Shifting Concepts: The Philosophy and Psychology of Conceptual Variability 0198803338, 9780198803331

Citation preview

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

Shifting Concepts

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

Shifting Concepts The Philosophy and Psychology of Conceptual Variability Edited by Teresa Marques and Åsa Wikforss

1

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

3

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © the several contributors 2020 The moral rights of the authors have been asserted First Edition published in 2020 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2020943036 ISBN 978–0–19–880333–1 Printed and bound in Great Britain by Clays Ltd, Elcograf S.p.A. Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

Contents List of Contributors

Introduction: Shifting Concepts Teresa Marques and Åsa Wikforss

vii

1

PART I. HOW CONCEPTS SHIFT: VARIATION ACROSS INDIVIDUALS, TIMES, AND CONTEXTS 1. Mapping Thoughts to Words: Cross-language Differences, Learning, and Communication Barbara C. Malt

17

2. How to Make Psychological Generalizations When Concepts Differ: A Case Study of Conceptual Development Gregory L. Murphy

34

3. When Does Communication Succeed? The Case of General Terms Peter Pagin

51

4. Investigating Differences in People’s Concept Representations James A. Hampton

67

5. Colour Categories in Context Yasmina Jraissati

83

6. The Myth of the Common-sense Conception of Colour Zed Adams and Nat Hansen

106

7. Variation in Natural Kind Concepts Daniel Cohnitz and Jussi Haukioja

128

PART II. TO SHIFT A CONCEPT: CONCEPTUAL REVOLUTION, AMELIORATION, AND PERVERSION 8. Conceptual Revolution Joshua Glasgow

149

9. The Folk Concept of Race Edouard Machery and Luc Faucher

167

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

vi



10. On the Conceptual Mismatch Argument: Descriptions, Disagreement, and Amelioration E. Díaz-León

190

11. Conceptual Fragmentation and the Use of ‘Race’ in Scientific Theorizing Robin O. Andreasen

213

12. How Not to Change the Subject Sally Haslanger

235

13. Amelioration vs Perversion Teresa Marques

260

Index

285

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

List of Contributors Zed Adams Associate Professor The New School for Social Research The New School, New York Robin O. Andreasen Associate Professor Department of Linguistics and Cognitive Science University of Delaware Daniel Cohnitz Professor of Theoretical Philosophy Department of Philosophy and Religious Studies Utrecht University E. Díaz-León Associate Professor Departament of Philosophy University of Barcelona Luc Faucher Associate Professor Department of Philosophy University of Québec at Montréal Joshua Glasgow Associate Professor Philosophy Sonoma State University James A. Hampton Professor Department of Psychology City, University of London Nat Hansen Associate Professor Department of Philosophy University of Reading Sally Haslanger Ford Professor of Philosophy Department of Linguistics and Philosophy Massachusetts Institute of Technology

OUP CORRECTED PROOF – FINAL, 31/7/2020, SPi

viii

  

Jussi Haukioja Professor Department of Philosophy and Religious Studies Norwegian University of Science and Technology Yasmina Jraissati Research Scholar Ronin Institute, USA Edouard Machery Distinguished Professor Director of the Center for Philosophy of Science University of Pittsburgh Barbara C. Malt Professor of Psychology Lehigh University Teresa Marques Assistant Professor and Researcher Department of Philosophy University of Barcelona Gregory L. Murphy Emeritus Professor of Psychology New York University Peter Pagin Professor of Theoretical Philosophy Department of Philosophy Stockholm University Åsa Wikforss Professor of Theoretical Philosophy Department of Philosophy Stockholm University Member of The Swedish Academy

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

Introduction Shifting Concepts Teresa Marques and Åsa Wikforss

Concepts stand at the centre of human cognition. We use concepts in categorizing objects and events, in reasoning and action, and in social interaction. It is therefore not surprising that the study of concepts constitutes a central and vital area of research within philosophy as well as psychology. Since the 1970s psychologists have carried out intriguing experiments testing the role of concepts in categorizing and reasoning. A wide range of concepts have been tested, such as biological kind concepts, artefact concepts, social kind concepts, and competing theories of concepts have been developed to account for the experimental data. A striking finding is that there is a great deal of variation, across individuals and cultures, in categorization behaviour. During the same period, philosophers of language and mind have done important work on the semantic properties of concepts. For instance, in the wake of the groundbreaking work of Kripke (1972) and Putnam (1975), many philosophers have proposed an externalist semantics for concepts, arguing that the concepts that individuals think with are determined by their social and physical environment. An important motivation behind this has been the idea that concepts must be shared, across individuals and cultures. However, there has been very little interaction between the psychology and philosophy of concepts. This is surprising, given that there are important points of contact and shared challenges. A possible explanation of the lack of interaction would be that, as some have suggested, there is not a common topic here since the philosopher’s notion of a concept is distinct from that of the psychologist (Machery, 2009). Ironically, then, this would be an example of the very phenomenon to be investigated: conceptual differences. There are reasons to think this is exaggerated. No doubt, there are differences in focus as well as in methodology. In psychology, the study of concepts often involves investigating patterns of categorization, and typically relies on the words people use in categorizing things into groups or ranking typicality features. In philosophy, authors tend to oscillate between talking of semantics and talking of concepts as psychological building blocks of cognition. Sometimes, also, philosophers are interested in concepts simply as abstract objects (as when they Teresa Marques and Åsa Wikforss, Introduction: Shifting Concepts In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Teresa Marques and Åsa Wikforss. DOI: 10.1093/oso/9780198803331.003.0001

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

2

   ˚  

investigate the concept of justice). However, it is one thing to recognize this theoretical variability in the study of concepts, quite another to suggest that there is not a common research topic—a suggestion that has been disputed both by psychologists and by philosophers (Hampton, 2010; Raffman, 2010). Moreover, even if there are some relevant differences in how concepts are understood in the two disciplines, it is important to explore the possibility of interdisciplinary gains. For instance, with the dawn of experimental philosophy of language, on proper names as well as kind terms, the proposal that the experimental data from psychology lacks relevance to semantics is increasingly difficult to defend. In the other direction, philosophy provides essential tools when it comes to evaluating the implications of these data for concept individuation and communication. In particular, when are we to say that there is a variation in concepts, as opposed to a mere variation in beliefs and intuitions? Time is therefore ripe, we believe, for the study of concepts to go beyond the disciplinary confines. This volume brings together leading psychologists and philosophers working on concepts. The inspiration came from a large transdisciplinary project on concepts and communication, CCCOM, funded through the European Science Foundation. The project illustrated that interaction between the philosophy and psychology of concepts can be extremely fruitful. However, the study of concepts also has a societal dimension. This is unsurprising, given the role of concepts in categorization and reasoning. For instance, concepts may contribute to preserving harmful social structures and they can be abused by demagogues to influence how we think and act. In the last decade, some of the most prolific work on conceptual variability and conceptual change has focused on semantic and conceptual amelioration, and this volume includes recent work in the area. As a result, the book is divided into two main parts: Part I focuses on how concepts in fact vary, along several dimensions, and Part II focuses on the societal role of concepts and attempts to deliberately shift them. Part I discusses questions relating to how concepts appear to vary in a variety of ways, inter- and intrapersonally, and the challenges such variation raises. For instance, what happens to the traditional philosophical assumption that successful communication requires shared concepts? And to what extent can we make psychological generalizations if there are constant shifts of this sort? The variability may be extensional—concerning the way things are categorized into different groups—or it may be intensional—involving differences in features attributed to a category, but also in ratings of importance of different features. How are these types of variations correlated? And there are further dimensions of variability. For instance, in the case of some concepts, the variability may be context-dependent and possibly interest-relative. Moreover, people sometimes appeal in philosophical arguments to the ‘folk concept’ of things, making arm chair assumptions about the nature of such concepts. Another important issue, therefore, is what empirical data tells us about the existence and nature of

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

3

philosophically relevant ‘folk concepts’ Finally, there is the essential metasemantic question concerning how facts about linguistic use map on to concepts. In particular, what role, if any, do external facts play in the determination of concepts (as suggested by content externalism) and how do we determine empirically that they play such a role? Part II focuses on questions relating to the societal role played by concepts. Some of the topics that are raised in Part I are also addressed here: intrapersonal and interpersonal variation; whether variability supports or undermines conceptual internalism or externalism; whether variability affects only beliefs and intuitions without conceptual change; whether a concept can shift in some of its dimensions while remaining the same concept; when folk conceptions purport to track natural kinds; and what implications there are when folk and theoretical concepts don’t match. The new set of questions raised include the following: given inter- and intrapersonal variation, under what conditions can categorizations be improved? Are the revisions and ameliorations of categorization changes in word meaning, or do they affect concepts in another, more psychological, way? And when should conceptual revision, or engineering, be pursued? Should it be pursued for epistemic or theoretical purposes and aims only? Or, if conceptual engineering is feasible, can it also be pursued for social, political, or moral reasons? Presumably, not all metasemantic or conceptual shifts are morally or politically legitimate. Can we delimitate the permissible and the impermissible revisionary projects (Burgess & Plunkett, 2013, Plunkett & Sundell, 2013, Plunkett, 2015)? Several of the chapters in this volume use the concept of race to illustrate issues that are raised by ameliorative projects. There are different reasons to pursue ameliorative projects about race. Some reasons are properly theoretic in nature. Given the interpersonal and intercultural variation, and the mismatch between folk and theoretical race concepts (in biology and in critical social theory, for instance), different theoretical options are available. One is to use different race concepts (not perfectly co-extensional, and not co-intensional) in research about population biology and in social theory. The theoretical improvement of taking this option would be mostly epistemic. But another option is to use race concepts only as social kind concepts, and to abandon talk of race in population biology. Population biology could change terminology, for instance use ‘clades’ instead of ‘races.’ It would thus avoid the social and political risks of regarding race as biological. Taking this option would have the advantage of explaining and addressing unjust racial discrimination (Haslanger, 2012). A further option is to take the social and political risks as a motive to abandon all talk of race, and to replace race terms with theory-specific terms that serve the aims and purposes in each theoretical domain. Conceptual ethics thus goes beyond the psychology, semantics, and metaphysics of conceptual variation, and ponders also on its ethical, social, and political dimensions.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

4

   ˚  

Part I. How Concepts Shift: Variation Across Individuals, Times, and Contexts In Chapter 1, Barbara Malt discusses how languages differ in the way everyday words, such as words for household containers, are mapped onto the world, suggesting that there are important semantic differences that often go undetected. For instance, while ‘bottle’ in English sounds a lot like the Spanish ‘botella’ and the French ‘bouteille’, native speakers apply these words in ways that suggest that they do not have the same extensions. Similar observations hold not only for a wide range of terms picking out human-made objects, but also for things in the natural world (for example, types of vegetables) and human body parts. According to Malt, this reflects the fact that things in the world do not neatly divide themselves into distinctive groupings. Instead, our categorizations are a result of a variety of contingent factors. Malt suggests that these semantic differences pose challenges to communication, particularly as they also arise within a native speaker community. It follows, she argues, that successful communication is not a simple matter of a given word reliably evoking a shared meaning, but requires engaging in social processes where contextual cues and gesturing play an essential role. In Chapter 2, Gregory Murphy examines another source of conceptual variation—the differences between children and adults. Murphy suggests that learning a first language is a process that involves constant change as a result of feedback from the environment. The child faces both the problem of excluding items that do not belong in a category and the problem of including items that are not obviously in the category. Murphy argues that this means that the child’s concepts change greatly over time. Nonetheless, the child manages to communicate. According to Murphy, the resulting conclusion is that communication does not require a perfect match in concepts. Another challenge is how to state psychological generalizations about the child’s concepts, given that these generalizations are defined in terms of adult concepts. Murphy suggests that the solution is to reformulate the generalizations. Instead of saying that children learn what we think of as basic categories, such as dog, we should say that whatever categories they learn first have the properties of being basic, in the sense of being more informative and distinctive than other categories. In Chapter 3, Peter Pagin gives a philosophical account of communicative success that is in line with the observations made by Malt and Murphy—that conceptual differences need not undermine communication. Pagin addresses two important questions. First, given that there are significant conceptual differences across individuals, why is this not commonly known? For instance, there is evidence that people have rather different concepts of body parts—some take the hand to be part of the arm, others don’t—and yet such differences are generally not discovered. Second, do such differences imply that communication fails? Pagin suggests that the reason we do not discover existing conceptual

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

5

differences in daily life is that they are hidden by implicatures. Even a speaker who takes the hand to be part of the arm would not report a hand pain by uttering ‘I have a pain in my arm’ since that would imply that the pain is not in the hand. In response to the second question, Pagin argues that the most fundamental level of communication concerns what is said, and that the success conditions are hyperintensional, but that this does not mean that the relation between the contents of the speaker and the hearer need be identity. On the contrary, Pagin argues, intensional similarity is sufficient. The differences in concepts discussed in the first three chapters focus on extensional differences. In Chapter 4, James Hampton discusses how such extensional differences, measured by typicality and membership judgments, relate to differences in intensions, measured by production of definitions and ratings of feature importance. Surprisingly, recent findings suggest that there is a disconnect between these two types of judgments. The upshot, according to Hampton, is that when one person’s concept differs from another’s, the dimensions of difference for extensional information are unrelated to the dimensions of difference for intensional information. This goes against an important assumption, in psychology as well as philosophy, that intensions and extensions should be strongly related. Hampton suggests that the disconnect found may indicate a dual process theory of categorization, one involving similarity-based associative learning (expressed in typicality judgments) and another involving rule-based categorizing on the basis of observed properties (expressed in feature importance judgments). Hampton also discusses a new study showing that a better correlation was found when people were asked to generate their own properties, rather than rank a fixed set. Another dimension of variation concerns the context sensitivity of some categorizations. In Chapter 5, Yasmina Jraissati discusses how context affects (lexical) colour categorization. Colour categorization is based on relevant similarity with other objects, but the notion of relevance, Jraissati suggests, is determined by the context. For instance, the very same wallet may be classified as red in one context (in the shop among brown wallets) and as brown in another (on a table among red wallets). Our colour experiences are organized into a structured whole, in terms of similarity and opponent relations, and this contributes towards accounting for the property of redness, but the dimensions of this structure, Jraissati argues, are not completely determined by biology, but by the purposes of classification as well. We do not always attend to the total set of colours when we categorize, but only to a subset, as when we categorize wines as ‘red’ or ‘white’ or skin as ‘white’ and ‘black’. This suggests that colour categorizations are not fixed for a given culture and time, but constructed online. At the same time, Jraissati argues, some colour subspaces may manifest a stable structure, yielding more permanent categories, given that some colour sets are often attended to together (the colours of wine or skin) and that the speaker purpose is stable.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

6

   ˚  

In Chapter 6, Zed Adams and Nat Hansen discuss another type of variation relating to the colour concepts. Philosophers arguing about the nature of colour often base their arguments on the assumption that there is a stable, single ‘common sense’ conception of colours. For instance, arguments in support of colour realism take as a premise that we ordinarily conceive of colours as if they were intrinsic, non-relational properties of objects. In their chapter, Adams and Hansen argue that, in fact, beliefs about colour vary along several dimensions and that there is no such thing as the ‘ordinary’ conception of colour. They discuss evidence from experimental studies, as well as history, and suggest that the proper conclusion is that conceptions of colour vary interpersonally, intrapersonally, and historically. For instance, whether we think of colour as a relational or a nonrelational property depends on what kind of object is being considered and what frame of reference we adopt—specifically, whether we adopt an ‘object frame of reference’ or a ‘stimulus frame of reference’. They conclude that we need to reconsider the widely shared assumption that metaphysical theories of colour should be assessed in terms of their fit with the ‘common sense’ conception of colour. Semantic externalism, as noted above, can be seen as a response to the worries raised by interpersonal variation: if people have such varying conceptions of things, then we need something external to anchor the concepts used in communication. At the same time, semantic externalism is an empirical theory and cannot be completely disconnected from the actual use of speakers. In Chapter 7, Daniel Cohnitz and Jussi Haukioja discuss externalist accounts of natural kind terms and how to square these with recent experimental data suggesting a great deal of conceptual variation. They argue that we need keep distinct two questions: what determines the meaning of natural kind terms, and what determines whether a term is a natural kind term. One might give an externalist reply to the first question, while insisting on an internalist answer to the second question, arguing that whether indeed a term is such that it has an externalist semantics is ultimately determined by our dispositions to use the term. Thus, Cohnitz and Haukioja suggest, one can accept first-order externalism combined with meta-internalism. However, they argue, this means that experiments on natural kind terms need be carefully designed, testing not only the dispositions to apply terms but also corrective dispositions of a certain sort.

Part II. To Shift a Concept: Conceptual Revolution, Amelioration, and Perversion In Chapter 8, Joshua Glasgow defends a version of dispositional internalism, and argues that meaning is disposition to use, not actual use. His view is a form of internalism because, according to it, meaning and reference are fixed by speakers’

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

7

commitments, in particular, by speaker’s behavioural dispositions to deploy semantic conversational stoppers. Semantic conversational stoppers reveal a speaker’s unwillingness to compromise, are motivated by the conviction that a dispute is merely verbal, and aim to end a conversation. Glasgow’s view is descriptivist because it holds that what a term refers to is fixed by a description, a set of properties that something has to have for the term to truly apply. Moreover, dispositions to use a term may include deferential dispositions, i.e. dispositions to defer to experts’ in a community. Glasgow argues that words can change meaning completely and radically in two different ways: by having its reference replaced, or by keeping its reference but having its meaning replaced. Word meaning is delimited by our dispositions to accept revisions of use. Glasgow replies to objections to descriptivism, and argues that what guarantees shared reference among speakers is the sharing of the same set of dispositions to deploy semantic conversation-stoppers, not the actual sharing of explicit descriptive beliefs. Hence, the fact that people don’t associate the same descriptions with a term is not an objection to descriptivism. He also agrees that deferential dispositions are not part of semantic content, but points out that what the speaker-source of a term is talking about is what users dispositionally intend to refer to when they parasitically use the term. Finally, Glasgow claims that his view can accommodate Quinean convictions about the open-ended way we use our words, and that any particular choice might be settled by extra-semantic factors. He also suggests that conceptual revolution—replacing a concept associated with a term for another one—might be pursued intentionally. Like other authors in this volume, Glasgow considers the possibility of redefining ‘race’ to refer to social kinds in order to advance moral or political purposes. This conceptual revolution would require people to be willing to negotiate how race terms are used. The consequences of such negotiations are discussed in this volume in the contributions of Díaz-León, Andreasen, and Marques. Whereas Glasgow is addressing shifts in meaning, Edouard Machery and Luc Faucher focus on concepts as psychological phenomena in Chapter 9, namely, concepts as bodies of information used in ‘processes underlying higher cognitive competencies.’ They characterize folk theories as sets of generalizations about a domain that can develop reliably across demographic groups, and which people know tacitly, even if they can’t verbalize such knowledge. Folk theories include folk biology, folk physics, and theory of mind. Machery and Faucher argue that there are constraints on the possible variation the concept of race may undergo. There are internal and external constraints on the concepts people have and share. External constraints are environmental constraints on the concepts people can acquire (for instance, shared observed properties), and there is a need to establish a common reference to a set of objects as a precondition on successful communication and concept formation. Internal constraints are formal and material constraints, e.g. on the structure of the concepts formed. Machery and Faucher

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

8

   ˚  

argue that folk theories are a material constraint that limits the variability of the concept of race that people have (both interpersonal and intrapersonal), and set out to test whether people think about races as biological or social groups. Biological and psychological concepts are observed to have a rich inductive potential, but artefact concepts are not. In a series of experimental studies, they looked for evidence to support the hypothesis that people’s race concepts are of biological groups in the sense of having a rich inductive potential, allowing people to infer that if something falls in a given category then it will likely have properties that other members of the same category also have. Machery and Faucher reported results support the claim that people’s folk concept of race is of a biological kind. They conjecture that if the concept of race is biological, then it is influenced by our folk biology. Since folk biology develops across cultures, its influence on racial concepts guarantees both intrapersonal and interpersonal stability. The remaining chapters in the volume that address issues about race concepts make it clear that some theoretical decisions must be made at this point. If the folk concept of race is influenced by our folk biology and is a biological concept, as Machery and Faucher argue, then there are a number of reasons to think that the semantic upshot of their work would be an error theory: there are no races as biological groups, as Glasgow indicated, and folk concepts of race don’t map onto anything in the world. However, finding that people’s folk concept of race is a biological concept in and of itself does not answer which theoretical race concept, if any, is more useful for specific theoretical aims, nor does it answer the normative question of whether we should continue to deploy any race concept at all, and if so, which one? Different ameliorated notions might be available and useful for different theoretical aims. But if moral and political reasons outweighed theoretical usefulness, maybe race talk should be abandoned entirely. In Chapter 10, Esa Díaz-León considers what we should do when we face a mismatch between folk concepts and various intended (ameliorated) concepts. Like Robin Andreasen’s and Teresa Marques’s chapters, her article is concerned with the normative constraints that should guide ameliorative projects. Díaz-León offers a summary of the dispute between internalist (or descriptivist) and externalist (or causal) theories of reference, and then shows how this dispute plays out in the race case. A conceptual mismatch occurs, she explains, when the intentions (or descriptions) that competent speakers normally associate with a word do not correspond to the extension or referent of the word. The intensional mismatch in the race case is pointed out by Glasgow (2009), where he argues that speakers’ intuitions or central beliefs about race don’t match biological properties or reproductively isolated breeding populations (against theorists like Andreasen (1998) or Kitcher (1999)), since speakers’ reference fixing beliefs only contingently pick reproductively isolated groups. Yet, there are standard externalist replies to the same problem in the case of natural kind terms like ‘water’. And the theoretical

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

9

identification between race and reproductively isolated groups could be, a posteriori, by analogy with the a posteriori theoretical identification between water and H₂O. Glasgow (2009) further argues that there is an extensional mismatch problem between groups that actually share the same visible features and reproductively isolated groups on the one hand, and an extensional mismatch between socially constructed properties and speakers’ folk concept of race on the other. But, also here, Díaz-León claims that philosophers of biology and social constructivists don’t have to agree that the theoretical identification must be a priori. Visible properties could be reference-fixing, not meaning-giving. Díaz-León further claims that a certain mismatch between the referent and the associated description is permissible. Ultimately, in the face of competing theories about race concepts, neither of which match the folk concept, when trying to answer the question of what ‘race’ refers to we should distinguish between merely descriptive and ameliorative projects, i.e. we should distinguish what concept of race people actually have from the concept of race that should be used (within a theory, given its purposes, or more generally). Machery and Faucher’s chapter in this volume addresses the descriptive project, not the ameliorative one. Díaz-León concludes that we need good reasons to prefer pre-theoretic folk concepts over theoretically more useful concepts. If there are good reasons for the ordinary concept to be deferential to the theoretical one, the existence of a mismatch should not be an objection to the theoretical concepts of race. In Chapter 11, Robin Andreasen focuses on the additional problem of conceptual fragmentation. This problem is addressed in order to raise the question what normative (moral and political) constraints should guide our theoretical decisions surrounding the permissibility of race talk. She starts by adopting Taylor and Vickers’ (2017) definition of conceptual fragmentation, and then argues that the concept race is fragmented. The notion of conceptual fragmentation is the idea that there is more than one legitimate meaning for a given word. Andreasen addresses several questions about what to do in the face of the fragmentation of a concept: should the term have more than one sense or should its use be eliminated on all senses but one? She also distinguishes, after Taylor and Vickers, between pluralism, selective eliminativism, and reconstructive eliminativism: the first allows for multiple meanings associated with a word, the second recommends abandoning one of the meanings and preserving only another (more useful) sense of a word, and the final recommends abandoning using the word and replacing the needed useful senses with other terms. The chapter offers an overlook of the various positions on race, and offers a sensible assessment of metaphysicalsemantic arguments about what race means: ‘The problem with the metaphysical-semantic strategy, at this point, should be clear. There is little independent agreement amongst reconstructive “race” eliminativists, selective “race” eliminativists, and “race” pluralists on a number of key semantic issues. As a result, this type of strategy often devolves into a semantic debate that falls

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

10

   ˚  

outside the question at hand—namely, how scientists ought to use “race” in theoretical contexts. Consequently, this type of argument strategy is not the best way to go.’ Andreasen’s conclusion focuses on the moral reasons to adopt reconstructive eliminativism about race, a conclusion that agrees with Marques’s pessimism about race-talk in her contribution to this volume. In Chapter 12, Sally Haslanger argues for conceptual amelioration, and makes the point that semantic contestations can shape ‘our agency and our lives together’ if it comes with political and material change. Her view, she argues, is a form of externalism. Haslanger further claims that we can ameliorate concepts themselves. This contrasts prima facie with Glasgow’s chapter, which argues for conceptual revolution, i.e. for the complete replacement of either the extension or the meaning of a word, and for dispositional internalism. For Haslanger, mental states and utterances have propositions as contents, where propositions are truthconditions or sets of possible worlds. Informational contents are partitions of logical space, they divide possibilities or possible objects, but concepts themselves are not part of the content of propositions. Rather, for her, concepts are modes of presentation of partitions of the world. Concept possession concerns how we articulate information in our environment, and shared concepts are important for responding to and coordinating with others. This is an externalist view because, Haslanger claims, concepts are not descriptive beliefs ‘in the head’, but rather capacities to process information about the world. One may wonder whether this is a sufficiently externalist view of concepts. To recall, Glasgow’s dispositional internalism does not equate meanings with descriptive beliefs, but with dispositions. Haslanger distinguishes two types of conceptual amelioration: epistemic and semantic. Whereas epistemic amelioration improves our understanding of informational contents of concepts, semantic amelioration changes the partition of logical space that serves as the content of a term or concept. It may be alethic, allowing us to better capture truths, or it may also be pragmatic. Like Glasgow, Haslanger considers the case of race concepts, and suggests that we may shift ‘race’ from a biological to a social term, which she proposes could ‘unmask’ ideology. Semantic amelioration may also be motivated by moral reasons. Yet, unlike Glasgow, Haslanger says that amelioration does not require conceptual replacement, and that concepts themselves can be improved. Haslanger replies to two objections to her view. One objection is conceptual essentialism—that changing the semantics ‘changes the subject’. The second objection concerns the impracticality of conceptual amelioration: whatever semantic amelioration requires may be epistemically inscrutable, and metaphysically out of our control (Cappelen, 2018). Haslanger’s account of semantic amelioration rests on a Stalnakerian two-dimensional metasemantic model: a concept can shift the partition of logical space which is its content, while remaining the same concept. So, she claims, conceptual change does not entail conceptual replacement. Concerning the impracticality objection, Haslanger concedes that

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

11

there is no algorithm for when or how to ameliorate. This depends on the point, purpose, and function of a concept. Cappelen (2018) objects that there is no meaningful sense of function, and the function of our use of words depends on human intentions and purposes. For instance, a screw driver can be used to open a can of paint if that’s the function we need it to perform. Amie Thomasson (2019) says that some dimensions of a concept’s function are independent of human intentions and purposes. Concepts, like artefacts, can have proper functions. But Haslanger’s reply to Cappelen differs from Thomason’s. Haslanger claims that something has a given function if it produces the selected functions of a system. Different concepts may provide different perspectives on the same content, which over time may lead to changes in the content itself relative to one system. If a system is bad (in some respect, for someone) then this gives reasons to disrupt that system by changing expectations and practices. Haslanger concludes by conceding that in any case semantic amelioration requires involvement in social movements and that philosophers ‘are sometimes engaged in working with social movements to determine what changes would result in a society that is more just and less harmful and how these changes might be brought about. Conceptual engineering can be part of such politically engaged activity.’ In Chapter 13, Marques draws attention to a neglected dimension of discussions on conceptual amelioration. Haslanger concluded her chapter inserting conceptual engineering in the context of broader movements of social engineering. Marques’s chapter narrows down the range of permissible answers to a question that Haslanger’s chapter raised. Haslanger says that she does not offer an algorithm to ‘guide us in deciding when and how to ameliorate’, and hence that she does not have an answer to the central question ‘how we should go on’. By proposing a way of understanding the mechanisms of meaning perversions, Marques’s chapter offers an answer to a related question—how we should not go on. Marques introduces the notion of a meaning or conceptual perversion as the dark side of conceptual amelioration. She draws from historical cases that should be seen as uncontroverted cases of purposeful meaning changes done for social and political purposes, but that were not ameliorations. They may have been advanced by social movements that purported to create a more just society, or pursued (for some time and by some people) in good faith. They were ultimately, politically illegitimate. Just like social engineering can be harmful, so can conceptual engineering. Marques emphasizes, thus, that the question ‘how can we assess the political and moral legitimacy of a meaning revision?’ must be added to the list of other epistemic, metaphysical, and metasemantic questions that are part of the assessment of conceptual engineering proposals. She argues that it doesn’t suffice to postulate good aims or purposes. Marques then considers two senses of conceptual or meaning perversion: a causal one, where a meaning perversion is a revision that has harmful effects, and a constitutive one. Hampton’s chapter in this volume mentions dual processes of categorization,

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

12

   ˚  

which involve typicality and rule (or norm) based considerations. Marques draws from related work on dual character concepts (Knobe & Prasada, 2011, Knobe et al., 2013) to analyse how a constitutive meaning perversion operates. To this effect, she modifies Jason Stanley’s (2015) definition of undermining propaganda, which includes different phenomena (dogwhistles, bullshitting, etc.). On Marques’s account, a use of a word is a perversion when it is false that the word applies to what the speaker intends to refer to, but the use of the word nonetheless has an illocutionary expressive effect—it expresses a conative state to the effect that what the speaker intends to refer to realizes a certain value or norm. Because of this, the use of the word undermines, while seeming to enforce, the value or norm expressed. Perversions are especially hard to resist and to reply to. For instance, it is difficult to resist calls for respecting the will of the people, since they seem essential to democracy. However, in the mouth of many demagogues, these uses of ‘the people’ exclude most of the people. These aspects of conceptual or meaning engineering are instrumental in social engineering. Marques’s chapter hence demarcates the conceptual revisions that should not be pursued.¹

References Andreasen, R. O. (1998). A new perspective on the race debate. British Journal of the Philosophy of Science, 49(2), 199–225. Burgess, A., & Plunkett, D. (2013). Conceptual ethics (I and II). Philosophy Compass, 8(12), 1091–1111 and 1102–10. Cappelen, H. (2018). Fixing language. Oxford: Oxford University Press. Glasgow, J. (2009). A theory of race. New York: Routledge. Hampton, J. (2010). Concept talk cannot be avoided. Behavioral and Brain Sciences, 33(2–3), 212–13. Haslanger, S. (2012). Resisting reality: Social construction and social critique. Oxford: Oxford University Press. Kitcher, P. (1999). Race, ethnicity, biology, culture. In L. Harris (ed.), Racism (pp. 87–117). Amherst, NY: Humanity Books. Knobe, J. and Prasada, S. (2011). Dual character concepts. In Proceedings of the 33rd annual conference of the cognitive science society. Boston, MA: Cognitive Science Society.

¹ Financial support was provided by the European Science Foundation EUROCORES project CCCOM, and FCT—Portuguese Government project EuroUnders/0001/2010; by the DGI of the Spanish Government, projects FFI2016-80588-R and FFI2015-73767-JIN; European Union’s Horizon 2020 programme under Grant Agreement no. 675415, DIAPHORA. We thank all the authors who participated in this volume, and the project participants in the EuroUnderstanding ESF EUROCORES project.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

13

Knobe, J., Prasada, S., and Newman, G. (2013). Dual character concepts and the normative dimension of conceptual representation. Cognition, 127(2), 242–57. Kripke, S. (1972). Naming and necessity. Cambridge, MA: Harvard University Press. Machery, E. (2009). Doing without concepts. Oxford: Oxford University Press. Plunkett, D., and Sundell, T. (2013). Disagreement and the semantics of normative and evaluative terms. Philosophers’ Imprint, 13(23), 1–37. Plunkett, D. (2015). Which concepts should we use?: Metalinguistic negotiations and the methodology of philosophy. Inquiry, 58(7–8), 828–74. Putnam, H. (1975). The meaning of ‘meaning’, language, mind, and knowledge. Minnesota Studies in the Philosophy of Science, 7, 131–93. Raffman, D. (2010). Can we do without concepts? Comments on Machery’s Doing without concepts. Philosophical Studies, 149(3), 423–7. Stanley, J. (2015). How propaganda works. Princeton: Princeton University Press. Taylor, H., and Vickers, P. (2017). Conceptual fragmentation and the rise of eliminativism. European Journal for Philosophy of Science, 7, 17–20. Thomasson, A. (2019). A pragmatic method for conceptual ethics. In A. Burgess, H. Cappelen, & D. Plunkett (Eds.), Conceptual ethics and conceptual engineering. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

PART I

HOW CONCEPTS SHIFT: VARIATION ACROSS INDIVIDUALS, TIMES, AND CONTEXTS

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

1 Mapping Thoughts to Words Cross-language Differences, Learning, and Communication Barbara C. Malt

The human experience of the world through our five senses is rich. We perceive objects and their parts and properties, actions, complex events, and relations among objects, parts, properties, actions, and events, and so on. To communicate about this rich experience, an observer must encode individual elements of the experience in words. For English speakers, such words include common nouns like tree, house, body, branch, window, and finger, adjectives such as blue, red, and soft, relational terms such as above and around, mental and physical action verbs such as thinking, biking, and kicking, and so on. To complete the communication, of course, an addressee must interpret the words and the utterances they form to understand the speaker’s message as it was intended. This chapter considers these issues from four different angles. First, how does the world get mapped into words, i.e. what principles or processes guide the way that a language divides up elements of the world into words, allowing its speakers to map experiences into words? Second, do languages differ in those mappings, and if so, how and why? Third, given that the answer to the preceding question is that languages do differ in their mappings (so the mapping of experiences to words is not entirely straightforward and universal), how is the complexity of mappings dealt with within a language? This question applies both to how children acquire word meanings and how speakers of a given language achieve successful communication. Last, it considers how the complexity of mappings is dealt with across languages, i.e. how do the differing mappings across languages affect second language learners and bilinguals? And how do they affect communicating in a single language when one or more of the participants is not a native speaker of that language?

Barbara C. Malt, Mapping Thoughts to Words: Cross-language Differences, Learning, and Communication In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Barbara C. Malt. DOI: 10.1093/oso/9780198803331.003.0002

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

18

 . 

1.1 How Does the World Get Mapped into Words? One possibility for how people map their experience of the world into words would be that they perceive the world as falling into natural groupings of things, and these groupings are distinctive enough that people feel the need to apply separate labels to these things. So, for instance, people might perceive cats and dogs as separate kinds of animals. They would also perceive fruits and vegetables as separate kinds of things, and likewise, bottles and jars, and so on. In each case, they would make a lexical distinction in the form of contrasting nouns to label the different groupings of things. A proposal along these lines was made by Rosch and colleagues in the 1970s (Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976) and was widely accepted as an account of both how people ‘categorize’ the world (perceive things in some general-purpose, non-linguistic way and group them mentally) and how they would apply names to things in the world. But a closer look suggests that, for at least some domains of the world, the situation is not so simple. Properties of things—especially human-made things— can come in many different combinations. For instance, in the domain of ordinary household containers, there are all sorts of combinations of sizes, weights, types of caps, types of contents, and purpose and manner of being used. Objects aren’t strictly limited to a set of those that are tall, narrow, and used for liquid and a set of those that are short, squat, and used for solid or dry goods. There are ones that are tall and narrow; short but skinny; ones that start out fat but then taper to thinner; ones with pump tops or squeeze sides; ones with straws; and ones that attach with a clip to a purse or knapsack. And some of the short ones hold liquids or gels and some of the taller ones hold dry items, like fish food or peppercorns (e.g. Malt, Sloman, Gennari, Shi, & Wang, 1999). So, when a person thinks about the English words bottle and jar, he or she may bring to mind typical instances of each word that are distinctly different. But starting instead by asking what is actually out there in the world, it seems that things in the world do not divide themselves so neatly into two distinctive groupings. The same is true in other domains. For instance, the words cup and glass may bring to mind a typical cup, which might be opaque and relatively short and used for drinking a hot liquid like coffee, and a typical glass, which might be transparent and taller and used for drinking a cold drink like water. Looking across all the familiar drinking vessels, though, it becomes apparent that many different combinations of properties exist. There are tall plastic cups, short paper cups, medium-sized children’s sippy cups, ceramic coffee cups, squat plastic wine glasses, and so on. Looking at the whole array of them together, it is not so clear where the lexical distinction should be drawn, or even how many different lexical distinctions would be useful to separate instances of the domain.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

19

The absence of obvious boundaries between things in the world may be most prominent when considering human-made objects, but such objects account for a large proportion of the things that humans in industrialized cultures interact with on a daily basis. A typical living room or office, for instance, contains objects spanning thirty or more different named types. The observation also holds true to at least some extent even for things in the natural world. For instance, the dog– wolf boundary is indistinct. Dogs and wolves are able to cross-breed and are considered a single species (e.g. Wilson & Reeder, 2005), with the high similarity of some dogs to wolves notable. They are differentiated based on domesticity, not on some larger constellation of properties that creates a clean break between the two. And although many people believe that there is a neat botanical distinction between things labelled fruit in English and those labelled vegetable, there actually is not. Many foods called vegetable are, botanically, the fruit of a plant (tomatoes, eggplants, zucchini, green peppers, cucumbers, among others; e.g. Live Science Staff, 2012). The label applied to them as foods has more to do with their taste (savoury or sweet) and the typical ways in which they are prepared and served. As such, there is no fixed boundary between the ones called fruit and the ones called vegetable. Sweet potatoes, sweet peas, and Sweet 100 cherry tomatoes are sweet, but are called vegetables, whereas apples, lemons, and cherries can be sour or tart but are called fruits and enter into main dishes, salads, and desserts (further examples are discussed later). How do the lexical distinctions that we have come about? There is a tendency to imagine that it happens in the way that researchers ask people to create categories in cognitive psychology laboratories. In the lab, participants are typically given an array of pictures to view and are asked to create groups. People survey the array and choose some basis to divide up the pictures. The basis they choose is sometimes a single salient property of the objects (e.g. separating all the tall ones from all the short ones). Sometimes it is a combination of properties (e.g. separating all the ones that are more rounded, lighter coloured, and smaller from those that are more angular, darker, and larger (Spalding & Murphy, 1996). Regardless, the important point for current purposes is that they choose some basis, having had the chance to first obtain an overview of the whole domain before making a choice. But in the real world, people rarely view whole sets of examples of a given domain and then create a way of dividing them from scratch. Instead, they make a choice of how to label each object individually as they encounter it and as the need for communicating about it arises. Each choice they make is constrained by the set of names available to them, which they learned from previous generations of language users. And those generations, in turn, usually did not view an entire array from a domain all at once and make a decision from scratch about what kind of lexical distinction was most important to apply before they made use of words from the domain.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

20

 . 

Instead, the typical way that a word may become associated with some set of objects within a domain is probably more that, at some point in a language’s history, some sufficiently novel object or objects came into existence such that a new word was coined to label them. Eventually, modest variations of those objects were created or became familiar by being brought into the culture from outside. At each point in time, the existing word was extended to cover the new instances as long as they had a recognizable similarity to the existing instances. This is much more parsimonious than inventing a new word every time some new variation comes along. Thus, over time, chains of instances that share a name may develop, where the early members of the chain may be rather dissimilar from the most recent members of the chain (e.g. Lakoff, 1990; Malt et al., 1999). Children will learn the range of instances that a word applies to from observing their parents’ use of the word, and they will in turn pass on the conventional uses of the word in their language to the next generation. The result is that patterns of naming may drift away from the pattern that would be predicted if it were possible to survey the entire domain at once and apply names to divide up the instances in the simplest or most compact way possible. A good example of this sort of process is for telephones and other electronic communication devices. Alexander Graham Bell’s first telephones in the 1870s were made up of only a handful of mechanical pieces that transmitted information as sound waves (‘Invention of the telephone’, 2016). Many changes followed over the decades, yielding much more sophisticated instruments with greatly improved function and relatively little resemblance in form to the first telephones (think ‘princess’ phones of the 1960s). After 100 years of evolution, the first cellular phones emerged, making a dramatic step in form and soon thereafter in function, adding capabilities such as text messaging. Modern smart phones have further evolved in form and now afford many functions of a laptop computer, including camera function and video recording, document display, internet browsing, and gaming capabilities. Yet they are still called a phone because of the historical evolution by which they emerged from prior devices with that name (Malt, 2010). Most recently, we have smart watches, too, that are very similar in function to the smart phone, but that are worn on the wrist. Existing along with them are other personal computing devices, including desktop computers, laptops, and tablets. The overlap of functions is substantial, with many of the differences relating to size, speed, storage capacity, and ability to carry out multiple resource-intensive functions at once. Imagine that a set of these objects were placed in front of someone along with a sampling of telephones from their earliest history to the modern smart phone, and suppose that person was asked to divide them according to their similarity or according to what kind of groupings intuitively made sense. We might very well find that people would create groupings that separated all of the modern computing devices, including smart phones,

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

21

from the earlier telephones including, perhaps, the first mobile phones that functioned only for voice transmission. If groupings like those were the basis for what we call these objects, then the pattern of how names are applied to the objects would be very different from what it actually is. The key point is that patterns of naming are strongly influenced by the historical sequence and are not based on any individual’s assessment of the entire domain at the time he or she chooses a name for a given object (Malt, 2010). A secondary point is also illustrated by these and examples of other consumer goods: Companies can influence the naming patterns by their choice of how to introduce new products into the market (Malt, 2010). Sometimes their goal is to position their new product as the latest, and best, instance of a kind of entity that already has a well-established name. In that case, they may carry the same name forward, emphasizing specific attributes of the product in their ads. In other cases, they may want to set the new product apart from other existing, related products. In that case, they may introduce a new label along with the product. Laptop computers share the name computer with desktop computers, for instance, with only a modifier providing a cue to the difference in size and portability. In contrast, tablets were given the more distinctively new name tablet that doesn’t obligatorily include the label computer (though sometimes they are called tablet computer), and that name perhaps signals more innovations than just the difference in size and portability. Another illustration of this influence would be with some eating utensils made for travel and camping. Things called fork and spoon, in everyday kitchenware, can each take on a wide range of forms and functions. The most typical version of each is for conveying solids or liquids to an individual’s mouth during a meal. But there are items with sharp, long tines and a long handle, intended for stabilizing meat while carving (carving forks) and items with slots to drain out liquids while serving (slotted spoons). Given the wide range of types of instances that each noun encompasses, one could imagine that an object intended to minimize the number of utensils needed for travel, combining a rounded bowl for conveying liquids to the mouth with a spiky end for spearing solids, could be called either fork or spoon. However, they are sold under the name spork, which highlights their dual function. Now one company also offers an object called a scork, which is extremely similar to a spork except that the handle end has a notch that can be used as a can opener. This object could have reasonably been labelled a spork, but the company chose to emphasize its subtle difference by giving it a different name (Malt, 2010). In sum, we’ve seen that how language users map experience in the world into words in any given situation is influenced by several facts. These include that (a) there aren’t always such natural and obvious groupings of entities within a domain that the number and type of lexical distinctions to make are fully dictated by structure in the domain itself; (b) individuals don’t get to consider from scratch what kind of lexical distinctions would make the most sense if all the entities

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

22

 . 

within that domain could be taken into account at once. Instead, they inherit a set of lexical distinctions from the preceding generation of language users, who, in turn, inherited it from generations before them; (c) as the preceding points imply, the development of naming patterns within a domain is a process that unfolds over a language’s history and is influenced by what entities within the domains speakers are exposed to and have needed to label at each step of the development, as well as which entities they have been exposed to and needed to name in the past; (d) name choices can be influenced by those distinctions that are important to flag with distinctive names (to whoever bestows the name), and those that aren’t. The preceding is not necessarily an exhaustive account of the complications that resulted in the naming choices that any individual makes for some portion of their experience at any given moment. Section 1.2 looks at how naming patterns are similar or different across languages. This discussion will raise some additional considerations about how naming patterns may be influenced.

