Modeling the Meanings of Pictures: Depiction and the philosophy of language 2020942961, 9780198847472

Citation preview

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

Modeling the Meanings of Pictures

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

Modeling the Meanings of Pictures Depiction and the philosophy of language JOHN KULVICKI

1

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, 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 © John Kulvicki 2020 The moral rights of the author 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: 2020942961 ISBN 978–0–19–884747–2 Printed and bound by CPI Group (UK) Ltd, Croydon, CR0 4YY Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

For Geraldine Caufield, Sooja Lee Park, and Soon Park

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

Table of Contents Preface and Acknowledgments List of Figures List of Tables

ix xi xiii

1. Pictures, Communication, and Meaning 1.1 Pictures and Language 1.2 The Structure of the Book 1.3 The Meaning Thread 1.4 The Parts Thread 1.5 What This Book is Not

1 2 6 7 14 17

2. Character, Content, and Reference 2.1 Kaplan’s Distinctions 2.2 Interpreting Pictures 2.3 Bare Bones Content as Pictorial Character 2.4 Pictorial Content 2.5 (In)definite Description and Reference 2.6 Worries about Indirect Pictorial Reference 2.7 Summary

19 20 22 25 28 32 34 35

3. Parts of Pictures 3.1 Syntax Without (Much) Grammar 3.2 Abstraction and Content 3.3 Two Clarifying Objections 3.4 The Parts Principle 3.5 Syntactic Parts and Semantic Roles 3.6 Revisiting Indirect Pictorial Reference 3.7 Summary

37 38 39 43 45 48 50 51

4. Pictorial Dthat 4.1 Attributive and Referential Use 4.2 Using Pictures Referentially 4.3 Dthat 4.4 Referential Use as Dthat 4.5 Worries about Deferred Ostention 4.6 Postcards and Portraits 4.7 Using Parts of Pictures Referentially 4.8 Individuals and Properties in Other Accounts 4.9 Summary

53 53 55 56 58 61 65 73 74 77

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

viii

  

5. Iconography 5.1 Introducing Iconography 5.2 A Semantic Mechanism 5.3 Labeling 5.4 Stories 5.5 The Practicalities of Iconography 5.6 Uses of Iconographic Interpretation 5.7 Iconographic Interpretation in Language? 5.8 Summary

78 79 83 86 88 93 94 96 98

6. Metaphor 6.1 Illustrated, Suggested, and Supplemental Metaphors 6.2 Some Non-metaphorical, Atypical Uses of Pictures 6.3 Strictly Pictorial Metaphors 6.4 Stern on Mthat 6.5 Mthat and Strictly Pictorial Metaphors 6.6 Stern’s Worries about Metaphor in Pictures 6.7 Summary

99 100 103 106 110 113 115 117

7. Direct Reference in Pictures and Maps 7.1 Presence in Photographs and Maps 7.2 How Objects are Involved 7.3 Two Worries about Locations as Names 7.4 Absence and Map Semantics 7.5 Why Maps Have Constant Characters 7.6 The Path from Pictures to Comics to Maps 7.7 Summary

118 119 120 122 123 126 127 129

8. Distinguishing Kinds by Parts 8.1 Syntax and Compositionality 8.2 Separable Syntactic Parts 8.3 Inseparable Syntactic Parts 8.4 The Main Claim 8.5 Three Objections to the Main Claim 8.6 Why the Objections Fail 8.7 Compositionality and Inseparability 8.8 Why Non-propositional? 8.9 Summary

131 132 133 135 136 137 138 144 146 148

References Index

151 157

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

Preface and Acknowledgments This book was unexpected. In 2016, I was asked to present a paper at a conference in honor of Josef Stern, who was one of my dissertation advisors. My plan was simple. Josef is well-known for his work on metaphor, and I have long liked, but been skeptical of, a paper of his that claims there are no distinctively pictorial metaphors. In the spirit of being difficult, I thought it would be fun to try and convince Josef that, in light of his own account, there are indeed pictorial metaphors. Metaphor, for Stern, is partly a semantic phenomenon, and he suggests that we explain it by appeal to a semantic operation akin to David Kaplan’s dthat, which is defined over the character and content of a definite description. So, all I had to do is explain what the character and content of a picture is. That turned out to be an interesting and largely unasked question. The paper was a sketchy mess, but the timing was perfect. For the 2017–18 academic year, I was scheduled to be on sabbatical, which is an excellent time to write a book, even if it was not quite the book I was supposed to be writing while on sabbatical. In the fall of 2017, I spent six productive weeks visiting Bence Nanay’s research group at the University of Antwerp. I thank him and audiences there for helpful feedback. Later that fall, Elisabeth Schellekens Dammann hosted me as a visitor for a month at the University of Uppsala, where I got to present a few chapters and, again, received much help. Thanks. This book also benefited from a fellowship at the Paris Institute for Advanced Study (France), with the financial support of the French State, program “Investissements d’avenir” managed by the Agence Nationale de la Recherche (ANR-11-LABX-0027-01 Labex RFIEA+). Conditions there were ideal. My colleagues were engaging, the Institute was supportive and beautiful and perfectly located. I thank Gretty Mirdal, the Director at the time, Simon Luck, and the rest of the staff for making my stay so wonderful. While in Paris, Roberto Casati and Jérôme Dokic organized parallel EHESS seminars on iconicity in which I was able to present the book in its entirety. They also brought in other people who specialize on pictures, like Catharine Abell, Katerina Bantinaki, and John Zeimbekis. Enrico Terrone arranged a presentation at the Institut Jean Nicod’s Aesthetics and Cognitive

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

x

  

Science Seminar. François Recanati made many helpful suggestions. I thank all of the participants at these seminars, and the members of the Institut Jean Nicod, who also made my stay in Paris productive and engaging. Also while I was in Paris, the London Aesthetics Forum provided a familiar and welcome place to discuss metaphor in pictures. Thanks to Andrew Huddleston for organizing. I presented parts of Chapter 4 at the philosophy colloquium at Goethe University in Frankfurt. Thanks to Martin Seel for the invitation. In between stints abroad I presented some material from Chapter 4 at the American Society for Aesthetics annual meeting. Thanks to the audience and especially to Thomas Adajian for helpful comments. And in the 2018–19 academic year I presented material to helpfully skeptical audiences at the University of Glasgow (thanks to Derek Brown and Robert Briscoe), NYU-Abu Dhabi (thanks to Gabe Rabin), and the University of San Diego (thanks to Jonathan Cohen and Matthew Fulkerson). Finally, Zed Adams organized a great conference on kinds of representation, at which I received very helpful comments on Chapter 8 from Liz Camp. Eliot Michaelson read through a good bit of the manuscript and provided very helpful feedback. Josef Stern read through the whole thing and offered very detailed and helpful comments. I am an amateur philosopher of language, so the messes here are my fault, but I owe Eliot and Josef for helping me avoid some real blunders. Paul Taylor offered very helpful criticisms of Chapter 5. I am much less than an amateur art historian, so his input was particularly welcome. Remaining mistakes, in all cases, are my own. Peter Momtchiloff and Sophie Robinson at Oxford University Press have been supportive throughout, and they found some very astute readers for the manuscript. Their comments have also improved things quite a bit, though I fear not as much as they would have hoped. Thanks to Sunny Park for her support, especially on the cloudy days.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

List of Figures 2.1 Three Chairs. Photo: the author.

23

4.1 Sketch. Drawn by Soo Sunny Park.

55

4.2 Two sketches. Drawn by Soo Sunny Park.

60

4.3 Hans Holbein, Sir Thomas More, 1527, oil on oak panel (cradled), 74.9  60.3 cm. © The Frick Collection.

69

4.4 Paul Cezanne, Madame Cezanne (Hortense Fiquet, 1850–1922) in a Red Dress, 1888–90, oil on canvas, 116.5  89.5 cm. The Metropolitan Museum of Art, CCO 1.0 Universal, Public Domain.

70

5.1 Raphael. Saint Catherine of Alexandria, 1508, oil on poplar, 72.2 x 55.7 cm. © The National Gallery, London.

80

5.2 Unknown, Spanish, St Catherine Delivered from the Wheel, ca 1375–1400, oil and tempera on panel, 35.8  33 cm. Art Institute of Chicago, CCO 1.0 Universal, Public Domain.

81

5.3 Palma Vecchio, Sacra Conversazione, ca 1516–18, tempera on wood, 84.5  106 cm. Collection of the Poznań Society of Friends of Sciences (deposited at the National Museum in Poznań), MNP Mo 24.

82

5.4 Albrecht Dürer, Great Triumphal Car (detail) 1522. British Museum, Public Domain.

87

5.5 Botticelli, Sandro. The Cestello Annunciation. 1489–90, tempera on panel, 150  156 cm, Uffizi Gallery. Photo: Scala/Art Resource, New York.

90

5.6 Leonardo da Vinci. Annunciation, Ca. 1472, oil on wood, 98  217 cm, Uffizi Gallery. Photo: Scala/Ministerio per I Beni e le Attività culturali/Art Resource, New York.

91

5.7 Rembrandt van Rijn, Historiestuk met zelfportret van de schilder, signed and dated 1626, oil on panel, 90.1  121.3 cm, Museum De Lackenhal, Leiden; long-term loan from the Cultural Heritage Agency of the Netherlands.

92

6.1 James Gillray, The Division of the World, 1805, colored engraving. Photo: bpk Bildagentur. Art Resource, New York. Public domain.

102

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

xii

  

6.2 Some Animals are Treated Like Garbage, 2017. Advertising Agency: Ruf Lanz, Zurich, Switzerland. CGI: Carioca Studio/Visualeyes International © Tier im Recht.

102

6.3 Pablo Picasso, Baboon and Young, 1955, Bronze, 53.3  33.3  52.7 cm. © Estate of Pablo Picasso/Artists Rights Society (ARS), New York. Digital Image © The Museum of Modern Art/Licensed by SCALA/ Art Resource, New York.

104

6.4 Nujoom, 2016. Advertising Agency: Memac Ogilvy. © Qatar Islamic Bank. 107 6.5 Child Labor, 2016. Advertising Agency: Agencia UM, Recife. © Agencia UN.

109

6.6 The Identity of Young Chennai, 2009. Advertising Agency: JWT, Chennai. © The Times of India.

110

8.1 Syntactic inseparability. The author.

140

8.2 Leaning tower illusion. Alteration of Komehakubutsukan Passage, Photographer: As6673. © CC-BY-SA 3.0. Source of original photo: Wikimedia Commons: https://commons.wikimedia.org/wiki/File: Komehakubutsukan-passage.jpg

141

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

List of Tables 1.1 The structure of the book 1.2 Character and content

7 9

1.3 Character and content 2

11

1.4 Character and content 3

12

1.5 Character and content 4

13

4.1 Attributive and referential use

58

4.2 Attributive, referential, denotative, non-denotative

66

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 25/7/2020, SPi

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

1 Pictures, Communication, and Meaning Pictures are important parts of communicative acts, along with language, gesture, facial expressions, and props. They express wide ranges of thoughts, make assertions, offer warnings, instructions, and commands. Pictures are also representations. They have meanings, which help explain the range of communicative uses to which they can be put. Modeling the meanings of pictures is accounting for the ways in which pictures manage to be meaningful, with an eye toward how those meanings let us use them as we do. Sentences have meanings too. “The weather is horrible!” can be used to express the thought that the weather is horrible, in part because of what it means. Used differently, that same sentence can communicate that the weather is wonderful, also in part because of what it means, even though it doesn’t mean that. One would be hard pressed to use a sentence about the weather to express the thought that Saturn has five rings but no chickens. It could happen, but only in an impressively strange set of circumstances. That’s because only under odd conditions could a sentence that means something about the weather communicate something about poultry and the rings of Saturn. Sometimes pictures are the only representational parts of communicative acts. They adorn boxes, showing what’s inside. They are placed on doors, suggesting a push or a pull. They are in churches, providing objects for contemplation and prayer. Other times, they are parts of acts that also involve language. “Get me one of these,” while gesturing at a picture of a screwdriver. “Find him,” gesturing at a mugshot. “The President” affixed to a photo. An account of pictorial meaning should help explain these modes of cohabitation. The philosophy of language is the most obvious place to look for tools that model meanings. If those tools do not shed light on pictures, one might wonder whether they are the right ones for language, given that pictures and language are so often partners in crime. If they do help, then the approach has all the advantages of theft, and it might shed new light on what they, developed with language in mind, are really tools for. This book offers an Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

2

, ,  

account of pictorial meaning, inspired by the philosophy of language, that does justice to the range of communicative uses to which pictures are put.

1.1 Pictures and Language Theorists of depiction have consistently borrowed concepts from the philosophy of language.¹ Both areas have articulate lives of their own, and they are both built around studying representation. They remain separate because the most salient questions in each don’t overlap with those in the other. The study of pictures focuses on the experiences they elicit, while the study of language focuses on syntax, semantics, and pragmatics. What is it about the word “wombat” that makes it about wombats? How does that configuration of letters conspire to deliver the little beasts as contents? Does the experience of letters, written just so, forge a special link to these animals? These questions sound fun, but they turn out to be uninteresting. Consensus is that just about any other word could have wound up meaning wombats, so no story linking features of letters to wombathood will be particularly illuminating. By contrast, what is it about a flat pattern of pigment that brings one into contact with a complex wombat scene? Is it a special kind of experience, the deployment of recognitional capacities, resemblance, or what? These questions are very interesting and definitely worth philosophers’ time. Pictures are understood, in part, because of the distinctive kinds of experiences they elicit, and there has been much said about what these experiences are like. With few exceptions (Goodman 1976, Kulvicki 2006, Greenberg 2013), it has been standard to argue that we can understand what makes pictures distinctive by understanding the special kinds of experiences they elicit. By contrast, very little work in the philosophy of language focuses on how reading or hearing inscriptions and utterances explains how they are understood.

¹ David Novitz (1977), Catharine Abell (2009), and Ben Blumson (2014) have made use of Gricean tools, among others, to help with understanding pictures. Elisabeth Camp (2007, 2018) has been working on modeling the meanings of maps and other representations. Gabriel Greenberg’s work (2013, 2018) is an interesting and helpful alternative to what is offered here. Dominic Lopes (1996, 2010), John Hyman (2012), John Zeimbekis (2010), and Raphael De Clercq (2015) have, in different ways, also leaned on some tools from the philosophy of language. Nelson Goodman (1976) spoke of Languages of Art, but didn’t so much use tools from the philosophy of language as build his own, which he hoped would deliver a theory of symbolization that transcends the picture–language divide. That approach has not taken hold.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

3

Central to the philosophical study of language are things pictures seem to lack. First, a rich grammar that, second, supports highly articulate modes of meaning, which, third, can be used in many ways conversationally. Syntax, semantics, pragmatics. Philosophers have disagreed about how to distinguish these three since Charles Morris introduced the terminology in 1938, but it’s worth sketching moderately uncontroversial working definitions for what follows. First, semantics is interested primarily in the conditions under which declarative sentences are true or false, in a way that abstracts from many aspects of conversational use. “I am in great shape” is true just in case the person who utters it is in great shape. When said while huffing and puffing one’s way up a mountain, that sentence can communicate the thought that one is in terrible shape. Nevertheless, semantically it says that one is in great shape. In that sense, semantics is interested in truth conditions in a manner abstracting from conversational use. To ignore some aspects of conversational use, however, is not to ignore completely the context in which the claim is made. Who, after all, is “I”? The meaning of the sentence is underdetermined when abstracted completely from context because in that case it is unclear who is supposed to be in great shape. These contextual aspects of semantics will be important in what follows. It is controversial how best to draw the line between contextual factors that matter for semantics and those that are relevant to the pragmatics of interpretation. This book will not offer a new way of doing that, so much as try to convince readers that, on just about any way of drawing the divide, there is an important place for pictorial semantics. Second, declarative sentences are complex, in an interesting way. It’s not just that they are composed of lots of noises, or letters, but that they have parts that contribute to whether they are true or false. The word “great,” for example, plays a role in determining the conditions under which the sentence above is true or false, as do the complex phrases “in great shape,” “am in great shape.” Moreover, all of these parts can occur in different complex expressions, like “The high priest is in great shape.” These parts follow rules about how they can be combined with other parts, and they affect a sentence’s truth conditions because of how they are meaningful, and combined with other parts. So, the indexical “I” has a meaning, and that makes a contribution to whether the sentence “I am in great shape” is true. Syntax is the study of what the meaningful parts of complexes are, and how they combine to yield meaningful wholes. One can describe the syntactic structure of a sentence, for example, without saying much of anything about its

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

4

, ,  

meaning, because one might just articulate the kinds of parts present and how they are combined. “I am in great shape” has a lot in common, syntactically, with “I am tall” and less in common with “The high priest might have been a contractor from Peoria.” This way of thinking about syntax, as the study of the inputs to semantics, was developed by Richard Montague (Thomason 1974), Barbara Partee (2004, e.g.), and with many variations by many others. Third, if syntax studies the substructures that contribute to semantic interpretation, pragmatics studies what people do with meaningful things. “I am in great shape” can communicate that I am in terrible shape and it communicates that in part because of what it means. Semantics is just one of many inputs, as it were, to pragmatics, or all-purpose interpretive activity. You say you are in great shape, as you struggle up the mild slope. Your interlocutors know you are struggling, they know that you aren’t generally deluded, and they assume you are trying to communicate something. So, they correctly take you to be communicating that you are in terrible shape. They would not have thought this if, while huffing and puffing, you said that the summit is near. In that case, the meaning of the sentence uttered would probably be taken to match what you intended to communicate. Sometimes what is communicated closely matches the semantic content of what is said, but often it does not. HP Grice (1989, e.g.) was a pioneer in unpacking what he called “the logic of conversation,” which gave shape to this way of uncovering communicative significance based on what someone says. It remains highly controversial just how one ought to draw a line between semantics and pragmatics, and whether that is even the important distinction (see, e.g., Szabo 2005). At one extreme, Dan Sperber and Diedre Wilson (1995, 2012) suggest that there is not much of an interesting role for semantics proper to play, because recovering meaning is so deeply dependent on highly variable contextual factors and general inference mechanisms. Others suggest that there is an interesting role for semantics that is minimally affected by contextual factors (Borg 2004, Cappelen and Lepore 2005). And others allow that much of what is identified as semantics depends on contextual factors and not just conventional meanings of words and the ways they are syntactically combined (Bach 1994, Recanati 2002, Korta and Perry 2008). Disputes about the line between semantics and pragmatics are not this book’s focus. Important for present purposes is that pragmatics makes use of all-purpose reasoning to figure out what someone is trying to communicate. Any feature of the context in which communicators find themselves can be

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

5

relevant to this interpretation. Things are much more limited with semantics. The thought about semantics is that the linguistic expressions carry meanings that they bring to the situations in which they are used, though some aspects of meaning might only get fleshed out in context. Knowing those meanings is a matter of understanding the language. Understanding a language allows one to use the meanings of its expressions, among other things, to figure out what someone intends to communicate. When interpreting pictures, much of the task falls uncontroversially within the realm of pragmatics, but another important part is semantic. The goal, starting with Chapter 2, is to show where semantic considerations seem important for understanding pictorial interpretation and communication. With this sketch in mind, note that pictures seem to lack articulate syntactic structure. They thus lack the ability to carry the range of meanings that sentences in language do. Without noun phrases, verb phrases, anaphors, demonstratives, proper names, and (in)definite descriptions, how can tools developed in light of such structures be deployed to understand pictures? In addition, it is far from clear that pictures are really the kinds of representations that are generally taken to be true, or false (see, e.g., Crane 2009). Pictures, according to a standard view, are meaningful to the extent that they elicit experiences of things and scenes. Those things and scenes are pictorial meanings. Unlike lexemes and the sentences they make up, pictures are not conventionally linked with meanings. Pictures of wombats could not just as well have been pictures of ducks. Tied as they are to perceptual abilities, pictures do not have their meanings in virtue of complex syntactic combinations of meaningful parts, but rather because they present viewers with arrays that yield these distinctive experiences. Perhaps, just as pictures lack interesting syntax, they lack interesting semantics. In fact, it might seem that the range of communicative ends pictures can serve may be explained in terms of just two things: the experiences they elicit, and general pragmatic communicative mechanisms. In the current literature, Catharine Abell’s (2009) position most clearly exemplifies this approach, but many adopt stronger and weaker versions of it. The point of this book is to suggest that the hypothesis is false and that syntax and semantics are helpful for understanding pictorial communication. Though pictures do not have the articulate grammar found in language, they are syntactically complex. They have a structure that helps explain what they mean and, by extension, the communicative uses to which they can be put. It’s just that their structure is radically different from that found in language. Second, though the semantic tools used here were developed with

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

6

, ,  

the structure of language in mind, they can be deployed quite effectively to show what can be done with pictures. And third, while philosophers working on pictures are right to think that they can elicit distinctive kinds of experiences, they should be open to locating those stories about understanding pictures within a broader semantic frame, either the one offered here or one of their own devising.²

1.2 The Structure of the Book Bringing philosophy of language into contact with the study of pictures requires two threads. The meaning thread shows that pictures have contextindependent meanings and context-dependent meanings. This will hopefully convince readers that there is important room for semantics in the study of pictures. Operations defined over such meanings can explain a wide variety of communicative uses to which pictures are put. These operations will help distinguish attributive, or descriptive, uses of pictures from those in which pictures seem to have singular contents, as well as iconographic and metaphorical uses of pictures. The parts thread is devoted to understanding the syntactic structure of pictures, how it differs from language, and how it’s similar to the structure of other kinds of representation like maps. Pictures have syntactic parts too, which are, like those of language, meaningful contributors to the overall meanings of pictures. This will lead, in Chapter 8, to a new way of distinguishing pictures, maps, and the like from language. The syntactic parts of pictures are quite different from the syntactic parts of languages, though all of them qualify for being syntactic parts. In addition, there are many cases in which different parts of pictures are used for different semantic ends. Without a detailed understanding of what picture parts are, this is bound to seem a mystery. Both of these threads are relevant throughout the book, though individual chapters can focus more on one than the other. Chapters 2 and 3 introduce the meaning and parts threads, respectively. Things fan out from there. The meaning thread is most prominent through Chapters 2, 4, 5, and 6, while the parts thread is most salient in Chapters 3, 7, and 8. Table 1.1 shows the logical structure of the book. After Chapter 3, readers can jump pretty much anywhere they like. ² Abell (2009), Blumson (2014), and Greenberg (2018) are examples of this, as is Cumming, Greenberg, and Kelly (2017) in the case of moving pictures.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

7

Table 1.1 The structure of the book

1

—

2

—

3

— — — —

4 5 6 7

—

8

The rest of this chapter offers an overview of each thread, with little in the way of defense. This is important because it’s easy, at any given point, to lose the forest for the trees. Readers will likely resist this story at many points, but keep in mind: this is an overview of the completed view, which will get its arguments and other support as the book progresses.

1.3 The Meaning Thread Pictures are representations with richly descriptive contents. So, pictorial content is akin to the content of a description, in that it specifies a complex set of features. One picture describes a pattern of light and shadow cast across a table, another a person looking through a window. The level of detail in each picture can be quite high. A person outfitted, positioned, illuminated, colored, and shaped just so is looking through just such a window, and is shown from a perspective far above his left shoulder. The detail might be so high, in fact, that no linguistic description of finite length could capture it. Some pictures say quite a bit less. They are sketches by comparison to their more detailed cousins. What’s important for present purposes is that in all cases the content is descriptive. If pictorial content is descriptive, what kind of descriptive content is it? Are pictures like definite descriptions or indefinite descriptions? Neither. In many languages, descriptions are not marked as definite or indefinite, and context of use makes it clear which way they should be interpreted (Ludlow 2018). Pictures work that way too. Sometimes they are used as definite descriptions, sometimes as indefinite descriptions. Outside of being used for this or that communicative end, there is no fact of the matter about definiteness. A context can involve the use of language—I need one of these (pointing at a picture of a hammer)! He (indicating a photo) stole my watch!—or not. Pictures on drawers in a hardware store show the kind of thing in the drawer, for example, and are thus understood in an indefinite manner, without linguistic help.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

8

, ,  

One immediate consequence of pictorial content being descriptive is that it does not include particular individuals. Obama, the number one, a particular scene, and redness cannot be constitutive of pictorial contents. A picture can refer to Obama if it correctly describes him, just as a description in language can. In neither case is Obama part of the descriptive content. This point is important because there are many cases in which it seems as though particular individuals or scenes are constitutive of pictures’ contents. One important goal of the book is explaining three ways in which that can happen, given that pictorial content is merely descriptive. Before that, however, it helps to answer a question about their descriptive contents. Specifically, do pictures have their descriptive contents independently of the context in which they are used, or not? David Kaplan (1989) suggested that demonstratives and indexicals show the need to acknowledge two kinds of meaning: character and content. When I say “I” I mean something different from what you mean when you say it, but in another sense we mean the same thing. Indexical expressions like “I” and “now” have non-constant characters. Character is a meaning that an expression brings to the context in which it is used. Within that context, the character is fleshed out to a full-blown content. You saying “I” makes it the case that the expression picks out you, while my saying it has different consequences. Character is, in effect, a rule that delivers a content in context. Many expressions in language, like “fire engine,” have constant characters, in that every context maps them onto the same contents. Recent work in philosophy of language explores many cases in which expressions that are not indexicals or demonstratives are nevertheless context sensitive. You can say “Humphrey is tall” and mean something quite different from what I mean because you are comparing him to wombats, I to elephants. Gradable adjectives, modal verbs, and many other expressions seem to depend on context to be fleshed out. By contrast to pure indexicals, these expressions manifest more complicated rules for delivering contents in context. “Tall,” for example, might mean something like having a great standing height for an X. A rule then fills in the X—for a wombat!—to give a full-blown content. The rule, for example, might be that you find the comparison class about which one is speaking and make that X.³

³ See, for example, Kennedy (2007, 7–8) and Braun (2015) for overviews. Kennedy does not unpack his account of gradable adjectives in terms of character and content, though he isolates both context-sensitive and context-insensitive aspects of their meanings. In general, the

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

9

Pictures are like these context-sensitive linguistic expressions. It’s a longstanding thought that pictures are interpretatively open. Does this depict a square tabletop half in shadow, or a well-illuminated trapezoidal part of a multicolored surface? A circular surface seen head-on, or an ellipse seen obliquely? Such openness is ubiquitous, but also constrained. Not anything goes. The picture might depict a circle oriented just so or an ellipse oriented differently, but it does not depict a square seen head-on. This shows that pictures have context-insensitive meanings that, in context, are fleshed out to something more determinate. It’s a centerpiece of my account of what makes pictures distinctive (Kulvicki 2006) that they have “bare bones” contents, which constrain their fleshed-out contents. The right way to understand this, Chapter 2 argues, is that bare bones contents are context-invariant meanings, or pictorial characters, which in context determine pictorial contents, or pictures’ context-sensitive meanings. So far, then, we have the description of pictorial meanings shown in Table 1.2. Some might be inclined to disagree with the claim that pictures have bare bones contents. For present purposes, that’s fine, though the defense of this view in Chapter 2 gives reasons for accepting it. Even if one disagrees, the distinction between invariant and dependent meanings will prove helpful because it can also model how pictures can sometimes have particular individuals as their contents. This is important because there are many cases in which it seems like that’s exactly what pictures do. For example, the photos on postcards are often meant to deliver a locale, about which one might say something. Portraits, photographic or handmade, might function to deliver the Lord of the estate, the President, or the monarch, as someone about whom something is said, or with respect to whom one might perform certain acts. Police sketches and mugshots Table 1.2 Character and content Character Bare bones content, descriptive

Content in context, determines a

Pictorial content, descriptive

character-content vocabulary is a bit strained here because of how it’s rooted in indexicals and demonstratives. I am using it because there are no other convenient ways of talking about this, and it will be important to talk about both of them a lot in what follows.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

10

, ,  

sometimes play a descriptive role—the offender looks like this—but sometimes just function to deliver an individual—this guy is a thief! Icons deliver Saints for acts of devotion. In many of these cases, the best way to interpret the picture is as having an individual as its content. Put differently, such pictures have singular contents. Interpreting the picture descriptively, or attributively, is incorrect. Kaplan (1978, 1989), following up on some thoughts by Keith Donnellan (1966), suggested a simple model for showing how a definite description can wind up with a particular individual as its content. Details are for later, but for now what’s important is that the model yields an individual as a content by applying a rule to its original descriptive content. In the pictorial case, the pictorial content, which is descriptive, refers to something or other in the context in which it was produced, let’s say Obama. In another context, it might be that a picture with such a descriptive content refers to someone else, who, for example, looks a lot like Obama. So, descriptive contents can denote different individuals depending on the context in which they are used. Now, take that descriptive content and treat it like we were treating bare bones content earlier. That is, treat it as part of a (new) rule, or character, that determines a new content for the picture. The new rule is: find what the picture refers to in the context in which it was produced. That rule delivers a new, dthat content, which, in this case, is Obama.⁴ Now Obama is the meaning of the picture, instead of some complex descriptive condition that happens to single Obama out among other people. This shows one way in which pictures, whose contents are first and foremost descriptive, can have individuals as their contents. This model works well despite the fact that it was developed with language in mind. Also, if one is tempted by the thought that such pictures have individuals as their contents, a semantic mechanism like this one is the best way to get that result. So, the discussion of dthat content is another way of supporting the claim that semantics is a useful tool for thinking about pictures. All of this will require more unpacking in Chapter 4, but for now Table 1.3 shows the expanded model of pictorial meanings. Dthat content consists of individuals that are denoted by the rich descriptive contents of pictures. Sometimes we have no clue what someone or

⁴ For now, the name just helps identify a new content. I will explain the name, which comes from Kaplan (1978), in Chapter 4.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

11

Table 1.3 Character and content 2 Character Bare bones content, descriptive Pictorial content, descriptive

Content in context, determines a

Pictorial content, descriptive

in context, determines a

Dthat content, singular

something looked like, and sometimes the individual we want to represent has no visual appearance at all. In those cases, there is no way for dthat to make such individuals the contents of pictures. No one knows what St Catherine of Alexandria looked like, and Fidelity does not look like anything, but some pictures are meant to have such contents. A different semantic rule, defined over pictorial content, does precisely this. Chapter 5 suggests that in iconographic interpretation one starts with a pictorial content, then searches it for attributes, which are associated with some individual. Attributes are the kinds of things that can be depicted, like keys, eyes on plates, men, women, animals, and so on. Upon finding such things in the pictorial content, the individual associated with those features becomes the new, iconographic content. A panel represents Saint Lucy because it depicts a woman holding a plate that has a pair of eyes on it. The older man with the keys is St Peter, while the scruffy guy in sack cloth is John the Baptist. Tellingly, when some aspect of pictorial content is used as an attribute to fix iconographic content, it ceases to be an attributive part of the picture. The picture does not depict Peter holding keys, because it was only insofar as a guy was depicted with keys that Peter became the content in the first place. This adds a way in which pictures have meanings to the list (Table 1.4) and deepens the case in favor of semantic rules being relevant to pictorial interpretation. Iconographic interpretation greatly expands the range of things that pictures can represent. No, iconographic content is not pictorial content proper, but iconographic interpretation depends on pictorial interpretation, so pictorial content is involved in generating it. The same is true for dthat content, but with iconographic interpretation pictures can represent vast ranges of things that might not even be perceptible. This proposal does not capture all of what art historians call iconographic interpretation, but it does capture one theoretically interesting part of it. Though easy examples of this can be found in European art history, it’s arguable that all cultures with rich picture-making traditions developed

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

12

, ,  

Table 1.4 Character and content 3 Character Bare bones content, descriptive Pictorial content, descriptive Pictorial content, descriptive

s in context, determines a

Pictorial content, descriptive

in context, determines a

Dthat content, singular

in context, determines an

Iconographic content, singular

iconographies, and that iconographic interpretation is at work even in a lot of contemporary practices. What’s more, something akin to iconographic interpretation happens in language, so the development of these tools to deal with pictures can help one spot interesting linguistic phenomena too. The most exploratory part of the book considers pictorial metaphor. Like iconography, it’s not a topic that has received a lot of attention from philosophers. It is far from settled what the best account of metaphor is in the first place, and this book does not aspire to offer one. Instead, Chapter 6 takes an account of metaphor due to Josef Stern (2000), articulated in terms of character and content, and shows how to apply it to pictures. The results are encouraging, and interesting, because Stern (1997) denies that there are distinctively pictorial metaphors. For Stern, interpreting something metaphorically amounts to applying a rule to its literal content in order to produce a new, metaphorical content. He denies that this applies to pictures in part because he thinks that semantic tools like character and content don’t work with pictures. Unlike dthat and iconographic interpretation, metaphorical interpretations do not yield individuals as contents. Instead, they yield new sets of features, which are metaphorically associated with the ones that they literally depict. Metaphorical associations can connect properties in pictorial content to pretty much any others. Metaphor is thus a way in which pictures can wind up representing properties that they could not otherwise represent. This chapter is exploratory because Stern’s account of metaphor is highly controversial, and the weight of opinion, to the extent that there is one, is that metaphorical interpretation is not semantic, but wholly pragmatic. Table 1.5 summarizes the kinds of contents pictures can have, according to the model pursued in Chapters 2, 4, 5, and 6. The left-hand column in Table 1.5 includes pictorial content in all rows except the first. In that sense, all of these meanings are thus determined

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

13

Table 1.5 Character and content 4 Character Bare bones content, descriptive Pictorial content, descriptive Pictorial content, descriptive Pictorial content, descriptive

Content in context, determines a

Pictorial content, descriptive

in context, determines a

Dthat content, singular

in context, determines an

Iconographic content, singular

in context, determines a

Metaphorical content, descriptive

partly by and depend upon pictorial content, the contexts in which they are used, and the communicative purposes to which they are put. What this means is that icons, for example, are not a separate kind of representation, but a use of pictorial representation. Ditto for metaphor and dthat. These are all pictorial meanings in the sense that pictures can have them, even though only one of them deserves to be called pictorial content proper. Those allergic to bare bones content will not find the first row plausible, but that is not prerequisite to finding the other rows appealing as a model of pictorial meanings. In each case, the claim is that a distinct rule determines some meaning in context. The rules deliver different results depending on the context in which they are invoked, which is what makes the meanings context dependent. But what triggers the application of a rule? Why apply Rule 2, as opposed to 3 or 4? In brief, the demands of communication. The goal here is a model of pictorial meanings that sheds light on how pictures are used communicatively. Sometimes pictures are used attributively, sometimes they deliver an individual, iconography is common to many picture-making cultures, as is pictorial metaphor. Pictures can do all of these things because of how their meanings can be deployed and transformed. The one rule that’s not optional, according to the current view, is Rule 1. Pictures are, fundamentally, representations that admit of multiple interpretations, built around a core character—bare bones content—that constrains their contents. Bare bones content explains the multiple ways pictures are interpreted pictorially across different contexts. And it is pictures, interpreted as such, that then go on to be used in all of those other ways. That’s why the inputs to all of the other processes—those that deliver dthat, iconographic, and metaphorical meanings—are pictorial contents.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

14

, ,  

1.4 The Parts Thread The meaning thread shows that some tools developed for understanding language can also help us model pictorial meanings. Those tools not only apply to whole pictures, but in some circumstances can apply individually to their syntactic parts. So, another important thread in this book, starting with Chapter 3, is explaining what the syntactic parts of pictures are, and showing that some practices with pictures depend on breaking them up into parts. The parts thread also shows, in Chapter 7, that pictures are syntactically much like maps. This leads, in Chapter 8, to a new way of understanding how both kinds of representation differ from language. The radical break between pictures and maps, on one side, and language on the other, makes it all the more interesting that the tools developed in the meaning thread work with them. They are much more broadly applicable than has been appreciated. The simplest way to think about picture parts is to imagine cutting them up. Elliot Sober (1976) suggested that when pictures are cut to pieces, the result is just more, albeit smaller pictures. This idea was a staple in discussions of mental imagery through the 1980s and 1990s and even as recently as Jerry Fodor’s LOT-II (2008). It also shows up in recent discussions of pictures (Abell 2005, Blumson 2014, Ch 6). Importantly, pictures do not come with instructions for how to cut them up, which means that there are many ways of doing so, all equally good at generating picture parts. Chapter 3, leaning on earlier work of mine, revises this way of thinking about pictures’ syntactic parts. Cutting yields parts because it is a way of focusing on some of a picture’s syntactic qualities to the exclusion of others. Once that becomes clear, it does not take long to notice that there are many non-spatial ways of breaking pictures into parts. For example, one can ignore the hues in a color photograph and focus on variations of light and dark. This applies to the whole picture and not just to a spatial part of it. One can ignore the high-frequency spatial features of pictures and focus on their low-frequency features. That’s spatial, but not in a way that cuts pictures to pieces. One can ignore specific colors to focus on less specific ones, and the same goes for shapes. What is interesting about pictures, maps, and other such representations is that systematically ignoring some of their syntactic features leaves one with interpretable representations. Nothing like this works with language. Leave out the verbs in this book and the result is (more of) a mess. Ignoring some of the syntactic features of pictures results in representations that say a

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

15

bit less about the scene than the original does. So, focusing on low-frequency spatial features leaves a picture that only speaks to the low-frequency spatial features of its scene. Ignoring color detail leaves a representation that says little about the specifics of color. Pictures represent many features of scenes, and they represent them across many levels of abstraction, from the specific to the very general. Because pictures have so many parts, it’s possible to interpret different parts of pictures differently. Parts of a picture, for example, have bare bones contents and pictorial contents. It’s possible for one such part to receive a dthat, iconographic, or even metaphorical interpretation, while the rest is interpreted attributively. Iconographic attributes might single out a saint, while other aspects of the picture say something about her. One part of the picture isolates a referent while the other works attributively. Seeing this in detail requires considering expressive acts in which pictures participate, such as being in sales catalogs, family photo albums, or stories meant to instruct the illiterate. This is why the parts thread, started in Chapter 3, runs through Chapters 4, 5, and 6 even if it is not their focus. Pictures do not come grammatically organized. No parts are singled out as playing a subject or object role. None are isolated as those whose role is to comment on something else. Fundamentally, pictorial content is descriptive, and can be transformed via other processes into other contents, but pictures are not structured like sentences. That doesn’t mean pictures are unable to express things like sentences express. It’s just that in order to do so pictures need to be embedded in conventionalized practices, or highly specific contexts, in which it becomes clear which parts are to play which roles. So, pictures provide syntactic and semantic structure that can, in the right communicative contexts, be expressively quite rich. Pictures in family photo albums, for example, can properly be interpreted as saying of those people it represents that they are in such and such a situation. The same photo, included in a study of tourism patterns at a specific site, says, of that place, that it is occupied with people like this.⁵ Ignore all of the details that show up at a particular spot on a picture or map surface, and what is left is a bare location. Those locations are parts of pictures or maps in much the same way that unspecific colors and shapes are, as Chapter 7 shows. Locations cannot play any attributive role of interest, but in many cases they are recruited as singular terms that refer ⁵ Novitz (1977, Ch. 5) makes similar points about pictures but he neither locates depicting within an articulate semantic frame nor offers an account of pictures’ syntactic parts.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

16

, ,  

directly to locations. Photographs, security camera footage, and maps are perhaps the most obvious cases in which this happens, but it’s possible for many handmade pictures to do this as well. In fact, the point about locations serving as referring terms has been a part of the maps literature since Roberto Casati and Achille Varzi’s Parts and Places (1999). In all of these cases, maps and pictures are readily understood as expressing propositions. They say, of some part of space, that it is thus and so. Maps differ from pictures in two important ways. First, they do not have bare bones contents, and they are thus representations with constant characters. Second, they have a richer syntactic and semantic structure than pictures. Simplifying a bit, pictures can say, of a given region of space, how it’s colored. Pictorial content is much richer than that, but pictorial content is built up from bare bones content, which is, fundamentally, a twodimensional chromatic pattern. So, in effect, all you can do to make a picture is place colors at locations. Maps, by contrast, place many syntactic features at each location, and thus can say of some region of space that it is dry, mountainous land crossed by a dirt road. Another map might say of a place that it is urban, with a given population density and average income. Though maps are structurally more sophisticated than pictures, they have syntactic parts of the same sort. In both cases, syntactic parts are found by abstracting over syntactic detail. Chapter 8 suggests that this is the only syntactic operation of interest defined over pictures and maps. Abstraction over detail is not an operation defined over linguistic representations, so this constitutes a new way of distinguishing the two. Sentences are broken up into parts grammatically. Or, to put things the other way around, they are assembled from lexemes into complex wholes that have grammatical structure. Lexemes fill places in these structures, which might be, for example, noun phrases, adverbial phrases, and so on. The parts of a linguistic representation are thus found by finding the lexemes that fill the spots required by the grammar. Those parts can show up in different places, within different structures. So, we can say John loves Mary or Mary loves John. In an important sense these two representations have the same smallest parts—John, Mary, loves—which have just been assembled differently. Each of those parts keeps its syntactic identity across the many grammatical structures in which it might find itself, much as a building block can be detached and refitted with others to form a new structure. That’s roughly what it means for syntactic parts to be separable. Pictures and maps don’t have grammatical structures into which separable parts can fit. Instead, their parts are found by abstracting over syntactic

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

    

17

detail. Syntactic parts of pictures, then, are syntactically relevant features that are less determinate than the most determinate feature the picture has. Therefore, any proper syntactic part of a picture will be such that it demands the presence of other syntactic parts. In that sense, none of the syntactic parts of a picture are separable in the way linguistic parts are. So, a very simple way to distinguish pictures, maps, and the like from language is that linguistic representations only have separable parts, while imagistic representations have no separable parts. The massive difference between imagistic and linguistic representations makes it interesting that the models described in the meaning thread work at all. The semantic tools developed for understanding language are not limited to modeling linguistic phenomena.

