109, 172, 239, 253n27, 326, 329, 335, 348

577

Index Barcan Marcus, Ruth 3, 222n2 bare noun phrases 538 Bean, T.E. 297–​298 Beck, S.R. 302 Beebe, James 536–​540 Berkeleyian idealism 511 Berndt, T.J.  298 Bezuidenhout, Anne 4, 55, 57, 62, 63, 130n5, 415 biases, conceptual and explanatory 298 Biggs, Stephen 266n14, 332n5, 342nn1&2, 343nn6&7, 353, 369n9, 553 binary relationism 465, 466, 467, 473 Blake, William 453–​454 Bob Dylan example 1, 2, 3, 5, 107–​109, 111, 114–​115, 165 Boethius 19 Borg, Emma 183 Bowie, David 3, 159, 161–​162 Braun, David 7, 10, 116, 118, 120nn9&14, 121n24, 144n14, 155–​157, 161, 164nn3, 5&12, 191n11, 192n43, 375, 376, 377–​378, 380, 381, 381nn1&2, 382nn4&9, 413, 466, 496n5 Brescoll, V.  301 Brock, Stuart 376, 389, 390, 393n4, 394nn23&41 buck-​passing objection  402 ‘bully’ example 314, 315, 316–​317, 318, 320, 321n11 bundles or files 116–​117 Burge, Tyler 80n1, 109, 199, 200, 201–​202, 208nn4, 5&10, 216, 251n4, 553, 555, 568–​569, 572nn1&6 Burgersdijk, Franco 21 Buridan, Jean 21 C-​intension 229, 244, 245, 250, 252n19, 261 C-​proposition 245–​246, 247 C-​world 244, 252n20 Caplan, Ben 10, 120n14, 376, 380 Cappelen, Herman 12, 183, 191n15, 192n29, 468, 470, 472, 476, 478n21, 496n1, 497n7, 512nn2&7, 530n37, 566–​567, 571, 573nn19, 25&27, 574nn36, 37&39; inscrutability of reference 555; Messy Superman 485–​486 Carnap, Rudolf 174, 238, 252n24, 253n26, 332n5, 529n29 Carstensen, A. 302 Castañeda, Héctor-​Neri 499, 512n1, 526 categories, as natural kinds 295–​297 causal-​descriptivism 5, 6, 78, 79, 82, 99, 100, 102, 169–​176, 172–​173, 234, 247, 248, 249, 250, 251, 253n27, 266n13, 276, 278n14, 348, 411; important insights of 90–​91; names 87–​89; non-​classical descriptivism 171–​172; saving important insights 90–​92; two-​dimensional 173–​175 causal-​historical approach, cross-​cultural semantics 535–​548 causal-​historical theory of reference for proper names 82, 174, 176, 234, 544, 548 causal relationship 4, 23, 26, 417 causal theory of reference 79, 82–​93, 97, 98, 249, 250 centrality 566–​567

centred information 229–​231 chain of uses 5, 79, 109, 173, 535 Chalik, L. 297, 299, 300 Chalmers, David 7, 177n18, 237, 242, 248, 250, 251, 251nn1&3, 252nn7, 20&23, 253nn24, 25, 26&33, 260, 261, 265n4, 277nn2&4, 278n14, 328, 341, 466, 467, 478nn3&4, 525, 530nn35, 36, 38&40, 539, 572n3; two-​dimensionalism 246–​247 character 517 Charles Manson example 413, 414 children, natural kinds and 295–​296, 298, 299–​300, 301, 302 Choi, I. 302 Chomsky, Noam 192n28, 365, 366, 417n2; Chomskian internalism 413 ‘Cicero’ and ‘Feynman’ counterexamples 391 Cicero/​Tully example 400–​401 Cimpian, A. 298, 299, 301, 302 Clark, E.V. 207 Clark, H.H. 300 classificatory pluralism 10, 360, 362–​364, 368 clustering 312, 403 Cogito 24 cognitive content 400–​401 cognitive neuropsychiatry 450 cognitive psychology 295–​302 cognitive significance 107–​121, 380–​381; causal theories and 5–​6; clarifying the challenge 109–​112; Frege’s challenge 107–​108; interactions 119–​120; New Theory of Reference 108–​109; taxonomy of approaches 112–​119 Cohnitz, Daniel 553, 557 Collins, John 181, 411, 413, 418nn13&14 combinatorial structures 25 common ground 440, 442–​443 common nature 21 competent rational speaker 109, 110 complex ideas 24 complex intension 319 compositionality of content 523–​524 conceptual structures 26 connotation 27–​28, 41, 212 considered-​as-​actual and considered-​as-​counterfactual 7, 174, 237, 241–​244, 253n25, 261, 262, 265, 266n17 content is evidence 562–​563, 567–​568 context 444–​445; sentences 20, 34, 56 context of utterance 113, 174, 213, 214, 218, 245, 246, 311, 516, 517, 521, 522, 523, 526, 530n37 context-​sensitive 11, 91, 132, 134, 143n1, 214, 252n17, 433n30, 451, 464, 518, 522, 523, 529n21 contextualists 118, 183, 192n34, 253n25, 529n25 conventionalist account 180, 183–​184, 189, 192nn34&45 conventions of speech 22 conversation vs. soliloquy, demonstrative reference 455–​456 Coppock, E. 53, 55, 64, 68 Core Thesis 246–​247