1.2 How Do Languages Differ in the Mappings? If people actually did perceive the world as falling into natural groupings that demanded distinctive names, we might expect that, to the extent that people around the world share experiences, they would create parallel mappings of those experiences into language. So, for instance, people everywhere might perceive bottles and jars as separate kinds of containers, and their languages would contain separate nouns for bottles and jars. The meanings would be parallel, although the words would differ in their form. The impression that this could be true is especially strong when noticing cognates, e.g. English has bottle, Spanish has botella, French has bouteille, and Italian has bottiglia. As we’ve already seen, though, the situation is more complicated than that. Things in the world don’t always fall into clear groupings, and multiple factors influence the particular distinctions that are labelled by speakers of a given language. Looking across languages, the result is that languages differ from one another in the distinctions they make, choosing different ways of mapping experience into words. Languages may differ not only in how finely they segment a given domain, but also in what the prototypical referents of the words are, where the boundaries of use lie, and even which dimensions of the domain they encode. In an early study revealing these differences, Malt et al. (1999) created a set of sixty pictures of common household containers and asked speakers of English, Spanish, and Mandarin to say what they would call each one. The extent of the differences across the three languages was surprising. English speakers divided the objects mainly into things labelled bottle, jar, or container (with a few being called by other names such as jug and box). Spanish speakers called almost half the objects by a single name, frasco, that was applied to all the objects that English

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

23

speakers called jar but also applied to a number of objects they called bottle or container. The remaining objects were distributed across fourteen other names (most of which were used for only a few objects). Surprisingly, Spanish botella was used for only three out of the sixteen objects called bottle by English speakers. (The remainder were spread across five other Spanish names). Mandarin speakers, on the other hand, applied a single label to forty out of the sixty objects. They used four additional names, one of which covered another ten objects. The remaining three names were applied to small numbers of objects. Other types of common household objects also show marked differences in how languages divide them up by name. Kronenfeld, Armstrong, and Wilmoth (1985) found that speakers of English, Hebrew, and Japanese divided drinking vessels differently. English and Hebrew each had two labelled distinctions but the instances that they separated differed, and Japanese had three labelled distinctions. Looking at drinking vessels in more detail, Pavlenko and Malt (2011) compared American English and Russian and found that the English cup vs glass distinction is similar to, but not fully mirrored in, Russian. English cup is used very broadly, covering not only coffee and tea cups, but also paper and plastic cups, children’s sippy cups, and others, whereas English glass is used narrowly for objects of a restricted shape made of glass. Although usually translated as if it were equivalent to cup, Russian chashka is a narrower term, used for coffee- and teatype cups. In contrast, stakan, typically translated as glass, is used broadly for a wide range of other types of drinking vessels. Even where the terms overlapped in their extension, the typicality of objects differed. The most typical Russian chashka was a teacup, but the most typical American English cup was a handle-less plastic cup. Different labelled divisions of objects for preparing, serving, and storing foods occur even when speakers of the languages compared largely share culture, as for Dutch- and French-speaking Belgians (Ameel, Storms, Malt, & Sloman, 2005). This phenomenon goes well beyond household objects and extends to domains where experiences are similar around the globe. For instance, Majid, Enfield, and van Staden (2006) show that languages vary in how they distinguish human body parts by name. Although joints may seem to provide an obvious demarcation of body parts, languages vary in which segments they distinguish from which. Japanese distinguishes the hand from the arm but not the foot from the leg, and Indonesian does not distinguish either. In a more abstract domain, the spatial relations that languages encode vary. Bowerman (1996) pointed out that the relations that English divides into in versus on are encompassed by a single preposition, en, in Spanish, but are divided across three prepositions in Dutch (in, op, and aan, with the latter two separating the relations exemplified by an apple being on a table and a handle being on a door). Thus, dimensions of support, containment, and contact have different degrees of involvement in the distinctions of the three languages. Similar observations have been made in many

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

24

 . 

other domains. Cross-linguistic diversity has been established for colour, causality, mental states, number, locomotion, acts of cutting and breaking, acts of carrying and holding, and geographic features, among others (see Malt & Majid, 2013; Majid, 2015, for review of many of these findings). As already noted, the development of naming patterns unfolds over the course of a language’s history and is influenced by what entities speakers have needed to label at each step of the development, as well as what ones they did in the past (providing the foundation for the next step in naming choices). Across languages, then, with their differing cultural and linguistic histories, it is evident that such processes could result in extensive cross-linguistic differences. For instance, as Malt and Majid (2013) discuss, the complexity of commerce or cuisines in a culture may influence their numeration or taste naming systems. Different physical environments may also contribute since different plants, animals, and landscape features will be present, and linguistic influences, such as different word formation opportunities afforded by different morpho-syntactic structures, along with differing patterns of contact with other languages, can also have an impact. One should not leave this discussion with the impression that languages can vary endlessly without constraint. Observations of diversity are balanced by observations of commonalities. Although domains may not always have perceptually distinctive divisions within them, some domains do, or do so within parts of the domain. For instance, at the genus level, some groupings of plants and animals may be fairly obvious, and there are shared aspects of naming patterns for them across languages (Berlin, 1992; Hunn, 1977; see Malt, 1995 for review). Human gaits such as walking and running are separated by constellations of biomechanical properties, also resulting in some commonalities in naming patterns across languages; Malt et al., 2008). Shared perceptual and conceptual systems, as well as shared needs to communicate with systems that are relatively simple and learnable, also constrain the variability (see Malt & Majid, 2013 for further discussion). Still, against this background of commonalities, there is room for sufficient diversity to provide a challenge for communication across individuals and for language learners of all stripes. Section 1.3 explores these implications.

1.3 How is Complexity of Mappings Dealt with Within a Language? It might be thought that word learning is the quick and easy part of native language acquisition. The traditional view says that an initial fast mapping of form to meaning occurs on initial exposure to a word (Carey, 1978), and understanding the strategies babies use in deciding what mapping to make is the interesting aspect of word learning. What happens after that is just a repeated process such that the vocabulary grows larger.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

25

Despite this general perspective, it was known that for some word classes, learning takes an extended period of time. Spatial terms, deictics (e.g. this, that), and verbs were considered trickier because what the word might mean is harder to identify in a discourse context. Common nouns were considered easy and targets for fast mapping on the basis of arguments harkening back to the ideas that their referents are easily segregated from their context, that they label perceptually obvious groupings, and that they may be closely equivalent across languages (e.g. Gentner, 1982). But the assumptions about common nouns are called into question by the observations about the non-obvious and language-specific nature of word meanings. Because the world-to-word mapping is, in part, language-specific, children learning a native language cannot rely entirely on some language-free, pre-existing conceptualization of a domain to bootstrap their way into the word meanings of their language. To test whether children require an extended learning period to fully master even common nouns meanings, Ameel, Malt, and Storms (2008) asked children ages five to fourteen to name large sets of pictures of common household containers and dishware for preparing and serving foods. They compared the name choices to those of adults. Children showed a substantial evolution in the use of names for the objects across the ages five to fourteen; it was not until age fourteen that application of terms to objects closely matched that of the adults. Even when the younger children had the same words as adults in their vocabulary, they did not always apply them in adult-like ways. They overextended some and underextended others, with these deviations being greater at the earlier ages. From there, continual reshaping of the initial word use patterns occurred until a match with adult usage was achieved. Other studies demonstrate that children also take an extended period of time to acquire full adult-like meanings for a wide range of other domains—again, even after they start producing the relevant vocabulary for the domain. Domains recently explored include colour, time, number, emotion, and locomotion (Tillman, Wagner, Chu, Imai, Malt, Whiden, & Shatz, 2016). Entwined with the facts that lexical distinctions are often language-specific, and that this cross-language variation is enabled by lack of obvious objective groupings of entities in some domains, there is a third fact: even adult native speakers of a language often show disagreement on what to call something. Statements about what speakers of a language call something really mean only what the most commonly given name is. The level of agreement among adult native speakers for entities can vary widely depending on the specific instance being named. A Coke bottle, for instance, will usually be named bottle by 100 per cent of native English speakers sampled, but a medium-height, narrow-width container of olives may be called bottle by some and jar by others. Likewise, some drinking vessels have one strongly dominant name, but others may be split across speakers

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

26

 . 

between cup and mug. This variability probably contributes to the learning challenge for children (and probably makes name choice for entities with higher name agreement easier to master than name choice for ones with low agreement, as Zinzser, Malt, Ameel, and Li, 2014, have shown for second-language learners). The observation of variable naming even within an adult native speaker community raises the question of how individuals succeed in communicating among themselves. If a given label can be applied to a diverse and to some extent arbitrary set of things, how does an addressee know what a speaker means on a given occasion of use? The answer lies in considering the more social aspects of communication. As Clark (e.g. 1996, 2006; Clark & Brennan, 1991; Clark & Krych, 2004) has argued and empirically demonstrated, speakers and addressees work together to establish reference. Language use in its most basic form is face-to-face conversation. This permits active engagement and collaboration of the parties in the interaction. Participants can take into account common ground in choosing their words. That is, a speaker has some sense of what the addressee knows about and will assume a given word means—and if they know each other well, this sense can be quite specific. Along with using general background knowledge about who knows what, speakers will respect more localized conversational precedents. If one participant in an interaction has called some object or other entity by a particular name in the course of the conversation, the other participant is likely to reuse that same name, keeping them on the same page about what is being referred to. Pointing and eye gaze also help indicate the intended referent if more than one candidate is physically present. Participants also sense when their meaning has been understood or not through cues such as head nods or shakes, and backchannel signals such as ok and hmm? and can offer clarifications if needed. Of course, not all interactions are face to face, and even when they are, a speaker’s intended referent may not be physically present. Speakers (or writers) can also adjust the amount of detail that they provide to help an addressee understand their intended referent. Malt (2013) showed people sixty common household containers (the same ones used by Malt et al., 1999) and asked them to name each object under three circumstances. In the first, participants were just asked to say what they would call the object, without any further context specified. In the second, they saw each object among pictures of dissimilar objects such as a hairbrush and a key, and they were asked to indicate to the experimenter that they wanted the container by saying ‘Please hand me the . . . ’ and then giving the object name. In the third version, they were asked to imagine that they wanted their addressee to fetch that specific item from a closet full of all the other fifty-nine objects in the containers set. They were to say ‘Go get me the . . . ’ and then specify the object. The nouns they used for each object rarely changed across contexts, but the number of modifiers attached to the nouns did. In the first case, modifiers were rarely attached to the nouns. In the second case, an average of one

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

27

modifier was used per request, and in the third case, two to three modifiers were routinely used. So, the use of disambiguating modifiers is another indication that speakers infer what help an addressee may need. The modifiers narrow the range of entities to which a word can refer and reduce the potential for speakers and addressees to make different assumptions about what the word is intended to mean. In short, the variability in naming patterns across languages, as well as the variation in what a word can mean within a language, produce challenges for children learning language and for discourse participants. Children take a long time to converge on adult usage, and adults must work to establish mutual reference, even when using common, familiar words. These observations harken back to the points discussed earlier about how naming patterns of a language can only be explained by including cultural and linguistic history. Children’s learning is more than a straightforward mapping of pre-linguistic concepts onto words; it also requires an extended process of working out conventional adult usage patterns for their language environment. Successful communication in discourse requires engaging in social processes rather than being a simple matter of a given word reliably evoking a shared meaning.

1.4 How is Complexity of Mappings Dealt with Across Languages? Many people across the globe speak two or more languages, not just one. There is a special challenge posed by complexity in the world-to-words mapping for children exposed to two languages from birth and for speakers of one language acquiring a second. That is, bilinguals (whether early or later learners) receive input giving different ways of grouping the same objects by name depending on which language they are currently listening to. What are they to make of the conflicting inputs? People who grow up with one language and later begin to learn a second already have one well-established set of word meanings and ways of mapping the world into words. When exposed to a second language, they use the first meanings and mappings to bootstrap their acquisition of the second (e.g. Jiang, 2000), resulting in treating words of the two languages as more equivalent than they really are. Second-language instruction typically does little to help learners reshape their understanding of the words toward the target language. It even encourages the mistaken assumption of complete translation equivalence by teaching vocabulary as if mastery were simply a matter of memorizing second-language form equivalents to first-language words. These tendencies in teaching and learning raise the questions of whether there is an extended learning period needed to master

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

28

 . 

native-like word use (even when relevant vocabulary words are known), and whether mastery is eventually achieved. Malt and Sloman (2003) addressed these questions by looking at non-native speakers of English attending two universities in the US. These English learners came from a variety of different language backgrounds. They were shown sets of common household containers and dishware used for serving and preparing foods and were asked, for each one, what they would call it in English. They were also asked to judge, for each object, how typical it was of the things called by English names (such as bottle or jar.) Participants were separated according to how long they had been immersed in the English language environment. Their responses were evaluated by crediting each choice they made proportional to the percentage of native English-speaking American college students who gave that same response to the picture. Average scores increased with greater years of immersion, showing that learners did make progress toward the target language standard. However, even students with ten or more years of immersion in English still differed from the native English speakers to a significant extent. Their judgments of what objects were most typical of the different names did, too. These results indicate that there is a long trajectory for late second-language learners toward the native language standard for patterns of word use, and that their evolving knowledge may not become fully monolingual-like. Many international students spend part of their day with students from their native country, speaking their native language, so this fact leaves open the possibility that complete immersion in the second language would shorten the trajectory and result in a closer match to monolingual-like. At the same time, perhaps mastering second language patterns would alter use of the native language—that is, perhaps speakers cannot acquire and maintain two separate ways of mapping the world into words. Malt, Li, Pavlenko, Zhu, & Ameel (2015) tested native Mandarin speakers in the US, assessing both their match to native English speakers’ naming patterns and to monolingual speakers of Mandarin back in China. All of the participants had arrived in the US no earlier than age fourteen, having full command of Mandarin. Malt et al. did find evidence of a bi-directional language influence. The students’ use of English words was influenced by their native Mandarin, and their use of native Mandarin words showed an impact of their English learning, reflected in deviations from the monolingual speakers in China. The extent of mastery of English patterns did not predict the extent of deviation from Mandarin, however, suggesting that higher motivation or verbal ability may allow acquiring and maintaining native-like use in both languages. Although we’ve considered late learners so far, many bilinguals grow up with two or more languages and use them throughout their lifespan. Could this early and more intensive exposure result in a match to monolingual speakers of each language? For English speakers who came to the US from Russia at ages ranging

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

29

from early childhood to early adulthood, and still used Russian at home, Pavlenko and Malt (2011) found that earlier immersion in English predicted greater deviation in naming from native monolinguals speakers in Russia. Ameel et al. (2005) looked at the case of bilinguals who might have the greatest likelihood of matching monolinguals of each language—those who were raised in a bilingual country, Belgium, by parents who were native speakers of different languages (Dutch and French), with each speaking to the child in their own native tongue. Even in this case, there was an impact of the two languages on each other. Rather than acquiring two completely monolingual-like sets of naming patterns, the adult bilinguals used naming patterns in each language that were a partial compromise between the patterns of monolingual speakers of the two languages. So, it seems that a bilingual’s word knowledge is likely to show cross-language influence in one or both directions, depending on age of acquisition and contexts of use. As a result, when native and non-native speakers of a language interact—or when monolinguals of one language interact with bilinguals whose use of that language has been influenced by another—differences in word choice can arise. These differences may lead to occasional miscommunications or at least short delays in comprehension, as in some observed instances of utterances by nonnative to native English speakers: Put the dishes in the closet; Do you have a chair on the plane?; Buy some paper glasses for the drinks; Get a bonnet for the swimming pool. Of course, we’ve already noted that native monolingual speakers of the same language sometimes also differ among themselves in word choice. Across dialects of the same language, differences can be equally or more acute. Native speakers of American vs British English, for instance, would interpret the nouns in these sentences differently: Put on a jumper; Lift the bonnet; Get some trainers; Serve the biscuits. In such cases, crossing languages or dialects, the same social processes can operate to achieve understanding that work for native speakers sharing a dialect: Taking into account common ground, adhering to conversational precedents, and using feedback, repair, pointing, eye gaze, and disambiguating modifiers to help convey the intended message. People who regularly communicate with speakers from a particular different dialect group or first language background may also learn to adjust their interpretation of some words and expressions when interacting with them. The benefits of bilingualism and communication across individuals from different language and cultural backgrounds clearly outweigh any disadvantages of the need for a bit of extra effort in understanding each other. Globalization brings more and more contact between individuals coming from different native language backgrounds or different combinations of bilingual backgrounds, raising the interesting question of whether mutual influences will eventually reduce differences. This issue has been considered most extensively in the case of English, which has become a global lingua franca. It is said that there are now more second-language speakers of English in the world than there are

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

30

 . 

native speakers (Crystal, 2003). The widespread use of English raises legitimate fears of losing other, less-used languages (Crystal, 2003). Whether this use will cause English to change is a separate issue. Schneider (2011) notes that English has become localized and indigenized in many different countries. Given that each region tends to develop its own distinctive dialect, the widespread adoption of English doesn’t, by itself, mean that English will become homogenized. A more likely source of impact would come from interactions in English between speakers of different first language backgrounds. It has sometimes been argued that when people acquire a language as a second language, they do so imperfectly and so some simplification is entailed—such as loss of inflectional morphology or reduction in the number of phonemes (e.g. McWhorter, 2011; Trudgill, 2009). The evidence on this point is debatable, however (e.g. Thomason, 2011). Even if true, it is not clear whether second-language users of English, when coming to live in predominantly English-speaking countries, will change the local variant. MacKenzie (2016) suggests that such speakers are more likely to accommodate to local speakers than vice versa, and their children will acquire the dialect of the community they grow up in. These outcomes will occur because the non-native variant is less likely to be perceived as high-status and worthy of imitation. Also working against change is the fact that (locally) standardized native varieties of English are dominant in education and the media. In this light, although there is room for debate about the future trajectory of English, it is likely that intercultural communication in English will continue to involve some differences in word usage patterns. As already noted, engaging social processes to achieve understanding is part of normal language interactions even among people from the same language and dialect background. Relevant processes for resolving such differences are part of every language user’s repertoire of language skills.

References Ameel, E., Malt, B. C., and Storms, G. (2008). Object naming and later lexical development: From baby bottle to beer bottle. Journal of Memory and Language, 58, 262–85. Ameel, E., Storms, G., Malt, B. C., and Sloman, S. A. (2005). How bilinguals solve the naming problem. Journal of Memory and Language, 53, 60–80. Berlin, B. (1992). Ethnobiological classification: Principles of categorization of plants and animals in traditional societies. Princeton, NJ: Princeton University Press. Bowerman, M. (1996). Learning how to structure space for language: A crosslinguistic perspective. In P. Bloom, M. A. Peterson, L. Nadel, and M. F. Garrett (Eds.), Language and space (pp. 385–436). Cambridge, MA: MIT Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

31

Carey, S. (1978). The child as word learner. In M. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality (pp. 264–93). Cambridge, MA: MIT Press. Clark, H. H. (1996). Using language. Cambridge: Cambridge University Press. Clark, H. H. (2006). Social actions, social commitments. In S. C. Levinson & N. J. Enfield (Eds.), Roots of human sociality: Culture, cognition, and human interaction (pp. 126–50). Oxford: Berg Press. Clark, H. H., & Brennan, S. E. (1991). Grounding in communication. In L. B. Resnick, J. M. Levine & S. D. Teasley (Eds.), Perspectives on socially shared cognition (pp. 127–49). Washington, DC: American Psychological Association. Clark, H. H., & Krych, M. A. (2004). Speaking while monitoring addressees for understanding. Journal of Memory and Language, 50, 62–81. Crystal, D. (2003). English as a global language (2nd edn). Cambridge: Cambridge University Press. Gentner, D. (1982). Why nouns are learned before verbs: Linguistic relativity versus natural partitioning. In S. Kuczaj (Ed.), Language development, Vol. 2: Language, thought and culture (pp. 301–34). Hillsdale, NJ: Lawrence Erlbaum. Hunn, E. (1977). Tzeltal folk zoology: The classification of discontinuities in nature. New York: Academic Press. Invention of the telephone (n.d.). Wikipedia. Retrieved 3 October 2016 from https:// en.wikipedia.org/wiki/Invention_of_the_telephone Jiang, N. (2000). Lexical representation and development in a second language. Applied Linguistics, 21, 47–77. Kronenfeld, D. B., Armstrong, J. D., and Wilmoth, S. (1985). Exploring the internal structure of linguistic categories: An extensionist semantic view. In J. W. D. Dougherty (Ed.). Directions in Cognitive Anthropology (pp. 91–113). Urbana, IL: University of Illinois Press. Lakoff, G. (1990). Women, fire, and dangerous things: What categories reveal about the mind. Chicago: University of Chicago press. Live Science Staff (2012). What’s the difference between a fruit and a vegetable? http:// www.livescience.com/33991-difference-fruits-vegetables.html MacKenzie, I. (2016). Will English as a lingua franca impact on native English? In C. Sanchez-Stockhammer (Ed.). VARIENG: Studies in Variation, Contacts, Change, and in English (vol. 16: Can we predict linguistic change?). Retrieved from http:// www.helsinki.fi/varieng/series/volumes/16/mackenzie/ Majid, A. (2015). Comparing lexicons cross-linguistically. In J. R. Taylor (Ed.), The Oxford Handbook of the Word (pp. 364–79). Oxford: Oxford University Press. doi:10.1093/oxfordhb/9780199641604.013.020. Majid, A., Enfield, N. J., & van Staden, M. (2006). Parts of the body: Cross-linguistic categorisation. Special issue of Language Sciences, 28(2–3), 137–360.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

32

 . 

Malt, B. C. (1995). Category coherence in cross-cultural perspective. Cognitive Psychology, 29, 85–148. Malt, B. C. (2013). Context sensitivity and insensitivity in object naming. Language and Cognition, 5, 81–97. Malt, B. C. (2010). Naming artifacts: Patterns and processes. In B. Ross (Ed.), The psychology of learning and motivation: Advances in research and theory (pp. 1–38). Amsterdam: Elsevier. Malt, B. C. and Majid, A. (2013). How thought is mapped into words. WIREs Cognitive Science, 4, 583–97. Malt, B. C., Gennari, S., Imai, M., Ameel, E., Tsuda, N., & Majid, A. (2008). Talking about walking: Biomechanics and the language of locomotion. Psychological Science, 19, 232–40. Malt, B. C., Li, P., Pavlenko, A., Zhu, H., & Ameel, E. (2015). Bidirectional lexical interaction in late immersed Mandarin–English bilinguals. Journal of Memory and Language, 82, 86–104. Malt, B. C. and Sloman, S. A. (2003). Linguistic diversity and object naming by nonnative speakers of English. Bilingualism: Language and Cognition, 6, 47–67. Malt, B. C., Sloman, S. A., Gennari, S., Shi, M., and Wang, Y. (1999). Knowing versus naming: Similarity and the linguistic categorization of artifacts. Journal of Memory and Language, 40, 230–62. McWhorter, J. H. (2011). Linguistic simplicity and complexity: Why do languages undress? Berlin: de Gruyter Mouton. Pavlenko, A., & Malt, B. C. (2011). Kitchen Russian: Cross-linguistic differences and first-language object naming by Russian-English bilinguals. Bilingualism: Language and Cognition, 14, 19–45. Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382–439. Schneider, E. W. (2011). English around the world: An introduction. Cambridge: Cambridge University Press. Spalding, T. L. & Murphy, G. L. (1996). Effects of background knowledge on category construction. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 523–38. Tillman, K., Wagner, K., Chu, J., Imai, M., Malt, B. C., Widen, S., & Shatz, M. (2016). Beyond the language explosion: What gradual word learning tells us about conceptual development. In A. Papafragou, D. Grodner, D. Mirman, & J. C. Trueswell (Eds.), Proceedings of the 38th annual conference of the cognitive science society (pp. 49–50). Austin, TX: Cognitive Science Society. Thomason, S. (2011). Does language contact simplify grammar? Working paper [available from author]. Trudgill, P. (2009). Sociolinguistic typology and complexification. In G. Sampson, D. Gil, and P. Trudgill (Eds.), Language complexity as an evolving variable (pp. 98–109). Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

33

Wilson, D. E. & Reeder, D. A. (Eds.). (2005). Mammal species of the world. A taxonomic and geographic reference (3rd edn). Baltimore: Johns Hopkins University Press. Zinszer, B. D., Malt, B. C., Ameel, E, & Li, P. (2014). Native-likeness in second language lexical categorization reflects individual language history and linguistic community norms. Frontiers in Psychology: Language Sciences, 5, 1–16. doi: 10.3389/fpsyg.2014.01203

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

2 How to Make Psychological Generalizations When Concepts Differ A Case Study of Conceptual Development Gregory L. Murphy

An important difference between many philosophical and psychological views of concepts has to do with whether concepts are universal or in some way personal, related to the question of whether they are external or internal. Is the concept of red something that can be defined so that when we say ‘John has the concept of red,’ we all know just what it is that John knows? Philosophical tradition tends to say that the answer to this is ‘yes,’ because red is red, and it is a natural or conventional fact as to what constitutes a concept. If we instead say that John’s concept of red is his own personal representation of the concept, then it is possible, and indeed likely, that his concept will not correspond to Jessica’s concept of red. Questions about whether people have the ‘same concept’ become a graduated question of similarity. In psychology, the study of concepts has focused almost entirely on the individual, internal level. My concept of ‘red’ or ‘dog’ can be measured by a variety of empirical means, such as classification or typicality judgments (e.g. Hampton, 1995; Rosch, 1973). We can measure other people’s concepts the same way. When we do so, we will generally find that two arbitrarily selected people do not have the exact same concept. I say that a light reddish hue is pink, whereas you say it is red, or rose; I think a tomato probably isn’t a fruit, whereas you think it is; I say that a German shepherd is only a moderately typical dog, whereas you say that it is very typical. With its interest in psychological representations, most of the psychology of concepts has simply said little about any way of specifying concepts apart from individuals’ judgments. Indeed, an empirical tradition has arisen to investigate just how concepts differ across people and situations (Barsalou, 1987). In philosophy, the approach is probably less monolithic, but there is nonetheless a strong tendency to identify concepts as being a definable set, with an extension and intension (reviewed by Laurence & Margolis, 1999). This was revealed in early responses to the work of Eleanor Rosch and other psychologists documenting the fuzziness of categories. A number of philosophers criticized this work, suggesting that psychologists simply didn’t know what a concept really was Gregory L. Murphy, How to Make Psychological Generalizations When Concepts Differ: A Case Study of Conceptual Development In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Gregory L. Murphy. DOI: 10.1093/oso/9780198803331.003.0003

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

35

(e.g. Rey, 1983; Sutcliffe, 1993). Perhaps the most vocal critic has been Jerry Fodor, whose reaction was succinctly described in the title of his book, Concepts: Where Cognitive Science Went Wrong (Fodor, 1998). Central issues in this debate surround communication and arguments. If people have different concepts, then isn’t trying to discuss things and come to an agreement pointless? In particular, isn’t philosophical analysis pointless? For example, if we debate the nature of the good or morality, it may all boil down to our not having the same thing in mind when we talk about ‘the good.’ If concepts are mental representations, we won’t even have a simple way of knowing whether we’ve got the same concepts. Even mundane communication will become problematic, according to Fodor (1998, p. 29): If everybody else’s concept [of] water is different from mine, then it is literally true that only I have ever wanted a drink of water, and that the intentional generalization. ‘Thirsty people drink water’ applies only to me.

That would clearly be a bad outcome. Fodor concludes that the concept of water must be externally defined in order for us to communicate with such statements. That is, he assumes that in order for people to communicate and coordinate behaviours, we must share the same concepts (emphasis on same). I have argued elsewhere that the argument of shareability is overly strong (Murphy, in preparation). Communication is in fact possible even in cases where it is obvious that we do not have identical concepts. Perhaps the most obvious example is differences between the concepts of children and adults, which is also the focus of this chapter. Consider the following interchange: : I want the horsie. [Points towards plush elephant] : You want the horsie? Here. Is it a nice horsie? : Soft horsie. : The horsie is soft! That’s a funny-looking nose on the horsie. This could go on for some time, so we will leave the tender scene now. I assume that Dad has a normal concept of horse, though it might not correspond to that of a rancher or an equestrian or a judge at a horse show. The child’s concept seems to deviate significantly from the adult norm, however. It is possible that she does have a near-adult concept of horse but is using the word incorrectly for local pragmatic reasons, e.g. not knowing the name elephant. However, much research on conceptual development (which I review below) suggests that young children do not know the correct extensions of common concepts and must learn them from observing adult word usage and from receiving feedback on their own labelling. Indeed, it is difficult to imagine how a young child would immediately come to the correct concept of horse, distinguishing horses from cows, goats,

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

36

 . 

donkeys, burros, zebras, giraffes, deer, wildebeest, and so on. Without sufficient experience in forming those concepts, the child would have no way of knowing where ‘horse’ begins, and where it ends. As a result, during the learning process, she must go through a series of stages in the development of ‘horse’. Perhaps the first stage is one in which all large quadrupeds are included. Then perhaps large zoo animals are excluded. One by one, goats, cows, zebras, and donkeys are learned and thereby excluded from the horse category. In this horse example, the main problem for the child is to exclude items that don’t belong in the category. In other examples, an equally difficult problem may be to know to include items that are not obviously in the category, e.g. that gnats are animals or that avocados are a fruit. During this process, the child does not lack an ability to communicate about horses and other animals. Most of the time, two interlocutors with different concepts figure out what the other is talking about, as constraints from the referent itself (the child clearly is pointing at the toy elephant), knowledge of personal conventions (she always calls cows horsie), and intelligent inference result in communication. Dad doesn’t actually have to know the details of his daughter’s concept of horse in order to understand what she means when she uses the word in many contexts. Even if her concept deviates significantly, her word uses may be correct in the most common and typical situations, leading to no problem in communication. In short, although I have not worked through the argument in detail, I suggest that there is reason to doubt the need for perfectly matching concepts in order for communication to work. Furthermore, as I discuss elsewhere (Murphy, in preparation), communication often is imperfect and requires negotiation between interlocutors over terms and assumptions. That is understandable if concepts are mental representations, but not so readily understandable if they are externally defined entities. However, that topic would take us too far afield from the main one of considering the relation between child and adult concepts (although I return to it briefly at the end).

2.1 Development and Concepts Although I’m a proponent of the psychological approach to concepts, I agree that it causes a conundrum with regard to development. The problem is that according to the story I have been telling, and that is told by most developmental psychologists who study concepts, children’s concepts are constantly changing. I don’t just mean that they continually add to their store of concepts, but that initially acquired concepts undergo content changes over development. Do children then have the same concept over time? The child’s concept of horse may start out as extremely broad. After further experience, its extension has changed to

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

37

exclude, say, elephants. After learning yet more, it has changed to exclude goats and also to include ponies. This process goes on for years. It is not crazy to say that the child does not have ‘a concept of horse’ but rather a hundred concepts, or maybe more, changing with each experience of using the concept. When should one state that a child has a acquired a concept or ‘knows the word’ that picks out a category? (Vocabulary acquisition is a bit easier to see and so is often what is used to measure category knowledge. I’m not going to be fussy in distinguishing concepts from word meanings here—see Murphy, 2002, Ch. 11.) There is little discussion of this issue in the field. An early writer proposed that a ‘reasonable criterion of acquisition [of category names] is simply that they can be used to denote at least some objects appropriately’ (Anglin, 1977, p. 33). This very weak criterion seems to be what most writers use. If a child correctly names a prototypical chair as chair, he or she gets credit for knowing the word, without much concern over whether the child correctly labels reclining chairs or bean-bag chairs or mistakenly calls a bench or a stool chair. Paul Bloom (2001, p. 1129) explicitly rejected the idea that refining the extension of a word is part of the psychology of word learning as it is usually understood: ‘The perspective adopted in [his book] is that children learn “bird” when they have a rough idea of what the word refers to.’ Thus, concepts and word meanings must change greatly over time, if their initial guesses and ‘rough ideas’ are sufficient for us to give them credit for knowing them. As an aside, these authors’ definitions of word learning both focus on extensions—whether children correctly label category exemplars. Most of the evidence I refer to in this chapter is also about extensions. This is the most straightforward evidence of conceptual difference, as any view of concepts should agree that if I consistently label different things as bird than my daughter does, we probably do not have the same concept of birds. However, as discussed by James Hampton in his chapter in this volume, one can also measure the intension of a concept, such as the properties that people associate to it. Although measures of intensions and extensions may not always agree [as Hampton shows], my default assumption is that differences in extension must be matched by differences in intensions that direct people’s labelling behaviour. However, measuring intensions in young children is problematic, as they are not very good at introspection and explaining their mental processes. Thus, differences in feature listing and the like would be ambiguous as to whether the children have different concepts or merely less metacognitive ability to describe their concepts. As a result, labelling behaviour seems a more reliable measure of conceptual difference. Answering the question of when people have the ‘same concept’ is knotty, and possibly not very necessary for the practice of the psychology of concepts. However, there is another conundrum that arises with this issue—namely, how to state psychological generalizations about concepts given the changes that occur

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

38

 . 

over development. This directly impinges on psychological theory and so is necessary to address.

2.2 Levels of Categorization The issue of conceptual variation has implications for another psychological question that is perhaps more substantive than the question of whether two people have the ‘same concept’—the issue of preferred levels of categorization. In a seminal paper, Eleanor Rosch and colleagues (Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976) pointed out that a given object can be classified into categories¹ ranging in specificity from highly abstract to extremely specific. For example, your pet might be an object, a living creature, a vertebrate, a mammal, a dog, a terrier, and a rat terrier. It might further fit into somewhat ad hoc categories such as a terrier born and raised on Rat Terrier Farm. Rosch and colleagues argued that one category in this hierarchy of abstraction was basic, in this case, dog. Basic categories are the ones we normally use to identify, think about, and talk about familiar objects. People could look at your pet and say, ‘Please get your rat terrier off my lawn,’ or ‘That is an attractive mammal,’ but they’re much likelier to talk about it as a dog. How does this relate to the developmental issues discussed above? There is a claim that children learn basic categories first and that they prefer to use them over more abstract and more specific categories, just as adults do. However, the problem is that the categories children have do not map exactly on to the categories that adults have, as just discussed. The child’s notion of mammals (if it has any) is very different from mine; she might not include as many dogs in her dog category as I do; she might not know what terriers are, or only have a very imperfect notion of them (e.g. some kind of dog). So, even if the child prefers to call these things dogs, does it follow that the child has a preference for basic-level categories over more general and specific ones, given that those categories are generally defined in terms of adult conceptual knowledge? To answer this question, I first summarize the theory and main phenomena involving basic-level concepts. Then I describe some of the developmental evidence that has led researchers to conclude that children also show a basic-level advantage (as it is termed). Finally, I address the issue of differences between adult and child concepts and discuss how we can decide whether children have a ¹ Readers may notice that I am shifting from primarily talking about concepts to talking about categories. This is in large part because the levels of categorizations literature is usually phrased about categories (as in the name levels of categorization). Obviously, people’s concepts of these categories are what we are ultimately interested in, but it is extremely unwieldy to have to write ‘one’s concept of the category of fish’ and the like, so I will leave it to the reader to understand that when I talk about a category, I am usually saying something about the concept people have of it as well.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

39

preference for basic-level concepts, even if their basic concepts are different from those of adults.

2.3 Basic-level Concepts Because of space constraints, I am going to simply assert some of the properties of basic concepts, usually without providing specific citations for the empirical evidence for them. Anything I state here has strong empirical support. I reviewed the topic extensively in Murphy (2002, Ch. 6), which may be consulted for a more detailed presentation and references. Little of substance has changed since then. Informally, basic categories seem to be a compromise between categories that are too abstract and those that are too specific. Rosch et al. (1976) showed that members of basic categories tend to have many properties in common. Dogs generally breathe, have four legs, bark, have ears, have fur, eat meat, and so on. In contrast, members of more abstract superordinate concepts like mammals tend to have relatively few features in common. Thus, basic concepts provide considerably more information. Why, then, aren’t subordinate concepts like terrier or rat terrier even better, given that they convey even more information? Rat terriers apparently have short hair, often have a piebald colouring, have a typical weight range, and so on, in addition to the usual dog features. So knowing that something is a rat terrier would carry even more information than knowing that it is a dog. The problem with more specific concepts is that they are usually very similar to one another and so require fine discriminations. A rat terrier is extremely similar to other short-haired terriers. The features of terriers overlap greatly with the features of poodles, beagles, or Labrador retrievers. To identify which kind of dog Fido is requires identifying particular distinctive features like exact size, colouring pattern, hair texture, and head shape. However, identifying Fido as a dog is easy, because dogs are so different from cows, fish, birds, or hedgehogs. As a result, classifying objects at the basic level is generally faster than classifying them at the subordinate level. Classifying something as a mammal, i.e. the superordinate level, is much slower than either (Murphy & Brownell, 1985; Rosch et al., 1976), due to the small number of features common to mammals and their abstractness. The basic level is ‘preferred’ in two senses. First, as noted earlier, it is the name that people generally give in a neutral context. When you see an object in an office, you’re more likely to call it ‘a chair’ than ‘a piece of furniture’ or ‘swivel chair.’ This difference in labelling is often thought to be paralleled by a difference in thought. That is, even absent any verbal response, when you see a tall plant in the park, you usually think of it as a tree, and not as a sugar maple. Second, basic level concepts are preferred in terms of performance on conceptual tasks. They are learned first in childhood (discussed in more detail later), they are learned more easily in

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

40

 . 

artificial category-learning experiments, they are faster in classification, and their names are often shorter and more distinctive. In some cultures, high-level superordinates such as animals or furniture do not even have a single-word name. The discovery of basic-level concepts is an important part of the psychology of concepts, which is generally accepted throughout the field. The theory behind it is also widely, though not universally, accepted—that basic concepts are high both in the amount of information they carry and in terms of their distinctiveness from other categories. These properties are exactly what makes identification easy—and useful.

2.4 Basic-level Categories and Development For just those reasons, it isn’t very surprising that children should learn basic categories first. Indeed, some of the first inklings of the existence of the basic level were given by Roger Brown (1958; and investigated more extensively by Anglin, 1977) studying language acquisition. He pointed out that parents label objects at a consistent intermediate level, which he believed probably aids word learning. Much research shows that children learn basic names first and that they prefer to label objects at the basic level. Parents talk to their children using basic names. More focused tests show that children can classify things together at the basic level from an early age (say, two years of age, though it depends on the task and categories), but take rather longer to group things at the superordinate level. For example, Rosch et al. (1976, Experiment 9) asked children to group together pictures into those that ‘are the same kind of thing.’ All but one kindergartner and one first grader sorted items correctly into basic-level categories like trains or shoes. Only half of the children at each grade could sort into superordinate categories like vehicles or clothing. There are dozens of studies showing that children are more accurate in naming, classifying, or grouping items at the basic level than at higher or more specific levels. I reviewed these recently (Murphy, 2016) and concluded that the basic-level advantage in children’s concepts is unassailable. Surprisingly, however, there is a completely different research tradition from the infancy literature that proposes that young infants are able to learn superordinates as early as basic categories, or even earlier (Quinn & Johnson, 1997, 2000; Rogers & McClelland, 2004; see Younger, 2010, for a review). There are many differences between the infant and child experiments that cannot be discussed in detail here (see Murphy, 2016). The most important one is that in the infant tasks, one is assumed to be teaching the baby a concept and then testing what it has learned. The tasks of most experiments with older children sample children’s knowledge of categories that they know when they enter the experiment.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

41

Because the infant and child literatures are to some degree separate, their conflicting results have generally not received much attention. However, they pose a severe problem to our theory of conceptual development. How can tiny two-month-old babies learn superordinate categories that four-year-old children are not able to handle and that adults are slower to process? The problems with superordinates—namely, that they carry very little information and are often based on abstract features—should surely pose problems for infants, too. The answer I arrived at in addressing this conundrum (Murphy, 2016) was that infants were not actually learning superordinate categories. In an important study, Behl-Chadha (1996) taught infants the concept of mammals by exposing them to pictures of mammals from different basic-level categories, that is to say, cats, deer, dogs, elephants, giraffes, hippopotamuses, horses, rabbits, squirrels, tigers, and zebras. Although these are a diverse group of animals, they are all four-legged land mammals and were usually presented in profile. The size and texture differences among them were not salient (all animals were depicted at about the same size). Missing from this list was humans, aquatic mammals, flying mammals, or nonhuman primates. Based on this experience the infants would have no way of including mammals that don’t fit this profile or excluding non-mammals that do, e.g. including porpoises and bats but excluding sharks and birds, as older children do. The infants would not think to include humans, as an adult with a fully formed concept of mammals would readily do. So, the infants did learn something impressive, but it was not really mammals—it was ‘mammals lite’—the easy mammals. But what level is the mammal lite category? The empirical question of what categories children learn first now collides with the philosophical question of how to identify concepts across people and ages. Is the infant’s (over)simplified version of the mammal concept close enough to mammal for us to be able to say that infants can learn superordinates first? Furthermore, the same issue arises, if not as obviously, with children’s basic concepts. Anglin (1977, Ch. 4) systematically investigated the labelling accuracy of 2.5 to four-year-old children for categories at different levels. At the superordinate level, children made large numbers of underextensions, i.e. withholding the label from an actual category member. For example, they underextended twenty-four per cent of animal pictures and thirty-seven per cent of fruit. However, basic categorization was not completely correct either. Children made about three to four per cent errors on dogs, but almost twenty per cent overextensions on flowers and twelve per cent overextensions on apples. (They made many more errors on subordinates tulip and collie.) So, this question is not simply hypothetical. If children make a lot of errors on superordinate concepts and some errors on basic concepts, how do we know that their concepts are truly superordinate and basic? Of course, if we wished to be picky, the same question could be raised about adult concepts. Indeed, Anglin (1977) also tested college students and found that they made at least five per cent

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

42

 . 

errors on collie, tulip, and fruit. However, differences among people who actually know the concept (i.e. ignoring people who have no idea what a tulip is other than a kind of flower) are likely to be slight and idiosyncratic, not changing the overall category structure. If I know about electric standing chairs and you don’t, but you know about Swedish tilting chairs and I don’t, the effect on our concepts of chairs is minimal. We agree on the features usually found in chairs and usually not found in sofas or tables. Indeed, it’s very unlikely that we’ll ever notice our differences. In contrast, the difference between infants and adults is more like my thinking that only upright, high-backed, armless wooden chairs are chairs, whereas you also think that armchairs, reclining chairs, rocking chairs, etc., are also chairs. At that point, the overall similarity profiles of our categories differ so much that it is reasonable to say that we have different concepts and also that they might be at different levels. Therefore, I suggest that we need to consider whether psychological generalizations about concepts and their development are undercut by this conceptual variation. In particular, I ask whether we should say that the first concepts are basic or superordinate, given that the first concepts don’t correspond to adult basic or superordinate concepts. (No one seems to think that the first concepts might be subordinate, so I don’t address that possibility.)

2.5 Developmental Changes in Category Levels Let’s agree for sake of argument that children’s early categories don’t correspond to those of most adults. Carolyn Mervis (1987) extensively tested her son’s extension of the word duck, which she argued, like most category labels, included some items that adults wouldn’t include and excluded some items that adults would have included. (Part of her test sample indicated representations of ducks, including toys and cartoons, but it also included photographs and real animals like swans.) These differences arise through lack of experience, lack of feedback (e.g. not yet being told that a swan is not a duck), and lack of theoretical knowledge that bears on the category. As an example of the latter, Mervis says that children don’t understand that the slot in a piggy bank is an important feature, because they don’t understand the purpose of banks to hold money, nor the role of the slot in achieving that purpose. Eventually they will learn that a solid pig statuette cannot be a bank. Anglin (1977) also documents that children make errors of both underextension and overextension. Perhaps surprisingly, underextensions are more common. What is the source of these errors? Within basic-level names, much research suggests that children tend to label typical members prior to atypical ones. Using a looking preference paradigm, Meints, Plunkett, and Harris (1999) found that twelve-month-old infants would look at a named typical category member

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

43

(e.g. a sedan when hearing ‘Look at the car.’) but not at an atypical member (e.g. a racing car). Anglin (1977) used more explicit categorization judgments, with twoto six-year-olds. Two of the tested categories were superordinates, but one was a basic category, bird.² Children extended the label to virtually all typical birds, even unfamiliar ones. However, they refused to extend it to thirty-two per cent of the atypical ones, even familiar ones like hen and duck. (Similar results were found for superordinates, only children were less accurate overall.) Rosch (1973) tested nineto eleven-year-old children on verbal categorization judgments such as ‘A duck is a bird’. She found that children were generally correct for typical category members (six per cent errors) but made twenty-three per cent errors for atypical ones. Errors in categorization can arise from unfamiliarity with infrequent contrast categories. American children may categorize a wolf as a dog or label tigers as cats, given that they do not have daily experience with wolves or tigers, and so have little experience in distinguishing them from their similar, more frequent contrast categories. As I’ve commented elsewhere (Murphy, 2002), domestic cats are a basic category in part because we don’t have to distinguish them from bobcats, pumas, tigers, fishing cats, mountain lions, and other felines. If the full population of felines were common inhabitants of suburban streets, then the basic-level concept would probably be feline rather than cat. But a young child may be in a similar situation for many categories, as its local environment probably has restricted diversity. Perhaps dog for a young child is a simple category that refers to mid-sized domestic dogs in the neighbourhood or at home, and extending it to mini-Chihuahuas, greyhounds, and Persian mastiffs has never been necessary. Thus, even if the child does not noticeably mislabel animals, its limited experience is likely to result in a narrower concept than would be found in adults. Experiments such as Anglin’s are necessary to reveal the concept’s underextensions, which do not reveal themselves within the child’s environment. In short, children’s concepts are not the same as adults’. Furthermore, they differ in systematic ways, omitting peripheral members and including some similar items that will later be excluded. In infant category-learning studies, babies who apparently learned superordinate categories like furniture or mammal probably learned simpler versions of subsets of those categories, namely, items that shared certain perceptual properties. Indeed, it isn’t really possible for infants to learn the category of mammals, given their lack of knowledge of the relevant features (gives birth to live young? four-chambered heart?). The problem this creates, then, is what the justification is for saying that children learn basic categories first. If the three-year-old’s category of dogs is not my category of dogs, then is it, in fact, basic? It’s not the basic category that adults in our culture know. Even though adults may differ to some degree in their familiar concepts, the ² Anglin’s work was done before the notion of basic-level categories was published, so I divide up his materials into the appropriate levels myself when reporting them.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

44

 . 

differences between concepts of adults and young children are much greater, and the deviations may be predictable and systematic rather than idiosyncratic gaps common with adults. To give a theoretically coherent answer to the question of what level of categories is usually acquired first, then, we cannot rely on the identity of specific concepts that are acquired early in life using the loose criterion voiced by Anglin or Bloom (correctly naming some members). For example, we cannot argue that children learn about chairs before furniture, and chairs are a basic category, therefore children learn basic categories first. That argument relies on chair’s and furniture’s status with adults, and we have just admitted that there are large differences between these and children’s categories. Instead, we must look at the child conceptual structures and see if they follow similar principles as those of adult basic concepts (as Mervis, 1987, observes). This analysis is easiest to do with category-learning experiments like BehlChadha’s (1996), because we know exactly what the stimuli are. As noted earlier, Behl-Chadha used predominantly (but not exclusively) four-legged mammals presented in a side view, all at about the same size. These pictures therefore shared an overall shape and a number of specific properties: a body with a roundish head arising to the left or right side, and legs descending from it. Facial features were also usually present. One of the properties of adult basic-level concepts is that they tend to share a common shape (Tversky & Hemenway, 1984). Not all, but many chairs have a similar profile. Cars have a generic long, rectangular shape with wheels underneath and outlines of windows and doors. Furthermore, Tversky and Hemenway (1984) argued that common shape arises from commonalities of parts. The mammals in the infant experiment also had common parts, including four legs, head, eyes, mouth, a torso. The result of omitting the more ‘difficult’ mammals from Behl-Chadha’s (1996) experiment (the aquatic mammals, primates, bats) was to make the learned category more like a basic category. If seals, whales, and porpoises had been included, this would have diluted the predominance of the legs, the higher head, and configuration of the facial features. By omitting them, the presented objects’ features and their resulting shapes became more consistent, just as in basic categories. The effect of omitting bats was probably even greater. Thus, it seems that the finding that infants can learn such categories in this experimental paradigm does not undermine the argument that basic categories are learned first. The changes that Behl-Chadha made to the categories—quite understandable changes from the perspective of making them learnable—resulted in them approaching basic categories in their structure. To conclude that infants learned a superordinate concept, one would have to test infants on items that are perceptually diverse, just as adults can classify bats and elephants as both mammals or a tricycle and a locomotive as both vehicles. This simplification is generally true of

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

45

other prominent experiments showing superordinate category learning in infants (e.g. Quinn & Johnson, 1997; see Murphy, 2016). What about the more normal case in which children acquire apparently basiclevel concepts first? Given that those are not the same as adult basic concepts, can we state what their level is? Again, children differ a great deal in exactly what they have learned, based on their experiences. If Rachel grows up in a New York City neighbourhood filled with Chihuahuas and pugs, her concept of dogs may differ significantly from that of Jessica, who grew up in a suburb filled with Labradors, huskies, and sheepdogs. However, the data of Rosch, Anglin, and others all suggest that it is atypical items that are most likely to be omitted from their categories, probably due both to environmental and cognitive factors. That is, St. Bernards and penguins are not very thick on the ground and so take longer to be experienced. Furthermore, a category member that is very different from the rest of the category may simply be excluded from the child’s concept even if it is experienced. Controlled learning studies show that atypical items are usually acquired after typical ones, even when their frequency is equated (e.g. Rosch & Mervis, 1975). Atypical items don’t share as many properties with the other category members, so the conceptual representation doesn’t match them as well as it does typical ones. I mentioned earlier that basic categories are believed to be more informative and also more distinctive from other categories. If children’s concepts differ from adults’, how does that affect those two factors? If they don’t call all birds bird but exclude penguins, ostriches, kiwis, etc., the remaining birds are more coherent and similar to one another than in the full category. Children have actually increased the number of common category features by eliminating examples that contradict those generalizations. So, if you think of birds as flying, the existence of kiwis and penguins weaken this generalization; if you don’t include them in the category, then in fact almost all birds do fly. If you think of birds as generally being small, then ostriches throw this out of whack. Banish ostriches from the category, and your size generalization is back in whack! The distinctiveness component of category structure is harder to specify with children’s simpler categories. For example, perhaps children must distinguish birds from ostriches, because they don’t think of ostriches as birds. Adults don’t have to do that. However, this seems a very minor issue, given that children apparently see little resemblance between ostriches and robins to begin with— that’s why they don’t want to call ostriches birds. (Indeed, they may not call them anything at all, which removes the need for distinguishing them from birds.) It is only adults, with their (often) sophisticated knowledge, who look at the beak, the feathers, and what passes for wings on ostriches to see a similarity with robins: Yes, they’re quite similar, if you ignore their size, shape, heads, neck, feet, and behaviours. So, from the child’s perspective, the exclusion of ostriches from the bird category probably increases within-category similarity and causes no decrease in distinctiveness. Furthermore, if atypical items are experienced less

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

46

 . 

often (definitely true for penguins and ostriches), then in practice, the child doesn’t have to distinguish them from the category containing typical items. If basic-level categories in general have more features in common and are also more distinctive than categories at other levels, the child-basic categories seem likely to be ‘more basic’ than the adult ones, or at least no less basic. That is, the generalization we made about basic concepts being formed first, seems to be true even with children’s different structure. Perhaps this analysis suggests a better way of forming the generalization. Rather than saying that basic-level categories are the first ones learned, it might be more accurate to say that children’s first categories fit the description of basic-level categories. That is, rather than saying that children learn what we think of as basic-level categories first, whatever categories they learn first have the psychological properties of being basic, even if they have the same name as adult superordinate categories. A similar analysis can be given of cross-cultural differences in categories. People in industrialized societies tend to have a basic level of biological categories that is higher than that of people in agricultural or foraging societies (Atran, Medin, & Ross, 2004). For example, there is good reason to think that sparrow and swallow were basic categories at one time in the history of our English-speaking ancestors, but now the basic category for most American college students is just bird. This is explained as a function of industrialized people’s loss of information about the categories. When they no longer knew much about birds, they couldn’t easily classify things as sparrows and swallows. They simply don’t know what properties sparrows in particular have. However, the features common to all or most birds are more consistent, more frequent, and more obvious (e.g. wings), thereby allowing easy classification at that level. Within a culture, experts tend to have a lower basic level of categorization, because they know more features that distinguish categories (Tanaka & Taylor, 1991). As a result, the claim that basic categories are preferred may be maintained even though different categories are preferred by different groups of people. What is basic is not the particular category but rather the concept people have of it. When their knowledge differs, the ‘basicness’ of the category may differ as well, following the principles of informativeness and distinctiveness. These well-accepted explanations of differences in preferred concept levels (expertise and cultural effects) show that my proposal for developmental differences is not an ad hoc attempt to ‘save’ the basic level advantage. The basic level differs in ways that reflect amount of knowledge one has about categories, not just for children, but for adults as well, and this assumption is incorporated into explanations of conceptual preference.