1.5 What This Book is Not There are many theories of depiction on the market, but this book is not another theory of depiction. Nor is it just an extension of my own theory. Most theories of depiction ask what is special about pictorial interpretation. It seems like a different game entirely than interpreting language because of the way pictorial perception exploits perceptual mechanisms. That, however, is not the same thing as giving a model of pictorial meanings and how they relate to one another. Chapter 2 claims that pictorial content is descriptive, and Chapter 4 offers a model for how particular individuals can become the contents of pictures. This will seem to conflict with many other theories of depiction. For many, the same mechanism that allows us to understand pictures’ attributive contents allows us to understand them as representing this or that particular individual. A recognition theorist, for example, will likely say that a picture represents Obama because it elicits recognition responses for Obama, which is the same way in which the picture comes to represent a pattern of qualities in space. Resemblance theorists of many sorts could say the same thing about their own views. No special song and dance is required for understanding how individuals become the contents of pictures. It’s worth noting that such theories do not obviously conflict with the claims being made here. One might recognize Nixon and a guy who looks thus and so in a picture, but that doesn’t yet answer whether the picture’s content is singular, attributive, both, sometimes one sometimes the other, or what. That’s not to say that all of these theorists should agree with the

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

18

, ,  

approach offered here. In offering these models, this book opens up a new topic for discussion. Yes, at the end of the day, the proposals made here work seamlessly with my own theory of depiction. That is by design, but it was easy to do that because my starting point all along was trying to understand pictures’ syntactic and semantic structure. This book does offer a new account of a general class of representations, which includes pictures, maps, many diagrams, and the like. But a theory of that general class can also be compatible with multiple theories of what makes pictorial representation special. A similar point holds for theories of maps. There are a number of accounts of cartographic representation on the market, but the general class into which these theories fit is compatible with multiple theories at the base level.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

2 Character, Content, and Reference The goal of this book is to offer an account of pictorial semantics. What do pictures mean? Can one picture mean different things when deployed in different communicative acts, in different contexts? If so, what explains the range of meanings pictures can have? It has been common in the philosophy of language since Frege to distinguish levels, or dimensions of meaning. Some meanings are contextindependent, others depend on the context in which an expression is used. The transcendent meanings can play a role in determining context-sensitive meanings. Indexicals like “I” and “now” are examples of this. You and I mean different things when each of us utters “I am leaving,” but in another sense we mean the same thing. Similarly, I describe a tall red oak on the ridge and you say your tall friend is standing next to it. In a sense, we mean the same thing by “tall”. But since none of your friends is tall for a red oak, in another sense we mean something different. The meanings of “I” and “tall” that transcend context play a role in determining what they mean in context.¹ Pictures have a similar structure, in that they have meanings that transcend context, which play a role in determining their context-sensitive meanings. This process—one meaning determining another when embedded in a context—will help explain many ways in which pictures can be meaningful. Moreover, noticing this about pictures is one important step to seeing why it’s important to think about pictorial semantics. If pictures act like these context-sensitive linguistic expressions, then the tools developed for dealing with the latter might also help with understanding the former. There is no unified vocabulary for talking about these kinds of meaning within the philosophy of language. David Kaplan (1989), focusing on indexicals and demonstratives, distinguished their characters from their contents. This terminology is helpful, even though pictures are neither indexicals nor demonstratives. Neither character nor content can be identified with a representation’s referent, so reference is yet another aspect of meaning. ¹ It’s not easy to unpack exactly how the meanings of such gradable adjectives depend on context. See, for example, Kennedy (2007). Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

20

, ,  

After introducing Kaplan’s distinction, this chapter argues that pictures have characters, which play a role in determining pictorial contents. Exactly what pictorial characters are, and exactly how they determine pictorial contents, will take some unpacking. The chapter ends by considering some issues related to pictorial reference. This machinery will prove quite helpful in explaining many other ways pictures can be meaningful in subsequent chapters.

2.1 Kaplan’s Distinctions Kaplan introduced character and content in order to distinguish the context-sensitive meanings of indexicals like “I” and “now,” and demonstrative uses of “that,” “she,” and “he,” from their context-independent meanings. “The meaning of the word ‘I’ does not change when different persons use it” (Kaplan 1989, 520), but when I say “I” I mean me, which is not what you mean when you say it. We understand what such expressions mean in a way that transcends contexts in which they are deployed, and so indexical expressions are not haphazardly ambiguous. When you say “I went to the store,” I don’t need to figure out which meaning of “I” you had in mind: the one that picks out me or the one that picks out you. Instead, these expressions have contents that depend in a regular manner on contexts of use. Character, for Kaplan, is a semantic feature of an expression that delivers a content, in context. So, roughly, “I” delivers the person who utters it, or writes it, as its content. It is prerequisite to understanding “I” that one grasp some such semantic rule (Kaplan 1989, 505–6). Demonstrative uses of “that” deliver the objects demonstrated as contents. Character is semantic, in that it concerns what an expression means, and it captures what users understand about a context-sensitive expression across contexts. It might be hard to unpack just what an expression’s character amounts to. Even the humble “I” works in strange ways when deployed in quotations, fiction, or even on answering machine messages (Michaelson 2014). Being able to give articulate voice to the rules governing these expressions is not prerequisite to understanding them, however, and thus using them effectively. Content is what character delivers in context. Just as character is a function from contexts to contents, content is a function from “circumstances of evaluation” (Kaplan 1989, 502) to extensions. Those less familiar with this material could notice at this point how content is semantic, in line

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

’ 

21

with the sketch of semantics offered in Chapter 1. An important part of understanding when a given declarative sentence is true is knowing what its parts apply to—their extensions—and how the sentence relates those parts to one another. Indexicals and demonstratives like those just mentioned have contents that are constant, or “fixed” functions to extensions, even though their characters are not fixed functions to contents.² Imagine unearthing the written fragment “I met a traveler in an antique land” in a ruined library, well out of the context in which it was produced. True, or false? Well, who was speaking? The character of “I” is a function to a content, which in this case is a constant function that delivers an individual as an extension. One needs to know which individual penned the phrase, and how she intended it, before evaluating the claim. For the indexical “I,” the individual is the same across circumstances in which the expression is evaluated, so one just needs to know who said it. Details of the circumstances, like finding old texts amid the rubble, have no bearing on what the expression picks out. By contrast, most definite descriptions like “the yellow book on the top shelf” deliver different extensions when they are evaluated in different circumstances. In one circumstance Amitav Ghosh’s In an Antique Land is the yellow book on the top shelf, while in another it is Shelley’s collected poems. The description’s content delivers different extensions across circumstances in which it is evaluated. Unlike the indexical, the description’s content is not a constant function from circumstances of evaluation to extensions. Some descriptions, like “the prime number between seven and 12,” have contents that are constant functions from circumstances to extension. These descriptions denote what they do because the features that constitute their contents isolate exactly the same thing across all circumstances. Though the character–content distinction was introduced to accommodate a curious subset of linguistic expressions, it trivially applies to all of them. Most expressions, for example, have constant or fixed characters, in that they deliver the same contents across all contexts in which they are used. The description “the yellow book on the top shelf,” for example, has a content that is a non-constant function to extensions, but its character is a ² Putting things this way is slightly misleading, but not in a manner that will get us into trouble. For Kaplan, the character of an indexical determines a referent, which is itself the content. So, the content is independent of the circumstances in which the expression is evaluated, which makes it a bit misleading to call the relation here a constant function from one to the other. In these cases, Kaplan says, the referent determines the content, while in cases of indirect reference, as with descriptions (see below), the content is what determines a referent.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

22

, ,  

constant function to contents. In all contexts, the content of the expression is the same, unless color adjectives are context sensitive. Over the last few decades more and more linguistic expressions have been analyzed as semantically context dependent, and in that sense as having non-constant characters. Gradable adjectives, like “tall,” discussed earlier, seem to have different meanings when used to assess mice as opposed to people, or mountains. “Yellow” in the above example might also be an example along these lines, since just what counts as the yellow thing depends on the context in which such an expression is deployed.³ For example, placed among books that are mostly gray, just about any yellowish book counts as a yellow book. But if every book is yellowish—some chartreuse, some a bit orange, and so on—only some of them will count as yellow. The extent of context dependence in language, and the proper account to give of it, are both matters of controversy.⁴ Usually, the term “character” is used specifically for meanings of indexicals and demonstratives. In what follows, however, the term is used generically to refer to meanings that are functions from contexts to contents. The goal is to show that pictures exhibit a semantic structure that needs something like the character–content distinction, in a way related to how Kaplan understood some demonstratives. Their semantic context dependence can be teased out by the careful consideration of cases. In fact, an idea that has been around for a while, that pictures have bare bones contents (Gombrich 1961, Haugeland 1998, Kulvicki 2006), is key to understanding the distinction between pictorial character and content, as the next sections show.

2.2 Interpreting Pictures Imagine an art installation, No Chairs, After JK (Figure 2.1). It’s inspired by Joseph Kosuth’s Three Chairs, A Chair (1965), Sherrie Levine’s appropriation art from the 1970s and 1980s, and Arthur Danto’s (1964) affection for indiscernables. The first thing to say is that these pictures are indistinguishable from one another, and thus seem to have the same meaning. Before rushing to judgment, however, consider the following description of the installation.

³ See, for example, Kennedy and McNally (2010) and Hansen (2011). ⁴ For some recent discussion see Borg (2004), Cappelen and Lepore (2005), Glanzberg (2012), King (2014a), Lepore and Stone (2014), and Silk (2018).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

 

23

Figure 2.1 Three Chairs. Photo: the author.

In keeping with the piece’s title, the picture on the left is not a chair, but, indeed, a photograph of a chair. Magritte could put it in his pipe, and smoke it. The ordinary way to interpret such a photo is to say that it represents a chair, or more carefully, a scene involving a chair situated in such and such a manner. So, this interpretation is not a stretch. The middle photo is clever, and steals a trick from Sherrie Levine’s After Walker Evans (1979). It is a photo of the leftmost photo of the chair, head-on, without remainder. It does not depict a chair. Instead, it depicts a picture of a chair, head-on, without remainder. At this point, some folks might want to get off the bus, but stay! This claim is hardly as controversial as the next one. The rightmost photo is the strangest of the lot. It is a photo of a papier mâché mess inspired by physiological optician Adelbert Ames. This Ames Chair is eight inches tall, but extends back a few feet. Bits of paper have been arranged so as to give an impression of a chair as long as one maintains a very specific perspective on them. Move a little, and the effect is gone. Ames deployed demonstrations like this (see Ittleson 1952), and his more famous Ames Room, in order to investigate the mechanisms of visual perception. It would make perfect sense to interpret all of these pictures in the same way if they were encountered absent any obvious communicative context. Each one represents a chair, and in most contexts that is the right way to interpret them. Most theories of depiction start by trying to explain how such patterns of pigment could represent things like chairs. Pictures, they say, resemble their objects (Hyman 1989, Abell 2009), elicit recognition responses (Schier 1986, Neander 1987, Sartwell 1991, Lopes 1996), experienced resemblances (Hopkins 1998), sui generis states of seeing-in (Wollheim 1980), and odder states still (Polanyi 1970).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

24

, ,  

In some contexts, however, the chair’s salience diminishes in favor of other options. Editors in art publishing houses spend hours dealing with so-called reproductions. These are usually just photos that depict other pictures, head-on, without context or remainder. They can be used as substitutes for the works they depict because the scenes photographed are quite simple. The picture is depicted head-on, without remainder, in even light. Instead of presenting a complex scene involving the work, these photos can be treated as substitutes for the works in question. Most of us, of course, do not spend a lot of time making or using reproductions, and thus the salience of such interpretive options is quite low. But that’s not to say they are unimportant. One can photograph other photos, with the intention of representing them. What should such a photo look like, if not like the middle one in No Chairs?⁵ Likewise, guests at Adelbert Ames’s dinner parties would have done well to avoid rushing to judgment about the photos there. The Ames cases are quite controversial, in a way that the middle case is not, but that point must wait until later. Art contexts demand full deployment of one’s interpretive faculties, and so uncommon openness to alternatives. Without prompting by a description of No Chairs, one might be at a loss for what to say about the three identical photos. But the same would be true for three identical expressions, like “I am here now,” arranged in a concrete poem: I am here now I am here now I am here now

Cases like No Chairs shed light on pictures by foregrounding alternative, if uncommon, interpretations of them. Nothing here depends on the fact that the work is composed of photographs, as opposed to handmade pictures. A draftsman, contracted to render a chair from a certain perspective, could very well produce the image on the left. And then ask yourself what he would do if charged to produce a similarly detailed drawing of all and only the drawing he had just produced, ⁵ Most accounts of depiction start with the most obvious cases of pictorial interpretation, and say little about the one stressed here. Wollheim (1980) and Polanyi (1970) are unable to accommodate these cases easily, while most of the other accounts mentioned, as well as Walton’s (1973, 1990), can be stretched to do so. That’s not to say their makers would be happy stretching them that way. My own account (Kulvicki 2006) was built around considerations like these. These odd cases are the cracks, I claim, where the light gets in.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

     

25

seen head-on in even light. Likewise, many depictions of the Ames Chair setup look nothing like a picture of a chair, but what about those drawn, with care, from the special perspective? In addition, nothing here depends on the fact that there is some actual chair, photo, or Ames mess that serves as the intended object of the picture. An artist could dream up a situation involving a chair, and set out to represent it pictorially. She could do the same for a picture of a chair, or an Ames mess, from a certain perspective. In all those cases, the resulting pictures would be indistinguishable from one another even though they lack mechanical, causal sources. What matters for No Chairs is that these pictures seem to have different contents. The one on the left has an extension that includes many chairs, the one in the middle many patterned planes, and the one on the right a variety of Ames demonstrations. Having come this far, it’s tempting to say that pictures are just ambiguous. If the three pictures in No Chairs have different contents, but are otherwise similar, then perhaps they are like the English noun “bank,” which admits of at least two salient readings. Ernst Gombrich, godfather of much work on these issues, suggested that different interpretations were due to the “ambiguities of the third dimension” (1961, Ch. 8), but ambiguity seems like the wrong tool for the job. No Chairs was invented for its art-historical resonances, but a drawing of a cat could have illustrated the point just as well. All pictures admit of these various interpretations. They are systematic, but there is nothing systematic about ambiguity in language. Also, there is an intuition, unpacked in the next section, that all of these pictures have something semantically in common despite their differences. Ambiguity requires no such commonalities. The character–content distinction is a better tool for explaining the meanings of pictures than ambiguity is, as section 2.3 suggests.

2.3 Bare Bones Content as Pictorial Character Each of the photos in No Chairs is indistinguishable from the others. But in what sense? Perhaps one uses ink made of ash, while the others use resin or minerals. Those differences do not matter. In that sense, they are irrelevant to the pictures’ syntax. Now, this is a strange use of the word syntax, especially considering it was introduced in the previous chapter in light of linguistic representations. There, the claim was that expressions are complex in that they are built of meaningful parts. Syntax was a way of understanding

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

26

, ,  

parts and the way they combine to yield meaningful wholes. The parts thread, mainly Chapters 3 and 8, will get a lot more specific about why calling these features syntactic makes sense.⁶ For now, it suffices to notice the distinction between features of a picture that are relevant to it representing what it does, and those that are irrelevant, and so the syntactic features of a picture are those relevant to it representing what it does. The chemical constitution or provenance of its inks, its mass, and its distance from the sun are irrelevant to that, while the pattern of light and dark on its surface matters. Given that, the pictures in No Chairs are syntactically indistinguishable. Keeping that in mind, the first very interesting point made salient by No Chairs is that it seems possible for pictures to be syntactically indistinguishable, and yet semantically different. If, in the minimal sense just sketched, syntactic features are those relevant to pictures representing what they do, then one should expect that semantic features do not differ between syntactically identical pictures. This strongly suggests understanding pictures on a model derived from indexicals and demonstratives. The three photos in No Chairs have different contents, but they all have something semantic in common too. These various scenes manifest the same pattern of light and dark from some perspective, namely, the perspective from which they have been represented. This perspectival pattern is two dimensional, so it abstracts from the depth that characterizes each scene. It’s also chromatically abstract, in a way that will be explained presently. That helps explain why the same photos result from pointing a camera at each scene, or why the same drawings should be expected if you situate the draftsman just so. Indefinitely many scenes are alike with respect to these perspectival features, so in that limited sense they are possible contents of the picture. Now, isolate all and only the features that each of these possible interpretations of the pictures has in common with the others. They are what the picture represents in a bare bones fashion. Ernst Gombrich (1961) was one of the first theorists to make much of this fact about depiction. John Haugeland (1998) never endorsed the claim that the three pictures in No Chairs could represent such different scenes, but he is the inspiration for the bare bones vocabulary. Using crime scene photos as his example, he says that all they “ ‘strictly’ represent is certain variations of incident light with ⁶ There is a history to this use of the term syntax. Nelson Goodman (1976) used it in much the way I am using it here. I followed his use (2003, 2006), as did Laura Perini (2005).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

     

27

respect to direction” (Haugeland 1998, 189). No blood, no mess, just patterns of color.⁷ Variations with respect to direction will, spatially, capture the perspectival features just discussed. A bare bones content might specify a trapezoid-shaped region, from a certain vantage point, but not specify that there’s a square thing there, at an oblique angle, or a trapezoid, seen headon. It might specify a region of streaked light and dark, but not specify whether it is a uniformly colored thing illuminated streakily or a streaky thing illuminated uniformly. In that way, bare bones contents are chromatically abstract.⁸ In other work (Kulvicki 2006, 2014, 2015, 2016), I have found the notion of bare bones content immensely helpful for understanding pictorial representation. It was unhelpful for Haugeland to say that bare bones content captures what photos and other pictures strictly represent. There is nothing loose about saying photos represent chairs, photos, or Ames situations. In fact, they pictorially represent such things. One of Haugeland’s goals in isolating bare bones content was his view that all modes of meaning-making and understanding can be deployed at once. So, without trying to get strict, just focusing on the many ways in which pictures can be meaningful— what he called their fleshed-out contents—one might miss what makes pictures distinctive kinds of representation. The problem with insisting that bare bones content is the only distinctively pictorial content is that no one ever interprets pictures as having just their bare bones contents. Some aspects of a fully fleshed-out content, in Haugeland’s sense, are clearly peripheral to what makes pictures distinctive. Iconographic conventions specify that a panel represents John the Evangelist, but those conventions are not central to what makes depiction different from other kinds of representation. By contrast, fleshing out the pictures in No Chairs to a chair, or a picture of a chair, seems absolutely central to depiction, and ought to be understood as such. The photo on the left strictly, pictorially represents a chair. In light of all this, consider the following two interesting facts about pictures. First, they admit of alternative interpretations in different contexts. Second, such interpretations all have something in common, a bare bones content. So, though these pictures have different interpretations in different

⁷ François Recanati (2002, 6), in discussing language, characterizes an expression’s sentence meaning as a “semantic skeleton,” which is “fleshed out” to a truth-evaluable what-is-said meaning in context. ⁸ See Hyman (2006) and Kulvicki (2020) for elaborations of this idea.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

28

, ,  

contexts, they also have something semantic in common. This makes it less puzzling to say that syntactically identical pictures can nevertheless represent different things. They have partly overlapping contents, in that each content includes the same set of perspectival features. Bare bones content captures something semantic about pictures that transcends the contexts in which they are deployed and interpreted. In addition, it is common to all of the properly pictorial interpretations of pictures. And within a context, bare bones content is precisely what constrains the many ways in which one might flesh things out. It might not be easy for viewers to unpack just what a picture’s bare bones content is, because it’s a set of typically non-salient perspectival spatial and chromatic features. Similarly, it takes sophisticated concepts to express just how a set of perspectival features might constrain the richer contents pictures seem to have, as will become apparent in section 2.4. But viewers’ willingness to accept alternative interpretations of pictures, subject to those constraints, suggests that bare bones content figures in pictorial understanding. The upshot is that bare bones content should be understood as a pictorial character, which in context delivers a full-blown pictorial content. That’s not to say pictures are just like pure indexicals, or even demonstratives, but they are tellingly similar. In fact, they work much as Kaplan (1989) suggested demonstrations work. In order to understand this point, the next goal is considering how bare bones content can act like a character: a rule that delivers pictorial contents in context.

2.4 Pictorial Content Pictorial content is an appropriate recognizable manifestation of bare bones content. That is, the way in which bare bones content determines pictorial content depends on those three things, two of which depend on the context in which a picture is used. Whether some candidate content is a manifestation of bare bones content does not depend on the communicative context. But whether that manifestation is recognizable and appropriate does. In the ordinary visual case, a manifestation of bare bones content is a way of filling out space in three dimensions from a point that is compatible with the bare bones content. In No Chairs, a chair with certain surface textures, situated thus and so from some point of view, would manifest the twodimensional pattern of light and dark that constitutes the picture’s bare bones content. Such a scene counts as a possible pictorial content. So would

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

 

29

a flat plane, colored and oriented thus and so, as well as Ames messes of many sorts. Interpreters of pictures are, in effect, looking for manifestations of bare bones content. Which manifestations they find—there are indefinitely many—depends on the tools they have for looking, which include the concepts they have for conceptualizing the world visually. Some manifestations of bare bones content are recognizable, but most of them are not. It’s only the recognizable ones that have a chance of being pictorial contents. Gombrich famously said that most possible manifestations of pictorial content have “no name and no habitation in the universe of our experience.” (1961, 249). His target, in fact, was Ames chairs, and he thought, plausibly, that they are not pictorial contents because of this. Imagine being at Ames’s house for a dinner party, and seeing the third photo from No Chairs on the wall. Yes, it’s possible that the photo was produced by pointing a camera, or a talented draftsman, at the Ames demonstration. But no one has a concept of which specific kind of Ames demo served as the object of the picture. Indefinitely many “criss-cross tangles,” as Gombrich called them, could result in that photo. Similarly, an artist could have imagined indefinitely many different Ames demos to serve as the object of his or her drawing. So, the specific kind of Ames mess—the one with paper arranged just so—is not a recognizable manifestation of the bare bones pattern. In fact, no Ames mess at all is recognizable, and so perhaps one shouldn’t say that the Ames chair is a pictorial content, at least not in this world. Failed art can be the best art. What about slightly different worlds? If there were a common kind of Ames demo, a mass-produced item wildly popular with kids, then that might work as a pictorial content because interpreters could readily recognize it. More locally, friends of Ames might have been familiar with specific instances of his demonstrations. At his house among friends, that interpretation might work. Other Ames-like options, indistinguishable from this particular Ames demo from that standpoint, would not be competitors because they have no place in the universe of his friends’ experience. For a range of scenes to count as pictorial contents, consumers need to be keyed in the right way (cf. Gombrich 1951). A chair placed just so is the ordinary pictorial content, but one easily shifts to thinking of the pictorial content as being a patterned plane, and in some contexts that shift is appropriate. In a stranger world, filled with Ames toys, the range of possible contexts increases, and so does the range of pictorial content. In this way, pictures’ dependence on context is akin to that of many other

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

30

, ,  

context-sensitive expressions in language. She said “Monty is tall,” and thereby communicated that he is tall, for a wombat. But she could not have done so if no conception of wombats were available to her and her audience. The point is not that Ames chairs can only be pictorial contents if they exist. If Ames demos are to be pictorial contents, they must somehow inhabit the world, and the toy suggestion is one way this could work. But it could also work if a vivid and popular description of an Ames demo wormed its way into our lives through literature. Dragons have been part of the popular imagination for centuries, so interpreters have robust conceptions of what such things might be like. This makes it possible to flesh out the bare bones content of a picture to a dragon, even though there are no dragons. Though one can resist the claim that an Ames chair is a pictorial content, it is not possible to resist the claim that another photo could be a pictorial content. At least not on the same grounds. There are lots of photos out there, and a simple flip of attention brings the kind of picture such a photo might depict into view. The question of whether a picture depicts a chair or a picture of a chair is answered by appeal to what is appropriate in the context, not to which manifestation is recognizable. Both are. In an art publishing house, the pictures depict other pictures. Snapping photos for the newspaper, out on the street, the pictures represent more ordinary scenes. Some contexts, like art spaces, make the standards for what would be appropriate quite open, and in those cases one might lean on what they are told about the work in question. Even in that case, not anything goes. There is much room for debate about the features that can characterize pictorial content. For example, is being a chair part of pictorial content, or merely being a chairishly arranged pattern of stuff? Many non-chair things could look just like chairs, after all. But chairs are recognizable manifestations of bare bones content, even if substitutes sometimes deceive us. What about a wooden chair? In a world where wooden chairs compete evenhandedly with indistinguishable plastic ones, most contexts leave a disjunction: wood or plastic. Some contexts, in which one or the other predominates— they use only the finest plastic chairs—make it reasonable to think that a plastic chair is part of the pictorial content. One further point is particularly important for what follows. As conceived here, pictorial content does not include particular individuals or scenes. It can include a chair, situated thus and so, but not the specific chair photographed. This is true even if the specific chair is part of a recognizable manifestation of bare bones content. Pictorial content is just a rich set of

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

 

31

qualities, and thus a more detailed descriptive content than the bare bones content is. Mindful that the range of features that can figure in pictorial content is open, one might suggest a special feature: having a John’s-chairappearance. Bracketing any worries about whether that’s a kosher quality, it’s possible that it can be a pictorial content. But that’s not the same thing as making the particular chair itself the content. There are three main reasons for restricting pictorial content to a rich pattern of qualities. First, in other work (Kulvicki 2006, 117–18) I suggest that the representation of particular individuals is not central to what makes pictures different from other kinds of representation. Pictures are distinctive because of (a) how syntactic features constrain their bare bones contents and (b) how bare bones content constrains this property-only pictorial content. Focusing on how to get a rich set of qualities out of a bare bones one tells a lot about what makes pictures a distinctive kind of representation. The ways in which pictures depict this or that particular individual do not. Others naturally disagree, but this is not the place to have such battles. If this way of understanding pictorial content is productive, as the rest of the book shows it to be, that will be one more reason for favoring an account of what makes pictures distinctive that limits pictorial content to complex patterns of qualities. Second, pictures can be used referentially, as Chapter 4 shows. Pictorial content can help explain the many referential uses to which pictures are put. In those cases, particular individuals will be contents of pictures, even though the individuals are not part of pictorial content proper. The ways in which particular individuals are represented by pictures mirror some referential uses of descriptions, but this will only become clear in Chapter 4. Chapters 5 and 7 will reveal yet more ways in which particular individuals can be the contents of pictures. So, the point here is not excluding individuals completely, but just suggesting that there is an important kind of content—pictorial content—that is purely attributive. This contrasts with how most theorists of depiction understand things, and there will be occasion to return to this issue in more detail at the end of Chapter 4. Third, it is always possible to interpret pictures in this rich-qualitypattern way, even those that are used to represent individuals. Yes, the picture is a portrait of Nixon, but those who don’t know Nixon can say a lot about what the picture represents, a lot that goes well beyond bare bones content. Nixon is a good fit for the descriptive pictorial content, perhaps the only one. In that sense, the picture might denote a scene involving Nixon, or Nixon himself. But that’s different from saying that Nixon, the individual, is

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

32

, ,  

the content of the depiction. It’s plausible that something fancy like having a Nixonish look could be part of pictorial content. It’s just that Nixon, on the view defended here, cannot. This section started by asking how bare bones content, as a pictorial character, acts as a rule for determining pictorial content. Interpreting pictures is finding appropriate, recognizable manifestations of bare bones contents. In this way, pictures have something in common with demonstrations. Just after his discussion of indexicals, Kaplan suggests that demonstrations, as Frege understood them, are ways of presenting individuals. Specifically, they present individuals as things that “look thusly from here and now” (Kaplan 1989, 525 original italics). In different contexts, of course, different things can look that way. So, demonstrations can occur in different contexts and isolate different individuals in each case, which is why he suggests that they have non-constant characters, like indexicals. On the view of pictures considered here, they work like demonstrations that single out, not individuals, but patterns of features.⁹ Pictures do not have particular individuals as contents, but they do have complex attributive contents, which are constrained by bare bones content, and fleshed out in context as described above. Showing a picture is doing something akin to the following: “It’s like this [mimicking something]!” In different contexts, the audience will latch on to different pictorial contents, sometimes chairs, sometimes pictures of chairs. In all cases, they are looking for appropriate recognizable manifestations of the bare bones content, which is what the picture brings to all such contexts of use.

2.5 (In)definite Description and Reference Pictorial content specifies a set of qualities that a scene might satisfy. Just as character is a function from contexts to contents, a content is a function from circumstances of evaluation to extensions. Understood this way, pictorial reference is indirect. That is, pictures refer to individual scenes only insofar as those scenes have the features the pictures specify. This is the same model of reference one finds with descriptions. Pictures are also different

⁹ Thanks to Josef Stern for this suggestion, and see his discussion of predicate demonstratives in Stern (2000, 85–8, 187–9). Dominic Lopes (1996, 101–7) develops some of Gareth Evans’s (1982) thoughts about demonstrations as a way of explaining pictures’ singular contents, but the point being made here only concerns pictures’ attributive contents.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

()   

33

from descriptions, and a few moments of compare–contrast will bring this out. First, though pictorial content is descriptive, there is no reason to think a linguistic description could readily capture it. The qualities represented by a typical picture might outrun the lexical resources of any given language. Even a lexically mighty language might fail to capture pictorial character or content fully because they are too rich for anything but indefinitely long descriptions (see Kitcher and Varzi 2000, e.g.). Second, linguistic descriptions come in at least two varieties, definite and indefinite, which function quite differently. In some languages, these differences are syntactically marked, as with English’s “a” and “the,” but in others they are not. There’s a lot of controversy in the philosophy of language concerning how we should understand the distinction between these two (see Ludlow 2018 for an overview). Is it, for example, a semantic distinction or are these two kinds of description semantically identical but pragmatically distinct? Pictures do not syntactically mark a distinction between definite and indefinite use, and so, while descriptive, pictorial content is neither definite nor indefinite. In different communicative contexts, pictures can be deployed with the force of either one.¹⁰ One way in which communication can make intended use clear is by bringing the picture into a linguistic act which specifies either an indefinite or a definite reading. “Bring me one of these” (indicating a picture of a widget). In that case, the pictorial content is understood indefinitely. There is no suggestion that only one thing satisfies the descriptive condition, though that might be true. Instead, the goal is to ask for something, anything, that satisfies it. By contrast, consider “Find him” (indicating a photo of a man). In this case, the suggestion is that a unique individual satisfies the content. Cases like this will recur in future chapters. The way in which pictures, and parts of the scenes they depict, are brought into linguistic acts via demonstrations is potentially problematic, but those issues will become clear only later on. Pictures don’t need linguistic contexts to distinguish between their definite and indefinite uses, however. Sales catalogs use pictures to show what products are like, and it’s usually precisely the point that many things satisfy their contents. Catalogues raisonnés, by contrast, carry the strong presumption that unique individuals satisfy the pictures in them. In these cases, the

¹⁰ Novitz (1977, 86) points this out as well.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

34

, ,  

picture is not brought into a linguistic act via demonstrative reference. It’s not language, so much as communicative context, which determines how pictures are to be understood in such cases.

2.6 Worries about Indirect Pictorial Reference Section 2.5 suggests that pictures refer in the indirect manner characteristic of descriptions. It’s also a commonplace that many pictures have impressively detailed contents. Because of that, one might worry whether interpreters can ever know what it is to which a given picture refers, if anything. First, if pictorial content is so rich then interpreters might never fully grasp that content. Second, given the richness of pictorial content, it’s hard to see how pictures could refer to scenes by correctly describing them. It is unlikely that any actual scene will have all of the features a detailed picture specifies. So, it is unlikely that such a picture will denote anything. This is even true for photographs, carefully done, in focus, because some surface features, albeit minor ones, will likely represent things as being different from how they are. If photos produced in that way misrepresent their subjects, then just about any other picture will misrepresent its scene too, or so one might suggest. And third, we often understand pictures as referring to individuals, and not necessarily scenes in which those individuals are embedded. The picture denotes Nixon, one might think, not Nixon, over there, situated in such and such a manner. But if denotation succeeds via accurate description, and the description in question is of a scene, then it is hard to see how the picture denotes Nixon, and not a Nixon-involving scene. These three worries all yield to the same kind of response. For now, a sketch will have to suffice, because the response depends on some material presented in Chapter 3. At the end of Chapter 3, these worries will return, and get a more complete answer. Pictures can denote because interpreters are willing to disregard many details of their contents. Yes, the photo suggests that the shirt is exactly this shade of vermillion, and it’s not, but the specific shade does not matter, so much as the fact that the shirt is vermillion. In different contexts, different levels of detail matter. So, in some contexts, pictorial denotation is possible. Similarly, the picture denotes a scene involving Nixon, but if the communicative point of the picture is to deliver a human being, the picture can certainly be understood as capturing his appearance in sufficient detail to isolate him, irrespective of the scene in which he is embedded.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



35

Similar things cannot be said about descriptions in language. “The vermillion-shirted man” does not denote the crimson-shirted man, despite the fact that both vermillion and crimson are shades of red. Interpreters might forgive someone’s ill-fated attempt to use fancy color words in describing someone, and understand him as picking out the crimson-shirted man. But that would not be to say he succeeds because the content is close enough to correct. The same charity could apply to the person who called the shirt chartreuse. The mechanism proposed here for pictures is not based on charity, so much as on shared, contextually determined standards, and it works because of the special way in which pictures have contents, and syntactic parts. That is why this claim cannot be defended until Chapter 3, in which picture parts are discussed at length. Before moving on, however, notice that these shared standards and how they work fall squarely within the realm of pragmatics. They concern figuring out what people intend to communicate when they use pictures. But they speak to the relevance of syntax and semantics, because what people can do depends in part on what the pictures mean, and what their parts mean.