578

Index corporate action 501, 504–​512; no de se requirement 502–​504 corporate agency 500–​502; four theories of de se content 502–​504 counterfactual worlds 229, 242, 243, 273 “counters” 22 Cova, F. 539 Crane, Tim 410, 411, 412, 418n11 Cratylus 17 Crimmins, Mark 117, 135–​136, 137, 143n5, 144n10, 381nn2&3, 382n10, 404 cross-​cultural difference, experimental semantics 535–​548 David, Marian 416 Davidson, Donald 94 Davis, Wayne 120n13, 180, 191n13 De Dialectica (Augustine) 19 de Rosset, Louis 312 de se attitudes/​beliefs 11, 12, 464–​479, 482–​497, 499, 500, 526; and action 482–​497; and action explanation 472–​474; cognitive role of 468–​472; corporate agency and 502–​504, 508–​512; Doctrine of Propositions 483–​484; essentiality 483; expanding the Doctrine of Propositions 486–​487; Frege’s puzzle 470–​472; immunity to error through misidentification (IEM) 475–​477; specification problem 484–​486; traditional theory and 464–​467 de se information 229, 234, 235 declarative sentences 373 Deeb, I. 297, 300 deferred demonstrative reference 11, 450, 453, 454, 455, 456, 462, 462nn2&6, 463nn7&8 defining concept 20, 21, 22 definite descriptions 4, 10, 13, 27, 38n9, 42, 43, 44, 46, 47, 49, 50, 53, 54, 55, 57, 60, 61, 63, 64, 65, 67, 68, 69nn2, 5&6, 74, 75, 77, 79, 80, 81n1, 82, 124, 125, 126, 130n4, 133, 138, 144n12, 171, 176n5, 200, 202, 203, 208n7, 213, 233, 238, 264, 283, 284, 325, 377, 384, 385, 386, 387, 388, 389, 392, 401, 402, 406n1, 409, 410, 414, 416, 450, 516, 538 demonstrative reference 11, 449–​462, 462nn2&6, 463nn7&8; deferred 453–​454; hallucinations 449–​463; hallucinations and general theory of 457–​458; hypnagogia 452–​453, 455; Kaplanian notion of 451; Macbeth’s Dagger 449–​450; numbers and fictional characters 454–​455; obvious question re hallucinations 453–​454; paradigmatically non-​paradigmatic cases 461–​462; provisional picture extended 457; related philosophical problems 458–​461; semantically irrelevant distinctions 455–​457; situation within general theory of demonstrative reference 457–​458; terminology 451–​452 demonstratives: alternative conception of semantics of 444–​446; semantics of indexicals and 442–​444 Dennett, Daniel 207 denotation 27, 28, 65, 88, 98, 100, 174, 212, 213, 284, 347, 350, 352, 353, 354, 356, 423, 454, 529n19

denoting concept 399 Descartes, René 4, 24, 572n9 description theory of reference 87, 227 descriptive identification 75, 77, 109 descriptive semantics 249, 250 descriptivism: cases against 73–​81; in cross-​cultural semantics 535–​548; Frege and names 43; Frege and reference fixing 48–​50; Frege and sense 43–​48; Kripke’s critique of 390–​393; in Kripke’s Frege 42–​43; names from fiction 388–​389 “designating impression” 27 designation of concepts/​objects 26 determinate and diffuse personal supposition 22 Deutsch, Max 567, 571 Devitt, Michael 57, 173, 260, 266n12, 285, 286, 292n3, 332n3, 337, 338, 360, 402, 458–​459, 463n10, 470, 552, 554, 572n11; cross-​cultural semantics 535, 536, 543, 546, 547, 548 diagonalization 251 Dialectic 21–​22 Dialectical Disputations 22 Dialectical Invention 22 diamonds case study 319 Diesendruck, G. 296, 297, 298, 301 disjunctivism 413 divine ideas 25 division of linguistic labor (Putnam) 296 Divisions of Dialectic 22 dominance 99 Donnellan, Keith 3, 4, 53, 54, 55, 56, 57, 58, 59, 64, 65, 66, 67, 68, 109, 124, 184, 188, 202, 240, 251n7, 273, 377, 399, 406n1, 415, 417n1, 418n10, 434n48, 447n7 Donovan, B.M. 298 Dummett, Michael 48, 49, 50, 51nn10&12, 79, 80, 94, 95–​96, 97, 102, 191n5, 238, 251n3 Dupré, John 9, 321n11, 363, 367, 368 Egan, A. 466, 526, 530n37 Elbourne, P. 53, 55, 64–​65, 66, 68 Emmons, N.A. 301 empty names 10, 51n10, 173, 373–​381, 382n9, 417n1; existence of 374–​378; Millian theory of 378–​380; objections to Millianism with gappy propositions 380–​381; problems for Millianism 374 epistemically rewarding (ER) relations 403–​404 epistemological markers, corporate action 505–​506 epistemology of reference 4, 8, 12; Centrality and “the Myth of the Intuitive” 566–​571; data and theories 560–​562; experimental philosophy 563–​566; Intuition-​Based Theory 562–​563; Myth of Quick and Easy Intuitions 571–​572 Estes, Z. 296, 301 Euclid 233, 234 evaluative circumstance 243 Evans, Gareth 5, 80n1, 86, 94, 95, 97–​100, 101, 169, 172, 173, 174, 208n1, 222, 233, 249, 251n2, 277n1, 301, 321n13, 399, 407n9, 412, 415, 417n1, 425,