2.6 Childhood’s End One might wonder what implications my discussion has for the general question of how to decide whether two people have the ‘same’ concept. Unfortunately,

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

47

space precludes addressing this question in any detail—or perhaps fortunately, because I really do not have an answer. However, I think it is unhelpful to consider this as a dichotomous issue in which two people have the same concept, or they do not. That framing leads immediately to the conclusion that those people can/ cannot communicate about that concept and therefore can/cannot understand one another. In real life, communication is often a matter of degrees. I might understand the literal message of your utterance but not detect that you are being ironic, or making an allusion. I might interpret a category name in a way that overlaps with yours ninety per cent, so when you tell me a new fact about that category, I correctly assign it to ninety per cent of your intended extension (plus some members that you didn’t mean it to apply to). Or our concepts might overlap eighty-five per cent. Or sixty-two per cent. Drawing a line somewhere to indicate when that closeness becomes enough to count as the same concept must be arbitrary. Of course, if one insists on 100 per cent agreement, that would not be arbitrary. This is essentially what Fodor (1981) wishes for in suggesting that we could not communicate about drinking water if our concepts differed. But communication is in fact possible between, say, children and adults, even when we know that children’s concepts don’t correspond to those of adults. If we use a behavioural criterion for having identical concepts, many or perhaps most of our concepts will not be shared with others under the 100 per cent agreement criterion (Barsalou, 1987). Such a criterion could therefore not be useful to tell us how people actually manage to communicate, and also how they fail.³ Of course, such arguments have led some writers to propose that the concept must be in the world rather than in the head, just so that such differences in knowledge and belief will not affect semantics and communication. Now we really are on a topic that space does not permit me to address (again, see Laurence & Margolis, 1999, for a review)! However, the obvious issue for such accounts is to explain how it is that people communicate, given that they do not have epistemic contact with that external concept. I know what dogs are, more or less, but I don’t

³ James Hampton points out two potential exceptions to me. One is formalized concepts such as addition. Studies of people’s actual categorization of such well-defined concepts find significant differences (most recently by Lupyan, 2013), but there is a normative quality to such categories that may make it reasonable to say that some people have the correct concept, which is therefore shared. Another is natural kind terms, in which people may defer to experts in determining the category extension, thereby achieving uniformity (assuming the experts themselves agree). Here I am not so sure, as deference to experts is something that happens very rarely, much more rarely than the use of categories and their labels in everyday life. You and I might disagree about whether sharks are fish, but we will probably never get around to consulting an expert. Yet we use the words shark and fish to communicate successfully. Furthermore, in my experience, people often refer to experts in an incorrect or naive way, perhaps only incompletely understanding the expert position or not understanding which expert should be qualified to judge. For example, students like to tell me that the US Supreme Court has determined that tomatoes are a fruit (or vegetable) and that biologists have determined that palm trees are not real trees, both of which are rank nonsense (for different reasons). In any case, even if people are willing to defer to experts in principle, experts are not available enough to get our concepts to converge in many cases in practice.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

48

 . 

pretend to be able to identify every dog or to know just what makes a dog a dog. If my own concept of dogs does not match that external concept, then the fact that there’s an external concept that includes all and only dogs can’t help me communicate, since I don’t know that concept. It’s a bit like saying that disagreements in word usage will be resolved by looking in a big dictionary in the sky, which we unfortunately will never be able to read in our lifetimes. It may be reassuring to think that there is a right answer somewhere, but it is not much help in everyday practice or in scientific explanations of language use. I suspect that the existence of external criteria act as a guiding force in category formation, as life will be much more successful if our categories correspond with natural groupings in the world. However, to assume that our concepts do in fact correspond to externally defined categories seems contrary to everyday observation as well as psychological evidence.

2.7 Conclusion The goal of my chapter has been to show that generalizations regarding concepts and word meanings are still possible if we admit that people may have significant differences in concepts, even systematic differences, like those between children and adults. Thus, this internalist approach does not prevent us from making the scientific generalizations that are part and parcel of cognitive science. However, doing so does require us to consider and analyse the differences in people’s concepts, rather than simply assuming that they are similar enough. The use of a common word can be misleading, especially if we use the loose criteria for knowing a word that language acquisition researchers generally have followed. One important result of allowing for individual differences in concepts is that we can then seek generalizations in those differences, including conceptual development, cultural differences, and expertise effects, rather than assuming that people who use the same word must have the same concept. However, those generalizations must be in terms of things like conceptual structures and information rather than in terms of shared concepts, which is a state that may seldom exist.

References Anglin, J. M. (1977). Word, object and conceptual development. New York: W. W. Norton. Atran, S., Medin, D., & Ross, N. (2004). Evolution and devolution of knowledge: A tale of two biologies. Journal of the Royal Anthropological Institute, 10, 395–420. doi:10.1111/j.1467-9655.2004.00195.x

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

49

Barsalou, L. W. (1987). The instability of graded structure: Implications for the nature of concepts. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and intellectual factors in categorization. Cambridge: Cambridge University Press. Behl-Chadha, G. (1996). Basic-level and superordinate-like categorical representation in early infancy. Cognition, 60, 105–41. Bloom, P. (2001). Précis of How children learn the meanings of words. Behavioral and Brain Sciences, 24, 1095–134. Brown, R. (1958). How shall a thing be called? Psychological Review, 65, 14–21. Fodor, J. A. (1998). Concepts: Where cognitive science went wrong. Oxford: Clarendon Press. Hampton, J. A. (1979). Polymorphous concepts in semantic memory. Journal of Verbal Learning and Verbal Behavior, 18, 441–61. Hampton, J. A. (1995). Testing the prototype theory of concepts. Journal of Memory and Language, 34, 686–708. Laurence, S., & Margolis, E. (1999). Concepts and cognitive science. In E. Margolis & S. Laurence (Eds.), Concepts: Core readings (pp. 3–81). Cambridge, MA: MIT Press. Lupyan, G. (2013). The difficulties of executing simple algorithms: Why brains make mistakes computers don’t. Cognition, 129, 615–36. doi:10.1016/j.cognition.2013.08.015 Meints, K., Plunkett, K., & Harris, P. L. (1999). When does an ostrich become a bird? The role of typicality in early word comprehension. Developmental Psychology, 35, 1072–78. Mervis, C. B. (1987). Child-basic object categories and early lexical development. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and intellectual factors in categorization (pp. 201–33). Cambridge: Cambridge University Press. Murphy, G. L. (in preparation). Building a theory of concepts for cognitive science. Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: MIT Press. Murphy, G. L. (2016). Explaining the basic-level concept advantage for infants . . . Or is it the superordinate-level advantage? In B. H. Ross (Ed.), Psychology of Learning and Motivation (Vol. 64, pp. 57–92). San Diego: Academic Press. Murphy, G. L., & Brownell, H. H. (1985). Category differentiation in object recognition: Typicality constraints on the basic category advantage. Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 70–84. Quinn, P. C., & Johnson, M. H. (1997). The emergence of perceptual category representations in young infants: A connectionist analysis. Journal of Experimental Child Psychology, 66, 236–63. Quinn, P. C., & Johnson, M. H. (2000). Global-before-basic object categorization in connectionist networks and 2-month-old infants. Infancy, 1, 31–46. Rey, G. (1983). Concepts and stereotypes. Cognition, 15, 237–62. Rogers, T. T., & McClelland, J. L. (2004). Semantic cognition: A parallel distributed processing approach. Cambridge, MA: MIT Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

50

 . 

Rosch, E. H. (1973). On the internal structure of perceptual and semantic categories. In T. E. Moore (Ed.), Cognitive development and the acquisition of language (pp. 111–44). New York: Academic Press. Rosch, E., & Mervis, C. B. (1975). Family resemblance: Studies in the internal structure of categories. Cognitive Psychology, 7, 573–605. Rosch, E., Mervis, C. B., Gray, W., Johnson, D., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382–439. Sutcliffe, J. P. (1993). Concept, class, and category in the tradition of Aristotle. In I. van Mechelen, J. Hampton, R. S. Michalski, & P. Theuns (Eds.), Categories and concepts: Theoretical views and inductive data analysis (pp. 35–65). London: Academic Press. Tanaka, J. W., & Taylor, M. E. (1991). Object categories and expertise: Is the basic level in the eye of the beholder? Cognitive Psychology, 15, 121–49. Tversky, B., & Hemenway, K. (1984). Objects, parts, and categories. Journal of Experimental Psychology: General, 113, 169–93. Younger, B. A. (2010). Categorization and concept formation in human infants. In D. Mareschal, P. C. Quinn, & S. E. G. Lea (Eds.), The making of human concepts (pp. 245–63). Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

3 When Does Communication Succeed? The Case of General Terms Peter Pagin

3.1 Intersubjective Conceptual Differences At the time of writing this chapter, there is a debate going on in Sweden as to whether esports (‘e-sport’ in Swedish) are sports. More precisely, should a confederation for esports become a member of The Swedish Sports Confederation (Riksidrotssförbundet)?¹ A crucial question is whether the criterion of administering a sport is satisfied. Those who argue for inclusion say that esports are sports (for instance, a multiple world champion of Counterstrike), and vice versa; proponents and opponents differ over the concept of sport. Or, more precisely stated, proponents and opponents either share a concept of being a sport and at least one of the parties misapplies the concept, or else they have different concepts of a sport. More precisely, again, in the second case they associate different concepts with the same noun: ‘sport’. If both parties apply their respective concept correctly, then the proponent of inclusion interprets ‘sport’ as expressing a concept that includes esports, while the concept of the opponent’s interpretation does not. As long as there is minimal risk of misunderstanding, I shall more briefly express this by saying that the proponent and the opponent have ‘different sport concepts’. Differences in sport concepts do not arise only with esports. James Hampton (2007) asked subjects whether they counted chess as a sport and whether they counted hiking as a sport. The former involves a competitive element but no physical effort, while the latter involves physical effort but no competitive element. Answers differed between subjects, depending on whether the respective concepts of the subjects involved, or didn’t involve, the condition of involving a competitive element, or physical effort, respectively. This case concerns culturally defined types of activity. We have similar examples for artefacts. In this category we have questions like whether slippers are shoes. English dictionaries (Websters, American Heritage, Oxford Advanced ¹ The question was eventually decided, after being postponed for two years, on 26 May 2019, in the negative. Peter Pagin, When Does Communication Succeed ? The Case of General Terms In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Peter Pagin. DOI: 10.1093/oso/9780198803331.003.0004

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

52

 

Learner’s) define slipper by saying that it is a kind of shoe with certain characteristics (e.g. can be easily slipped on and off). In spite of this, we find several websites (e.g. http://blog.clifford-james.co.uk/slipper-vs-shoe/) where slippers are contrasted with shoes. Hence, there are different shoe concepts among English speakers. As a comparison, the Swedish dictionary Svensk Ordbok of the Swedish Academy defines ‘toffel’, which is the standard translation of ‘slipper’, to be a kind of ‘fotbeklädnad’ (translation of ‘footwear’), but does not say it is a kind of ‘sko’, where ‘sko’ is the standard translation of ‘shoe’. In the area of natural categories the noun ‘tree’ provides an interesting example. According to an early July 2012 Wikipedia article on Horsetail, the giant Horsetail (Equisetopsida) in prehistoric times (about 350 million years ago) was a tall tree (the current article still says so; February 2017). However, according to an early July 2012 Wikipedia article on Tree, giant Horsetail was not a tree, because it did not have woody tissue. Clearly, assuming no misinformation, there was a difference between the respective tree concepts of the two Wikipedia authors. One of them required woody tissue in the definition of tree and the other did not. The current Wikipedia entry on Tree opens by recording differences between definitions on the market: In botany, a tree is a perennial plant with an elongated stem, or trunk, supporting branches and leaves in most species. In some usages, the definition of a tree may be narrower, including only woody plants with secondary growth, plants that are usable as lumber or plants above a specified height. (Wikipedia, entry on Tree, 4 February 2017)

Clearly, with such a variety of definitions offered, speakers adhering to the different definitions will associate very different extensions with the noun ‘tree’. A final, and most remarkable, example, concerns words for body parts. Asifa Majid (2010) compares words for main body parts across three languages: Dutch, Japanese, and Indonesian. She asked subjects to colour the part of a drawing of a human body that corresponds to the word. In one case, she asked subjects to colour the part of the drawing that corresponded to the word for arm: Dutch ‘arm’, Japanese ‘ude’, Indonesian ‘tangan’. There are differences between the languages. As a matter of fact, there were variations also between speakers of the same language. In particular, half of the Dutch speakers coloured the arm from the shoulder down to the fingertips and half from the shoulder down to the wrist. So, under reasonable assumptions about the subjects, half of the Dutch speakers interpreted the Dutch word ‘arm’ to express a concept of a body part that includes the hand, while the other half a part that does not include the hand. The result is particularly surprising, since the term ‘arm’, in Dutch or in English, is very frequently used, and is used to denote something familiar to all

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ? 53 speakers. Nevertheless, they don’t all mean the same. The differences are big and concrete. What shall we make of it? This chapter asks and tries to answer two questions concerning communication with expressions that are understood differently by speaker and hearer: first, why are the differences not commonly known, and second, because of these differences, does communication fail? It turns out that the first question is rather easily answered. Section 3.2 provides an explanation and Section 3.3 discusses the more difficult, second question.

3.2 Pragmatics Hides Differences Suppose that speaker A has an arm concept that includes the hand, while speaker B has an arm concept that excludes the hand, and suppose that they are unaware of this difference. Consider the spoken phrase: (1) I have a pain in my arm. Clearly, if B makes the statement, the pain experienced will be located somewhere between the wrist and the shoulder in one of B’s arms. Similarly, this is how B will interpret the same statement as made by A. But what will A mean by stating (1)? Could A sincerely state (1), in a normal, everyday context, to report a pain that A has in his hand? Intuitively not. Even someone who does count the hand as part of the arm would find it misleading to report a hand pain by (1). Rather, the phrase instead would be: (2) I have a pain in my hand. Similarly, someone who counts slippers as shoes would not report the acquisition of a pair of slippers by saying: (3) Today I bought a pair of shoes. Again, that would be misleading. What is the explanation? It seems that there is a preference for being more specific when there is a good occasion. Even if the hand counts as part of the arm, reporting a pain in the arm would seem insufficiently specific, because it is insufficiently informative. The pattern is therefore straightforwardly explained as an application of Grice’s maxim of Quantity. The first maxim of Quantity says: (Quantity 1) Make your contribution as informative as is required (for the current purposes of the exchange) (Grice, 1975/1989, p. 26).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

54

 

Reporting a pain in the hand by means of (1) would be insufficiently informative, and hence a violation of Quantity 1. Under the assumption that the speaker obeys the general Cooperative Principle (Grice, 1975/1989, p. 26), it can be inferred that the speaker does not violate Quantity 1, and hence does not report a pain in the hand. Hence, by means of saying (1), in accordance with the mechanism described by Grice, the speaker implicates that the pain in the arm is not in the hand. This is independent of differences over the parthood question. Presumably, by stating (1), the speaker implicates that the pain is not in any saliently named part of the arm, such as the wrist or the elbow, but when it comes to the hand, the implication will conceal the conceptual difference. Similarly, by stating (3), the speaker implicates that she has bought a pair of shoes that are neither slippers, nor sandals, nor (high) boots, and especially not rubber boots, or shoes intended for sport. Conceptual differences come to light in abstract contexts like that proposed by Majid (2010), where the influence of other contextual factors is minimized and where the question concerns the entire arm, as opposed to a part of, or location in, the arm. Because of the latter feature, there is no quantity effect: colouring the arm down to the fingertips is exactly as informative as colouring the arm down to the wrist. This observation raises the question of the nature of those contexts in which such conceptual differences will come to light. Before answering that, we need to know what coming to light amounts to. Suppose there are two semantic functions μ and μ0 for a language L and an atomic predicate F of L over which the two semantic functions differ. For the difference between μ and μ0 to come to light, minimally there must be a sentence in L where the truth value will differ between the two functions, and the difference depends exactly on the interpretation of F. To spell this out a little more precisely, let μi;c ðSÞ be the truth value of sentence S according to the semantic function μ, at the index i with respect to the context c. The index can be a possible world, a pair of a time and a possible world, or something yet different. Let α be the actual index, meaning the actual world, or the actual world and a privileged time, or something yet different. Let S½F=G be the result of replacing the predicate G in S by the predicate F in zero or more occurrences. Then we require the following: The difference between μ(F) and μ0 (F) comes to light semantically iff there is a predicate G, a sentence S, and a context c such that μα,c (S ) = μ0 α,c (S ), but μα,c (S[F/G]) ≠ μ0 α,c(S[F/G]).

S and S½F=G then forms a minimal pair, differing only with respect to the substitution of F for G, and this substitution generates a difference in truth value between the two interpretations.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ? 55 Semantic coming to light is hardly sufficient for real noticing. Because of the implicature involved in (1), no difference would be noticed between the inclusive and the exclusive interpretation of ‘arm’. We should also consider how the difference in truth value survives or does not survive pragmatic operations. We need to distinguish between primary and secondary pragmatic processes (roughly following the terminology of Recanati, 2004). The primary operations involve loosening, semantic shift, and free enrichment.² The secondary processes consist of the implicatures. We can add these two steps in turn. Let ‘P1;c ’ stand for the primary pragmatic operation on the semantic meaning in context c. We say that a difference that comes to light after primary pragmatic process are taken into account comes to light in what is said. We define: The difference between μ(F) and μ0 (F) comes to light in what is said iff there is a predicate G, a sentence S, and a context c such that P1,c (μα,c (S)) = P1,c(μ0 α,c(S)), but P1,c(μα,c(S[F/G])) ≠ P1,c(μ0 α,c(S[F/G])).

Let ‘P2;c ’ stand for the secondary pragmatic operation on what is said in a context c. We say that a difference that comes to light after secondary pragmatic processes are taken into account comes to light in what is communicated. We define: The difference between μ(F) and μ0 (F) comes to light in what is communicated iff there is a predicate G, a sentence S, and a context c such that P2,c (P1,c (μα,c (S))) = P2,c(P1,c(μ0 α,c(S))), but P2,c(P1,c(μα,c(S[F/G]))) ≠ P2,c(P1,c(μ0 α,c(S[F/G]))).

Using this terminology, we can say that difference between the inclusive and the exclusive interpretation of ‘arm’ comes to light semantically, and also in what is said, by stating (1), but does not come to light in what is communicated. That explains why it goes unnoticed. What characterizes contexts where such conceptual differences would come to light also with respect to what is communicated? The crucial feature is, I think, coercion: the freedom for the speaker to choose expression is significantly limited. Consider the question: (4) Is this a shoe (yes or no)? when a slipper is displayed. In answering the question, the speaker is not given the choice of the alternative noun ‘slipper’. The answer ‘yes’ clearly does not carry the implicature that what is demonstrated is a non-slipper shoe, because the choice of the more informative term is not allowed. This effect goes not only for a listener ² Recanati also includes saturation, which is assigning values to context dependent expressions. I here take that to be given by the context argument in the semantics.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

56

 

that sees the demonstrated object, but also for a listener around the corner who does not see it and who does not independently know what kind of object it is. The listener around the corner who takes the speaker to have an inclusive interpretation of ‘shoe’, and who believes the speaker, is still in doubt as to whether the object is a slipper. There can be other means of blocking the implicature. If you are asked: (5) How many pairs of shoes are there on the stairs? and you see five pairs of standard shoes and one pair of slippers, the answer: (6) There are five pairs of shoes and one pair of slippers. is cumbersome, not fully congruent with the question, and may seem overly pedantic. The more informative answer is then dispreferred for reasons other than being false.³ Hence, if your interpretation of ‘shoe’ is inclusive, you do not answer ‘five’ in order to implicate that there are that many non-slipper shoes but that there also may be additional footwear. Rather, the strong preference for the shorter answer makes you say ‘six’ in order to communicate something true. Some pragmatic phenomena can prevent conceptual differences to come to light with respect to what is said. Pragmatic enrichment is a case in point. Consider: (7) a. He had taken off the jacket and thrown it on the floor. The shoes were lying next to it. b. He had taken off the jacket and thrown it on the floor. The shoes [that he had worn with it] were lying next to it. A typical reading of (7a) includes a so-called bridging inference that raises the coherence of the discourse (Clark, 1975; Levinson, 2000, pp. 37–38, Wilson & Matsui, 1998/2012, Pagin 2014). Since we normally do not wear jackets with slippers, the default reading would be that the shoes in question are not slippers, even on an inclusive interpretation of ‘shoe’. This latter inference, in turn can be seen as generated from Grice’s second maxim of Quantity: (Quantity 2) Don’t make your contribution more informative than is required. (Grice, 1975/1989, p. 26).

The idea is that we can infer from normal conditions that slippers are not worn with jackets and that this does not have to be mentioned. This idea is elaborated in ³ From a Gricean point of view, the maxim of Manner trumps the maxim of Quantity.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ? 57 Stephen Levinson’s (2000) theory of I-implicatures. As part of that theory, Levinson proposes a general interpretation heuristic: (Heuristic 2) What is simply described is stereotypically exemplified (Levinson, 2000. p. 32).

The use of the simple noun ‘shoe’ without elaboration indicates, according to this heuristic, that the kind of shoe reported is of the kind stereotypically used together with jackets, thus excluding slippers.⁴ As for forcing the differences to come to light, we can block the triggers of enrichment by being maximally explicit and by avoiding indication of typical wearing situations. However, conceptual difference may also be blocked from coming to light with respect to pragmatic factors that affect the semantics itself. In the context where the topic of discussion is preparing for a hike, slippers are not relevant. The answer ‘five’ to the question can then be true, even under an inclusive interpretation. The mechanism here would be quantifier domain restriction (cf. Westerståhl, 1985): the range of quantification is limited to those shoes that are fit for hiking. Hence, there are five. Quantifier domain restriction can be blocked in turn by creating an abstract context where, among other things, no particular practical considerations play a role, or by creating an imagined context where the domain restriction is controlled for. This can normally be achieved in experimental settings. We have seen here how various kinds of pragmatic operations can serve to conceal intersubjective differences in word meaning. We have also seen how these pragmatic operations can be blocked. They do get blocked in somewhat artificial and unusual contexts. We thus have explanations of how these differences can be noticed and why this nevertheless does not easily happen.

3.3 The Nature of Communicative Success In communication there is always a signal of some kind, a sender of the signal, and a receiver of the signal (usually distinct from the sender). When communication is successful, some particular success-related correlation obtains between the cause of the signal and the effect of the signal. For instance, in the case of an animal alarm call, the sender has observed a predator, which

⁴ In conversation, Francois Recanati has stressed the role of so-called prototypes, in the sense of Rosch (1978), in connection with divergence-hiding pragmatics. I am inclined to believe, however, that the notion of a stereotype, which can depend on type of situation described, is more useful in this context.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

58

 

causes it to emit the alarm call, and the call in turn causes the receiving conspecific to seek appropriated shelter. In the alarm call case, we have two main choices for the relata of the success relation. We can take them to be the behaviour of the animals. In this case, we take the emitting of the call as the behaviour of the caller, and the shelter-seeking as the behaviour of the hearer. In case the hearer does not react, or reacts by taking the wrong kind of shelter, or by some other behaviour, the communication has failed. Communication has taken place, since the signal was observed by the hearer, but it was not successful. We can also take the relata to be the inner states of the animals. In this case, we take the caller’s relevant inner state to be the state that is generated by observing the predator, and the hearer’s inner state to be the state that is triggered by observing the signal, and which then typically causes the shelter-seeking response. Transposed to human linguistic communication, the first option is broadly behaviourist: communicative success depends on whether the appropriate kind of behaviour is executed by the hearer, a behaviour of a kind that in a relevant respect matches the behaviour of the speaker. It is natural then to take the matching to consist in coordination of subsequent behaviour, such as meeting up at a later time. The second option is broadly mentalistic: success depends on whether the inner state of the hearer that results from observing the utterance matches, in the relevant way, the inner state of the speaker that triggered the utterance. In case the utterance was an assertion, it is natural to take the relevant inner states to be, in the standard case, beliefs.⁵ I have argued extensively (Pagin, 2008) against the behaviourist alternative, mainly that behaviour can be coordinated in the relevant way also in cases that are intuitively examples of communicative failure. In those cases, the intuitive misunderstanding is compensated for by false beliefs, typically of the hearer, such that the effects of the misunderstanding and the false belief cancel out, leading to the same overall effect as successful communication combined with true beliefs. I shall not repeat the discussion here, but proceed with the mentalist alternative that what matters for success are the modes and the contents of the propositional attitudes of speaker and hearer. These are the success relata. What is, then the success relation? When does communication by means of a linguistic utterance succeed? The default, with respect to content, is identity.⁶ The

⁵ Crucially, communicative success does not depend on whether the hearer actually believes that what the speaker says is true. Rather, what matters is that the hearer comes to entertain the appropriate proposition in thought, whether or not that results in believing the proposition. ⁶ I shall not here discuss the success conditions with respect to mode. Clearly, if the speaker makes an assertion and the hearer interprets it as a question, or as the expression of a wish, there is misunderstanding, and in one respect, communicative failure. The problem here is to accommodate the intuition that there must be a match without requiring, too, the theoretical notions of the hearer. For instance, a hearer who does possess the theoretical concept of an assertion cannot be required to

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ? 59 content of the hearer’s belief that relevantly results from observing the utterance is the same as the content of the belief that relevantly causes the utterance. For instance, when the speaker asserts: (3) Today I bought a pair of shoes. communication is successful just in case the hearer comes to think the very same proposition, that the speaker the same day bought a pair of shoes. However, we noticed in Section 3.2 the complications that pragmatics gives rise to when it comes to hiding conceptual differences. It seems reasonable to say that, as far as the semantic content of the utterance of (3), after saturation, i.e. fixing the values of indexicals, there is a failure of communication between the inclusive speaker (who counts slippers as shoes) and the exclusive hearer (who does not). It is not equally clear what to say about what is said and what is communicated. In general, it can easily happen that speaker and hearer converge on standing meaning while diverging on occasion meaning, or what is said. For instance, when you text me I am here I do understand the standing meaning (character) of the sentence, but I take you to be referring to the cafeteria by ‘here’, while you are referring to the subway station. Intuitively, this is a communicative failure with respect to what is said, but not with respect to standing meaning. On some semantic theories, you can also get the converse result. For instance, applying Keith Donellan’s (1966) distinction between referential and attributive uses of definite descriptions, a misunderstanding of linguistic material inside a referentially used description might have no significance. When A says: (8) The man in the corner drinking a martini is a philosopher. on Donellan’s view, the description, if used referentially, on the occasion refers to the man in the corner, even if what he is drinking is not a martini. By the same token, if B believes that a martini is a kind of gin, but does take the description to have been used referentially, A and B will be thinking about the same man in the corner, and on Donellan’s account, this is enough for communicative success with respect to what is said. With respect to semantics (standing meaning), on the other hand, there is communicative failure, because of the divergence on the interpretation of ‘martini’.⁷

believe that the speaker made an assertion, but can nevertheless react differently to assertions and questions, in adequate ways. I shall leave this issue aside. ⁷ In Kripke’s (1977) alternative view of the matter, there is a clear difference between the meaning of it and what is said by means of it under the referential use.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

60

 

We can easily have communicative success with respect to what is said combined with failure with respect to what is communicated, for instance, if the hearer reads an implicature into an utterance that was not intended (and not believed to be true) by the speaker. We can also easily have success with respect to what is communicated, even though it is based on failure with respect to what is said. (9) A: Would you like to go to the movies? B: I have to go to the bank. The implicature is that B does not have time to go to the movies, regardless of whether he means river bank or financial institution by ‘bank’. Hence, a misunderstanding between A and B on this score has no effect on success with respect to the implicature. The upshot is that we should distinguish between communicative outcomes with respect to level of the communicative device: semantics (standing meaning), what is said (occasion meaning), and what is communicated (implicated). The remainder of the chapter focuses on the first two.⁸ We can then return to the question of whether identity of content is required for success, with respect to semantics, or with respect to what is said. But again, there is a further distinction to be made.

3.4 Structured Contents In what follows, I will, unless otherwise noted, speak of communicative outcomes (success or failure) with respect to what is said. That is, I think, the most fundamental level of communication. Language is a means for effecting communication at the level of what is said, even though as we have seen linguistic understanding and communicative success can come apart. I take it as intuitively true that if speaker A successfully communicates a particular content to hearer B, then B is in a position to truthfully report what A has said. Thus, if A utters: (10) There is a traffic jam on the Brooklyn bridge. and thereby communicates successfully with B, in accordance with the standard meaning of the sentence, then B can truthfully report:

⁸ Things are somewhat complicated by the fact that implicatures sometimes have an effect on what is said. This is what Levinson (2000, p. 186) has called ‘Grice’s circle’. In fact, (3) may be a case in point. It can be argued that the quantity implicature that the speaker does not report buying a pair of slippers is projected back on what is said, leading to its being correct to report the speaker as having said that he had bought a pair of non-slipper shoes (even on an inclusive interpretation). I shall, however, try to avoid this complication in the remaining discussion.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ? 61 (11) A said that there is/was a traffic jam on the Brooklyn bridge.⁹ It is a consequence of this relation between communicative success and indirect discourse that whatever fineness of grain is required in indirect discourse, and similarly in belief reports, carries over to success conditions.¹⁰ In particular, if indirect contexts and belief contexts are hyperintensional, then so are success conditions. They are hyperintensional in the sense that whatever similarity of content is ultimately needed, similarity of intension is not in general enough. This agrees, I think, with direct intuitions. For suppose that A utters (12a), while B interprets A as having said what (in A’s idiolect) would have been expressed by (12b): (12) a. Trump won the 2016 election and brothers are male siblings. b. Trump won the 2016 election and bachelors are unmarried men. The intension of (12a) is the same as the intension of (12b), namely, the proposition true at every world where Donald Trump wins the 2016 US presidential election. Nevertheless, communication intuitively fails with respect to semantics and with respect to what is said. B misunderstands A despite getting the right possible-worlds proposition. The remainder of this chapter assumes both that the condition of communicative success with respect to what is said agrees with the truth-conditions of indirect discourse and belief reports, and that they are both hyperintensional. That does not, however, entail that the relation between the contents of speaker and hearer (and the condition of the speaker’s content and the content later reported by the hearer) must be identical. Neither does it determine exactly what the model of the content relata should be. In earlier work (Pagin, 2007, Pagin & Pelletier, 2007, Pagin, 2019), I have appealed to structured meanings. Alone and together with Jeff Pelletier I have argued that a semantics of structured meanings is needed in order to get appropriate fine-grained arguments for pragmatic operations. Pragmatic operations often take parts of contents as argument as in the classical enrichment example from Robyn Carston (2002, p. 138): (13) a. He ran to the edge of the cliff and jumped. b. He ran to the edge of the cliff and jumped [over the edge of the cliff].

⁹ Choose your preferred tense. I am setting the tense and temporal issues aside here. ¹⁰ The idea that truth-conditions of belief attributions should be coupled with success conditions of communication has also been defended by David Chalmers (2011, pp. 619–21). Chalmers proposes a relation of coordination between propositions (of a certain kind), both for the truth-conditions of belief sentences and for communication. On this issue, see also Pagin 2019.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

62

 

Here, as in many other examples, it is just unclear what an operation would look like that, when applied in this case, takes the entire unstructured proposition expressed by (13a) as argument and gives the entire unstructured proposition expressed by (13b) as value. By contrast, the operation that takes the intension of an action verb as argument and gives as value that intension restricted by the intension of a preposition phrase is straightforwardly specified. I have also proposed a structured meanings approach to belief sentences (Pagin, 2019), which relies on switching mechanisms to introduce structured meanings in belief contexts, as ingredients of unstructured meanings. I have argued that structured meanings, as models of how subjects represent unstructured propositions to themselves, provide the adequate means for combining intensions, as belief contents, and hyperintensional entities. Structured entities with ordinary intensions as simple elements are well-understood (in contrast to ideas of unstructured Fregean senses), and paired with a given recursive projection function, determine the corresponding unstructured intensions. I propose that, in the case of communication as well, we model the hyperintensional success conditions by means of structured meanings. We shall thereby require that for communicative success between speaker A and hearer B, the respective content CA and CB be communicatively isomorphic, in a sense corresponding to Carnap’s notion of intensional isomorphism (1947/1956, p. 59): (CIS) Communicative isomorphism. Contents C₁ and C₂ are communicatively isomorphic iff i) C₁ and C₂ are both unstructured and they are similar, or ii) C1 and C2 have the same structure of immediate parts, and each immediate part C1,i of C₁ is communicatively isomorphic to the corresponding immediate part C2,i of C₂. The remaining question then is what similarity should amount to.

3.5 Intensional Similarity Let us return to varying lexical interpretations of ‘tree’ from Section 3.1. Consider the following example. We suppose that Woody requires trees to have woody tissue, while Wendy does not. For all items they ever actually come across, they classify them the same with respect to being a tree or not. No lexical difference actually comes to light. They would have classified Giant Horsetail differently, but are not aware of the difference. Does it induce communicative failure with respect to an utterance such as: (14) John has five trees in his garden.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ? 63 Despite the lexical divergence, I would count communication by means of (14) as successful (assuming the other elements do not induce failure). Their divergence over ‘tree’ does not seem to matter for the current topic. At least, it does not matter enough. The trees in the garden are birches and maples. How could it matter that Woody and Wendy disagree over Giant Horsetail? In a sense, the answer is easy: it would have mattered if John had had a Giant Horsetail in his garden. The plausible explanation of why the lexical divergence does not seem to matter is that it could not easily have happened that John had a Giant Horsetail in his garden, or any other tree-sized plant without woody tissue. Given the actual natural history of the Earth, this happens only in remote possible worlds. Hence, only in remote possible worlds does Woody’s interpretation of (14) differ in truth value from Wendy’s interpretation of (14). We can take this intuition about the tree example as our guiding idea about similarity of intension. We assume that we have fixed a standard of normal possibilities, i.e. a set of possible worlds that, for some given similarity metric between worlds, are close to the actual world. Call this set N. Two unstructured intensions C1 and C2 are then similar just in case their respective extensions in normal worlds are the same. That is, for any world w 2 N, C1 ðwÞ ¼ C2 ðwÞ. We can immediately note that this similarity relation is transitive, as long as we stick with one fixed set of normal worlds.¹¹ This idea of similarity between contents thus makes it into an equivalence relation. Its application to intersubjective comparison was suggested first in Pagin (2006). With this basic notion of similarity between unstructured contents, and the idea of communicative isomorphism, we have the elements of an account of communicative success despite lexical divergence. There are complications, however, because in some contexts, small differences do matter. For instance, consider: (15) a. 350 million years ago, Giant Horsetail was a tree. b. Trees have woody tissue. ¹¹ There is a simple argument for transitivity: we start with the natural condition that communication fails if the speaker proposition has a different truth value from the hearer proposition. Then we consider a sorites-like chain of communication events where the hearer in event i is the speaker in event i+1, and where in each exchange the difference between speaker and hearer proposition is small enough for success, but where the difference between the first proposition in the chain and the last is large enough to preclude sameness of truth value. Such a chain is possible. There must then be a change of truth value somewhere in the chain, violating the natural condition. To conclude the argument for transitivity we also need to exclude the possibility of simply combining the natural condition with a non-transitive success relation. On such a combination, communication where speaker and hearer beliefs have very similar but non-identical propositions as their contents succeeds in all worlds except in worlds where the two propositions differ in truth value. This an unacceptable reliance on semantic luck, the luck of being in a world where this is not the case (more about this on another occasion). We would lose transitivity only if the set of normal worlds would vary with the concepts involved. Thus if N is relevant for comparing C1 and C2 , while the larger set N’ is relevant for comparing either with C3 , then it may be that for any world w 2 N, C1 ðwÞ ¼ C2 ðwÞ, for any world w 2 N’, C2 ðwÞ ¼ C3 ðwÞ, while there is some world w 2 N’ such that C1 ðwÞ≠C3 ðwÞ.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

64

 

Clearly, the lexical divergence between Woody and Wendy results in Woody’s interpretation of (15a) being false and Wendy’s interpretation being true. In this case, partly because of the temporal operator, the difference between the two tree concepts induced a difference in truth value between the two resulting propositions. Suppose (15a) is asserted by Wendy and, as a result, is believed by Woody, who does not know that Giant Horsetail did not have woody tissue. In this case, communication clearly fails: the combined failure to know background facts and to be aware of the lexical difference results in Woody having a false belief. Believing that (15a) is true, from his own interpretation of ‘tree’, Woody can infer the false proposition that Giant Horsetail had woody tissue. We also get a truth value difference between the Woody and Wendy interpretations of (15b), provided the statement is understood as having some kind of lawlikeness or modal strength, such that not only actually, locally (on Earth), and presently existing trees are relevant. Under the implicit modal reading, the existence of Giant Horsetail in relevant non-actual worlds would make (15b) false under Wendy’s interpretation. Two issues arise with respect to this example. The first is that the modally strengthened reading of (15b), if that reading is shared, would lead, or could easily lead, to the discovery of the lexical difference. Once the difference is discovered, communication is repaired. Woody can interpret Wendy’s utterances by means of her tree concept, and thereby get the interpretation right.¹² Still, even if communication is dynamically repaired because of the discovery, before the discovery has taken place, there is still failure. If Woody asserts (15b), the proposition mentally entertained by Wendy will be the proposition induced by her interpretation. She will not believe that proposition, since it will strike her as false on hearing it. But as noted above, whether communication succeeds or fails only depends on the proposition that is presented as a candidate for being believed, not on whether the hearer actually believes it. There can be many reasons why a hearer does not believe speaker in a particular case.¹³ The second issue concerns what to count as normal worlds with respect to modal statements. Compare the explicitly modal: (16) Necessarily, trees have woody tissue. For the truth value of (16), clearly all accessible possible worlds are relevant. If there are Giant Horsetails in any accessible possible world, Wendy’s proposition is

¹² It can also go wrong again because of double compensation. If Wendy adapts the semantic intention of her own utterance to Woody’s tree concept, while Woody interprets Wendy’s utterance according to her tree concept, there is still communicative failure. For communication to succeed, there has to be an asymmetry: only one of them can adopt the other’s concept. The problem here is that there is no general strategy as to who that should be, for instance, the speaker or the hearer. ¹³ I discuss the main cases in Pagin (2011).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ? 65 false, since the embedded proposition (that [Wendy] trees have woody tissue) are false in those worlds. That does not, however, mean that we need to expand the set of normal worlds. What we need to consider is whether the modal propositions, of Woody and Wendy, agree in all normal worlds. Since they disagree already in the actual world, the answer is no. But suppose there is no Giant Horsetail at the present time in any world accessible from the actual world, and that there is a world w’, accessible from a normal world w 2 N, where there are Giant Horsetails. Under those assumptions, it would be right to say that the Woody and Wendy interpretations of (16) actually agree in truth value, but that they could easily have disagreed, namely, in the non-actual but normal world w. Thus we can reason exactly as we did with non-modal statements, considering the normal alternative worlds. The same holds with the implicitly modal statement. Communication fails if the modally strengthened propositions that result from the Woody and Wendy interpretations, respectively, disagree in truth value in any normal world. The modal strengthening simply takes care of itself, given the initial set N of normal worlds. The outcome of communication with a particular sentence will depend on the relevant spatio-temporal region, the domain of quantification, and the modal character of the sentence. But the basic idea of success can remain invariant: the respective propositions of speaker and hearer should not easily diverge in truth value, which is to say that they should agree in all normal worlds. Exactly which those are, however, is a question for another time.

References Carnap, R. (1947). Meaning and necessity. Chicago, IL: University of Chicago Press. Second edition published in 1956. Carston, R. (2002). Thoughts and utterances. Oxford: Blackwell. Chalmers, D. (2011). Propositions and attitude ascriptions: A Fregean account. Noûs, 45, 595–639. Clark, H. H. (1975). Bridging. In R. C. Schank and D. L. Nash-Webber (Eds.), Theoretical issues in natural language processing. New York: Association for Computing Machinery. Donellan, K. (1966). Reference and definite descriptions. Philosophical Review, 75, 281–304. Grice, H. P. (1975). Logic and conversation. In P. Cole and J. L. Morgan (Eds.), Speech acts, Vol. 3: Syntax and semantics (pp. 41–58). New York: Academic Press. Reprinted in Grice (1989), Ch. 2, pp 22–40. Page references to the reprint.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

66

 

Grice, H. P. (1989). Studies in the way of words. Cambridge, MA: Harvard University Press. Hampton, J. (2007). Typicality, graded membership, and vagueness. Cognitive Science, 31, 355–83. Kripke, S. (1977). Speaker’s reference and semantic reference. In P. A. French, T. E. Uehling, Jr, & H. K. Wettstein (Eds.), Studies in the philosophy of language (pp. 255–96). Minneapolis, MN: University of Minnesota Press. Levinson, S. C. (2000). Presumptive meanings: The theory of generalized conversational implicature. Cambridge, MA: MIT Press. Majid, A. (2010). Words for body parts. In B. C. Malt and P. Wolff (Eds.), Words and the mind: How words capture human experience (pp. 58–71). Oxford: Oxford University Press. Pagin, P. (2019). Belief sentences and compositionality. Notional part. Journal of Semantics, 36, 241–84. doi:10.1093/jos/ffy018. Pagin, P. (2006). Intersubjective externalism. In T. Marvan (Ed.), What determines content? The internalism/externalism dispute (pp. 39–54). Newcastle upon Tyne: Cambridge Scholar Press. Pagin, P. (2007). Pragmatic composition? In M. Zouhar and T. Marvan (Eds.), Svet jazyka a svet za jazykom (pp. 11–26). Bratislava: Institute of Philosophy. Pagin, P. (2008). What is communicative success? Canadian Journal of Philosophy, 38, 85–116. Pagin, P. (2011). Information and assertoric force. In J. Brown and H. Cappelen (Eds.), Assertion (pp. 97–136). Oxford: Oxford University Press. Pagin, P. (2014). Pragmatic enrichment as coherence raising. Philosophical Studies, 168, 59–100. Pagin, P., & F. J. Pelletier (2007). Content, context and communication. In G. Preyer & G. Peter (Eds.), Context-sensitivity and semantic minimalism. New essays on semantics and pragmatics (pp. 25–62). Oxford: Oxford University Press. Recanati, F. (2004). Literal meaning. Cambridge: Cambridge University Press. Rosch, E. (1978). Principles of categorization. In E. Rosch and B. Lloyd (Eds.), Cognition and categorization (pp. 27–48). Hillsdale, NJ: Lawrence Erlbaum Associates. Westerståhl, D. (1985). Determiners and context sets. In J. van Benthem and A. ter Meulen (Eds.), Generalized quantifiers in natural language (pp. 45–71). Amsterdam: Foris. Wilson, D., & Matsui, T. (1998). Recent approaches to bridging: Truth, coherence, relevance. In: UCL working papers in linguistics 10, pp. 1–28. Reprinted in Wilson and Sperber 2012, pp. 187–209. Page references to the reprint. Wilson, D. & Sperber, D. (2012). Meaning and relevance. Cambridge: Cambridge University Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

4 Investigating Differences in People’s Concept Representations James A. Hampton

4.1 Introduction Language serves many purposes in human society, and one of its prime functions is as a tool for communication. As you read this sentence, you are taking the words and their grammatical marking and deriving an understanding of what the author is trying to convey. A key element in this system of communication is the actual words that are used, and it can only work successfully if people have a common understanding of what those words mean. The words of a given language have a conventionally agreed meaning, or set of meanings, which, for languages such as English, can be found in a dictionary. Dictionaries allow us to determine when someone is using a word correctly or in error. They act as a brake on language evolution, slowing down the natural process by which word meanings change over generations, and at the same time they improve the likelihood that we will understand one another. Given the crucial importance of a shared understanding of word meanings for underpinning successful communication, it is interesting to ask just how strong is that shared understanding. People appear generally to understand each other most of the time, but are there actually stable individual differences in how people grasp any given concept and the meaning of its associated term? The fact of successful communication within a language community has been argued by some philosophers to be evidence for the identity of concepts held by different agents. The so-called publicity constraint (e.g. Rey, 1983) proposes that if two individuals were not using the same identical concept, then it would no longer be possible to resolve any argument of fact between them that depended on that concept. They would necessarily be talking at cross-purposes if they had different meanings for a key word in the debate. While different individuals may have different beliefs or ideas about a particular kind or class of thing, the concept about which their ideas are in conflict itself has to be the same concept. The extent to which this publicity constraint works in practice is the subject of much debate. This chapter’s focus is to review some empirical research into the question of shared understandings. As such, it leaves aside the question of whether James A. Hampton, Investigating Differences In People’s Concept Representations In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © James A. Hampton. DOI: 10.1093/oso/9780198803331.003.0005

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

68

 . 

it is possible to investigate people’s concepts directly, or whether in fact empirical enquiry into people’s concepts is actually tapping into something rather different—for example, their conceptions of those concepts (Rey, 1983). The chapter reviews a set of studies that have looked directly at how much agreement and stability there is in people’s reports of their concepts (or conceptions of concepts). It then describes some research recently published with Alessia Passanisi and others that tries (and fails) to map the similarity of people’s concept intensions (the properties they consider important for a concept) onto their concept extensions (the degree to which they consider that different exemplars are typical of the concept). It also describes some subsequent attempts to follow up this disturbing null result.