2.7 Summary Pictorial character specifies a potentially rich but nevertheless fairly abstract set of features that determine a pictorial content in context. Bare bones content has played a role in thinking about pictures, though perhaps not by that name, at least since Gombrich’s Art and Illusion. Interpretive intuitions concerning pictures point strongly to the fact that any picture, given the right context, can be understood to represent a range of things, even though there is something common to those contents across contexts. This fits precisely with the motivations for understanding linguistic expressions as having non-constant characters. Pictorial content is an appropriate recognizable manifestation of bare bones content. What counts as a pictorial content depends in part on concepts interpreters can be relied upon to have, which in turn depends on the world they inhabit. Which things have been the kinds of thing worth making concepts for? All of this affects what counts as a possible pictorial content, and similar thoughts apply to context-sensitive linguistic expressions. Pictorial content is richly descriptive. It specifies a complex set of features arrayed in space. The degree of richness can vary across pictorial practices

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

36

, ,  

from line drawings and coarse mosaic to color photography, but, in all cases, pictures manage to say quite a bit. The specific range of qualities pictures can capture is open to discussion, though it’s plausible that many odd features, like having the appearance of John’s chair, can be included within it. Pictorial contents do not include particular individuals or scenes. This undoubtedly runs contrary to some intuitions about pictures—that photo represents Nixon—but the force of those intuitions needs to be reconsidered in light of the subtleties of the semantic apparatus introduced here. Certainly, a photo can denote Nixon, and its content can include a person with a Nixonish appearance. And in some cases, as Chapter 4 shows, Nixon can be a content of a picture, even if he is not a pictorial content. Thinking of pictorial semantics in this multi-level manner allows one to apply operations defined over character and content to pictures. For example, Kaplan understands some referential uses of descriptions by appeal to what happens if the content of a description becomes a new, constant character. That move will be the focus of Chapter 4. After that, other operations defined over character and content will be explored. Chapter 5 offers another such mechanism that helps explain iconographic interpretation of pictures, and shows that it also applies to some linguistic phenomena as well. And Chapter 6 suggests, following some of Josef Stern’s (2000) work, that metaphorical interpretations of pictures can be modeled in this manner. Though particular individuals do not figure in pictorial content, pictures can still refer to individuals. Like any linguistic expression with a descriptive content, pictures can refer via denotation. That is, a picture can refer to things included in its extension, the things that satisfy the descriptive conditions the picture presents. Because pictures are so richly descriptive, one might worry that they are not useful in their denotative capacity. They will almost always fail to denote anything, and in any case users of pictures will typically fail to understand exactly what the picture’s descriptive conditions are. Chapter 3 addresses these worries in some detail. Finally, this chapter has focused on whole pictures and the ways in which they can be meaningful, but pictures have parts. Moreover, many practices involving pictures suggest that different parts of pictures can be treated differently, semantically speaking. Sometimes a part of a picture identifies a subject while another part of the picture says something about it. Before any of that can be clear, however, picture parts, and why they count as syntactic parts, need some explaining.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

3 Parts of Pictures Pictures have rich descriptive contents. These contents are determined to be what they are by pictorial characters, in context. Syntax focuses on what the parts of complex representations are and how they can be combined to form meaningful wholes. This chapter suggests that pictures, like many linguistic expressions, are syntactically complex. They have meaningful parts, and these parts, and the way they are put together, play a role in determining the meanings of whole pictures. The syntax story will not be complete until Chapter 8, but this chapter provides enough to make the following discussions of pictorial semantics more articulate. In the right contexts, different parts of a single picture can play different semantic roles, thus making pictures semantically much more complex and expressive than one might expect. Pictures, unsurprisingly, don’t have the kinds of complex grammar that linguistic expressions have, but that’s not to say they lack parts that contribute to their meanings. This chapter’s goal is to unpack and explain what the syntactic parts of pictures are, and illustrate how they can be deployed expressively.¹ Informally, pictures can be broken into parts by slashing them to pieces, dissolving them in acid, or burning them to ash and smoke. These violent partitions do not capture the notion of syntactic parts suggested here, but they are relevant because philosophers have claimed that cutting pictures to pieces is one way of generating their syntactic parts. Understanding the details, virtues, and limitations of that proposal will be very important in what follows. The next three sections clear some ground, and set up the main claim, in section 3.4, about the nature of picture parts. Sections 3.5, 3.6, and 3.7 then apply this idea to pictures and set the stage for the chapters that follow.

¹ The main theoretical tools deployed in this chapter are drawn from earlier work of mine, specifically Kulvicki (2007, 2010, 2014 Ch. 8, 2015a). I’ll be more specific about these debts as the chapter moves along. Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

38

  

3.1 Syntax Without (Much) Grammar Some features of pictures are relevant to their representing what they do, while others are not. It was not important, for example, that the pictures in No Chairs were made of different materials, printed in different inks, or different distances from the sun. Those features of pictures are incidental, while the disposition of pale and dark patches on their surfaces is syntactic. They are relevant to the picture representing what it does. Change them, and the result can be a picture with a different content, but changing any of the other features just mentioned will not do this, unless the process also happens to change those surface features. Though this does not yet yield a useful notion of picture parts, it at least gives a sense of the syntactically relevant properties of pictures: which matter, and which do not, when it comes to what a picture represents. Different features are syntactically relevant to different kinds of picture. Black and white photos might be tinted pink here and there, but hue does not affect what they represent and thus is strictly speaking incidental. Line drawings differ from those built around patches of pigment. That’s not to say it is always easy to know which features matter. In many cases, consumers have to make an educated guess. Line drawings tend to stand out as such, just as photographs do, but it is possible to photograph line drawings, and trace line drawings atop photos. A photo of a line drawing has a different set of syntactic features than a line drawing does, so interpreting a picture requires that one know what kind of picture it is. These decisions are akin to what Kaplan called “pre-semantic” judgments, in which one must decide what language is being spoken before being able to interpret it (Kaplan 1978, 296–7). In the pictorial case, these decisions are also presyntactic, since different kinds of picture make use of surface features differently in determining content. We can apply this thinking about syntax to linguistic representations, and ask, in effect, which aspects of letters or words are relevant to them being the letters and words they are. But this is not terribly interesting. It doesn’t reveal much about what makes linguistic representations distinctive. The interesting parts of sentences are the words. Those words are also assembled into important grammatical parts. Some phrases are in the subject position, some are verbs, some are objects of this or that sort. Some, like the words “the” and “a,” might not have a meaning absent their use among other words. These grammatical positions affect sentence interpretation. Does John love Mary or Mary love John?

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

39

Pictures do not have grammar in the manner just sketched. There is no subject, verb, or object position within the picture. Talk about the subject of a picture is not grammatical, so much as a way of singling out the important things depicted. Lack of grammar doesn’t mean there is nothing worth saying about pictorial syntax, but one should expect the parts to be fairly uniform. They will not be grammatically marked as playing different roles in constituting what the picture as a whole means. To the extent that pictures’ syntactic parts play different roles in constituting the meaning of a picture, it will not be because they come grammatically packaged as doing so. Instead, context and conventional practices can make some parts salient as the things that play one role, rather than another. That point will become clear in subsequent chapters, but for now the goal is understanding just what pictures’ syntactic parts are.

3.2 Abstraction and Content Section 3.1 distinguishes the syntactic features of pictures from their representationally incidental features like mass, distance from the sun, and the like. There is no distinctive pictorial grammar to speak of, so one should expect the syntactic parts of pictures to be uniform in that they can all contribute in the same range of ways to what a picture as a whole represents. So, how do those syntactic features break up into parts that are relevant to what the whole represents? The basic story is simple, though it will take some time to unpack it in detail. Specifically, because syntactic features play their own, systematic roles in determining what a picture represents, instantiations of those features constitute the syntactic parts of pictures. There is a systematic connection between the syntactic features of pictures and their bare bones contents (Kulvicki 2006). By extension, there is an interesting, albeit more attenuated, relationship between pictures’ syntactic features and their pictorial contents. The motivation for calling something a syntactic part is that it is meaningful and plays a role in determining the meaning of a complex expression. And for these reasons, instantiations of those features are pictures’ syntactic parts. Now for the unpacking. First, notice that different kinds of syntactic features seem responsible for representing different properties of scenes. Colors of the picture represent chromatic features of their subjects, for example. Black and white photos just represent patterns of light and dark, and they themselves tend to be monochrome. Spatial features of pictures represent spatial features of scenes. This

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

40

  

division of labor makes a lot of sense, and is deeply related to what makes pictures a distinctive kind of representation.² Different syntactic features play different semantic roles. These features play the semantic roles they do in a manner mediated by pictorial character. Character determines a pictorial content in context. The most straightforward connection between syntactic features of the picture and features that figure in its content will be found with respect to the bare bones content. So, the picture is dark green, but represents a bright green surface, in shadow. The pictorial content includes being bright green, and in shadow. The bare bones content, by contrast, includes only a more abstract feature, which is compatible with many combinations of surface color and illumination: bright green in shadow, dark green in bright light, and so on.³ Similarly, a region of a picture is trapezoidal, even though its pictorial content includes a rectangular table. The picture manifests a trapezoidal shape, but also an abstract shape property that is shared by trapezoids, seen head-on, and rectangles, seen obliquely. The aspect of the picture responsible for it representing what it does is the more abstract one, which the table and picture can share. All of the details about how this works are open for argument and further articulation, but for now the point is just that there is a fairly systematic division of labor among pictures’ syntactically relevant features. The second important point is that abstractions over syntactically relevant features map readily onto abstractions over what they represent. Let’s say reds on a picture surface represent reddish chromatic features of scenes, and greens do a similar job for greenish features. If all one knows is that a syntactic feature is red-or-green, then one knows that the picture represents either a reddish or a greenish chromatic feature of the scene. Imagine also that a specific shade of vermillion represents a specific shade of red. If one knows that vermillions are shades of red, then one also knows that the vermillion picture represents a feature that is a shade of red. This point should not strike anyone as controversial. In fact, same can be said for linguistic representations. If one is confident that some word is “red” or “green,” then one knows that it represents red or it represents green. This ² In fact, I’ve argued (Kulvicki 2006, Ch. 3) that pictures are instances of their bare bones contents. That is, pictures manifest all of the features included in their bare bones contents. That’s quite different from saying that pictures instantiate all of the features they represent pictorially, of course. ³ I discuss such chromatically perspectival features, in relation to their spatial counterparts, in Kulvicki (2020).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

41

uncontroversial point plays an important role in the more controversial claim that follows. When a picture surface is red, it is eo ipso red-or-green. Does the picture’s being red-or-green represent a red-or-green feature of a scene? Similarly, any vermillion picture is red, so does any vermillion picture represent its scene as being red? The uncontroversial point made in the previous paragraph does not commit to an answer here. Just because all vermillion things are red doesn’t mean that a representation of something as vermillion is also a representation of it as red. In fact, in the linguistic case, it very tempting to deny these claims. One can have the concept of vermillion, after all, without having a concept of red, so it’s hard to see why one should build redness into the representational content of a claim about vermillion things. The third, and controversial, point is that in the pictorial case, things work exactly that way (Kulvicki 2007, 2010). A picture that represents something as vermillion does so, let’s say, in virtue of being a specific shade of red.⁴ But such a picture is a member of a class of pictures that can represent a large range of colors: blues, greens, yellows, and so on. Across the range of such pictures, many different shades of red are responsible for representing the range of red things. And because of this, the fact that a picture is red, rather than this or that shade of red, means that the picture represents something red. Similar remarks apply to spatial qualities of depicted scenes. A region of a picture is trapezoidal, and it depicts a rectangular table, from an oblique angle. That trapezoidal region of the picture is also quadrilateral, and the picture thus also depicts a quadrilateral shape. Why think this is so? Perhaps pictorial content is quite specific, and anything less specific is not so much included within the content, but an inference away from it. It seems like things work this way in language. The concept of vermillion does not represent red, even though anyone with a basic understanding of the concept should readily infer from the fact that something is vermillion that it is red. In pictures, things work differently, and a few points should suffice to show this. First, ordinary practice with pictures suggests that they represent features across levels of abstraction. They can be aids for finding the orange things in

⁴ Remember that the main point here applies most straightforwardly to bare bones content, and only by extension to pictorial content. Because it’s easier to talk about pictorial content, I’ll do that here, and get more careful when it’s important to do so.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

42

  

a depicted scene just as readily as for finding specific shades of orange. The reason for this is that the picture helps (Kulvicki 2010). It is both orange and a specific shade of orange, and both of those features are relevant to what the picture represents. Not all words for shades of orange wear such semantic links on their sleeves, but all the syntactically relevant features that represent shades of orange do. This point will return in the context of objections considered in section 3.3. Second, more abstract color features, like being green rather than chartreuse, work in the same way to determine pictorial content as the more specific shades do. Red pictures, per se, represent red things, and vermillion pictures represent vermillion things. So, red and vermillion pictures represent red and vermillion things. Similar remarks apply to spatial features that scenes are represented as having. Third, it’s unlikely that pictures would be interpretable at all absent such a mechanism being in place. Many pictures, though not all of them, have contents that depend in an indefinitely subtle manner on their syntactic features. And many more pictures have contents that depend on their syntactic features in a manner too subtle for ordinary consumers to appreciate. Observers simply don’t have access to such subtleties, even under conditions one might think of as ideal for viewing the picture. But if observers cannot determine what the syntactic features of pictures are, how on earth do they manage to interpret them? Put differently, how could someone interpret a representation without knowing which representation it is? If abstractions over syntactic features—I’m not sure which shade it is, but it’s definitely red—map readily onto abstractions over the most determinate contents, then interpretation can be accurate, and blameless, even though it’s not capturing all of the details (Kulvicki 2015a). Remember that the points about abstraction and content apply, first and foremost, to bare bones contents. That is, pictures have syntactic features responsible for their having the bare bones contents that they have. Some abstract chromatic features, spatial features, and so on, specify a bare bones content, which in context yields a fleshed out pictorial content. Abstractions over the most determinate features responsible for this bare bones content map onto abstractions over bare bones contents, which, in turn, yield differing fleshed out, or pictorial, contents. This is important because without restricting attention to bare bones content, the claim about abstractions seems implausible. Pictorial content might include a feature like being a turkey. But it’s implausible that any picture of a red thing also represents that thing as being red, or a turkey. It is plausible, by contrast, that the

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

43

picture represents the thing as being red or warm-colored, as well as representing it as being red, warm-colored, vermillion, and so on. This feature of pictorial content is not limited to pictures. Diagrams, graphs, maps, and all sorts of non-pictorial images work this way as well, and I have suggested that this is the true mark of analog representation (Kulvicki 2015a). Most of the non-pictorial examples discussed in Chapter 2 have constant characters, so in those cases there is no bare bones content of interest. In many ways, those representations are easier to understand than pictures. Their non-constant characters are what distinguish pictures from all other non-pictorial, imagistic representations.

3.3 Two Clarifying Objections Wait a second, anything that is red is eo ipso red-or-Martian. So, any picture that is red is also red-or-Martian. Does any picture of a red thing thus depict it as being red-or-Martian? That’s madness, but it’s not obvious how to avoid the conclusion, given what has been said so far. The way forward is to look to the most specific things a picture in some system can say as a way of limiting the most abstract things it can say. Imagine a given kind of picture can represent a range of specific color properties. Any abstract feature that can be built up by disjoining those properties is also part of the pictorial content. So, red-or-green works, as does red in the vermillion case, given that the system of depiction can represent the full range of reds. But red-or-Martian is not a possible content because the system of depiction cannot represent being Martian. The idea is that not any abstraction over a picture’s syntactic features yields a ready abstraction over its content. The limitation here is built into the most specific syntactic features the picture might possess. Abstractions constructed from those are also represented, while those not constructible in such a fashion are not. Okay, but if one must look at the system as a whole before determining what abstractions are legitimate, it seems as though languages work in exactly the same manner. Consider all of the names for colors in English. They constitute the lexical resources devoted to representing colors, and now imagine abstractions over those words, written or otherwise. “Blue,” “green,” and “puce” are all representations of colors. Any word that is “blue” is eo ipso “blue”-or-“green.” Ditto for “green.” So, being “blue” or “green,” in English, represents being blue-or-green, and so on for any abstractions that

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

44

  

can be built up from the most specific terms for color in the language. If languages work this way, then the feature seems not to distinguish pictures from other kinds of representation. But, more importantly, it doesn’t seem as though just making the abstraction yields a representation of a new, abstract content in language, so perhaps it doesn’t do that in the pictorial case, either. Yes, it is possible to think of languages in this manner. But no, doing so is not helpful for understanding how they are interpreted. By contrast, this fact is central to pictorial understanding, what makes them a distinctively perceptual kind of representation, and what makes them useful in ways linguistic representations are not. Trivially, with a range of representations one can generate abstractions over their contents by considering abstractions over the representations themselves. But learning a language amounts to learning meanings of terms. These terms are the meaningful units of more complex expressions, and abstractions over those lexical expressions typically play no significant role in linguistic interpretation.⁵ Language provides others tools for generating abstractions over the contents of representation, like disjunction. It’s easy to represent something as being red-or-green in language: “Monty is red or green.” Unlike language, in the pictorial cases the key to interpretation is found in making abstractions over syntactic features of representations. As suggested earlier, without leaning on abstractions over the most specific syntactically significant features of pictures, they might not be interpretable at all. But how do these abstractions help in the pictorial case, if they do not in the linguistic case? Pictures’ syntactically relevant perceptible features are such that perceptually salient abstractions over them are the ones that are important to interpreting them. Many shades of blue are visible, as shades of blue. Ditto quadrilaterals. Perceptual access to things’ properties extends across levels of abstraction (Kulvicki 2007, Green 2017). Pictures are readily interpretable because those perceptually salient abstractions map readily onto abstractions over content. Some linguistic practices exploit this feature of pictures, and when seen in this light they seem like the most pictorial parts of language. Using dashes or lines to count, for example, is somewhere between a linguistic and a pictorial practice, and all of the points about abstraction made above apply to it. ⁵ Fully unpacking this point will help us distinguish representations with propositional content from those that have non-propositional content. So, if this is compelling, or the opposite, head right to Chapter 8 after this one.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

45

Arabic numerals are more complex than dashes, but they are built to allow ready abstractions over their features that map onto abstractions over their contents. Focusing on the ten’s place in a list of numerals, or the 100’s place, provides a sense for how values vary within that range. If all that matters is whether a value falls between 100 and 200, it’s easy to find this out by ignoring the ten’s and one’s places.

3.4 The Parts Principle The points about abstraction and content are essential to understanding what the syntactic parts of pictures are. It will also help to contrast the proposal made here with a related one built around cutting pictures to pieces. Elliot Sober (1976) suggested that one interesting thing about pictures is that parts of pictures are themselves pictures. He imagines cutting up a picture into smaller and smaller pieces, and notices that “successive snippings never destroy representationality.” (Sober 1976, 124). This is one of the earliest, explicit suggestions to this effect, but the idea has been implicit in talk about pictures in the West going back at least to Alberti’s Della pittura (1435/1991). This idea was also important to participants in the imagery debates from the late 1970s through the 1990s. This book is not focused on mental images, what they are, or whether they exist, but those who were focused on that topic needed to know what might make something in the mind or brain an image. One thing that suggested itself is the way in which spatial parts of an image are images themselves.⁶ Not only are spatial parts of pictures themselves pictures, they seem to be pictures of spatial parts of whatever the uncut representation depicts. Jerry Fodor proposed a “picture principle” that almost makes these points explicit: “If P is a picture of X, then parts of P are pictures of parts of X.” Fodor (2007, 2008, 173) Just specify that “part” means spatial part, and the result is the Sober point just made. It’s clear from his text that this is what Fodor has in mind. At first glance, the Sober proposal is an excellent way to think about pictures’ syntactic parts. He identifies parts of a whole that are representations, and whose meanings contribute to the meaning of a whole picture. ⁶ See Kosslyn (1980, 33, 1994, 5), Tye (1991, 44), and also Kulvicki (2014, Ch. 8), where the ideas presented here are developed in some detail.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

46

  

Moreover, these parts can be recombined in different ways to give different pictures, even though the results of haphazard recombination are often only semi-intelligible. These parts thus can serve as the inputs to an overall interpretation of a picture, and the way that they are combined seems like a syntactic operation. The semantic identity of the whole depends on the semantic identities of the parts and their method of combination.⁷ Elisabeth Camp goes so far as to say that “insofar as we can discern syntactic ‘parts’ of a picture at all, these are either just points in a two-dimensional array, or else regions whose boundaries are given by salient boundaries in the scene represented” (Camp 2007, 156). This proposal captures the thought from section 3.1 that the syntactic parts of pictures, if any, will look quite uniform, in the sense that there are no grammatical markers to single out this or that part as playing this or that role in the whole. All of these spatial parts represent spatial parts of the scene, and the way in which they might combine to form a whole depends on nothing more than how they are put next to each other. This proposal also reveals an interesting feature of picture parts: there is no privileged way of forming them. Pictures can be cut up in many ways. All such divisions yield bona-fide syntactic parts. So, not only do pictures not come grammatically prepackaged, with set roles for the different parts, they do not come pre-divided. There are indefinitely many, equally good, choices for how to break up a picture into parts without remainder. The account offered here agrees that certain spatial parts of pictures are syntactic parts, but suggests that they are so for a specific reason. And that reason shows that pictures have many other syntactic parts too. What makes spatial parts syntactic parts is that they only represent a limited set of features compared to the whole representation. Instead of the whole picture, which represents features across a given range of space, a spatial part just represents things happening in a subspace of the whole. And it does so because its syntactic features are a subset of all the picture’s spatially relevant syntactic features. So, the core notion behind syntactic parts of pictures is that they represent part of the whole, and they can be found by focusing on

⁷ See Blumson (2014, Ch. 6). In one sense, Blumson seems to suggest that the parts of pictures are spatial parts. In another sense, he suggests that parts might be non-spatial. He compares pictures to chess diagrams and maps, on which one can place markers here and there. The markers are parts, but they are not spatial parts, per se. See Chapter 7 for a discussion of maps. Blumson never comes out and says exactly what a picture part is. Abell (2005, §6) suggests that the parts of pictures are their syntactic features, which is more in line with what I have in mind here.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

47

some syntactic features to the exclusion of others. To put it a little differently: one finds syntactic parts by abstracting over syntactic detail. Doing so yields an interpretable representation, even if it is just a representation that represents part of what the whole does. In that sense, instantiations of syntactic features are good candidates for being syntactic parts. Once this point is clear, many non-spatial syntactic parts of pictures reveal themselves. For example, one can focus exclusively on high-spatial-frequency features of a picture or low-frequency features, depending on one’s interests. One can ignore the hue of a color photograph to focus more on what one would ordinarily find in a black and white photo. One can ignore color details in favor of focusing on where the scene is depicted as warm, or cool. These are all ways of breaking the picture into parts that matter interpretatively and those that do not. Cutting up a picture is just a vivid example of this, which has a simple physical correlate: you slice the picture into pieces that can be handled independently of one another. But being a syntactic part of a picture does not require being separable in that physical fashion. It is impossible to cut a picture into its high-frequency and low-frequency parts, even though in the sense just outlined those are syntactic parts of the whole. The Parts Principle, then, just says that abstractions over the syntactic features of a representation, R, are representations of abstractions over R’s content (Kulvicki 2015a). Cutting in Sober’s sense is a special instance of this principle, because cutting up a picture amounts to focusing on the parts of it that represent only some spatial parts of the whole scene. The Parts Principle is superior to the previous claims because it focuses on abstractions over syntactic features of all sorts. They need not be spatial, so one need not understand the chief syntactic operation as cutting something to bits. Pictures have high-spatial-frequency parts and low-spatial-frequency parts, they have hue parts and brightness parts, but those parts completely overlap one another spatially. The appeal of Sober’s claim is that it identifies parts in a manner that’s perhaps more familiar from other cases, as well as from the philosophy of language. Parts, for Sober, are the kinds of things that have lives of their own. They can be assembled with other parts in indefinitely many ways. This sounds a lot like the lexemes of a language. By contrast, the parts identified by appeal to abstraction are not necessarily separable from one another. It’s important to liberate the notion of a syntactic part from the more ordinary notion of a separable part, but the details will not be important until Chapter 8.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

48

  

Let’s say this point about abstraction is unappealing. For much of what follows, this won’t be a disaster. I’ve defended the parts principle and its usefulness elsewhere (Kulvicki 2015a), but the more restricted claim from Sober (1976) and those who followed him, like Fodor (2007), will serve many of the purposes served by the Parts Principle. The goal is to convince readers that the more expansive notion of a pictorial part is more useful than the Sober version, and examples from the following chapters work toward that end.

3.5 Syntactic Parts and Semantic Roles Chapter 2 claims that pictorial content is descriptive, though not marked as definite or indefinite. As such, pictures are not themselves expressive of propositions for the same reason that the average (in)definite description is not. Properly embedded in a larger communicative act, however, perhaps one involving a linguistic utterance, the picture can be part of the expression of a proposition. While that point seems plausible enough, one is likely left with the intuition that sometimes pictures themselves are treated as expressing propositions. But if pictures are not embedded within linguistic contexts, and, by themselves, are just rich descriptions, how could a picture be the kind of thing that expresses a proposition? The short answer: pictures have parts, and in the right contexts those parts can be deployed to play different semantic roles. In some cases, for example, one part of a picture plays the role of identifying a subject, while other parts predicate things of that subject. In that case, the picture is expressive of a proposition. Consider the joyful ritual of viewing friends’ vacation photos. There they are at the Taj Mahal! There they are at India Gate! There they are in Connaught Place! Tedious, but revealing. The family is represented by each of these photos. In fact, a part of each of these photos is devoted to representing the family. The pictorial content specifies a set of features that the members of the family, and presumably only the members of the family, satisfy. The rest of the picture represents other things, like patterns of features satisfied by the Taj. These different picture parts can play different semantic roles. When the context is the presentation of vacation photos, the part representing family members plays a subject role, while the other parts of the picture play the role of predicates. In such contexts, the pictures say something like “the people of this sort were at a/the place like that.” This is expressed in terms of definite and

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

    

49

indefinite descriptions, even though the pictures themselves have no way of marking such things. They have no grammar for identifying one part as a subject, another an object, or predicate. All they provide is an articulate, and complex descriptive content. Because the complex content is articulate, in the sense that there are many parts that contribute to the complex whole, different parts can play different roles. They do play different roles when the pictures are used within a highly conventionalized practice, such as the display of vacation photos. Find the family, and notice what the rest of the picture says about it. In one sense, the picture doesn’t do this all by itself. But then again linguistic expressions are not, all by themselves, responsible for what is expressed linguistically. Some languages do not mark a difference between definite and indefinite descriptions, for example. And all languages are such that linguistic content can depart significantly from what is expressed in context. It might be tempting to say that these contexts force pictures into something like a linguistic mold. But why suppose that predication, description, and the like are uniquely linguistic? Part of the point of modeling meanings is explaining the range of communicative uses to which representations can be put. Understanding pictures as having articulate, rich descriptive contents does just that. Parts of these contents can serve different roles in communication. Now consider what happens when the context changes. The local tourism office collects photos of people posing at nearby monuments. In their hands, the parts of the picture that isolate subject and predicate are reversed. They are not concerned with this or that family, in particular, but with the place, and those who visit it. Is it mostly families? Singles? Tourists? Or locals? In such situations, these photos say of the place like this, that it has visitors like that. With vacation photos, it is tempting to think that the relevant subject parts can be understood on the Sober model. Cut them out, and we have pictures of family members, which then occupy the places in question. But that does not quite do justice to the phenomenon. After all, the family occupies the space, and cutting them out doesn’t preserve those features of how they occupy it. Similarly, the tourist board photos can’t quite cut out their subject parts. After all, the place in question is represented in different ways by every part of the picture including the ones occupied by representations of the families. Indeed, this thought should make it clear that just cutting out the people from the vacation photos won’t suffice to identify them as parts, since it’s unclear what remains of the predicative part. The

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

50

  

family members, for example, are represented as standing in full sun, but the illumination is a feature of the scene as a whole. Product catalogs are keen to show what, for example, clothing looks like when worn by different people in different contexts. Those photos take the clothing as a subject, and predicate much of it. Advertisements for paint show it in situ. They treat the paint, in fact usually its color, as the subject, and show what it is like in situations like these. Here too it’s hard to think about the relevant part of the picture in terms of something one could cut out of the whole. The parts are the features responsible for representing the colors on the walls, or the features of the clothing. The other parts play a role predicating things of them. This theme will reappear. Because the syntactic parts of pictures are uniform, they all have the same range of possible uses. So, if a picture as a whole can be used referentially (Chapter 4), so can a part of a picture, to the exclusion of the rest. Iconographic (Chapter 5) or metaphorical (Chapter 6) interpretation can apply to all, or just some, of a picture. But which parts should be used this way? That is up to contexts, in a one-off manner, or highly conventionalized practices. They not only make it clear which parts are treated in which way, but which parts play the role of subjects and objects. They can, that is, allow pictures to go well beyond their simple descriptive contents and be expressive of propositions, commands, and so on.

3.6 Revisiting Indirect Pictorial Reference Chapter 2 ends by considering worries about indirect pictorial reference, or pictorial reference via denotation. One worry was that pictorial contents are so rich and detailed that they might always fail to denote because any scene will depart in some way from such detailed content. The other was that pictures are often used to refer to individuals that happen to make up scenes and not the scenes themselves. The response to these worries was that denotation might work in both cases if one were to ignore some of that descriptive detail. But it was unclear how that might happen, since, for example, there is no analog of this in language. The discussion of parts hopefully clarifies how one might ignore the most specific things a picture says. The most specific things a picture says are, after all, but a part of what the whole represents. That’s so because pictures also represent things at many less detailed levels of abstraction.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



51

Pictures can denote, even if the most determinate things they say fail to match the scene. This is so because pictures can be embedded within a practice that happily disregards those most determinate details. Sometimes, the natural thing to do is to disregard pictorial detail. A blurry photo is not usually interpreted as representing a blurry scene, for example. Understanding the photo despite its blurriness is, in part, being willing to ignore some of its most specific details in favor of focusing on its less determinate features. Such a picture depicts a person with a reddish shirt, from the front in some kind of light, but it doesn’t go into further detail. Precisely the same kind of interpretive move is available when interpreting photographs in perfect focus. Just like the blurry photos, they can be read as going into lots of detail about a scene, but they needn’t be. Likewise, the photo on the outside of the box accurately describes the thing inside, even though it does not capture a scene in which the thing is involved. Aspects of the picture that are specific to the scene, aside from the object involved in it, are irrelevant to the communicative task to which such a picture was put. In other cases, of course, lots of details do matter and pictures can be used to denote scenes, as they might on postcards, but see Chapter 4 for a discussion of postcard pictures. Chapter 7 will explore another way in which pictures might refer to a scene even though they do not accurately describe it.

3.7 Summary Pictures have syntactic parts, namely instantiations of syntactically relevant features. Those parts are found by abstracting over the most specific syntactic features a picture manifests. Those parts are also representations, just like the syntactic parts of linguistic expressions are. They are different from linguistic representations, however, because these parts are not grammatically marked as contributing in this or that manner to the whole. All the parts of a picture are, in that sense, the same. This explains the appeal of earlier discussions that suggest parts of pictures are just themselves pictures, of parts of a scene. The fact that all picture parts are grammatically the same does not mean that all parts must play the same role in communicative acts involving those pictures. Instead, it means that all parts have the same range of possible contributions to the whole. Within suitably conventionalized practices, certain parts can play the role of singling out a subject for comment, while in others those same parts might constitute the comment on a subject

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

52

  

singled out some other way. Because there are no grammatical cues, practices must be highly conventionalized for any such separation of roles to take place. Alternatively, as when this happens in a one-off fashion, a fairly elaborate setup of the context is required. In such cases, pictures can even succeed in expressing propositions. One part identifies a subject, while the other constitutes a comment on it. Yes, these parts fail to do such work absent the right communicative context, but this is not to say that pictures only manage to express propositions when combined suitably with linguistic representations. Within the practice of showing vacation photos, for example, it’s the pictures themselves that say the family is in such and such a situation. The foregoing only scratches the surface by considering a couple of cases in which parts of pictures play different semantic roles. The following chapters consider different ways in which pictures, understood primarily as unspecified descriptions, might acquire contents. The operations in question will all be defined over character and content, and all of the operations defined in what follows can apply differently to different parts of pictures, once a suitably rich practice is developed.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

4 Pictorial Dthat Police sketches, postcards, and portraits are pictures. They have characters and rich pictorial contents. Under some circumstances, it’s tempting to say that these representations have particular individuals as contents. What is a portrait, one might think, but a representation of this or that person? A postcard picture shows a specific locale, which is more than offering a description that some locale happens to satisfy. Examples like this might make one worry that Chapter 2 is wrong to insist that pictorial content never includes particular individuals. This chapter develops tools for understanding such uses of pictures by stealing some of Kaplan’s thoughts about how this works for descriptions. Chapters 5 and 7 will consider yet other ways individuals can be the contents of pictures. To borrow Keith Donnellan’s (1966) terminology, pictures can be used attributively or referentially. This distinction has generated a lot of literature in the philosophy of language. Kripke (1977) famously suggests that referential use is wholly pragmatic. Kaplan (1978, 1989), by contrast, modeled some referential use semantically, in terms of the mechanics of character and content. This chapter suggests that Kaplan’s model applies neatly to some uses of pictures. After unpacking the distinction and Kaplan’s account of it, the chapter shows that it is important for understanding some communicative uses of pictures.

4.1 Attributive and Referential Use Donnellan (1966) introduced the referential–attributive distinction to account for the fact that one and the same definite description can be used in two quite different ways in different communicative contexts. Sometimes, one is trying to say something about whoever or whatever satisfies the description. In other contexts, satisfying the description is incidental to what one intends to communicate. The description, in the latter cases,

Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

54

 

manages to refer to someone, and one intends to communicate something about that individual. Imagine someone saying, “The winner of the race is insane!” (cf. Donnellan 1966, 233).¹ When the race is a particularly grueling affair, the point of the sentence might be to suggest that whoever could triumph under those conditions must somehow be crazy. In this attributive use of the description, one intends to say something about “whoever or whatever is the so-and-so” (Donnellan 1966, 233). By contrast, imagine seeing a goldmedal-wearing individual at the race after-party doing insane things. In that case, one might also say “The winner of the race is insane.” But this claim is directed at the individual spied wearing the gold medal. The fact that the person won the race is incidental to the claim about insanity, and the description is just being used to establish reference to the individual in question. In this referential use, the speaker “uses the description to enable his audience to pick out whom or what he is talking about and states something about that person or thing” (Donnellan 1966, 233). For Donnellan, referential uses can succeed in picking someone out even if that individual does not satisfy the description, and thus fails to be denoted by it. For example, imagine that the person wearing the gold medal is the partner of the race winner. In that case, one can succeed in referring to the person acting crazily via the description “the winner of the race” even though the real winner is the sober, slightly mortified person off to the side. Or perhaps the race was cancelled because of weather and the organizers are just wearing the medals at the party. In that case, “the winner of the race” fails to denote anyone, because no one satisfies the condition it lays out, but the referential use still manages to single out someone for comment. Attributive uses of the description provide no such wiggle room, because they are deployed when one wants to say something about the individual satisfying those descriptive conditions, as such. In the first example, if no one actually completes the race the description fails to isolate someone for comment. Importantly, Donnellan is not claiming that a description used referentially always fails to isolate its object via ordinary descriptive means, viz. denotation. Highlighting referential uses that fail in this regard only serves to illustrate how different each use is. Kaplan tries to accommodate some cases of referential use by appealing to the distinction between character

¹ Page numbers here are to the version of the paper anthologized in Martinich (1996).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

55

and content. His story does not handle all referential use as Donnellan understood it, but it can accommodate many uses of pictures. Specifically, Kaplan suggests that descriptions can be used demonstratively, and that this involves the ordinary descriptive content becoming a character that fixes a new content. It will be easier to understand Kaplan’s story, and how it applies to pictures, once some examples have been presented.

4.2 Using Pictures Referentially Imagine that Figure 4.1 is a police sketch, generated from witness descriptions of a suspect. Let’s assume that it’s a good sketch. That is, assume that the witness did a great job of remembering the person’s appearance and that the artist did a great job of capturing that appearance with the sketch. This picture has a character, as articulated in Chapter 2, and a pictorial content, which involves a detailed set of qualities including a human form with such-and-such features. When such sketches are good in the way just described, their pictorial contents, in the context in which they were made, are typically satisfied by a unique individual, so in that sense they denote that individual. Such sketches can be used attributively or referentially. Imagine a police press conference at which the officers hold up the picture and say “This guy took candy from a baby.” The sketch is being used to

Figure 4.1 Sketch. Drawn by Soo Sunny Park.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

56

 

communicate something here. It provides a set of features that identify someone who allegedly stole some candy. In this case, pictorial content is essential to understanding the officers’ claim. Yes, the perp satisfies the pictorial content, but the point of the utterance was to indicate something about what that guy looks like. They are saying something about whoever, in the relevant context, satisfies that pictorial content. If, for example, the perp is the infamous Harry Lollipops, they could say “Harry took candy from a baby,” but this would be unhelpful even though it’s true because it would not inform the crowd about his appearance. So, in this attributive use of the picture “the speaker wishes to assert something about whatever or whoever fits” the pictorial content (Donnellan 1966, 233). Now imagine that a skeptical reporter asks why the police are putting so much effort into catching candy thieves, given how many other problems— armed robbery, assault, murder—trouble their city. An incredulous officer holds up the picture again and says “This guy took candy from a baby!” Here the picture is used referentially, not attributively. The point is not to communicate anything about what he looks like, but rather to single him out as someone who stole candy, from a baby. The officer could just as well have said “Harry stole candy from a baby,” and communicated the same thought. In this case, then, the picture contributes an individual to the thought expressed by the officer. Notice that in the first case, if the picture is not good then the officers fail to do what they hoped to do, namely, tell the audience what the perp looks like. They say, in effect, that the thief is thusand-so, but he is not. In both the attributive and referential cases as originally described, the picture refers to an individual, but in the latter its contribution to the expressed thought is meant to be the individual, while in the former the descriptive pictorial content is important. Notice that this is not a case in which a picture, like a description, is used referentially even though it fails to denote the object referred to. Those cases are not the focus for present purposes.