579

Index 427, 429, 431–​432, 432n19, 433nn27&30, 434n48, 53&56, 445, 447n8, 462n1, 477, 483, 497n14, 502, 526, 530n34, 536 Everett, Anthony 376, 382n9, 411, 418n16 experimental philosophy 12, 340, 535, 539, 563–​566 experimental semantics 535–​548 Expertise Defence 552, 558 explanation, role in cognition 321n7 Explanatory Account 311, 313, 314, 315, 316, 318, 319, 320, 321nn8&9 explanatory value 316 expressible (dicibile) 19 externalism 309–​323 fallacies 4, 19–​20 Fast, A.A. 297 Feyerabend, Paul 10, 359, 368n1 fictional characters, demonstrative reference to 454 fictional names 10, 414 fictional terms 10, 33 fictive sentences 384, 385; reference-​to-​sense view 386–​387; sense without reference 388 Fine, Kit 181, 189, 277n1, 312, 404–​405 first-​level intuitions 557–​558 first-​order internalism  553 First-​Order Logic  182 first-​person pronouns  11, 12 Fleming, Ian 414 Fodor, Jerry 116, 121n24, 265n11, 277n1, 405, 406n2 folk uses 9 “form of language” 26 formal deductive validity 21, 22 formal science instruction 297 formal validity 22 Foster-​Hanson, Emily 8, 9, 300, 301, 343n14 foundational semantics 179, 249 fourth component of speech 19 Frege, Gottlob 2, 54, 68n1, 73, 77, 78, 80, 80n1, 91, 94, 132, 133, 139, 143n2, 154, 157, 169, 174, 177n18, 181, 213, 222n6, 237, 238, 239, 247, 248, 251, 251nn3&4, 252n24, 269, 273, 277n1, 293n23, 368n1, 400, 401, 402, 403, 404, 405, 406, 406nn2, 5&7, 407n9, 412, 450, 460, 483, 484, 486, 496, 496n2, 497n14, 528n2, 535, 572n9; context principle 20, 30, 34, 35, 37; Fregean descriptivism 41–​51; Fregeanism and Russellianism 409–​410; meaningfulness 34–​35; reference and definition 35–​37; reference to numbers 31–​32; sense 32–​34; theories of names from fiction 384, 385, 386, 387, 388, 389, 390, 391, 392, 393nn5, 6&11, 394nn24, 35&39 Frege-​pairs 108–​109, 110, 113, 115, 116, 117, 119, 120n12 Frege-​Russell theory 390–​392 Fregeanism, propositions 409–​410 Frege’s Constraint 401, 403 Frege’s context principle 20, 30, 34, 35, 37

Frege’s Puzzle 3, 5, 6, 84, 107, 109, 110, 112, 153, 164n11, 222n6, 223n11, 238, 464; De Se attitudes and 470–​472 French, J.A. 301 fully competent speaker 189, 190, 247, 248 function of names 10–​11, 19, 422, 425–​427, 428, 429, 430, 432, 433n32 gappy propositions 10, 378–​381, 382n10; objections to Millianism 380–​381 García-​Carpintero, Manuel 6–​7, 179, 181, 183, 184, 185, 186, 187, 188, 189, 191nn4, 5, 10, 14&19, 192nn31, 41&44, 251n1, 252n13, 418n9, 478nn20&22, 479n29 Gauker, Christopher 184 Geiger counter example 229, 230 Geirsson, Heimir 6, 7, 117, 164nn2&12, 251n2, 332n2, 376, 432n18, 434nn40, 51&55 Gelman, Susan 295, 296, 297, 298, 299, 300, 301, 302, 321n7 gender 297, 298, 300, 301 generics 301–​301 Gentner, D. 302 Gilmore, Cody 380 God 23, 24, 25, 459, 470, 472, 511; god’s-​eye perspective 464, 469 Gödel and Peano counterexamples 391 Gödel case 12, 101–​102, 170, 239, 248, 253n33, 390, 392, 418n15, 541, 542, 543, 544, 545, 546, 547, 548, 556; cross-​cultural semantics 535–​548 Goodman, Jeremy 118 Goodman, Rachel 10–​11, 421, 423, 424, 429, 430, 431, 432nn8, 13&18, 433n36, 434nn40, 44, 48&52 Gopnik, Alison 298 Gorgias of Leontini 17 grasping 32, 33, 34, 41, 43, 44, 47, 48, 51n9, 61, 67, 108, 109, 118, 138, 143, 169, 227, 238, 239, 247, 248, 253n33, 271, 352, 353, 355, 356, 379, 402, 406, 412, 423–​424, 430, 431, 433nn21, 25&26, 447n7, 467, 469, 470 Gray, Aidan 5, 201, 208nn3, 6&7, 209nn17&18, 404 Gray, R. 336 Grice, H.P. 56, 59, 65–​66, 124, 127, 128–​129, 179, 180, 181, 185, 186, 187, 189, 190, 191nn14&16, 250–​251, 403, 423, 433n26, 434n48, 440, 447nn4&7 Gricean definition of speaker meaning 440 Guises Absolutism 466–​467, 470, 474 hallucinations 11, 376, 413, 415–​416, 418n19, 462nn5&6, 463n7, 473; demonstrative reference and 449–​463; hypnagogia 452–​453, 455; Macbeth’s Dagger 449–​450; terms and terminology 450–​452 Hanks, Peter 181 Harman, Gilbert 94, 96, 97 Harnish, R.M. 180, 181, 185, 191n15 Hawthorne, John 192n32, 202, 209n16, 409, 412, 417nn1&2, 418nn8&11, 433n21, 434n54 Heck, R.G. 541, 542