4.2 Individual Differences in Concept Tasks First, let us outline the kind of conceptual differences that are the focus of this chapter. We will not, for example, be considering differences in word meanings between US and UK English (pants as underwear or pants as trousers), nor the well-known problems of finding exact translations between languages (English chair versus French chaise and fauteuil). The differences we will be considering are more subtle and emerge from methodology aimed at eliciting people’s personal meanings for words in various more or less direct ways. They are found within samples of speakers within the same language community, and relate to an aspect of word meaning that has been variously described as a stereotype (Putnam, 1975) or prototype (Rosch, 1975; Lakoff, 1987). The notion of a prototype for a concept has been proposed to explain some very widespread phenomena to be found in studies of people’s understanding of concepts (Hampton, 2006). Briefly, there are two primary aspects to any concept that can be explored psychologically. The intension of a concept is the set of associated features or properties that people consider to be a part of the concept, and that determine category membership. For example, for the concept bird it would include features such as having feathers, hatching from eggs, and having two legs. The extension of a concept by contrast is the set of entities (objects, creatures, situations) in the world to which the concept term can be applied—it is the category of things that ‘fall under’ the concept. The two aspects, intension and extension, should in principle be closely related. It is in virtue of possessing the intensional properties of a concept that an individual entity will be included in the extension of that concept. Any feathered creature that hatches from eggs and has two legs and a beak will be a bird, and any creature lacking any of these features will be something else. It should be noted that the use of the terms intension and extension here differs from the classic meaning of the terms introduced by Frege (1892), and as

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ’  

69

commonly used in philosophy and semantics. The chapter uses the terms to refer to behavioural measures of what a person may represent in their mind about a concept. Traditionally, intension is what determines extension, so there is no possibility that the two do not precisely correspond. When we consider the psychology of an individual possessor of a concept however, it is possible that the beliefs they hold about an intension (how they think they define a concept in terms of its properties) may not, in fact, map well onto the beliefs that they hold about the reference or extension of the concept (what they think the concept term refers to in the world). An intriguing example of this mismatch was a study by Levitis, Lidicker, and Freund (2009) in which they showed that members of scientific societies concerned with the study of ‘behaviour’ frequently had selfcontradictory views about what features defined the term for them, and what examples in the world they would count as falling under the term. The use of intension and extension here should be understood then as a short-hand for ‘information that people are able to provide about their understanding of a concept’s defining features and their beliefs about its application or reference in the world.’ Prototype theory (Hampton, 2006, Rosch & Mervis, 1975) takes this difficulty in defining terms to be characteristic of most of our everyday concepts. Following Wittgenstein (1953), Rosch and Mervis (1975) argued that the notion of all and only category members possessing a common set of defining features simply does not work for a large number (indeed a majority) of semantic concepts in everyday language. Hampton (2006) summarizes the evidence. On the intensional side, when people are asked to generate features of a concept, they make no differentiation between features common to the class (birds having feathers) and those which are only found in typical instances (for example, birds flying). In fact determining a good set of defining features for any given concept has been a major intellectual challenge going back to Plato’s Socratic dialogues. Furthermore, the case of birds is exceptional. Most other categories have no clear set of defining features which taken together can determine the correct extension of the concept. Wittgenstein (1953) famously pointed out that if we consider the category of games, there is no common element running through them all, but rather a set of family resemblances, or attributes, which are common to the majority of cases, though not necessarily to all. Research in the 1970s confirmed this intuition by showing how it is often impossible to generate clearcut common feature definitions from people’s descriptions of their conceptual categories (Hampton, 1979; McNamara & Sternberg, 1983). Turning to extensions, there are similar problems. On the one hand, there are often borderline cases to membership in the extension. The concepts are vague in much the same way that adjectives such as tall or bald are vague. There seems to be no higher authority to which we can appeal to decide if an

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

70

 . 

avocado should be called a fruit, or if trampolining should be counted as a sport. People disagree with each other, and are prone to change their opinion from one week to the next (McCloskey & Glucksberg, 1978), or depending on the context in which the question is asked (Hampton, Dubois, & Yeh, 2006). Together with this vagueness at the category boundary, Rosch (1975) also discovered the phenomenon of typicality. The members of a category differ from each other in terms of how typical they are of the concept. Robins and sparrows are more typical as birds than penguins and ostriches, even though all four are bona fide members of the category. Typicality has been found to affect a wide range of tasks such as perceptual processing, ease of recall, and inductive reasoning (Murphy, 2002). Hampton (1988; 1995) showed that, for many categories, the degree to which an item belongs in a category (its graded membership) and its typicality in the category should be treated as two measures based on the same underlying scale of closeness or similarity to the concept prototype. Taken together, the evidence for prototypes is overwhelming. The individual differences in concepts that are the focus of this chapter relate to the prototypes possessed by different speakers. To access someone’s prototype for a concept (be it bird, fruit, or science), there are several tasks that have been commonly used. In production tasks, people are asked to list the properties that they consider important or defining for a concept (the intension) and people are asked to list the exemplars that they consider to fall within the named category (the extension). Because of the rather idiosyncratic data that production tasks can generate, lists of features and exemplars are usually then submitted to rating tasks in which people rate the degree to which each property is an important part of the concept, and the degree to which each exemplar is typical of the category. In addition to typicality judgments, we also can obtain categorization judgments where lists of items including clear members, borderline cases, and clear non-members are categorized as in or out of the category, and the probability of a positive categorization can be measured. For this range of different tasks probing people’s semantic memories, it is possible in each case to determine whether people have their own personal take on a concept that differs from other people’s. The procedure requires that people perform that same task on two separate occasions, usually a week or more apart. The average consistency with which someone gives the same responses on each occasion can then be compared with the average amount of agreement between one person and another. If a person’s responses on the second occasion are closer to their responses on the first occasion than they are to those of some randomly chosen other individual, then there is prima facie evidence that people have what we might term stable ‘idio-prototypes’ in their personal idiolect, which differ from those of other people.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ’  

71

4.3 Evidence for Idio-prototypes Interest in exploring idio-prototypes began in the 1980s with a series of studies by Barsalou (1987), who investigated the stability of typicality judgments, and Bellezza (1984a; 1984b), who tested the stability of responses in a range of different semantic tasks. To take an example, Barsalou provided students with lists of category members to be judged in terms of their typicality in a category. For example, they may have ranked different activities such as baseball, swimming, athletics, and archery for their typicality as examples of sport. Participants worked through several category lists, and then returned after a certain period of time to do the task again. Correlating their judgments over time, Barsalou found that the correlations declined over the first twenty-four hours, but then levelled off at around 0.8. On the other hand, he reported that the average correlation between any two individuals doing the same task was much lower, between 0.3 to 0.6. Similar results were found by Bellezza for production tasks where students generated either category exemplars or category definitions on two occasions a week apart. Within-participant consistency (correlation) was around 0.7 for category exemplars, and 0.5 for definitions, while between-participant agreement or consensus was just 0.4 and 0.2, respectively. The finding that people’s semantic judgments are stable and replicable and that the level of agreement with others is much lower than the level of consistency for an individual over time provides good evidence that people have idiosyncratic meanings for these concepts. It is possible that at least some of the effect may reflect more general individual differences, such as the way that people choose to interpret the instructions, or differences between individuals in the strategy that they may use to perform the tasks. However, the size and consistency of the effect across different tasks suggests that there is more to the result than this.

4.4 Using Similarity to Assess Individual Differences in Concepts Theories of semantic memory and concepts in psychology share a common fundamental assumption. Intensions and extensions should be strongly related. Extensional categories of reptiles and amphibians should be distinguished by their different intensions. Similarly, the intensional differences between reptiles and amphibians should be discoverable (perhaps with some difficulty) by examining their extensions. Scientific categorization typically evolves through a cyclical process of refining intensions/definitions of classes and extensions/reference of terms until a satisfactory classification has been achieved. The two sides of a concept, intension and extension, are thus tightly bound. The question then arises

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

72

 . 

as to whether the individual differences that we see in prototype production and judgment tasks will match across the two types of semantic information. We know from Barsalou, Bellezza, and others that there are quite strong, and constant, individual differences in extension (typicality and membership judgments) and in intension (production of definitions and ratings of feature importance). So, it should therefore be possible to show that individual patterns of similarity and difference between individuals in how they judge extensions should be mirrored in patterns of similarity and difference in how they judge intensions. Consider a simple example from Verheyen and Storms (2013), who explored whether concepts such as sport might show individual differences in conceptual contents. After all, there are some sports involving individual skilled activities that take place indoors, such as darts, billiards, chess, or snooker. High levels of skill are involved, but little in the way of physical exertion. On the other hand, there are outdoor sports which are tests of strength or endurance more than skill, e.g. marathon running, triathlon, and long-distance swimming. So, it is possible that when activities are borderline to the category (like chess or hill walking) some people may favour one type of sport, and other people may favour the other. This was in fact what Verheyen and Storms found. Fitting a quantitative model to categorization probabilities, they discovered that dividing their participant group into subgroups produced a significant improvement in fit to the data, with each subgroup providing a different set of categorization probabilities to the items at the borderline. They also showed that for some categories it was possible to speculate about the underlying intensional basis for the group split, although their evidence for this was weak. They did not have data on individual views about intensions, and so could only provide post hoc accounts of the different categorization profiles. Hampton and Passanisi (2016) followed up an earlier set of studies conducted in the 1980s (Hampton, 1988), and looked for more direct evidence that individual or group differences in extension would mirror differences in intension. The method used was quite simple. A group of students were asked to make extensional judgments for six semantic categories—in this case, judgments of typicality of a list of category members (exemplars). The same group also made intensional judgments about the same categories, namely, to judge the importance of a list of features for defining the category. Correlations within each set of judgments were then used to produce two similarity matrices for the group of students, showing for each individual a measure of how similar their responses were to each of the other individuals in the group. One similarity matrix was produced for the extensional (typicality) judgments, and another for the intensional (feature importance) judgments, and results were studied to see if they corresponded. If a pair of individuals had similar ideas about which sports were most typical, did they also share similar ideas about what features of sports were most important for deciding membership?

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ’  

73

The strange result obtained was that there appeared to be zero relationship between these two sets of similarities. A pair of participants who had similar ideas about typicality showed no greater similarity than average for feature importance ratings, and vice versa. There appears to be a disconnect between the ways in which people vary in their extensional beliefs and the way in which they vary in their intensional beliefs. Is this null result to be believed? What if the variation between individuals reflects random responding and they all have fundamentally the same concept prototypes? That could then account for the lack of correlation between the similarity matrices. To test this and other challenges, several controls were present in the study. First, people’s judgments were tested again a week or two later and the similarity matrices recomputed. If there is little or no difference between people, and the similarities reflect randomness, then the similarity in typicality (or feature importance) at week one should not correlate with similarity for the same measure at week two. In fact, there was a significant positive correlation across occasions in the similarity matrices. So, the similarity and difference between individuals in each task was stable over time. There are genuine differences in how people make the two judgments, but the differences on the extensional task do not correspond to the differences on the intensional task. A second control was to consider that differences between people may reflect some general strategy for responding to the question. Perhaps different groups interpret each task in different ways, or they use the rating scale in different ways, and so the similarities reflect general effects, and not the contents of people’s prototypes. To answer this possibility, the correlation between the similarity matrices for typicality (or feature importance) was calculated for different pairs of categories. The expectation was that general strategic differences should show up as positive correlations between the similarity matrices for different categories. Results showed that there was indeed a small positive correlation across categories for each task, but this was much lower than the correlation for the similarity matrices for the same category over time, confirming that the stable individual differences related to specific categories. A pair of individuals with similar ideas of rating typicality in sports were much less likely to be similar when rating typicality in another category, e.g. science. Finally, researchers questioned whether the right set of features and exemplars were used to provide a good test of the hypothesis. If the features were not relevant to determining degrees of typicality in the category, that could explain the null result. Using data from the Leuven concept database (De Deyne, et al., 2008; Verheyen & Storms, 2013), results showed that the number of features an item possessed did indeed correlate well with the mean typicality of that item. The correlation averaged .80 based on the twelve features sampled, compared with a correlation of .82 using the full set of twenty-nine to thirty-nine features in the norms. So the features were representative of the full set.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

74

 . 

4.5 A Further Study: Connecting the Two Tasks In an attempt to bring the two aspects of conceptual representation closer together in our data, Hampton & Passanisi conducted a new study (as yet unpublished). This study had 188 participants divided into eight groups of between twenty-one and twenty-five, each on the basis of three between-subjects factors. All participants judged the typicality of exemplars, and the importance of features for two different categories, sports and fish. All participants did all four tasks. Across groups we varied the order of doing the four tasks, and the instructions they were given. The design is shown in Table 4.1. While counterbalancing whether the fish or sports category was done first, the two tasks were always done for one category before moving on to the other category. This procedure was a marked change from the earlier method, where participants performed the first task on six categories before moving on to the other task. This study deliberately had the extension and intension tasks for a particular category placed next to each other, in order to increase the influence of one on the other. In addition, the order of the two tasks for each category was counterbalanced: when the two tasks are set side-by-side in this way, would there be a greater influence of one upon the other? Having just decided which sports are most typical, the expectation was that there would be a better chance that people would then also judge the features of those typical sports more important. If that turned out to be the case, it would show whether the influence was

Table 4.1. Design of the unpublished study by Hampton & Passanisi. Group

First

Second

Third

Fourth

Instructions

A

Typicality Sports Typicality Fish Importance Sports Importance Fish Typicality Sports Typicality Fish Importance Sports Importance Fish

Importance Sports Importance Fish Typicality Sports Typicality Fish Importance Sports Importance Fish Typicality Sports Typicality Fish

Typicality Fish Typicality Sports Importance Fish Importance Sports Typicality Fish Typicality Sports Importance Fish Importance Sports

Importance Fish Importance Sports Typicality Fish Typicality Sports Importance Fish Importance Sports Typicality Fish Typicality Sports

Standard

B C D E F G H

Standard Standard Standard Connected Connected Connected Connected

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ’  

75

symmetrical (extension on intension, and vice versa), or not. For example, if people primarily represent concepts by their exemplars (Storms, 2004) then extension should affect intension more than the reverse. The third factor was the instructions that were given. One set of four groups had ‘standard’ instructions as used in the earlier studies, asking the participants to rank the typicality of twelve exemplars and then to rank the importance of twelve features as two separate tasks. The ‘connected’ instructions gave the instructions for both tasks at the beginning. After instructions for one set of rankings (e.g. exemplar typicality), the booklet then explained that participants would subsequently be asked to do the other (e.g. feature importance). In the version where the feature importance task came first, instructions (translated here from Italian) were as follows: First, we want you to tell us how important you think each property is for deciding whether something is in the category. An important property is one that you find often in members of the category—most members have it, and that is also not so often found in other kinds of thing. So it is distinctive to that category. [An example was given here] After this task, we want you to give us a judgment of how representative or typical you think each word is of its category. A typical example of a category is one that has lots of properties in common with the other members of the category—thus it would be a good example to represent what that category is normally like. Judge typicality in terms of the features that you have chosen as most important for the same category.

Equivalent instructions were given for those doing the tasks in the opposite order, again referring to the first task when giving instructions for the second. We hoped that this combination of the two instructions into a single page before the start of either task would encourage participants to connect the two tasks. In that case, the pairwise similarity matrices for the two tasks may be expected to show a positive correlation. The tasks were administered as paper booklets to a large class of students at Kore University in Enna, Sicily. Five participants were excluded on the basis of strong evidence of collusion with other participants. Reliability (Cronbach’s alpha) for each of the ranking scales was high (mean = .944, range from .83 to .99). The data were analysed to generate similarity matrices for each of the eight conditions for each of the four tasks (extension and intension rankings for sports and for fish). Correlations between the similarity matrices were then calculated. Following an earlier study (Hampton & Passanisi, 2016) this study compared the correlation between similarities for typicality and importance within the same

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

76

 . 

category to the same correlation calculated across different categories. The latter correlation shows the possible effect of general strategic differences between individuals, so the difference between these two correlations will show the extent to which concept-specific individual differences in extension and intension can be mapped. Table 4.2 shows the results for sports and fish and for the two types of instruction. The first thing to note is that the effect of instructions was in the wrong direction. Asking people to think about both tasks at the same time and to coordinate their responses led to reduced correlation between the similarity patterns for the two tasks, particularly for the sports category. Otherwise, there was some evidence here for the within-category correlations to be higher than the between-category control. To get an idea of the statistical accuracy of these figures, an estimate of the standard deviation of the correlations was obtained using a randomization procedure where the participants’ data in the two similarity matrices were set out of alignment. The resulting mean correlation was zero, and the standard deviation was .09. In sum, having participants do the two procedures one after the other for the same category did have some positive effect on the correlation between extensional and intensional similarity. Instructing the participants to deliberately consider one task when performing the other, however, did not have the expected result. If anything, the two tasks were less related under these conditions. The unexpected differences between the two categories calls for a further study using a larger number of categories.

4.6 Implications The results of these experiments represent something of a paradox. There is little reason to doubt that the conceptual categories that we form are based on intensional properties. It is a commonplace since Aristotle that we judge whether a concept applies to an item by considering shared properties. The whole tradition of scientific discovery has been based on the classification of the world on the basis of properties, which are then selected and redefined to provide categories that Table 4.2. Results of the study. Correlations of the Similarity matrices for Exemplar Typicality and Feature Importance rankings, within and between the two categories.

Sports within category Fish within category Between categories (control)

Standard Instructions

Connected Instructions

.25 .29 .16

.03 .24 .12

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ’  

77

optimise our ability to understand and predict the processes involved. The results from Hampton and Passanisi (2016) do not undermine this general principle. In fact, the degree to which exemplars possess the category features was correlated at 0.8 with judgements of typicality. After all, what else could determine whether some activity is a sport or a science other than its having enough of the right kind of properties? What the results suggest is that when one person’s prototype concept differs from another’s, the dimensions of difference for extensional information (typicality and graded membership of categories) are unrelated to the dimensions of difference for intensional information (importance of properties for defining the category). Only by placing the tasks side by side was any evidence found for correspondence in the individual similarities and differences in responses, and that was far from strong. One way to understand this result is in terms of a dual process theory of categorization (Ashby, et al. 1998). It has been hypothesized that the brain has two mechanisms for learning to categorize. One involves a similarity-based associative learning in which people learn to associate category labels with particular regions of similarity space (Gärdenfors, 2000). The other involves the induction of an explicit rule for categorizing on the basis of observed properties. When a category can be readily discriminated with a rule (e.g. members of category A tend to be larger in size than category B) then an explicit rule can be learned. But when two dimensions need to be integrated to determine the category (for example, members of category tend to be larger but also rounder) then associative processes are more likely to be used. We can speculate therefore that when judging typicality of category exemplars, such as whether snooker or wrestling is a better example of sport, people are basing their judgment on an associative similarity between these activities and their general prototype for sport. The position of these items within a similarity space relative to the prototypical centre of the category would be used to make the ranking. On the other hand, when asked to judge whether exercise or skill is more important as a feature of sport, people are retrieving (or attempting to construct) a rule-based representation of the category. They are thinking of what they have heard or read about the value and meaning of sport within a socio-cultural context. Do people enjoy it? Is it healthy? What should count as a sport in general? These two modes of thinking both show systematic and stable individual variation (Hampton & Passanisi, 2016). However, they tap into different parts of the semantic memory system. While intensions and extensions have to be strongly coordinated at a social level (that is, at the level of group data and common understanding), the individual variation is overlaid onto this coordination in a way that is largely independent for the intensional and extensional representational system.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

78

 . 

4.7 A More Positive Result To conclude, I mention some results from a recent study initiated by Farah Djalal and Tom Heyman with Gert Storms at the Catholic University in Leuven, to which I have subsequently contributed (Djalal, Heyman, Storms & Hampton, 2018). In a conceptual replication of Hampton and Passanisi (2016), the tasks for intension and extension were changed. The extension task used was a binary category membership judgment rather than a typicality judgment. For each of eight categories, participants were shown a set of fifteen pictures and had to click on pictures of exemplars that were category members. Prior to this, they completed a property generation task. Rather than rating or ranking the importance of features, the participants were asked to generate their own properties as if explaining the terms to someone who did not understand them. The properties generated were then used for measuring intensional similarity, while correlation of the binary category membership judgments was used for measuring extensional similarity. Initial results confirmed the earlier findings of Hampton and Passanisi (2016). The correlation between the extensional and intensional similarities was close to zero (mean of .03 and .07 for two different ways of computing intensional similarity). However, a different analysis did show a connection between the two sides of concept representations. The properties generated to the category names were combined with the exemplars from the extensional task into an exemplar by property matrix. Separate groups of four students then filled in each matrix putting a 1 if each exemplar (in rows) had each property (in columns), and a 0 otherwise. From this matrix it was then possible to generate predicted values for a graded membership scale, based on the sum of feature applicability scores for each exemplar. Simply put, the more features that an exemplar possessed, then the better member of the category it should be. (As in Rosch and Mervis, 1975, no attempt was made to provide differential weighting to the features). That was indeed the case. The next step was to consider each individual in the original group, and calculate a predicted degree of membership for the set of exemplars based on two separate sets of properties. One set contained those properties that the individual themselves had generated (individual properties) and the other set contained those properties that they did not generate (residual properties). A logistic regression analysis was then used to predict the person’s own categorization of exemplars on the basis of the two different sets of properties. This analysis was run for each individual participant, taking their own set of individual properties (and residual properties) to predict their own categorization judgments. It was found (as expected) that both individual and residual properties played a role in the prediction for any given individual. However, when the data were shuffled so that individual A’s properties were used to predict other

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ’  

79

individuals’ (B, C, D, etc.) categorizations, the weight in the model for individual properties declined, while the weight for residual properties increased. When the individual model parameters were averaged across all the participants in the sample, this pattern was seen in over ninety-eight per cent of 1,000 random shuffles of the data. This analysis is quite complex, and is also different in many ways from the comparison of similarity matrices used before. It involves showing that the properties that a given person generated for a category were more predictive of their own categorization than of other people’s categorization of the exemplars in that category. (Recall also that the extensional measure here was a yes/no membership decision about pictures of exemplars). What has been shown by this new study? While the use of the previous method of analysis—correlating similarity matrices for extensions and intensions— showed the same lack of correspondence as before, a better correlation was found when an individual’s own generated properties were used to predict their own exemplar categorization than when either the properties or the categorization behaviour were from a different individual. It is possible (but probably unlikely) that the use of different intensional and extensional measures was responsible for the positive result this time. It is possible that asking people to generate their own properties rather than rank a fixed set will require them to access a deeper level of personal meaning. It is known that what people say and what people do are often at odds (Nisbett & Wilson, 1977). So, people may explicitly rank a particular property as important on the basis of some higher-level theory or belief that they hold, whereas when it comes to generating their own features they may rely more on their extensional beliefs about exemplars. However, the generation task alone was not sufficient to show a link between intension and extension. When similarity matrices based on generated properties were compared with similarity for extensions, they showed no more correspondence than before. It was only when those properties generated by an individual were used to predict their own categorization of exemplars that a greater level of correlation between predicted and actual categorization was seen. At this stage, the most likely explanation is one of statistical sensitivity or power. It may be that measuring similarity between individuals is not a sensitive enough measure. Reliability of the similarity matrices over time in Hampton and Passanisi varied around .3 to .4, which is quite low, even though clearly greater than zero. Time may tell whether more powerful designs can be found to show a significant effect with this method of analysis.

4.8 Conclusions This chapter set out to show that the two aspects of prototype concept representations, intensions and extensions, can be linked at the level of individual

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

80

 . 

differences. Repeated attempts to show that similarity between individuals for intensions can be mapped to similarity between individuals for extensions have produced null results. Hampton and Passanisi (2016) reported four experiments with no positive result. Our replication with sports and fish showed that when the two tasks are done one immediately after the other, a low level of correlation can be seen, although instructing people to explicitly make the connection had a negative impact (for some reason only for sports). Djalal et al. (2018) confirmed the lack of a correlation between similarity matrices, based this time on a property generation and exemplar categorization task. However, a different analysis, possibly more sensitive to individual variation, was able to show a significant link between a person’s own individually generated properties and their own categorization. How close then is the coordination of different individuals’ conceptual representation? There is clear evidence that people differ from each other, both in judging the typicality and membership of category items, and in judging the importance of, or generating category features. A next step in the research should seek to know how these differences relate to other domains, such as social categories, understanding of political positions, or judgments about crime and morality. How labile are the individual differences that we have seen? They persist over time, but can they be easily influenced by recent experience? In a year when a male British tennis player became World Number One, how does the typicality of tennis change for British, as opposed to American (or Australian), people? From a more philosophical viewpoint, can these differences in prototypes be kept apart from differences (or identity) in beliefs about the essential nature of conceptual categories? Time may tell.

References Ashby, F. G., Alfonso-Reese, L. A., Turken, A. U., & Waldron, E. M. (1998). A neuropsychological theory of multiple systems in category learning. Psychological Review, 105(3), 442–81. Barsalou, L. W. (1987). The instability of graded structure: implications for the nature of concepts. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and intellectual factors in categorization (pp. 101–40). Cambridge: Cambridge University Press. Bellezza, F. (1984a). Reliability of retrieval from semantic memory: Common categories. Bulletin of the Psychonomic Society, 22, pp. 324–6. Bellezza, F. (1984b). Reliability of retrieval from semantic memory: Noun meanings. Bulletin of the Psychonomic Society, 22, pp. 377–80.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   ’  

81

De Deyne, S., Verheyen, S., Ameel, E., Vanpaemel, W., Dry, M. J., Voorspoels, W., et al. (2008). Exemplar by feature applicability matrices and other Dutch normative data for semantic concepts. Behavior Research Methods, 40, 1030–48. Djalal, F. M., Hampton, J. A., Storms, G., & Heyman, T. (2018). Relationship between extensions and intensions in categorization: A match made in heaven? Journal of Experimental Psychology: Learning, Memory, and Cognition, 44, 655–66. Frege, G. (1892/1952). Über Sinn und Bedeutung. Translated in (P. Geach & M. Black, eds.) Translations from the philosophical writings of Gottlob Frege. Oxford: Blackwell. Gärdenfors, P. (2000). Conceptual spaces: The geometry of thought. Cambridge MA: MIT Press. Hampton, J. A. (1979). Polymorphous concepts in semantic memory. Journal of Verbal Learning and Verbal Behavior, 18, 441–61. Hampton, J. A. (1988). Overextension of conjunctive concepts: Evidence for a unitary model of concept typicality and class inclusion. Journal of Experimental Psychology: Learning, Memory and Cognition, 14, 12–32. Hampton, J. A. (1995). Testing the prototype theory of concepts. Journal of Memory and Language, 34, 686–708. Hampton, J. A. (2006). Concepts as prototypes. In B.H. Ross (Ed), The Psychology of Learning and Motivation: Advances in Research and Theory, Vol. 46 (pp. 79–113). London: Academic Press. Hampton, J. A., Dubois, D., & Yeh, W. (2006). The effects of pragmatic context on classification in natural categories. Memory & Cognition, 34, 1431–43. Hampton, J. A., & Passanisi, A. (2016). When intensions don’t map onto extensions: Individual differences in conceptualization. Journal of Experimental Psychology: Learning, Memory, and Cognition, 42, 505–23. Lakoff, G. (1987). Women, fire, and dangerous things. Chicago, IL: Chicago University Press. Levitis, D. A., Lidicker, W. Z. Jr., & Freund, G. (2009). Behavioural biologists don’t agree on what constitutes behaviour. Animal Behavior, 78, 103–10. doi:10.1016/j. anbehav.2009.03.018. McCloskey, M., & Glucksberg, S. (1978). Natural categories: Well-defined or fuzzy sets? Memory & Cognition, 6, pp. 462–72. McNamara, T. P. & Sternberg, R. J. (1983). Mental models of word meaning. Journal of Verbal Learning and Verbal Behavior, 22, 449–74. Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: MIT Press. Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84, 231–59. Putnam, H. (1975). The meaning of ‘meaning’. Minnesota Studies in the Philosophy of Science, 7, 131–93. Rey, G. (1983). Concepts and stereotypes. Cognition, 15, 237–62.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

82

 . 

Rosch, E. R. (1975). Cognitive representations of semantic categories. Journal of Experimental Psychology: General, 104, 192–233. Rosch, E. R., & Mervis, C. B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 573–605. Storms, G. (2004). Exemplar models in the study of natural language concepts. The Psychology of Learning and Motivation, 45, 1–39. Verheyen, S., & Storms, G. (2013). A mixture approach to vagueness and ambiguity. PLoS ONE, 8(5), e63507. Wittgenstein, L. (1953). Philosophical investigations. New York: MacMillan.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

5 Colour Categories in Context Yasmina Jraissati

That categories have a graded structure is well known (see namely the work of Rosch: Rosch, 1973; Rosch & Mervis, 1975; Rosch, 1999). Robins are typical of birds, while ostriches are not. But robins are not typical of birds across all contexts. For Chinese people, swans and peacocks are typical birds; though for Chinese people robins are also birds, they are less typical of birds than swans or peacocks. Categorization, in this sense, has been shown to be context sensitive, where context includes conversations, culture, point of view, and more. From these observations, it was concluded that category structure depended on constraints, not of the category itself, but of specific situations (Barsalou, 1987; Barsalou et al., 2010). That categories are context sensitive is also true of sensory categories, and colour categories more specifically, to the extent that some cross-cultural variations are expected to occur (see Roberson et al., 2000; Cook et al., 2005; Regier et al., 2010). But the role of contextual factors is usually not specified or even explicitly considered in the colour categorization literature. Furthermore, colour categorization studies make assumptions pertaining to the nature of colour categories that are at odds with their context sensitivity. These assumptions also impact colour categorization accounts. This chapter aims to spell out the way in which context affects colour categorization, and how a framework that is based on context sensitivity can offer a powerful account of colour categorization. There is a longstanding interest in why we categorize colour the way we do. This question involves the relationship of language to perception. Also, to the extent that categories are the ‘building blocks’ of cognition (Harnad, 2005), the question of colour categorization more generally serves as a case study to better understand sensory categorization. An important body of empirical work is available, and data has been extensively collected cross-culturally towards providing an answer to this question. Most of the time, participants are shown colour samples in a random order, and asked to name them (or alternatively, to sort them). In this way, the colour lexicon of a given linguistic community is identified, and so are the categories to which the colour terms refer (more on this below). Several inferences can be drawn from such results. Importantly, by comparing how various languages partition the same array of colours, one can either observe similarities or dissimilarities across languages. The results of such comparisons Yasmina Jraissati, Colour Categories in Context In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Yasmina Jraissati. DOI: 10.1093/oso/9780198803331.003.0006

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

84

 

support either the claim that there are colour universals (the so-called ‘universalist’ claim famously supported by proponents of the basic colour terms theory), or the opposing claim that there are no colour universals (the so-called ‘relativist’ claim, famously contended by Whorf (1940) and his disciples). Cross-linguistic similarity (universalism) or dissimilarity (relativism) in colour categorization is differently accounted for. Traditionally, proponents of the relativistic view have argued that there are cross-linguistic dissimilarities, and that these result from language use (Brown & Lenneberg, 1954; Roberson et al., 2000). On the other hand, proponents of the universalistic view have argued that there are no (unconstrained) cross-linguistic variations, and therefore that colour categorization cannot be taken to result from language use (Berlin and Kay, 1969). Specifically, it has been suggested that colour categorization would, in this case, most probably result from cognitive and perceptual mechanisms that are common to human beings (Kay & McDaniel, 1978; Kay et al., 1991; Kay et al., 1997; Kay & Regier, 2007). Both these views however, have important limitations, namely, that it is true there are important regularities across languages, but that there also are irregularities that cannot be ignored. For example, granting that some colour terms are ‘basic’, and the categories they refer to are universal, different languages feature different numbers of basic colour terms. It has been suggested that the number of basic colour terms depends on the industrialization, and therefore the needs, of the linguistic community (Berlin & Kay, 1969; Cook et al., 2005), which suggests an important degree of cross-cultural variation. Also, even in languages featuring the same number of basic colour terms, boundaries and focal points of categories vary. For this reason, recent categorization models have sought to make room for the flexibility implied by linguistic and cultural influences (Regier et al., 2007; Abbott et al., 2016). Does such flexibility also make room for the context sensitivity of colour categories? This question is not explicitly addressed, as the context sensitivity of colour categories is not usually factored in. In fact, colour categorization studies make an assumption regarding the nature of colour categories that conflicts to some extent with context sensitivity. Indeed, as the aim of these colour categorization studies is to understand the mechanisms that determine the sensory categories we have, such studies therefore require that the colour categories in question be identified. The lexical categories thereby identified are taken to correspond to the average lexicon of the community under study, at that moment in time, where the ‘moment’ in question is generally assumed to extend over a given historical period (Kay, 1975, is particularly clear here). Implicit to colour studies in the literature therefore, is the assumption that colour categories are there to be identified in the first place, that is, these categories are there prior to the study. This implies that an assumption of colour categorization studies is that colour categories are rather fixed. They are set at least for a given historical period. Presumably, these categories, whether they result from

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

85

language use or precede it, are somehow represented in long-term memory, and the representation retrieved when required, for example in a colour-naming study. A few important characteristics of colour categories seem to support this assumption. Colour categories have focal points or areas, or prototypes that organize it for some (Kay & Cook, 2005); centres resulting from the category’s boundaries for others (Roberson et al., 2000; also see Regier et al., 2005); or a mixture of both—it was indeed recently suggested that the focal colour results from representativeness, as determined by the extension of the given colour category, ranging over an irregular colour space. That is, categorization results from a trade-off between the high similarity among members of the category, and the low similarity to members of other categories; the focal colour being the most representative of the category, results from its extension (Abbott et al., 2016). In all three cases, categories are organized along a similarity principle and are structured in the following way: naming consensus concentrates around the focal area, and consensus decreases as distance to the focal area increases. The graded structure of the category also has important cognitive implications (typicality effects and categorical perception effects, for example). Thus, as mentioned earlier, colour categories have a structure that is assumed to result from constraints of the category, thereby reifying it further. Connectedly, though colour categories are referred to in the literature, most of the time, lexical colour categories are examined. Whether there are categories independent of lexicon is an important, difficult, and unresolved question (see Kowalski & Zimiles, 2006; Davidoff, 2006; Roberson et al., 1999). Do lexical terms pick up pre-existing non-lexical colour categories, or do categories result from language use? This question is also connected to the nature of the focal colour mentioned in the previous paragraph: historically, those who have argued that focal colours pre-exist lexical categories and organize their structure have also argued that there are therefore reasons to believe that categories pre-existed language use. Furthermore, those who have contended that focal colours rather result from the extension of the lexical category have also contended that there are therefore reasons to believe that categories result from language use. Today, categorization models converge, overcoming what seems to be an oversimplified way of framing the problem (Regier & Kay, 2009). This chapter concentrates on the notion of lexical categories for three reasons. First, the wide corpus of data that is at the heart of the over-century-long discussion is essentially lexical. Therefore, the questions raised here (to wit: how do we lexically categorize colour in our daily interaction with it? And can perceptual and cognitive constraints be reconciled with contextual factors in lexical categorization?) also stem from the data on lexical categorization. Second, in order to provide an answer as precise as can be to the question of context sensitivity in colour categorization, the scope of our reflection realistically had to be narrowed down. Finally, the answer to the question of how we lexically

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

86

 

categorize in context is bound to have implications pertaining to our understanding of the relationship of lexical categories to mental representations, and to the nature of these representations. For these reasons, examining the role of context in lexical categorization strikes us as an important preliminary question. Thus, this chapter keeps the important, but separate, question of the relationship of lexical categories to non-lexical representations open. It briefly examines it in the concluding sections of this chapter since, as we shall see, our take on context sensitive lexical categorization inevitably has implications for colour cognition more generally. In what follows, the notions of ‘colour category’ and ‘lexical colour category’ are used interchangeably, unless otherwise specified. Thus, as mentioned, colour categorization studies seem to assume that lexical colour categories are there to be uncovered. Yet, the context sensitivity of colour categories is difficult to reconcile with a view according to which colour categories are fixed. To this extent, and also because these assumptions shape the accounts of colour categorization, it is precisely this assumption that I am interested in here. In what follows, I suggest we take a new, and somewhat different start. Putting the wealthy colour categorization literature and the corresponding discussion aside, I propose that we genuinely address the following question: how, when confronted with a given object of a certain colour in everyday life, do we categorize its colour? An answer to the more traditional question of why we categorize the way we do, as a result of what mechanisms, follows.

5.1 What Makes All These Objects ‘Red’? 5.1.1 Context Let us consider the following different situations in which one experiences colour: (A) Walking in a garden exclusively planted with roses, I am taken by the colours of these fragrant flowers. Some of these roses I describe as being ‘pink’, some as ‘white’, and some as ‘red’. (B) As I stand on my balcony looking down the street, a ‘red’ car speeds by. (C) At the restaurant, the waiter asks me whether I would like some ‘white’ or ‘red’ wine. (D) In a shop, different, natural-looking, brownish leather wallets are on display, one of which interests me. I say: ‘I would like the red wallet, please’. These common different situations instantiate one of the several possible understandings of the notion of ‘context’ in colour naming that I am particularly interested in examining here—also of interest are the notions of colour constancy

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

87

(e.g. Olkkonen et al., 2010) and relativity of colour experience to surrounding colours, both of which will not be examined here for the sake of simplicity. All of the objects above are labelled ‘red’ without ambiguity. There is no doubt that the term ‘red’ applies to each of these objects. As I identify these specific objects by their colour in these contexts respectively, and say ‘the red rose’, ‘red car’, ‘red wallet’, ‘red wine’, etc., people immediately understand what object it is I am referring to. This rose is typically red, and so is the car, the wine, and the natural leather wallet. Yet, there is also no doubt that the redness of the rose is different from that of the car, the wine, or the wallet. If the surface colours of these different objects were compared, out of context, if they were somehow detached from the object and presented separately, as if on a Munsell paper chip, perhaps only the redness of the car would be close to what we consider to be ‘typically red’ in a colour-naming study. Though these are typically red roses, they are of a darker shade compared to the car’s so-called ‘focal red’. The same is true of the colour of wine, which could be labelled ‘purple’ in a different context, and of the wallet, which could also be labelled ‘brown’ in a different context. What then, makes all of these objects ‘red’? What motivates my labelling them ‘red’? Clearly, beyond these differences, there is something in common to all these surfaces. Perhaps this common property is what we call redness: the property of being ‘red’. But what does that mean? What is this property? And how do we choose redness as the common chromatic property relevant to a particular set of objects? This is the heart of the question of colour categorization. This chapter proposes that what we call the property of redness is a reference point in a structured colour space, the structure of which is context dependent; redness is a relevant property only in this particular context.

5.1.2 Relevant Similarity: Determined by Context Wine could be described as ‘purple’, and the wallet as ‘brown’ (the difference between these cases will be developed in Section 5.3). Presumably, if the speaker labels both these objects ‘red’, it is because these objects bear some similarity to other objects labelled ‘red’. This is not what is in question. Rather, the question is why the speaker chooses to label these objects ‘red’—instead of ‘purple’ or ‘brown’, respectively. If the speaker opts for ‘red’ in these cases, it is doubtless because categorizing these objects as ‘red’ in these particular circumstances is somehow more advantageous. Take the brown or red wallet above, and imagine the following: (i) I am shopping for a wallet, and on display are a few wallets that are all natural leather looking. Seeking to get a specific wallet that I like most, I ask the

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

88

 

salesperson to hand me the ‘red’ wallet, as this shade differentiates it from the other wallets on display. (ii) I am at a café, and several of my friends and I have our wallets on the table. Interestingly, my friends all have reddish (bright red, red-orange, red-pink) wallets. Mine, which I just bought at the shop in (i), being in natural leather is rather of a brownish shade. To differentiate it, I say: ‘The brown wallet is mine!’ Now imagine different circumstances: (iii) I need a bunch of more or less similar wallets for gifts I intend to make to a group of people. At the same shop, I look at the same few wallets in natural leather on display in (i), and say to the salesperson ‘These five brown wallets will do, thank you’. (iv) At the café where my friends and I happen to all have our wallets on the table in (ii) I say: ‘It’s funny how all our wallets are red!’ In the four situations described above, the same object is labelled differently in different contexts, determined both by the colours the object of interest is surrounded by, and what I need to say. It is known that colour categorization is influenced by the colour of surrounding objects; there are both perceptual (Brown & MacLeod, 1997) and semantic (Mitterer et al., 2009; Kubat et al., 2009) effects. Purpose also has an influence—whether I decide to include the wallet’s colour in the category brown, or exclude it, depends on what I need to say and do: do I need to get the whole bunch of wallets, or do I need to get just one? Contextual effects on the category’s inner structure and boundaries have been observed before in categorization (Barsalou, 1987, and in particular Barsalou & Sewell, 1984). To cite one famous example: ‘cow’ and ‘goat’ are judged to be more typical when processed in the context of milking, than ‘horse’ and ‘mule’—which are judged to be more typical when processed in the context of riding (Roth and Shoben, 1983). It was also shown that some of these contextual effects are precisely determined by the anticipated action (Borghi et al., 2004). While there is no doubt that categorization is determined by a relation of similarity to other coloured objects, what I suggest is that perceptual similarity is not, in a sense, an ‘absolute’. Similarity is more or less relevant in a given context. The wallet is either ‘red’ or ‘brown’ depending on its relevant degree of similarity to other ‘red’ or ‘brown’ objects, and on whether I need to differentiate or assimilate it to these other objects (Tribushinina, 2008, Ch. 3). In (i) and (ii), I need to differentiate the object from other similar objects, while in (iii) and (iv) I need to lump them together. Thus, when a certain object is deemed similar in colour to another object, it is deemed similar to a degree that is necessary or sufficient in the given context, that is, depending on the colour of the surrounding objects and on the purpose of the speaker.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

89

5.1.3 A Note on Relevance Of particular interest to us here is one illustration of Sperber and Wilson’s (1998) principle of relevance: Mary refuses Peter’s invitation to go to the cinema on the grounds that she is tired. Although all Mary says is ‘I am tired’, Peter understands that she is tired to a degree that is sufficient for her not to want to go to the cinema. Peter enriches Mary’s answer on the basis of the expectation that Mary aims for relevance when she says that she is tired. Sperber and Wilson articulate the principle of relevance as being both a cognitive and a communicative one. The cognitive principle of relevance is geared towards maximization of relevance, that is, more cognitive effect for less effort, hence Mary being content with just the expression ‘I am tired’. The communicative principle of relevance states that every act of ostensive communication ‘communicates a presumption of its own relevance’ (Sperber & Wilson, 1998), hence Peter’s enriching Mary’s utterance ‘I am tired’, based on his expectation that Mary’s utterance seeks relevance. While the notion of relevance is here used more loosely, it does run, still, along similar lines. When, at the shop (i) I label the wallet ‘red’, my categorization behaviour is guided by what could be considered ‘cognitive’ and ‘communicative’ principles. Labelling the wallet ‘red’ produces more cognitive effect for less effort, to the extent that I identify the wallet I am interested to buy with a simple colour expression. Labelling the wallet ‘red’ also assumes the listener expects my utterance to be relevant. Given that the shades and shapes of the wallets are rather similar, I know that when the salesperson hears me use the label ‘red’ in reference to the specific wallet I need to isolate from the rest, she expects that I maximize the difference between the wallet I am interested in and the wallets surrounding it to help her identify it. Conversely, in (iii), when I label the same wallet ‘brown’, I know she expects me to minimize the difference between all the wallets on display. In this sense, colour categorization depends on a relevant degree of similarity between colours, where relevance is determined by context, here defined by surrounding colours and purpose of the speaker.