4.3 Dthat Kaplan (1978, 1989) suggested a way of understanding part of Donnellan’s referential–attributive distinction in terms of character and content. Typically, descriptive contents are non-constant functions to extensions, as “the yellow book on the top shelf” made clear in Chapter 2. Consider:

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



57

The yellow book on the top shelf is written by the author of In an Antique Land.

If Amitav Ghosh’s The Circle of Reason is the yellow book on the top shelf, then this claim is true, but it’s false if the book is Shelley’s collected poems. These non-constant functions to extensions are typical of attributive uses of descriptions. The important thing is the set of qualities used to isolate something or other. When used referentially, Kaplan suggests, the description delivers an object as the content, or, to put it differently, its content is a constant function to an extension. This is a demonstrative use of a definite description, which Kaplan indicates as follows: Dthat [the yellow book on the top shelf] is written by the author of In an Antique Land.

Here, the description is being used in a manner akin to the way a demonstration might be. Its role is singling out an individual, but the individual, and not the features that single it out, constitutes the representation’s content. This process can be modeled in terms of an operation involving character and content. Such referential use treats the descriptive content as a non-constant character, or a new function from contexts to content. In the context of utterance, that non-constant character singles out an object. The content of the dthat-description is just the object singled out, and thus its content is a constant function to an extension. The description is thus used to single out an object in the way a demonstrative use of “that” might do—hence the operator “dthat,” though in these cases it is the description, and not an associated demonstration, that does the singling out. The previous sentence expresses just what the following claim does: That [pointing at the yellow book on the top shelf] is written by the author of In an Antique Land.

In this case, a demonstrative “that” is completed by a non-linguistic demonstration—pointing, some other gesture, and so on—and that complex act delivers a particular individual. In the pictorial case, the picture and its descriptive content are used as something akin to a demonstration to complete a demonstrative “dthat” as well. The picture is non-linguistic, and used as a demonstration to single out an individual.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

58

 

Understood attributively, the claim that the yellow book on the top shelf is written by the author of In an Antique Land will vary in truth value across circumstances in which it is evaluated, because the content of the description is a non-constant function to extensions. False when it’s Shelley, true for Ghosh, and so on. But used referentially, the context of utterance fixes the object that figures in content. As a result, the claim will have the same truth value in all circumstances in which it is evaluated. In the context of utterance, the truth values of both the attributive and referential uses must coincide. (Kaplan 1978, 301–2 in Martinich 1996). As section 4.4 shows, some expressions involving pictures keep their truth values across circumstances in which they are evaluated in a manner that strongly suggests that pictures are being used referentially along the lines of Kaplan’s dthat model.

4.4 Referential Use as Dthat Following Kaplan’s dthat model, let’s say that some referential uses of pictures treat their pictorial contents as non-constant characters. That is, the pictorial content becomes a new rule, one which takes the rich description, finds what it picks out in context, and delivers that individual as the picture’s content. Remember that ordinary pictorial character, as described in Chapter 2, is also partly descriptive, but it works differently. A bare bones attributive content is fleshed out to an appropriate recognizable manifestation, which is also descriptive. The particular individual denoted by the pictorial content in context becomes the content of the picture when used referentially. Table 4.1 helps keep track of this. As in the linguistic case, the context of production is akin to the context of utterance, and can be distinguished from the circumstances in which the expression is evaluated. Pictures, on this view, are like descriptions in that Table 4.1 Attributive and referential use Character

Content

Attributive use

Bare bones content

Referential use

Pictorial content, understood in a definite way

Pictorial content: an appropriate, recognizable manifestation of bare bones content Dthat content:the thing referred to by the pictorial content, in context

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

59

they are representation types, instances of which can be deployed in any number of contexts. One can thus ask, of a picture that was used in a particular context, what its interpretation should be in a number of circumstances. “The man in the red hat” gets one interpretation in the context in which it was produced, and perhaps others in other circumstances. Used referentially, the picture has a different content than it does when used attributively, but it is unhelpful to call that a new pictorial content, because the latter expression is a term of art in this book. Henceforth, the content of a picture in these kinds of referential use is its dthat content. The following argues that in many cases pictures have dthat contents. Imagine that two sketches like the one in Figure 4.1 were produced independently, one, say, in Georgia (GA) and the other in New Jersey (NJ). Similar crimes have been committed. The NJ police suspect a shadowy figure, alias Vinnie Candies, while the GA police suspect Harry Lollipops. Both sketches can be good in the manner described above. In addition, they have the same pictorial contents. That is, the context makes it clear that the sketches are being used to pick out people, not pictures of people. These identical sketches, in their original contexts of production, can have different referents. The GA sketch refers to the GA perp, while the NJ sketch refers to the NJ perp. And for all the audience knows these two people are distinct. There could be twins or just similar-looking people committing these crimes, and each of these sketches refers to the one in the vicinity of where it was produced.² What happens if these pictures are used referentially? On the Kaplanian model, each sketch’s pictorial content becomes a character, which determines a dthat content, potentially distinct from the other picture’s. So, the NJ perp becomes the content of the NJ sketch and the GA perp is the content of the GA sketch. Each sketch, even though it is descriptively indistinguishable from the other one, has a different content. This makes mechanical sense, but does this phenomenon show up in practice?

² On this model, the pictures are working as incomplete definite descriptions, much as most definite descriptions are. “The man in the red hat” refers when there’s only one red-hatted guy in the room, even if there are plenty of others in the world. Just how this is supposed to happen is a matter of significant controversy. It might be that we have to treat descriptions as involving hidden demonstrative elements that tie them to the times and places they are made. It might be that we can somehow restrict the quantifiers that are involved in a definite description so that they do not capture things beyond a certain place/time (Stanley and Szabó 2000; Recanati 2004b). The point for now is that, however that gets resolved in the descriptions case, it will get resolved in the pictorial case. At least I see no deep reason for thinking that the two cases will not stand/fall together.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

60

 

Imagine the GA cops fax their sketch to NJ. Eager detectives think they’ve stumbled upon a multi-state crime spree. They show the press two indistinguishable sketches, as in Figure 4.2, and say “We think this guy is that guy.” They might be right. In this case, it matters that the police show two sketches, produced in different contexts, even though they are otherwise identical. Identical sketches can have different dthat contents when shown to the press in NJ. They are being used referentially, and thus it is an interesting question whether their contents are identical. It is not an interesting question whether the pictorial contents of those two sketches are identical. Anyone can see that the pictures are indistinguishable from one another, and there is nothing suggesting that one of them represents, for example, another sketch. Since there is no reason to think that the police are trying to make a redundant claim—someone looks like this and someone looks like this—it’s implausible that the contents of these pictures, deployed like this, remain descriptive. The police could have skipped the pictures and said, instead, “We think Harry Lollipops is Vinnie Candies.” This would have communicated the same thought. As evidence for their claim, they could then use the pictures descriptively. Vinnie looks like this [indicating the NJ sketch], while this [indicating the GA sketch] captures Harry’s appearance.

Figure 4.2 Two sketches. Drawn by Soo Sunny Park.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

61

The police sketch example is meant to show that pictures can be used referentially, and that referential use can be modeled as a case of pictorial dthat. The pictorial content becomes a non-constant character, which determines an individual as content, fixed in the context of production/utterance. It’s important that the sketch, used in this way, keeps its content across circumstances in which it is evaluated. That is, it acquires a content in the context in which it is produced, and remains rigidly latched onto it. That’s the key point about dthat operations for Kaplan, and something like that seems to be at work here. It might be, for example, that someone in Belgium fits the sketch’s pictorial content very well, and, if such a person had committed a crime, resulting in a Belgian sketch like the one above, it would have referred to him. But this law-abiding individual is not the one being talked about by the American police, and he would not be even if the police held their news conference in Belgium. Things are quite different with sketches used attributively. Talent scouts need doubles for a famous actor. The film will be shot in GA, NJ, and Belgium, and they use sketches just like those above when looking for actors. “We need someone who looks like this.” In those cases, it is the ordinary pictorial content, and not any possible dthat content, that makes sense of how the pictures are being used.

4.5 Worries about Deferred Ostention Is such an elaborate model really necessary? Perhaps the following is a better, and simpler, explanation of what is going on. We have linguistic mechanisms for securing reference, and in these cases demonstrations—this guy, that guy—do the work. A subtle understanding of demonstration, the objection goes, obviates the need for any special story about referential uses of pictures. The officers say “We think this guy is that guy.” This involves two demonstratives, used in a deferred manner (Quine 1968, 194). Use is deferred in the sense that the immediate objects demonstrated are sketches, while the intended targets of the demonstration are at (at least) one remove from those things. As Emma Borg (2002) has shown, there are many ways in which deferred ostention can work. It certainly does not demand that the intermediary be a thing referring to or in any way representing the thing ostended. Consider, for example, three demonstrative uses of pronouns:

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

62

 

She [pointing at the empty spot where she usually hangs her coat] is usually late. She [pointing at the house where she lives] is an avid gardener. She [pointing at one of her artworks] understands space much better than I do.

All of these examples use demonstrations to pick someone out via an intermediary. The intermediary’s relationship to the intended target, however, is different in each case. None of these intermediaries represents the target of the demonstration, and these examples hardly cover the range of options. In general, it’s unclear whether an interestingly unified account of deferred ostention, in all of its forms, can be given.³ Even if a unified account is forthcoming, these cases raise worries about the analysis offered in section 4.4. If it is so easy to get deferred ostention off the ground, why rely on a story about pictorial dthat to explain referential uses of pictures? Perhaps there are no properly referential uses of pictures, even though they can be used as mediums for deferred ostention. The heavy lifting is done by mechanisms in the language that demand no such fanciness on the part of pictures. These considerations foreground many ways in which the police officers could have accomplished much the same end, without using pictures at all. Perhaps, for example, the detectives just speak of the causes of the sketches, irrespective of whether we treat the sketches themselves in a referential manner. After all, in another briefing they could have just shown us things left behind by the thieves and said, “We think this guy is that guy.” They could have shown pictures of the crying candiless babies, or even brought out the babies themselves, and said much the same thing to much the same effect. So, this example involving pictures does not establish that they are being used referentially, absent further argumentation. Yes, the case gets a rather tidy treatment on the Kaplanian model. Because the picture is used referentially, it delivers an individual, which is just what the claims about “this guy” seek. But local tidiness is no virtue, one might think, in the context of so many ways of achieving reference. Though there are many routes to deferred demonstrative reference, each one does have a mechanism behind it. To see this, imagine variations on the ³ Borg (2002) is pessimistic about the prospects for a unified account, for example. Jeffrey King (2014) works out an account in terms of solving a coordination problem, and thus situates the problem as one of many involving communication.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

63

press conferences, and the follow-up questions each use of deferred ostention licenses. Possible follow-ups are useful because they can reveal the mechanism that allows for deferred ostention in each case, and some of these clearly show that pictures are being used referentially. The police say “We think this guy is that guy” while pointing to things left behind at the crime scenes. This remark invites the obtuse, but sensible, follow-up: “You mean the guy who left that [indicating artifact 1] behind is the same as the guy who left that [indicating artifact 2] behind?” Well, yes. Now imagine the police show the reporters crying candiless babies. Obtuse, but sensible: “So you mean the guy who made that baby [indicating baby 1] cry is the same as the guy who made that baby [baby 2] cry?” Indeed. The follow-ups are obtuse because reasonable people will understand what’s going on without the aid of follow-ups. But the fact that they are sensible shows something about the understood mechanism behind deferred ostention in each case. Now consider what kind of follow-ups are sanctioned by deferred ostention via the indistinguishable sketches. First: “So you think the guy who looks like that [indicating picture 1] is the same as the guy who looks like that [indicating picture 2]?” This response is worse than merely obtuse. Anyone with eyes can see that the sketches’ pictorial contents are the same, so it’s an open question whether “the guy who looks like that” is even a referring expression. There might be two, or more, or none. In addition, singling out appearances does not require demonstration of both of the pictures. One suffices, since their pictorial contents are manifestly the same. Though this response is confused in lots of ways, one virtue of it is that it treats the mediators as representations. They are treated as things that inform the crowd about appearances, perhaps among other things. Specifically, this response treats the mediators as representations with (descriptive) pictorial contents. Second: “So, you mean the GA perp looks like this [indicating picture 1] and the NJ perp looks like that [indicating picture 2]?” This misses the point of the announcement. They are not making a point about the perps looking alike, though if their guess is correct one can be quite confident that they look alike. They are suggesting that one individual is behind the robberies in both states. As with the earlier response, this leaves it a mystery why each sketch is used, rather than just one of them, since they have the same pictorial contents. Third: “So you think the guy sketched in GA is the same as the guy sketched in NJ?” Yes. This response fits the obtuse-but-sensible pattern seen

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

64

 

above. This follow-up doesn’t require referential use of the pictures because it just appeals to the sources of the sketches, and says nothing about whether the sketches even succeed in denoting or otherwise referring to the perps. One might think reference is involved, because one might think that reference is fixed via causal chains. But this follow-up, by itself, just appeals to the individuals sketched, not the individuals referred to by the pictures. Thus, this response doesn’t really make use of the pictures as representations. Yes, it appeals to sketching, but a similar follow-up would make sense if they had been ostending printouts of descriptions made by witnesses, fingerprints, or anything else that the perps might have caused. For example, “So you think the guy who made those [indicating one set of fingerprints] prints is the same as the guy who made those [indicating the other set]?” In effect, this response treats the uses of pictures along the lines of things which stand in some cause–effect relation with the things they are interested in referring to. It’s perfectly coherent to understand the example in this way, but there are also ways of understanding the use of these pictures as pictures. Fourth: “So you mean the guy picked out by this sketch [indicating picture 1] is the same as the guy picked out by that sketch [indicating picture 2]?” This treats the mediators as representations, because it appeals to what they “pick out.” Since the pictorial contents are obviously the same, the question cannot be one about the denotation of each picture, understood as the thing that satisfies the descriptive pictorial content in the present circumstances of evaluation. Anything satisfying one of them, here and now, satisfies the other and vice versa. A presupposition of this follow-up, then, is that the sketches can pick out different things, even though they are indistinguishable. This makes no sense if the pictures are understood in an attributive way. So, the sensibility of this response strongly suggests that the pictures are understood referentially. Since this seems like a reasonable way to understand things, it gives good reason to think the officers can use the pictures in that way. Like pictures generally, police sketches represent complexes of properties that can be instantiated by scenes. But such sketches are used to focus on the individual people rendered, and this is why other represented aspects of the scene, like lighting and background situation are ignored. The sketch artist leaves out such details, and doing so is in line with focusing on the individual depicted. Photos could be used similarly, however, even those that depict lots of detail about the scene in which someone is embedded. Imagine two photos that are different, except for the features of the face that they depict: we think this guy is that guy. In that case, the officers’ use of “this guy”

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

65

focuses viewers on the aspects of the photos that matter. And they could still be using those pictures referentially. But the reference is not to a scene, so much as an individual that partly constitutes each scene. In yet other cases, the scene could be the relevant content. Two kidnapping victims describe the view from their cell, and the sketch artists’ renderings match. We think this scene is that scene. The foregoing does not show that the pictures must be understood referentially for their use by the police to make sense. At best, it shows that it is reasonable to understand them as being used referentially, in accord with the model sketched in section 4.4. This is so despite the many possible routes to deferred ostention that make giving a general account of the phenomenon so difficult. It will certainly be possible to refer to the perps, in a deferred manner, via the pictures, without insisting that we understand them referentially. But there is no obvious way to do so that treats the mediating pictures as representations. To the extent that pictorial representation, as such, can serve as a medium for such deferred ostention—and why not, given all the other routes?—then the examples above demonstrate a referential use of pictures. They thereby demonstrate that there is some semantic complexity to pictures, beyond what one might have expected to find.

4.6 Postcards and Portraits There is more to life than police sketches. This section broadens things out by considering postcards and portraits.⁴ The claim is not that all uses of such pictures are referential. In fact, most of them are not. Some are, however, and one cannot understand these practices without keeping referential use in mind. Because most uses of portrait and postcard pictures are not referential, this section will hopefully complicate the discussion of those topics going forward. Practice with portraits and postcards involves two things: a subject, and some comment about the subject. The subject can be a person, place, animal, or specific thing, and exactly what is said about the subject, how it is said,

⁴ See Lopes (1996, 152). Portraiture, proper, is a much more restricted set of activities, worthy of investigation on its own (see, e.g., Freeland 2007, Spinicci 2009, Maes 2015). One thing all of these views of portraiture have in common is the claim that portraits isolate individuals, and say something about them.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

66

 

and why, seems to be rather open-ended. But how is it that such acts with pictures isolate individuals in the first place? Sometimes, the picture does the isolating work. It can do this and be used in either a referential or an attributive manner. In other cases, the picture amounts to a comment on the subject, which has been provided by a title, caption, context of use, or even history of production. On still other occasions, part of the picture isolates a subject, while another part does the commenting. The range of such acts is vast: family photo albums, vacation photos, postcards of many sorts, check-ins, selfies, Presidential portraits in post offices, and so on. This section merely illustrates how the frame provided by this account of pictures can be helpful for understanding these acts. An exhaustive discussion is well beyond reach. All of these acts can be understood within a matrix of possibilities, presented in Table 4.2. One dimension is the distinction between attributive and referential use, while the other is whether the picture is denotative or not. Some portraits are referential parts of acts of portrayal, some are attributive parts. Some are denotative, while others are not. Sometimes the picture plays the role of isolating a subject for comment—and in these cases it can be used either referentially or attributively—while sometimes it constitutes the comment made about a subject identified otherwise. And all of the previous claims can also apply, differently, to parts of pictures. Framing things like this leaves a spot for pictures that are used referentially, even though they are not denotative. Chapter 5 shows that those cases correspond to iconographic uses of pictures. Two standard communicative acts carried out by postcards are captured by “I (bet you) wish you were here!” In the “I wish you were here” cases, it’s often plausible to read such postcard pictures referentially. They make locations available for the thought expressed, in the following manner. Their pictorial contents include a detailed scene, which, in the context in which the picture was produced, denotes some locale. That rich pictorial content becomes a pictorial character, and the locale to which it refers, in the context in which it was produced, becomes the picture’s content. The “here” Table 4.2 Attributive, referential, denotative, non-denotative Attributive Denotative

Referential

Photo portrait Post office presidents/royalty Photo roster Non-Denotative Abstract portrait/Child’s “portrait” Icon (see Chapter 5)

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

67

in “wish you were here” is used demonstratively, in a deferred manner. It points first at the picture, and thereby at its content, which is the location of interest. The referents of such postcard pictures don’t shift as they’re sent through the mail, and thus evaluated in different circumstances (see Michaelson 2014). This isn’t just because postcard pictures are uniquely satisfied by certain locations. It’s possible to occupy a bland location that is not qualitatively much different from many others. Any postcard picture made there will have a pictorial content that many other boring locations also satisfy. Thus, those locations will be candidates for what the picture denotes, given its pictorial content. Despite all of this, it seems as though postcard pictures can make specific locales available for thought—wish you were here! The mechanism is just the dthat maneuver detailed above. In the context in which they were produced, their pictorial contents refer to those bland locations, and that referent becomes the dthat content. A humorous example of a bad postcard picture is the trope, “[Location] at night.” A marketing genius decided to sell uniformly black postcards to many locales, mostly on the more Podunk side of the aisle, simply explaining that each postcard represents a different place, at night. This makes sense, along the lines of the identical police sketches, and it has the bonus of violating all other kinds of norms for postcard pictures. Such postcards don’t have these locations as their pictorial contents, of course, since particular individuals don’t show up in pictorial content. Also, with the pictorial contents they have—uniform expanses of black, or dark—it is unlikely that they denote the locations in question in the circumstances in which they are evaluated. Such pictures cannot isolate the location, even at night, that it apparently represents, because any dark closet is an equally good candidate for satisfying that descriptive content. But these pictures can be used referentially in the manner described above, assuming one finds a particularly dark patch of each location as the context of production/utterance. The idea is that the picture is produced as a representation of a particularly dark patch of some location. In the context in which it is produced, the black picture succeeds in referring to a dark part of that town. The pictorial content is then treated as a character. That character delivers the location in question as a content because, in that context, the character succeeds in picking out that location descriptively. Understood referentially, the picture keeps this content—the specific location—across circumstances of evaluation. Wish you were here!

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

68

 

It doesn’t do justice to this phenomenon to suggest that the captions of these postcards do “all the work,” whatever that means. On the model here, what the captions do, in theory (I get it), is indicate something about the context of production. Not all expressions wear their contexts of production on their sleeves, so sometimes proper interpretation requires the use of other representations, like those that name a place. To be fair, the joke in the location-at-night postcards also works if the pictures are read attributively, with the caption isolating a location for comment. For example, read the caption as part of a claim that the picture completes: [Location], at night, is like this [indicating the picture].

That is, the postcard says that this is one seriously Podunk place to spend one’s time. That’s somewhat funny, and perhaps as funny as such postcards could get. From this philosopher’s perspective, it’s much more fun to think of the pictures as working referentially in those contexts. Both readings are available, which is all that’s needed for present purposes. Other uses of postcards, such as inciting envy in one’s friends, do not support referential readings of their pictures. “I bet you wish you were here!” In these cases, the picture is used attributively, as follows: I am in an enviable situation, one like this [indicating the picture].

In that case, the picture is used attributively, filling in the “like this” part of the thought in a way that’s more striking than most descriptions. It’s important to choose a picture that not only captures the appearance of some locale, but captures its best face. The location-at-night postcards, for example, don’t work at all for the envy-inciting crowd. Portraits, too, can denote individuals, or fail to do so, and be used either in a referential or an attributive manner, though purely referential uses of portraits proper are fairly uncommon. Portraiture involves the representation of particular individuals, especially persons, but how those individuals make their way into acts of portraiture varies. In some cases, a name accompanies the portrait, or its placement in a house, or atop a casket, makes it clear which person is involved. In some cases, the picture denotes the individual, capturing enough of her properties to single her out among others. Because pictorial content can be impressively rich and detailed, it’s possible for a portrait to denote an individual under an impressively broad

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

69

range of circumstances. In other cases, the pictorial content might single someone out, but only do so in a rather limited range of circumstances. It would probably be possible to recognize Sir Thomas More based solely on Holbein’s 1527 portrait (Figure 4.3). Such a picture denotes More across time and space, in that it might turn out that no one else satisfies its content. Many photographic portraits are like this too. Other practices are less keen on detail, though that doesn’t mean they fail to denote their subjects, in the right contexts. “The book on the table” denotes a book when it is said in a context that includes only one table, which has just one book on it.⁵ Such a description doesn’t travel well, because the world is full of solitary books atop tables. We can say similar things about portraits. Many in Cezanne’s

Figure 4.3 Hans Holbein, Sir Thomas More, 1527, oil on oak panel (cradled), 74.9  60.3 cm. © The Frick Collection.

⁵ See, for example, Strawson (1950). It is far from obvious how one ought to account for such “incomplete” definite descriptions. I’m not offering a new account of them here, but just indicating where the topic is relevant to pictures.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

70

 

Figure 4.4 Paul Cezanne, Madame Cezanne (Hortense Fiquet, 1850–1922) in a Red Dress, 1888–90, oil on canvas, 116.5  89.5 cm. The Metropolitan Museum of Art, CCO 1.0 Universal, Public Domain.

circle, for example, would have a fairly clear conception of Marie-Hortense Fiquet (Figure 4.4), even though most now do not. So, it’s plausible that his portraits (1891, etc.) singled her out descriptively among the members of that circle. In fact, her acquaintances could have had a concept of a Fiquet-looking person, and Cezanne’s portraits, without going into precise physiognomic detail, could nevertheless represent a Fiquetlooking person. Perhaps it’s fair to criticize Cezanne for treating his subjects more like objects than portraiture calls for (see Freeland 2007, 98), but it helps to remember that denotation can be a fairly local phenomenon. Portraits, because they are pictures with potentially rich contents, can denote individuals. As such, they can also be used referentially. But it’s implausible that acts of portraiture typically use pictures referentially. And it’s also implausible that when used attributively, the point of the picture is to isolate an individual for comment. Instead, the picture is more plausibly a

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

71

part of the comment made about the individual, who has been isolated by other means. To see this, consider: The man wearing red is wearing red.

This sentence singles someone out, by hypothesis, via a denoting definite description. And it says of that individual that he is wearing red. Far from an interesting communicative act, this claim attributes a feature to someone that was used to isolate him in the first place. If pictorial content is used primarily as a means of singling out a subject, what is the comment that the act of portraiture makes about her or him? The obvious choice in such a case would be that the non-pictorial parts of the act of portrayal constitute the comment made about the subject. Non-pictorial parts of a portrait are not terribly appealing as comments about the subject, however. They might include a name, or a title, for example. In that case, the portrait says something like: The individual like this is named so-and-so.

Sometimes pictures are used in precisely this manner. A photo roster, for example, treats the pictures, with their descriptive contents, in a definite manner, and says, of what satisfies those descriptions, that they are called soand-so. Rosters help teachers associate names with faces. Slightly more interesting: The individual like this is an ancestor of ours.

This act can be accomplished by placing the photo in the right place within a house. Similarly: The individual like this (who was like this) is entombed here.

Such acts use the picture as a denoting definite description, which singles someone out for comment. Portraits proper, by contrast, are better understood as comments on subjects, singled out otherwise, perhaps with names or by the contexts in which they are presented. So-and-so is like this [indicating the picture].

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

72

 

Portraits’ descriptive contents are often thus treated indefinitely, in that we say, of a subject, that she is a person like this. Even though portraiture proper might aim for pictures that manage to single out their subjects via denotation, it’s not essential, since the larger act, of which the picture is but a part, singles out the subject anyway. This allows us to see how children’s drawings and caricatures are genetically related to, even if distinct from, portraiture. The act isolates an individual—via a caption, context, or what have you—and the picture constitutes a comment on it. The comment might be wildly inaccurate, or simply not terribly detailed. The act identifies the individual extra-pictorially, so that the picture can then be a comment on the subject. Because it’s rare for the picture to play the role of isolating a subject in acts of portraiture, it’s also rare for pictures to be used referentially in portraiture. All they would contribute to the acts in that case would be the individuals. Interestingly, some practices in the neighborhood, even though not portraiture proper, seem to work in exactly this manner. Pictures of presidents, royalty, and the like, placed in official venues, seem to work referentially. They are not meant to inform viewers of what the important figure looks like, so they do not aspire to say, for example: The President of the United States is like this.

After all, those pictures can be quite out-of-date, and it would be odd to regard them as inappropriate once, for example, the President gets a haircut, or goes gray. Such a reading also does little to explain why official venues find it so important to show what important people look like. Such pictures, placed and titled just so, do seem to say something like: This is the President of the United States.

The picture delivers the individual who is the President, and in that sense makes him or her present for comment, consideration, and so on. The comment in this act just says, of that individual, that he or she is the President. But making the President, King, Queen, and so on, present for comment and consideration is, for many reasons, an important act, so perhaps in these cases pictures are used referentially. This relates closely to what some people do with icons. They are meant to make a Saint or other important person available for comment, consideration, entreaties, and so on. Icons, however, cannot be understood attributively or denotatively,

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

    

73

because viewers typically have no idea whatsoever what kinds of visible features their subjects had. It misses the point to think that an icon reveals St Anthony’s appearance, and there will be a lot more to say about icons in Chapter 5. As a quick summary, let’s reconsider Table 4.2. This table does not specify whether an attributive use of a denotative picture isolates a subject for comment (as a photo roster might), or constitutes the comment on a subject isolated otherwise (more typical of a portrait). That constitutes another dimension of this matrix that becomes salient once pictures become parts of communicative acts.

4.7 Using Parts of Pictures Referentially The cases considered so far involve pictures, as wholes, used referentially and attributively, perhaps embedded within communicative acts involving language. But pictures are syntactically and semantically complex. They have parts of their own, even if these parts are not as grammatically diverse as the parts of linguistic representations. And Chapter 3 suggests that in the right contexts different parts of pictures can play different roles in expressing a thought. The same considerations apply here. In some cases, a part of a picture is treated referentially, while the rest of it works descriptively, just as sometimes parts are treated indefinitely while others are treated definitely. Well-established practices, or highly specific contexts, allow for such partwise uses of pictures. In lieu of attempting to give a full description of the range of such acts, let’s return to the vacation photos. At some point or other, everyone has endured slideshows of friends’ vacation photos. The family members are subjects, along the lines discussed in Chapter 3. So, the parts of the photo responsible for picking out the family members, in effect, have things predicated of them. “There we are at the beach!” “There we are in the forest!” Ugh. Bracket, for now, just how the other parts of the photo are used in such acts. In most cases, the family parts—that is, the parts responsible for representing family members—are used referentially. The pictures in a vacation album cannot pick out some other family, even if that family happens to look like the one that took the pictures. What constitutes a family part is also open to contextual, pragmatic negotiation. It’s not as simple as cutting out the spatial part of the picture

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

74

 

that represents a family member, because sometimes the point is to show the person wearing this or that, with this or that kind of makeup on, and so on. This should make sense because the syntactic parts of a picture need not all be spatially disjoint. Some descriptive elements are deployed with the goal of securing reference to a family member, while other descriptive elements might be used to say, of this individual, that they are thus-and-so. The non-family parts of these pictures are typically understood attributively: “We were in such-and-such a situation.” They can also, in part, be understood referentially. We were in such-and-such a situation, near the Eiffel Tower. The vacation photo trope of standing in front of famous places seems to work this way. Practices like these have a good bit in common with those in which iconographic conventions allow one to tell stories about important people, with pictures, even though viewers are in the dark about what those people look like. More on that in Chapter 5.

4.8 Individuals and Properties in Other Accounts While there are many accounts of what makes pictorial representation distinctive, few spend a lot of time trying to model pictorial meaning along the lines pursued here. Nevertheless, the difference between attributive content and singular content has been quite important to past theories, and it will help to set that work within the frame suggested above. Despite the fact that this will be a fairly summary discussion, it will prove valuable for three reasons. First, it contextualizes those views in a new way, even though more work would be required to flesh out these points fully. Second, it shows what aspects of the current proposal are open to those with different accounts of depiction. And third, it helps show just what a book like this is trying to accomplish. This book is not offering a new theory of what makes pictorial representation distinctive, so what, exactly, is it doing? Though this doesn’t capture every view, the following three points constitute a decent amount of common ground between otherwise competing theories. First, they account for depiction by isolating a mechanism that accounts for our understanding of pictures. Since that mechanism is not operative in linguistic understanding, the stories go, these mechanisms are identified as those which make depiction a distinctive kind of representation. Second, they suggest that much the same mechanism is responsible for interpreting pictures as representing patterns of features as is responsible for them representing particular individuals. Third, they distinguish the

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

     

75

representation of particulars from the representation of properties by appealing to what are usually cast as different levels of pictorial content. Dominic Lopes (1996, Ch. 5, cf. Gombrich 1951, 1961, Schier 1986, Neander 1987, Podro 1987, Sartwell 1991) suggests that pictures represent by exploiting recognition responses, and there are two kinds of recognition—content and subject—that correspond to the representation of properties and individuals, respectively. Robert Hopkins (1995, 1998) suggests that we understand pictures via experienced resemblances in outline shape, but that we can experience the picture as representing either a thing with such-and-such features—the “content of seeing-in”—or a particular individual, “pictorial content” (Hopkins 1998, 128). John Hyman (2000, 24–5; 2006, 65–7) thinks that the “internal subject” of a picture is a rich set of features, while its “external subject” is the individual, if any, the picture represents. For Hyman, the mechanism behind interpretation of both sorts is the noticing of perspectival features that the picture shares with its contents.⁶ Catharine Abell (2009, cf. Blumson 2014) thinks that pictures work by getting cooperative viewers to notice resemblances, though the kinds of resemblances can vary quite a bit, and the resemblances can relate the picture to patterns of properties or to particular individuals. Kendall Walton (1973, 1990) suggests that we understand pictures by imagining, of our seeing them, that it is a seeing of something else. That state can include seeing particular individuals as well as patterns of features. All of these theorists tend to agree that the properties represented can play a role in determining which particular individuals are also represented. Because of this, these two aspects of content can be called levels. Gabriel Greenberg claims that while pictures have “unified” contents that involve singular and attributive aspects, “we may also, more or less at will, track purely attributive content as well. The point is that there is a level of content which incorporates singular content.” (Greenberg 2018, 867 n. 3).⁷ ⁶ In more recent work, Hyman (2013) suggests that the “referent” of a picture, which in the current context should be understood as a singular content, is not closely connected to its attributive content, or “sense.” It is likely inappropriate to think of these different contents, then, as occupying different levels. ⁷ Lopes (1996, Ch5) is perhaps the most helpful extended discussion of this. As noted, the levels talk might not work for Hyman, and Greenberg (2018) does not commit to how these two kinds of content are related. Richard Wollheim (1980, 1987), and in a different way Edmund Husserl (2006) and Lambert Wiesing (2009), think that there is a single mechanism at work—a special perceptual state—but it’s harder to find the point about levels in their work. Nelson Goodman (1976) thought that pictures had both attributive and singular contents, but he did not articulate these in terms of levels of content. In fact, he thought that the two were independent.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

76

 

My approach (Kulvicki 2006) has levels too, but not in the same way. Pictures are distinctive because of how their syntactic features relate to their bare bones contents. So, for me, the interesting levels are bare bones and fleshed-out content. Within the present book, I’ve suggested that we think of bare bones content as a pictorial character, which delivers a fleshed-out or pictorial content in context. In effect, I have put my account of what makes pictures distinctive within a semantic frame. That frame makes the question of how pictures represent particular individuals interesting. We have seen one way in which pictures can acquire singular content in this chapter, while Chapter 5 and Chapter 7 will offer two more. The semantic frame, however, is available to anyone. In fact, it’s quite tempting to locate other theories within it as follows. Everyone wants an account of how to get from the picture surface to a pictorial content. For Lopes the rule might involve explicit mention of recognition, for Walton the imagination, for Hopkins, experienced resemblances, for Abell noticed conversationally appropriate resemblances, and so on. The semantic model is not about these mechanisms behind pictorial understanding. It is about what pictures are about, and how the different things they are about might relate to one another. When the police present the indistinguishable sketches, do they contribute particular individuals to what the police are expressing? If so, how do they manage that? Yes, one might recognize a particular individual when looking at the sketch, but why think that numerically distinct but otherwise identical sketches might deliver different individuals? If the sketches do not accomplish this end, then one needs to explain why it seems appropriate to follow up by asking whether the person picked out by one is the same as the person picked out by the other. To be sure, there are other ways of trying to make sense out of these phenomena, but being a recognition theorist, or an experienced resemblance theorist, does not prevent one from adopting a model like this. The claim being made here is not that no other accounts of these phenomena are possible, but that the one involving character and content is compelling. Admittedly, the character–content distinction is particularly helpful and plausible if pictures uniformly admit of multiple interpretations depending on the contexts in which they are deployed. If pictures are representations with non-constant character, it is a simple extension of the view to account for the representation of particular individuals in terms of operations defined over character and content. Chapter 2 makes a case for that claim, while this chapter extends the case for the usefulness of these semantic tools. As a whole, this book is meant to present an overall view that derives some

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



77

of its support from the fact that it works to explicate a whole range of things we can do with pictures, maps, and other representations.

4.9 Summary The machinery introduced in Chapters 2 and 3 opens up a way of thinking about how pictures might be used referentially. Examples presented above suggest that there are referential uses of pictures, even if referential use does not dominate the ways in which pictures are deployed in communication. Pictorial contents are descriptive. In referential use, those contents work as characters for determining new contents. This is strictly analogous to the way in which a description can be used referentially according to Kaplan’s dthat model. The new dthat content is a particular individual, the one that satisfies the descriptive content in the context in which the picture was produced. Because parts of pictures, understood as Chapter 3 suggests, are also representations, different parts of pictures can play different roles in expressing a thought. Some parts can play the role of isolating a subject, while others play the role of commenting on it. Which parts play which role, of course, is not marked grammatically, so it is only in fairly specific contexts, or within fairly well-established practices, that this partwise division of labor manifests itself. Just as this works with parts of pictures understood attributively, it works with the parts when some of them are understood referentially. Dthat contents do not exhaust the range of referential use. One important lacuna is iconography. Iconographic contents can be understood as the result of an operation on pictorial character and content, but they cannot be understood as instances of dthat contents. This is primarily because iconographic conventions have no denotative aspirations. They are not aimed at capturing the specific appearances of the individuals to which they refer, while doing so is essential to the rule that yields a dthat content.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

5 Iconography Societies fond of picture-making typically also include rich iconographies. Erwin Panofsky, one of the important early theorists of iconography, understood what we have been calling pictorial content as the “natural subject matter” of pictures. Anyone with eyes to see and basic familiarity with picture-making ought to be able to understand that the picture depicts a chair, or a person, and so on. The more interesting thing for Panofsky was pictures’ “conventional subject matter,” which is “apprehended by realizing that a male figure with a knife represents St Bartholomew, that a female figure with a peach in her hand is a personification of veracity” (Panofsky 1955, 28–9) This is the iconographic content of a picture, which is distinct from its pictorial content, and cannot be understood as a dthat content, either. Panofsky rightly characterized iconographic content as conventional, for reasons that will become apparent. This chapter will not do justice to the full range of interpretive practice that art historians call iconography, but it captures a semantically interesting part of it. Interpreted iconographically, pictures have particular individuals, broadly speaking, as contents. That is, iconographic content constitutively involves people, specific places, fictional entities, abstracta like virtues, numbers, and so on. The mechanism that accomplishes this is, like dthat, defined over characters and contents. After introducing the phenomenon, we’ll consider a simple semantic model for iconographic content. The simplicity of this model fails to capture the actual ways in which iconographic conventions have developed, however. I suggest that pragmatic concerns, combined with the semantic mechanism, make a lot of sense out of iconographic practice. Crucially, though theoretical work like Panofsky’s was developed in response to practices in the European Renaissance, iconography shapes our dealings with pictures across a wide range of times and places, including our own. So, on this view iconography is a central mechanism for explaining the communicative uses to which pictures are put.

Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

 

79

5.1 Introducing Iconography Consider Figure 5.1. Raphael represents St Catherine of Alexandria, but it’s a challenge to understand how he does this for two reasons. First, pictorial content does not include particular individuals. Second, no one knows what Catherine looked like.¹ It would thus be impossible for Raphael to have knowingly created a picture that denotes her, or, has a pictorial content that she (perhaps uniquely) satisfies. How does this image manage to be so clearly about a specific saint, even though it probably does not bear her likeness? As the story goes (de Voragine 1993, Ch. 172), Roman Emperor Maxentius tried to execute the young princess Catherine in a particularly gory fashion, using a bladed wheel. Miraculously, the wheel broke apart. This must have been a great relief to Catherine and those who knew her. The next thing they tried to do was cut off her head. Contrary to Catherine’s and everyone else’s expectations, that worked. Lots of people have been beheaded by Emperors, but the wheel is distinctive, if not unique, to Catherine’s story.² Iconographically, it’s the presence of the wheel that shows the picture to be about her, rather than any of a number of other saints. The halo—it’s subtle in the reproduction—shows her to be a saint, and the wheel alone would not usually suffice to represent Catherine, absent a depiction of a young, well-dressed woman with a halo nearby. A man standing by a bladed wheel, for example, or an old woman, would not be Catherine. Following art-historical practice (see Taylor ms, e.g.), we can call the wheel an attribute of St Catherine. Ideally, attributes are unique to iconographic subjects, lest we wind up with ambiguous icons. As mentioned, the wheel does not suffice, but attributes like this are an important pictorial phenomenon and so deserve special attention when thinking about the semantics of such pictures. Both Catherine and Maxentius’s attempts to execute her—the event—are particular individuals, and so both are possible iconographic subjects. The fourteenth-century Spanish panel shown in Figure 5.2 seems to go into more ¹ It’s possible that Catherine did not exist (Thurston and Attwater 1956 vol. 4, 420–1). There are well over 10,000 saints, some of whom might have been invented (Taylor ms). I am bracketing any special issues that might arise, broadly speaking, in relation to fictional subjects of pictures. ² As Taylor (ms) points out, St Euphemia (de Voragine 1993, Ch. 139) was also victimized by a wheel.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

80



Figure 5.1 Raphael. Saint Catherine of Alexandria, 1508, oil on poplar, 72.2 × 55.7 cm. © The National Gallery, London.

detail about the event. We only can guess at Maxentius’s identity because of the wheel, and the depicted haloed woman. Maxentius never had a distinctive attribute associated with him, and not just because he was no saint. The wheel is what marks the panel as picking out Catherine, and the rest of the stuff, including the Emperor and the axe-wielding angel, comes along for the ride. This panel gives a sense of the scene of the first attempt to execute her. Later on, we will have occasion to ask whether this panel iconographically represents Catherine, the important event, or both. This choice will matter, but the stakes are far from clear at this point. Consider also the sixteenth-century sacra conversazione by the Venetian painter known as Palma Vecchio in Figure 5.3. Palma’s painting involves Mary, Christ, John the Baptist, and St Sebastian. Paintings like this are late developments of a form in which figures important in the Church were presented together, usually surrounding the Virgin and Child. We know who’s who because of iconographic conventions. The shaggy guy at center in

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

 

81

Figure 5.2 Unknown, Spanish, St Catherine Delivered from the Wheel, ca 1375–1400, oil and tempera on panel, 35.8 × 33 cm. Art Institute of Chicago, CCO 1.0 Universal, Public Domain.

sackcloth with a stick cross is the Baptist. The unfortunate guy, hands bound, with the arrow protruding from his chest is Sebastian, and the modest woman and child combo is the Virgin and Christ. Unlike the Spanish panel involving Catherine, there is no story behind this presentation of important figures in the Roman Catholic Church. This is apt to be a bit confusing, because this painting is made with the seemingly effortless realism of the High Renaissance. That is, though this is a sacra conversazione, and painted in a vividly realistic style, it is not meant to represent a conversation involving these individuals. The pictorial content does involve four people occupying the same space, within earshot of one another. In that sense, we could understand this picture as depicting a conversation, but what could the topic of conversation possibly be? The only one that suggests itself is the arrow protruding from that guy’s chest. In this MST3000 interpretation of the painting, the Baptist

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

82



Figure 5.3 Palma Vecchio, Sacra Conversazione, ca 1516–18, tempera on wood, 84.5 × 106 cm. Collection of the Poznań Society of Friends of Sciences (deposited at the National Museum in Poznań), MNP Mo 24.

looks away, disgusted, while Mary tries to stop Jesus from trying to grab it. Fun, perhaps, but inaccurate, even though the pictorial content clearly includes a guy with an arrow protruding from his chest. The iconographic conventions—guy with arrow, for example—establish reference, and the viewer is invited to consider these three important figures, together, and their relevance to the Church. Raphael and Palma are not using their pictures to say much of anything about Catherine, Mary, John, and Sebastian. We cannot even interpret Raphael as depicting Catherine standing next to a wheel. The pictorial content is certainly, inter alia, a woman standing next to a wheel, but the only way Catherine manages to get represented in the first place is via the depiction of the wheel. Similarly, Palma does not depict Sebastian as having an arrow protruding from his chest, even though the pictorial content of the painting includes a man with an arrow protruding from his chest. The depiction of the arrow and a man serves to establish reference to Sebastian,

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

83

not to predicate anything of Sebastian. In order to do the latter, we would need some way of isolating Sebastian, independently of the depicted arrows. Unlike Palma and Raphael, the Spanish panel does seem to predicate things of the subjects depicted. There is Catherine, one might think, surrounded by the wheel, Maxentius, a guardian angel, and some extra who will certainly be killed before the episode is over. One might think that this is just a more complex version of what we find in Raphael, or Palma. As we will see, a lot rides on just how the panel is more complex than Raphael or Palma. European painting, especially from the fourteenth through the sixteenth centuries, served as the impetus for Panofsky’s discussion of iconography, so the concept applies most obviously there. But pictorial practices from many times and places are iconographically rich, and iconography is also quite salient in contemporary practice. For now, the European historical examples suffice as exemplars of the phenomenon, but we will have occasion to discuss more examples as the chapter proceeds. What all of this suggests is that the pictorial contents of these pictures are not their contents understood iconographically. At best, the pictorial content helps establish what Panofsky called the conventional, or iconographic, subject matter. And “subject matter” is a good way of putting things. Iconographic conventions deliver individuals—people, events, and even abstracta like virtues—as contents, and we are then, depending on the context, invited to do something with them. That is, iconographic interpretation is referential. It provides a subject matter, about which something or other could be said, with respect to which some action could be performed, and so on.

5.2 A Semantic Mechanism The character–content mechanism behind dthat content provides a model for one that explains iconographic interpretation. Ordinary pictorial content becomes a character, which, given the iconographic conventions in place, delivers an individual as a new content. That is, elements within the content have been assigned referring roles conventionally. The new character— originally the pictorial content—given the conventions, assigns individuals as the new content. In Raphael’s case, the pictorial content includes a young woman beside a bladed wheel. Interpreted iconographically, given the conventions in place at the time Raphael was painting, those represented features deliver Catherine

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

84



of Alexandria as the content. A man, bound and shot through with arrows, is Sebastian. (Like the wheel, by the way, the arrows didn’t work. Diocletian had to have him beaten to death.) In Palma’s painting, we get the Virgin and Child, Sebastian, and the Baptist. In general, the strategy is to find ordinary pictorial contents—men and women, wheels and arrows—and use them to refer to something else. This is a semantic mechanism like dthat—were it not so trendy we could call it ithat—because it takes a pictorial content as input, in context, and delivers a new content as a result. But this mechanism works quite differently than dthat. Yes, it delivers individuals as contents, but no, it does not do so by finding what the pictorial content denotes in context. Instead, it appeals to something much like a code, or a set of iconographic conventions. These codes can differ from place to place and change over time, so that fifteenth-century European iconographic rules can differ from those used in Middle Kingdom Egypt or contemporary Japan. Icons, understood in this manner, are not their own kinds of representation, but a kind of interpretation of certain kinds of representation, especially pictures. At the end of the chapter we will consider cases in which something like this seems to happen in language. Pictorial contents can be quite complex, and this leads to two questions about this iconographic semantic mechanism. First, do the Catherine paintings refer to Catherine, or to the event involving Catherine and the wheel? Both are good candidates. Second, how do we handle the simultaneous representation of more than one set of attributes, as in the Palma painting? Just as it is a convention to connect Catherine with the depiction of a young woman and a wheel, it is a convention whether this or that sort of painting singles out the event of her execution, as opposed to Catherine the saint. Which conventions were operative at any given time is an interesting art-historical question that is beyond the scope of this study. In section 5.3, we will consider an interesting fact about the period that points to a related, but non-iconographic way of interpreting some of these pictures. For now, the point is that the semantic mechanism proposed above allows for a great range of referential uses of pictures, with the details depending on the governing conventions at any given time and place. What about the four important figures in the Palma painting? First, note that aspects of pictorial content can be grouped in different ways, and this matters for iconographic identification. St Philomena was also shot at with arrows three times—as with Sebastian, at Diocletian’s behest—though they only managed to hit her once. That didn’t work, so they also tried to drown

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

85

her with an anchor, but that didn’t work either, so they had to cut off her head.³ Palma depicts a woman, and arrows, but this painting does not depict Philomena. The man is shot through with arrows, and when Philomena is depicted with arrows (the anchor is more common), she is holding them. So, iconographic interpretation involves finding features depicted as standing in the right relationships with one another. When the attributes and other features for more than one individual are presented in a single painting, the painting refers to each of them, unless it thereby manifests the iconographic attributes for something else. So, the Palma painting refers to Mary, Jesus, Sebastian, and John the Baptist. It’s as if we conjoined the names like I did in the previous sentence. Imagine, however, that the attributes for all four of them were used iconographically to single out St Bartholomew. There is nothing in the semantic rule to stop that from happening. In that case we cannot unambiguously establish iconographic reference to that collection of individuals in the way Palma does. Perhaps another convention exists for that group. If not, then the painting is ambiguous. Either it picks out the group, or Bartholomew. Because it would be crazy to establish iconographic schemes that work this way, we don’t, and so these messes are not a salient part of iconographic interpretation. One could imagine more mundane cases of overlap, however. Sometimes, saints have the same attributes, or an artist paints a figure with attributes for distinct individuals. In those cases, we are left with a mess because the mechanism does not unambiguously deliver a referent.⁴ This model shows how iconography is related to pictorial content. We cannot interpret a picture iconographically without also being able to interpret it pictorially. But the move to iconographic content is not a move to, for example, a different way of correlating patches of color with qualities in space. The mechanisms that generate pictorial content, whatever they are, remain. Pictorial content becomes part of a character that delivers, potentially, a range of particular individuals or abstracta as contents. None of these things can figure in ordinary pictorial contents, as we have been

³ See Borrelli (2011). Whatever lessons one might learn from the lives of the saints, the lesson of their deaths is “keep it simple.” ⁴ Apropos St Catherine, the church of Santa Illuminata in Montefalco houses a peculiar painting. Apparently, they originally wanted a Catherine but then later decided that they wanted a St Illuminata. The artist had already painted the wheel, but no matter: it was easy to paint over it. Later, however, the painting was “restored” so that the wheel shows, making a holy mess of the iconography. Thanks to Paul Taylor for drawing my attention to this example. See http:// www.museodimontefalco.it/it/antoniazzo-romano-e-montefalco_12.html.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

86



understanding them over the last few chapters, because pictorial content is only descriptive. The semantic mechanism behind iconographic interpretation is impressively simple, and one might worry that it cannot do justice to the phenomenon. Iconography is a rich practice, after all, realized in multiple ways across time and space. I think that worry is misguided. Given the communicative goals of iconographic representation and interpretation, there are many constraints on how iconographic interpretation works in practice, and these pragmatic considerations show how a simple mechanism can yield a wealth of practices. Sections 5.3 and 5.4 describe phenomena that do not count as iconographic interpretation. This shows that while the semantic mechanism sketched in this section is simple, it does a decent bit of work isolating an interesting phenomenon.

5.3 Labeling Imagine labeling a picture “St Sebastian.” The picture otherwise just looks like a picture of a young man, with none of the so-called attributes of Sebastian figuring in its pictorial content. This labeling certainly allows the representation (with label) to be used in a referential manner, and indeed to pick out St Sebastian. We’re allowed to use pictures in any way we like, and it’s hard to see why such a label wouldn’t allow us to use it to represent the saint of our choice. For a more elaborate example, consider just two parts of Dürer’s gloriously over-the-top Great Triumphal Car (1522), in which Reason drives a chariot on wheels of glory and dignity, with goodness and constancy along for the ride (Figure 5.4). None of the human figures Dürer depicts has anything like an attribute for those abstracta. They are all just labeled. None of this is iconographic interpretation, as I suggest we understand it. The end result of labeling can be functionally similar to iconographic interpretation, viz. a picture interpreted referentially. Taylor (ms) points out that in many cases only a label allows us to single out to whom the picture is meant to refer. But while iconographic interpretation deploys conventions defined over pictorial contents to fix reference, labeling uses, well, labels. Labeling, so understood, can be thought of as an extreme case that allows one to use any picture to stand for anything one likes. Label a picture of a pipe “St Sebastian” and you wind up with a representation of Sebastian. Of course, in such cases, one wonders what the point is of including the picture in the first place. Wouldn’t the label do all by itself? If all that matters is

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



87

Figure 5.4 Albrecht Dürer, Great Triumphal Car (detail) 1522. British Museum, Public Domain.

delivering an object, then of course the label would suffice. But the ways in which this is done with pictures open up expressive possibilities otherwise absent from language. We miss the potential here if we focus on possible one-off correspondences like that between Sebastian and a pipe. The people, wheels, and other objects in Dürer’s Great Triumphal Car bear about as much relation to the virtues as the pipe bears to Sebastian. But by pairing depictions of those objects with such virtues one opens up the possibility for important allegorical and other figurative interpretation. In those cases, the label fixes the reference, and the pictorial content is freed to contribute to one’s comment on the subjects so identified. Worries about labels and pictures go back quite a long way in the Western world. For example, art historian Liz James discusses a striking passage from the Libri Carolini, most likely written by Theodulf of Orleans between 790 and 793. It is surprisingly contemporary in its use of indistinguishables to serve a philosophical end: He was shown two pictures of beautiful women without captions. The painter supplied one picture with the caption “Virgin Mary” and the other

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

88

 with the caption “Venus”. The picture with the caption “Mother of God” was elevated, venerated, and kissed; while the other, because it had the caption “Venus” was maligned, scorned, and cursed, although both were equal in shape and color and made of identical material and differed only in caption. (IV, 16, 528–9; in James 2005, 99)

Theodulf was writing on behalf of Charlemagne in response to what turned out to be a bad translation of the proceedings from the Second Council of Nicaea in 787 (James 2005, 98). James goes on to suggest that Theodulf ’s example misses the important fact that no one at the time would make a picture meant to refer to Venus that looked anything like one meant to refer to the Virgin. That is, there are conventional constraints on which pictorial contents have which kinds of referential significance, and we will discuss some of them below. For now, the point is that the semantic mechanism outlined here for iconography does not capture the labeling phenomenon. The Dürer print ought also to be interpreted allegorically. And one might want to fit allegorical interpretation under the larger banner of metaphorical interpretation. Those issues can wait until Chapter 6, but it is worth noting that the semantic mechanisms behind pictorial metaphor might play in complicated ways with those behind iconographic interpretation. Once the many kinds of semantic and pragmatic mechanisms of picture interpretation are on the table, it will become easier to think about how they might all interact. So far, we have only seen that iconographic interpretation is incompatible with pictorial dthat, in the sense that you can’t do both at once.

5.4 Stories The second non-iconographic bit of interpretation is typically included within iconography proper by art historians. Charles Hope (1986) reminds us that Renaissance patrons and painters drew an important line between images, on the one hand, and stories, on the other. Though the terminology is unhelpful in a contemporary context—both the images and the stories Hope discussed are pictures, and thus images given current use—the distinction is important. A story is “a narrative, showing an event, and its main purpose was didactic.” (Hope 1986, 804). Of the two St Catherine pictures shown above, only one of them (Figure 5.2) is plausibly a story. The panel shows Catherine beside the wheel, being sentenced by Maxentius while a guard

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



89

sleeps and an angel axes the wheel. The other, Raphael’s (Figure 5.1), is an image: “An image was a representation of a person, for example a saint, and it was often used as a focus for devotion” (Hope 1986, 804). Raphael presents nothing but Catherine. There is no action in the painting. The point is not so much to show anything Catherine did, so much as to make Catherine available for prayers, thoughts, and the like. Earlier, we noted that it would be inappropriate to interpret the Palma sacra conversazione as depicting a conversation, because the best candidates for topics were fairly inappropriate to the uses for which the picture was made. Palma’s painting is an image, in Hope’s sense, even though it involves more than one individual. Likewise, it’s inappropriate to interpret the Raphael as being about a woman hanging out with a spiked wheel. Yes, the pictorial content involves a woman and a wheel, but that’s not the point of the picture. The picture, via the iconographic mechanism sketched in section 5.3, delivers Catherine as a content. If one were to continue, “No! It shows Catherine alongside a wheel,” we would be hard pressed to understand how, since the only way the picture delivers Catherine in the first place is by, inter alia, depiction of a wheel. Standing-by-a-wheel is not predicated of Catherine, because the representation does not independently isolate Catherine so that such a comment can be made about her. Images, in Hope’s sense, are meant to deliver individuals, and in Renaissance practice, the individual was typically an important person. Stories are about important events. Do they deliver events for contemplation in the way that images deliver people? Not in the same way. As Hope suggests, stories served didactic purposes. The pictorial content was meant to show events in such a way that people could learn from them. The pictorial content, then, played a role not so much in fixing reference to some event as in predicating features of it. The event in question was singled out for comment otherwise, either by a label or by locals in the know who could point this out to those who engaged with the picture. Many of these people were illiterate (Baxandall 1988), so the picture served as a way to engage with the events that were described in otherwise inaccessible texts. Alberti devotes much of the latter part of Della pittura (1435/1991) to explaining how to present stories. Late in the fifteenth century, Leonardo would criticize what’s probably a Botticelli annunciation (1489, below: Leonardo didn’t name names) because it looked as though Gabriel had scared the daylights out of Mary (see Baxandall 1988, Kemp 1989). Let’s be honest: Gabriel probably did just that. The proper frame of mind for

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

90



someone contemplating this event, however, is less slapstick, more resigned (see Figure 5.5). Contrast that with Leonardo’s famous annunciation (ca 1472, Figure 5.6), completed when he was only about twenty years old, which is a model of restraint by comparison. Pictorial content, for Alberti, has to be responsible for delivering the proper emotional responses to things. One of the chief avenues for this was the representation of facial expressions and bodily comportment. In effect, these annunciations tell a story. To treat them as simply referring to, and thus delivering, the event in question is to miss the point. They have pictorial contents that are importantly related to what they want us to do with those events. The interpretation of stories, in Hope’s sense, can work in two related ways, neither of which is wholly iconographic. First, reference to the event in question is established independently of the picture, and the picture is thus

Figure 5.5 Botticelli, Sandro. The Cestello Annunciation. 1489–90, tempera on panel, 150  156 cm, Uffizi Gallery. Photo: Scala/Art Resource, New York.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



91

Figure 5.6 Leonardo da Vinci. Annunciation, Ca. 1472, oil on wood, 98  217 cm, Uffizi Gallery. Photo: Scala/Ministerio per I Beni e le Attività culturali/Art Resource, New York.

an illustration of the event. It works attributively, offering a perspective on what happened. Second, parts of the picture work iconographically to establish reference to individuals. The rest works attributively to offer a perspective on the events in question. The second is the more plausible option for many annunciations, as Gabriel and Mary are both presented along with their standard iconographic markers (wings, lilies, halos, etc.). The second option might also apply, albeit in a more attenuated fashion, for what Hope calls images. After all, Raphael had to decide what facial expression was fitting for Catherine. Her facial expression can change even though all of the iconographic markers—wheel, young, well-dressed woman—stay the same. In that sense, it might be that images, too, are meant to be understood via a combination of iconographic reference-fixing and feature attribution. This seems right, even though there is no event in Catherine’s life that such images are meant to capture. The same mix of interpretive tools applies to images and stories, which just suggests that the Renaissance distinction emphasized by Hope is not built to carve representation at the semantically interesting joints. This isn’t a criticism, since neither Hope nor Renaissance painters, nor patrons, were trying to do that. Sometimes paintings come to us without any labels or any other obvious way of figuring out what event, if any, they were meant to illustrate. For art historians, one goal in such cases is finding some story in the literature that might have served as the subject matter for the painting. This is not iconographic interpretation, in the specific sense articulated above. It is

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

92



more of a matching exercise. The painting has significant elements like people engaging with one another in a specific fashion. The goal is to find a story that such a painting could plausibly be an illustration of. Taylor (2011), for example, discusses an early Rembrandt (The Injustice of Piso, 1626, Figure 5.7) that has accumulated no fewer than fifteen published attempts to connect it with a story it might be illustrating. Taylor makes a convincing case that Seneca’s De ira is the source of the Rembrandt. Doing this is a subtle process—his was the fifteenth attempt— which involves sorting out the details that matter from those that do not. Muskets, for example, were not part of Seneca’s world, though they are clearly depicted in the lower left. They are not a detail that Taylor’s reading accounts for, but he gives reasons for leaving them out. The painting was commissioned with a subject matter in mind, and was thus used as an illustration of some story. In that sense, though not a Renaissance work, this painting is more of a story than an image.

Figure 5.7 Rembrandt van Rijn, Historiestuk met zelfportret van de schilder, signed and dated 1626, oil on panel, 90.1  121.3 cm, Museum De Lackenhal, Leiden; long-term loan from the Cultural Heritage Agency of the Netherlands.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

93

5.5 The Practicalities of Iconography The simple mechanism behind iconography is compatible with many possible iconographic practices. Successfully using pictures iconographically, however, requires that iconography answer to some practical constraints. To begin, it seems as though iconographic interpretation typically requires an impressive disregard for pictorial detail. Pictorial content is very rich and fine-grained. The features of pictorial content relevant to iconographic interpretation, by contrast, are usually fairly abstract. It’s not a bladed wheel, of exactly this sort, arranged in just this manner with respect to the ambient light, but just a bladed or spiked wheel that matters. That doesn’t suffice for St Catherine, since one typically also needs to represent a woman in the vicinity of the wheel. Like the wheel, the woman need not be depicted wearing exactly these clothes, with this hairstyle, facial features, and so on. So, while the pictorial content of a St Catherine icon might be quite rich, the aspects of it that secure reference to Catherine are quite thin, and this is completely typical of iconographic interpretation. Semantically speaking, it’s perfectly possible to imagine a practice in which only a scene of such-and-such a highly detailed sort picks out Catherine, and so on for the other individuals represented iconographically. But it’s hard to imagine the motivation for such a practice, and there are many reasons to avoid it. First, choosing fairly abstract features of scenes to do iconographic work makes it easy to interpret pictures this way. If highly specific contents were the only bearers of iconographic weight, then it would be hard, without doing a lot of work, to determine whether a given picture deserves any particular iconographic reading. So, as consumers, iconography would be a needlessly difficult endeavor. Second, the range of things we would like to represent iconographically is usually rather limited: saints, martyrs, gods, virtues, vices, and so on. Each group is open-ended in principle, but at any given time picture-makers have a modest, possibly expanding range of things in mind. Pictorial contents can be vastly specific, and so the number of possible pictorial contents outruns any class of individuals or abstracta we might decide we want to represent iconographically (see Kulvicki 2006, Ch. 3). Given that, we could choose just a handful of highly specific pictorial contents to bear iconographic weight, but why do that? Third, from the perspective of producers, if only highly specific contents had iconographic significance, making such pictures would be both difficult

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

94



and boring. It would amount to producing copies, or near copies, of a given pictorial representation. European painting made much of the freedom afforded by abstract choices for iconographic attributes. Building iconography around abstract features of scenes leaves artists with much freedom to depict things as they like. And it makes it similarly easy for viewers to understand the iconography with which they are presented. All of this suggests that iconographies are like languages, in at least two ways. First, aspects of depicted scenes—man with keys, woman with wheel— can conventionally be used to stand for particular individuals, virtues, and vices. Attributes are like the lexemes of a language, which can make pictures stand for individuals they could not pictorially represent. There are no pictorial constraints on which attributes can be used to stand for which individuals, and so the attributes, vis-à-vis their referents, are much like words. Second, attributes are chosen so that they can be represented, or, we might say, expressed, in line with practical constraints on linguistic representations like words and sentences. Written and spoken words are such that it is relatively easy to identify and produce them. If it were not, using language as a communicative tool would be unnecessarily difficult. Abstract sets of features are easy to recognize in pictorial contents, and they leave picture-makers many options for expressing them. It’s not off base to say that a system of attributes works more like a digital system of representation than one that is classically analog (see Kulvicki 2015a, 2017). Each iconography is, in a way, its own language. As mentioned earlier, pictures are distinctive because of how their syntactic features constrain their skeletal contents, and how those characters constrain pictorial content. Iconographic conventions are language-like in that they are completely open-ended. Any feature that could be included in pictorial content can be used to stand for any particular individual, place, idea, or what have you. There need be nothing particularly visual, for example, about the iconographic content of a painting, even though its pictorial content must be visual. Iconographies are thus more like languages than systems of depiction are.

5.6 Uses of Iconographic Interpretation Though fifteenth- and sixteenth-century European painting was the inspiration for Panofsky’s discussion of iconography, the phenomenon has long been and continues to be an important one for understanding pictures. The

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

95

aims of this chapter are theoretical, but I hope the following convinces readers that the scope of iconographic interpretation is quite broad. Venerated icons have been a subject of debate, on and off, since well before the eighth century when Theodulf wrote the Libri Carolini. Idolatry is an enduring worry for anyone who hopes their prayers are directed toward the proper objects, and one might worry that venerating icons sends prayers the wrong way. Regardless of where you stand on the veneration of icons, note that when an icon is venerated, it is treated referentially. The targets of such icons are visually unknown to their venerators, so it is through iconographic means that reference is secured. One hopes that the reference, so secured, is to the proper object. No one uses an icon of Joseph to think about what he was thinking when the picture was made. It’s not a portrait, so no interesting descriptive interpretations are salient.⁵ Surprisingly, computer icons work in a related fashion. They are little pictures, but they do not function attributively. One icon depicts a yellow manila-style folder at an oblique angle, illuminated from the upper right. That image is used generically to refer to a kind of computer file, called a folder, which is not yellow, manila-style, or at any specific kind of angle with respect to the user, much less illuminated. Because these icons are used generically to pick out folders, while others pick out files, and so on, there is some descriptive content to them, but it has nothing to do with the icons’ pictorial contents. Technically, we could have just used roman letters for each of the things iconically represented. If we did so, however, it would be difficult to learn the rules correlating letters with things. Leaning on pictorial interpretation, and analogies with the functions of the things referenced, works much better. Comic books tell stories, but they tell stories about specific, usually fictional, individuals. What’s more, they typically use pictorial styles that are light on depicted detail. So, you can invent a character, but not one with a particularly detailed physiognomy, because the style typical of comics will not permit you to render such details pictorially. Comic book characters are often thus identified via something very much like iconographic markers: uniforms, hair styles, hats, glasses, and so on. Since comics tell stories, it is

⁵ Things are more complicated than this lets on. In light of Hans Belting’s (1994) work, Cynthia Freeland (2010) points out that Eastern icons in particular were thought to capture the appearances of their subjects and, in many cases, to be causally related to them in ways similar to photographs. See also Maynard (1997, 238–9). There is much more for philosophers to say about iconography.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

96



merely parts of those pictures that work iconographically. The rest can be deployed attributively, in the service of the story. When iconographic mechanisms are used, artists pay an attributive price. To the extent that some depicted feature plays a role in fixing reference, it is not also a candidate for being something attributed to a character. It’s very difficult to pictorially represent St Catherine beside a wheel because the wheel is the thing that fixes reference to Catherine in the first place, and putting Catherine beside two wheels (one to fix reference, the other to attribute some feature of her) would be odd at best. The same worry applies to your favorite comic book character. If a special hat is used to establish reference to the character, it’s hard to depict that character as wearing such a hat, just like it’s hard to depict the character as not wearing a hat. In those cases, we want to use some aspects of a picture in two incompatible ways.

5.7 Iconographic Interpretation in Language? Given that iconographic interpretation is modeled as a simple operation with character and content, one might want to know whether such a thing shows up in language. This specific operation does not seem common because language has ready access to referring terms like proper names, which are absent in pictures. But sometimes even though it’s easy to say something, semantically speaking, it is not appropriate to say it. Or, it’s not appropriate to say it in a certain manner. In these cases, we find examples of a similar mechanism, or so it seems to me. Coded speech can include what seem like iconographic elements, in which aspects of the ordinary content turn an expression into a referring term. The Anti-Defamation League maintains a database of hate symbols, many of them are ordinary bits of language recruited to nefarious ends. Coded speech need not always be a bad thing. Sometimes it might be the only way important messages can be communicated. In the current political climate, however, almost all of the salient examples seem negative. Sarah Palin, for example, sent out a tweet on July 7, 2017, which read “Trump Gives Speech to the People of Poland, Says 14 Words That Leave Americans Stunned.” Though it did not get much press, Rebecca Schoenkopf at Wonkette responded on the same day with an article entitled “Sarah Palin literally just went full Nazi. This is not a drill.” Schoenkopf said of expressions like “14 words,” “These are not ‘secret’ codes. They are red

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

    ? 97 neon lights coupled with tsunami alert horns.”⁶ And they are, in a way. The expression “14 words” stands for a white supremacist claim that doesn’t deserve repeating in the present context. These expressions are often called “dog whistles” because they are, on the surface, respectable expressions in ordinary language, but carry a specific meaning for those in the know. For present purposes, they look like iconographic uses of language because they take an ordinary content, and use it to establish reference to something else, like a white supremacist ideology. It turns out things were more complicated than the tweet and Schoenkopf ’s response suggest, by the way. The tweet was drawn from a Facebook teaser for an article posted by the website for the Young Conservatives group. They suggested that they often use expressions of the form “xx words” in their titles, and that they were referring to a statement at the end of a speech Trump gave in Poland the previous day. Talking Points Memo asked Todd Gutnick of the Anti-Defamation League about the affair, and he responded that “We don’t believe she meant those 14 words. Not a story here.”⁷ Fair enough. But this response misses part of the point. It seems as though Palin was working in ignorance of something like iconographic conventions. She was someone who painted a woman standing near a bladed wheel, but had no idea about the St Catherine thing. “We don’t think she meant that woman and wheel.” There are many ways in which expressions acquire idiomatic and even metaphorical meanings. Iconography is significant for pictures because pictorial content is descriptive, and limited to the sensible. Iconographic interpretation makes pictures referential in a manner untethered to the sensible. We can do similar things, semantically, with language, even though languages have many other means of securing reference. Panofsky contrasted the natural subject matter of a picture with its conventional subject matter. The idea was that anyone with eyes to see could unpack what we have been calling pictorial content, but that understanding iconography was akin to knowing a code. One might think that all of language is conventional in this sense. After all, there are few constraints on what words need to be like in order to pick out this or that thing. All language speakers are in on the code. But there can be codes within codes. The distinction between iconographic and ordinary interpretation of linguistic ⁶ Schoenkopf 2017: https://wonkette.com/619957/sarah-palin-literally-just-went-full-nazithis-is-not-a-drill. ⁷ Shuham 2017: http://talkingpointsmemo.com/livewire/sarah-palin-tweet-14-words.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

98



expressions is very much like the distinction as it applies to pictures, so it seems plausible that this kind of interpretation is a part of language as well.

5.8 Summary To the extent that the term “iconography” is familiar, it is associated with art history, and even specific groups of art historians. This chapter isolates a phenomenon that does not fit precisely with that used by art historians, even though it largely overlaps with it. Iconographic interpretation is a way of making particular individuals the contents of pictures. It is useful when denotation is unavailable as a means for securing reference. In some cases, denotation is unavailable because the thing to which one wants to refer has no qualities that could be part of pictorial content. Some pictures represent integrity, but you cannot depict integrity. In other cases, we have lost any descriptive sense of the thing to which we want to refer, even if we once had it. If St Catherine really existed, she could have sat for a portrait, and there is no doubt that painters in Maxentius’s time may have captured her appearance quite well. But we have no such records, and we’re not even sure she really existed. So, if we want a picture to be about Catherine, we need another means to that end. Structurally, iconographic interpretation shares much with the dthat interpretations described in Chapter 4. A pictorial content becomes, in this case, a character, which fixes a new content based on iconographic conventions. As with any analysis of content in pictures, it pays to remember that pictures have parts, and that sometimes parts of pictures can serve to establish reference iconographically, while others play an attributive role. We find this in some stories, and it is plausible that some uses of images in the sense Hope described work this way as well.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

6 Metaphor The character–content distinction has proven helpful for understanding a number of ways in which pictures manage to be meaningful. Once the machinery is up and running, of course, it’s worth exploring just how far it can go, and metaphor presents an interesting case for two reasons. First, Josef Stern (2000) has an account of linguistic metaphor that has been worked out in terms of operations over character and content. And second, Stern (1997) claims that there are no pictorial metaphors. This sounds like a challenge! Given that we have been able to use character and content profitably to understand pictures, it should be possible at least to apply Stern’s thoughts about metaphor to the pictorial case as well. Because Stern is skeptical that there are pictorial metaphors, the first few sections of this chapter are devoted to isolating the phenomenon of interest. The goal is to find cases of metaphor in pictures that seem independent of language. Because the existence of linguistic metaphor is taken as given, pictorial phenomena that relate closely to linguistic metaphor are bad examples. They might, after all is said and done, only seem like metaphorical pictures because of closely related linguistic metaphors. In addition, it’s common to label many non-standard uses of pictures “metaphorical,” but we are looking for something quite specific. Once strictly pictorial metaphors have been identified as a phenomenon of interest, Stern’s account of metaphor will be introduced, and then applied to these cases. His account works quite well, and thus further vindicates the value of character and content as useful tools for modeling pictures’ meanings. Stern’s objections to metaphorical pictures are rooted partly in his views about pictures, partly in his views about character and content. In the former case, he thinks it makes no sense to talk of picture types, and in that sense follows Nelson Goodman (1976). In the latter, he thinks character and content are endemic to language. This book is built around the suggestion that they are not. There are a number of things this chapter does not, and could not, aspire to do. A small but mighty literature on metaphor in pictures already exists. Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

100



It will only come up in passing, because I am not trying to show that the approach to metaphor in pictures presented here is the best one. The goal is to isolate a pictorial phenomenon, and then use the character–content tools to explain it. That can happen even if, at the end of the day, some other approach wins out as the best explanation. In addition, showing that the approach to pictorial metaphor unpacked here is best should also involve trying to defend a Stern-style semantic approach to metaphor against its able competitors. That, too, would take us much further afield than the focus of this book permits. The results of this chapter might be deployed in a defense of Stern’s account of metaphor, or be used against it, but that’s a project for another day. Chapters 4 and 5 explored ways in which particular individuals can wind up being the contents of pictures. With iconography, the range of particular individuals that could be contents of pictures is not limited by the fact that ordinary pictorial content involves only perceptible features. Pictorial metaphor, if it works as suggested, provides a mechanism for making a wide range of imperceptible properties the contents of pictures.