580

Index Heim, Irene 61, 179, 182, 191nn19&21, 218, 529n24 Hermione Granger example 10, 384–​390, 392–​393, 393nn9&14, 394nn19, 22&31, 395n47 ‘Hesperus is Phosphorus’ example 90, 91, 154, 164nn2&14, 174, 237, 238, 241, 250, 251, 269, 270, 284, 325, 470–​471 Hirschfeld, L.A. 301 historical-​chain model of reference-​fixing 5, 82, 85–​87, 88, 92n5, 96, 259, 403 Hobbes, Thomas  4, 23 Horn, L. 300 Horowitz, Tamara 252n9, 414 human social categories, natural kinds and 301 human speech 17; three components 18 human understanding 17, 18, 23, 26, 27 Humboldt, Wilhelm von 4, 26–​27 Hume’s Principle 35 Hunter, Julie 179, 191n22 ‘hunter’ example 318 Hussak, L.J. 302 hypnagogia 452–​453, 455 identity statements 9, 34, 41, 108, 109, 174, 175, 237–​253, 253n26, 324, 326, 327, 328, 330–​331, 450, 477, 479n31 ignorance vs. knowledge, demonstrative reference 456–​457 illocutionary 82, 91, 180–​181, 184, 186, 187, 189, 190, 440 immunity to error through misidentification (IEM) 475–​477 implicated normalcy, principle of 127–​129, 130 indexical attitudes 464–​467 indexical expressions 516, 521 indexical knowledge argument 468–​469 indexical shift 520–​523 indexicals: alternative conception of semantics of 444–​446; semantics of demonstratives and 442–​444 inferentialism 561–​562 initial baptism 5, 45, 78, 99, 101, 102n1, 109, 172, 239, 253n27, 326, 329, 335, 348 inner linguistic form 26 Insolubilia 20 intelligibility of names 17 intension: proper names 374; of water 312–​313 intentional artifact theory 101, 375–​376, 377 intentionalist view 180, 183–​184, 187, 189, 190, 192nn34&45 intuition-​based epistemologies  12 intuition-​based epistemology of reference intuition-​based theory 562–​563; content is evidence 562–​563; naïve realism 563; presentationalism 562; quasi-​perceptualism  562 intuitions: about natural kinds 296–​297; different kinds 555–​556; first-​level 557–​558; and intuition reports 556; of laypersons 552; and meta-​ externalism 553–​555; and meta-​internalism 555–​558 intuitive capacities 556

irrealism 26 Izumi, Y.  538 Jackson, Frank 5, 7–​8, 9, 87–​88, 92nn8&10, 174, 177n18, 237, 238, 242, 245–​246, 247, 248–​249, 250, 251, 252nn14&19, 253nn27&33, 258, 260, 261, 262, 265nn3&4, 266nn15, 19&21, 277nn2&4, 278n14, 468; two-​dimensionalism 245–​246 Jackson, Mike 453 Jackson’s causal-​descriptivist response 78, 79, 82, 88, 90, 99, 247, 248, 250, 251, 278n14 James Bond example 413, 414 Jeshion, Robin 176n2, 201, 207, 209n17, 222n7, 411, 412, 417n1, 418nn8&11, 429, 432nn5, 6, 10, 17&19, 433nn21, 25&38, 434nn41, 43, 45, 46&49 John of St. Thomas 21 Jonah case, cross-​cultural semantics 539–​540, 547 ‘Julius,’ proper name example 249–​250 Julius Ceasar problem 35–​36 Kalish, C. 296, 297, 299, 301 Kant, Immanuel 26, 27, 77, 252n24, 325 Kaplan, David 11, 12, 73, 84, 109, 121n24, 132, 143n1, 171, 174, 177n18, 179, 181, 182, 183–​184, 186, 191nn6&23, 192n32, 217, 239, 240, 249, 251nn5&7, 252nn13, 15&17, 268, 270–​271, 272, 273, 277n6, 325, 376, 377, 378, 381n2, 399, 402, 406, 413, 415, 418n8, 423, 432nn5&10, 444, 445, 447n2, 452, 453, 454, 455, 457, 464, 466, 521, 522, 523, 524, 525, 526, 528nn1, 2, 3, 7, 8&15, 529nn16&27, 530n34; context 444–​445; Kaplanian notion of demonstrative reference 451; semantic monsters 515–​519 Katz, Jerrold 554 Keil, F.C. 296, 298, 299 King, Jeffrey 183, 192nn32&34, 209n13, 380, 417n2 Kinzler, K.D. 297 Known Distinct Predicament (KDP) cases 473 Kornblith, Hilary 554 Kripke, Saul 3, 4, 6, 7, 8, 9, 10, 12, 41, 42, 43, 44, 50n2, 51n3&4&5&6, 56, 57, 59, 60, 66, 73, 85, 86, 87, 89, 90, 92n5, 94, 109, 112, 125, 143n3, 154, 163, 164n14, 169, 171, 172, 173, 174, 175, 176, 176n1, 177n20, 179, 187, 189, 198, 199, 202, 213, 214, 222n2, 234, 237, 240, 242, 243, 245, 246, 247, 248, 249, 251, 251nn2&3, 253nn25&31, 259, 266nn12, 16&23, 269, 271, 295, 299, 320n2, 322n20, 324, 325, 326, 328, 330, 332n5, 335, 336, 337, 340, 342, 359, 368n1, 375–​376, 377, 378, 381n1, 382n6, 390, 394nn19, 22, 29, 39&40, 399, 402, 404, 406n6, 413–​414, 418nn10, 15&17, 421, 422, 433n21, 447n15, 450, 497n14, 526, 530n38, 555, 556, 569, 570, 572n6; anti-​descriptivist treatment of names in Naming and Necessity 83, 84, 85; cross-​cultural semantics 535–​548; Frege and Reference-​Fixing Descriptivism 48, 49, 50; ‘Godel’-​’Schmidt’ scenario 92n5; and his opponents 95, 96, 97, 98, 99, 100, 101, 102, 102n1; modal argument 5, 77, 79–​80, 82, 84, 170, 239, 241, 252n14, 418n16;