5.2 The Structured Nature of Colour Experience 5.2.1 Similarity Relations in Colour Experience Clearly, that is not to say that similarity is all determined by relevance. If among the five wallets on display, one was a fully saturated bright blue, I wouldn’t have been able to say: ‘these five brown wallets will do’, since in these viewing conditions there is no way in which the fifth uniformly blue wallet could be labelled ‘brown’. Though there is an important sense in which context affects judgment of

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

90

 

perceptual similarity, the latter remains ultimately determined by our biological mechanisms. These relations of similarity result in part from the overlapping sensitivity ranges of our receptors. Humans usually have three cones: L, M, and S cones, respectively sensitive to long, medium, and short wavelengths. There are more or less important overlaps between these three ranges. Though there is no possible linear transformation of receptors’ signals to perceptually uniform representation, the sensitivity ranges across receptors and their overlaps participate in our experience of colour similarity. Thus, if the speaker labels both the wine and the wallet ‘red’, it is because these objects bear a relevant degree of similarity to other objects labelled ‘red’, where similarity is not only ultimately biologically determined, but also context dependent. However, this does not answer the question why specifically redness emerges as a property that determines categorization behaviour? Clearly, to the extent that objects are labelled X because of their relevant degree of similarity to other objects labelled X, categorization depends on the colour of the objects under consideration. However, this constraint alone is not sufficient. It may explain why some chromatic property is not used to categorize the colours of a given set of objects. But it does not explain why specifically this chromatic property X is used to categorize the colours of a given set of objects. Also, assuming there were more objects of various colours to categorize: what degree of redness is required in a given object for it to be labelled ‘red’? How do we determine the degree at which a given object should be similar to other objects for it to carry the same label (for a related discussion, see Hansen & Chemla, 2017)? Part of the answer to this question rests on the fact that when I use the label ‘red’ in reference to the colour of an object, I am importantly not using a different colour label. In choosing the label ‘red’ for a given object, I imply that this object is not, for example, ‘blue’. It is understood that the objects to which I apply the term ‘red’ are different from some other objects that are not labelled ‘red’. Such objects are not labelled ‘red’ because other more adequate terms from my colour lexicon apply. Thus, specifically this chromatic property labelled X is used to categorize the colours of a given set of objects, in contrast to other chromatic properties that carry different labels. Similarity relations alone do not tell us the degree of redness required for an object to be labelled ‘red’; they do not offer limitations to what can be labelled ‘red’. What is also required is the connected notion of dissimilarity.

5.2.2 Structured Similarity and Dissimilarity Relations I experience some colours as being similar to greater or lesser degrees, but I also experience colours as being most dissimilar. This experience of maximal dissimilarity organizes colour experience beyond relations of similarity, into a structured

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

91

whole. This chapter’s proposal is precisely that such a structure ultimately contributes towards accounting for the property of redness (Section 5.2.3). Meanwhile, a few words on structured similarity and dissimilarity relations. The notion of orthogonal dimensions in the case of colour is not new, and the opponent nature of colour experience has a long history in colour science. Starting at least with Goethe (1970), and his interest in after-images, the idea that some colours are ‘opposed’ to each other made its entry in the literature. This notion of opponency implies that not only are there similarity relations, but also that there are opponent relations: hues cannot only be arranged rather smoothly on a rainbow-like gradient—they also can be organized in a circle, determined by orthogonal axes at the poles of which respectively lie opponent colours. Any two colours diametrically opposed on this circle are maximally dissimilar. The idea of colour experience being structured along orthogonal dimensions is mostly associated to Hering (1964), who argued that some colours are unique while others are binary (in other words, some colours are perceived as being simple while others are perceived as being mixtures). Hering’s proposal is based on introspection, and rests on his impression that while orange is experienced as a mixture of yellow and red, there is no sense in which red is experienced as a mixture of orange and purple. Hering argued that this was true of four chromatic sensations: red, green, yellow, and blue. Furthermore, while I can experience a reddish yellow, or a bluish red, I cannot experience a yellowish blue, or a reddish green. Thus, according to Hering, the dimensions that structure colour experience are none other but red/ green, yellow/blue, and black/white. Two of the dimensions, red/green and yellow/ blue, are opponent, the third achromatic dimension black/white is not, in the sense that I can represent grey as a mixture of black and white. This means that any colour is more or less similar to white or black (i.e. is bright or dark), to yellow or blue, and to red or green. So, for example, according to Hering, a bright orange is a point in that space that lies somewhere in the red-yellow quadrant, closer from white than from black. Given that white and black are not opponents in this system, each colour also bears a similarity to their mixture, which is grey, at various degrees of brightness. This is the third dimension that usually goes by the name of chroma (more commonly ‘saturation’. Chroma and saturation are not exactly the same property, but for our purpose this distinction is not critical, and we use these two notions interchangeably). Thus the perceptual distance of the bright orange point to grey will determine the extent of its saturation.

5.2.3 Redness Results from a Relevant Similarity Degree to Other Objects and to One Specific Part of Structured Space I suggested above that if the speaker labels both wine and the wallet ‘red’, it is because these objects bear a relevant degree of similarity to other objects labelled

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

92

 

‘red’. As we have seen, however, this characterization of the categorization behaviour raises the question of why, in the first place, the property of redness determines the categorization behaviour. We have also seen that maximal dissimilarity, or orthogonal relations, structure colour experience. In response to the question of why redness determines categorization behaviour in the first place, I therefore adjust this first claim and suggest that both wine and the wallet are labelled ‘red’ because these objects bear a relevant degree of similarity to other objects labelled ‘red’, and to one pole of our structured colour space. It should be noted that similarity to the ‘red’ pole is differently constraining than similarity to other ‘red’ objects. What defines the ‘red’ pole is not only its own chromatic characteristics, but also importantly its dissimilarity to the ‘blue’ and ‘yellow’ poles, and its maximal dissimilarity to the ‘green’ pole. Take the wallet example above. The ‘redness’ or ‘brownness’ of the wallet in the cases (i) and (iii), is determined by a relevant degree of similarity to other ‘red’ or ‘brown’ objects, with relevance resulting from context. However, this variation from ‘brown’ to ‘red’ is constrained: given the English lexicon used by the speaker and listener, and in similar viewing conditions, there is no context in which that wallet can be described as being ‘yellow’, ‘green’, or ‘blue’. The terms ‘blue’ and ‘yellow’, and mostly ‘green’, refer to colours that are maximally dissimilar from the colour of the wallet in that colour space structured by the three axes blue/yellow, red/green, and white/black. Thus, granting that these three axes structure colour experience, the term ‘red’ refers to that area at the pole of one of the axes. The boundaries of this area are determined by a relevant degree of similarity to the pole English speakers label ‘red’, and to relevant degrees of dissimilarity to the poles English speakers label ‘blue’, ‘yellow’, and ‘green’. Specifically, take the colours extending from the ‘red’ to the ‘yellow’ pole, for example, gradually decreasing in redness and increasing in yellowness: there will be a point at which the similarity to ‘yellow’ will be such that it is no longer informative for me to categorize this colour as being ‘red’. Therefore, when labelling an object as being of a certain colour, I am referring to a region of the structured colour space, which means that I am importantly not referring to other regions of that colour space. Thus, when I label an object as being ‘red’, I am importantly differentiating it from the ‘yellow’, ‘blue’, and ‘green’ areas of the structured space. What makes all of the things relevantly similar to one given pole is both their similarity to this pole and their dissimilarity to the other poles of the structured space. In other words, the property of redness, as we call it, is a characteristic a set of colours have in common, and that they have in common in contrast to other colours that do not have it. These other colours are non-‘red’ because they in turn have a different property in common. This is what the notion of a space structured by three axes boils down to: Critically, the property of X-ness, is to be understood from ‘the point of view of the whole rather than of a single part’ (Garner, 1974,

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

93

p. 2). The property of X-ness will vary depending on the structure of the space. That is, the property of X-ness depends on how many, and to which colour properties X is contrasted.

5.2.4 Structure: ‘The Point of View of the Whole Rather Than of a Single Part’ Garner’s work sheds an interesting light on this notion of structured colour space. Imagine a set of eight objects, varying in shape (elliptical or diamond), orientation (top tip oriented to the left, or the right), and size (big or small). These three dimensions (shape, orientation, and size), with their two levels (elliptical or diamond; left or right; big or small), represent all the ways in which each stimulus can vary. Thus, given a total set of stimuli, based on the relations of similarity between the stimuli, definite dimensions will emerge. All the shapes within this total set are similar with respect to them being ellipse- or diamond-shaped, big or small, pointing to the left or to the right. The dimensions of a given set of stimuli exhaust all possibilities in that set. In this structured space, a given shape cannot be both diamond and elliptical; it cannot both point to the left and to the right; it cannot be both big and small. The dimensions of the set represent the features, or properties of the stimuli, and therefore the way in which they individually vary. Thus, a given shape in the set provides information in terms of the properties of the set that it instantiates, and those that it does not. Objects lying at the poles of the dimensions of the set are experienced as most dissimilar. Garner’s approach to information and structure can be applied to colour. According to Garner, given a total set of stimuli, dimensions along which the individual stimuli can be compared and vary are extracted. These dimensions structure the set and in it consists the information that the individual stimuli provide. In other words, the way in which individual stimuli instantiate or not each pole of the dimension, or the way in which each stimuli relates to other stimuli of the set, is the information that perceivers extract from individual stimuli. Interestingly, as in the case of the set of objects above, it can be argued that we also have a ‘total set’ of colours, in the sense that there is a limited set of colours we can and do perceive throughout our lifetime, made available by our sensory apparatus. The colour we experience results from the interaction of light sources, with surfaces, and our sensory apparatus. The electromagnetic radiations emitted by light sources, strike the surfaces of our environment. These radiations are modified by the surfaces’ microstructures, or reflectance properties. Some of these radiations are absorbed by the surfaces, while others are reflected. Our visual system captures the reflected radiations and processes the information that we experience as the surface’s colour. Because of the nature of our sensory apparatus,

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

94

 

we only experience colour when the wavelengths of these radiations are between 400 and 700 nanometres. Thus, given the limited number of light sources we encounter, and given the determined structure of our receptors, and their limited number (usually three types of cones and one type of rods), our sensory apparatus makes available to us a limited number of colours throughout our lifetime. True, we can discriminate a great number of colours, but given the sensory and cognitive apparatus with which we are endowed, the boundaries of the set of colours we can experience is fixed: it cannot expand. Thus, given that there is a total set of colours we can and do experience, we should be able to extract dimensions of this set along which individual colours vary, and that structures our experience of colour, in which consists the information colour provides. Note that there are important differences between the diamonds and ellipses example above, and colour. Importantly, colour is not a complex visual object featuring a number of attributes. It is a simple experience, and an attribute of objects. Also, colour is experienced as continuous, in contrast to objects that are discrete entities. Applying Garner’s view to colour: if the colours X and Y lie at the poles of one axis, and the colours p and q lie at the poles of the other axis, their being orthogonal means that one single colour cannot be X and Y or P and Q at the same time. Like in the case of the shapes above, a given shape can be a diamond or an ellipse, as these two shapes are the two alternative levels of one of the dimensions of the set. Any given shape cannot be both a diamond and an ellipse at the same time. Now we can imagine that in the example above, there are many more than eight shapes. We can imagine that ellipses and diamonds are the extreme cases of a range of possible objects that gradually change from a diamond to an ellipse shape. In this case, instead of discretely instantiating or not a given level of one dimension, a given, discrete, object will more or less instantiate a given level of one dimension. In other words, a given object will be similar to a given level of one dimension to a certain degree. This seems to be the case in colour, where any given colour bears similarity relations to a certain degree to P or Q, and X or Y at the same time.

5.3 About Redness 5.3.1 A Context-dependent Colour Space Structure Thus, when labelling the wallet ‘red’, I do so in virtue of a relevant degree of similarity to other ‘red’ objects, which together bear a relevant degree of similarity to one pole of structured colour space—and dissimilarity to the other poles of that colour space. The structure of colour space results from the dimensions along

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

95

which the total set of colours varies, and constrains my colour categorization behaviour. Importantly, when labelling an object ‘red’, I refer to one pole of colour space, and do not refer to other poles of colour space. This raises the important question of the origin of such a structure. The red/ green, yellow/blue, and black/white axes have been identified as dimensions structuring the total set of colours. Where do these specific dimensions come from? Are they determined by biological mechanisms? The answer is yes, and no. Specifically, I argue that, to the extent that these dimensions are not biologically determined, they are also context dependent. In other words, not only would we categorize objects based on a relevant degree of similarity to other objects in relation to a pole of the structured colour space, where relevance is determined by context as explained above, but the structure of the space would itself be also context dependent. The following explains the sense in which the structure of colour space is biologically constrained and context dependent. Since the middle of the twentieth century, we have known that some of the cells processing light information are opponent at post-receptor (De Valois et al., 1966; Abramov & Gordon, 1994) and cortical (Shapley, 2011) levels: cells that are excited by a signal are inhibited by another kind of signal. Some cells are also double-opponent, with a centre and a surround that are excited and inhibited by opposing signals. So, if the centre of such a cell is excited by a signal, its surround is inhibited by it, and vice versa. This mechanism that is arguably responsible for our edge detection is presumably also why we experience some colours as maximally dissimilar. It is assumed that the orthogonal structure of our colour experience results from this characteristic of visual processing of light. However, unlike what was long conjectured, this mechanism is not responsible for any particular orientation of the opponent axes (see, for example, Mollon, 2006; Abramov & Gordon, 1994). Colour processing is highly non-linear, and the responses of opponent and non-opponent cells yield various spaces: no perceptually uniform representation can result from linear transformations of receptor’s signals (Horwitz & Hass, 2012; Parraga & Akbarinia, 2016). Thus, specifically, this opponent mechanism is not what determines unique hues. The latter do not correspond to any known biological mechanism of colour vision. This means that, while our experience of colour is indeed orthogonally structured for biological reasons, the nature of these orthogonal axes is not. Why then, do these specific dimensions, instead of others, emerge? I suggest that the nature of these axes is precisely what is context dependent.

5.3.2 Context Dependent Sub-spaces of Colour Take the case of wine. Despite its rather purplish shade, wine is never labelled ‘purple’ in English. What makes it the case that purplish wine is always labelled

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

96

 

‘red’ in English? Specifically, wine is labelled either ‘white’, ‘red’, or ‘rosé’—a French word for pinkish. In fact, wine comes in a variety of colours, which, in a different context, could be described as ranging from ‘bright yellow’, through ‘bright yellowish pink’, to ‘dark reddish purple’. Yet, and regardless of the available English colour lexicon, English speakers, when it comes to wine, will not categorize its colour in terms of ‘yellow’ or ‘purple’. In the case of wine, English speakers label its colour ‘white’ and ‘red’, respectively. When the speaker labels a wine ‘red’, she knows the listener will identify what wine she is referring to. As the expression ‘red’ is applied to wine, the ‘presumption of relevance’, to borrow the notion from Sperber and Wilson (1998), implies that the listener and speaker both attend a specific subset of colours, namely wine colours. The case of wine reveals that we do not always attend the total set of colours made available to us by our perceptual system throughout our lifetime when we categorize. In some cases, only a specific subset of colours that is directly concerned by what the speaker needs to say or do is attended to. If, with Garner, we accept the view that the dimensions extracted for a given set of objects depends on the set of objects attended to, then, in the case of wine, we attend a subset of colour and therefore extract context-dependent dimensions of this subspace. Why, then, would speakers label the dimension of wine-colour sub-space white/red rather than purple/yellow? Why would the dimension of wine-colour subspace carry labels usually used in reference to colours that are precisely not in the set of colours attended to? The labels ‘white’ and ‘red’ applied to the subspace of wine have different extensions than the labels ‘white’ and ‘red’ applied to the total colour space. For one thing, wine, which is a transparent medium, is labelled ‘white’, while in English, ‘white’ is usually used exclusively for non-transparent objects (as famously argued by Wittgenstein, 1977, also see Casati, 2009). To the extent that structure is a complex system ‘considered from the point of view of the whole rather than of a single part’ (Garner, 1974, p. 2), the labels ‘white’ and ‘red’ applied to wine should be understood in light of the contrast they mark in the subset of colours attended to. It could be that ‘white’, when applied to wine, precisely marks the absence of the chromatic property observed in the other kinds of wine (labelled ‘red’ in English), rather than what we consider to be whiteness in total colour space. In this respect, there is a relevant similarity between such wines and whiteness: in comparison to other purplish wines, such wines are bright and poor in colour. In this sense, they are relevantly similar to the pole of total colour space that is labelled ‘white’. In the same way, the pole of structured total colour space to which purplish wines bear the most relevant similarity is labelled ‘red’. It would seem that what best characterizes the dimension of wine-colour subspace is the contrast labelled white/red in English. It would seem this specific contrast, presumably between the highly coloured nature of so-called ‘red’ wine, and the poorly coloured nature of the so-called ‘white’ wine, is most relevant to English

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

97

speakers. Given the limited variety of colours in that subset, there is no need for finer-grained labels. Another interesting illustration is skin colour lexicon. Though all possible skin colours are shades of what we could call ‘pink’, the label ‘pink’ is rarely used in reference to the colour of skin. Like in the case of wine, instead of using labels that are available in the English lexicon to refer to the colour of human skin, speakers of English will extract dimensions of skin-colour subspace, and rely on the similarity relation between these dimensions and those of total colour space. It would seem that the labels ‘white’ and ‘black’ capture the contrast between skin colours that is most relevant to English speakers. The ‘white’ and ‘black’ labels applied to skin colour presumably emphasize the brightness of some skins, in contrast to the darkness of others, knowing that, in total colour space, white/black is the achromatic dimension along which brightness is taken to vary. When compared to other objects usually labelled ‘white’, skin is not white. Similarly, when compared to other objects usually labelled ‘black’, skin is not black. Nevertheless, so-called ‘white skin’ is more similar to ‘white’ than so-called ‘black skin’; and the way in which ‘white skin’ differs from ‘black skin’ is comparable to one way in which other ‘white’ objects are taken to differ from other ‘black’ objects in total colour space (Gardenförs, 2000). That is: they differ in brightness. This choice of labels suggests that this is precisely the difference between skin colours that the English speaker intends to mark. What contrast speakers want to mark in the subset of colours they choose to attend to depends on their purposes. The notion of ‘subspace’ is similarly a useful way to characterize what takes place in labelling the wallet ‘red’ or ‘brown’: As I label the wallet ‘red’, I do so by attending a subspace of colours, i.e. the colours of the wallets displayed at the shop. As mentioned, the structure of colour subspace depends on surrounding colours and purpose of the speaker. In this case, the purpose is to differentiate one wallet from a set of rather similar wallets: a red/brown dimension is extracted that serves this purpose and marks a contrast, which, in other contexts, is completely absent. Unlike the case of wine, the rather similar shades of the subset of wallet colours calls for labels that are of a finer-grain than the labels carried by the poles of total colour space (Tribushinina, 2008). There is a difference though, between wine or skin colour and the wallet case: in the case of the wallet, the subset is determined by the co-occurrence in space and time of several wallets. In the case of wine and skin, the surrounding colours, or the other colours of the subset that are attended to, are not necessarily cooccurring in space and time. In the case of wine and skin, surrounding colours are rather the colours of a set of objects to which the given object is considered to belong. What this naturally implies is that purpose also partly determines the subset of colours, or objects, attended do.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

98

 

In sum, according to the framework sketched in this paper, the suffix ‘-ness’, in the ‘redness’ of roses or of wine and the ‘whiteness’ of skin, refers to a relevant similarity degree instantiated by a given object to a pole in subcolour space, and to relevant dissimilarity degree to the remaining poles of subcolour space. The subset of colours attended to and the extracted dimension(s) depend on the speaker’s purpose. Thus, though partly biologically determined, the structure of colour subspace is flexible so as to accommodate various contexts. Presumably, the extent to which the different subspaces are themselves flexible is a matter of degree.

5.3.3 Cross-cultural Differences and Context For example, though the subset of wine colours the speaker attends to rarely varies, the contrasts the speaker wants to mark does. Purplish wines that English speakers label ‘red’, are labelled ‘black’ in Catalan (‘vi negre’).* We have hypothesized that the red/white axis in the English subspace of wine colour marks the highly coloured versus poorly coloured nature of the various wines. In Catalan, it seems that the contrast speakers intend to mark is not how highly or poorly coloured the wines are, but how dark or bright they are. In Catalan then, speakers choose to use the contrast marked by the labels white/black, the poles of the axis that marks the brightness contrast in total colour space. According to this view, then, the highly or poorly chromatic nature of the wine is relevant to English speakers in the subset of wine colour, while the brightness or darkness of the wine is what is relevant to Catalan speakers. These different linguistic and cultural communities choose to mark different contrasts in the same colour subspace, and are a good illustration of the way purpose determines its structure. These considerations bring us back to what we have called ‘total colour space’, or the space that results from our attending to all colours that it is possible for us to perceive—regardless of whether they co-occur in space and time. We have seen that no biological mechanisms from which these specific three axes (white/black, red/green, blue/yellow) could result were identified. The question of why these axes emerge remains open. According to the view defended in this chapter, these dimensions would result from context. Specifically, these dimensions emerge because speakers choose to attend a given set of colours (the total set of colours in this case), and, importantly, choose to mark similarities and contrasts in this total set that fit their purpose. For example, yellow and blue correspond to the two predominant illuminants in the world: yellow is the colour of direct sunlight, and blue the colour of skylight. It is possible that the yellow/blue axis results from this particularity of natural

* I thank Nat Hansen for this useful comment.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

99

illuminant. It indeed appears that the colour of sunlight, and the geographically opposing colour of northern skylight vary along a line running from unique blue at 476 nm and unique yellow at 576 nm, in the MacLeod–Boynton chromaticity diagram (which represents human cone excitation; see Mollon, 2006). From this perspective, the yellow/blue axis would result from speakers attending the total set of colours, as well as attending the contrast in the predominant colours of light— while the red/green axis could, for example, result from its being orthogonal to the first (for a different ecological account of the unique hues see Mollon, 2006; Shepard, 1992; Philipona & O’Regan, 2006; but see Witzel et al., 2015). Thus, there is no clear-cut answer as to why the colours labelled ‘red’, ‘green’, ‘yellow’, and ‘blue’ have a particular status in our relation to colour. In the absence of physiological mechanisms that could account for unique hues and the orientation of the orthogonal axes, characteristics of the colour of light might account for why these contrasts are marked in certain cultures. Such an ecological approach provides support to a context-dependent view on the structure of colour space, to the extent that contextual (ecological) factors are what determine the orientation of the orthogonal axes of colour space. This implies that the specific orientation of the orthogonal axes is not necessary, in the sense developed above that different contrasts can be attended to in different contexts. The history of colour science is another valuable indication of how colour experience can be represented as varying along different dimensions: ranging from Aristotle’s ordering of colours from white to black along a single dimension, to the appearance of colour circles in the seventeenth century, Munsell’s three-dimensional sphere, and other numerous contemporary colour models (Westfall, 1962; Parkhurst & Feller, 1982; Gage, 1993; Kuehni, 2007. For a connected argument, see Chapter 6 by Adams and Hansen in this volume, Section 6.7). That is not to say that the axes of total colour space cannot be oriented in the same way across cultures. But if they are, the universality of the axes’ orientation should be traced back to the universality of contextual factors (characteristics of sunlight, for example), and of the contrasts speakers of these different communities need to mark.

5.3.4 Colour Categories Are Not Set Entities Are the dimensions of total colour space cross-culturally the same then (albeit for contextual factors), or do they vary? If these dimensions are always the same, and given that lexicons vary, this would mean that the poles of structured colour space are not always labelled. This brings us back to the question we have so far willingly avoided regarding the relationship of lexical colour categories to non-lexical representations of colour. We have seen above that when I label a wallet ‘red’ in English, I mark the similarity this object bears to other objects and to a specific pole of structured

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

100

 

colour space. In doing so, I importantly do not label it ‘green’, ‘blue’, ‘yellow’, ‘white’, ‘black’, etc. What about languages that do not have such terms? There is data on languages featuring mostly two colour terms (Dani is possibly one such language, Rosch, 1972), partitioning the space into roughly ‘bright’ and ‘dark’ areas; or languages featuring three, four, or five terms (MacLaury, 1992; The World Color Survey, http://www1.icsi.berkeley.edu/wcs/). These cases are different from English in at least one important respect: they do not have terms that individually refer to the six poles English speakers take total colour space to have. In languages where the wallet would be described as being ‘dark’, it would importantly not be described as ‘bright’. It would be understood that the colour of the wallet is either (to borrow from the English lexicon) blackish, brownish, dark bluish, dark greenish, dark purplish; but it is not ‘bright’, that is: it is not yellowish, reddish, or whitish. Does this mean that speakers of such languages do not represent the space as being structured by the yellow/blue, red/green, and black/ white dimensions? They certainly do not represent this structure in language and categorization behaviour. Do they not represent it in thought either? A precision on the nature of colour categories is first required. In order to spell out the role of context in lexical colour categorization as precisely as can be, this chapter focuses on the notion of category in a very narrow sense. The working hypothesis has been that lexical colour categories result from the psychological process of classifying properties and labelling them. Thus, in a narrow sense, lexical colour categories are sets of properties that are experienced in the world. Why we take a given colour to belong to category X, and why we have a category X in the first place, are precisely the questions in which we are interested here. The suggestion is that colour categorization is determined both by context (objects attended to and purpose) and the structure of colour space, which is itself both biologically determined and context dependent. In this sense, colour categorization is both variable and constrained. Context is where variability in colour categorization comes from, while colour similarities and maximal dissimilarities, or the fact that colour space is structured, constrains colour categorization. This implies that, unlike what seems to be assumed in the colour categorization literature, colour categories are not fixed. Indeed, as the introduction to this chapter shows, colour categorization studies aim at ‘identifying’ the colour categories common to a given linguistic community. In this sense, colour categorization studies seem to imply that colour categories are rather set entities: their extensions within a linguistic community are fairly stable (with some interpersonal variations, Webster & Kay, 2007; Kuehni, 2004) over a given historical period. However, if as argued above, categorization is context dependent, colour categories are best understood as constructed online (for a connected study on colour memory, see, for example, Lucy & Shweder, 1988). Note that this view is not completely at odds with the widespread assumption in colour categorization studies. According to the view developed here, the difference

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

101

between colour categories that are more permanent than those that are more temporary in categorization behaviour is a matter of degree. To the extent that some colour sets are often attended to together (i.e. the total set of colours; the colours of wine; the colours of skin), and some contrasts are generally marked, some colour (sub)spaces manifest a stable structure, which in turn yields more permanent categories, provided the purpose of the speaker is also stable. Thus, some categories are more permanent than others because the context in which they are created is more stable and occurs more regularly. Assuming that lexical colour categories are not set entities and are constructed online as we have offered here: what is their relation to thought? How are they represented in thought? Conversely, are all representations lexicalized? Though in a different domain, it has been shown that not all mental representations correspond to lexical categories (Malt et al., 2015). Applied to the case of colour, this could mean that, despite the fact that some languages do not linguistically mark the same contrasts as in English (between ‘red’ and ‘green’, ‘blue’ and ‘yellow’, ‘white’ and ‘black’), these languages could still represent these contrasts, and therefore these dimensions, in thought. This chapter being essentially based on lexical categories does not offer an answer to this question. However, what this chapter has shown, I hope, is the dynamic nature of colour cognition. This is bound to affect the way the question of the relation of language to thought is framed in the case of colour: that the dimensions of colour space and subspaces, which determine the structure of colour lexicon and lexical categories, are extracted in context. What this implies is that some categories that occur more regularly and are more permanent might be stored differently in memory than categories that occur very rarely, or even uniquely (see for example Barsalou’s ad hoc categories: Barsalou, 1983; Barsalou, 1987; also see Barsalou, 1999). In other words, different categories might be differently represented in thought. More to the point of cross-cultural differences: the approach to colour categorization presented here makes room for the possibility of cross-contextually varying mental representations of colour space structure. Whether this is indeed the case is an empirical question. The framework sketched in this chapter sought to delineate the way context determines colour categorization. Indeed, though often considered a factor, the way context is supposed to influence colour categorization is never spelled out in the colour categorization literature. The account of context sensitivity presented here resulted in an approach to colour categorization that accounts both for crosscultural regularity and variability. It should be noted that the view presented here uses several notions that are familiar to the reader (importantly, conceptual spaces; see Gardenförs, 2000, but also context sensitivity in graded membership, context sensitivity in language use, etc.), with the important difference that this unified framework was developed with the particular case of colour in mind. It therefore specifically addresses the

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

102

 

question of the role of context in categorization in relation to the assumed role of biology and perception. Whether such a framework offers a better account of colour categorization than the views it combines remains to be determined. Meanwhile, I hope to have offered a convincing alternative take on colour cognition, useful both for philosophical and psychological concerns. One outcome of this view that would be interesting to examine in the future is the role of objects in colour cognition: we normally do not encounter abstract colour samples in our daily life. We use colour labels in reference to a variety of objects, and as argued here, in the context of sets of coloured objects. What determines these sets are our purposes which themselves depend on the nature, shapes, and functions of these objects. Finally, to the extent that context is expected to play a role in sensory categorization, the generalizability of this account to other sensory domains is an interesting question. Whether the framework introduced here has the required explanatory power is another question worth pursuing in future work.

References Abbott, J. T., Griffiths, T. L., & Regier, T. (2016). Focal colors across languages are representative members of color categories. Proceedings of the National Academy of Sciences, 113(40), 11178–83. Abramov, I. & Gordon, J. (1994). Color appearance: On seeing red—or yellow, or green, or blue. Annual Review of Psychology, 45, 451–85. Adam, Z., & Hansen, N. (2020). The myth of the common sense conception of color. In A. Wikfross, & T. Marques (Eds.), Shifting concepts. Oxford: Oxford University Press. Barsalou, L. W. (1983). Ad hoc categories. Memory & Cognition, 11(3), 211–27. Barsalou, L. W. (1987). The instability of graded structure: Implications for the nature of concepts. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and intellectual factors in categorization (pp. 101–40). Cambridge: Cambridge University Press. Barsalou, L. (1999). Perceptual symbol systems. Behavioral and Brain Sciences, 22, 577–609. Barsalou, L. W., & Sewell, D. R. (1984). Constructing Representations of Categories from Different Points of View. Technical Report. Emory University, Atlanta. Barsalou, L. W., Wilson, C. D., Hasenkamp, W., Mesquita, B., Feldman Barrett, L., & Smith, E. (2010). On the vices of nominalization and the virtues of contextualizing. In: B. Mesquita, L. Feldman Barrett, & E. R. Smith (Eds.), The mind in context (pp. 334–60). Guilford Press: New York. Berlin, B., & Kay, P. (1969). Basic color terms: Their universality and evolution. Berkeley, CA: CLSI Publications.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

103

Borghi, A. M., Glenberg, A. M., & Kaschak, M. P. (2004). Putting words in perspective. Memory & Cognition, 32(6), 863–73. Brown, R. & Lenneberg, E. (1954). A study in language and cognition. Journal of Abnormal and Social Psychology, 49, 454–62. Brown, R. O. & MacLeod, D. I. A. (1997). Color appearance depends on the variance of surround colors. Current Biology, 7(11), 844–49. Casati, R. (2009). Are shadows transparent? An investigation on white, shadows, and transparency in pictures. RES: Anthropology and Aesthetics, 55(1), 329–35. Cook, R. S., Kay, P., & Regier, T. (2005). The World Color Survey database: |History and use. In H. Cohen & C. Lefebvre (Eds.), Handbook of categorisation in the cognitive sciences (pp. 224–42). Amsterdam: Elsevier. Davidoff, J. (2006). Color terms and color concepts. Journal of Experimental Child Psychology, 94, 334–8. De Valois, R. L., Abramov, I., & Jacobs, G. H. (1966). Analysis of response patterns of LGN cells. Journal of the Optical Society of America, 56, 966–77. Gage, J. (1993). Colour and culture: Practice and meaning from antiquity to abstraction. Berkeley, CA: University of California Press. Gardenförs, P. (2000). Conceptual spaces: the geometry of thought. Cambridge, MA: MIT Press. Garner, W. R. (1974). The processing of information and structure. New York: Psychology Press. Goethe, J. (1970). Theory of colours. Cambridge, MA: MIT Press. Hansen, N. & Chemla, E. (2017). Color adjectives, standards and thresholds: An experimental investigation. Linguistics and Philosophy, 40(3), 239–278. Harnad, S. (2005). To cognize is to categorize: Cognition is categorization. In H. Cohen & C. H. Lefebvre (Ed.), Handbook on categorization. Amsterdam: Elsevier. Hering, E. (1964). Outlines of a theory of the light sense. Cambridge, MA: Harvard University Press. Horwitz, G. D. & Hass, C. A. (2012). Nonlinear analysis of macaque V1 color tuning reveals cardinal directions for cortical color processing. Nature Neuroscience, 15(6), 913–19. Kay, P. (1975). Synchronic variability and diachronic change in basic color terms. Language in Society, 4, 257–70. Kay, P. & McDaniel, C. K. (1978). The linguistic significance of the meanings of basic color terms. Language, 54(3), 610–46. Kay, P., Berlin, B., & Merrifield, W. (1991). Biocultural implications of systems of color naming. Journal of Linguistic Anthropology, 1(1), 12–25. Kay, P., Berlin, B., Maffi, L., & Merrifield, W. (1997). Color naming across languages. In C. L. Hardin & L. Maffi (Eds.), Color categories in thought and language (pp. 21–56). Cambridge, MA: Cambridge University Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

104

 

Regier, T., Kay, P., & Cook, R. S. (2005). Focal colors are universal after all. Proceedings of the National Academy of Sciences, 102(23), 8386–91. Kay, P., & Regier, T. (2007). Color naming universals: The case of Berinmo. Cognition, 102(2), 289–98. Kowalski, K., & Zimiles, H. (2006). The relation between children’s conceptual functioning with color and color term acquisition. Journal of Experimental Child Psychology, 94(4), 301–21. Kubat, R., Mirman, D., & Roy, D. K. (2009). Semantic context effects on color categorization. In N. A. Taatgen & H. V. Rijn (Eds.), Proceedings of the 31st Annual Cognitive Science Society Meeting (pp. 491–5). Austin, TX: Cognitive Science. Kuehni, R. G. (2004). Variability in unique hue selection: A surprising phenomenon. Color Research and Application, 29, 158–62. Kuehni, R. G. (2007). Development of the idea of simple colors in the 16th and early 17th centuries. Color Research and Application, 32(2), 92–9. Lucy, J., & Shweder, R. (1988). The effect of incidental conversation on memory for focal colors. American Anthropologist, 90(4), 923–31. MacLaury, R. (1992). From brightness to hue. Current Anthropology, 33(2), 137–86. Malt, B. C., Gennari, S. P., Imai, M., Ameel, E., Saji, N., & Majid, A. (2015). Where are the concepts? What words can and can’t reveal. In E. Margolis & S. Laurence (Eds.), The conceptual mind: New directions in the study of concepts (pp. 291–326). Cambridge, MA: MIT Press. Mitterer, H., Horschig, J., Musseler, J., & Majid, A. (2009). The influence of memory on perception: It’s not what things look like, it’s what you call them. Journal of Experimental Psychology, 35(6), 1557–62. Mollon, J. (2006). The Verriest Lecture, Lyon, July 2005. Visual Neuroscience, 23, 297–309. Olkkonen, M., Witzel, C., Hansen, T., & Gegenfurtner, K. R. (2010). Categorical color constancy for real surfaces. Journal of Vision, 10(9), 1–22. Parkhurst, C., & Feller, R. L. (1982). Who invented the color wheel? Color Research and Application, 7(3), 217–30. Parraga, C. A., & Akbarinia, A. (2016). NICE: A computational solution to close the gap from colour perception to colour categorization. PloS One, 11(3), e0149538. Philipona, D., & O’Regan, K. (2006). Color naming, unique hues, and hue cancellation predicted from singularities in reflection properties. Visual Neuroscience, 23, 331–9. Regier, T., Kay, P., & Cook, R. (2005). Universal foci and varying boundaries in linguistic color categories. In Proceedings of the annual meeting of the Cognitive Science Society, (pp. 1827–1832). Merced, CA: California Digital Library. Regier, T. & Kay, P. (2009). Language, thought, and color: Whorf was half right. TICS, 13(10), 439–46.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

   

105

Regier, T., Kay, P., Gilbert, A. L., & Ivry, R. B. (2010). Language and thought: Which side are you on, anyway? In B. Malt & P. Wolff (Eds.), Words and the mind: Perspectives on the language-thought interface. New York: Oxford University Press. Regier, T., Kay, P., & Khetarpal, N. (2007). Color naming reflects optimal partitions of color space. Proceedings of the National Academy of Sciences, 104(4), 1436–41. Roberson, D., Davidoff, J., & Braisby, N. (1999). Similarity and categorization: neuropsychological evidence for a dissociation in explicit categorization tasks. Cognition, 71(1), 1–42. Roberson, D., Davies, I., & Davidoff, J. (2000). Color categories are not universal: Replications and new evidence from a Stone Age culture. Journal of Experimental Psychology: General, 129(3), 369–98. Rosch, E. (1972). Probabilities, sampling, and ethnoraphic method: The case of Dani colour names. Man, 7(3), 448–66. Rosch, E. (1973). Natural categories. Cognitive Psychology, 4(3), 328–50. Rosch-Heider, E., & Mervis, C. B. (1975). Family resemblances: Studies in the internal structure of categories. Cognitive Psychology, 7, 573–605. Rosch, E. (1999). Principles of categorization. In E. Margolis & S. Laurence (Eds.), Concepts: Core readings (pp. 189–206). Cambridge, MA: MIT Press. Roth, E. M., & Shoben, E. J. (1983). The effect of context on the structure of categories. Cognitive Psychology, 15(3), 346–78. Shapley, R., & M. J. Hawken. (2011). Color in the cortex: Ssingle-and double-opponent cells. Vision Research, 51(7), 701–17. Shepard, G. H. (1999). Pharmacognosy and the senses in two Amazonian societies. PhD dissertation. Berkeley: University of California. Sperber, D. & Wilson, D. (1998). The mapping between the mental and the public lexicon. In P. Carruthers & J. Boucher (Eds.), Thought and language (pp. 184–200). Cambridge: Cambridge University Press. Tribushinina, E. (2008). Cognitive reference points: Semantics beyond the prototypes in adjectives of space and colour. Utrecht: LOT. Westfall, R. S. (1962). The development of Newton’s theory of color. Isis, 53, 339–58. Whorf, B.-L. (1940). Science and linguistics. Technological Review, 42(6), 229–31. Wittgenstein, L. (1977). Remarks on colour. Oxford: Basil Blackwell. Witzel, C., Cinotti, F., & O’Regan, J. K. (2015). What determines the relationship between color naming, unique hues, and sensory singularities: Illuminations, surfaces, or photoreceptors? Journal of Vision, 15(8), 19.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

6 The Myth of the Common-sense Conception of Colour Zed Adams and Nat Hansen

6.1 Overview This chapter challenges the very idea of a single, ahistorical, ‘common sense’ conception of colour, on grounds that beliefs about colour vary along several different dimensions: they vary interpersonally, intrapersonally, and historically. Section 6.2 begins by discussing the central role played by one way of thinking about the ‘common sense’ conception of colour in metaphysical debates about colour eliminativism vs colour realism. That way of thinking holds that the ‘common sense’ conception of colour treats it as a non-relational property of objects, ontologically on par with shape or mass. Then, Section 6.3 shows how recent experimental investigations that evaluate that particular way of thinking about the ‘common sense’ conception reveal interpersonal variation in how colours are conceived. Section 6.4 puts pressure on the idea that conceiving of colours as non-relational is a stable component of the ‘common sense’ conception of colour by showing how there is intrapersonal variation in how colours are conceived, depending on what type of object is being considered, and on what intentional ‘frame of reference’ adopted. Section 6.5 discusses a recent genealogical argument (Chirimuuta, 2015) that uses observations about historical variation in how colours are conceived to debunk the idea of an ahistorical, ‘common sense’ conception of colour. Section 6.6 discusses a more subtle conception of the ‘common sense’ conception of colour due to Johnston (1997), who holds that there are five ‘core beliefs’ about colour to which we are ‘susceptible’, simply by virtue of experiencing colour and grasping ordinary colour language. Section 6.7 argues against the idea that we are ‘susceptible’ to these ‘core beliefs’ on the grounds that there is substantial historical variation in the content of those beliefs. Finally, Section 6.8 concludes that the ‘common sense’ conception of colour, as discussed by philosophers, is a myth.

Zed Adams and Nat Hansen, The Myth of the Common-sense Conception of Colour In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Zed Adams and Nat Hansen. DOI: 10.1093/oso/9780198803331.003.0007

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

107

6.2 The ‘Common Sense’ Conception of Colour in Colour Metaphysics The ‘common sense’ conception of colour is relevant for debates in colour metaphysics because central arguments for colour realism, as well as arguments for colour eliminativism, begin from claims about how we ordinarily conceive of colours.¹ They then go on to argue either that such a conception is true of the world (for colour realists), or that it is not (for colour eliminativists). Colour Eliminativism is the denial that physical objects have the colours that we attribute to them, in our everyday thought and talk: the visually conspicuous features that we naively and pre-reflectively attribute to physical objects. And Colour Realism (at least, in one form) is the natural opponent for eliminativism, is the denial of this denial. (Maund, 2011, p. 363)

The eliminativist argument begins with a conceptual and semantic claim about ‘everyday thought and talk,’ specifically about the ordinary concept of colour.² The claim is that we conceive of colours, and talk about them, as if they were intrinsic properties of objects. An intrinsic property of an object is a non-relational property, where shape and mass are exemplars of intrinsic properties. Eliminativists claim that in ordinary thought and talk, both colours and (e.g.) shapes are represented as being intrinsic, non-relational properties of objects: [O]ur dominant ordinary view gives [colours and shapes] much the same status. (Mackie, 1976, p. 16)³

The second step in the eliminativist argument is the claim that colours are not intrinsic properties of objects, because there is no role for such properties in modern scientific explanation. In contrast, other properties that are represented as intrinsic in ordinary thought and talk, like shape, do play a role in scientific explanation (see Mackie, 1976, p. 18; Levine, 2006, p. 276). Combining these two steps produces the following argument for eliminativism: Premise (i):

We ordinarily conceive of colours as intrinsic properties of objects.

¹ This and the following section draw on material in (Adams, 2015, Ch. 1). ² See (Maund, 2011, p. 363) for the ‘conceptual and semantic’ characterization of the first step in the eliminativist argument. ³ In this characterization of the ‘ordinary’, or ‘common-sense’ conception of colour, some realists about colour are in agreement with eliminativists. For example, ‘Pre-reflective common sense is robustly realist about colours and does not distinguish colour from other fundamental properties of external objects such as shape or weight. Our pre-reflective attitude towards colours takes them to be properties of the things they are seen to qualify, in just the same sense as the shape of an object is a property of it’ (Hilbert, 1987, p. 2).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

108

    

Premise (ii): Modern science has discovered that colour is not an intrinsic property of objects. Conclusion:

Colours (as we ordinarily conceive them) do not really exist.

One prominent response to the eliminativist argument is to deny premise (i) by claiming that ordinary thought and talk about colours do not represent colours as being intrinsic properties of objects. Instead, this response involves the claim that the ordinary conception of colour is relational, specifically, that we conceive of colours as dispositions to produce certain kinds of visual experiences. Allen (2007, p. 137) characterizes this response to the eliminativist argument: [P]roponents of this approach claim that the view of colour implicit in common sense is already the view that colours are mind-dependent dispositional properties. . . . [T]his view has proved especially popular at Oxford, where in one form or another it has passed through successive generations of ‘Oxford philosophers’. (Allen, 2007, p. 137)⁴

Allen dubs this rejection of eliminativism ‘Oxford Realism.’⁵ One expression of Oxford Realism declares: When Locke said that the secondary qualities were powers in things to produce sensations in us, he stated the facts correctly, but he did not realize that his statement was only an analysis of the plain man’s use of secondary quality adjectives. . . . When in ordinary life we say ‘The paper isn’t really red,’ we always intend to imply that the paper has some other colour as a dispositional property. (Kneale, 1951, p. 123)

while a second suggests: A secondary quality is a property the ascription of which to an object is not adequately understood except as true, if it is true, in virtue of the object’s disposition to present a certain sort of perceptual appearance: specifically, an appearance characterizable by using a word for the property itself to say how the object perceptually appears. Thus, an object’s being red is understood as something that obtains in virtue of the object’s being such as (in certain circumstances) to look, precisely, red. (McDowell, 1985, p. 111)

⁴ Allen himself does not adopt this approach to arguing against eliminativism. ⁵ For a detailed discussion of the theoretical commitments of Oxford Realism, see (Adams, 2015, §1.2). For a broader history of Oxford Realism, see (Travis & Kalderon, 2013).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

109

The debate between Oxford Realists and eliminativists essentially hinges upon whether one accepts or rejects premise (i). In either case, however, the arguments for or against this premise draw upon considerations that are thought to derive from a ‘common sense’ conception of colour. Eliminativist and others have argued for accepting premise (i) by invoking the ‘transparency’ of colour experience (see, e.g. McGinn, 1996, pp. 541–2), the phenomenon of colour constancy (see, e.g. Allen 2007, p. 145), the fact that we teach the meaning of colour terms by pointing to objects (see, e.g. Gert, 2006, p. 568), the possibility of imagining colours that we are constitutionally unable to perceive (such as ‘killer yellow’) (see, e.g. Broackes, 1992, p. 444), and the claim that colours do not (or cannot) look like dispositional properties (see, e.g. Johnston, 1997, pp. 226–7; Averill & Hazlett, 2010, pp. 142–3). Oxford Realists and others have argued against premise (i) primarily by invoking additional considerations that they think speak on behalf of it (as opposed to arguing directly against the considerations just mentioned). Perhaps most strikingly, some have argued against premise (i) on the grounds that it is incoherent to think of colour is an intrinsic property of objects (see, e.g. Evans, 1985, p. 272; McDowell, 1985, p. 113). Others have also invoked the ineliminable role that human perceptual faculties play in characterizing colours (see, e.g. Dummett, 1993, p. 394), analogies between colour experience and the experience of manifestly dispositional properties (see, e.g. Langsam 2000, p. 74), the ease with which we entertain the possibility of behaviourally undetectable variation in colour experience (see, e.g. Levin, 2000, p. 161), and tacit knowledge of the dependence of colour appearance on line-of-sight and illumination conditions (see, e.g. Adams, 2012, p. 77). In an attempt to break the argumentative deadlock over how to characterize the ‘common sense’ conception of colours, some philosophers have begun to use the tools of the experimental cognitive sciences to investigate how non-philosophers think about colour, in the hope that this might be a better guide to whether or not our ordinary conception of colour involves representing it as an intrinsic property of objects.