6.1 Illustrated, Suggested, and Supplemental Metaphors Metaphor is a kind of use. It’s tempting to think of some expressions as being metaphorical in and of themselves, while others are literal. But it’s only when an expression is used in the appropriate way that it makes sense to say it is literal, or metaphorical. A deranged individual might claim that Juliette is the sun and mean what he says in a completely literal fashion. Likewise, most ordinary statements can be used metaphorically. In fact, whenever one encounters an expression that seems strange if taken literally, it is tempting to say that the statement itself is just metaphorical. But that would be a mistake. Calling the statement metaphorical, when presented with it out of its context of use, is at best a bet concerning the contexts in which it might be used felicitously. The same lesson applies to pictures. It’s not that some pictures are metaphorical, some not, but that some pictures are used to express metaphors in some contexts, some not. One and the same picture might be understood metaphorically in one context of use and literally in another. To show that there are metaphorical pictures, it helps to focus on a specific set of cases. There are many ways in which pictures might be used metaphorically, but only some uses point to there being strictly pictorial

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

, ,   

101

metaphors. A strictly pictorial metaphor is, for now, understood negatively. It is a metaphorical use of a picture that is neither an illustrated metaphor, nor a suggested metaphor, nor a supplement to a metaphor. Metaphors are common in language, and one thing that makes them interesting is that they also have literal contents. So, a statement understood metaphorically also has a literal meaning, even if that is not what one intends to communicate in using the claim metaphorically. Sometimes, the literal content of a metaphorical statement is the kind of thing that can be depicted. “He is burning up,” said of someone who is quite angry, has the literal content that someone is on fire, and that situation can be depicted. Doing so is illustrating a metaphor. Figure 6.1 shows one of James Gillray’s many engravings illustrating a metaphor (cf. Stern 1997, 261). It suggests, not so subtly, that British Prime Minister William Pitt and Napoléon Bonaparte are carving up the world into different regions of influence. The pictorial content captures the literal content of the metaphor, by putting the world on the table, to be cut up by the statesmen. The ad in Figure 6.2, from Ruf Lanz, for Foundation Tier im Recht, also seems like a straightforward case of illustrated metaphor. In this case, the tag line suggests the metaphor that is being illustrated: animals are garbage. Indeed, “x’s are garbage” is pretty much a dead metaphor at this point, so this rich image might manage to reenergize it.¹ This case also might count as one of suggested metaphor, as described below. Often, it is plausible that illustrated metaphors are genuinely metaphorical uses of pictures. Such examples are poor candidates for starting one’s investigation, though, because they are all open to a simple objection. Since there are linguistic metaphors, and these examples illustrate the literal contents of such metaphors, it’s open to think that the picture is just an ordinary illustration. It’s not meant to be interpreted metaphorically, so much as bring a linguistic metaphor to mind, reinvigorate it, and so on. Because these examples do not require there to be genuinely metaphorical pictorial interpretation—though they allow it—they are not good for convincing anyone who is a bit skeptical that there are genuinely metaphorical uses of pictures. Suggested metaphors are a similar case. There are always obvious and less obvious things to say when trying to articulate the contents of pictures. Sometimes, the natural things to say sound a lot like the literal contents of

¹ Thanks to Josef Sten for the suggestion.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

102



Figure 6.1 James Gillray, The Division of the World, 1805, colored engraving. Photo: bpk Bildagentur. Art Resource, New York. Public domain.

Figure 6.2 Some Animals are Treated Like Garbage, 2017. Advertising Agency: Ruf Lanz, Zurich, Switzerland. CGI: Carioca Studio/Visualeyes International © Tier im Recht.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

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

103

metaphors. “His head’s on fire!” True: the picture depicts someone with a burning head. Or, in light of seeing the Ruf Lanz picture (Figure 6.2): “That animal is garbage!” In most contexts in which someone would say such a thing, they mean it metaphorically. In cases like this, a picture can suggest a linguistic metaphor. It might be that such a picture can be used metaphorically, and so is a genuine example of pictorial metaphor. Maybe not. The problem with focusing on such cases is the same as with illustrated metaphors. They might be parasitic on metaphor in language, and thus are not good cases for convincing skeptics. The goal is to find examples of metaphor in pictures that are not simply derived from their linguistic cousins. A third case worth excluding involves pictures used as parts of more complex linguistic expressions. Showing a picture of an atom bomb explosion, someone says, “This was my mind when I heard the news.” She’s not demonstratively referring to the picture, let’s say, but to the picture’s content, in a deferred fashion (see, e.g., Borg 2002). She could have said, instead, “My mind was a violent conflagration.” The picture represents a violent conflagration, though, so it can fill in the sentence instead of the words. Again, cases like this are not implausible examples of metaphorical uses of pictures, but they are not ideal for convincing someone that there is a phenomenon here that is independent of language. In effect, cases like this might be examples of either illustrated or suggested metaphors. And the case for these supplemental metaphors being that way is strengthened by the fact that the picture completes a linguistic utterance. What your friend said, in language, is metaphorical. But that should not convince anyone that the picture is used metaphorically.

6.2 Some Non-metaphorical, Atypical Uses of Pictures When it’s hard to make sense out of some expression in ordinary terms, it is tempting to say that it is intended metaphorically. This can be completely innocent, in that the target remark might genuinely express a metaphor. But there are ways of not making literal sense that have nothing to do with metaphor. Similar remarks apply to the concept of expression. When a painting fails to be representational in any obvious sense, people often say that it’s expressive, but this obscures what might be many nonrepresentational uses of painting. These are what William Wimsatt calls wastebasket uses of concepts. Things that don’t fit one concept are put under another heading. When trying to get clear on the notion that had been used

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

104



as a wastebasket, however, it is important to keep in mind that the things put there need not all be alike. With that in mind, consider the fact that Stern begins his article on metaphors in pictures with a quote from Picasso: “My sculptures are plastic metaphors.” (Gilot and Lake 1964, 297).² He then follows Noël Carroll (1994) in using Picasso’s Baboon and Young (Figure 6.3) as an example of a visual metaphor. The Museum of Modern Art in New York used one of these bronzes in a 2008 show Focus: Picasso Sculpture, for which the gallery

Figure 6.3 Pablo Picasso, Baboon and Young, 1955, Bronze, 53.3  33.3  52.7 cm. © Estate of Pablo Picasso/Artists Rights Society (ARS), New York. Digital Image © The Museum of Modern Art/Licensed by SCALA/Art Resource, New York.

² In fact, Carroll (1994, 191) misquotes the Gilot and Lake text as “My sculptures are visual metaphors.” Stern follows him in this. The book first appeared in English, so it’s not a translation issue, and in any case the French uses “plastique” Gilot and Lake (2006, 318). Virgil Aldrich (1968) got the quote right, but Aldrich’s paper is entitled “Visual metaphor,” which might explain the later misquotations.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

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

105

label reads that the piece “shows Picasso’s power of metaphorical transformation at its height.”³ Picasso did talk about both painting and sculpture effecting a “metaphorical” transformation (Gilot and Lake 1964, 297f.). As Stern points out (1997, 256), however, it is not obvious that Picasso was using the term “metaphor” in anything but a very loose, even metaphorical sense; perhaps the same is true of the Museum of Modern Art curators, who probably should have done more than parrot Picasso. This is a clever bit of sculpture, to be sure, but what makes it metaphorical? Carroll suggests that the way it identifies a car with a head is metaphorical (1994, 191–192). Shorn of a context of use, however, it’s hard to say that this sculpture is or is not metaphorical. What, precisely, is the metaphor here? Baboon’s heads are not cars, even though a car can work as one. This is not a standard way to represent a baboon, but that is a long way from saying the sculpture is meant metaphorically. In some contexts, the expression “The baboon’s head is a car” can express a metaphor, as can “Cars are baboons’ heads.” Likewise, in some contexts Picasso’s sculpture can express a metaphor. But it does not seem as though a proper understanding of this artwork requires it to be read metaphorically. Picasso’s sculpture is an assemblage work. These are quite common in the arts, and they are fun. The head is made of two toy cars, the ears are mug handles, the body a jug. The original (1951), which was later cast in bronze (1955), is in the collection of the Picasso Museum in Paris, alongside other similar works such as an owl made of nails and scrap steel. Where is the metaphor here? Relatedly, John Kennedy’s interesting work unpacks different ways in which pictures can represent, and he pioneered the study of tactile pictures by the blind. He suggests that there are pictorial metaphors, but includes what seems like an implausible range of phenomena within the category. For Kennedy, motion marks in comics are pictorial metaphors, because the picture suggests movement, even though there is nothing in the picture that moves. At one point he suggests that Ernst Gombrich (1963) discusses “four types” of metaphor: “personification, simile, allusion, and the pun in depiction.” (Kennedy 1982, 589). All of these devices are interesting, and one might think that many of them somehow operate outside of the normal range of pictorial interpretation. Even if that’s true, it’s a stretch to call them metaphorical in any but

³ https://www.moma.org/collection/works/81119 (accessed 3/4/18).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

106



the loosest sense. Conventions such as motion marks are definitely worth studying. In fact, Chapter 7 suggests they constitute an added dimension of variability on a picture surface, of the sort found in maps. For now, the point is that, while they are worth studying, it might be unhelpful to think of them as metaphors. Artists can bring about clever transformations, and they have developed many tricks for conveying content with images. Metaphor seems to be one such transformation, but it’s just one. The reason it is helpful to focus on advertising is precisely because it presents an interesting communicaticontext within which pictures can be discussed. This happens in the fine arts too, but Picasso’s baboon, treated as central by Carroll, doesn’t even seem to be metaphorical. Going forward, then, keep in mind that it’s tempting to call things metaphorical, even though use of that word is sometimes loose, or even metaphorical (Stern 1997, 256).

6.3 Strictly Pictorial Metaphors The goal is to find strictly pictorial metaphors, which are not illustrated, suggested, or supplemental. Such cases involve pictures used metaphorically, in such a manner that it is difficult to attribute their metaphorical status to some nearby linguistic expression. First, consider the award-winning ad in Figure 6.4, from Memac Ogilvy in Qatar. The spatial design is drawn from decorative patterns common in Qatar and other nearby parts of the Arab world. This is a public service campaign aimed at encouraging people to use their seatbelts. The pattern is closely associated with the local cultural heritage, and one of the aims of the campaign is to combat the view that wearing seatbelts is just not something real Qataris do. This ad thus associates seatbelts with Qataris’ cultural heritage. The tag line is “Seat belts save lives. Let’s make them a part of our culture.” This fails to be an illustrated metaphor. Well, let’s just say I am making a bet here. Unfamiliar with the local culture, I cannot assert with confidence that there is no linguistic metaphor in the driver’s seat, but I am confident that there is not. I am also fairly confident that no obvious metaphor is suggested. “Seatbelts are a part of Qatari culture” can be used metaphorically, but in this conversation it would not be. And this ad also fails to simply supplement the tag line. This ad is an impressive metaphorical use of a picture. The tag line could even be left out, and one could imagine careful,

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

Figure 6.4 Nujoom, 2016. Advertising Agency: Memac Ogilvy. © Qatar Islamic Bank.

107

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

108



patient viewers understanding the message conveyed. This and the other examples will be discussed more later in the chapter. For now, the goal is to present some plausible examples of strictly pictorial metaphors. Second, Agencia UM, a firm based in Recife, Brazil launched the ad in Figure 6.5 as part of a public service campaign in 2016 Without reading the text, it pays to try and figure out what the picture is trying to say. It takes some time to unpack this one. The campaign is aimed at illegal child labor. More specifically, it is aimed at those who might enable the practice by not speaking up about it. This ad depicts a picture of a child burdened by blocks of stone. But the child is burdened only because the individual’s mouth is depicted as being closed. This is a very clever and touching use of depiction to make a point, and it is in no way related to any obvious linguistic metaphor one might use, again, unless a local metaphor is in play. The tag line is not metaphorical, and it seems like nothing but the non-metaphorical message behind the metaphor so cleverly presented here. “Child labor: If you don’t speak up, it doesn’t stop.” Notice how much more powerful the Memac Ogily and Agencia UM pictures are, when understood as metaphors, than Picasso’s bronze baboon. As advertisements, they come with a communicative context and goal, and they clearly seem like metaphorical uses of pictures. Third, the Chennai office of JWT launched an ad campaign for The Times of India in 2009, which depicts a bunch of newspapers from the side, rolled up and framed in such a way that they look like a fingerprint (Figure 6.6). Fingerprints are associated with individual identity both in its mundane every-flower-is-different sense and in the more profound sense that we are all distinctive selves, different groups have their own identities, and so on. Making the fingerprint out of newspapers suggests that that the newspaper is constitutive of identity. The ad’s tag line is “The identity of young Chennai.” This is not just an illustrated metaphor because there is no longstanding metaphor relating a population’s identity to its newspapers, or to fingerprints. It is also a poor candidate for being a suggested metaphor. Though one can wring some metaphorical expressions out of what the picture presents, they hardly capture what the picture does. The picture is more than just a means for presenting a linguistic metaphor. The advertisements reviewed in this section constitute metaphorical uses of pictures, and one can find more examples by consulting the website Ads of the World.⁴ These examples are not illustrated, nor suggested, nor

⁴ https://www.adsoftheworld.com/.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

109

Figure 6.5 Child Labor, 2016. Advertising Agency: Agencia UM, Recife. © Agencia UN.

supplemental. Pictures are deployed metaphorically, in a manner that is not simply parasitic on linguistic metaphor. Section 6.4 unpacks Stern’s account of metaphor, which is then applied to these examples in section 6.5.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

110



Figure 6.6 The Identity of Young Chennai, 2009. Advertising Agency: JWT, Chennai. © The Times of India.

6.4 Stern on Mthat Stern (2000) believes we cannot understand metaphor without modeling it partly as a semantic phenomenon. One motivation for this is just how complicated metaphorical interpretation can be. Semantics studies the

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

111

meanings that expressions, as it were, bring to the conversation. Figuring out what someone intends to communicate, based on what she says, can be constrained in part by the semantic content of her expressions (Stern 1997, 259). You can figure out I am remarking that the weather is great by saying “What terrible weather!” because the semantic content of that remark bears an interesting relation to the way things are (not). Stern thinks that appeal to semantics is required to explain how we interpret metaphors successfully. Others, naturally, disagree. Perhaps the majority of theorists reject the claim that metaphor is partly semantic.⁵ If metaphor is partly semantic, how does it manage to be that way? Stern suggests that the semantic aspects of metaphor are operations genetically related to Kaplan’s dthat. Stern’s mthat operation takes an ordinary content, and transforms it into part of a new character, which delivers a new, metaphorical content in context. Specifically: Mthat: For every context c and for every expression E, an occurrence of ‘Mthat(E)’ in a sentence S(= . . . Mthat(E) . . . ) in c (directly) expresses a set of properties P presupposed to be m-associated with E in c such that the proposition < . . . P . . . > is either true or false in the circumstance of c. (Stern 2000, 115)

A linguistic expression, E, has an ordinary, literal meaning. “He is on fire!” Take that meaning, and find properties that are, in context, presupposed to be metaphorically associated (m-associated) with E. Fire, for example, is often presupposed to be associated with things like anger, speed, danger, and warmth. Some of those properties become the new, metaphorical, content of the expression in context. In many contexts, for example, we say he is on fire to indicate that he is angry. The mechanics of the operation should make sense in light of the discussion of dthat and iconographic interpretation in Chapters 4 and 5. A content becomes a character, which in turn determines a new content. What remains is to explain presupposition and metaphorical association. Stern understands presuppositions as belief-like states that play an important role in determining the appropriateness of what we say (2000, 117). It’s appropriate for the motorist to say to me simply “I’m out of gas” when he might want to communicate that he needs help of a certain sort, ⁵ As mentioned earlier, this is not the place to be arguing about which theory of metaphor is the best.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

112



such as directions to a gas station. In part, that is because of facts about me. I look like a local, not a tourist. I don’t look like I am a reporter looking for an interviewee. I look like someone likely to speak his language. In part, the motorist’s choice of words is appropriate because of facts about other aspects of the situation. It doesn’t look like the set of a gameshow—Who’s out of gas?—or a speed dating session, or a movie set. Though my interlocutor might not run through these things in his head, they all seem to be what allow us to carry on the conversation as we do. If the unlucky individual thought I was an actor on a set, he might not have approached me in the first place, let alone said that he’s out of gas. And if the guy thought I believed I was an actor on a set, even though, transparently, I am not, he also would not have approached me. Moreover, if the guy thought I would think he is actor on a set, he couldn’t be confident I would answer him in a helpful fashion. So, presuppositions are complexes of beliefs about the situation, including beliefs about what people believe about each other (cf. Grice 1989, Stalnaker 1984). And Stern’s point is that, in conversations, properties can be presupposed to be m-associated with certain meaningful expressions.⁶ So what, exactly, is m-association? Metaphorical association intentionally covers a fairly broad class of relations, and I will not run through all of them here (see Stern 2000, 108–17). Metaphors work for different reasons, in that they can succeed because of properties that things in their literal contents have, have in an exemplary way, do not have, do not have in an exemplary way, are structurally related to, and so on. The covering term for this fairly broad set of relations is massociation. For example, fingerprints can be m-associated with personal identity in some contexts, and with criminality in others. Redness can be massociated with anger, or with joy. Being a newspaper can be m-associated with being informative, with regularity (they show up every day!), with commuting, and so on. The idea behind mthat is that in conversation it is often presupposed that certain features are m-associated with a given expression, E. When an expression is used metaphorically, the mthat operation delivers those presupposed, m-associated properties as a new content. Because presupposition is pragmatic, one’s range of metaphorical potential is shaped by context. So, when considering whether this model applies to pictures, we need to keep in mind that it is not the pictures, shorn of any conversational use, that are at ⁶ Stern credits Max Black (1962) with being the first to recognize the importance of presuppositions to metaphors (Stern 2000, 108).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

    

113

issue. It is pictures being used for a certain purpose, aimed at a certain audience. This should be unsurprising because metaphor, as mentioned earlier, is itself a kind of use. And throughout, this book has been at pains to stress that we can understand the meanings of pictures only when we consider how they are deployed communicatively.

6.5 Mthat and Strictly Pictorial Metaphors Recall the fingerprint made of newspapers (Figure 6.6). A fingerprint pattern is part of the pictorial content, as is being a pattern formed from newspapers. And this is an advertisement. So those encountering the image are expected to try and get a point. This is especially important in the pictorial case, because pictorial contents are indefinitely rich and varied. Communicative context makes it appropriate for us to focus on the fact that the pattern is like a fingerprint and that it is made of newspaper. In addition, when searching for a metaphorical meaning, we are going to need to decide which aspects of pictorial content are intended metaphorically, which not. In this case, the fingerprint pattern drives the metaphorical interpretation of the picture. Fingerprints are m-associated with identity, for example. They are supposed to be unique to each individual, and in that sense they are a stand-in for an individual. In a trivial sense, they make each of us unique—we don’t have the same fingerprints—but this distinctive feature is associated with the fact that we are distinctive in many other ways. So, given the features massociated with the fingerprint pattern, the picture manages to communicate a claim about the relationship between newspapers and identity. In a different context, fingerprints have different metaphorical potential, because different features are m-associated with them. Perhaps the very same image could be used in a political campaign aimed at the “Lying Press.” In that context, it is the association between fingerprints and criminality that drives interpretation of the picture. Notice that, in this and other cases, metaphorical meaning is not a completely semantic affair. The picture, like any other metaphor according to Stern’s account, helps with metaphorical interpretation by delivering some features m-associated with the features that are part of pictorial content. Also notice that linguistic metaphors work this way as well. Literal contents can be deployed with myriad metaphorical significance, depending on what features are presupposed to be m-associated with the literal meanings being used.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

114



We can apply the same thinking to the Qatari seatbelts ad (Figure 6.4), and the Brazilian ad condemning child labor (Figure 6.5). The campaign from Qatar Islamic Bank shows a pattern that is m-associated with the local cultural heritage. Because the pattern is made out of seatbelts, Memac Ogilvy cleverly connects them to the heritage. One can imagine a different context in which the seatbelts and pattern are used differently. For example, a different use of a similar picture could suggest that a strong cultural heritage is a bulwark against depravity. In that context, the picture would not suggest using seatbelts, but would use seatbelts metaphorically. Being protective is m-associated with being a seatbelt, and that fact could matter in a different context of use. Remember, metaphor is a use of a meaningful expression, so it is expected that one and the same picture can be used to communicate different metaphors. Similarly, no one thinks that literally closing one’s mouth has anything to do with child labor. But mouths are m-associated with speech and expression generally, and a closed mouth, in that sense, is associated with a lack of it. The mechanics of the picture do the rest of the work: It’s the silence that leaves the burden on the depicted child. In this way, Agencia UM helps the Ministry of Public Works bring the problems of child labor to the consciousness of people in Recife. But the same image, in a different context, could communicate something very different. Since mouths are m-associated with consumption—we eat with them, for example—this picture could be part of an ad campaign suggesting that consumption by the wealthy enables child labor, or that consumption is a painful burden. In the latter case, child labor per se is left out entirely. Neither of these imagined examples reflect how the picture is actually used, but they show the expected possible variation in use. Overall, then, it seems straightforward to apply Stern’s thoughts about metaphor to pictures. So, this constitutes one more reason for thinking that the character–content tools discussed here are valuable when we are thinking about pictures. Where does this leave us with the illustrated, suggested, and supplemental metaphors? Are they truly metaphorical uses of pictures, or something else? It depends on the communicative act of which the picture is a part. A picture can illustrate a metaphor, in the sense described above, but be intended only to convey its ordinary pictorial content. In fact, it’s possible to deploy a picture as nothing but an illustration of the literal content of a metaphor. In that case, of course, it’s inappropriate to read it metaphorically. On the other hand, a picture can be used to express the metaphor that it also illustrates. It might not work in every case, because audiences might not find the relevant

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

’     

115

parts of the picture salient, for any number of reasons. The same holds for suggested and even supplemental metaphors. One can use a picture to suggest a metaphor, even if the picture is not intended to convey it. Or one can intend to convey it. If the communicative context cooperates, in those cases the picture expresses a metaphor. The same kind of thing happens with linguistic metaphors. They can fail, of course, and that is typically because the audience is not attuned to the associations the speaker intends them to notice. The previous chapter on iconography shows one way in which semantic mechanisms can bring us well beyond the ordinary contents of pictures. Notably, iconographic conventions can be used to make pictures represent virtues, vices, values, numbers, and anything else one might care to represent. In that sense, the semantic mechanism behind iconography expands the range of contents different uses of pictures can have well beyond the perceptible. The same is the case here. M-association can take many forms, and there is no reason to think that it respects the boundaries of ordinary pictorial content. If something like Stern’s semantic account of metaphor works for pictures, then it shows another way in which pictorial meaning can extend well beyond ordinary pictorial content, in an open-ended fashion. In this case, however, we are considering an expanded range of properties, not individuals. An ever-changing palette of features can be massociated with any given feature, especially as contexts of use change, and in this way the range of metaphorical meaning is unbounded.

6.6 Stern’s Worries about Metaphor in Pictures Section 6.5 suggests that we can apply Stern’s way of thinking about metaphors to what are pretty convincing examples of strictly pictorial metaphors. Why does Stern insist that there are no pictorial metaphors? If metaphor were purely a pragmatic affair, by Stern’s lights, there should be no problem with having pictorial metaphors. Pragmatics deals with general purpose interpretive mechanisms that apply to things people say, do, depict, and so on. So, if the path to a metaphorical interpretation, even in language, were solely a matter of pragmatics, we should expect there to be pictorial metaphors. But Stern’s account is distinctive in suggesting that metaphor has a semantic component—the mthat operation—and he thinks that this semantic component is native to language. So, we can get things that look similar to metaphor with pictures because metaphor is partly

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

116



pragmatic. But there is no truly metaphorical pictorial interpretation because real metaphor involves a semantic mechanism unavailable in pictures. To the extent that Chapters 2–5 have been successful, they diminish any commitment one might have to character and content being native to language. And to the extent that the earlier discussion in this chapter is convincing, character and content can help us think through pictorial metaphor too. Stern’s (1997) worries run deeper than that, though. His paper is not built around his own account of metaphor, so much as showing that other attempts to establish that there are metaphorical pictures fail, for internal reasons. Given that two promising proposals fail, and his own account of metaphor suggests that the phenomenon is fundamentally linguistic, we should be skeptical about the possibility of metaphor in pictures. It’s worth bringing up a couple of the specific criticisms Stern directs at Nelson Goodman (1976), because they seem to carry over to his own reasons for being skeptical about metaphor in pictures. By Goodman’s lights, and on this point many agree with him, pictures do not come with grammatically marked off parts. In fact, the identity of a picture potentially depends in an indefinitely sensitive manner on the details of its surface. Because of this, neither pictures as wholes nor their spatial parts are repeatable types. Pictures can be quite similar to one another, but they are not, like linguistic expressions, the kind of things that allow for perfect copying (Goodman 1976, Ch. 1). Since Goodman insists metaphor concerns the deployment of an expression type in a new domain, he cannot accommodate pictorial metaphor (Stern 1997, 276–8). Goodman might also give us a reason for thinking that character and content are operations native to language. Specifically, character and content are understood to apply to types of expressions deployed in different contexts. So, to make sense of a picture’s character, or content, we would need to know what type of picture it is. And in an interesting sense, we can never know exactly what type any given picture belongs to. This book has endeavored to convince readers that character and content are useful ideas for understanding pictorial interpretation. Another important goal has been to refashion the way we understand parts of pictures. They are not grammatically marked off as playing distinctive roles in determining pictorial meaning. Also, there are indefinitely many ways in which any given picture can be broken up into parts. But the parts of pictures are meaningful, and they can be deployed in many different contexts. When we try to

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



117

interpret pictures charitably, as parts of communicative acts, we can isolate aspects of their contents and decide how we should understand them. If things work in this way, there should be no worry about understanding pictorial metaphor on the model offered here.

6.7 Summary Pictures can be used to express metaphors, even given a semantic account of metaphor. Metaphor has not been the most prominent topic in theoretical or art historical discussion of pictures. Perhaps this is because metaphor has not been as popular a mode of pictorial expression as it has been in language. Some advertising images are used in a way that makes their conversational goals clear enough to help pin down metaphorical uses of pictures. Another reason for using advertising, however, is that it is difficult to find clear-cut examples of pictorial metaphor in the history of art. Language is thick with metaphor, pictorial practice less so. In fact, even in advertising there are not as many examples of pictorial metaphor as one might expect. Clever uses of pictures are quite common, but strictly pictorial metaphors are not. Defending a semantic account of metaphor is well beyond this chapter’s aspirations. Stern’s account is particularly relevant to present purposes because it suggests that metaphors owe their meanings to operations defined over character and content. In fact, it was precisely this account of metaphor, and some thoughts about how one might apply it to pictures, despite Stern’s remarks to the contrary, that got this book project started. If a semantic account of metaphor is the wrong way to go, so be it. Making such an account work with pictures is, as Stern (1997) showed, quite difficult. So, in effect, this chapter shows that in the most difficult case we can still make sense out of pictorial metaphor. If pragmatic theories win the day, well, then we already knew there could be pictorial metaphors, and the phenomenon floats somewhat free of the mechanisms articulated in the earlier chapters of this book.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

7 Direct Reference in Pictures and Maps Pictures, the focus of this book so far, are part of a larger class of representations that includes graphs, diagrams, charts, and, of course, maps. The final chapter (Chapter 8) will offer a new account of what makes this broad class of representations distinct from linguistic representations. Maps merit special attention because they are an especially well-developed kind of representation and they differ from pictures in two enlightening ways. First, maps have constant characters. Deciding what a map is about, attributively, involves nothing more than figuring out which syntactically relevant properties are located at each of its locations. Second, maps isolate individuals spatially. So, maps attribute properties to individuals, and they do so in the following manner. Locations on the map work like names that isolate locations for comment. This important feature of maps shows up in some pictures, too. Security camera footage is associated with specific places. More generally, many ways of understanding photographs, including video, treat their locations, that is, locations on the photos, as fixing reference to places, and, thereby, the things that occupy those places. The discussion thus far has focused on the fact that pictures have descriptive contents, and mechanisms that can model how such contents might be transformed. This chapter, in effect, adds directly referring expressions to the mix, and, I suggest, completes a model of the many ways pictures can have meaning. Maps give up on one very interesting feature of pictures—their nonconstant characters—in order to gain the ability to represent many features of scenes simultaneously and unambiguously. The rule that gets you from a pictorial character to a content appeals to recognizable manifestations of bare bones contents. But recognizable manifestations of things will be perceptually available, and so in an important sense what we can depict is limited by what we can see, hear, and so on. Maps can represent any range of features you care to collect, and they can do so to a high degree of precision. Section 7.1 considers strong intuitions that maps, photos, and videos have objects. Section 7.2 suggests that these intuitions are due to the fact that in all such cases we treat locations as referring terms akin to names. Section 7.3 Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

    

119

raises and dismisses some worries about this proposal. Section 7.4 reviews proposals about map semantics, and then section 7.5 suggests that pictures are actually a simple kind of map. Finally, section 7.6 relates maps to pictures and other kinds of representation, like comics.

7.1 Presence in Photographs and Maps A large literature following Kendall Walton’s (1984) famous photography paper asks whether photos enable indirect perception of the scenes they are about.¹ What animates that discussion is the strong sense that photographs somehow just have objects. This seems true even when they are blurry or poorly composed. Photos don’t seem to make existential claims, like “There is something that is like this.” Instead, they say, of something, that it is thus and so. Photos, that is, work like linguistic expressions that isolate an individual, perhaps by using a name, and say something about that individual. Interestingly, such intuitions do not readily extend to handmade sketches or paintings, though there are some exceptions to this rule (Lopes 2009). The sense of presence evoked by photos only increases when they are videos, and in particular videos that are displayed in real time. Security cameras produce live images on a guard’s computer screen. The Big Game is broadcast to the local pub. We know, independently, that there are individuals that these moving images are about, but more importantly the representations elicit a sense of that. Something similar, but more attenuated, happens when you find an old family photo in a shoe box. Mechanically produced pictures stand the greatest chance of evoking these feelings, and philosophers have speculated as to why this might be the case.² Maps have not been discussed in this connection, but they elicit a sense of individuals at least as strong as what we find with mechanically produced pictures. This is so despite the fact that maps generally seem handmade, and they are almost never photo-realistic. Imagine finding a map sandwiched between the pages of an old book. What place is this? Yes, maps can be traces of nothing but overactive imaginations, just as some invented pictures can ¹ See Kulvicki (2014 Ch. 9, forthcoming) for both an overview and my own critical take on that debate. ² André Bazin (1971) might be an important point at which this kind of thinking crystallized (see Friday 2005), which in turn influenced Sontag (1977), Barthes (1984), Walton (1984), and others.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

120

     

look like photos. But neither possibility takes away from the sense that some individual or other is importantly related to these representations. Likewise, both photos and maps can be wildly incorrect or lack detail and still evoke feelings of presence. These intuitions about photos and maps thus float free of their attributive functions. While the intuition about photographs has launched a thousand articles, no one seems to have mentioned maps in this connection. Perhaps the intuition is not as universally robust as it is for photos, so consider this an invitation to think about how you feel. Informal surveys suggest a mix, though a common response is that it is something folks have not considered. In sum, photos and maps seem to include objects as parts of their contents. This sense does not depend on their attributive contents. Perhaps, then, these representations involve directly referring expressions. If so, how? A longstanding suggestion about maps provides an answer.

7.2 How Objects are Involved In some cases, locations on a map or picture are treated as referring terms.³ Locations don’t refer in the denotative way descriptions do. That is, they do not pick out something because they accurately describe it. Instead, they refer directly, and their contribution to the content of a picture or map is a set of locations. Other features, like the colors at those locations, attribute properties to the represented locations. In earlier chapters we saw that some parts of pictures can isolate subjects while others make comments about them. Doing so required highly conventionalized practices—vacation photos, photo rosters—or highly specific contexts of use, because pictures are not grammatically complex. Practices and context, not grammar, let one part of a picture do one job while some other part does another. The same is true here. Locations are parts just as colors at those locations are. In photos and maps, typically, and in other contexts, occasionally, the former isolate individuals, while the latter comment on them. Locations are parts of pictures in the same way that other aspects like colors and shapes are. We find them by abstracting over syntactic detail. ³ Casati and Varzi (1999, Ch. 11) suggest that the locations on maps work like names to isolate individuals. Kitcher and Varzi (2000), Michael Rescorla (2009), Elisabeth Camp (2007, 2018), and I (Kulvicki 2015b) all follow their lead. See Blumson (2010) for worries about this.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

121

Ignore the scarlet color that occupies some region and focus instead on the fact that it is red. Likewise, ignore all chromatic features at a location and one is left with a bare location. The location works attributively, in the sense that it isolates some place, distinct from the others isolated by the picture. In photos and maps, locations also refer directly to locations.⁴ This suggestion fits the phenomena quite well. We treat the live video of the game as saying what’s going on, at a place, now. It can be blurry or otherwise inaccurate, but nevertheless tied to a set of locations at some time. The photo connects to some place, presumably in the past. A map shows the layout of a city. Its locations stand for locations, and its other features tell us what those locations are like. In this way, maps, photos, video feeds, and the like seem to express propositions. It might not always be easy to express that proposition in language, and often one might not know, for example, which locations are singled out. That does not interfere with understanding the photo to be saying of some place that it is thus and so. The picture uses names, as it were, but we are not always familiar with the names being used. Strictly analogous things happen in language when unfamiliar names are deployed. Harry is wearing a gray three-piece suit, pink shirt and tie, and a fedora. Yes, but who is Harry? Locations in pictures are sometimes treated as directly referring terms, but they do not need to be used that way. In maps, locations are always used that way, and we are more prone to understand photos in this manner than handmade pictures. This interpretive move, however, is available for any kind of picture. When pictures fail to do this, they might still express propositions. Earlier chapters give examples. Why are we more prone to think of photos in this way than handmade images? The natural thought is that photos have causal sources. Cameras were pointed at locations, so there is a sense in which it makes sense to say that pictures are guaranteed, in some sense, to have objects.⁵ The earlier discussion of photos and maps, however, does not mention causal sources. Instead, locations are treated as directly referring terms. These ideas relate. Being the right kind of causal source for something is one way in which

⁴ Camp worries about Casati and Varzi’s proposal about names on the grounds that “maps typically lack marks that function just to denote locations” (2018, 28) and it’s hard to see what the syntactic difference would be between a map that does and one that does not denote would be. My suggestion is that the location-denoting markers are the bare locations, which are parts of a map, given the understanding of parts we have been working with since Chapter 3. ⁵ Patrick Maynard (1997) stresses the way in which photos are traces of their sources. See also Dawn Wilson’s (Phillips 2009) helpful discussion of this issue.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

122

     

people have speculated directly referring expressions to get their referents.⁶ So, in one’s meta-semantic view, which should explain how directly referring terms refer, one might make use of causation, among other possible explanatory tools. For now, meta-semantic debates can be left aside. It’s likely that intuitions about photos having objects are based partly on their causal relations to those sources. Semantically speaking, however, such cases suggest that photos make use of directly referring terms. Maps are made quite differently, and one might think that their locations get their referents in a manner distinct from photos.

7.3 Two Worries about Locations as Names It’s plausible that the different locations on a map indicate different locations within a scene, but two features make locations seem quite unlike ordinary names in language. First, they seem to have no independent syntactic identity. “Gertrude” is a different name than “Susan.” How might we distinguish map and picture names from one another, when in each case they are just locations, otherwise featureless? Second, even though locations are names, one is not free to assemble them in any way one likes when making a map or picture. Which part of a scene is named by a given location is not independent of what the other locations do. A small region here, for example, situates a location with respect to the others that it names. Nothing at all like this happens with names in language. These two points are related. It seems as though the only thing distinguishing one location name from another is its place relative to those others, and it seems like the constraints on what a name can be about relate only to its presence among those others. Those constraints come in many sorts. Some of them preserve metric structure of what is represented, some just preserve topological structure, and some do something in between (Casati and Varzi 1999, Ch. 11). First note that even in language we do not have an obvious way to distinguish all names from one another. There are many Gertrudes, for example, and it’s a matter of controversy whether they all, in some sense, have the same name or whether they all have different names that ⁶ See, for example, Kripke (1980), Devitt (1981), Evans (1982), and Dickie (2011) for views that make use of cause as at least part of an explanation for the reference of names, and see Reimer and Michaelson (2014) for an overview and critical discussion.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

123

nevertheless sound and look alike (see Kaplan 1990, e.g.). Without resolving this general issue here, I’ll just point out that it’s not a special problem for the proposal about pictures. Second, it seems there are structural constraints on what a given location of a map or picture can represent, given what some other location represents. That does not interfere with the point that these locations refer directly. It just suggests that names in pictures come in groups, and that the reference of each member is decided in a holistic fashion. It’s not in virtue of describing some part of space that a location on a map refers to it, even though which parts of space each location refers to are globally constrained.

7.4 Absence and Map Semantics Maps tell you what things are like at various places. A street crosses a river, then heads into woods ringed by mountains. Whether an area is represented as water, swamp, mountains, plane, or canyon depends on the features placed at the corresponding locations on the map. A certain shade of blue indicates shallow water, another shade deep water. Green is for forests, brown for scrubland. Maps thus seem to place features at locations, and those features say what the locations are like. To know what a map is about, one must know which features are at which locations, and how the space on the map correlates with space in the world. Things are not quite as simple as the foregoing suggests, and it is easy to show why. When the feature for representing water is at a given location on the map, it says that the corresponding part of space is watery. But what should we say about regions of the map that do not have the feature for water? On the standard way of interpreting maps, those regions are represented as not having water there. Features should be placed at all locations that represent places with the corresponding property, and not elsewhere. Failure to do this results in a faulty map. Michael Rescorla (2009) calls this the absence intuition. Much of the recent literature on maps concerns how best to handle the absence intuition. Rescorla adopts Casati and Varzi’s (1999) semantics for maps. Colors are placed at locations. The presence of a color at the location indicates that the represented location has the feature represented by the color. Blue for water, green for land. Absence of a color at a location indicates the absence of the represented feature at that location. Now, saying

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

124

     

Marge is a doctor in no way commits you to saying anyone else is or is not a doctor. In general, failing to predicate a feature of someone does not eo ipso represent that person as not having the feature. So, by Rescorla’s lights, it cannot be predication at work in maps, but something else. Linguistic representations predicate features of objects, while maps suggest the presence and absence of features in a more holistic fashion. Others, like Camp (2007), Blumson (2012), and Bronner (2015) account for the absence norm pragmatically, not semantically. Users expect certain things from maps, namely, that they predicate the presence of water of every region that has it, as long as that region is represented by the map. People also expect, however, that when someone says she has two apples that she doesn’t have eighteen. There is little need to insist that not having eighteen apples is part of what “I have two apples” means. The claim about apples is true just in case she has two, irrespective of whether she has more, even though, in saying what she does, she can communicate that she has exactly two. My own view preserves predication, but also insists that the phenomenon is part of map semantics, not pragmatics (Kulvicki 2015b).⁷ The features we place on a map are organized into incompatibility classes. The members of an incompatibility class are such that no two of them can be put at the same place on a map. The presence of one excludes all of the others. So, for example, maps often have markers for cities of different sizes. No two of these can be put at the same place on the map. Colors for land, marsh, and sea are typically incompatible, in that they may not be placed at the same location. Maps can have many incompatibility classes. Features within such a class cannot occupy the same place, but any two features in different classes can. You get into trouble if the members of an incompatibility class represent perfectly compatible features of a scene. Roads cover land, but if the marker for a road excludes the one for land, the map could not say of a given region that it has a road, and is land. So, one constraint on good maps is that incompatible features represent incompatible properties, while compatible features represent compatible ones.