581

Index notion of rigidity 283, 284, 285, 286, 287, 289, 290, 291, 292n4, 293nn18&23; objections to descriptivism about names from fiction 390–​393; semantic externalism 320 Kripke-​Putnam view of theoretical identity statements 9, 331, 339, 360 Kroon, Fred 78, 99, 172, 173, 234, 253n27, 278n14, 411, 418n7 Kuhn, Thomas 1, 10, 331–​332, 345, 356, 359, 368n1, 463n10 Lam, B. 538–​539 Lancelot, Claude 24 Le Verrier, Urbain 376, 377, 381 learning argument 470 Lefèvre d’Étaples, Jacques 22 Leibniz, Gottfried Wilhelm von 4, 25–​26 lekta 18–​19 Lepore, E. 183, 184, 189, 191n15, 192n29, 530n37 Leslie, S.J. 295, 297, 299, 300, 302 Levine, Joseph 235n3 Lewis, David 5, 78, 87, 88, 92n8, 99, 172, 177n13, 180, 181, 182–​183, 186, 188, 220, 231, 252nn15&17, 253n27, 278n14, 411, 417n3, 418n17, 447nn5&15, 465–​466, 470, 471, 478nn5, 12, 14&15, 483, 490, 497n12, 499, 500, 509, 512n1, 522, 524, 526, 529nn24, 25&26, 530n33, 535; de se attitudes 495–​496; Indexical Knowledge Argument 468–​469 Li, J. 539, 543, 544 linguistic competence 552, 553, 555 Linskey, Bernard 290, 292n10 Linsky, Leonard 238, 251n3 Locke, John 4, 23–​24, 230, 329, 349, 402 logic of induction 22 Logical Introductions 22 Lowell, B. 296 Lycan, William 262, 573n28 Lynch, E.B. 296 Lyons, J. 302 Macbeth’s Dagger 449–​450, 461 Machery, Edouard, cross-​cultural semantics 535–​548 Maier, Emar 179, 188, 191nn19&21, 528n3, 529n17, 530n37 Mallet, J. 297 Mallon, R. 551, 552 Malt, Barbara 295, 296, 299, 339, 365–​366 Mandalaywala, T.M. 297, 298, 299, 301 Mark Twain example 380–​381, 410, 484 Marmor, Andrei 180 Martí’s criticism, theories of reference 542–​543 material supposition 21 Mates, B. 405 maximally simple-​fiction hypothesis 413 Mayr, E. 295, 301 McDowell, John 417n1, 497n14 McGinn, Colin 253n27, 400 McLaughlin, Brian 321n6, 322n19

meaning: demonstrative and indexicals 439–​440; meaningful speech 17, 18; non-​primary speaker reference 441–​442; speaker meaning 440; speaker reference 440–​441 meaning-​shifting operators 524–​526 medieval philosophy 4, 19–​21 Medin, D.L. 296 Meinong, Alexius 54, 410, 451–​452 Meinongians 410–​411 mental files 10, 59, 157–​161, 164n13, 172, 277n1, 403–​406, 406n7, 412, 418n12, 425, 427, 428–​429, 434n41, 466; conception of singular thought (MFC) 428 messy shopper 465, 470, 471, 472, 478n22, 484–​485, 500, 510 meta-​externalism: contrasted with meta-​internalism 553; and intuitions 553–​555 meta-​internalism: contrasted with meta-​externalism 553; and intuitions 555–​558 metafictive sentences 384, 385 metasemantics 7, 179, 180, 182, 183, 184, 189, 190, 237, 248, 249–​251, 277n6, 335, 336, 535, 548 metasemantics/​semantics distinction 7, 179–​190, 249, 251 Mill, John Stuart 4, 27–​28, 41, 212–​213, 237–​238, 295, 362, 373–​381 Millianism 7, 83, 107, 108, 109, 120n5, 154, 155, 157, 171, 190, 208n1, 212–​215, 216, 217, 218, 220, 221, 222, 222nn6, 7&9, 223n12, 237–​240, 338, 339, 341, 342, 373–​381, 382n8, 430; abstract object Millianism 376; artifactual Millianism 376; attitude ascriptions 381; belief attributions 381; objections with gappy propositions 380–​381; possibilist Millianism 376; problems raised by empty names 374 minimalists 183 modal argument 5, 77, 78, 79, 80, 82, 84, 170, 239, 241, 252n14, 418n16 modes of presentation: aspects of content 404–​406; cognitive content 400–​401; descriptions 401–​403 Moltmann, Friederike 180 Moty, K. 300 multivocal 9 Murphy, G.L. 301 Myth of Quick and Easy Intuitions 565, 571–​572 naïve realism 460–​461, 563 name-​based referential communication 426, 427–​428, 430–​431 name-​based singular thought thesis (NBT) 422–​423; assumptions behind 423–​425; pure testimony cases 425–​427 name-​using practices  431 names: bound 218–​219; connotation 27; function of 425–​427, 429; predictivist account of 198–​209; shifted use 7, 215, 220, 223n11 names from fiction 375; Frege on 384; sense without reference 387–​388; stipulationism 384–​386