6.3 Interpersonal Variation in Two Experiments on the Ordinary Conception of Colour There are two experimental studies that aim to directly address the question of whether non-philosophers conceive of colours as relational or non-relational properties: Cohen and Nichols (2010) and Roberts et al. (2014). Cohen and Nichols find that only a minority (35.5 per cent) of participants consistently judge in a way that is compatible with colour being a non-relational property, in contrast with a majority (65.5 per cent) who judge in a way that is compatible with

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

110

    

shape being a non-relational property. Cohen and Nichols argue that their findings pose a problem for arguments that assume that the ‘common sense’ conception of colour represents it as an intrinsic, non-relational property. But their results also do not put them in a position to claim that the ‘common sense’ conception of colour is clearly relational, either—they found that forty-seven per cent of responses were consistent with relationalism about colour, while fifty-three per cent were consistent with anti-relationalism about colour.⁶ Though they themselves do not draw this conclusion, Cohen and Nichols’ results call into question the assumption that there is a single, univocal ‘ordinary’ or ‘common sense’ conception of colour as either intrinsic or relational. Roberts et al. (2014) criticize the earlier Cohen and Nichols experiment on the grounds that the anti-relationalist response option they offered does not adequately distinguish linguistic, perceptual, and ontological claims about how people ordinarily think about the nature of colour and shape. What might look at first glance like a non-relationalist judgment about colour might instead indicate, for example, a claim about linguistic disagreement. Roberts et al. asked their participants a series of preliminary questions before asking the target ontological questions about the existence of ‘absolute’ (non-relational) facts of the matter about colour and shape; these preliminary questions were intended to distinguish the target ontological question from alternative interpretations. They found that 72.3 per cent of their participants gave anti-relationalist responses to their target prompt about colour, and 85.9 per cent gave anti-relational responses to their target prompt about shape, but the difference between these responses was not statistically significant. When they analysed the responses in two separate groups, one composed of participants with postgraduate training (‘Philpost’) in philosophy, the other without postgraduate training in philosophy (‘Not-philpost’), Philpost did respond to the target colour question and the target shape question in statistically significantly different ways (64.7 per cent non-relational responses to the target colour prompt, and 92.9 per cent non-relational responses to the target shape prompt), whereas Not-philpost did not (seventy-five per cent nonrelational responses for colour, and eighty-four per cent for shape). Roberts et al. (2014, Sect. 5) gloss their results as follows: [C]ompared with 53.00% of Cohen and Nichols’ participants, 72.30% of our participants, and 75.00% for Not-Philpost, indicated agreement with antirelationalism about colour. Thus, our results suggest that things are not nearly as ambiguous as Cohen and Nichols’ study would have us believe. In fact, our data suggests that anti-relationalism is clearly the pre-theoretically intuitive position. ⁶ They do not report what percentage of participants gave consistently relationalist responses to their survey.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

111

There are two reasons why that conclusion is too hasty. First, there is still a substantial minority of relationalist responses to the target statement for colour, even among ‘ordinary’ participants who have not had postgraduate training in philosophy. Roberts et al. are assuming that there is a single, univocal ‘common sense’ conception of colour that their results are uncovering, but that is not supported by their data. Second, Roberts et al. are measuring levels of agreement with an affirmative target statement that expresses an anti-relationalist claim. There is a well-known acquiescence (or ‘yea-saying’) bias in responses to surveys, which can lead to artificial inflation of rates of agreement with claims like the target prompt, irrespective of their content (see, e.g. Jackman, 1973). A standard way of controlling for acquiescence bias is by asking for responses to both positive and negative statements. For example, consider the relative appeal of Roberts et al.’s anti-relationalist prompt and its relationalist negation: Target [original, anti-relationalist]. In reality, there is an absolute fact of the matter about the colour [or shape] of the object regardless of how it appears to Alex and Harry and regardless of what they think, say, or do. Target [reversed, relationalist]. In reality, there is not an absolute fact of the matter about the colour [or shape] of the object; its colour depends on how it appears to Alex and Harry, or on what they think, say, or do.

If participants’ responses to these two prompts were mirror images of each other, generating high levels of agreement with the original target, and high levels of disagreement with the reversed version, that would rule out the presence of acquiescence bias and generate much stronger evidence in favour of Roberts et al.’s conclusion. While Cohen and Nichols (2010) and Roberts et al. (2014) are important first attempts to experimentally investigate aspects of the ‘common sense’ conception of colour, neither study decisively establishes that the ‘common sense’ conception of colour is best characterized as either relationalist or anti-relationalist, and we think there is good reason to take the central lesson of these two studies to be that there is interpersonal variation in whether colour is thought of as an intrinsic or relational property. That is, the studies should make us wonder whether there is a single univocal way that colours are represented in ‘common sense’.

6.4 Intrapersonal Variation in the ‘Common Sense’ Conception of Colour Cohen and Nichols’s experiment asks for colour judgments about a tomato, and Roberts et al.’s experiment asks for colour judgments about a generic, unspecified

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

112

    

‘object.’⁷ We think that the type of examples employed in both experiments contributes to an illusion of univocality in the ‘common sense’ conception of colour, even bracketing concerns about interpersonal variation raised in the previous section. The focus on an extremely limited set of objects is an instance of what Austin criticized as a typically ‘philosophical’ obsession with a narrow range of examples: My general opinion about this doctrine is that it is a typically scholastic view, attributable . . . to an obsession with a few (and nearly always the same) halfstudied ‘facts’. (I say ‘scholastic’, but I might just as well have said ‘philosophical’; over-simplification, schematization, and constant obsessive repetition of the same small range of jejune ‘examples’ are not only not peculiar to this case, but far too common to be dismissed as an occasional weakness of philosophers). (Austin, 1962, p. 3)

Austin goes on to discuss the class of ‘familiar items’ that philosophers rely on in their discussion of the beliefs of what the ‘ordinary man’ would say that he perceives, which Austin famously characterizes as ‘moderate-sized specimens of dry goods’, saying ‘we are given, as examples, “familiar objects”—chairs, tables, pictures, books, flowers, pens, cigarettes . . . ’ and observes that this list of items leaves out many other items that people say that they see or hear or smell, which would include at least ‘ . . . people, people’s voices, rivers, mountains, flames, rainbows, shadows, pictures on the screen at the cinema, pictures in books or hung on walls, vapours, gases . . . ’ (1962, p. 8). Any comprehensive assessment of the ‘common sense’ conception of colour should not stop at beliefs about ‘moderate-sized specimens of dry goods’ (including tomatoes), however, but also ask about objects from Austin’s expanded list. Consider rainbows. Try plugging in ‘rainbow’ as the object in Roberts et al.’s target statement:⁸ Target [rainbow]. In reality, there is an absolute fact of the matter about the colour of a rainbow regardless of how it appears to Alex and Harry.

Asking about the colour of a rainbow immediately draws attention to a number of considerations that are overlooked by Roberts et al.’s discussion of generic objects. Consider how rainbows are only visible under specific lighting and atmospheric conditions, and from a specific point of view, as well as how those conditions

⁷ On the methodological significance of ‘generic’ objects, see (Cavell, 1979, pp. 52–7). ⁸ Here we are following the advice of (Wilson, 2006, p. 79): ‘A useful form of experimentation to employ in such cases is to lower the level of abstraction by replacing the programmatic “thing” throughout by some suitable exemplar . . . ’.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

113

affect the rainbows themselves: if Alex moves to the left, and Harry moves the right, the rainbow(s) that they see will appear to move in opposite directions; if something comes in between Alex and the light source behind him, but not between Harry and the light source, then the rainbow that Alex sees, but not the rainbow that Harry sees, will cease to exist. Also, the vividness of the rainbow depends directly on the lighting and atmospheric conditions, as does the very existence of the rainbow itself. These sorts of considerations immediately raise the question of whether we ordinarily think of rainbows themselves as stable, nonrelational objects, which thereby complicates the question of whether we think of their colours as stable, non-relational properties. Or consider the back of a compact disc. Johnston (1997) observes how the ‘unsteady colours’ on the back of a CD, unlike the ‘steady’ colours of moderatesized dry goods, seem manifestly relational: A basic phenomenological fact is that we see most of the colours of things as ‘steady’ features of those things, in the sense of features which do not alter as the light alters and as the observer changes position. . . . A course of experience as of the steady colours is a course of experience as of light-independent and observerindependent properties, properties simply made evident to appropriately placed perceivers by adequate lighting. Contrast the highlights: a course of experience as of the highlights reveals their relational nature. They change as the observer changes position relative to the light source. They darken markedly as the light source darkens. With sufficiently dim light they disappear while the ordinary [i.e. ‘steady’] colours remain. They wear their light- and observer-dependent natures on their face. Thus there is some truth in the oft-made suggestion that (steady) colours don’t look like dispositions; to which the natural reply is ‘Just how would they have to look if they were to look like dispositions?’; to which the correct response is that they would have to look like coloured highlights or better, like shifting, unsteady colours, e.g. the swirling evanescent colours that one sees on the back of compact discs. (Johnston, 1997, pp. 226–7)

Johnston’s discussion makes clear that his ‘common sense’ conception of colour is not uniform in terms of how it represents colour as intrinsic or relational—it admits of variation across different objects. For another type of object not typically invoked in philosophical discussions of the nature of colour, consider the following thought experiment concerning mirrors: You are a professional truck driver. You are making a delivery that involves backing your truck down a narrow alley up to a loading dock. In order to see what is behind the truck, you must rely entirely on the use of your side view mirrors. Doing so, by using your mirror, you see that someone standing on the dock is wearing a red Santa Claus outfit. (Adams, 2015)

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

114

    

Ask yourself the following questions: Q1: Do you think the Santa Claus outfit would continue to be red in the dark? Q2: Do you think the mirror would continue to be red in the dark?
 Q3: Do you think that the Santa Claus outfit would continue to be red if you move in such a way that you are no longer able to see it? Q4: Do you think that the mirror would continue to be red if you move in such a way that you are no longer able to see it? It seems (to us, and to Maund, 2016 at least) that the intuitive answer to the odd numbered questions is clearly yes while the answer to the even numbered questions is clearly no. But those different responses are indicative of different attitudes about the intrinsic vs relational nature of the colour of the Santa Claus outfit and the mirror: it seems we judge that the colour properties of the mirror are dependent on lighting conditions and viewing angle, while the colour properties of the Santa Claus outfit are not. Again, consider plugging in ‘mirror’ into a suitably modified version of the Roberts et al. target statement: Target [mirror]. In reality, there is an absolute fact of the matter about the colour of a mirror regardless of how it appears to Alex and Harry.

By expanding the range of examples beyond standard moderate-sized dry goods (or generic objects), the existence of intrapersonal variation in the ‘common sense’ conception of colour is revealed: even when it may seem appropriate to judge that certain objects (tomatoes, Santa Claus outfits) are red non-relationally, it may seem, to the same person at the same time, appropriate to judge that other objects (rainbows, compact discs, mirrors) are red only relationally. This is another reason to doubt the idea that there is going to be a univocal ‘common sense’ conception of colour. In addition to object-based variation in whether colour appears relational, there is another form of intrapersonal variation in whether colour seems relational, which can even affect judgments about one-and-the-same object. Evans (1974) distinguishes different ‘frames of reference’ from which the colour of an object can be considered. In the ‘object’ frame of reference, An observer is always aware of the illumination though seldom consciously. He has simply trained himself to take such action, automatically, as will maximize his perception of the object properties. We see this in everyday commonplace actions. If a glossy surface reflects light into his eyes, he will move his head, change his position, or, if possible, pick up the object and hold it so the reflection disappears. If the light is too dim, he will take an object to the window or to a

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

115

light. If the illumination is too bright, he may shade it with his hand. In other words, he deliberately manipulates the illumination to see the object colour; they are two separate things to him but he is usually interested only in the object. (Evans, 1974, p. 198, quoted in Wilson, 2006, pp. 458)

Switching from the object frame of reference, observers can intentionally adopt what Evans calls the ‘stimulus frame of reference.’ In the stimulus frame of reference, an observer is interested in ‘seeing, or trying to see, the stimuli as such without regard to the situation’ (Evans, 1974, p. 205). Evans (1974, p. 206) refers to passage in John Ruskin’s Elements of Drawing to illustrate the stimulus frame of reference: [W]hen grass is lighted strongly by the sun in certain directions, it is turned from green into a peculiar and somewhat dusty-looking yellow . . . if there were primroses near, we should think that the sunlighted grass was another mass of plants of the same sulphur-yellow colour . . . Very few people have any idea that sunlighted grass is yellow. (Ruskin, 1876, pp. 27–8, n 1)

Adopting the stimulus frame of reference involves treating colours of objects as explicitly relational—as how the object looks from this angle, at this distance, under these lighting conditions. We can intentionally switch between the object and stimulus frames of reference, which means that there is no settled univocal ‘common sense’ conception of the colour of objects as relational or non-relational, even within a single subject and with regard to a single object. Whether we are inclined to think of colour as a relational or non-relational property of an object varies depending on what kind of object is being considered, and on what ‘frame of reference’ we adopt. Those two forms of intrapersonal variation in how we conceive of colour put further pressure on the idea that there is a univocal ‘common sense’ conception of colour.

6.5 Historical Variation in the ‘Common Sense’ Conception of Colour Chirimuuta (2015, Ch. 2) develops another dimension of criticism of the ‘common sense’ conception of colour, offering a genealogical account of the intuition that colours are equally as objective as shape or mass.⁹ She argues that that intuition—which is endorsed by both eliminativists and some colour realists (albeit not Oxford Realists)—is not an immutable part of ‘common sense,’ ⁹ For some recent reflections on the genealogical method in philosophy more generally, see (Dutilh Novaes, 2016) and (Plunkett, 2016).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

116

    

where ‘common sense’ is understood as ‘intuition, part of our perceptual and cognitive endowment, no matter how scientifically sophisticated we are’ (p. 32). Instead, she proposes that the idea that colour and shape have the same ontological status (that is, that they are both non-relational) is ‘actually the remnant of scholastic theory,’ and thus potentially can be changed. She then undertakes the task of ‘reshaping our intuitions about colour sensation’ along different lines (p. 34). Chirimuuta’s genealogy begins with a description of the scholastics’ ‘robustly realist epistemology of sight,’ which derives from their theory of intentional species. According to that theory, perception involves the sensory organs coming to have properties possessed by the perceived object—namely, the object’s form or species (p. 21). If colour is understood as a form or species of a perceived object, then a ‘strong colour realism,’ according to which colour belongs to the perceived object (as well as the perceptual organ), follows from the theory of intentional species. It is this strong colour realism that is challenged by the new scientific world view that emerges in the seventeenth century. Chirimuuta summarizes the relation between the scholastic conception of colours and the scientific rejection of that conception, saying: In the late Middle Ages colours are very much in the world. There is no mismatch between perceptual phenomenology and scientific ontology. Therefore, there is no problem of colour. At the same time scholastic realism set colour up for a fall. By intimating so strong a notion of the objectivity of vision—that the world is as we see it to be—it was inevitable that colours would be found wanting, thus inviting the Galilean criticism that we are entirely wrong in our association of colour experiences with physical objects. (p. 22)

The seventeenth century emergence of an eliminativist ‘error theory’ about colour, which persists in contemporary philosophizing, rests on the idea that ‘common sense’ represents colours as belonging to objects non-relationally. But once we come to see that the purportedly ‘common sense’ conception of colour is ‘is not just unschooled intuition, but part of a sophisticated theoretical edifice’ (p. 26), then we won’t be moved to deny the existence of colours as such. We are sympathetic with Chirimuuta’s historical criticisms of the idea that the ‘common sense’ conception of colour is non-relational, but Section 6.2 and Section 6.3 give reasons (from interpersonal and intrapersonal variation in the contemporary ‘common sense’ conception) to think that it is a mistake to think that there is a univocal contemporary ‘common sense’ conception of colour as non-relational in the first place.¹⁰ Given our arguments, her account might better ¹⁰ Adams (2015) offers a genealogy of our contemporary concept of colour according to which it is made up of multiple conflicting but intertwined strands of thought.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

117

be thought of as explaining the origin and persistence of certain philosophers’ assumption that ‘common sense’ represents colour as non-relational. But her genealogical approach is a fruitful way of investigating another dimension along which the ‘common sense’ conception of colour may vary. Section 6.6 and Section 6.7 adopt a genealogical approach to a more nuanced, but still inadequate, way of thinking about the ‘common sense’ conception of colour.

6.6 A Wider Array of ‘Core Beliefs’ about Colours The debate central to the first part of this chapter hinged on a single, purportedly ‘common sense’, belief about colour, namely, whether we ordinarily believe that colours are intrinsic or relational properties of objects. The eliminativist argument rests upon the presupposition that if that single belief turns out to be false, then colours do not exist. That presupposition involves the contentious idea that there is a single essential belief at the centre of the ‘common sense’ conception of colour. Mark Johnston (1997) offers a more nuanced framework for exploring the ‘common sense’ conception of colour than the experimental studies discussed in Section 6.3 (even though it was published nearly twenty years earlier). Johnston proposes that the ‘common sense’ conception of colour is a ‘cluster concept’ composed of five ‘core beliefs.’ Significantly, Johnston’s proposal thus explicitly rejects any attempt to tie the ‘common sense’ conception colour to any particular single belief (a tendency which extends beyond the debate outlined above to include other articles, e.g. Strawson, 1989 and Boghossian & Velleman, 1989). Rather than focusing solely on the issue of whether colours are represented as intrinsic or relational, Johnston explores the extent to which each of his five ‘core beliefs’ about colour may or may not be ‘at odds with discovered facts’ (Johnston, 1997, p. 137), as a way of determining how many (if any) of the five ‘core beliefs’ can be retained by a defensible metaphysical theory of colour. His conclusion is that an ‘ever so inclusive’ conception of colour—one that incorporates all five ‘core beliefs’—has to be abandoned, but that a ‘more or less inclusive’ conception of colour—one that only incorporates some of the five ‘core beliefs’—can be retained (Johnston, 1997, p. 137). Retaining some of the ‘core beliefs’ is sufficient to prevent wholesale elimination of the concept of colour—though it may require revision of some parts of the ‘common sense’ concept. Here are the five ‘core beliefs’ about colour that Johnston proposes (with ‘canary yellow’ standing in for all colours) (Johnston, 1997, p. 138): Paradigms. Some of what we take to be paradigms of canary yellow things (e.g. some canaries) are canary yellow.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

118

    

Explanation. The fact of a surface or volume or radiant source being canary yellow sometimes causally explains our visual experience as of canary yellow things. Unity. Thanks to its nature and the nature of the other determinate shades, canary yellow, like the other shades, has its own unique place in the network of similarity, difference and exclusion relations exhibited by the whole family of shades. Availability. Justified belief about the canary yellowness of external things is available simply on the basis of visual perception. That is, if external things are canary yellow we are justified in believing this just on the basis of visual perception and the beliefs which typically inform it. Revelation. The intrinsic nature of canary yellow is fully revealed by a standard visual experience as of a canary yellow thing.

For Johnston, the most important questions are how many, if any, of these beliefs can be retained, and what sort of metaphysical view of the nature of colours they are best explained by. He concludes that a dispositional theory of the colours offers the most inclusive account of the ‘core beliefs’. Johnston’s project depends upon two major assumptions. First, it assumes that the success or failure of metaphysical theories of colour is measured by the extent to which they accommodate ‘core’ aspects of the concept of colour. Second, his project assumes that we are ‘susceptible’ to the ‘core beliefs,’ in a very specific sense. According to Johnson, we find ourselves with the ‘core beliefs’ about colour solely in virtue of (a) visual experience of colours, and (b) a grasp of ordinary colour language (Johnston, 1997, p. 137). In short, human beings with normal perceptual abilities who are competent speakers of a language should simply find themselves with the ‘core beliefs’ about colour. Call this the susceptibility assumption. If it turned out that adherence to the ‘core beliefs’ was highly variable (across individuals, cultures, or historical periods), then the idea that metaphysical theories of colour should be assessed in terms of how well they satisfy such beliefs would have to be reconsidered. Johnston’s first assumption is very contentious. The assumption has been defended by advocates of contemporary ordinary language philosophy (including contextualists about ‘knows’ and some experimental philosophers), who argue that revisionary theories of philosophically significant concepts like yellow or knowledge fail because they change the subject—they are not theories of what they purport to be about. Hansen (2014, p. 561) formulates the argument behind this position as follows: Premise (i): If we want our philosophical theories (e.g. of knowledge [or colour]), to address the concerns that led us to philosophical investigation in the

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

119

first place, then they need to be about the relevant objects of investigation as they are ‘ordinarily understood outside of strictly philosophical discourse’ (Alexander & Weinberg, 2007, p. 58).
 Premise (ii): We do want our philosophical theories (e.g. of knowledge), to address the concerns that led us to philosophical investigation in the first place. Conclusion: So our philosophical theories need to be about the relevant objects of investigation (e.g. knowledge [or colour]), as they are ordinarily understood outside of strictly philosophical discourse. 
 One standard criticism of Johnston’s first assumption is that it is simply a mistake to assume that the relevant objects of investigation can only be understood via whatever ‘folk categories’ first led us to be interested in them. Here is how Byrne and Hilbert (2003, p. 4) motivate this criticism: Consider an analogy. From the point of view of the biologist, the word ‘food’ is applied by ordinary people in a somewhat arbitrary way. According to them, the synthetic cooking oil Olestra, which has no nutritional value at all, is a food, but vitamin tablets and beer are not. An investigation of how ordinary people use the word ‘food’ is not particularly relevant to biology. What is relevant is an investigation into the sorts of substances human beings can digest, whether or not the biological category of the digestible lines up exactly with the folk category of food. The problem of colour realism is like the investigation of what humans can digest, not the investigation of the folk category of food. The enquiry concerns certain properties that objects visually appear to have, not how ordinary people use colour words, or how they conceptualize colour categories (emphasis in original).

This chapter does not enter into this ongoing debate over the first assumption. Our concern is with Johnston’s second assumption, the susceptibility assumption. We think there is evidence that at least some of Johnston’s ‘core beliefs’ are not simply the result of visual experience and grasp of ordinary colour language. Our evidence is historical.¹¹ We think that a genealogical approach to thought about colour provides evidence that many of our ‘core beliefs’ about colour are historically contingent, which means that we are not ‘susceptible’ to them in Johnston’s sense.

¹¹ We think that some of Johnston’s ‘core beliefs’ are accurate characterizations of aspects of the contemporary ‘common sense’ conception of colour. There is experimental evidence (Roberts and Schmidtke, 2019) that provides some support to the claim that the ‘core beliefs’ correctly characterize our contemporary ‘common sense’ conception of colour. In particular, it seems difficult to dispute that both Paradigms and Unity are aspects of contemporary ‘common sense’ ways of thinking about colour.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

120

    

6.7 A Genealogy of Some ‘Core Beliefs’ about Colour Our goal in this section is to make the case that although some of Johnston’s ‘core beliefs’ might correctly characterize our contemporary ‘common sense’ conception of colour, they probably would not have been ‘core beliefs’ in earlier historical periods. To make our case, we focus on the ‘core beliefs’ of Explanation and Unity. In brief, we argue that the notion of ‘causal explanation’ invoked in Explanation has varied widely across its history, just as beliefs about the similarity and difference relationships invoked in Unity have varied. In both cases, the notions that Johnston is working with appear to be modern developments, and hence cannot be something that humans are ‘susceptible’ to in virtue solely of visual experience of colours and grasp of ordinary colour language.¹² First, consider Explanation. According to Johnston, it is part of our ‘common sense’ conception of colours that an object’s being (for instance) canary yellow causally explains visual experiences of it as being canary yellow (Johnston, 1997, p. 138).¹³ Johnston is presumably thinking of colours as efficient causes, as opposed to their being material, formal, or final causes.¹⁴ But merely raising the question of which of these four types of causation Johnston has in mind immediately raises the larger question of whether there is a single, shared, ahistorical ‘common sense’ conception of efficient causation itself. By looking at the history of thought about efficient causation, however, we can see that there is not a single, shared conception of efficient causation.¹⁵ On the contrary, the notion of efficient causation has gone through a series of transformations from Aristotle’s time to our own. Perhaps most strikingly, for Aristotle an understanding of efficient causation was not something one could have in isolation from an understanding of the three other sorts of causal understanding just mentioned; in particular, an understanding of something’s efficient cause ¹² Our goal in this section is not to argue that there is no ‘common sense’ conception that does not change over time, or even that there is no possible way of weakening Johnston’s ‘core beliefs’ so that they would correctly characterize ancient (e.g.) beliefs about colour. In this respect, we agree with Chirimuuta (2015, p .32, n. 16), who writes: ‘Note that I remain neutral over the issue of whether there might be any universal folk theory of color, one that makes weaker claims about the nature of color than those typically formulated under the heading of common sense.’ Our criticism of the ‘common sense’ conception of colour is targeted at the way that idea has been developed and deployed by philosophers. ¹³ Other philosophers who share Johnston’s commitment to thinking that Explanation is one of our ‘core beliefs’ about colour include Hacker (1987); Broackes (1992); Campbell (1993); Jackson (1996); Watkins (2005); and Maund (2006). Notable dissenters include White (1961) and Hyman (2006). As Hyman puts it, ‘an analysis of the basic conception of color that is implicit in the simplest color statements we make’ reveals that ‘colors are inert, in the sense that they cannot influence what happens’ (p. 19). As with the debate between eliminativists and Oxford Realists outlined in Section 6.2, we think it is telling that there is such stark disagreement among philosophers about the content of putatively ‘common sense’ beliefs. ¹⁴ For a useful overview of the distinction between efficient, material, formal, and final causes, see Falcon (2015). ¹⁵ Schmaltz (2014) is a useful overview of the history of thought about efficient causation.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

121

essentially depends upon an understanding of its final cause.¹⁶ The notion of efficient causation has since been transformed by being wrenched from its internal relationship to these other sorts of causation. Schmaltz summarizes this history, saying: Somewhere along the road that leads from Aristotle to our own time, then, material, formal and final aitai [causes] were lost, leaving only efficient aitia to serve as the central element in our causal explanations. Indeed, there is reason to think that the journal has transformed Aristotle’s efficient aitai into something he could not have anticipated. . . . There is in fact no straightforward connection between Aristotle’s concept of an efficient aitai and our concept of a cause. (Schmaltz, 2014, p. 5)

For our purposes, the takeaway point from this history is simply that ‘Aristotle’s notion of cause, then, is quite different from ours’ (Frede, 1980, p. 218).¹⁷ This observation, in itself, should make us wary of thinking that Johnston’s own understanding of efficient causation is a deliverance of ahistorical ‘common sense.’ There is an additional benefit to be gained from looking at this history. It helps us to see how a shared language or culture is not sufficient for having a shared understanding of causal explanation. For instance, as the result of surveying both ancient Greek and ancient Chinese writers, G. E. R. Lloyd notes that ‘the diversity of views expressed by different authors within each ancient civilization is remarkable’ because ‘there was no uniformity about what was to count as a cause’ (Lloyd, 2007, p. 128, 123). This is a telling reminder of how sensitivity to the history of a concept can make us more attuned to the possibility of conceptual variation not just across times and places but even within a particular time and place, variation that our own situatedness in our present time and place might lead us to overlook. Now consider Unity. Johnston gives a clear expression to his susceptibility assumption in relation to the ‘core belief ’ of Unity in the following passage: But is it really a matter of scientific discovery that canary yellow is not a shade of blue? No: such similarity and difference principles surely have a different status. We take ourselves to know these principles just on the basis of visual experience ¹⁶ As Alan Code puts this aspect of Aristotle’s view, with specific reference to the efficient causes involved in an animal’s conception and growth, ‘Knowledge of the efficient causes by means of which animals are generated is posterior to knowledge of their final causes’ (Code, 1997, p. 143). ¹⁷ Summarizing the difference between Aristotle’s understanding of efficient causation and our own contemporary conception is difficult not just because of debates within Aristotle scholarship about how to interpret his view, but also because there is widespread disagreement about what ‘our’ view is. That said, Hulswit glosses the difference, saying: ‘All in all, the complex evolution of the concept of cause from the seventeenth century on is marked by the interplay between, at least, two radically different conceptions of cause: the Aristotelian-scholastic conception, according to which causes are the active initiators of a change, and the scientific conception, according to which causes are the inactive nodes in a law-like implication chain’ (Hulswit, 2002, p. 44).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

122

    

and ordinary grasp of colour language. No one had to wait until the end of the second millennium A.D. to find out whether or not canary yellow is a shade of blue. (Johnston, 1997, p. 150)

By looking at ancient attempts to organize colour space, however, we can appreciate how our own understanding of the ‘internal’ relations among colours, captured by Unity, is a historical development, the result of centuries of work reflecting on and refining our contemporary models of how colour is represented in visual experience. Consider, for example, Plato’s discussion of relations among colours from the Timaeus, and the visual representation of the resulting relations in Figure 6.1 (quotation and figure from Kuehni, 2003, pp. 22–3): . . . we ought to term white that which dilates the visual ray, and the opposite of this black . . . in (the eye) the fire, mingling with the ray of the moisture, produces a colour like blood, to which we give the name red. A bright hue mingled with red and white gives the colour auburn [xandon]. The law of proportion, however, according to which the several colours are formed, even if a man knew he would be foolish in telling, for he could not give any necessary reason, nor indeed any tolerable or probable explanation of them. Again, red when mingled with black and white, becomes purple, but it becomes umber [orphninon] when the colours are burned as well as mingled and the black more thoroughly mixed with them. Flame colour [pyrron] is produced by a union of auburn and dun [phaion], dun by an admixture of black and white; and pale yellow [ochron] by an admixture of white and auburn. White and bright meeting and falling upon a full black,

Primary

Secondary

Tertiary

Quaternary

Bright Dun Flame Color White Auburn

+ White

Pale Yellow

+ Black

Leek Green

Red Purple

Black

Dark Blue

+ White

Light Blue

Figure 6.1. Plato’s colour mixture scheme. The four primary experiences are bright, white, red, and black. Additions of these in various combinations from the secondary and later stage mixture colours. After Kuehni (2003)

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

123

become dark blue [kyanoyn], and when the dark blue mingles with white a light blue [glaykon] colour is formed as flame colour with black makes leek-green [prasion]. There will be no difficulty in seeing how and by what mixtures the colours derived from these are made according to the rules of probability.

Notice that, according to Plato’s system of colour ordering, leek green is derivable from red (whereas, on our contemporary scheme, green and red are opposites). Pale yellow and light blue are both tertiary colours that are both produced by mixing white with secondary colours. Plato’s scheme represents a radically different conception of the relation among shades than our contemporary understanding of colour space. Describing Plato’s colour system, Kuehni (2003) writes: There is clearly an awareness of lightness from the sequences ‘auburn plus white makes light yellow’ and ‘dark blue plus white makes light blue’, as well as ‘purple burned makes umber.’ But there is no indication of a systematic arrangement by lightness. (p. 25)

Roughly seven hundred years later, a Medieval commentary on Plato’s Timaeus by Chalcidius still employed a one-dimensional colour scale, with five basic colours: white, yellow, red, blue, and black. Multi-dimensional colour spaces can be traced back to Avicenna (980–1037) (Kuehni, 2003, p. 27), but the first ‘explicit’ hue circle representing the familiar relations of similarity, difference, and exclusion that Johnston has in mind was created by Isaac Newton (Kuehni, p. 43). Modern colour ordering systems are a historical development, the result of centuries of work refining the representation of colour space. Distinguishing hue from lightness (the white–black scale), for example, is not something we are ‘susceptible’ to in Johnston’s sense—black and blue can seem phenomenologically similar, just as white and yellow can seem phenomenologically similar. Distinguishing the relevant dimensions of similarity and difference requires experimentation and theory building.¹⁸ In short, we think that whatever ‘susceptibility’ there is to certain conceptions of relations of similarity, dissimilarity, and exclusion that might constitute a ‘core belief ’ of Unity is not due solely to visual experience and grasp of ordinary colour language—it is a historical, theoretical development.¹⁹

¹⁸ See Kuehni (2003) for a detailed historical survey of the development of colour ordering systems. See also Mausfeld (2003). ¹⁹ For a challenge to the idea of a unified colour space from cultural variation, see Gage (2012), and Lloyd (2007, Ch. 1).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

124

    

6.8 Conclusion: The Myth of the ‘Common Sense’ Conception of Colour Our aim in this essay has been to challenge the univocality of ‘the common sense’ conception of colour. Ordinary, non-philosophical, conceptions of colour vary in several different dimensions: a) There is experimental evidence of interpersonal variation in whether nonphilosophers conceive of the colour of objects as relational or non-relational properties (moreover, there is no disputing the fact that there is interpersonal variation in how philosophers conceive of colours!). b) Once the set of examples under discussion expands from generic objects to the more diverse objects that Austin encourages us to think about, the fact of substantial intrapersonal variation in whether colour is conceived of relationally or non-relationally becomes hard to deny, in two ways. First, certain types of objects (rainbows, the backs of CDs, mirrors) seem to have colours in manifestly relational ways, while other objects (tomatoes and other ‘moderately-sized dry goods’) seem to have colours in relatively ‘stable’ non-relational ways. Secondly, with an intentional shift in ‘frame of reference’, even those objects that seem to have ‘stable’ colours can come to appear to have colours in manifestly relational ways. c) Some of Johnston’s (1997) ‘core beliefs’ about colour, namely Unity and Explanation, are historical developments, rather than ur-beliefs that we are ‘susceptible’ to simply in virtue of having colour experiences and being competent with colour vocabulary. This historical variation in supposedly ‘core beliefs’ makes it difficult to maintain that metaphysical theories of colour should be assessed in terms of how well they fit with commonsensical ‘core beliefs’ unless the existence of such ‘core beliefs’ is backed up with evidence indicating that they are aspects of an ahistorical (and nonculturally relative) conception of colour. In light of these dimensions of variation, we conclude that the idea of a univocal, ahistorical, ‘common sense’ conception of colour, as it has been discussed by philosophers, is a myth.

* This work was supported by an External Faculty Fellowship at Stanford University’s Humanities Center. Thanks to Jake Browning, Yasmina Jraissati, and Mark Phelan’s philosophy of mind class for comments.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

125

References Adams, Z. (2012). Seeing is knowing. Review of Metaphysics, 66(1), 61–88. Adams, Z. (2015). On the genealogy of colour: A case study in historicized conceptual analysis. New York: Routledge. Alexander, J., & Weinberg, J. (2007). Analytic Epistemology and Experimental Philosophy. Philosophy Compass, 2(1), 56–80. Allen, K. (2007). The mind-independence of colour. European Journal of Philosophy, 15(2), 137–58. Austin, J. L. (1962). Sense and sensibilia. Oxford: Oxford University Press. Averill, E. W., & Hazlett, A. (2010). A problem for relational theories of colour. Philosophy and Phenomenological Research, 91(1), 140–5. Boghossian, P. A., & Velleman, J. D. (1989). Colour as a secondary quality. Mind, 98, 81–103. Broackes, J. (1992). The autonomy of colour. In K. Lennon & D. Charles (Eds.), Reduction, Explanation, and realism (pp. 191–225). Oxford: Oxford University Press. Byrne, A., & Hilbert, D. R. (2003). Colour realism and colour science. Behavioral and Brain Sciences, 26(1), 3–21. Campbell, J. (1993). A simple view of colour. In J. J. Haldane & C. Wright (Eds.), Reality: Representation and projection (pp. 257–68). Oxford: Oxford University Press. Cavell, S. (1979). The claim of reason. Oxford: Oxford University Press. Chirimuuta, M. (2015). Outside Colour: Perceptual science and the puzzle of colour in philosophy. Cambridge, MA: MIT Press. Code, A. (1997). The priority of final over efficient cause in Aristotle’s Parts of animals. In W. Kullmann & S. Föllinger (Eds.), Aristotelische Biologie (pp. 127–43). Stuttgart: Franz Steiner Verlag Wiesbaden GmbH. Cohen, J., and Nichols, S. (2010). Colours, colour relationalism and the deliverances of introspection. Analysis, 70(2), 218–28. Dummett, M. (1993). Common sense and physics. In The seas of language (pp. 376–410). Oxford: Clarendon Press. Dutilh Novaes, C. (2016). Conceptual genealogy for analytic philosophy. In J. A. Bell, A. Cutrofello, & P. M. Livingston (Eds.), Beyond the analytic-continental divide: Pluralist philosophy in the twenty-first century (pp. 75–110). New York: Routledge. Evans, R. M. (1974). The perception of colour. New York: John Wiley and Sons. Evans, G. (1985). Things without the mind—A commentary upon chapter two of Strawson’s Individuals. In G. Evans, Collected Papers (pp. 249–90). Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

126

    

Falcon, A. (2015). Aristotle on causality. In E. N. Zalta (Ed.), The Stanford Encyclopedia of Philosophy (Online, Spring 2015 Edition). Available at https:// plato.stanford.edu/archives/spr2015/entries/aristotle-causality/. Accessed 17 January 2017. Frede, M. (1980). The original notion of cause. In J. Barnes, M. F. Burnyeat, & M. Schofield (Eds.), Doubt and dogmatism: Studies in Hellenistic epistemology (pp. 217–49). Oxford: Oxford University Press. Gage, J. (2012). Colour ordered and disordered. In F. G. Barth, P. Giampieri-Deutsch, & H.-D. Klein (Eds.), Sensory perception: Mind and matter (pp. 299–311). Vienna: Springer. Gert, J. (2006). A realistic colour realism. Australasian Journal of Philosophy, 84(4), 565–89. Hacker, P. M. S. (1986). Insight and illusion: Themes in the philosophy of Wittgenstein. Oxford: Oxford University Press. Hansen, N. (2014). Contemporary ordinary language philosophy. Philosophy Compass, 9(8), 556–69. Hulswit, M. (2002). From cause to causation: A Peircean perspective. Dordrecht: Springer. Hyman, J. (2006). The objective eye. Chicago: University of Chicago Press. Jackman, M. R. (1973). Education and prejudice or education and response-set? American Sociological Review, 38(3), 327–39. Jackson, F. (1996). The primary quality view of colour. Philosophical Perspectives, 10, 199–219. Johnston, M. (1997). How to speak of the colours. In A. Byrne & D. Hilbert (Eds.), Readings on colour, Volume 1: The philosophy of colour (pp. 137–72). Cambridge, MA: MIT Press. Kneale, W. (1951). Sensation and the physical world. The Philosophical Quarterly, 1(2), 109–26. Kuehni, R. G. (2003). Colour space and its divisions: Colour order from antiquity to the present. Hoboken, NJ: John Wiley and Sons. Langsam, H. (2000). Why colours do look like dispositions. The Philosophical Quarterly, 50(198), 68–75. Levin, J. (2000). Dispositional theories of colour and the claims of common sense. Philosophical Studies, 100(2), 151–74. Levine, J. (2006). Colours and colour experience: Colours as ways of appearing. Dialectica, 60(3), 269–82. Lloyd, G. E. R. (2007). Cognitive variations: Reflections on the unity and diversity of the human mind. Oxford: Oxford University Press. Mackie, J. L. (1976). Problems from Locke. Oxford: Clarendon Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    -   

127

Maund, B. (2006). The illusory theory of colours: An anti-Realist theory.’ Dialectica, 60(3), 245–68. Maund, B. (2011). Colour eliminativism. In L. Nolan (Ed), Primary and secondary qualities: The historical and ongoing debate (pp. 362–85). New York: Oxford University Press. Maund, B. (2016). Review of Zed Adams, On the genealogy of colour: A case study in historicized conceptual analysis. Notre Dame Philosophical Reviews. Available at: http://ndpr.nd.edu/news/70511-on-the-genealogy-of-color-a-case-study-inhistoricized-conceptual-analysis/. Mausfeld, R. (2003). Colour’ as part of the format of different perceptual primitives: The dual coding of colour. In R. Mausfeld (Ed.), Colour perception: Mind and the physical world (pp. 381–436). Oxford: Oxford University Press. McDowell, J. (1985). Values and secondary qualities. In T. Honderich (Ed), Morality and objectivity: A tribute to J. L. Mackie (pp. 110–29). London: Routledge & Kegan Paul. McGinn, C. (1996). Another look at colour. Journal of Philosophy, 93(11), 537–53. Plunkett, D. (2016). Conceptual history, conceptual ethics, and the aims of inquiry: A framework for thinking about the relevance of the history/genealogy of concepts to normative inquiry. Ergo, 3(2), 27–64. Roberts, P., Andow, J., & Schmidtke, K. (2014). Colour relationalism and the real deliverances of introspection. Erkenntnis, 79(5), 1173–89. Roberts, P., & Schmidtke, K. (2019). Folk core beliefs about colour. Review of Philosophy and Psychology. Ruskin, J. (1876). The elements of drawing. New York: John Wiley and Sons. Schmaltz, T. M. (Ed.) (2014). Efficient causation: A history. New York: Oxford University Press. Strawson, G. (1989). Red and ‘Red’, Synthese, 78, 193–232. Travis, C. and M. Kalderon (2013). Oxford realism. In M. Beaney (Ed.), Oxford handbook of the history of analytic philosophy (pp. 489–517). Oxford: Oxford University Press. Watkins, M. (2002). Rediscovering colours: A study in Pollyanna realism. Dordrecht: Springer. White, A. R. (1961). The causal theory of perception, Part II. Proceedings of the Aristotelian Society, 153, 153–68. Wilson, M. (2006). Wandering significance: An essay on conceptual behavior. Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

7 Variation in Natural Kind Concepts Daniel Cohnitz and Jussi Haukioja

7.1 Introduction On a textbook internalist conception of meaning, concepts¹ seem to be very unstable. If a term’s meaning is determined by the speaker’s beliefs, associated (bundle of) descriptions, or dispositions to apply the term to objects in the world, it would seem that conceptual variation is quite commonplace, between speakers as well as between different temporal slices of the same speaker. We change our beliefs about all kinds of things relatively frequently, and the same holds for any descriptions we might associate with terms. Already Frege noticed that different speakers well might, and often probably do, associate different descriptions with a name. We know from empirical research that people vary in their dispositions to apply terms, between cultures, within cultures, and even interpersonally (for the latter, cf. Hampton & Passanisi, 2016, and Hampton, Chapter 4 in this volume). This kind of instability leads to well-known puzzles. How can speakers ever mean the same by their terms, if meaning is subject to such variation? Doesn’t that show that we speak past one another almost all the time? How is communication then possible? And, relatedly, how is rational theory change and scientific progress possible, if the terms of scientists in the past had a different meaning from those of scientists today? Externalism about meaning promises a solution to these problems. Meaning is not determined by our variable psychological states; instead, it is determined by external factors—perhaps facts about the underlying kinds and their essences or properties,² perhaps facts about the linguistic community as a whole. Facts of the ¹ We follow the convention of philosophers of language and will use ‘concept’ for the meaning of predicate expressions, which determine the reference (extension) of these expressions. If two expressions differ in extension, they must also differ in meaning. Because of that latter connection, philosophers often discuss variations in extension when discussing conceptual variation in this sense. By identifying concepts with meanings, it is left open whether they are psychological entities or closely related to such. Note that in psychological literature ‘concepts’ are often by definition psychological entities. ² It is sometimes assumed that externalism is already committed to the existence of essences, such that if there are no properties that could play the role of essences, then externalism is false (cf. Häggqvist & Wikforss, 2018). But this is a misunderstanding. First of all, essences might be conferred, and thus not require a metaphysical foundation (Sveinsdottir, 2008). Secondly, the dispositions of speakers to defer to whatever gives rise to the stereotypical properties of a natural kind in the actual Daniel Cohnitz and Jussi Haukioja, Variation in Natural Kind Concepts In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Daniel Cohnitz and Jussi Haukioja. DOI: 10.1093/oso/9780198803331.003.0008

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

129

latter kind might prevent us from talking past one another, facts of the former kind might moreover guarantee that we use terms with the same meaning even if our beliefs about their referents change radically over time. Externalism of this kind we will call first-order externalism (and it contrasts with first-order internalism as sketched above). But is first-order externalism true? How can we find out? What kind of facts would make it true? What kind of facts, in general, make metasemantic theories true or false? As we argue in Section 7.2, one can have two radically different views about this. One may think that the metasemantics of natural kind terms is determined by certain dispositional states of speakers (a view that we will call ‘metainternalism’) or hold that it is determined by speaker-external facts, such as certain metaphysical facts or perhaps supra-individual facts about the linguistic community as a whole (a view that we will call ‘meta-externalism’). In Section 7.2 we also argue that the latter view is implausible. Metasemantics must be tied to speaker dispositions of a certain kind in order to allow our metasemantic theory to be of explanatory value. Section 7.3 argues that this view has methodological implications. If certain dispositional states of speakers determine metasemantics, then we should inquire what these dispositions are, in order to find out whether the externalist metasemantic view sketched above is correct and can indeed solve the puzzles described. Investigating these dispositions can be done empirically. However, the kinds of experimental studies that have so far been performed to evaluate the truth of externalism are insufficient for deciding the matter. Section 7.4 discusses how existing methods could be improved.