⁷ My way of doing things does not preserve the Absence intuition, not fully. It shows why it is generally true, in virtue of map semantics, even though in some cases, which are for many reasons bad maps, the absence intuition is violated. This is not directly relevant to current concerns, but see Kulvicki (2015b, §5) for a detailed discussion.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

125

Finally, a map must commit to putting some member of each incompatibility class at each location. Sometimes, as with city markers, this is a null value, in that its presence does not make a noticeable difference to the map’s surface. In other cases, the incompatibility class is such that some visible marker corresponds to each member. In effect, the number of incompatibility classes gives you a sense for how many things a map can say about any given region. Five incompatibility classes, and you are indicating five things about each point on the map. The overall picture of maps, then, involves dimensions. Maps take a space, and, at each location in the space, commit to things being a certain way along a number of dimensions, each one determined by an incompatibility class. Maps vary with respect to the number of incompatibility classes they have. A typical example might involve the following: Land–Sea–Marsh–etc. Topographical features: plains, hills, mountains, etc. Roads: Interstate, county, local, rail Cities: Huge, big, medium, small

Some of these classes have empty null values. For example, the class of markers for cities is such that you can leave out a marker, which is, in effect, saying that there is no huge, big, medium, or small city there. Ditto for roads. The idea is that each class is such that it fully covers the range of options. Every part of a map must have an interstate, county, or local road, or none of the preceding. Sometimes, classes don’t need empty null values. Land and sea, or land–sea–marsh, can cover the range of options, as long as you are willing to tolerate forcing some borderline cases. Each choice of incompatibility class involves design challenges. Remember that any features in different incompatibility classes are compatible with one another, and there must thus be a way of understanding the layering of features from each: An interstate cuts through a huge city on hilly land. Easy enough, given current mapmaking practice. The design challenge is independent, as it were, of the logic of these kinds of representation. In the abstract, maps can layer as many features as you like (Casati and Varzi 1999, Ch. 11), but practice puts interpretable millefeuille maps out of reach. The range of features an incompatibility class can represent is openended. It’s typical for maps to represent roads, lakes, mountains, and the like, but they are just as capable of representing incomes, rents, laws, natural

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

126

     

resources, kinds of pets, climate, growing zones, and so on. Pretty much anything that can be distributed across space can be mapped. Doing things this way manages to preserve a lot of intuitions about absence. The reason not putting blue somewhere means that there is no water is because instead of blue you had to put another color, say, one for land, or marsh. By saying something incompatible with the presence of sea, the map by extension indicates a lack of sea. According to Casati and Varzi’s (1999) semantic approach to maps and absence, the reason not putting blue somewhere means a lack of water is because of a special way colors work in maps. That means, in turn, that we cannot think of maps as making use of predication in a manner akin to language (Rescorla 2009). By insisting that maps organize their syntactic features into incompatibility classes, we can preserve the thought that maps predicate features of places (Kulvicki 2015b). Having these accounts on the table will help us with three things. First, it helps show how maps have, and should be expected to have, constant characters. This makes them quite different from pictures. Second, it will help us see how pictures might be, in another sense, a simplified kind of map. And third, this proposal will fit maps neatly into the account of what makes non-propositional representation distinctive. We won’t get to that last point until Chapter 8.

7.5 Why Maps Have Constant Characters Maps have varying numbers of incompatibility classes, and they specify exactly what feature each member of an incompatibility class represents. Cities that have between 50k and 150k inhabitants get one symbol, those with between 150k and 500k get another. These properties are not specific in the sense that they include a determinate number of individuals, but in the sense that they involve a clearly delineated range. In other contexts, the features represented need not be matters of degree. Interstate roads differ from county roads, but there are no degrees of being either kind of road. One of the virtues of maps is that they can precisely specify just what properties different aspects of the map represent. Precisely specifying just what something represents is exactly what we can do with representations that have constant character. When this feature shows up, the map represents that. One might think that representing something specific isn’t logically incompatible with having a non-constant character. Why not design maps such that in one set of circumstances they represent one specific feature,

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

       

127

while in another they represent something else? Logically speaking, this is manageable. We could design maps with non-constant characters, but we won’t, for two reasons. First, while one virtue of maps is the way in which they can specify what properties map features represent, another is the fact that they can layer such features indefinitely. If we are to design a map whose features pick out different things in different contexts, and which also uses many incompatibility classes of features, we will quickly wind up with uninterpretable maps. It’s hard enough to read complicated maps with constant character. And all maps are expected to come with legends that specify the significance of each feature. Making the significance of more than one feature depend on context, in any but the simplest fashion, would likely make the map useless. Second, remember the rule that gets us from bare bones content to pictorial content. It leans heavily on ordinary perception. Find the appropriate recognizable manifestation of bare bones content. No such rule could be in play with maps. They can use incompatibility classes to represent all kind of features, many of which are imperceptible, and they can represent many such features all at once. So, while maps are perceptual in the sense that they exploit the way we see to make their contents available to us, they are not perceptual in the sense that their contents depend on what we can readily recognize in the wild. Of course, some other rule could be the one that allows pictures to have non-constant characters, but that rule would run into the first worry just mentioned. So, while maps share a lot of features with pictures, they are not pictures. They have constant characters, which means that there is no interesting notion of the bare bones content of a map, or a rule that brings you from it to a map content. Maps, in a certain sense, wear their contents on their sleeves, and if you are unsure about what some aspect of a map means, you just need to consult the legend. This move away from pictures is an enormous boon, because we now have representations that can represent indefinitely many ranges of features, all at once, while also enjoying some of the benefits of how pictures present information spatially (see Camp 2007).

7.6 The Path from Pictures to Comics to Maps Now let’s return to pictures, and try to think of them in light of what has been said about maps. Assume that locations play a role identifying subjects, and that the features placed at those locations say something or other about

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

128

     

them. Using maps as a model, two questions come up. How many incompatibility classes do we find in pictures, and what are they responsible for representing? These questions have easy, but revealing answers. Pictures typically have only one incompatibility class, and those features specify aspects of bare bones content. In effect, pictures are maps of features that figure in their bare bones contents. Some pictures, like those in comics, add one or more incompatibility classes to the mix, as we will see. Which marks on a picture surface are syntactically significant? The answer depends on the kind of picture. For color photographs, there are many options, but they are very simply organized. Put some color or other there. It is not possible to leave out the colors. Doing so would be tantamount to leaving a hole in the picture. And no two colors can be at the same place. This isn’t to say that a color photograph couldn’t represent a colored thing behind colored glass. It’s just that when you are choosing to place features at locations all you can do is put one color at each location. And because you have to put some color or other, it looks like such pictures work like maps. They have to commit to some value along the color dimension for each location that counts as part of the picture. None of the foregoing has anything to do with the fact that photographs are mechanically produced. Handmade full color paintings work in much the same way. You can leave the canvas blank, but unless viewers are meant to interpret the blank canvas as white, the blank part is basically a hole in the picture. Not all pictures are color photos or paintings, of course, but the rest work in similar ways. Line drawings are built out of lines, so you can mark part of the picture as having a line, or you can leave it blank. In this case, we have one incompatibility class: you put a line there or not. So, failing to commit to some value within this class amounts to cutting a hole in the picture. Otherwise, the blank part is read that way, just as lines are read as lines. Some pictures work with volumes of light and dark, rather than lines per se. In these cases, we can imagine a gradient from black to white as the incompatibility class. Each part of the picture commits to some value(s) or other along this dimension. Failure to commit constitutes a break in the picture. In yet other cases, there is a mix of line, volume, and color. Mixed cases can work in a number of ways. The goal here is not to unpack all of them, but mentioning some can help us get a feel for the phenomenon. In some pictures, obvious lines mark off the boundaries of objects, while colors fill those boundaries in. The colors can, but need not, be shaded to indicate features of light. It’s plausible in such cases that (line, no-line) is one

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



129

incompatibility class, layered with another for the colors of volumes. This accounts for the fact that we tend to interpret a dark region constituted by intersecting lines differently than a similarly dark region in the center of a volume. In many comics, special marks indicating motion, speech, and so on, can peacefully cohabit with the marks responsible for the most basic depictive content. Failure to place a motion mark at a location, indicates a lack of that kind of motion, but the motion marks are compatible with the regions where they are put being any color you like. That is, even though the motion marks make some parts of the picture black, they are not indicating that there is a black surface, or an object boundary there. What the last few paragraphs have done is trace a path through a number of kinds of pictures that shows two things. First, it seems as though pictures make use of incompatibility classes. They have features that must be placed at locations, and pictures must commit to values for each such class. They are, in that sense, a lot like maps. Second, even in pictures, we can sometimes see a layering of more than one incompatibility class. This is especially salient in comics. Now, what makes pictures pull apart from maps is the fact that one or more of these incompatibility classes is devoted to representing bare bones content, not pictorial content. Patterns of object-boundary lines and fill colors establish a bare bones content, which can be interpreted pictorially in a number of ways. The layered motion marks are more map-like, in that they seem to have constant character. Lines like this, attached to one side of an object, indicate this kind of motion. Those fixed-character features can help with interpreting the pictorial part. For example, since nothing moves in a flat pattern of pigment, the presence of many motion marks makes it much more plausible that the picture should be interpreted as representing people instead of pictures of people.

7.7 Summary Maps are important for this book in three ways. First, they have long been understood to involve directly referring parts. So far, this book has said nothing about whether pictures involve such parts. Adding such spatial reference to the mix accounts for a number of ways in which some pictures, especially photos and video, seem to represent things. The fact that this model works for pictures is especially interesting in light of the fact that maps seem to evoke much the same sense of having objects that photos do.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

130

     

So, the semantic connection between photos and maps seems tracked by the ways in which we understand them. Second, maps have constant character. Pictures are distinctive in part because of how they have non-constant character. Some representations, like comics, are hybrids between pictures and maps that include parts with constant character alongside those with non-constant character. Third, maps and pictures are members of a larger class of representations. What makes this class distinctive is that they all seem to have syntactic parts in the same way. In an important structural sense, pictures are just like maps, albeit rather simple examples of them. Given that, we now have the tools for making a surprisingly simple suggestion about the distinction between these kinds of representation and language. This distinction cuts deep. And it’s interesting, in light of it, that pictures, maps, and the like can be semantically modeled using the same tools that work so well for language.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

8 Distinguishing Kinds by Parts Pictures are syntactically complex. Because of that, different parts of pictures can, under the right circumstances, play different semantic roles when pictures are used communicatively. Maps have parts in the same way pictures do, and in both cases this is different from linguistic representation. This chapter proposes that the distinction between linguistic and imagistic, propositional and non-propositional representation comes down to how they have syntactic parts. Linguistic representations are such that all of their syntactic parts are separable. Imagistic or nonpropositional representations are such that none of their syntactic parts are separable. The challenge is explaining what separability is, and then arguing that the rather extreme all-none claims hold for each kind of representation. Separability is a new notion, though, like any idea, it has ancestors.¹ When language-minded philosophers have turned their attention to pictures and other non-linguistic representations, they looked for separable syntactic parts. They found some good candidates, too. One goal of this chapter is to move away from parts talk that seems, on the surface at least, plausible and much more in line with the linguistic case than the present proposal says it is. The chapter begins by clearing ground, reviewing what it means to talk about syntax, what a separable part is, and how philosophers have tried to force this notion onto pictures. Worries about compositionality turn out to be involved here. After that, the notion of an inseparable part is introduced, and with it the claim that all syntactic parts of imagistic representations are inseparable. This will reveal how questions of compositionality look different when asked of pictures, and why it makes good sense to say that pictures and their ken are non-propositional.

¹ As we will see, John Haugeland (1998) came closest to articulating the point I make below.

Modeling the Meanings of Pictures: Depiction and the philosophy of language. John Kulvicki, Oxford University Press (2020). © John Kulvicki. DOI: 10.1093/oso/9780198847472.001.0001

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

132

   

8.1 Syntax and Compositionality It’s somewhat rare for philosophers working on pictures to use the notion of syntax at all, but when they do they focus on which features of a representation can make a difference semantically.² For example, one might say that some feature is syntactically relevant to a representation because representations that differ with respect to it can differ in meaning. In some pictures, hue does not matter, while in others it does. When it does, changing a picture’s hue can change what it means. The same is true of shapes, and any other candidate for being a syntactic feature. Thought of this way, syntactically relevant features are a special subset of a representation’s features overall. The mass of a photo is completely incidental to what it represents, as are the colors on its backside (Kulvicki 2014, Ch. 4). Chapter 3 suggests that instantiations of such features deserve to be called the syntactic parts of pictures. Syntactic because they are relevant to what pictures represent. Parts because they make their own contributions to what whole pictures represent. What makes languages interesting as representational systems is that words can be combined into grammatically structured expressions. In pictures, by contrast, there are no noun phrases, relative clauses or verbs. This leaves one wondering how the syntactic parts of a picture conspire to form a whole. If there are no interesting grammatical operations, what is it that makes pictures whole? Or, put differently, what are the syntactic operations defined over pictures? Chapter 3 suggests that there is only one operation of interest: abstraction over detail. This chapter takes that point and unpacks it. The end result is an account of picture parts that reveals them to be structured fundamentally differently than linguistic syntactic parts. The best way into see the contrast between image parts and language parts is to think about compositionality. Since syntactic parts are supposed to be the things that, together, make an expression whole, it is common for philosophers to suggest that languages are compositional. Barbara Partee unpacks what philosophers of language and mind mean when they say a representational system is compositional: The meaning of an expression is a function of the meanings of its parts and of the way in which they are syntactically combined. (Partee 2004, 153) ² See, e.g., Goodman (1976), Bach (1970), Stern (1997), Perini (2005), Kulvicki (2006), and Greenberg (2013).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

133

This definition does not require that the expression be a bit of language (see Szabó 2017 for a brief discussion of road signs), though languages are its main target. Key here is that some expressions are complex, in that they are built out of other expressions that can be meaningful. Some expressions are simple in that they are meaningful, but have no meaningful parts. Parts are meaningful on their own, and those meanings, plus rules based on the ways in which they are combined, yield the meanings of complex expressions. “John loves Mary” and “Mary loves John” have the same basic constituents but they are combined differently in each case, resulting in complex expressions with different meanings. A representational system is compositional when the meanings of its complex expressions are determined by the meanings of their parts and the ways in which they are combined syntactically. Parts of a linguistic expression are those things one combines to make complex wholes. They can occur in many different expressions, often playing different roles in the many expressions they constitute. It’s natural to formulate questions about whether pictures are compositional by using language as a model. This has motivated others to look for syntactic parts of pictures that resemble the syntactic parts of language. That promising idea turns out to be problematic.

8.2 Separable Syntactic Parts Remember that syntactic parts are the kinds of things over which syntactic operations are defined. In language, all the operations are such that they work with separable parts. Building blocks and puzzle pieces are a nice model for what I mean by separable parts. They can be independently handled, and then assembled into many different wholes. The wood block keeps its shape whether it is part of a wall, a car, a letter, or whatever else you make out of it. It cannot be combined into a stable structure with any other block, but as long as the other blocks are of the right shape, they can conspire to make a whole. So, a syntactic part is separable just in case (a) it keeps its syntactic identity across the many well-formed expressions in which it can occur, and (b) it does not demand the presence of any other specific part in those well-formed expressions. This seems to capture what Elisabeth Camp has in mind when she says that operations over syntactic parts in language are digital and universal (Camp 2018, 25). Separable syntactic parts are meaningful, and

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

134

   

they are combined via a kind of structured assembly. Any given part can occur in indefinitely many structured assemblies, and being a part is precisely being one of those meaningful things that can be so assembled. When a putative syntactic part seems to demand the presence of another, as “The” does in “The Bronx,” the ensemble is regarded as an indivisible whole. That is, it doesn’t seem like “the” in “The Bronx” has the same syntactic identity as “the” in “The birthplace of hip-hop.”³ Export this linguistic thinking to the pictorial case, and the result is the Sober (1976) line of thought about picture parts, which was discussed in some detail in Chapter 3. What could a part of a picture be, after all, but the kind of thing that can figure in many other pictures and keep its syntactic identity across those many uses? Cut up a bunch of pictures, and the pieces can be mixed and matched to form new ones. The pieces are just the same as they were when they were parts of other pictures. So, pictures’ syntactic parts are, like those in language, separable parts. We also see this line of thinking at work in recent discussions of maps (Camp 2007, 2018, Rescorla 2009, Blumson 2012). There, marks that indicate cities and towns, mountains, and the like, are treated like separable elements that can be placed anywhere, moved around, and so on. The Sober line of thought is quite powerful, in that it suggests at least three convincing points of contrast between pictures and language. First, being a picture is, to borrow a term from Nelson Goodman (1951, 38), a dissective property, in that their parts are also representations, of the same sort as the whole is. In a fairly idealized sense, the spatial parts of pictures are pictures in their own right. Linguistic representations do not work this way. A part of a sentence need not be a sentence, parts of noun phrases need not be noun phrases, and so on. Second, there is only one rule for combining parts, connection, which makes it plausible that pictures lack an interesting grammar. Make the border of one part congruent with the border of another and the result is another picture. The new picture is, in a way, a conjunction of the earlier two (Sober 1976). There is no rule of overlay. Something like overlay is important to understanding maps, but it is not overlay of spatially disjoint parts of

³ In claiming that separable parts keep their syntactic identities, I am not also insisting that they are semantically alike across all the expressions they constitute. That is a matter of controversy beyond the scope of this chapter. Fodor (1998a,b), Robbins (2005), and Johnson (2006).

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

135

maps. As with pictures, one can break a map into parts, by cutting it up, reassemble the pieces and wind up with a new map. Third, any spatial part of a picture is a part. There is no privileged way to find its constituents. These three features play well together. Without a special way of breaking up pictures, it’s hard to imagine how to implement different ways of combining the parts. Camp (2018, 25) suggests that, in addition to being digital and universal, propositional systems of representation are asymmetric and recursive. This means that, inter alia, the system comes with more than one type of part, that there are rules for which kinds of part can combine with which other kinds, and that such procedures can be applied recursively, to form more and more complicated representations. By contrast, she claims maps and pictures are structurally “flat,” which both seems true and fits well with the Sober line of thought. Because the Sober proposal makes pictures seem so different from language, one might think that it provides an account of the distinction between these representational families. There is thus good reason for this line of thought animating the imagery debate from the late 1970s until the early 1990s. Section 8.3 challenges the view that finding the syntactic parts of pictures amounts to finding separable parts. This is just another way of expressing the main question motivating this chapter. What is it that brings the many syntactic parts of a picture together to form a whole?

8.3 Inseparable Syntactic Parts Chapter 3 presented a Parts Principle according to which abstractions over a representation’s syntactic features are representations of abstractions over the original representation’s content. Ignore the hue of a picture surface, focusing instead on patterns of light and dark, and the result is also a representation, though one with a different content than the original. The content is different in that it is an abstraction over the first. It says nothing about hue, but still says quite a bit about patterns of light and dark in space. Similar syntactic operations—leave out the high-frequency spatial detail of the picture, focus exclusively on its hue, excluding saturation and brightness, and so on—also result in (syntactically and semantically diminished) representations. Admittedly, it’s a bit odd to say that an abstraction over the syntactically relevant features of a representation is itself a representation. Abstraction is

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

136

   

not a procedure that produces any new objects in the ordinary sense. Even though abstraction doesn’t yield distinct objects, it can yield different representations because the result has a different set of syntactically relevant features than the original, and those remaining features have their own semantic jobs to do. In the original, let’s say, all aspects of color matter, while in the abstracted one hue is irrelevant. Why think that giving a formula for making new representations out of a given one is helpful for understanding its parts? After all, a rule for transforming an inscription of the word “green” into an inscription of the word “red” would not suggest that “red” is in any sense a syntactic part of “green.” Abstracting over syntactic detail yields syntactic parts because the abstract features—red instead of scarlet, quadrilateral instead of square—play their own roles in determining the content of the representation one started with. Pictures and maps represent qualities across levels of abstraction. For example, a picture represents a scarlet surface, while at the same time representing the surface as red and many other shades of color less specific than scarlet. Simplifying a bit, red is responsible for the picture representing things as red. Scarlet, is responsible for representing things as scarlet. Pictures and maps have “vertically articulate” contents (Kulvicki 2007). As such, all of these more abstract features are syntactic, and thus deserve to be called syntactic parts. The Parts Principle suggests, therefore, that some syntactic parts of pictures and related kinds of representation are inseparable. They are parts, because they make their own contribution to the meaning of the whole expression, but they are not the kinds of parts that can occur independently of whether any other specific part is present. You can’t peel the red off of scarlet. So, insofar as a picture is scarlet, it represents its object as being scarlet, and insofar as it is red, it represents red, but it’s hopeless to try and separate the scarlet from redness. In this sense, being red is not a separable syntactic part of the whole representation.

8.4 The Main Claim With separable and inseparable syntactic parts in view, the main claim of this chapter is easy to state, even if it will be a challenge to defend: Imagistic representations only have inseparable syntactic parts, while linguistic representations only have separable syntactic parts.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

     

137

One of the most fundamental distinctions between kinds of representation relates to how they have parts. Because parts are parts in virtue of the operations defined over them, one could also say that these two representational families differ in the syntactic operations defined for them. In languages, grammatical operations demand separable parts, while in images, maps, and so on, the operation is abstraction over determinate syntactic features. This operation demands inseparable parts. This also, by the way, answers the other question that animates this chapter, though it will only become clear later on. The syntactic parts of a picture don’t need to be held together by anything. They are all inseparably bound to one another. So far, no argument has been given for the main claim. It is bound to strike readers as unnecessarily strong. Why not just say that pictures are distinctive because they have some inseparable syntactic parts? There are surprisingly compelling reasons for the strong claim, as section 8.5 argues. With a defensible account in hand, questions about compositionality and propositionality look different. As will become clear, it is trivial that imagistic representations are compositional, and there is an interesting sense in which they are non-propositional, even though they can sometimes express propositions.

8.5 Three Objections to the Main Claim Recall the exemplar of an inseparable syntactic part. Nothing can be scarlet without also being red. Both being scarlet and being red have their own semantic roles to play in determining a picture’s content. So, they both count as syntactic parts, albeit inseparable ones.⁴ There are three examples of what appear to be separable parts of pictures, maps, and the like. First, cut up a picture as Sober suggests, and the result is new pictures. Those spatial parts can be combined with other picture parts in any way one likes, so those parts are separable. This is as close to the building block analogy as one can get, and it applies to both pictures and maps. Pictures and maps might have some inseparable parts, but they also have separable ones, or so the objection goes.

⁴ We can certainly imagine kinds of representation in which being red is syntactically important but being scarlet is not. And we can imagine cases in which scarlet matters but not red. In pictures, however, we find that that each one plays its own role in determining content. So, in the pictorial case, both are parts, albeit inseparable ones.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

138

   

Second, colors are distinct from the locations they occupy. Some color or other needs to be at each location in a picture or map, but the color’s location plays a different role in determining content than the color itself does. Chapter 7 suggested, too, that locations can play the role of referring terms, while colors and the like play an attributive role. So, this seems like another, fairly basic, example of separable syntactic parts. Third, Casati and Varzi (1999) offer a map semantics in which different colors are given distinct semantic values and can thus be evaluated independently of one another. One can layer as many colors as one likes at any given location on a map. On my view (Kulvicki 2015b), maps have incompatibility classes of features. One of the important things about such classes is that they can be added to or taken away from a map. Some maps tell you about cities, some do not. Both proposals yield fully compositional semantics for maps. So, it seems as though colors, in Casati and Varzi’s sense, or incompatibility classes of features, in mine, are separable syntactic parts. These cases suggest that the main claim is false. While pictures might have some inseparable parts, they also have separable ones. Separability might therefore be less theoretically interesting than advertised. Despite appearances, these cases all collapse to the exemplar of inseparability found with scarlet and red.

8.6 Why the Objections Fail Section 3.3 voiced a worry about the Parts Principle. Say that a picture, at a given location, is scarlet. That part of the picture is red, but it is also redor-a-wombat. This might suggest that the picture represents the corresponding part of the scene as being red-or-a-wombat, which is deeply implausible. One avoids this conclusion by insisting that abstractions are constructed out of an image’s most determinate possible syntactic features. The picture can say that a scene is scarlet, because the picture can be scarlet. It can say a scene is red because, in addition to scarlet, the system to which the picture belongs can represent the full range of determinate colors that count as red. But it cannot say anything about wombats because there is no feature of the picture that represents wombats. The picture can represent a specific shape, but not that shape or a magnetic potential.⁵ Abstractions are built out of the ⁵ A graph (or map) of magnetic fields, which somehow also represents color properties of a scene, could be used in this way, but that’s not what pictures do.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

139

most specific things that a picture can say, which correspond to the most determinate syntactic features a picture can have. In this very important sense, imagistic representations are holistic. Schematically, the responses to all three objections in section 8.5 go like this. The abstractions that matter are those built from the most determinate syntactic features of representations. But the most determinate syntactic features are typically much more specific than one might have thought. It’s not being a specific shade of color, but being such a shade, at such and such a location, in such and such an ensemble of points. Likewise, on a map, it’s not a specific color, but a specific combination of specific features, at a specific location, relative to all the others on the map, that is the most determinate syntactic feature. This renders all of the putatively separable syntactic parts inseparable. The process of generating the exemplar of inseparable syntactic parts was simple. Find the most determinate chromatic features that can characterize the surface of a picture. Then ask whether less determinate features, understood as abstractions over those, also carry semantic weight. When the answer is positive, instances of those features are working as syntactic parts. These parts are inseparable because anything that has the highly determinate property is also, for example, red. Now apply that reasoning to locations. Locations are syntactically relevant features of pictures and maps. Where some color shows up is quite important to what the picture or map represents. Move all the colors around and you get a different picture. So, what is the most determinate location property on a picture surface? A location is syntactically significant to the extent that it is distinguished from and spatially related to all the other locations on the picture surface. The picture says nothing about locations that have no points corresponding to them on its surface. This is not a claim about the grain of the picture, either syntactically or semantically. Whatever the grain, the smallest syntactically significant spatial part identifies a location vis-à-vis all and only the other spatial parts of the picture. When a picture is cut in two, each half says nothing spatial about the other one. Locations on the new part no longer have anything to do with the rest of the picture from which they were excised. Semantically, each new picture is different in that it speaks of a smaller range of locations, but it has different syntactic qualities too. Locations on the surface are now part of a smaller ensemble, so they are distinct from and related to a mere subset of the points that they used to be distinguished from. The most determinate spatial feature on the cut-off part is less specific than the most determinate

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

140

   

features on the whole picture (cf. Heck 2007, 127–8). So, when a piece is cut from a picture or map, the syntactic properties of both parts change. Consider the simple example in Figure 8.1. The Sober line of thought is that the smaller rectangle in the upper right is a separable part of the whole. It keeps its syntactic identity when separated, and can be used, like a building block, to make new pictures. But when this part is attached to the whole it has a different syntactic as well as semantic identity than when it is on its own. Consider the point C in the upper right. And remember that the syntactic operation of interest for pictures is abstraction from determinate detail. When the upper rectangle is part of the whole, point C is also point C-orA, C-or-A-or-B, and so on. So, if the picture says things are red at point C, then it also says that things are red at point C-or-A, C-or-B, C-or-D, and so on. The abstractions over syntactic detail that matter are those defined over the most specific features the representation has. In this case, the point C is distinguished from all the other points in the whole array, and so abstractions like C-or-B matter to the content of the whole. But when the upper rectangle is excised, it says nothing about any of the points outside of the excised part. So, when excised, the point C is no longer C-or-B in a way that matters for content. As a result, the point C is syntactically distinct from what it was before excision. A similar claim applies to all the points in the extracted part of the picture, as it does for the picture from which it is

Figure 8.1 Syntactic inseparability. The author.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

141

extracted. Remove a spatial part of a picture, and its syntactic identity changes. In that sense, such parts are inseparable. Arguably, examples of what has been called the Leaning Tower Illusion illustrate this (Kingdom, Yoonsei, and Gheorghiu 2007). Consider the two conjoined pictures in Figure 8.2. People are generally quite surprised to learn that the picture on the left is identical in its surface features to the one on the right. In fact, convincing oneself that they are the same might require rotating them. Because these pictures are close together, it is tempting to see them as one image (Kingdom, Yoonsei, and Gheorghiu 2007). But, as a whole, the spatial parts of the two images have quite different significance, syntactically and semantically, than those same locations do when they are in the separate pictures. With one picture, there are apparently two hallways related to each other in space. With two pictures, neither has anything to say about the spatial parts of the other. Put them close together, however, and one might see the two pictures as one, thus affecting its interpretation. It’s hard to see the surface features of each picture independently of what they might represent. Though these cases are particularly vivid, much the same is

Figure 8.2 Leaning tower illusion. Alteration of Komehakubutsukan Passage, Photographer: As6673. © CC-BY-SA 3.0. Source of original photo: Wikimedia Commons: https://commons.wikimedia.org/wiki/File:Komehakubutsukanpassage.jpg

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

142

   

true of any picture from which spatial parts are extracted. Doing so changes the syntactic and semantic identity of the part removed. All of this should make sense, given the discussion that started this section. In interpreting pictures, one looks for abstractions over the most determinate things a picture can say. That’s why scarlet pictures do not represent being red-or-a-wombat, even though they do represent being red. Similarly, a picture does not represent a color patch as being there, or on the rings of Saturn, or at any other location one might imagine. At its most specific, the picture says something about locations with respect to the other locations that the picture represents. Spatial parts of pictures are thus inseparable syntactic parts.⁶ One can certainly cut off parts of pictures and reassemble them into new ones, just like one can cut written words into letters, and sentences into words. But to know whether such parts are separable requires more than that. Does the mechanical operation of cutting correspond to a syntactic operation that breaks the representation into parts that retain their syntactic identities? In the pictorial case, no. Excision changes the syntactic identity of the thing excised. Remember, having an operation that shows how to transform one representation into others is not the same thing as showing what the syntactic parts of that representation are. One can change “red” to “green” without revealing much of interest about either lexeme. So much for the view that Sober parts are separable. What about the claim that colors are separable from locations? The exemplar of inseparable parts suggests one cannot peel the red off of scarlet. Insofar as that example focuses on the chromatic features that characterize representations, it already abstracts from a more determinate syntactic property the picture or map can have. The picture has the property of being this or that color at location x. Syntactically speaking, the picture is a collection of colors at locations. Every location gets a color, on pain of the picture or map having a hole in it. This is a minimal, but important, condition on well-formedness for pictures and maps. As far as the representation is concerned, anything with a color is a location, and all locations have colors. One can no more peel the color off of a color–location pair than one can peel the red off of scarlet. Colors are not separable from locations. It is easy to think about color independently of location, and vice versa, but the same is true for red and scarlet. Articulate understanding of syntactic ⁶ Camp (2018, §4.4) also notes that maps are syntactically holistic in the way they represent space, though that’s not to say she agrees with the point about separability.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

   

143

parts is not the same thing as those parts being separable. Colors are exemplars of inseparability because they form a perceptual quality space. It’s easy to think of color as a determinable with various indeterminates and determinates. The scarlet–red case is gripping because it’s metaphysically impossible to have something scarlet that is not red. Adding locations to the mix yields amalgamated properties that are not usually talked about as such. Syntactically speaking, however, these odd properties are the most determinate ones possessed by the picture. It’s less gripping, perhaps, that it is syntactically impossible to have a location without a color, but in this case that’s the kind of possibility that matters. Now imagine another feature like texture, and that this feature is independent of the other color properties at a given location. In effect, imagine a map. Any texture can be combined with any color at any location. It might seem as though texture is a syntactic quality separable from color. So, even if colors cannot be separated from locations, perhaps textures separate from colors. This move fails for the same reason the simpler proposal about color and location does. If texture is another feature, even if it’s independent of color in some metaphysical sense, the map commits to both color and texture values for each location, on pain of being ill-formed. So, the most determinate syntactic features of the picture surface are amalgams of color, texture, and location. Every point has both, so it’s combinations of these features that fill locations. You can’t have some color–texture combination without both a color and a texture, and so in that sense color and texture are, within this system, inseparable. Schematically, this argument applies in the same way to cases that include many more than two distinct colors and textures. The three objections to the main claim, though at first glance plausible, turn out to be misguided. It is easy to think about the parts of pictures and maps independently of one another, but that’s not enough to show that those parts are separable.⁷

⁷ John Haugeland suggested that we can distinguish what he called logical representations from iconic ones by appeal to the nature of their contents. The elements of the contents of logical representations “can enter into atomic contents one-by-one, without depending on their concrete relations to one another, if any” (Haugeland 1998, 191). The elements of iconic contents “might be conceived as variations of values along certain dimensions with respect to locations in certain other dimensions” (Haugeland 1998, 192). This has much to recommend it, but here I need to repeat a complaint about this that I have made before (Kulvicki 2006, Ch. 6): Haugeland should have considered the ways in which syntactic structure relates to semantic structure. By focusing exclusively on contents he missed much of importance. My discussion above focuses on syntax, and the way it affects semantics: structure and content. Cf. Heck (2007,

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

144

   

An interesting result follows from these considerations. Pictures and maps are profoundly holistic. At the most specific level, all of their features coalesce into a highly specific whole. Color is inseparable from position, positions here are inseparable from those elsewhere. In reality, it’s misleading to speak of a collection of most specific features for any given representation. It makes perfect sense to talk of a multitude of syntactic features, some more specific than others, but these features form an amalgamated whole, which is the sole most determinate syntactic feature of the picture, or map. Before moving on, let’s return to a worry raised in Chapter 3. It should always be possible to define abstractions over representations, regardless of whether they are linguistic, map-like, or pictorial. Take all the color terms of English, for example, and then imagine abstracting over features of those words. Anything that happens to instantiate “red” also instantiates “red”-or“blue.” So, an abstraction over these shape properties has predictable semantic consequences, viz. the abstraction represents red-or-blue. We can say the same for numerals, words for animals, and any mix of these we like. This is a worry because it seems as though exactly the same syntactic process identified as distinctive of pictures and maps works well in languages too. In some cases, like color terms, it’s relatively easy to imagine playing this abstraction game. In other cases, like animal names, it’s hard because the list is huge and ever-expanding. In all cases, however, this kind of abstraction is not the syntactic operation at work in language. It’s easy to represent an abstraction over red and blue linguistically: red-or-blue. Alternatively, one could invent a new lexeme like “blured” (don’t do this). In neither case is an abstraction over “red” and “blue” eo ipso a new lexeme. Yes, the expression “blue” is “blue”-or-“red,” but no, being so is not a syntactically interesting feature of that lexeme. There is nothing within the linguistic system to suggest that such an abstraction gives you another representation.

8.7 Compositionality and Inseparability Compositionality is a claim about meanings of complexes and how they depend on syntactic combinations of their meaningful parts. Inseparability

135), who says, “We should instead regard the content of a mental state S as encoding not just what S represents, but also how, that is, as encoding S’s compositional structure.”

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

  

145

makes questions of compositionality look rather different than they might at first seem. Remember the definition: The meaning of an expression is a function of the meanings of its parts and of the way in which they are syntactically combined. (Partee 2004, 153)

The one syntactic operation of interest in pictures and maps is abstraction over determinate syntactic detail. No syntactic parts of pictures are separable. Fix the most specific features of pictures or maps, and one thereby fixes the rest. Each picture or map has just one most specific syntactic feature anyway. So, though pictures have syntactic parts, there are not really any options concerning how to combine them. Every change is global. Questions of compositionality seem interesting only when there are parts that can be combined in many different ways. That is exactly what separable parts are like, and not at all what one finds with inseparable parts. So, in a trivial sense, pictures are compositional. A picture cannot have the most determinate syntactic part it has without also having all the other parts. And pictures are meaningful, so of course they are compositional. If pictures and maps are compositional, however, they certainly don’t seem to be so trivially. Moreover, it is common to talk of painters, photographers, cartographers, and the like composing pictures and maps. The process of doing so typically involves building wholes from marks and color. In some cases—chess diagrams (Blumson 2012), maps (Camp 2007, 2018), architectural diagrams (Haugeland 1998)—it feels very much like one is given separable parts to assemble. Casati and Varzi (1999) break maps down into atomic stages, one for each color, which together yield a meaning for the whole. My view of maps is that we need to break them up into incompatibility classes of features, which together give you a whole. Maps can have many different sets of incompatibility classes. Pictures are assembled by putting lines and color, in different combinations, on a surface. Do not mistake recipes for making and breaking representations for descriptions of separable syntactic parts. To make a map of a certain sort, one must place features of this or that sort at every point on its surface. To make a line drawing, lines must be located, weighted, and the like. None of that speaks to whether the syntactic parts are separable. Making a map or a picture is, in effect, deciding what its most determinate syntactic feature is. Such determinate syntactic features are assembled from some properties that, metaphysically speaking, can occur independently of one another, and

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

146

   

for which people have independent concepts. But that’s different from saying that they are separable.⁸ Also, don’t mistake having an articulate understanding of a kind of representation for understanding its separable syntactic parts. One learns a lot about maps by trying to find their incompatibility classes, and one can even derive a compositional semantics from those classes. But an articulate description of incompatibility classes does not mean that they constitute separable syntactic parts of maps. There are pictures and maps of many sorts, and each kind has its own way of recruiting syntactic features to make a whole. So, while it might be uninteresting that pictures and maps are compositional, it is very important to try and give articulate compositional semantics for these different sorts of representations. Doing so shows just how these different kinds of pictures and maps are different.