582

Index Naming and Necessity (Kripke) 3, 73, 77, 81n1, 82, 95, 96, 98, 154, 164n14, 171–​172, 237, 239, 240, 251n3, 252n15, 283, 291, 335, 390, 392, 536, 569, 570; anti-​descriptivist treatment of names in 83–​85 natural kind concepts 8–​9, 296, 297, 300, 343n7, 347 natural kind representations, origins 297–​299 natural kind sense 9, 310, 311, 314, 315, 316, 317, 318, 320, 321nn5&13, 342n1 natural kind terms 1, 3, 4, 5, 82, 83, 84, 85, 87, 88, 89–​90, 231–​233, 238, 239, 241, 242, 245, 249, 252n11, 253n26, 266n14; antecedently important kinds 311; carving nature at its joints 296; categories as 295–​297; concepts exist only in people’s minds 296; externalism 309–​323; inferences can be problematic 301; intuitions about animal species 296–​297; language shapes cognition, development, and behavior 301; psychology of 295–​302; representation of particular categories 299–​300; and rigidity 8–​10, 283–​293; semantics of 309–​311; status in psychology and language 297–​300; variability in beliefs 297 natural languages 25, 27, 33, 44, 46, 47, 179, 180, 183, 185, 189, 413, 416, 515, 528n1 natural phenomenon, reference as 554–​555 Nazzi, Thierry  321n7 Neale, Stephen 11, 56, 180, 183, 187, 189, 190, 191n8, 192n34, 202, 447nn3, 7, 9&10 negative existentials 11, 74, 75, 380, 450, 459–​460, 462 neutral view of truth 416 Nicole, Pierre 24–​25 Ninan, Dilap 12, 191n5, 223n11, 465, 466, 478nn23&24, 486, 497nn7, 10&14, 512nn1&2, 525, 526, 529nn25&26 nominalism 21 nominatio 19 non-​primary speaker reference 441–​442 nonsemantic singularity hypothesis 418n10 noumenon 27 Nowak, Ethan 183, 184 Noyes, A. 299 numbers, demonstrative reference to 454 Nunberg, Geoffrey 57, 183, 519, 528n10 Occam’s Razor 461 Odysseus 389–​390, 394n24 Okasha, Samir 321n11 “On Sense and Reference” (Frege) 2, 3, 4, 38n12, 43, 80n1, 384, 393n6 Organon 22 Pagin, Peter 176n3, 183, 190, 343n9, 406n5 paradigmatically non-​paradigmatic cases 449–​450, 461–​462 paradox of fiction 385 paranoid schizophrenia, auditory hallucinations of 457 Parmenides of Elea 4, 17 Parsons, Terence 376, 377, 382n5, 393nn2, 9, 12&17

Partee, B. 521 Peacocke, Christopher 265n5, 399, 497n14 Peano, G. 86–​87, 391, 392 ‘pediatrician’ example 317–​318 Pelletier, J. 183, 299 perception 18, 26, 76, 77, 100, 402, 416, 450, 452, 460, 469, 470 Perry, John 135–​136, 143n5, 180, 191n8, 381nn2&3, 382n10, 402, 404, 466, 467, 478nn3, 6&18, 482, 496, 496n2, 497nn6&9, 499, 503, 512n1, 526; de se attitudes 496; Doctrine of Propositions 483–​484; messy shopper 465, 470, 471, 472, 478n22, 484–​485, 500, 510 personal supposition 21 perspectives argument 469–​470 pervasive externalism 317–​318; Putnam and 317–​318 Peter of Spain 20, 22 phenomenological markers, corporate action 508–​509 Pierre example (Kripke) 6, 112, 134–​139 Plantinga, Alvin 80 Plato 17–​18, 42, 43, 60, 84, 109, 186, 253n27, 295, 389, 572n9 Platonism 554 Playful Dialectics [Dialectica Ludicra] 22 pluralism 467 Port-​Royal Grammar 24 Port-​Royal Logic 24, 25 positive theory 5, 85, 239 pragmatic ontological commitment 413 Prasada, S. 302 Predelli, Stefano 150–​151, 524, 528n15, 530n27 Premack, D. 302 Prentice, D.A. 297 presentationalism 562 probe question 537–​540 proper names: empty names 373–​381; fictional characters are artifacts 375; referring expressions 28; rigid designators 374 propositional concept 245 propositions, Perry’s doctrine of 483–​484; expanded 486–​487; problem for expanded doctrine 487–​494; specification problem 484–​486; truth and motivation 494–​495 propositions, Russellian 373 Pryor, James 404, 479n31 pure testimony cases 425–​427 Putnam, Hilary 1, 3, 8, 9, 109, 239, 240, 241, 251, 251n7, 266n23, 291–​292, 293nn19&23, 295, 299, 309, 310, 311, 313, 314, 319, 320n2, 321nn5, 6, 9, 10, 14&15, 322n19, 324, 325, 326, 328, 331, 337, 338, 339, 359, 360, 361, 362, 363, 365, 366, 368, 368nn1&7, 463n10, 553, 555, 556, 570, 572nn2&6; division of linguistic labor 296; pervasive externalism 317–​318; semantic externalism 321n10; ‘Twin Earth’ thought experiment 231–​232, 239, 243, 248, 252n18, 330, 340, 341, 342, 558, 568, 572nn3&10 Pylyshyn, Z. 402