7.2 Two Stories about Meta-metasemantics Putnam’s standard Twin Earth argument for externalism is very well known. We introduced the term ‘water’ in order to talk about a certain liquid we find in rivers, lakes, and oceans. The term then applies in our world (but also in other possible worlds) to liquids that share the same chemical composition as that of water. Thus, on Twin Earth, where the ‘watery stuff ’ is XYZ, we find no water. Compared to the externalist story about proper names, we have now two places where external factors determine the reference of a term like ‘water’. First, just as in the case of names, there is the causal-historical chain of usage that connects our world, are probably highly flexible and might well settle for bundles of properties, or disjunctions (or for whatever contemporary scientists, if well-informed, would consider the most salient realizers for the assumed stereotypical properties, or for some of them). The point of externalism is precisely that the properties (be they essential, intrinsic, or highly disjunctive) that do determine the extension of a natural kind term are not necessarily known to a speaker who uses a natural kind term with that meaning.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

130

    

usage of the word with ourworldly water. Just as with names, we don’t need to know anything about this chain in order for ‘water’ to refer to water. But the extension of ‘water’ is also determined by the chemical composition of water.³ Whatever shares this microstructure is water, even if we don’t know what that microstructure is and even if we don’t know that there is such a thing as chemical composition. As Putnam argued in the case of gold: It is possible (and let us suppose it to be the case) that just as there were pieces of metal which could not have been determined not to be gold prior to Archimedes, so there were or are pieces of metal which could not have been determined not to be gold in Archimedes’ day, but which we can distinguish from gold quite easily with modern techniques. Let X be such a piece of metal. Clearly X does not lie in the extension of ‘gold’ in standard English; my view is that it did not lie in the extension of ‘χρυςος’ in Attic Greek, either, although an ancient Greek would have mistaken X for gold (or, rather, χρυςος). (Putnam, 1975, p. 235)

But what makes it the case that ‘gold’ or ‘χρυςος’ have their extension determined in this way? After all, not all terms work this way. The extension of ‘bachelor’ is not determined by microstructure and neither is the extension of ‘pencil’, or countless other general terms. What is it about natural kind terms that makes them behave differently? The possible options for answering that question can also be best illustrated with a case from Putnam’s Meaning of ‘Meaning’: Imagine that we someday discover that pencils are organisms. We cut them open and examine them under the electron microscope, and we see the almost invisible tracery of nerves and other organs. We spy upon them, and we see them spawn, and we see the offspring grow into full-grown pencils. We discover that these organisms are not imitating other (artifactual) pencils—there are not and never were any pencils except these organisms. It is strange, to be sure, that there is lettering on many of these organisms—e.g. BONDED Grants DELUXE made in U.S.A. No. 2—perhaps they are intelligent organisms, and this is their form of camouflage. (Putnam, 1975, p. 242)

As we said above, the extension of ‘pencil’ seems to be determined by a certain functional description. According to Wikipedia, pencils are artefacts that one can write with in virtue of a ‘narrow, solid pigment core inside a protective casing

³ Or to whatever it is that accounts for the stereotypical properties of water in the actual world. See the previous footnote for a brief explanation why externalism is not committed to the view that the extension of ‘water’ is determined by some unique microstructure. The examples we use pretend that there is such a unique microstructure, for ease of exposition.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

131

which prevents the core from being broken or leaving marks on the user’s hand during use’. Could we then discover that pencils are organisms? Putnam’s own answer is that this is indeed an epistemic (although not metaphysical) possibility, because ‘we intend to refer to whatever has the same nature as the normal examples of the local pencils in the actual world’ (Putnam, 1975, p. 243). But what if this is a misdescription of our intention? What if we indeed intend to refer to whatever serves the same function thanks to a similar mechanism (by having a narrow, solid pigment core inside a protective casing which prevents the core from being broken or leaving marks on the user’s hand during use, for example)? Could the fact that what we call ‘pencils’ all turn out to be organisms reveal that ‘pencil’ does and always did refer to whatever shares the same DNA with pencils (whether or not you can write with them in other possible worlds), or would it still be the case that ‘pencil’ refers in this and other possible worlds to objects with the help of which one can write on paper and similar surfaces? In other words, is the metasemantic question of whether an expression is a natural kind term or a functional kind term a matter of our intentions and dispositions to use the term, or a matter of metaphysics? One can be an externalist or an internalist about what determines metasemantics. In earlier work (Cohnitz & Haukioja, 2013) we labelled these opposing positions ‘meta-internalism’ and ‘meta-externalism’: Meta-Internalism: How a linguistic expression E in an utterance U by a speaker S refers and which theory of reference is true of E is determined by the individual psychological states of S at the time of U. Meta-Externalism: How a linguistic expression E in an utterance U by a speaker S refers and which theory of reference is true of E is not determined by the individual psychological states of S at the time of U.

‘How a linguistic expression refers . . . ’ is our abbreviation for the different metasemantic options: is the expression a natural kind term (such that its extension in other possible worlds is determined by having the same nature as the things in its extension in the actual world), or is it an artefact/functional kind term (such that its extension in other possible worlds is determined by having the same function in these worlds as the things in its extension in the actual world), and so on. We take it that most authors, also those who favour a first-order externalist view on concepts, subscribe to meta-internalism. The typical story is that it is because we are disposed to defer to underlying microstructure or the knowledge of experts in our usage of a term, that our term refers via this microstructure, or according to that expert judgment. We have just seen this in the quote by Putnam; he seems to hold that it is due to our ‘intentions to refer to whatever has the same

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

132

    

nature as the normal examples of the local pencils in the actual world’ that ‘pencil’ refers to a natural kind, if what we thought were pencils turned out to be organisms. Thus, the question of whether ‘pencil’ is a natural kind term or a functional kind term (or whether it has a more complex, conditional structure) is then a question of whether speakers indeed have those dispositions or intentions. Such questions should then, in principle, be accessible to empirical investigation.

7.2.1 Two Examples of Meta-externalism for Kind Terms Since meta-internalism seems to be relatively common, one might wonder whether there are actually any instances of meta-externalism. We discuss two examples. On both examples, the deferential dispositions of speakers come out being largely irrelevant for the metasemantics of natural kind terms. Consequently, it becomes also less clear how one should go about determining the actual semantics for such terms.

7.2.1.1 Reference Magnetism Hilary Putnam famously argued that we can find for any global theory infinitely many interpretations that satisfy that theory. This is Putnam’s model theoretic argument against ‘global descriptivism’. The theoretical role of an expression, even in a global theory, doesn’t fix the extension for that expression. Moreover, Putnam added, any constraint other than truth (i.e. satisfaction of the theory by a model) would be just another piece of theory, hence not a way out of the embarrassment of riches. David Lewis noticed in his ‘Putnam’s Paradox’ (Lewis, 1984) that the last inference doesn’t hold. Putnam’s argument seems to overlook the fact that there is a difference between a theory satisfying a constraint and the constraint holding because we accept a theory stating it: [An additional constraint] C is not to be imposed just by accepting C-theory. That is a misunderstanding of what C is. The constraint is not that an intended interpretation must somehow make our account of C come out true. The constraint is that an intended interpretation must conform to C itself.

(Lewis, 1984, p. 225) But Lewis was also aware that this observation alone doesn’t help. The distinction between satisfying C-theory vs conforming to C can only be made if constraint C is not established merely by stipulation, by our referential intentions: The main lesson of Putnam’s Paradox, I take it, is that this purely voluntaristic view of reference [viz. that whatever theory of reference is true, it is true because of our referential intentions] leads to disaster. If it were right, any proposed

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

133

constraint would be just more theory. Because the stipulation would be something we say or think, something we thereby add to total theory. Referring isn’t just something we do. What we say and think not only doesn’t settle what we refer to; it doesn’t even settle the prior question of how it is to be settled what we refer to. Meanings—as the saying goes—just ain’t in the head.

(Lewis, 1984, p. 226) As the last sentence of the quote already indicates, Lewis concluded from the observation that the constraint C (whatever it is) shouldn’t hold just because we intend it to hold. The constraint must be something that is entirely independent of us and what is in our heads. Lewis then suggests that the constraint is externally provided by the objective naturalness of certain elite properties. These elite properties are, for Lewis, fundamental physical properties, like mass, charge, quark colour, and flavour. Obviously, most of our words do not refer to these, but Lewis believes that they do refer to derivatively eligible referents that are connected to the elite properties via chains of definitions. Properties that stand in such chains to the elite properties of fundamental physics are reference magnets: because they carve nature at the joints better than rival interpretations, they become the referents of our terms. Moreover, they become the referents of our terms, independently of our intentions concerning what our terms should refer to: It is not to be said that our theorising makes the joints at which the world is to be carved. That way lies the ‘just more theory trap’. Putnam would say: ‘very well, formulate your theory of “objective joints in nature” [ . . . ] and stipulate if you will that your referents are to be “eligible”. But total theory with this addition goes the way of all theory, it is satisfiable with the greatest ease in countless ways.’ [ . . . ] No: the proposed constraint is that referents are eligible, [ . . . ] not that the referents of ‘cat’ etc., are to be included among the referents of ‘eligible’.

(Lewis, 1984, p. 228) Now, of course, our intention might be that ‘cat’ refers to an eligible referent, like a natural kind, but the fact that ‘cat’ refers to a natural kind is not because of these intentions; rather, it is solely because there is a natural kind and the assignment of this kind to the word ‘cat’ is part of an interpretation that maximizes the eligibility of referents overall. It is clear why this is an instance of what we have defined as meta-externalism. Again, as Lewis says: ‘what we say and think not only doesn’t settle what we refer to; it doesn’t even settle the prior question of how it is to be settled what we refer to’ (Lewis, 1984, p. 226, our emphasis).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

134

    

7.2.1.2 Reference Communitarianism A second view that would count as meta-externalist takes less issue with the metainternalist idea that semantic properties (like the metasemantics of an expressions) must ultimately be grounded in the way speakers use a language (which seems to be a very common idea anyway), but rather with the individualism that we built into our definition. We said that it should be a matter of the dispositions/ intentions of a particular speaker at the time of a particular utterance that determines the metasemantics of the expressions in that utterance. This is denied by views which would accept that these intentions occur at other times (for example, at some point in the past when the speaker first acquired the relevant terms, cf. Devitt, 2011). It is also denied by views which would hold that semantic facts supervene on the usage of expressions in the linguistic community as a whole and hold that the entrance ticket to the linguistic community is obtained through, for example, linguistic interaction with that linguistic community (Williamson, 2007). Thus, if the usage in the linguistic community is such that ‘pencil’ would turn out to be a natural kind term, if we made the discovery that Putnam describes, then the unwillingness to change usage and retract the former claims about pencils being artefacts, etc., of any individual speaker who previously engaged linguistically with the linguistic community that Putnam envisages in his example, wouldn’t betray that speaker’s deviant meaning of ‘pencil’ in her idiolect, but simply display that speaker’s ignorance of the right metasemantics of that term.

7.2.2 Against Meta-externalism Language, especially when used in linguistic communication, helps us to coordinate our activities. It allows us to share knowledge about the world, on the basis of which we can then take action, but it also allows us to make and share plans that we can then carry out together. That our words express the concepts they do will somehow have to be a systematic part of any explanation of the role of language in this. The problem with meta-externalism is that it makes that systematic contribution superfluous.⁴ The impact of language on our joint plans and activities (but also on our individual beliefs) is (at least) a matter of our dispositions to react to information of certain kinds. Let us assume that the relevant dispositions of a linguistic community L are in harmony. Their linguistic interaction, their dispositions to correct and change their usage of expressions in the light of new information is as if the expression e in their language was, say, a functional kind term, and had

⁴ The argument below is presented in much more detail in Cohnitz & Haukioja (2013).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

135

functional kind f as its referent. Let us call f the ‘schmeferent’ of e. Let us assume though that some meta-externalist story is true for e. And, in fact, the dispositions of our harmonious linguistic community are actually out of step with e’s actual referent. Let us assume, for example, that, in fact, e is a natural kind term and the natural kind n is the referent of e. If we want to explain how the speakers of our hypothetical linguistic community manage to coordinate their plans and actions with the help of e, we will need to talk about the schmeference of e. There will be no reason to bring in the reference of e. The latter will only be reasonable if the relevant dispositions of the speakers of the linguistic community in question are in pre-established harmony with the meta-externalistically determined facts about reference. But even when they are, these latter facts will not contribute to the explanation of the coordination achieved through linguistic communication.⁵ This is always so on the meta-externalist picture. Because the relevant metaexternalistic facts are in the past, in the future, or outside your head, they might not have any impact on how you coordinate with an expression (i.e. how you adapt your linguistic and non-linguistic behaviour and how you update your beliefs in reaction to information phrased in terms of that expression). There might be more than one speaker for whom that is so, there might be more than two, and, ultimately, this might be so for all members of a linguistic community. In the latter, extreme case it is obvious that reference is not contributing to an explanation of linguistic communication. The cases in which meta-externalist facts seem to make a contribution are those in which schmeference and reference coincide.

7.2.3 Meta-internalism of the Dispositionalist Variety A dispositionalist version of meta-internalism claims that the metasemantics of a referring expression, as used by a speaker, is supervenient on that speaker’s dispositions to apply and interpret the expression in question.⁶ The range of dispositions that will have to be included in the supervenience base is, however, ⁵ In his 2011 work, when defending his own version of reference magnetism, Ted Sider claims without argument that ‘clearly’ the meta-externalistically determined referents of expressions would have to play a role in semantic explanations of thought, behavior and understanding (Sider, 2011, p. 28). Not only isn’t this ‘clearly’ so, this just isn’t so. ⁶ Above we talked a bit vaguely about dispositions and intentions and somewhat pretended that there is a way to characterize individual mental states in a purely internalist way. One does not need to endorse individualism or internalism about the mental in order to make sense of meta-internalism or the meta-internalism/externalism distinction. From our point of view it seems to make most sense to be a (first-order) externalist about the mental and drop the reference to intentions. The intention to use ‘water’ for water and ‘pencil’ for pencils will then be intentions to use either expression as a natural kind term, depending on how the world turns out to be. Meta-internalism and meta-externalism defined in terms of such intentions just collapse into one another.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

136

    

quite wide. It will of course include the speakers’ dispositions to apply the term to entities in the world, including her dispositions to apply and interpret the term in conditions considered as non-actual. However, the supervenience base will also have to include the speaker’s dispositions to revise and reconsider her own applications of the term. There are at least three kinds of such corrective dispositions that will have to be considered. The first two kinds of corrective dispositions, when present, can make it the case that a given expression is to be given an externalist metasemantics. Dispositions of these kinds are the speaker’s dispositions to revise her application and interpretation in response to empirical information about her surroundings, and in response to information about how relevant experts in one’s speech community use the same expression. In our view, when a first-order externalist theory of reference is true of a term, it is made true by the fact that the relevant speakers are, in their application of the term, suitably sensitive to empirical information concerning such facts. The third kind of corrective disposition is more general, and has to be brought in to account for the distinction between correct and incorrect applications of terms, regardless of whether a first-order internalist or externalist view is true of the term in question. The first kind of corrective disposition is sensitive to empirical information about the entities that putatively belong in the extension of a term. When speakers have such corrective dispositions, and they are systematic enough, physical externalism (or natural kind externalism) is true of a term. To illustrate, let us look at some examples. For some expressions, such as ‘bachelor’, our patterns and dispositions of application and interpretation appear to be unaffected by contingent features of the natural world around us. The properties that speakers— individually or collectively—associate with the expression are taken to be sufficient for determining whether the expression applies to a given individual or not. But with other expressions we have dispositions to ‘shift the burden’ of determining their applicability partly to external factors. In the case of ‘water’, for example, we have dispositions to evaluate the correctness of actual and counterfactual applications of the term according to whether or not the term is applied to samples which share the underlying structure of the substance that is causally connected in the appropriate way to our actual usage of ‘water’. We also have dispositions to re-evaluate our application of such terms in the face of new empirical information about what the world is like. To illustrate, suppose that we took a representative sample of bachelors, studied them empirically, and found out that every single one of them has a certain neural

Both notions should thus be defined in terms of dispositions to apply terms to objects and to correct previous applications in the light of new information. The characterization of these dispositions can refer to the things in the world that they are sensitive to, it can also be applied to mental types (expressions in the lingua mentis, if you like).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

137

structure—call it N—while no individuals outside the sample have N. Would we then start to categorize people as bachelors, in the actual world as well as in other possible worlds, according to whether or not they have neural structure N or not? No, we wouldn’t: we would go on categorizing people as bachelors, in the actual world as well as in counterfactual ones, according to their age, gender, and marital status. Or, consider another version: suppose that we found that, say, almost all unmarried adult males were found to have N, while a tiny proportion of them do not; at the same time, we find N present in a very small number of married females. Would we then revise our categorization of this small minority of men as bachelors and instead include the married females? No, we wouldn’t. But were we to make a similar discovery about the golden stuff, say that we find some of the metal we categorised as gold to have a different atomic number than 99.9 per cent of the rest, while a small sample of a greenish looking metal turns out to have the same atomic number as the other 99.9 per cent, we would, we think, revise our categorization of the 0.1 per cent and consider the greenish stuff as gold. In the above, we have assumed that ordinary speakers in fact do have the kinds of dispositions on which externalist thought experiments typically turn. Of course, this is an empirical assumption, and one that can be (and has been) questioned. We return to this issue in Section 7.3. What matters here is that, on our dispositionalist meta-internalist view, such dispositions are what makes first-order externalism true (in contrast to meta-externalism, where they would have to be thought of as tracking, more or less successfully, an independently determined semantic reality). On our view, the truth or falsity of natural kind externalism turns on the existence and systematicity of precisely such dispositions. The second kind of corrective disposition is sensitive to how other speakers use the expression in question—in particular, these are dispositions to defer to other, more expert, speakers. When present, and when systematic enough, such corrective dispositions make it the case that social externalism is true of an expression. For example, I can refer to elm trees with my term ‘elm’, even though I cannot tell them apart from beech trees, on the basis of my dispositions to defer to people who can actually tell elms apart from other trees (e.g. botanists or gardeners). Should I find that my classification of trees into elms and beeches differs from that of an expert, I would be disposed to immediately revise my earlier application of the terms and align my usage with that of the experts. The third kind of corrective disposition is, as noted above, more general in that such dispositions need to be assumed to be present regardless of issues having to do with internalism and externalism. These are dispositions to reconsider and possibly take back one’s own earlier applications of a term, based on closer consideration of the issue at hand, and not prompted by new information concerning one’s surroundings or the linguistic usage of experts. All speakers make mistakes every now and then: not all instances of applying and interpreting expressions are correct. But theories of reference try to account for correct

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

138

    

application and interpretation. One cannot, for example, argue against the view that the reference of ‘bachelor’ is determined by the associated properties of being unmarried, adult, and male, simply by pointing out that speakers are, in some situations, disposed to call women, or married men, bachelors. In order for such an argument to have any force, it would have to be established that such applications are, in fact, correct. As we have learned from the extensive debates concerning the Kripkensteinian problem of rule-following, it is by no means obvious how the distinction between correct and incorrect instances of application should be drawn. We cannot hope to settle this huge issue here, but it seems clear to us that corrective dispositions will have to have a central role here, either as constituents of correctness, or as paradigmatic evidence of incorrect use. Suppose, for example, that two speakers, A and B, both apply ‘bachelor’ to married men, on a few isolated occasions. One of the speakers, A, would instantly take back her classifications of these men as bachelors, were she to reconsider the cases (for example, in response to other speakers’ challenging her use). Speaker B, on the other hand, would be unmoved, not at all disposed to take back her classifications, and insist that these men are bachelors, even though they are married. Obviously, in such a case, A’s applications of ‘bachelor’ to married men would not pose a serious threat to the claim that the reference of ‘bachelor’, as used by A, is determined by the associated properties of being unmarried, adult, and male. Not so in the case of speaker B: if she really were to insist on calling some unmarried men bachelors, and not show any inclination to correct herself, we should conclude that the reference of ‘bachelor’, as B uses it, is not determined in the same way; she is not using the term with its standard meaning. Obviously, a lot more should be said about all three kinds of corrective dispositions. The above sketch is, however, sufficient for our aims in this chapter. The resulting view is one on which the reference of a given linguistic expression, as used by speaker S, is determined by how S is disposed to use the expression, how she is disposed to re-evaluate and revise her application in the light of new information about the world and about other speakers, and in the light of closer inspection of her own usage. Externalism and internalism concerning a given expression are then to be thought of views about which kinds of new information can lead to the relevant kinds of re-evaluations and revisions.⁷ ⁷ This has a lot in common with the view recently proposed by Michael Johnson and Jennifer Nado, independently. On their view, ‘a linguistic expression E means some object, property, kind, relation, etc., X, in the mouth of speaker S, in virtue of the fact that S would be disposed to apply E to X if S had all the relevant information’ (Johnson & Nado, 2014, p. 81), where ‘relevant information’ is said to consist of ‘the facts F that would, were S apprised of F, influence S’s dispositions to apply E’ (ibid.). The difference between their view and ours appears to be mainly one of emphasis—we focus on S’s dispositions to change her dispositions to apply E, whereas they focus on the end-result of such (idealized) dispositions. A closer examination of the relationships between our views is, however, a topic for another occasion.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

139

7.3 What Empirical Data Has Shown So Far Let us now return to the issue of conceptual variation. If meta-externalism were true, systematic variation in how speakers apply natural kind terms would, perhaps, not be a huge problem. For example, if one accepted reference magnetism, our natural kind terms would refer to the most natural of the plausible candidate referents (however exactly the class of such candidates is determined), and divergent applications could simply be dismissed. But we reject metaexternalism, so we need to meet the challenge head on. On our meta-internalist view, referential relations are determined by competent speakers’ patterns of dispositions to apply the term to objects and to revise his or her applications in the light of new information. Thus, to appeal to first-order externalism about natural kind terms to explain successful communication and scientific progress, it has to be established that our dispositions in fact do support the purported referential stability of our natural kind terms. And, since the nature of our actual dispositions is clearly an a posteriori matter, such stability could be denied on empirical grounds, should it turn out that there is considerable variation in how we apply natural kind terms. And indeed, this is precisely what some experimental studies claim to have shown. A number of theorists have claimed that, due to such variation, we should reject the (first-order) externalist view of natural kind terms that is predominant in philosophy of language, and replace it with a ‘representational change’ theory according to which the reference of natural kind terms is context-dependent (Braisby et al., 1996), a ‘hybrid’ theory that includes both causal-historical and descriptive factors (Genone & Lombrozo, 2012), or an ‘ambiguity’ theory according to which natural kind terms can take on a causal-historical reading or a descriptive reading, depending on conversational setting (Nichols et al., 2015).⁸ According to these authors, the empirical results that establish that there is widespread variation in speakers’ use of natural kind terms all have to do with test subjects’ verbal responses to questions concerning imagined scenarios that are fairly closely modelled after the standard externalist thought experiments by Putnam and Burge. The questions take a variety of forms: in some cases, subjects were asked whether they agree with existence claims (Braisby et al., 1996, Nichols et al., 2015) or classifications of individuals as belonging to a natural kind (Braisby et al., 1996) in response to Twin Earth-style scenarios, and in others they were asked whether two speakers in Burge-inspired scenarios were using terms (such as ‘tyleritis’, a term denoting a disease) co-referentially or not (Genone & Lombrozo, 2012).

⁸ A fourth study, by Jylkkä et al. (2009) finds roughly similar results, but does not take them as evidence for an ambiguity view. We briefly return to the results of this study in Section 7.4.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

140

    

All studies show a similar pattern: in some situations, subjects were using and interpreting natural kind terms as a causal-historical theory would predict, while in others their usage was in line with a descriptivist theory. Moreover, none of the studies found subsets of subjects using terms exclusively according to one of the theories; all subjects appeared to be switching between the two patterns, even to the point that they accepted apparently contradictory pairs of sentences as true (Braisby et al., 1996, Nichols et al., 2015). As indicated above, the authors of the three studies come to different conclusions regarding how to best explain the variation. We will not enter this discussion here (cf. Martí, 2015 for discussion of the relative merits of a ‘hybrid’ theory vs an ambiguity theory). Rather, we argue that it is at best premature to conclude, on the basis of the kinds of results that we have so far seen, that a mainstream externalist, causal-historical theory of reference for natural kind terms should be replaced by a theory that posits systematic ambiguity, or context-dependence. The variation is certainly interesting, and deserves closer attention, but at the moment it is far too early to conclude anything about the fate of the causal-historical theory, and of (first-order) externalism more generally. First of all, the studies do not look directly at the test subjects’ dispositions to apply and interpret natural kind terms. Asking subjects to make metalinguistic judgements about the reference of terms, or about co-reference, or asking them to make truth value judgments, are plausibly sources of indirect data about their linguistic dispositions, but they introduce various possible sources of error (cf. Cohnitz & Haukioja, 2015). We return to this issue in Section 7.4. Secondly, the studies only inform us about the subjects’ initial inclinations to apply (or, form metalinguistic judgements about others’ application of) natural kind terms. While the variation in such judgements is, apparently, robust and in need of explanation, so far we have no reason to think of it as data to be explained, rather than as the discovery of a range of circumstances where ordinary speakers are prone to a certain kind of (fairly systematic) error. That is, we are not forced by the data to conclude that natural kind terms are, in some contexts, correctly applied as the descriptivist theory would have it. The data is equally consistent with the conclusion that, in certain contexts, speakers are inclined to classify things incorrectly as belonging, or not belonging, to a natural kind, or to make mistaken judgements about existence claims using natural kind terms, and so on. To emphasize, we are not claiming that the causal-historical theorist can simply choose to ignore the variation that has been found, by dismissing the nonexternalist applications as erroneous. On the contrary, we think it has to be taken seriously, and that it cries out for an explanation. But it is not obvious that the variation should be taken at face value. We know that speakers make mistakes; we know that speakers can be prone to making systematic mistakes in their application of language. The data that we have been presented with cannot, by itself, determine whether the variation is evidence of systematic ambiguity, or

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

141

of systematic speaker errors. To make progress, we need to reach a better understanding of what makes some applications of terms correct and others incorrect. To illustrate, and to take the first steps towards developing our own view, consider a bit of anecdotal evidence. One of the authors of this paper vividly remembers his first encounter—as a first-year philosophy student—with the Twin Earth thought experiment. His first reaction (after a brief period of initial puzzlement) was to say that XYZ on Twin Earth is water. But very soon thereafter (possibly in response to a comment by an older student—here, recollections are unfortunately quite hazy) he realized that XYZ on Twin Earth is not water (given, of course, that the watery substance on Earth is H₂O and not XYZ). This realization came with a clear sense of having made an error in the initial judgment concerning the thought experiment. That is, he wanted to take back his earlier application of ‘water’ to the imaginary substance on Twin Earth; he judged that, contrary to his first reaction, ‘water’ does not correctly apply to XYZ. Based on our experience as teachers, and based on an informal survey of our colleagues, this experience is far from uncommon. But note that had the past timeslice of one of the authors been, prior to this realization, transported forward in time to take part in one of the empirical studies discussed above, he would very likely have contributed towards the result that natural kind terms are contextdependent, ambiguous, or governed by a hybrid theory. This suggests to us that there is a very real possibility that a number of the subjects who were used in the experiments referred to earlier were similar in this respect. Again, we are not claiming that this is the obvious explanation of the variation in judgments, and that the results can therefore be ignored. We are merely claiming that this is one possible explanation, and that the issue deserves further study. On the dispositionalist meta-internalist view we described above, the correctness or incorrectness of particular applications of terms is determined precisely by the kinds of second-order dispositions to correct one’s own past usage of a term that the above anecdote illustrates. Such dispositions can be studied experimentally, but to do that, experimental semantics would have to distance itself further from the thought-experiment paradigm that has so far been dominant. The kinds of thought experiments that have figured prominently in the internalism/externalism debates are philosophers’ tools: experiments formulated by philosophers for other philosophers who are already familiar with the competing theories and who are thereby able to focus on the relevant features of the scenarios that are (often sketchily) presented. Gathering first reactions to such scenarios from nonphilosophers is unlikely to get us very far. To get more informative data, more sophisticated experimental setups should be developed, but it is not at all clear how we should go about this.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

142

    

7.4 What Would Need to be Done We argued above that, in addition to a speaker’s dispositions to apply terms to things in the world, three kinds of corrective dispositions are crucial for determining the meanings of her terms. To make further progress, experimental work on natural kind terms should focus on such corrective dispositions, in addition to the application dispositions studied in the experimental work referred to above. The two first types of corrective dispositions had to do with speakers’ inclinations to reconsider and revise her application of a term in response to new empirical information about the putative extension of the term, and about other speakers. As we pointed out, if physical externalism and social externalism are true of natural kind terms, they are made true by the presence and systematicity of such dispositions in speakers using the term. Experimental work that aims to settle whether one or another form of externalism holds for natural kind terms, as used by ordinary speakers, should then focus on the relevant kinds of corrective dispositions. First, concerning physical externalism: are speakers, in fact, disposed to reconsider and revise their applications of natural kind terms, should it turn out that they were wrong about the underlying features of things putatively belonging to the extension of the term? The only existing study that tries to tackle this question directly is Jylkkä et al. (2009). In this study, test subjects were presented with Twin Earth-style scenarios in two stages. In the first stage, they were given a story about newly discovered samples with a similar appearance to known samples of a natural kind, and told either that the new samples are found to have a similar underlying nature as the old samples, or that they are found to have a different underlying nature, despite the similarity in appearance. At this point, subjects were asked to judge whether the new samples belong to the same natural kind as the old ones, testing their dispositions to apply natural kind terms. At the second stage, the subjects were told that the previously given information concerning the underlying nature of the old samples has turned out to be incorrect: in the scenarios where the old and new samples were first thought to differ in underlying nature, they turn out to share the same underlying nature, and vice versa. At this point, the subjects were asked whether they still agreed with their earlier classification of the new samples or not, testing the kinds of corrective dispositions we have claimed to make physical externalism true. The results gave some support for physical externalism: most subjects were disposed to revise their earlier judgment in the light of the new empirical information presented in the second stage of the scenarios. The results were, however, inconclusive, and a similar split between seemingly externalist and internalist responses was found in this study as well. Second, concerning social externalism: are speakers, in fact, disposed to reconsider and revise their applications of natural kind terms, should it turn out that the

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

143

relevant experts in their linguistic community would not apply a term to things they are disposed to apply it to, or vice versa? Third, concerning correction based on closer consideration: would subjects, for example, persist in the kinds of judgments that have been taken as evidence for systematic ambiguity, or context-dependence, were they encouraged to think more about the cases, or exposed to more cases of the same kind? Or, would they eventually converge on judgments that are more clearly in line with either the causal-historical theory, or descriptivism? Note further that the three kinds of corrective disposition can work in tandem. If, for example, subjects who are first exposed to scenarios similar to those used by Jylkkä et al. (testing the presence of the first kind of corrective dispositions), and who had responses similar to the ones reported in that study, were further to be told that the relevant experts in their linguistic community would uniformly categorize entities in one way, would they revise their application (testing the presence of the second kind)? And finally, would these judgments survive closer examination of the cases, by the subjects (testing the presence of the third kind)? These are, then, the kinds of dispositions that should be looked at in order to be able to say, with any confidence, whether one or another form of externalism is true of natural kind terms, and whether there is substantial variation in natural kind concepts. So far, we have not said anything in this section about how such dispositions are to be studied. First, it should be a consequence of the meta-internalist view presented that one should favour evidence about linguistic usage over evidence about (tacit) metalinguistic beliefs (Cohnitz & Haukioja, 2015). In other words, what should primarily be studied is actual production and interpretation of linguistic expressions of the relevant kind. In many studies so far, test subjects have been asked to make metalinguistic judgments of some form or other, which is at least inviting certain forms of systematic mistakes (cf. Cohnitz & Haukioja, 2015). We take it that the interpretation of linguistic expressions, as well as the production of utterances are both prone to be influenced by contextual, pragmatic factors. We might interpret others by taking their perspective, instead of interpreting them simply in accordance with what their expressions mean, and the same can, of course, happen in an experiment in which we ask the test subject to tell us how to interpret certain expressions in utterances of third parties (Sytsma & Livengood, 2011). These potential confounds can be avoided by either moving to neutral contexts and to studies of production (this could be realized through studies of elicited production, self-paced reading, eye-tracking, etc.; for some suggestions along these lines, see Cohnitz, 2015). Thus, secondly, one should favour production in contexts that are unlikely to be tainted by pragmatic factors. Since theories of reference concern the use of expressions already introduced into the linguistic community, and our corrective dispositions with respect to

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

144

    

these, experiments should, thirdly, ideally use entrenched expressions. Most studies conducted so far instead introduce new expressions⁹ (and then present conflicting information about the referents of these new expressions), but this at best tests the corrective dispositions of speakers at the point of reference fixation of a new expression, rather than how an already introduced expression refers. The worry is that with new expressions it will be difficult to distinguish situations in which a speaker changes her usage because she believes that she was mistaken about its meaning and situations in which she changes her usage because she is using the expression in question with a deferential meaning.¹⁰ However, using entrenched expressions does bring some additional complications with it. In a pilot study that led to Jylkkä et al. (2009), Jylkkä, Railo, & Haukioja used a similar setup to the one used in the published work (described earlier), but with entrenched natural kind terms in some of the scenarios, including a variant of Putnam’s Twin Earth case. The subjects’ judgments for the entrenched expressions were highly variable, and did not exhibit clear patterns. In informal post-experiment interviews with the subjects, it became clear that the subjects were very reluctant to go along with the stories and to accept them as true when they featured familiar natural kinds such as water. The best way to make sense of a story where ‘scientists have found out that the clear, odourless liquid in the lakes and rivers is not H₂O but XYZ’ was, for many subjects, to assume that the scientists had suddenly lost their minds, rather than to bracket all their empirical knowledge about the world.¹¹ Needless to say, such an interpretation of the scenario made their responses worthless as data about externalism. Jylkkä et al. resolved this problem by using only invented examples, but for the reasons mentioned above, it would be ideal to find other ways of circumventing this problem. Fourth, since meta-internalism considers reference not just to supervene on usage, but on counterfactually robust usage, i.e. usage that includes our dispositions to correct what we perceive as mistakes, tests should include possibilities for self-correction. ⁹ Of the studies mentioned in the previous section, only Braisby et al. (1996) uses entrenched rather than invented natural kind terms. ¹⁰ There is, of course, a general problem of distinguishing cases in which a speaker changes how she applies an expression, because she uses the expression with a deferential meaning and received new information about its proper application, and cases in which a speaker changes the meaning of one of the expressions in her repertoire, not because she used that expression with a deferential meaning but because she wants to align the way she uses the language with the wider linguistic community to which she belongs. The theoretical distinction between the two situations is clear: only in the former type of situation will previous applications of the expression now be considered mistaken (such that they were mistakes all along). To tease this out empirically is not easy, but we don’t see why it should be impossible. ¹¹ We should add that when the reluctant subjects were explicitly told that they should accept the improbable story as true, they had no problems doing so. This reaffirms the point made earlier: the classic thought experiments were written by philosophers, for other philosophers who already know what is at stake. Studies on non-philosophers where such stories are used only with minor variations are unlikely to give us very informative data.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

145

7.5 Conclusions Section 7.4 formulated what we consider to be desiderata for empirical investigations into reference. We find it difficult to even sketch concrete suggestions how these desiderata could be met by specific experimental set-ups. Reference is an abstraction from a complex set of dispositions which together constitute our use of language. Studying such an abstraction empirically is difficult, but we don’t think that it is impossible. Although we argued that the empirical results so far do not cast serious doubt on externalist theories of reference, we do nevertheless think that reference is an empirical phenomenon, and should be approached by empirical methods.

References Braisby, N., Franks, B., & Hampton, J. (1996). Essentialism, word use, and concepts. Cognition, 59, 247–74. Cohnitz, D. (2015). The metaphilosophy of language. In J. Haukioja (Ed.), Advances in experimental philosophy of language (pp. 85–108). London: Bloomsbury Academic. Cohnitz, D., & Haukioja, J. (2013). Meta-externalism vs. meta-internalism in the study of reference. Australasian Journal of Philosophy, 91, 475–500. Cohnitz, D., & Haukioja, J. (2015). Intuitions in philosophical semantics. Erkenntnis, 80, 617–41. Devitt, M. (2011). Deference and the use theory. ProtoSociology, 27, 196–211. Genone, J., & Lombrozo, T. (2012). Concept possession, experimental semantics, and hybrid theories of reference. Philosophical Psychology, 25, 717–42. Hampton, J. A., & Passanisi, A. (2016). When intensions don’t map onto extensions: Individual differences in conceptualization. Journal of Experimental Psychology: Learning, Memory, and Cognition, 42, 505–23. Häggqvist, S., & Wikforss, Å. (2018). Natural kinds and natural kind terms: Myth and reality. British Journal for The Philosophy of Science, 69, 911–33. Johnson, M., & Nado, J. E. (2014). Moderate intuitionism: A metasemantic account. In A. R. Booth & D. Rowbottom (Eds.), Intuitions (pp. 68–90). Oxford: Oxford University Press. Jylkkä, J., Railo, H., & Haukioja, J. (2009). Psychological essentialism and semantic externalism in lay speakers’ language use. Philosophical Psychology, 22, 37–60. Lewis, D. (1984). Putnam’s paradox. Australasian Journal of Philosophy, 62, 221–36. 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 (pp. 157–72). London: Bloomsbury Academic. Nichols, S., Pinillos, A., & Mallon, R. (2015). Ambiguous reference. Mind, 125, 145–75.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

146

    

Putnam, H. (1975). The meaning of ‘meaning’. Philosophical papers, Vol. 2: Mind, language, and reality (pp. 215–71). Cambridge: Cambridge University Press. Sveinsdóttir, Á. (2008). Essentiality conferred. Philosophical Studies, 140, 135–48. Sytsma, J., & Livengood, J. (2011). A new perspective concerning experiments on semantic intuitions. Australasian Journal of Philosophy, 89, 315–32. Williamson, T. (2007). The philosophy of philosophy. Oxford: Blackwell Publishing.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

PART II

TO SHIFT A CONCEPT: CONCEPTUAL REVOLUTION, AMELIORATION, AND PERVERSION

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

8 Conceptual Revolution Joshua Glasgow

8.1 Introduction What terms mean and refer to can change. Every term acquired the meaning or reference it has. Not long ago, the term ‘sexual harassment’ was unavailable to English speakers, and with its introduction came meaning and referent acquisition (Fricker, 2007: 149–52). Once established, new meanings can get added to preexisting words, in cases of conceptual expansion. ‘Sick’ used to just mean ill; now it—like ‘bad’—also means good. It seems we can now use ‘literally’ to mean figuratively. And then there are terms whose old meaning or referent is simply lost and taken over by a new one—outright replacement. Famously, ‘Madagascar’ had been the name for part of the Somali peninsula, not that big island off of Africa’s eastern shore. But Marco Polo got the location wrong. When map-makers subsequently doubled down on his mistake, the referent of that word underwent its change, and we now know Madagascar to be the island, not the peninsular region (Burgess, 2014). This instance of referential replacement is hardly unique. ‘Awful’ once meant inspiring awe. It now instead means really bad. Even our hallowed example of ‘bachelor’ meant young knight before it referred to a marital status, pausing along the way to acquire rank of university achievement of a certain kind. ‘Fantastic’ (from imaginary to wonderful), ‘silly’ (from worthy to foolish), ‘senile’ (from the senescent generally to those with dementia specifically), ‘tool,’ and on and on: language is littered with conceptual change. This chapter explores the criteria for conceptual change: what has to happen for a word to change meanings? On the broadly Carnapian picture to which I am drawn, concepts are fixed by some of our behavioural dispositions (Carnap, 1955; see also Chalmers, 2012). More specifically, what a term means and refers to depends in part on when we would be willing or unwilling to negotiate on how that term is used, where this willingness or unwillingness is motivated by a certain kind of reason. If that is the correct picture of how meanings and referents get attached to words, then conceptual shift happens when our dispositions to negotiate change. This approach to conceptual replacement is familiar but sometimes contested. In what follows I aim to help show how it can be both plausible and useful.

Joshua Glasgow, Conceptual Revolutions In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Joshua Glasgow. DOI: 10.1093/oso/9780198803331.003.0009

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

150

 

Some of us have made a replacement recommendation for the term ‘race,’ and of course replacement has been proposed for a host of other terms as well. The view of conceptual shift fleshed out here is meant to paint a picture of what this sort of linguistic action looks like. It also serves other purposes. I have briefly argued elsewhere that the dispositional account can save descriptivism from an alleged problem that has been raised by Sally Haslanger and Scott Soames. What follows more fully defends the key premise of that response, and I also explain how that premise insulates descriptivism from another objection that Soames lodges. Finally, I show how this theory also can accommodate the Quinean convictions that we have an open-ended set of options about how to use our terms and that our choice of which option to pursue is largely dictated by extra-semantic factors such as utility and psychological proclivity. In what follows, Section 8.2 articulates the outlines of the dispositional approach, highlighting how it alone can accommodate certain facts and how it can answer an important challenge. Next the chapter demonstrates the role of this approach in solving some of the putative problems for descriptivism that Haslanger and Soames raise. It then examines whether, on this picture or any other, terms change meanings or merely have their meanings evolve; I argue that we should accept full-on change, what I call conceptual revolution. Finally, the chapter details how the dispositional approach to conceptual revolution can accommodate those Quinean convictions.

8.2 What Would You Say? If anything has widespread acceptance in philosophy of language, it is that use dictates meaning. But like most slogans, this is a caricature. Language use doesn’t dictate meaning. Instead, dispositions to use language dictate meaning. Actual use is beside the point, except insofar as actual use is the authentic expression of disposition. When dispositions outstrip or contradict actual use, dispositions have semantic authority. This view is not new, but it will pay to unpack it a bit. Meaning and reference are fixed by certain commitments. We are semantically committed to ‘bachelor’ referring to men. ‘Man’ is part of the meaning of the term and therefore partially fixes the reference of ‘bachelor.’ Such commitments, in turn, are nothing more than certain behavioural dispositions. In particular, they are the dispositions to deploy semantic conversation stoppers. Semantic conversation stoppers are acts— speech acts and non-speech acts—that end a dialogue for semantic reasons. Of course, conversations are killed by all sorts of things. Sometimes the bell rings and class is over. Sometimes we fall asleep. Sometimes we are interrupted by a stranger. Sometimes it is last call. But these are all non-semantic conversation stoppers. A semantic conversation stopper is any behaviour that aims to end a

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

151

conversation, motivated by the thought, be it explicit or unconscious, that we are talking past one another, or having a merely verbal dispute where ‘merely’ signifies that nothing hangs on which side is using the verbiage correctly, that is, where the person doing the stopping believes that each usage is equally legitimate. When we are disposed to deploy a semantic conversation stopper, we have an antinegotiation disposition: we are unwilling to negotiate on how a key term can be used, relative to the context and conversation—I’m using the term in my way, you are using it in yours, and the two usages are not to be reconciled.¹ Or at least we were unwilling to negotiate. We might at that point change our dispositions. We might change the conversation and use the term in a way that deviates from our current or past commitments, but up to any such point in this conversation, that meaning was fixed for that term for that speaker. (Henceforth I will mostly ignore the relativity parameters; they do not impact the arguments that follow). We are making camp together, and I tell you, ‘The tent is too light.’ Being conscientious of my sleep troubles, you kindly say, ‘I’ll take care of it.’ So while I go off to catch a fish for dinner, you put the fly tarp over the tent to darken it, at which point you join me at the water’s edge. When we return, our tent is missing; the howling winds have sent it tumbling down the canyon, out of reach. ‘What happened?’ I ask, ‘I thought you were taking care of it.’ ‘What do you mean?’ you protest, ‘I said I would darken the tent! Which I did—I put the fly over it.’ ‘Darken it?’ I reply in confusion, ‘You were going to weigh it down. It was too . . . light.’ And then, realizing that we talked past each other, using the ambiguous word, ‘light,’ in two different senses, we have finished. There is no arguing about what happened, no judging right or wrong, no fault assigned, and most importantly for our purposes, no correcting one another. We did not disagree; we simply miscommunicated. As we sit there pondering what to do next, our silence is the semantic conversation stopper, ending the dialogue for semantic reasons—we are motivated to end the conversation because, for this conversation, ‘light’ for you is opposed to dark, and for me, it is opposed to heavy. Other times we have to be more explicit when stopping a conversation on semantic grounds. Sometimes people don’t recognize ambiguity, and sometimes people try to capitalize on ambiguity by deploying words in a weaselly manner. As your lover is ending your relationship, you say, ‘I can’t believe you are breaking up with me.’ Your now-ex-partner replies, ‘I’m not really breaking up with you. Technically I’m just restructuring our relationship.’ You’d be within your linguistic rights to say, ‘I guess we just mean different things by “breaking up” ’, as a way of ending that not-so-pleasant segment of the conversation.