8.8 Why Non-propositional? Some have suggested that linguistic representations are propositional while pictures and maps are not. Some use the term because they don’t think pictures or maps can express propositions, while linguistic representations can (Crane 2009). Others find the term apt even though they think maps or pictures can express propositions (Grzankowski 2014, Camp 2018). Previous chapters show many ways in which pictures and maps can express propositions, and this chapter yields a new way of thinking about what nonpropositionality amounts to. Giving articulate voice to the meaning of a declarative sentence comes quite close to unpacking its truth conditions. Understanding the meaning of a declarative sentence is in part having a good sense for the conditions under which it would be true or false. Propositions have traditionally been thought of as the bearers of truth or falsity (see, e.g., Hanks 2009). Declarative sentences give voice to, or express, propositions.⁹ So, what makes linguistic ⁸ As mentioned in Chapter 3, both Abell (2005b) and Blumson (2014, Ch. 6) seem on the fence as to whether it is spatial parts or features like colors that count as the syntactic parts of pictures. They don’t have the same notion of part in mind as presented here, but there is nothing objectionable about finding parts to be features, spatial or otherwise. It’s just that in doing so one is abstracting over determinate syntactic detail. ⁹ Unsurprisingly, there is much controversy over the nature of propositions, and especially about how finely they should be individuated. Do two sentences, for example, with the same truth conditions express the same proposition? “Hesperus is a planet.” “Phosphorous is a planet.” Set these debates aside for present purposes.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

 - ? 147 expressions propositional? Generally, each syntactic part and the way it’s combined with others makes a distinctive contribution to those truth conditions. That’s why unpacking the meaning looks a lot like listing truth conditions. Raised on thinking about syntactic parts in language, this seems almost obvious. This is precisely what fails when a representation’s parts are inseparable. Pictures and maps have exactly one most determinate syntactic feature each. The parts are inseparable abstractions from it, and their role is representing abstractions over the most determinate content. As a result, the most determinate syntactic feature fixes truth conditions. The inseparable parts can only make a redundant contribution to them. For example, if the truth-maker for some representation must be scarlet, it’s given that it must also be red. Abstractions over the most determinate content in no way add to, or change the truth conditions of the whole. The picture represents someplace as being that specific shade of scarlet, and red. But anything that shade of scarlet is also red, so fixing the first fixes the other as well. No inseparable syntactic part of a picture or map can add anything to truth conditions already determined by the most determinate syntactic feature. Many philosophers have pointed out that pictures tend to have much more specific contents, as a matter of course, than linguistic representations do.¹⁰ Language, the thought goes, can choose the level of abstraction at which to represent things, while pictures must be quite specific. The point here is not just that pictures are specific. In fact, as Lopes (1996) points out, many pictures can be quite unspecific. The point is that pictures represent features at many levels of abstraction, all at once (Kulvicki 2007, 2010, 2015a). Most of what they represent articulately has little to do with fixing truth conditions. So, while it makes sense to say that pictures and maps have truth conditions, and even that they express propositions when used in certain ways, there is much more to pictorial content than that. Of course, it’s strange to say that there is more to pictorial content, because abstractions over determinate content say less than the whole does. But the point is that pictures are articulate about the abstractions, so they say less, informationwise, in an articulate fashion, and in that sense they say more. It is helpful to consider linguistic examples that are articulate across levels of abstraction. Imagine describing something as scarlet, red, and reddish, all at once in a declarative sentence. “That’s scarlet, red, and reddish!” What ¹⁰ For example, Chisholm (1942) and Dretske (1981) in the philosophy of perception, and Schier (1986) and Hopkins (1998) on pictures.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

148

   

conversational contexts make saying that a reasonable thing to do? The most plausible cases are those in which one is not trying to communicate anything about the color of an object, but about one’s understanding of a concept. “I know what scarlet is! It’s a color that’s red, reddish, and a bit yellowish as well” (See Kulvicki 2007, 366). Otherwise, norms about giving the right amount of information kick in (Grice 1989, e.g.), and it would seem problematic to say so many things. If a conversation demands specific colors, say it’s scarlet, but if less will do, go ahead and say less. Pictures and maps are not as selective in the information that they deliver. Yes, they say quite specific things, but they also say so many non-specific things as well. In that way it makes sense to say that they are non-propositional.

8.9 Summary Perhaps the deepest divide between kinds of representation concerns how representations have parts. Despite appearances to the contrary, pictures, maps, and the like have no separable syntactic parts while linguistic representations only have separable parts. Parts come along with operations defined over them, so another way of putting this is that abstraction over syntactic detail is the only syntactic operation of relevance to pictures while grammatical (de)composition is the one relevant to languages. Linguistic representations are built from parts, while images are broken down into parts. This book began by asking whether semantic tools developed in the philosophy of language could profitably be applied to the study of pictures. Chapters 2, 4, 5, 6, and 7 constitute an affirmative answer. Pictures have non-constant characters, which explains intuitions about how they are interpreted. They have attributive contents that are not marked as either definite or indefinite. In communication, they can be used with either definite or indefinite force. Sometimes, particular individuals are constitutive of the contents of pictures, in at least three ways: via dthat operations, iconographic interpretation, or as the contents of locations, which can serve as directly referring terms. Pictorial metaphor can even be modeled well within this frame. In communication, different parts of pictures can play different roles, which makes pictures particularly useful kinds of representation. Sometimes, one part is responsible for singling out a subject, sometimes it’s attributive, sometimes a part delivers a particular individual, while the

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi



149

other parts constitute a comment on it. In maps and photographs, locations play the role of delivering particular individuals while the other features tell us what’s going on there. Chapter 3 made it clear, however, that picture parts were a bit strange. One finds them by ignoring syntactic detail. As a result, the contents of pictures and maps are distinctive in being articulate across levels of abstraction. This last chapter uses that point to suggest a deep division between kinds of representation. The parts of pictures and maps are fundamentally different from those of linguistic representations. But both are profitably discussed using the semantic tools sourced in language. Semantic tools are thus not local to language. They apply as long as there is some kind of articulate syntax with meanings attached. Though I have a theory of what makes pictorial representation distinctive, this book doesn’t fully presuppose it. You might think that I get many things wrong in that theory and still think that the model introduced here for discussing pictures’ meanings works quite well. There are probably other ways to model the phenomena discussed in this book. In fact, I hope there are. Greenberg’s (2013, 2018) recent work is a promising alternative. I’ve not attempted to build many models and choose the best one because that task goes well beyond what a single book should do.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 28/7/2020, SPi

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 5/8/2020, SPi

References Abell, C. 2005. Against depictive conventionalism. American Philosophical Quarterly 42(3): 185–97. Abell, C. 2009. Canny resemblance. Philosophical Review 118(2): 183–23. Alberti, L.B. 1991/1435. On Painting. C. Grayson, trans. London: Penguin. Aldrich, V. 1968. Visual metaphor. Journal of Aesthetic Education 2(1): 73–86. Bach, K. 1970. Part of what a picture is. British Journal of Aesthetics 10(2): 119–37. Bach, K. 1994. Conversational implicature. Mind and Language 9(2): 124–62. Barthes, R. 1984. Camera Lucida. R. Howard, trans. London: Flamingo. Baxandall, M. 1988. Painting and Experience in Fifteenth-Century Italy, 2nd edn. Oxford: Oxford University Press. Bazin, A. 1971. The ontology of the photographic image. In What is Cinema? Vol. 1. H. Gray, trans. University of California Press. Belting, H. 1994. Likeness and Presence: A History of the Image Before the Era of Art. E. Jephcott, trans. Chicago: University of Chicago Press. Black, M. 1962. Metaphor. In Models and Metaphors. Ithaca, NY: Cornell University Press. Blumson, B. 2010. Maps and meaning. Journal of Philosophical Research 35: 123–28. Blumson, B. 2012. Mental maps. Philosophy and Phenomenological Research 85(2): 413–34. Blumson, B. 2014. Resemblance and Representation. Cambridge: Open Book Publishers. Borg, E. 2002. Pointing at Jack, talking about Jill: Understanding deferred uses of demonstratives and pronouns. Mind and Language 17(5): 489–12. Borg, E. 2004. Minimal Semantics. Oxford: Oxford University Press. Borrelli, A. 2011. Santa Filomena de Roma. http://www.santiebeati.it/dettaglio/65825 (accessed June 10, 2018). Braun, D. 2015. Indexicals. In The Stanford Encyclopedia of Philosophy E. Zalta, ed. https://plato.stanford.edu/archives/sum2017/entries/indexicals/ (accessed April 16, 2020). Bronner, B. 2015. Maps and absent symbols. Australasian Journal of Philosophy 93(1): 43–59. Camp, E. 2007. Thinking with maps. Philosophical Perspectives 21(1): 145–82. Camp, E. 2018. Why maps are not propositional. In Non-Propositional Intentionality. A. Grzankowski and M. Montague, eds. Oxford: Oxford University Press, 19–45. Cappelen, H. and E. Lepore. 2005. Insensitive Semantics. London: Blackwell. Carroll, N. 1994. Visual metaphor. In Aspects of Metaphor. J. Hintikka, ed. Kluwer: 189–18. Casati, R. and A. Varzi. 1999. Parts and Places. Cambridge, MA: MIT Press.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 5/8/2020, SPi

152



Chisholm, R. 1942. The problem of the speckled hen. Mind 51(204): 368–73. Crane, T. 2009. Is perception a propositional attitude? Philosophical Quarterly 59(236): 452–69. Cumming, S., G. Greenberg, and R. Kelly. 2017. Conventions of viewpoint coherence in film. Philosophers’ Imprint 17: 1–28. Danto, A. 1964. The artworld. Journal of Philosophy 61(19): 571–84. De Clercq, R. 2015. The lazy person’s approach to depiction. Journal of Aesthetics and Art Criticism 73(2): 95–104. Devitt, M. 1981. Designation. New York: Columbia University Press. Dickie, I. 2011. How proper names refer. Proceedings of the Aristotelian Society 111: 43–78. Donnellan, K. 1966. Reference and definite descriptions. Philosophical Review 75(3): 281–304. Dretske, F. 1981. Knowledge and the Flow of Information. Cambridge, MA: MIT Press. Evans, G. 1982. Varieties of Reference. Oxford: Clarendon Press. Fodor, J. 1998. Concepts: Where Cognitive Science Went Wrong. New York: Oxford University Press. Fodor, J. 2007. The revenge of the given. In Contemporary Debates in Philosophy of Mind. B. McLaughlin and J. Cohen, eds. Malden, MA: Blackwell: 105–16. Fodor, J. 2008. LOT II. Cambridge, MA: MIT Press. Freeland, C. 2007. Portraits in painting and photography. Philosophical Studies 135: 95–109. Freeland, C. 2010. Photographs and icons. In Photography and Philosophy, S. Walden ed. Oxford: Blackwell, 50–69. Friday, J. 2005. Andre Bazin’s ontology of photographic and film imagery. Journal of Aesthetics and Art Criticism 63(4): 339–50. Gilot, F. and C. Lake. 1964. Life with Picasso. London: Virago Press. Gilot, F. and C. Lake. 2006. Vivre avec Picasso. 10x18 Press. Glanzberg, M. 2012. Not all contextual parameters are alike. Unpublished ms. https://sites.northwestern.edu/mjg904/research/ Gombrich, E.H. 1951. Meditations on a hobby horse. In Aspects of Form L. L. Whyte, ed. London: Lund Humphries: 209–24. Gombrich, E.H. 1961. Art and Illusion, 2nd ed. Princeton: Princeton University Press. Gombrich, E. 1963. Meditations on a Hobby Horse and Other Essays on the Theory of Art. London: Phaidon. Goodman, N. 1951. The Structure of Appearance. Cambridge, MA: Harvard University Press. Goodman, N. 1976. Languages of Art, 2nd ed. Indianapolis: Hackett. Green, E.J. 2017. A layered view of shape perception. British Journal for the Philosophy of Science 68(2): 355–87. Greenberg, G. 2013. Beyond resemblance. Philosophical Review 122(2): 215–87. Greenberg, G. 2018. Content and target in pictorial representation. Ergo 5(33): 865–98. Grice, H.P. 1989. Studies in the Way of Words. Cambridge, MA: Harvard University Press.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 5/8/2020, SPi



153

Grzankowski, A. 2014. Pictures have propositional content. Review of Philosophy and Psychology 6(1): 151–63. Hanks, P. 2009. Recent work on propositions. Philosophy Compass 4(3): 469–86. Hansen, N. 2011. Color adjectives and radical contextualism. Linguistics and Philosophy 34: 201–21. Haugeland, J. 1998. Having Thought. Cambridge, MA: Harvard University Press. Heck, R. 2007. Are there different kinds of content? In Contemporary Debates in Philosophy of Mind. B. McLaughlin and J. Cohen, eds. Malden, MA: Blackwell: 117–38. Hope, C. 1986. Religious narrative in Renaissance art. Journal of the Royal Society of Arts 134(5364): 804–18. Hopkins, R. 1995. Explaining depiction. Philosophical Review 104(3): 425–55. Hopkins, R. 1998. Picture, Image, and Experience. Cambridge: Cambridge University Press. Husserl, E. 2006. Phantasy, Image-Consciousness, and Memory. J. Brough, trans. Dordrecht: Springer. Hyman, J. 1989. The Imitation of Nature. Oxford: Blackwell. Hyman, J. 2000. Pictorial art and visual experience. British Journal of Aesthetics 40(1): 21–45. Hyman, J. 2006. The Objective Eye. Chicago: University of Chicago Press. Hyman, J. 2012. Depiction. Royal Institute of Philosophy Supplement 71:129–50. Hyman, J. 2013. Depiction. Royal Institute of Philosophy Supplement 71: 129–50. Ittleson, W.H. 1952. The Ames Demonstrations in Perception. Princeton: Princeton University Press. James, L. 2005. Seeing is believing but words tell no lies. In Negating the Image. A. McClanan and J. Johnson, eds. Farnham: Ashgate: 97–112. Johnson, K. 2006. On the nature of reverse compositionality. Erkenntnis 64(1): 139–68. Kaplan, D. 1968. Quantifying in. Synthese 19(1–2): 178–14. Kaplan, D. 1978. Dthat. In Syntax and Semantics. Academic Press: 221–43. Kaplan, D. 1989. The logic of demonstratives. In Themes from Kaplan. J. Almog, J. Perry, and H. Wettstein, eds. Oxford: Oxford University Press. Kaplan, D. 1990. Words. Proceedings of the Aristotelian Society (Supp.) 64: 93–119. Kemp, M. ed. 1989. Leonardo on Painting. New Haven, CT: Yale University Press. Kennedy, C. 2007. Vagueness and grammar: the semantics of relative and absolute gradable adjectives. Linguistics and Philosophy 30: 1–45. Kennedy, C. and L. McNally. 2010. Color, context, and compositionality. Synthese 174(1): 79–98. Kennedy, J. 1982. Metaphor in pictures. Perception 11: 589–05. King, J. 2014. The metasemantics of contextual sensitivity. In Metasemantics: New Essays on the Foundations of Meaning A. Burgess and B. Sherman, eds. New York: Oxford University Press, 97–118. Kingdom, F.A.A., A. Yoonsei, and E. Gheorghiu. 2007. The leaning tower illusion: A new illusion of perspective. Perception 36(3): 475–77. Kitcher, P. and A. Varzi. 2000. Some pictures are worth 2‫א‬o sentences. Philosophy 75: 377–81.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 5/8/2020, SPi

154



Korta, K. and J. Perry. 2008. The pragmatic circle. Synthese 165(3): 347–57. Kosslyn, S. 1980. Image and Mind. Cambridge, MA: Harvard University Press. Kosslyn, S. 1994. Image and Brain. Cambridge, MA: MIT Press. Kripke, S. 1977. Speaker’s reference and semantic reference. Midwest Studies in Philosophy 2: 255–76. Kripke, S. 1980. Naming and Necessity. Cambridge, MA: Harvard University Press. Kulvicki, J. 2006. On Images: Their Structure and Content. Oxford: Clarendon Press. Kulvicki, J. 2007. Perceptual content is vertically articulate. American Philosophical Quarterly 44(4): 357–69. Kulvicki, J. 2010. Knowing with images: Medium and message. Philosophy of Science 77(2): 295–313. Kulvicki, J. 2014. Images. London: Routledge. Kulvicki, J. 2015a. Analog representation and the parts principle. Review of Philosophy and Psychology 6(1): 165–80. Kulvicki, J. 2015b. Maps, pictures, and predication. Ergo 2(7): 149–74. Kulvicki, J. 2017. Recording and representing, analog and digital. In Giving a Damn: Essays in Dialogue with John Haugeland. Z. Adams, ed. Cambridge, MA: MIT Press. Kulvicki, J. 2020. Chromatic Perspectives. In The Routledge Handbook of Philosophy of Color. D. Brown and F. MacPherson, eds. London: Routledge, 91–106. Lepore, E. and M. Stone. 2014. Imagination and Convention. New York: Oxford University Press. Lopes, D. 1996. Understanding pictures. Oxford: Clarendon Press. Lopes, D. 2009. Drawing in a social science: lithic illustration. Perspectives on Science 17(1): 5–25. Lopes, D. 2010. Picture this: Image-based demonstratives. In Philosophical Perspectives on Depiction C. Abell and K. Bantinaki, eds. Oxford: Oxford University Press, 52–80. Ludlow, P. 2018. Descriptions. The Stanford Encyclopedia of Philosophy (Summer 2018 Edition), E. Zalta, ed., URL = . Maes, H. 2015. What is a portrait? British Journal of Aesthetics 55(3): 303–22. Martinich, A. P. ed. 1996. The Philosophy of Language. New York: Oxford University Press. Maynard, P. 1997. The Engine of Visualization. Ithaca: Cornell University Press. Michaelson, E. 2014. Shifty characters. Philosophical Studies 167(3): 519–40. Neander, K. 1987. Pictorial representation: a matter of resemblance. British Journal of Aesthetics 27(3): 213–26. Novitz, D. 1977. Pictures and their Use in Communication. The Hauge: Martinus Nijhoff. Panofsky, E. 1955. Meaning in the Visual Arts. Garden City, NY: Doubleday. Partee, B. 2004. Compositionality in Formal Semantics: Selected Papers of Barbara Partee. London: Blackwell. Perini, L. 2005. Explanation in two dimensions: diagrams and biological explanation. Biology and Philosophy 20: 257–69.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 5/8/2020, SPi



155

Phillips, D. 2009. Photography and causation. British Journal of Aesthetics 49(4): 327–40. Podro, M. 1987. Depiction and the golden calf. In Philosophy and Visual Art A. Harrison, ed. Dordrecht: Reidel, 3–22. Polanyi, M. 1970. What is a painting? British Journal of Aesthetics 10: 225–36. Quine, W.V.O. 1968. Ontological Relativity and Other Essays. New York: Columbia University Press. Recanati, F. 2002. Literal Meaning. Cambridge: Cambridge University Press. Recanati, F. 2004. Descriptions and situations. In Descriptions and Beyond. M. Reimer and A. Bezuidenhout, eds. Oxford: Oxford University Press: 15–40. Reimer, M. and E. Michaelson. 2014. Reference. The Stanford Encyclopedia of Philosophy, Edward N. E. Zalta, ed., URL = . Rescorla, M. 2009. Predication and cartographic representation. Synthese 169(1): 175–200. Robbins, P. 2005. The myth of reverse compositionality. Philosophical Studies 125(2): 251–75. Sartwell, C. 1991. Natural generativity and imitation. British Journal of Aesthetics 31(1): 58–67. Schier, F. 1986. Deeper into Pictures. Cambridge: Cambridge University Press. Schoenkopf, R. 2017. https://wonkette.com/619957/sarah-palin-literally-just-wentfull-nazi-this-is-not-a-drill Shuham, M. 2017. Did Sarah Palin really reference white supremacy on Twitter? (Probably not.) Talking Points Memo: http://talkingpointsmemo.com/livewire/ sarah-palin-tweet-14-words Silk, A. 2018. Discourse Contextualism. Oxford: Oxford University Press. Sober, E. 1976. Mental representations. Synthese 33: 101–48. Sontag, S. 1977. On Photography. New York: Farrar, Strauss, and Giroux. Sperber, D. and D. Wilson. 1995. Relevance, 2nd ed. London: Blackwell. Sperber, D. and D. Wilson. 2012. Meaning and Relevance. Cambridge: Cambridge University Press. Spinicci, P. 2009. Portraits, some phenomenological remarks. Proceedings of the European Society for Aesthetics 1: 37–59. Stalnaker 1984. Inquiry. Cambridge: Cambridge University Press. Stanley, J. and Z. Szabó. 2000. On quantifier domain restriction. Mind and Language 15: 219–61. Stern, J. 1997. Metaphors in pictures. Philosophical Topics 25(1): 255–93. Stern, J. 2000. Metaphor in Context. Cambridge, MA: MIT Press. Strawson, P. 1950. On referring. Mind 59: 320–34. Szabó, Z.G. ed. 2005. Semantics versus Pragmatics. Oxford: Clarendon Press. Szabó, Z.G. 2017. Compositionality. In The Stanford Encyclopedia of Philosophy. E. Zalta, ed. URL = https://plato.stanford.edu/archives/sum2017/entries/compositionality/. Taylor, P. 2011. Rembrandt’s “Injustice of Piso”. Oud Holland 124(2/3): 112–18. Taylor, P. ms. Iconography. Thomason, R. ed. 1974. Formal Philosophy: Selected Papers of Richard Montague. New Haven and London:Yale University Press.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 5/8/2020, SPi

156



Thurston, H.J. and D. Attwater, eds. 1956. Butler’s Lives of the Saints, vol. 4. New York: P.J. Kenedy and Sons. Tye, M. 1991. The Imagery Debate. Cambridge, MA: MIT Press. De Voragine, J. 1993. The Golden Legend. W. Ryan, trans. Princeton: Princeton University Press. Walton, K. 1973. Pictures and make-believe. Philosophical Review 82(3): 283–19. Walton, K. 1984. Transparent pictures. Critical Inquiry 11(2): 246–77. Walton, K. 1990. Mimesis as Make-Believe. Cambridge, MA: Harvard University Press. Wiesing, L. 2010. Artificial Presence. N.F. Schott, trans. Stanford, CA: Stanford University Press. Wollheim, R. 1980. Art and its Objects, 2nd ed. Cambridge: Cambridge University Press. Zeimbekis, J. 2010. Pictures and singular thought. Journal of Aesthetics and Art Criticism 68(1): 11–21.

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 26/7/2020, SPi

Index For the benefit of digital users, indexed terms that span two pages (e.g., 52–53) may, on occasion, appear on only one of those pages. Abell, C. 2 n.1, 5, 6 n.2, 14, 23, 46 n.7, 75–6, 146 n.8 absence intuition see maps, and the absence intuition abstraction and pictorial content 9–10, 14–15, 35, 39–48, 50–1, 93–4, 138–44, 147–8 as a syntactic operation 14–17, 39–48, 120–1, 132, 135–6, 138–45, 146 n.8, 147–8 acts, communicative see communicative acts adjectives, gradable 8, 8 n.3, 19 n.1, 22 advertising 50, 106, 108–9, 113, 117 Alberti, L.B. 45, 89–90 Aldrich, V. 104 n.2 ambiguity 20, 25, 79, 85, 118 Ames demonstrations 23–5, 27–30 analog see representation, analog vs digital Attwater, D. 79 n.1 babies, taking candy from 55–6, 62–3 baboon 104–6, 108 Bach, K. 4, 132 n.2 Barthes, R. 119 n.2 St Bartholomew 78, 85 Baxandall, M. 89–90 Bazin, A. 119 n.2 Belguim 61 Belting, H. 95 n.5 Black, M. 112 n.6 Blumson, B. 2 n.1, 6 n.2, 14, 46 n.7, 75, 120 n.3, 124, 134, 145, 146 n.8 Borg, E. 4, 22 n.4, 61, 62 n.3, 103 Borrelli, A. 85 n.3 Braun, D. 8 n.3 Bronner, B. 124 Camp, E. 2 n.1, 45–6, 120 n.3, 121 n.4, 124, 127, 133–5, 142 n.6, 145–6 Candies, V. 59–60

Cappelen, H. 4, 22 n.4 captions see labels Carroll, N. 104–5, 104 n.2, 106 Casati, R. 15–16, 120 n.3, 121 n.4, 122–6, 138, 145 Catherine of Alexandria 10–11, 79, 83–4 Cézanne, P. 69–70 character constant vs non-constant 8, 16, 22, 32, 35–6, 43, 57–8, 61, 76–7, 118, 126–30, 148 vs content 7–14, 19–22, 25–8, 32–3, 35–7, 40, 52–9, 61, 66–7, 76–8, 83, 85–6, 94, 96, 98–100, 111, 116–17, 126–7, 148 in maps 118, 126–30 chartreuse 22, 35, 42 Chisholm, R. 147 n.10 Christ, J. 80–1 chromatic features see features, chromatic circumstances of evaluation 20–1, 32–3, 58–9, 61, 64, 67, 111 coded speech 96–8 Cohen, L. 24 n.5 comics and iconography 95–6 and maps see maps, and comics and metaphor 105 communicative acts 1–7, 12–13, 15, 19, 23, 33–4, 48–9, 51–4, 66–7, 71, 73, 76, 78, 86, 94, 106–8, 110–17, 131, 147–9 compositionality 131–3, 137–8, 144–6 content bare bones 9–10, 12–13, 15–16, 22, 25–32, 35, 39–40, 40 n.3, 41 n.4, 42–3, 58, 76, 118, 127–9 vs character see character, vs content and denotation 10–11, 21, 31–2, 34–6, 50–1, 54–6, 58, 63–4, 66–73, 77, 79, 84, 98, 120

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 26/7/2020, SPi

158



content (cont.) descriptive (attributive) 6–11, 13, 15–18, 30–3, 32 n.9, 35–7, 48–51, 53–60, 63–4, 66–7, 70–5, 75 nn.6,7, 77, 85–6, 90–1, 95–8, 118–21, 138, 148–9 dthat see dthat fleshed out see content, pictorial iconographic see iconography levels of 14–15, 19, 34, 41–2, 44, 50–1, 74–6, 136, 147–9 mthat see mthat pictorial 7–13, 15–17, 20, 22, 28–37, 39–43, 41 n.4, 48, 50, 53, 55–6, 58–61, 63–4, 66–9, 71, 74–9, 81–7, 89–90, 93–5, 97–8, 100–1, 113–15, 127, 129, 147 propositional vs non-propositional 44 n.5, 126, 131, 135, 137, 146–8 and reference 8, 10, 15–16, 19–20, 21 n.2, 32–5, 50–1, 53–4, 56, 59, 59 n.2, 61–7, 73–4, 75 n.6, 77, 81–6, 88–91, 93–8, 119–23, 129–30, 138, 148 singular (particular) 6, 8–10, 13, 15–18, 30–2, 32 n.9, 36, 53, 57–8, 67, 74–9, 75 nn.6,7, 85–6, 94, 98, 100, 148–9 context 3–10, 12–13, 15, 19, 37, 39–40, 42–3, 48–59, 61–2, 65–73, 76–7, 83–4, 100–3, 105–6, 108, 111–17, 120, 126–7, 147–8 and character see character and content see content of production 58–9, 61, 65–8 of use 7, 20, 32, 65–6, 100, 105, 114–15, 120 conversational acts see communicative acts copying 116 Crane, T. 5, 146 Cumming, S. 6 n.2 Danto, A. 22 De Clercq, R. 2 n.1 demonstration 28, 32, 57, 61–2 demonstrations, Ames see Ames demonstrations demonstrative 5, 8, 8 n.3, 19–22, 26, 28, 32 n.9, 33–4, 57, 59 n.2 denotation see reference, denotation depiction, theories of 17–18, 58–61 description see content, attributive and accuracy 34, 42, 51, 120 attributive vs referential use of 6, 53–8

definite vs indefinite 7, 32–4, 48–9, 71–3, 148 incomplete 59 n.2, 69 n.5 Devitt, M. 122 n.6 Dickie, I. 122 n.6 digital see representation, analog vs digital Diocletian 83–5 dogwhistles see coded speech Donnellan, K. 10, 53–6 Dretske, F. 147 n.10 dthat 10–13, 15, 56–62, 67, 77–8, 83–4, 88, 98, 111, 148 Dürer, A. 86–8 elephants 8 Evans, G. 32 n.9, 122 n.6 expression types 58–9, 99, 116 Fiquet, M-H. 69–70 features, spatial 14–15, 26–8, 39–43, 46–7, 135, 139–42, 146 n.8 features, chromatic 16, 26–8, 39–41, 40 n.3, 42–3, 120–1, 139, 142 Fodor, J. 14, 45, 48, 134 n.3 Freeland, C. 65 n.4, 70, 95 n.5 Friday, J. 119 n.2 Georgia 59 Gheorghiu, E. 141–2 Ghosh, A. 21, 57–8 Gillray, J. 101 Gilot, F. 104–5 Glanzberg, M. 22 n.4 Gombrich, E. 22, 25–7, 29–30, 35, 75, 105 Goodman, N. 2, 2 n.1, 26 n.6, 75 n.7, 99, 116, 132 n.2, 134 grammar 3, 5–6, 16, 37–9, 48–9, 120, 134–5 Green, E.J. 44 Grice, H.P. 2 n.1, 4, 111–12, 147–8 Grzankowski, A. 146 Hanks, P. 146–7 Hansen, N. 22 n.3 Haugeland, J. 22, 26–7, 131 n.1, 143 n.7, 145 Heck, R. 139–40, 143 n.7 Hip-Hop 133–4 Holbein, H. 69–70 Hope, C. 88–93 Hopkins, R. 23, 75–6, 147 n.10

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 26/7/2020, SPi

 Husserl, E. 75 n.7 Hyman, J. 2 n.1, 23, 27 n.8, 75, 75 n.6, 75 n.7 icons 9–10, 12–13, 72–3, 79, 84, 93, 95 iconography 6, 10–13, 15, 27, 36, 50, 66, 74, 77–8, 100, 111, 115, 148 as digital representation 94 ithat™ 84 and language 96–8 indexical 3–4, 8, 8 n.3, 19–22, 26, 28, 32 Ittelson, W.H. 23 James, L. 87–8 Johnson, K. 134 n.3 Kaplan, D. 8, 10, 10 n.4, 19–22, 28, 32, 36, 38, 53–9, 61–2, 77, 111, 122–3 Kelly, R. 6 n.2 Kemp, M. 89–90 Kennedy, C. 8 n.3, 19 n.1, 22 n.3 Kennedy, J. 105 King, J. 22 n.4, 62 n.3 Kingdom, F.A.A. 141–2 Kitcher, P. 33, 120 n.3 Korta, K. 4 Kosslyn, S. 45 n.6 Kosuth, J. 22 Kripke, S. 53, 122 n.6 labels 86–9, 91–2, 104–5 Lake, C. 104–5 Lepore, E. 4, 22 n.4 letters 2–4, 38, 95, 142 Levine, S. 22–3 lexeme 5, 16, 47, 94, 142, 144 locations at night 67–8 as syntactic parts 15–16, 118–23, 125, 127–8, 138–44, 148–9 Lollipops, H. 55–6, 59–60 Lopes, D. 2 n.1, 23, 32 n.9, 65 n.4, 75–6, 75 n.7, 119, 147 Ludlow, P. 7, 33 McNally, L. 22 n.3 Maes, H. 65 n.4 maps and the absence intuition 123–6 and character 16, 76–7, 118, 126–7

159

and comics 118–19, 127–30 and pictures 2 n.1, 6, 14–18, 43, 46 n.7, 105–6, 118–20, 127–9, 131, 134–6, 144–7 and propositions 146–8 and separable syntactic parts 137–44 and singular content 15–16, 118, 120–3 Martians 43 Martinich, A.P. 54 n.1, 58 Maxentius 79–80, 83, 88–9, 98 Maynard, P. 95 n.5, 121 n.5 metaphor illustrated 101, 114–15 and m-association 111–13, 115 mthat 110–15 vs other non-literal use 88, 97, 103–6 as semantic 12–13, 36, 110–12, 148 strictly pictorial 100–1, 106–10, 113–14 suggested 101–3, 114–15 supplemental 103, 114–15 as use 6, 12–13, 100 Michaelson, E. 20, 67, 122 n.6 Montague, R. 3–4 More, Sir Thomas 69–70 Morris, C. 3 MST3000 81–2 mthat see metaphor, mthat Neander, K. 23, 75 New Jersey 59–61, 63–4 Nicaea, Second Council of 88 Nixon, R. 17–18, 31–2, 34, 36 non-propositional content see content, propositional vs non-propositional Novitz, D. 2 n.1, 15 n.5, 33 n.10 numerals 44–5, 144 oak, red 19 Obama, B.H. 8, 10, 17 ostention, deferred 61–5 Palma Vecchio 80–5, 89 Panofsky, E. 78, 83, 94–5, 97–8 Partee, B. 3–4, 132, 145 parts, syntactic and abstraction 14–17, 39–45, 135–6 in language 3–4, 6, 16, 20–1, 25–6, 38, 132–7 of pictures 5–6, 14–17, 25–6, 35, 37, 73–4, 116–17, 120–1, 129–30, 132, 136–7

OUP CORRECTED AUTOPAGE PROOFS – FINAL, 26/7/2020, SPi

160



parts, syntactic (cont.) and semantic roles 15, 48–50, 65–6, 71, 73–4, 77, 90–1, 95–6, 98, 113, 120 separable vs inseparable 133–44 and syntactic operations 16, 45–7, 134–5, 148 Parts Principle 45–8, 135–6, 138 Perini, L. 26 n.6, 132 n.2 Perry, J. 4 Phillips, Dawn see Wilson, Dawn St Philomena 84–5 photographs rosters 71, 73 and singular content 9–10, 15–16, 26–7, 30–1, 36, 65–6, 118–22, 129–30, 148–9 vacation 48–52, 65–6, 73–4 Picasso, P. 104–6, 108 Picture Principle 45, see also Parts Principle places, Podunk 67–8 Podro, M. 75 Polanyi, M. 23, 24 n.5 police sketches 9–10, 53, 55–6, 59–65, 67, 76 portraits 9–10, 31–2, 53, 65–73, 95, 98 postcards 9–10, 51, 53, 65–73 pragmatics 2–6, 12, 33, 35, 53, 73–4, 78, 86, 88, 112–13, 115–17, 124 predication 48–50, 73, 82–3, 89, 123–4, 126 pre-semantic judgments 38 presence in maps and photographs see maps, and singular content presupposition 111–13 propositional content see content, propositional vs non-propositional puce 43–4

Schier, F. 23, 75, 147 n.10 Schoenkopf, R. 96–7 St Sebastian 80–7 security cameras 15–16, 118–19 semantics see content vs pragmatics 4–5 sentence 1, 3–4, 15–16, 20–1, 27 n.7, 38, 54, 57, 94, 103, 111, 134, 142, 146–8, 146 n.9 Shelley, Percy Bysshe 21, 57–8 Shuham, M. 97 n.7 Silk, A. 22 n.4 Sober, E. 14, 45–50, 134–5, 137, 140, 142 Sontag, S. 119 n.2 spatial features see features, spatial Sperber, D. 4 Spinicci, P. 65 n.4 Stalnaker, R. 112 Stanley, J. 59 n.2 Stern, J. 12, 32 n.9, 36, 99–101, 104–6, 108–17, 132 n.2 stories vs images 88–93 Strawson, P. 69 n.5 syntax 2–6 see also parts, syntactic Szabo, Z. 4, 59 n.2, 133

Quine, W.V.O 61

Varzi, A. 15–16, 33, 120 n.3, 121 n.4, 122–6, 138, 145 Venus 87–8 Vermillion 34–5, 40–3 De Voragine, J. 79

Raphael 79, 82–4, 88–9, 91 Recanati, F. 4, 27 n.7, 59 n.2 recognition theory 2, 17, 23, 75–6 reference see content, and reference Reimer, M. 122 n.6 representation, analog vs digital 43, 94, 133–5 reproductions 24 Rescorla, M. 120 n.3, 123–4, 126, 134 resemblance theory 2, 17, 23, 75–6 Robbins, P. 134 n.3 room, wiggle 54

Taylor, P. 79, 79 nn.1, 2, 86, 92, 95 n.5 Theodulf of Orleans 87–8, 95 Thomason, R. 3–4 Thurston, H.J. 79 n.1 truth conditions 3–4, 27 n.7, 58, 146–7 Tye, M. 45 n.6 universe of our experience 29

Walton, K. 24 n.5, 75–6, 119, 119 n.2 Wiesing, L. 75 n.7 Wilson, Dawn 121 n.5 Wilson, Dierdre. 4 Wimsatt, W. 104–5 Wollheim, R. 23, 24 n.5, 75 n.7 wombat 2, 5, 8, 29–30, 138–9, 142 Yoonsei, A. 141–2

Sartwell, C. 23, 75 scarlet 120–1, 136–9, 142–3, 147–8

Zeimbekis, J. 2 n.1