583

Index Qua-​Problem 9, 11, 338, 339, 359–​362, 458–​459, 462 quantified modal logic 240 quasi-​perceptualism 562, 565–​566 Quine, Willard Van Orman 80, 92n2, 158, 187, 198, 208n4, 222, 240, 252n23, 295, 299, 332, 357, 459 race and ethnicity 1, 297 Rafetseder, E. 302 Ramus, Peter 22 Recanati, François 10, 57, 59, 130n8, 158, 164nn12&13, 191nn9&10, 208n11, 218, 277n1, 403, 405, 406nn3&7, 411, 412, 417n1, 418n11, 433n21, 434nn43&48, 466, 476, 477, 503, 526 Recipe, semantic argument and 247–​249, 251 reduplication argument 5, 75, 76–​77 Reed, N. 556 Reference and Existence (Kripke) 390 reference borrowing 249, 259 reference change/​preservation 94–​102; Evans and 97–​100; Kripke and 95–​97, 101–​102 reference-​fixing 179–​193; by definite description 377 referential-​attributive distinction 53–​69; cases of misdescription 66–​67; classic treatments of 54–​57; contemporary linguistic treatments of 64–​66; contrasted with epistemic access 57–​62; referential versus attributive uses 62–​63 referential neutrality 413 referential use 4, 56, 57, 58, 125, 130, 201 referring expressions 1, 3, 23, 28, 48, 54, 56, 57, 59, 69n2, 399, 401, 402, 406n1, 553 Reimer, Marga 11, 57, 59, 184, 382nn5&9, 462n1, 463n9 relativism 465–​466, 467, 474 Renaissance philosophy 21–​22 representational properties 238, 248 res 19 res extensa 460 res significata 22 Retrenching of Dialectic and Philosophy 22 Rhetoric 21–​22 Rhodes, Marjorie 8, 9, 296, 297, 298, 299, 300, 301 Richard, Mark 6, 118, 135, 136, 137, 143n2, 144n16, 326, 381nn2&3 Richland, L.E. 302 rigid designator 8, 42, 43, 49, 51n6, 80, 83, 171, 176, 177n18, 202, 203, 218, 229, 239, 241, 242, 243, 252n11, 253n31, 266n18, 269, 283, 289, 291, 293n18, 335, 374 rigidified descriptions 84, 92n4, 234, 245, 335 Rips, L.J. 296, 301 Roberts, Craige 181, 184, 185, 188 Roberts, S.O. 297, 301 Rosch, E. 296 Rostworowski, W. 54, 66–​67, 68 Rothbart, M. 297 Routley, Richard 411, 417n3 Rowling, J.K. 384, 386, 388–​389, 392, 395n47 Russell, Bertrand 4, 5, 28, 41, 42, 54, 55, 56, 57, 59, 67, 68n1, 73, 75, 76, 77, 78, 79, 94, 113, 114, 115,

116, 117, 118, 119, 120, 126, 130n6, 132, 133, 134, 137, 139, 143, 143n2, 174, 187, 198, 200, 202, 208n4, 213, 229, 238, 240, 368n1, 373, 390, 391, 392, 399–​400, 412, 417n2, 432n3, 433n22, 439, 440, 444, 450, 452, 459, 478n8; Paradox 30, 37; pretense 446; propositions 409–​410; theorist of singular thought 421 Russellian Absolutism 467, 470, 474, 478n15 Ryle, Gilbert 461 Sacks, Oliver 11 Sainsbury, Mark 140, 141, 144n14, 405, 411, 417n1 Salmon, Nathan 116, 120nn11, 14&18, 121n24, 164n11, 208n1, 239, 251nn2, 3, 4&5, 252nn7&11, 277n1, 286, 289, 292n1, 293nn12, 13&14, 375, 376, 377, 378, 381, 381nn1&2, 382n6, 394nn24, 25&29, 422, 466, 478n16 Salmon, Wesley 312, 337 ‘Sam Smith’ counterexample 391–​392 ‘Santa Claus’ example 375 sayable [dicibile] 19 Schiffer, Stephen 11, 117, 118, 134, 136, 138, 139, 144nn10, 11&12, 180, 183, 187, 190, 401, 442, 445, 447nn3, 5, 7&9 Scholasticism 27 Schwartz, Stephen 288, 291, 292nn7&10, 293n15, 295, 321n10, 325 scientific realism 9, 338, 339 scope ambiguity 240 Searle, John 176n5, 179, 180, 181, 185, 186, 238, 251n3, 278n14, 403, 535 SELF-​file 404, 406 semantic content: accessibility 428, 430–​431; of proper names 373, 376 semantic hypocrisy 416 semantic-​metasemantic distinction 7, 179–​190, 249, 251 semantic monsters 12; content compositionality 523–​524; four notions of 519–​520; indexical shift 520–​523; Kaplan on 515–​519; meaning-​shifting operators 524–​526 Semantic Relationism 115–​116 semantics and pragmatics 183, 190 sentence token 234–​235 sentences, names from fiction 384 set theory 416 Seuss, Dr. 415 shareability 465, 466, 467, 471, 473, 474 Sherlock Holmes example 10, 375, 376, 377, 378, 380, 381 Short, S.D. 298 Shpall, Samuel 505, 506–​507 Shtulman, A. 295, 298, 301 signification 18, 19–​20, 21, 22, 23, 24 signs of notions 22 similarity 312 simple ideas 23–​24 simple intension 319 simple nontology 410–​411