¹ Reliance on non-negotiable commitments can be found in a variety of other places (e.g. Joyce, 2001).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

152

 

We have recently seen a lively discussion about verbal disputes.² I want to sidestep most of the issues in that conversation and focus on one thing: when interlocutors are disposed to say that a dispute is merely verbal or that they are just talking past one another, and that they are therefore not really disagreeing, it is that disposition that sets the meaning or reference of a term, at least partially.³ To put it Euthyphronically, whether a disagreement is merely verbal is not determined by the terms’ meanings; rather, terms’ meanings get determined by what we (are disposed to) maintain is a merely verbal disagreement. So on this view, when we are disposed to say that we have landed in a situation where we are talking past one another, our disposition establishes the meaning or reference of the relevant verbiage, at least in our own idiolects (and again, relative to conversation and context). The ‘we’ here is, of course, all of us. If you determine that you are unwilling to use a word in the way that the rest of us do, then you have started your own conversation with distinct meanings or references attached to our common terms. There is nothing necessarily bad about that—as we will see below, taking this step of siloing a distinct usage of a term might even be used for good. In fact, it reveals the wonderfully democratic nature of conceptual boundaries: we can, each of us, draw them wherever we like. The only potential price to pay in doing so is sequestering ourselves in our own little conversations, if others don’t adopt the new boundaries with us. In some cases we might be tempted to stop a conversation on semantic grounds, but things turn out to be murky, and we surprisingly find ourselves willing to talk a bit more, to negotiate over language use. Once upon a time, our predecessors said, ‘atoms are indivisible.’ Competent language users widely appeared unwilling to negotiate on this commitment. Then, in 1917, Rutherford split the things that his linguistic community had been calling ‘atoms’ (this is a dramatization, but bear with me). The community was at a semantic crossroads, forced to compromise on at least one of two commitments: should they negotiate on the commitment that atoms are indivisible; or should they negotiate on the commitment that those things Rutherford had split really were atoms? History, of course, loudly announced our verdict. We compromised on the claim that atoms were indivisible and preserved both the observation that those things were divisible and the referential commitment that it was appropriate to call those things ‘atoms.’ We kept reference and revised our surprisingly flawed theory of the term’s meaning.⁴

² See, e.g. Chalmers, 2011; Hirsch, 2013; Jackson, 2013; Jackson, 2014; Jenkins, 2014; Sidelle, 2007. ³ What are the other parts? They arguably include at least the external world and the social context, insofar as these factors help determine reference for some terms (Putnam, 1975; Burge, 1979; 1993). My views are anti-externalist only to the extent that such externalities do not wholly determine meaning and reference. ⁴ Some think that from Democritus to Rutherford, there was a conceptual shift in ‘atom.’ (That is, the two inhabited different conversations with respect to that term and it etymological ancestor,

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

153

Sometimes we go the other way, giving up reference and keeping our theory of what a term means. When the observation that people don’t have supernatural powers forced us to choose between the commitment that witches have supernatural powers and the commitment that those people were witches, we kept the former (our theory of the meaning of ‘witch’) and dispatched the latter (reference). When empirical observation or even mere thought experiment drives us to these crossroads, there appears to be no principle to tell us which way to go. Do we sacrifice our understanding of what a term means, or do we sacrifice our understanding of what it refers to? Later I show some pragmatic factors that can steer these choices. What matters for the moment is that when we are confronted with incompatible commitments, how we are disposed to resolve the incompatibility decides where our conversation-stoppers lie, which themselves shape reference and meaning. When we would not be willing to negotiate, then that in that unwillingness we lock some content into the concept. ‘Witch’ has to mean someone with supernatural powers in that conversation, but not in the conversation where ‘witch’ just means practitioner of Wicca. Our unwillingness to negotiate about bachelors being men means that being a man is part of the concept . When you insist that your partner is breaking up with you, you in that act assign meaning to ‘break up.’ The flip side of this phenomenon is that when we would be willing to negotiate, that willingness excludes some content from the concept. Our willingness to allow that atoms might be divisible means that indivisibility is not part of the concept of atom—no matter how surprising that might have been. Again, these negotiation dispositions are not mere concept indicators; they actually do the work of determining which concepts, and constraining which referents, are assigned to which terms. To return to the sometimes neglected key point, it is worth emphasizing that it is dispositions to refuse to negotiate, rather than actual refusals to negotiate, that ultimately set meaning. If actual use dictated meaning, then it would be impossible for meaning to stay constant when we accept surprises and use old terms in new ways without changing the subject. Not only is this infelicitous, but it also unacceptably breaks off what are clearly dialogues. Imagine that prior to Rutherford, everyone in his linguistic community actually insisted that atoms are indivisible (again, please excuse the oversimplification in this dramatization). In that case, if actual use dictated meaning, then only two possibilities present themselves. First, ‘atom’ might have been defined as indivisible, and those things they were calling ‘atoms,’ those things that Rutherford showed to be divisible, would have been shmatoms, not atoms. The theory of what the word meant would have trumped the theory of what the word referred atomos.) That may be. It is difficult to know what Democritus would have been willing to negotiate on when confronted with the history of ‘atom’ talk up through Rutherford’s advances.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

154

 

to. But that is not what happened. Those language users were referring to atoms with their use of the word ‘atom.’ The second option, if actual use dictated meaning, would be to say that the reference of ‘atom’ changed. On this option, in 1916 ‘atom’ refers to something indivisible and therefore was widely misapplied to things that were in actual fact divisible, while by 1918, it does not refer to something indivisible and is widely and correctly applied to divisible objects. This diagnosis of the linguistic landscape makes ‘atom’—like ‘Madagascar’ or ‘awful’— a case of semantic replacement. But in this case it is a false diagnosis. The change in ‘Madagascar’ means that we are talking past medieval users of that word, on no more shared linguistic ground than you and I are when we use the word ‘light’ in two different senses during our camping trip. This is not the case with ‘atom.’ People before Rutherford are not talking past people after his experiment. Besides the bald counter-intuitiveness of that idea (those two groups of people include many of the same people who take themselves to be in the same conversation), also 1918ers can disagree with 1916ers about whether atoms are divisible. This kind of successful communication requires shared semantic commitments. In this case, at least, the conversation is not stopped. The natural way to capture the fact that 1916ers and 1918ers subscribe to a common meaning for ‘atom’ is to move from actual uses to dispositions as the fixers of meaning and reference. We simply say: ‘Had the 1916ers known about Rutherford’s experiments, they would have agreed that atoms are divisible.’ Competent users of ‘atom’ in 1916 had a (projected) disposition to uses the term ‘atom’ in certain ways under certain conditions that were at the time counterfactual, namely, having evidence that the kind of thing they’d been calling ‘atoms’ could be split. Even for them being indivisible is not part of the concept , their explicit and firmly held beliefs to the contrary notwithstanding.⁵ So, on this account, dispositions to use words in certain ways are what dictate meaning. Specifically, dispositions to not negotiate, to deploy semantic conversation stoppers, supply a term with its meaning and constraints on its reference; dispositions to negotiate exclude candidate meanings and references. We wouldn’t negotiate or linguistically collaborate with medieval time travellers about where Madagascar is or whether Big Sur is awful. Instead, we would say, ‘You and we mean different things by those terms.’ In being so disposed to not negotiate, we set the actual meaning. Then we’d either stop the conversation or introduce new terms that explicitly relativize the meanings: ‘Big Sur is awful-in-your-sense-ofthe-word, but it is definitely not awful-in-our-sense. Madagascarus is right here; Madagascaryou is over there.’

⁵ Some differing understandings of a concept arguably are disagreements (sometimes concealed disagreements) about how a term should be used (Plunkett, 2015). But clearly many cases where terms are being used in different ways are not disputes of this sort; rather they are different attempts to zero in on what a concept is.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

155

From the dispositional theory of how concepts (and referents) get assigned to words, a natural picture of conceptual shift takes shape. If anti-negotiation dispositions set meaning, then words take on new concepts when our negotiation dispositions change. Speakers used to be (dispositionally and actually) committed to the content of the concept  being awe-inspiring—the relevant linguistic community would not have negotiated on that. Now we are committed to it being really bad—we would rather say that the meaning and conversation have changed than negotiate on this. These dispositions give meaning to ‘awful,’ and when the dispositions change, we now have a conceptual shift. In other words, this view neatly answers the Quinean challenge to provide a principled distinction between cases where revising a statement does and does not involve conceptual change: conceptual change happens when our semantic conversation-stoppers change (cf. Chalmers, 2012: 203–5).⁶

8.3 Dispositions and Descriptions In addition to providing a tidy picture of conceptual shift, the dispositional theory of meaning and reference-fixing disposes with an apparent problem for descriptivists. For our purposes let ‘descriptivism’ be the view that what a term refers to is fixed by a description, a set of properties that something has to have for the term to truly apply to it (Jackson, 1998). If the descriptivist is going to accommodate the fact that we truly communicate with one another, that we can agree and disagree with one another rather than simply talk past each other (as in the tent-is-too-light case), then descriptivism must allow that every language user competently participating in such conversations possesses the reference-fixing description for that term. But, one worry goes, competent speakers in the real world possess descriptions that do not overlap enough to secure the same, conversation-enabling reference (Haslanger, 2010; Soames, 2005, p. 418). You might think that whales are fish, while I think they are non-fish. If so, it seems doubtful that we share the same description for ‘whale.’ But, surely, we are both successfully co-referring to whales when we apply the word ‘whale’ to that humpback breaching fifty feet off the bow of our vessel. And surely we are talking about the same thing when we start debating whether whales are fish. So, descriptivism appears to be in trouble: if language users can possess different and conflicting descriptions while converging on identical references for a term, then any description attached to the term by language users is not what fixes reference. A dispositional approach to descriptions provides the materials to reply to this objection (Glasgow, forthcoming): sharing the same set of dispositions to deploy

⁶ Conceptual shift can partial or whole (Quine, 1951: 25–7).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

156

 

semantic conversation-stoppers, as opposed to sharing actual usages or explicit descriptive beliefs, is what overlaps enough to guarantee shared reference. To spell out this approach in more detail, the description that counts as the operative one in a successful conversation is the one that we would all converge on when presented with the same (actual and merely possible) conflict-of-commitment scenarios. As we saw with ‘atom,’ what competent language users would be disposed to negotiate on when confronted with their own incompatible beliefs can diverge from what they explicitly believe or say about the term in question— even when those beliefs are cherished or deeply held. So if you say that whales are fish, I might ask you, ‘What if that thing over there, that thing you’re calling a “whale,” turns out not to be a fish, on your operative definition of “fish”?’ When you negotiate, and say, ‘Then whales would be nonfish,’ we will have uncovered your disposition to say that being a fish is not part of your reference-fixing description for ‘whale,’ even if you were previously certain that being a fish actually was part of the description. This is how, for descriptivism, two people can talk about the same thing while having radically different and firmly held beliefs about whales. Your latent willingness to negotiate shows that being fishy is not part of the reference-fixing description of ‘whale’ for you, just as 1916ers would have been willing to negotiate on the divisibility of atoms without realizing it. Dispositions to negotiate exist even when the speakers are unaware of them. Of course, this does not vindicate a descriptive analysis of ‘whale’ or any other term. To vindicate that, we would need to show that there are some properties that all competent users of ‘whale’ are dispositionally committed to non-negotiably, to fix the reference of that term. But we can already see that the Haslanger/Soames challenge only poses a problem for certain kinds of descriptivism, and not the kind of descriptivism characterized here. Their objection supposes that for descriptivism, the speaker must explicitly believe the description that is true of the term, or at least have some sort of intellectualized, conscious access to that description: what you explicitly believe about whales departs from what I explicitly believe about whales. But dispositional descriptivism de-emphasizes explicit belief in favour of dispositions, when it comes to possessing a description. We need not adopt the standard that we are aware of these reference-fixing descriptions. Instead, dispositional descriptivism can just say that competent speakers must be disposed to have a non-negotiable commitment to the description. If speakers are not aware of relevant information (such as the circulatory facts about whales and the smashability of atoms), they might not be aware of the descriptions they themselves are committed to, because they haven’t tested those commitments to uncover their dispositions. We do not always know what we are disposed to say, so our explicit beliefs might contradict our tacit dispositions, and on this account the latter trumps the former (Horwich, 1998; Jackson, 1998).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

157

This is why Galileo can say that the earth is flat and the patient can say that she has arthritis in her thigh (Burge, 1979), while still being implicitly committed to conceptual contents for ‘the earth’ and ‘arthritis’ that, jointly with observed fact, are inconsistent with those claims. In such cases, we think we mean one thing by a term, or refer to one object with that term, when really and unknowingly our dispositions commit us to the term meaning or referring to something else. Such commitments may not be revealed for centuries, and may be revealed only under extraordinary discoveries or thought experiments about very remote possibilities. Consequently, we can quite easily share overlapping, conversation-enabling descriptive reference-fixers. We just have to share the same dispositions about when a use of a term is negotiable, and when it is not. It is also old news that an important disposition can be the disposition a language user has to defer to the dispositions held by experts in her linguistic community, even when those linguistic authorities would reject the way that that user uses the relevant terms (Horwich, 1998, p. 86; Searle, 1983). These deferential dispositions explain how we can acquire a word and use it competently without fully understanding its meaning, which arguably is the case for just about every word we use: we are disposed to defer to some (perhaps ideal) speaker and borrow whatever they mean by the word. Deferential dispositions also explain how we can misuse words while still possessing their meaning, as in Burge’s ‘arthritis’ case: we are (dispositionally) more committed to their proper meaning, which again may be possessed only by other (ideal or actual) agents, than we are to our own misuses. It is here that Soames (2005, pp. 419–20) levels another objection to descriptivism: these so-called parasitic intentions are not part of the semantic content of terms. Soames is correct that the parasitic semantic intention, I defer to whatever they mean by ‘t’, is not itself part of the semantic content of ‘t.’ Instead, those intentions direct us to semantic content, which waits elsewhere in the linguistic field. Assume for simplicity that the relevant description of ‘arthritis’ is something like, chronic joint pain. Soames is correct that chronic joint pain is not semantically identical to the causal source of the term ‘arthritis,’ which is something like the relevant medical authorities (or perhaps, Archibald Garrod’s first usage of ‘arthritis’). So we have to further distinguish between two kinds of dispositional descriptivism. One kind says that the speaker-source of the term is what the rest of us dispositionally intend to refer to when we use the term parasitically. This is exposed to Soames’ objection: the source is not itself the meaning or referenceconstraint. The other kind says that whatever the speaker-source is talking about is what we dispositionally intend to refer to when we parasitically use the term. This is not exposed to Soames’ objection. Dispositional descriptivism can in these ways survive any putative counterexamples where competent speakers seem to reject the description for a term (Jackson, 1998). Whenever we think someone has found a counter-example to

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

158

 

descriptivism, really we have just landed on a flawed theory of what our dispositions are and therefore what the description is. The problem is not with descriptivism; rather, the problem is that we have misidentified the description, because we have misidentified our dispositions. With that limited defence of descriptivism in place, we can now return to the dispositional approach to conceptual change in general, leaving descriptivism in particular to the side. We need to more closely examine the possible forms that conceptual change might take, for what I have said so far is consistent with at least two possibilities.

8.4 Revolution, Evolution, Revelation The referential change for ‘Madagascar’ was inadvertent. Other times we more deliberately change the referent or meaning of a term. Certain language users chose for ‘sick’ to mean good, and their usage caught on. And conceptual change has frequently been recommended to advance social justice—‘queer’ is sometimes cited as a reclamation case where the term was changed, with some degree of intention, from a pejorative to a term of pride. Sometimes meaning drifts, and sometimes we give it a shove. Haslanger (2000) and I (2009, Ch. 7) have recommended—in different ways— such a change with ‘race,’ redefining it for certain purposes.⁷ My own version of this prescription is to redefine ‘race’ to refer to social kinds, in order to advance ethics, justice, and epistemic respectability, on the premise that ‘race’ currently purports to refer to biological groups that turn out to be non-existent.⁸ If you wanted to be an activist about these things, trying to make Haslanger’s or my proposal a reality, then on the view articulated above, you would have to change people’s willingness to negotiate on how terms can be used. If we want to redefine ‘race’ as a social category, instead of a biological category, then we have to accept certain possibilities that we currently refuse to accept. (I underscore that this is just my view, not Haslanger’s.) For example, imagine that people forgot to classify one another racially for one minute due to a drug in the global water supply that caused temporary racial amnesia. Would that mean that Hillary Clinton stops being white for that minute? Clearly not, on the current, ordinary concept of race. We refuse to accept this possibility, and I believe that the same holds for other

⁷ For a different redefinition agenda, see Millstein (2015). Some other commentators (e.g. Blum, 2002; McPherson, 2015) recommend adopting different terms to replace the term ‘race’—shifting terms (and consequently concepts, too) rather than shifting the concepts that attach to unchanging terms. ⁸ I am now less enthusiastic about this proposal because there is arguably a more plausible way to construe current, non-revolutionary race-talk in non-biological terms (Glasgow & Woodward, 2015). See Glasgow, Haslanger, Jeffers, & Spencer (forthcoming: Ch. 4) for an examination of the relative merits of these two approaches.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

159

changes in social practices regarding race (Glasgow, 2009, Ch. 6). A disposition to not negotiate on the idea that race persists across changes to social practices means that ‘race’ is presently defined in a non-social way. Nevertheless, we could start being open to these possibilities, and if we did change our dispositions in that way, then Hillary Clinton could stop being white for a minute while we forgot about race. We would have a conceptual shift. There is something liberating about this possibility, and also something troubling. The liberating thing is that we can do whatever we want with our dispositions! The troubling thing is that we can do whatever we want with our dispositions. This tool to reshape language, to replace which concepts attach to which words, might not only be used to advance justice, but also to violate rights and do grievous harm. For example, it can manifest in what Teresa Marques calls ‘meaning perversions’ (Chapter 13, this volume): trying to (intentionally or unintentionally) hijack liberatory and just language, like the language of ‘freedom’ or ‘justice’ or ‘truth’ or ‘news,’ in a way that makes us worse off. Language is a weapon that can be used for good or evil. That, anyway, is the view on the table: we can change the meanings and references of our words, which creates an opportunity to make things better . . . or worse. Call this phenomenon conceptual revolution: the concept is replaced, such that one concept used to attach to a term, but now a new one does, due to a shift in our dispositions. This phenomenon is, I believe, common, and pursuing it is arguably recommended for certain terms, like ‘race.’ That is, the shift in how we use racial terms like ‘white’ that I endorsed above should be framed as a realignment of our dispositions to negotiate over usage of ‘white.’ ‘White’ undergoes an outright conceptual revolution in the relevant linguistic community, where the concept that attaches to the word has been, or is being, replaced. But a second, different possibility for our shifting practices is that ‘white’ has always failed to refer to people like Hilary Clinton in (very rare, possibly counterfactual) social contexts, but many people are and have been unable or unwilling to recognize that usage. Once that sort of application is more widely revealed as compatible with our dispositions, it will be explicitly used more widely. Call this sort of phenomenon conceptual revelation. When revelation happens, there is no change in the language; instead the change is to our awareness of what our nonchanging language can do. Clearly revelation is common with both scientific and lay terms. It had to be revealed to us, for example, that whales are not fish—instead of changing the concept of  or  in some linguistic revolution, we simply changed our theory of what the unchanging contents of the concept of  were all along. Haslanger and other social constructionists often say this is exactly the case with racial language—its social contingency is concealed to many language users, and needs to be revealed to us. Then there is a third possibility, where changes are being made to the concept (unlike revelation’s unchanging concepts) but the concept is not changing into a

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

160

 

new one (unlike revolution’s changing concepts). Instead, there is one concept there throughout that simply evolves. Call this conceptual evolution: the concept is dynamic, and it remains but changes shape.⁹ When a concept evolves, the identity of the thing whose shape is in question remains static. The term’s concept is not replaced by a new concept. Nor is a new possibility for an old concept with fixed content freshly revealed to us. Rather, on this third possibility, the concept’s content changes, without changing its identity. Like revolution, ‘race’ will be able to do new things—things that it has historically been incapable of doing— but unlike revolution (and like revelation), we’re still talking about one thing— race—all the way through. If we go with the first, revolutionary, analysis, then like ‘Madagascar,’ the ‘race’ or ‘white’ of tomorrow will not be continuous with or translatable into the ‘race’ or ‘white’ of today. This non-translatability might seem strange—it means that we will have to relativize the meanings of terms, and that cross-era dialogue becomes impossible. If we think that is an unhappy diagnosis, then in a scenario where racial terms seem to be socially contingent, either the concept that attaches to ‘race’ has a changing shape, admitting new possibilities; or pre-existing possibilities for the concept are being newly revealed to language users. Let us put aside revelation for the moment—it is common enough, and it can coexist with both revolution and evolution. Of the remaining two possibilities, why might we analyse any given word in the revolutionary or the evolutionary way? After all, each kind of analysis seems equally capable of explaining the same phenomena. Where revolutionary theory says that the static concept has been replaced, evolutionary theory will just say that the concept’s dynamic content has changed shape, and vice versa. The two theories can be made extensionally equivalent. It seems obvious that at least some words, like ‘Madagascar’ and ‘awful,’ have undergone revolutions, not evolutions. (Equally obviously, some other terms may evolve—we don’t need to say there is only One Possible Universal Form of Conceptual Change.) This is obvious, I believe, because of the ways those words’ references have transformed. They used to refer to one set of things; now those words refer to completely different things, which amounts to a revolution rather than evolution. What is interesting and tricky are cases that manifest only a partial reference shift. If we start accepting that people could lose their races if the social community around them forgets to categorize people racially, then those people in those contexts will get different racial categorizations. But everyone in social contexts where people still classify themselves and each other racially (including much of the world today) will still have the same racial identities we’ve always thought they had. Only some of the reference of ‘race’ will change. So the more

⁹ Something like this view of conceptual change can be found in Richard (ms).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

161

specific and more difficult version of our question is: are cases with only partial reference shift evolutions, or revolutions? Quineans like to remind us that we should be open to using our terms in surprising ways without changing meaning. Haslanger (2010) argues that this is a problem for descriptivism. But it is hopefully clear now how revolutionary theory—in both descriptivist and non-descriptivist guises—can allow for discovering surprising things about our terms without meaning changing. It is simply revelation, on the account I am defending: we uncover dispositions we did not know that we had. Such surprises do not require the concepts themselves to change; they just require change to our knowledge of the concepts’ contents. Our theories of what terms mean can change with new discoveries, without meaning itself changing: the static concept is revealed to us by the surprising experiment, and so our theory of what content that concept contains will change, even as the concept itself remains unaltered. Thus, a pro-revolutionary approach to concepts will require us to say that when there is no revolution but there is linguistic surprise, the explanation for the surprise is revelation: those working under a mistaken theory of a term’s definition did not know what concept they were working with, as when 1916ers used ‘atom’ without knowing what concept attached to it. On the dispositional view, longstanding conceptual content that has been hidden can be exposed by new evidence (actual or counterfactual cases). ‘Atom’ underwent neither revolution nor evolution on the story told here. Rather, there was just a new conceptual awareness: prior to Rutherford the concept included divisibility, but it took his and his colleagues’ work to make folks realize it. The idea that we have systematic, profound, and widespread ignorance about our concepts might seem unlikely or unfriendly. But it is neither. People get confused. It happens. If we can be confused about what time we were supposed to meet at the restaurant or whether the pencils we use for work are tax deductible, then surely we can be confused about the contents of our concepts, too. In fact, because language is often socially powerful—as with ‘race’ and a host of other socially charged terms—and because we can get easily confused when the stakes are high, language users may be more likely to end up confused about combustible concepts.¹⁰ So, I have a difficult time seeing why we should oppose a revolutionary diagnosis, when it allows both for changes to meaning (revolution) and for surprises in how our pre-existing concepts work (revelation). And the revolutionary view seems to have the clear advantage of enabling us to cleanly map ¹⁰ It might be thought that it is uncharitable to attribute contradictions to people in the way of a 1916er who said that atoms must be indivisible, since that amounts to accusing them of bald irrationality. But there is no such accusation. The speakers who have not yet had the full range of conceptual possibility revealed to them are not aware of the inconsistency (Horwich, 1998, p. 101). They are not irrational.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

162

 

conceptual boundaries. We talk past one another, exclude one another from a conversation on linguistic grounds, and change the subject. A natural way to make sense of these phenomena is to have the same word refer to or mean different things.¹¹ So the pressing question for the evolutionary approach is: if concepts merely persevered while changing shape, then how can we say that with ‘awful’ or ‘Madagascar,’ or in the ‘light’ tent case, the interlocutors are talking past one another, rather than sharing a conversation with a rapidly shifting shape? What is changing the subject, on the evolutionary approach? Conceptual revolution seems essential to making sense of the phenomenon of changing the subject. To have a change in subject, or for two interlocutors to talk past one another, there must be some sort of conceptual boundary—the speakers must attach different concepts to the same word. And where there is a boundary, there can be revolutions: a revolution will take place when the boundary moves. And the boundary will move when dispositions shift. When a sufficient amount of divergence takes hold, meaning and reference can get replaced, partially or wholly. Again, this allows for concepts to evolve, too—some words might evolve while others revolve. But if what I am saying here is on track, we don’t need evolving concepts to explain apparent changes in reference; nor to explain how different speakers have radically different beliefs while sharing a language; nor to explain how scientific discovery can make us start using old words in new and surprising ways. Allowing for revolution in which concepts attach to which words, and allowing for language users to undergo conceptual revelation, seems to be all we need.

8.5 Our Fickle Dispositions Lastly, a final Quinean concern about this sort of view comes from Mark Richard (ms). Richard argues that our dispositions about how to respond to surprising scenarios, such as atoms being split or cats having robot innards, are open to many possibilities, and the answers we give are unpredictable and shaped by nonsemantic factors. If so, then these dispositions cannot be meaning-makers, according to his argument. Here is Richard’s illustration. If the first time you hear the word ‘along’ is when you hear Dylan’s lyric, ‘All along the watchtower, princes kept the view,’ you will be disposed to accept the possibility that one can be along something that is vertical. If instead you first read Dylan’s critic saying that this is ludicrous, that someone can only go along horizontal objects, then you will not be disposed to

¹¹ At least, this is a natural way to make sense of these phenomena when we are assuming that our statements are true and false. There are other ways to make sense of them when we merely presuppose (Stalnaker, 2002).

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

163

accept that claim. Whatever we’re exposed to first impacts our negotiability dispositions! And this is just the tip of the cognitive iceberg: primacy is only one of many non-semantic factors that drag around our negotiation dispositions. So do our biases towards conservatism, simplicity, convenience, efficiency, and so on. Following Richard, call these pragmatic factors (cf. Appiah, 1991). The claim, then, is that pragmatic facts determine our dispositions. Richard thinks that this phenomenon is especially vivid when we think about far-flung scenarios like Putnam’s (1962) case of cats turning out to be robots: whether we would decide in that situation to keep calling them ‘cats’ and negotiate, surprisingly, on the proposition that cats have to be animals, is not reflective of any semantic commitment the ordinary person might have now, prior to entertaining the thought experiment. The idea is supposed to be that there are no semantic rules about ‘cats’ that determine whether saying in the robot case that cats are not animals constitutes a change of meaning or a surprising revelation (or an evolution). Meaning in these cases is underdetermined by semantic dispositions. And because we can imagine radical scenarios about any term, we might change how we use any term without changing its meaning. Given the many factors that impact our reactions to surprising cases, we can react in many ways, and all of these can potentially preserve meaning. Thus, Richard (ms, Sec. 11) thinks that meaning facts and non-linguistic facts dictate our dispositions, rather than our dispositions dictating meaning. Richard has other axes to grind here, but I think that this particular argument mischaracterizes the issues. On the Carnapian picture, anything can impact our dispositions to use, or refuse to use, some given name for an object. What matters are what the dispositions are, not where they come from. So what is important for this view is not whether the disposition is guided by pragmatic or non-pragmatic factors. What is important is that if we are not willing to negotiate in refusing to call a couch ‘a horse,’ then that refusal partially sets the meaning of ‘horse.’ This is true even if our disposition is shaped by primacy. Or inertia. Or anything else. Whatever the cause, the resulting dispositions are what dictate meaning. In an earlier exploration of the extra-semantic pressures on meaning, Kwame Anthony Appiah (1991) highlights an odd arbitrariness that arises when we find ourselves at the kinds of semantic crossroads we see with ‘along’ or ‘atom.’ We chose to say that those divisible objects were atoms; but then why did we not also accept that the ions that travel along (!) our nerve pathways were the referent for Descartes’ term, ‘animal spirits’? Why did we not say that mermaids are manatees, that werewolves are people with rabies, and that witches are people who are oppressed for being thought to have the power to cast spells? Appiah also suggests that pragmatic factors determine our choices. And here the dispositional account can agree. All that it has to say is that, in those choices, we fix meaning. Dispositions shape meaning regardless of how the dispositions themselves are shaped. The two questions are orthogonal.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

164

 

With that we can turn to the underdetermination claim that we have no such dispositions prior to them being tested or evoked by a thought experiment or an actual-world surprise. I believe such dispositions simply lie there undetected. The choices we would make—and unless we are sceptics about dispositions, we have to accept that there are choices we would make—are just choices that we are not aware of yet, prior to being made to think about divisible atoms or cats with robot innards.¹² At one point, Richard (ms, Sec. 12) suggests that at t₁, there is no fact of the matter as to what uses of a term a speaker might license or non-negotiably reject at t₂. I think that since there is some reaction the speaker will have at t₂, we can say that, barring evidence to the contrary, she would at t₁ have had that reaction, had she been confronted at t₁ with the evidence that generates her actual reaction at t₂. Of course, these choices may be wholly unpredictable until we get the test— until we get to t₂. If we don’t know whether we would say that the cattish objects with robot innards are cats, then we don’t know whether it is true by definition that cats are animals. And if we would never have suspected that we would allow divisible atoms, then we would not have known that the concept  did not include being indivisible. But equally clearly, to say either that we are ignorant of some fact or that we have misjudged the fact is not to say that there is no such fact. As Carnap (1955, p. 36) emphasized, identifying the content of a natural-language concept is an empirical science. We can err in that endeavour as much as in any science. And it is a science that is far from complete.

References Appiah, K. A. (1991). Social forces, ‘natural kinds.’ In A. Zegeye, L. Harris, & J. Maxted (Eds.), Exploitation and exclusion: Race and class in contemporary US society (pp. 1–13). London: Hanz Zell. Blum, L. (2002). ‘I’m Not a Racist, but . . . ’. Ithaca, NY: Cornell University Press. Burge, T. (1979). Individualism and the mental. Midwest Studies in Philosophy, 4, 73–121. Burge, T. (1993). Concepts, definitions, and meaning. Metaphilosophy, 24, 309–25. Burgess, J. P. (2014). Madagascar revisited. Analysis, 74(2), 195–201. Carnap, R. (1955). Meaning and synonymy in natural languages. Philosophical Studies, 6, 33–47. Chalmers, D. J. (2011). Verbal disputes. Philosophical Review, 120, 515–66. Chalmers, D. J. (2012). Constructing the world. Oxford: Oxford University Press.

¹² Putnam himself certainly seemed to think there were intuitive answers to these questions, though he recognized that his intuitions might not be shared by all.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

 

165

Fricker, M. (2007). Epistemic injustice: Power & the ethics of knowing. Oxford: Oxford University Press. Glasgow, J. (2009). A theory of race. New York: Routledge. Glasgow, J. (Forthcoming). ‘Race’ and description. In Q. Spencer (Ed.), The race debates. Oxford: Oxford University Press. Glasgow, J., Haslanger, S., Jeffers, C., & Spencer, Q. (Forthcoming). Four views on race. Oxford: Oxford University Press. Glasgow, J., & Woodward, J. (2015). Basic racial realism. Journal of the American Philosophical Association, 1(3), 449–66. Haslanger, S. (2000). Gender and race: (What) are they? (What) do we want them to be? Noûs, 34, 31–55. Haslanger, S. (2010). Language, politics, and ‘the folk’: Looking for ‘the meaning’ of ‘race.’ The Monist, 93, 169–87. Hirsch, E. (2013). Charity to charity. Philosophy and Phenomenological Research, 86, 435–42. Horwich, P. (1998). Meaning. Oxford: Clarendon Press. Jackson, B. B. (2013). Metaphysics, verbal disputes, and the limits of charity. Philosophy and Phenomenological Research, 86, 412–34. Jackson, B. B. (2014). Verbal disputes and substantiveness. Erkenntnis, 79, 31–54. Jackson, F. (1998). Reference and description revisited. Philosophical Perspectives, 12: Language, Mind, and Ontology, 201–18. Jenkins, C. S. I. (2014). Merely verbal disputes. Erkenntnis, 79, 11–30. Joyce, R. (2001). The myth of morality. Cambridge: Cambridge University Press. McPherson, L. K. (2015). Deflating ‘race.’ Journal of the American Philosophical Association, 1, 674–93. Millstein, R. L. (2015). Thinking about populations and races in time. Studies in History and Philosophy of Biological and Biomedical Sciences, 52, 5–11. Plunkett, D. (2015). Which concepts should we use? Metalinguistic negotiations and the methodology of philosophy. Inquiry, 58, 828–74. Putnam, H. (1962). It ain’t necessarily so. The Journal of Philosophy, 59, 658–71. Putnam, H. (1975). The meaning of ‘meaning.’ Minnesota Studies in the Philosophy of Science, 7, 131–93 Quine, W. V. (1951). Two dogmas of empiricism. The Philosophical Review, 60, 20–43. Richard, M. (Ms). Quine and the species problem. In Meaning as species, Ch 1. Draft of June 2016. https://markrichardphilosophy.files.wordpress.com/2011/07/meaningsas-species-chap-1-3-21-june-16.pdf Accessed 8 September 2016. Cited with the permission of the author. Searle, J. R. (1983). Intentionality. Cambridge: Cambridge University Press. Sidelle, A. (2007). The method of verbal dispute. Philosophical Topics, 35, 83–113.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

166

 

Soames, S. (2005). Reference and description. In F. Jackson and M. Smith (Eds.), The Oxford handbook of contemporary philosophy (pp. 397–426). Oxford: Oxford University Press. Stalnaker, R. (2002). Common ground. Linguistics and Philosophy, 25, 701–21.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

9 The Folk Concept of Race Edouard Machery and Luc Faucher

Concepts vary in myriad ways.¹ Much of the recent psychological literature focuses on intraindividual variation (Bloch-Mullins, 2015). In particular, cognitive psychologists and philosophers of psychology disagree about the contextsensitivity of concepts. Invariantists hold that within a given individual, tokens of a given concept (e.g. ) are constituted by the same context-insensitive body of information retrieved from long-term memory (Machery, 2009, 2015; Mazzone & Lalumera, 2010); by contrast, contextualists argue that the tokens of a given concept are context-sensitive through and through, and that the tokens of a given concept can have little in common across some contexts (e.g. Barsalou, 1999; Faucher, 2014; Casasanto & Lupyan, 2015; Löhr, 2017). Philosophers have also discussed interindividual variation, examining in particular whether interindividual variation is consistent with communication, and if so, how (e.g. Fodor, 1998; Prinz, 2002; Machery, 2010b). This chapter examines the issue of inter- and intraindividual variation from a new angle, particularly among philosophers: What constrains variation of concepts? Instead of addressing this issue abstractly, we focus on a case study and examine the constraints that limit variation of the concept of race. We argue that people’s folk biology constrains the concept of race and thus limits how much concepts of race can vary within and across individuals. Section 9.1 introduces our views about folk theories and concepts in general. Section 9.2 presents two competing hypotheses about our case study: the biological and the social hypotheses about the concept of race. Sections 9.3 to 9.6 review our experimental work on the lay conceptualization of races. Section 9.7 brings these results to bear on how folk theories constrain variation in concepts.

¹ This chapter characterizes concepts as bodies of information used in the processes underlying higher cognitive competencies such as categorization, inductive and deductive reasoning, metaphorical thinking, etc. (for discussion, see Machery, 2009, Ch. 1; 2010a; 2017a). Edouard Machery and Luc Faucher, The Folk Concept of Race In: Shifting Concepts: The Philosophy and Psychology of Conceptual Variability. Edited by: Teresa Marques and Åsa Wikforss, Oxford University Press (2020). © Edouard Machery and Luc Faucher. DOI: 10.1093/oso/9780198803331.003.0010

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

168

    

9.1 Folk Theories and the Stability of Concepts 9.1.1 Folk Theories Folk theories (or ‘intuitive theories’) are sets of generalizations about what is often called a ‘domain,’ that is, a particular type of objects, processes, and events (Murphy & Medin, 1985; Gopnik et al., 2004; Carey, 2009; Machery, 2009). For instance, folk biology is about (i.e. its domain consists of) the objects (e.g. animals), processes (e.g. growth), and events (e.g. death) that are related to the living world. Some of the generalizations that constitute a given folk theory are causal; some relate observables to unobservable entities or processes postulated by the relevant folk theories. Thus, according to proponents of the ‘theory theory’ about mindreading, the theory of mind includes causal generalizations between lines of sight (an observable) and the formation of perceptual beliefs (which are unobservable entities posited by people’s theory of mind). People know these folk theories tacitly: by and large, people cannot verbalize the content of the theories they are supposed to know. Psychologists have studied people’s folk theories in detail for decades. They have identified theories that develop reliably across demographic groups in neurotypical individuals. These theories also often start developing early in neurotypical children and their development seems to follow a more or less strict timeline. Folk biology, folk physics, and the theory of mind are prominent examples of such folk theories (e.g. Carey, 2009). Psychologists often say that these folk theories are, at least partly, innate, but we need not use the outdated notion of innateness to characterize their reliable and early development (Griffiths & Machery, 2008). Furthermore, while psychologists, particularly developmental psychologists, have often paid greater attention to the aspects of folk theories that are cross-cultural and develop early, some folk theories are likely to vary partly across cultures, even when other aspects are universal, while other folk theories are likely to be acquired by cultural learning (Fessler & Machery, 2012). Indeed, there has recently been much attention to the aspects of folk theories that vary across cultures and to the environmental and social circumstances that promote such variation (e.g. Atran & Medin, 2008; Medin & Bang, 2014). Folk biology, for instance, consists of both cross-cultural components and components that vary across cultures.

9.1.2 The Stability of Concepts Several factors contribute to explaining why people tend to share similar concepts, e.g. why people tend to think of, e.g. dogs, justice, and heat similarly. Most evidently, the environment constrains the concepts people acquire. Dogs have

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

169

properties in common, and these constrain the way people represent dogs in thought (see Malt, 1995 for an influential discussion). Communication, too, constrains the way people form concepts about newly encountered categories of objects. Markman and Makin (1998) have examined the role of communication and of the need to establish common reference to a set of objects (‘referential communication’) in the stability of concepts. In Study 1, dyads of participants were asked to name Lego parts before building an object out of them. The concepts they had acquired were then probed by a sorting task. Their performance was compared to a control condition examining the sorting of participants who built the object on their own instead of as a dyad. It was found that on average participants acquired similar concepts in the communication and control conditions, but that there was more consistency (i.e. stability) among the concepts of participants in the communication condition than in the control condition. In addition to the constraints imposed by the external world and the requirements of successful communication, the structure of the mind also constrains the concepts that people form and ensure a form of stability. These constraints are both formal and material. Formal constraints determine the structure of the concepts people form. For instance, according to recent theorists who have emphasized the role of causation in cognition (e.g. Gopnik et al., 2004; Sloman, 2005), the concepts people form encode some causal knowledge about the relevant categories. The constraints can also be material: material constraints determine the content of concepts, what information they encode. For instance, Carey (1985) proposed that children’s folk biology emerged from their theory of mind: that is, on her views, young children’s concepts of biological processes, events, and things are patterned on their concepts of persons. Our guiding hypothesis in this chapter is that folk theories, whether culturally acquired or not, are an important source of material constraints on the concepts people acquire (Murphy & Medin, 1985). This claim may not be very surprising for some concepts. For instance, it is not very surprising that concepts of animals (e.g. ) are influenced by our folk biology. This claim may however be more interesting for other concepts: For instance, Carey’s (1985) claim about young children’s biological concepts is surprising. Similarly, Griffiths, Machery, and Linquist (2009) have claimed that the concept of innateness is the expression of central components of folk biology (see also Linquist, Machery, Griffiths, & Stotz, 2011). To return to the stability of concepts, folk theories are one of the constraints that limit the variability of concepts. To the extent that folk theories are shared and materially constrain concepts, people will share similar concepts, limiting interpersonal variability. They will also constrain intrapersonal variability and thus the context sensitivity of concepts. For instance, if the concept of innateness really is the expression of core components of folk biology, people will tend to think of development in line with folk biology when they conceptualize development through the lens of the concept of innateness (for other examples, see

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

170

    

Shtulman, 2015). The remainder of this chapter illustrates this claim by focusing on the concept of race.

9.2 The Concept of Race 9.2.1 Races: Social or Biological Groups? How do lay people think about races? What does it mean, on their view, to belong to a particular race, to have a racial identity? Philosophers have addressed this question at length (e.g. Hardimon, 2003), but they have often only backed their claims with historical and textual evidence (e.g. Appiah, 1996). While historical and textual evidence is clearly valuable and important, it is also limited. Relying on this type of evidence amounts to focusing on the writings of a few, highly educated scholars (public intellectuals, historians, anthropologists, philosophers, politicians, biologists, etc.), whose views may differ from lay people’s. In addition, it focuses attention on the explicit characterization of the concept of race, which may differ from the way people spontaneously and implicitly think about race. We can distinguish two competing hypotheses about how lay people think about race. They may think about races as social groups (‘the social hypothesis about the concept of race’) or as biological groups (‘the biological hypothesis’). As a first pass, we can spell out this contrast as follows: to think about races as social groups is to think of them the way we think about cops and senators; to think about races as biological groups is to think of them the way we think about dogs and poodles (as species and subspecies). One reason to favour the social hypothesis about the concept of race over the biological hypothesis is that, for at least seventy years, anthropologists, biologists, philosophers, and social theorists have argued repeatedly that races are not biologically real (e.g. Montagu, 1942; Lewontin, 1972). This scientific and philosophical consensus, broadcast in official declarations, books for the lay public, articles in newspapers and journals dedicated to scientific vulgarization, may have influenced the lay concept of race. Furthermore, some influential theories about the psychological origins of racial classification predict that people should think about races as if they were social groups. In particular, Kurzban, Tooby, and Cosmides (2001) have argued that racial classification is a mere by-product of our evolved sensitivity to coalitions in our social environment (for review, see Machery & Faucher, 2017): ‘[t]o the human mind, race is simply one historically contingent subtype of coalition’ (2001, p. 15391). If they are correct, then we would expect people to think about races in a way similar to the way they think about coalitions, e.g. neighbourhoods, the mafia, trade unions, i.e. stereotypical social groups. Finally, some recent evidence suggests that many Americans conceive of race membership as being social. Glasgow, Shulman, and Covarrubias

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

171

(2009) show that racial identity is not uniformly determined by the one-drop of blood rule or, more generally, by ancestry. Rather, social factors (affiliation or ascription of racial identity by others) often matter. Furthermore, Shulman and Glasgow (2010) use a questionnaire to collect evidence about the folk conception of race and found that many people do not think that races are real, and that among race realists, many believe that social factors contribute to racial identity. On the other hand, from a historical point of view, races seem to have been conceived biologically for a very long time. An important tradition (‘polygenism’) even viewed different races as being literally different species (e.g. Gossett, 1963). Furthermore, several influential theories about the psychological origins of racial classification tie the lay conceptualization of race membership to folk biology. GilWhite (2001) has proposed that the concept of race is a misfiring of our evolved sensitivity to identify ethnic groups (roughly, groups that are normatively homogenous and with whom group members identify) in our social environments (see also Machery & Faucher, 2005; Machery, 2017b). Furthermore, he proposes that we have evolved to think about ethnic groups as being biological groups (perhaps as some kind of species). This evolutionary hypothesis predicts that races are conceptualized as biological entities. It is worth mentioning a subsidiary issue. If the concepts of race are biological, what properties happen to be treated as the expression of the biological nature of races? Consider for instance dogs—while there is no doubt that people conceive of dogs biologically, we do not think that all the properties of dogs express their biological nature; we recognize that the shortened tails of many breeds of dogs and their particular haircuts are socially determined. No one mistakes poodles’ sometimes silly haircut for biological traits! Thus, if the concept of race is biological, which properties are viewed as the expression of race members’ racial nature? It could be that only bona fide biological properties (e.g. anatomy, physiology, etc.) are biologicized, or it could be that other properties, in particular psychological and social properties, are biologicized, too. Historians of racism have emphasized that many social and psychological properties (e.g. intelligence) have long been viewed as the expression of the racial nature of race members, a phenomenon called ‘racialism.’ Is that still the case currently? How widespread is the biologicization of social and psychological properties among lay people?

9.2.2 Operationalizing Biologicization So far, we have only treated the contrast between the social and the psychological hypotheses about the concept of race in vague terms, and we now need to operationalize it. To do so, we rely on the research showing that biological categories are viewed as having a ‘rich inductive potential’ (e.g. Gelman & Markman, 1986; Gelman, 1988). In contrast to other categories (e.g. artefacts),

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

172

    

people expect the members of biological kinds to have many properties in common. As a result, they are more likely to induce a property from one member to another for a biological category. In what follows, we examine whether people treat races as having a rich inductive potential, and are thus willing to generalize to a new category member when they find out that a category member possesses a given property. We also examine which properties, if any, are biologicized in this operationalized sense, that is, which properties, if any, people are willing to project. In the remainder of this chapter, we thus provide empirical evidence in support of the following hypotheses: 1. Races are viewed as being biological groups in the sense that races are treated as having a rich inductive potential. 2. Bona fide biological properties, but also psychological properties, are biologized in the sense that they are projected. 3. Gender is not viewed the way race is.

9.3 Study 1 The goal of this study and of those that follow is to examine how people use concepts of races, concepts of social groups, and concepts of gender when they are asked to decide whether to generalize a property from an individual to another individual. Study 1 focuses on the induction of an unknown biological property. We were concerned that our findings would merely reflect how race membership happens to be conceptualized in the US. To control for this, we ran the study in the US and in Québec, a North American cultural context that has many similarities with the US, but has a different racial history (Mackey, 2013).

9.3.1 Study 1: Participants and Materials The first group of participants consisted of 267 students taking classes at the University of Pittsburgh. Their mean age was twenty-one (range: eighteen to seventy-eight years of age), and forty-two per cent of them were male. Eightythree per cent of them were Caucasian, five per cent African American. The second group of participants consisted of 138 students taking classes at the Université du Québec à Montréal (UQAM). Their mean age was twenty-five (range: nineteen to seventy years of age) and fifty-nine per cent of them were male. Data about ethnic and racial membership of UQAM participants were unreliable and are not reported here. The individuals involved in data collection report that most of them were Caucasians.

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

    

173

American participants were randomly assigned to one of eight conditions: twenty-five per cent of participants were told about an individual identified by its membership in a social group (a football team), fifty per cent about an individual identified by its race (half of those were told the individual was African-American, half that he was Caucasian), and twenty-five per cent about an individual identified by its gender. Half of the participants had to assess an induction from the individual to another individual belonging to the same group (e.g. same race or same gender), while the other half had to assess an induction across groups (e.g. from one race to another race or from one gender to another gender). Participants from Québec were randomly assigned to one of six conditions: thirty-three per cent of participants were told about an individual identified by its membership in a social group, thirty-three per cent about an individual identified by its race (e.g., Caucasian), and thirty-three per cent about an individual identified by its gender. Half of the participants had to assess an induction from the individual to another individual belonging to the same group while the other half had to assess an induction across groups (e.g. from a Caucasian character to an African-American character). All participants were presented with a scenario describing an individual who has been found to possess a particular gene. The gene was identified by an imaginary, scientific-sounding name. Participants were asked to assess the strength of the induction from this character to another character on a sevenpoint scale anchored at 1 with ‘extremely unlikely,’ 2 with ‘clearly unlikely,’ 3 with ‘unlikely,’ 5 with ‘likely,’ 6 with ‘clearly likely,’ and 7 with ‘extremely likely.’ For instance, some participants were presented with the following vignette: Steve is a young African-American male. Recently, doctors have found that Steve carries a previously unknown gene, XOR, on his chromosome 17. Carl is another young African-American male. On the scale below, indicate the likelihood that Carl also carries the gene XOR on his chromosome 17. 1 means extremely unlikely, 2 clearly unlikely, 3 rather unlikely, 5 rather likely, 6 clearly likely, 7 extremely likely.

Other participants were presented with a nearly identical vignette, except for the race of the second character, described as follows: ‘Carl is a young Caucasian male.’ For the race scenarios, in the US, but not in Québec, we examined the induction from African-American to Caucasian, and from Caucasian to AfricanAmerican. In Québec, we only examined the induction from Caucasian to African-American. The gender of the characters (always male) was identified by their names. For the gender scenarios, we only examined the induction from a male character to another male character (‘Steve is a young American male. ( . . . )

OUP CORRECTED PROOF – FINAL, 3/8/2020, SPi

174

    

Carl is another young American male.’) or to a female character (‘Steve is a young American male. ( . . . ) Jane is a young American female.’). The race of the character was not identified, although names carry racial information. For the social group scenarios, we used a coalition, inspired in part by the views of Kurzban, Tooby, and Cosmides (2001) about the origins of racial classification. Some participants read that ‘Steve is a young football player. ( . . . ) Carl is one of Steve’s teammates’ while other read that ‘Steve is a young football player. ( . . . ) Carl is a member of another football team.’ The gender of the character (always male) was identified by the names of the football players, while their race was not explicitly identified. The scenarios were presented in English in the US, and in French in Québec.

9.3.2 Study 1: Results Among American participants, participants did not assess the strength of the induction equally across the eight conditions (F(7, 259) = 5.52, p