584

Index simple sentences 109–​112, 114, 116, 145–​151, 154–​157, 163–​165 simple supposition (suppositio simplex) 20 singular identities 241 singular linguistic content 430–​432 singular propositions 399–​400, 416–​417 singular thought 421–​434; about nonexistent things 409–​418; assumptions behind NBT 423–​425; function of names 425–​426; name-​based 421–​423; pure testimony cases 426–​427 singularity, empty and nonempty 413–​416 Sinn (sense) 28, 32, 38n21, 41, 238, 251n3 SML 240, 241 Smyth, K. 297, 298 Snowdon, Paul 413 Soames, Scott 5, 80, 89, 92nn7, 8, 10, 11, 12&13, 114, 118, 120nn4, 11&12, 136, 144n10, 164n11, 171, 175, 176nn6&11, 191n2, 208n1, 251nn3&5, 252nn11, 12&13, 284, 285, 286, 287, 288, 289, 290, 291, 292, 292n8, 336, 337, 381, 422, 430, 466 Sobel, David 321n7, 325 Socrateity 21 soliloquies 450, 455–​456 Sophismata 20 sophisms 19 Sophist (Plato) 17 Sophistical Refutations 19 “souls of words” 22 “sound-​form”  26 speaker meaning 440 speaker reference 440–​441 speaker’s reference vs. semantic reference 540–​542 Speaks, Jeff 460 speech-​act potentials 180, 189, 190 Spelke, E.S. 302 Sperber, D. 60 stability 312 Stalnaker, Robert 8, 11, 179, 181, 184, 188, 243, 245, 249, 250, 251n2, 252nn8, 15&17, 253nn28, 30&32, 268, 270–​271, 274, 275, 276, 277nn2, 3, 4, 6, 7, 8, 12&13, 278n14, 468, 470, 471, 478nn8, 10&11, 497n9, 524, 525, 526, 528nn4&5, 529nn27&31; conception of common ground 440; context as common ground 445–​446; Pragmatic Theory of Assertion 272–​273; semantics of demonstratives 444 Stanley, Jason 80, 128, 176n5, 182, 183, 191n5, 209n16, 252n14, 418n16, 477 Stardust, Ziggy see David Bowie Stenius, Erick 181 Stewart, J. 302 stipulationism 384–​386 Stoics 18–​19 Stojnić, Una 184, 192n31 Stone, M. 184, 189 Strawson, Peter 3, 5, 54–​55, 56, 59, 75, 76, 77, 176n4, 180, 187, 190, 403, 425, 432n5, 433n38 structural markers, corporate action 506–​508 structured propositions 410

substantial form 21 Substitution implicatures 181 Summulae Logicales 20, 22 supposition theory 4, 19, 20, 21, 22 suppositum 19 Swiney, L. 298 syllogisms 21 syncategorematic terms 19 syntactic ambiguity 127 syntactic structures 26, 27 System of Logic 27 Sytsma, J. 536–​538, 541, 542 Tarskian-​Davidsonian truth-​conditional approach  182 Taverna, A.S. 296, 298 Taylor, Kenneth 378, 418n7, 434n52 Taylor, M.G. 297, 298 Tessler, M.H. 301 ‘that’-​clauses  374 ‘The Philosophical Lexicon’ 207, 208 “The Thought”  390 theoretical identity statements 9, 324, 326, 327, 328, 330, 331 theories of direct reference 9, 24, 28, 73, 74, 75, 77, 154, 222, 239, 251n5, 325–​326, 361, 373, 402, 418n8, 423, 516, 520, 524, 528n2 Thomasson, Amie 331, 375, 376, 382n6 token sentence 234–​235 traditional theory of attitudes 464–​467; three arguments against 468–​470 transfictive sentences 384, 385; reference-​to-​sense view 386–​387; sense without reference 388 truth-​conditional content 400–​401 truth-​conditions 2, 5, 34, 35, 82, 83, 108, 109, 118, 132, 134, 213, 245, 283, 288, 289, 290, 354, 355, 356, 357, 400, 406, 446, 488, 492, 495, 496, 516, 569 truth-​correspondence  416 truth-​value 1, 2, 6, 7, 11, 18–​19, 31, 33, 38n9, 55, 108, 111, 118, 130, 133, 134, 145, 147, 148, 149, 154, 155, 156, 157, 160, 161, 162, 163, 180, 201, 202, 208n2, 242, 243, 245, 250, 252n17, 268, 269, 270, 271, 273, 274, 348, 352, 353, 356, 373, 374, 379, 380, 381, 382n10, 385, 386, 388, 390, 409, 411, 413, 414, 416, 417n1, 418n13, 454, 456, 458, 465, 466, 467, 490, 491, 495, 517, 518, 525, 539, 542, 548; truth-​value judgements 543–​547 truth-​value judgments, cross-​cultural semantics and 543–​545 Twin Earth example 231–​232, 239, 243, 248, 252n18, 330, 340, 341, 342, 558, 568, 572nn3&10 two-​dimensional semantics 227–​236 univocality 410 unreal, terminology 451–​452 utterance (verbum) 19 ‘vacuum’ example 315, 316 Valla, Lorenzo 22

585

Index Van Dongen, N. 539 Van Inwagen, Peter 375 variabilism 209n11, 212–​223; empirical challenges 215–​217, 218–​222; Millianism and 212–​214 Vasilyeva, N. 301, 302 Venus 33, 90, 240–​241, 269, 270, 271 Vlach, Frank 526, 530n33 Vulcan example 376, 377–​378, 380, 381 W-​conditionals 310, 311, 312, 319 Walker, Caren 321n7 Water example 9, 91, 228, 231, 232, 241, 245, 247, 250, 253n26, 284, 285, 288, 291, 315, 325, 556, 560, 561, 567, 568 Waxman, S.R. 296, 298, 299, 301 Weinberg, Jonathan 565, 572nn7&16, 573n19 Wellman, H.M. 298 Wettstein, Howard 184

William of Ockham 4, 20–​21 Williams, M.J. 301 Williamson, Timothy 2, 3, 185, 187, 251n2, 571, 572n7, 573n25 Wilson, George 505, 506–​507 Wilson, Jessica 286 Wilson, Mark 253n33 Wilson, R.A. 295, 299, 347 Wittgenstein, Ludwig 1, 191n3, 392, 403, 452, 571 word-​world relationship 19, 21, 23, 26 Wright, Crispin 477 Yagisawa, Takashi 218, 222n9, 376 Yalcin, Seth 180, 191n7, 192n28 Zalta, Edward 376, 377 Zeus example 376 Zimmermann, T.E. 464, 524

586